diff --git a/cts/lab/CTStests.py b/cts/lab/CTStests.py index c5c7393bdc..649489cf6f 100644 --- a/cts/lab/CTStests.py +++ b/cts/lab/CTStests.py @@ -1,3142 +1,3142 @@ """ Test-specific classes for Pacemaker's Cluster Test Suite (CTS) """ __copyright__ = "Copyright 2000-2021 the Pacemaker project contributors" __license__ = "GNU General Public License version 2 or later (GPLv2+) WITHOUT ANY WARRANTY" # # SPECIAL NOTE: # # Tests may NOT implement any cluster-manager-specific code in them. # EXTEND the ClusterManager object to provide the base capabilities # the test needs if you need to do something that the current CM classes # do not. Otherwise you screw up the whole point of the object structure # in CTS. # # Thank you. # import os import re import time import subprocess import tempfile from stat import * from cts import CTS from cts.CTSaudits import * from cts.CTSvars import * from cts.patterns import PatternSelector from cts.logging import LogFactory from cts.remote import RemoteFactory, input_wrapper from cts.watcher import LogWatcher from cts.environment import EnvFactory AllTestClasses = [ ] class CTSTest(object): ''' A Cluster test. We implement the basic set of properties and behaviors for a generic cluster test. Cluster tests track their own statistics. We keep each of the kinds of counts we track as separate {name,value} pairs. ''' def __init__(self, cm): #self.name="the unnamed test" self.Stats = {"calls":0 , "success":0 , "failure":0 , "skipped":0 , "auditfail":0} # if not issubclass(cm.__class__, ClusterManager): # raise ValueError("Must be a ClusterManager object") self.CM = cm self.Env = EnvFactory().getInstance() self.rsh = RemoteFactory().getInstance() self.logger = LogFactory() self.templates = PatternSelector(cm["Name"]) self.Audits = [] self.timeout = 120 self.passed = 1 self.is_loop = 0 self.is_unsafe = 0 self.is_docker_unsafe = 0 self.is_experimental = 0 self.is_container = 0 self.is_valgrind = 0 self.benchmark = 0 # which tests to benchmark self.timer = {} # timers def log(self, args): self.logger.log(args) def debug(self, args): self.logger.debug(args) def has_key(self, key): return key in self.Stats def __setitem__(self, key, value): self.Stats[key] = value def __getitem__(self, key): if str(key) == "0": raise ValueError("Bad call to 'foo in X', should reference 'foo in X.Stats' instead") if key in self.Stats: return self.Stats[key] return None def log_mark(self, msg): self.debug("MARK: test %s %s %d" % (self.name,msg,time.time())) return def get_timer(self,key = "test"): try: return self.timer[key] except: return 0 def set_timer(self,key = "test"): self.timer[key] = time.time() return self.timer[key] def log_timer(self,key = "test"): elapsed = 0 if key in self.timer: elapsed = time.time() - self.timer[key] s = key == "test" and self.name or "%s:%s" % (self.name,key) self.debug("%s runtime: %.2f" % (s, elapsed)) del self.timer[key] return elapsed def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not name in self.Stats: self.Stats[name] = 0 self.Stats[name] = self.Stats[name]+1 # Reset the test passed boolean if name == "calls": self.passed = 1 def failure(self, reason="none"): '''Increment the failure count''' self.passed = 0 self.incr("failure") self.logger.log(("Test %s" % self.name).ljust(35) + " FAILED: %s" % reason) return None def success(self): '''Increment the success count''' self.incr("success") return 1 def skipped(self): '''Increment the skipped count''' self.incr("skipped") return 1 def __call__(self, node): '''Perform the given test''' raise ValueError("Abstract Class member (__call__)") self.incr("calls") return self.failure() def audit(self): passed = 1 if len(self.Audits) > 0: for audit in self.Audits: if not audit(): self.logger.log("Internal %s Audit %s FAILED." % (self.name, audit.name())) self.incr("auditfail") passed = 0 return passed def setup(self, node): '''Setup the given test''' return self.success() def teardown(self, node): '''Tear down the given test''' return self.success() def create_watch(self, patterns, timeout, name=None): if not name: name = self.name return LogWatcher(self.Env["LogFileName"], patterns, name, timeout, kind=self.Env["LogWatcher"], hosts=self.Env["nodes"]) def local_badnews(self, prefix, watch, local_ignore=[]): errcount = 0 if not prefix: prefix = "LocalBadNews:" ignorelist = [] ignorelist.append(" CTS: ") ignorelist.append(prefix) ignorelist.extend(local_ignore) while errcount < 100: match = watch.look(0) if match: add_err = 1 for ignore in ignorelist: if add_err == 1 and re.search(ignore, match): add_err = 0 if add_err == 1: self.logger.log(prefix + " " + match) errcount = errcount + 1 else: break else: self.logger.log("Too many errors!") watch.end() return errcount def is_applicable(self): return self.is_applicable_common() def is_applicable_common(self): '''Return TRUE if we are applicable in the current test configuration''' #raise ValueError("Abstract Class member (is_applicable)") if self.is_loop and not self.Env["loop-tests"]: return 0 elif self.is_unsafe and not self.Env["unsafe-tests"]: return 0 elif self.is_valgrind and not self.Env["valgrind-tests"]: return 0 elif self.is_experimental and not self.Env["experimental-tests"]: return 0 elif self.is_docker_unsafe and self.Env["docker"]: return 0 elif self.is_container and not self.Env["container-tests"]: return 0 elif self.Env["benchmark"] and self.benchmark == 0: return 0 return 1 def find_ocfs2_resources(self, node): self.r_o2cb = None self.r_ocfs2 = [] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rtype == "o2cb" and r.parent != "NA": self.debug("Found o2cb: %s" % self.r_o2cb) self.r_o2cb = r.parent if re.search("^Constraint", line): c = AuditConstraint(self.CM, line) if c.type == "rsc_colocation" and c.target == self.r_o2cb: self.r_ocfs2.append(c.rsc) self.debug("Found ocfs2 filesystems: %s" % repr(self.r_ocfs2)) return len(self.r_ocfs2) def canrunnow(self, node): '''Return TRUE if we can meaningfully run right now''' return 1 def errorstoignore(self): '''Return list of errors which are 'normal' and should be ignored''' return [] class StopTest(CTSTest): '''Stop (deactivate) the cluster manager on a node''' def __init__(self, cm): CTSTest.__init__(self, cm) self.name = "Stop" def __call__(self, node): '''Perform the 'stop' test. ''' self.incr("calls") if self.CM.ShouldBeStatus[node] != "up": return self.skipped() patterns = [] # Technically we should always be able to notice ourselves stopping patterns.append(self.templates["Pat:We_stopped"] % node) # Any active node needs to notice this one left # (note that this won't work if we have multiple partitions) for other in self.Env["nodes"]: if self.CM.ShouldBeStatus[other] == "up" and other != node: patterns.append(self.templates["Pat:They_stopped"] %(other, self.CM.key_for_node(node))) #self.debug("Checking %s will notice %s left"%(other, node)) watch = self.create_watch(patterns, self.Env["DeadTime"]) watch.setwatch() if node == self.CM.OurNode: self.incr("us") else: if self.CM.upcount() <= 1: self.incr("all") else: self.incr("them") self.CM.StopaCM(node) watch_result = watch.lookforall() failreason = None UnmatchedList = "||" if watch.unmatched: (rc, output) = self.rsh(node, "/bin/ps axf", None) for line in output: self.debug(line) (rc, output) = self.rsh(node, "/usr/sbin/dlm_tool dump 2>/dev/null", None) for line in output: self.debug(line) for regex in watch.unmatched: self.logger.log ("ERROR: Shutdown pattern not found: %s" % (regex)) UnmatchedList += regex + "||"; failreason = "Missing shutdown pattern" self.CM.cluster_stable(self.Env["DeadTime"]) if not watch.unmatched or self.CM.upcount() == 0: return self.success() if len(watch.unmatched) >= self.CM.upcount(): return self.failure("no match against (%s)" % UnmatchedList) if failreason == None: return self.success() else: return self.failure(failreason) # # We don't register StopTest because it's better when called by # another test... # class StartTest(CTSTest): '''Start (activate) the cluster manager on a node''' def __init__(self, cm, debug=None): CTSTest.__init__(self,cm) self.name = "start" self.debug = debug def __call__(self, node): '''Perform the 'start' test. ''' self.incr("calls") if self.CM.upcount() == 0: self.incr("us") else: self.incr("them") if self.CM.ShouldBeStatus[node] != "down": return self.skipped() elif self.CM.StartaCM(node): return self.success() else: return self.failure("Startup %s on node %s failed" % (self.Env["Name"], node)) # # We don't register StartTest because it's better when called by # another test... # class FlipTest(CTSTest): '''If it's running, stop it. If it's stopped start it. Overthrow the status quo... ''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "Flip" self.start = StartTest(cm) self.stop = StopTest(cm) def __call__(self, node): '''Perform the 'Flip' test. ''' self.incr("calls") if self.CM.ShouldBeStatus[node] == "up": self.incr("stopped") ret = self.stop(node) type = "up->down" # Give the cluster time to recognize it's gone... time.sleep(self.Env["StableTime"]) elif self.CM.ShouldBeStatus[node] == "down": self.incr("started") ret = self.start(node) type = "down->up" else: return self.skipped() self.incr(type) if ret: return self.success() else: return self.failure("%s failure" % type) # Register FlipTest as a good test to run AllTestClasses.append(FlipTest) class RestartTest(CTSTest): '''Stop and restart a node''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "Restart" self.start = StartTest(cm) self.stop = StopTest(cm) self.benchmark = 1 def __call__(self, node): '''Perform the 'restart' test. ''' self.incr("calls") self.incr("node:" + node) ret1 = 1 if self.CM.StataCM(node): self.incr("WasStopped") if not self.start(node): return self.failure("start (setup) failure: "+node) self.set_timer() if not self.stop(node): return self.failure("stop failure: "+node) if not self.start(node): return self.failure("start failure: "+node) return self.success() # Register RestartTest as a good test to run AllTestClasses.append(RestartTest) class StonithdTest(CTSTest): def __init__(self, cm): CTSTest.__init__(self, cm) self.name = "Stonithd" self.startall = SimulStartLite(cm) self.benchmark = 1 def __call__(self, node): self.incr("calls") if len(self.Env["nodes"]) < 2: return self.skipped() ret = self.startall(None) if not ret: return self.failure("Setup failed") is_dc = self.CM.is_node_dc(node) watchpats = [] watchpats.append(self.templates["Pat:FenceOpOK"] % node) watchpats.append(self.templates["Pat:NodeFenced"] % node) if self.Env["at-boot"] == 0: self.debug("Expecting %s to stay down" % node) self.CM.ShouldBeStatus[node] = "down" else: self.debug("Expecting %s to come up again %d" % (node, self.Env["at-boot"])) watchpats.append("%s.* S_STARTING -> S_PENDING" % node) watchpats.append("%s.* S_PENDING -> S_NOT_DC" % node) watch = self.create_watch(watchpats, 30 + self.Env["DeadTime"] + self.Env["StableTime"] + self.Env["StartTime"]) watch.setwatch() origin = self.Env.RandomGen.choice(self.Env["nodes"]) rc = self.rsh(origin, "stonith_admin --reboot %s -VVVVVV" % node) if rc == 194: # 194 - 256 = -62 = Timer expired # # Look for the patterns, usually this means the required # device was running on the node to be fenced - or that # the required devices were in the process of being loaded # and/or moved # # Effectively the node committed suicide so there will be # no confirmation, but pacemaker should be watching and # fence the node again self.logger.log("Fencing command on %s to fence %s timed out" % (origin, node)) elif origin != node and rc != 0: self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.debug("Waiting for fenced node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.Env["nodes"], 600) self.logger.log("Fencing command on %s failed to fence %s (rc=%d)" % (origin, node, rc)) elif origin == node and rc != 255: # 255 == broken pipe, ie. the node was fenced as expected self.logger.log("Locally originated fencing returned %d" % rc) self.set_timer("fence") matched = watch.lookforall() self.log_timer("fence") self.set_timer("reform") if watch.unmatched: self.logger.log("Patterns not found: " + repr(watch.unmatched)) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.debug("Waiting for fenced node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.Env["nodes"], 600) self.debug("Waiting for the cluster to re-stabilize with all nodes") is_stable = self.CM.cluster_stable(self.Env["StartTime"]) if not matched: return self.failure("Didn't find all expected patterns") elif not is_stable: return self.failure("Cluster did not become stable") self.log_timer("reform") return self.success() def errorstoignore(self): return [ self.templates["Pat:Fencing_start"] % ".*", self.templates["Pat:Fencing_ok"] % ".*", self.templates["Pat:Fencing_active"], r"error.*: Operation 'reboot' targeting .* by .* for stonith_admin.*: Timer expired", ] def is_applicable(self): if not self.is_applicable_common(): return 0 if "DoFencing" in list(self.Env.keys()): return self.Env["DoFencing"] return 1 AllTestClasses.append(StonithdTest) class StartOnebyOne(CTSTest): '''Start all the nodes ~ one by one''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "StartOnebyOne" self.stopall = SimulStopLite(cm) self.start = StartTest(cm) self.ns = CTS.NodeStatus(cm.Env) def __call__(self, dummy): '''Perform the 'StartOnebyOne' test. ''' self.incr("calls") # We ignore the "node" parameter... # Shut down all the nodes... ret = self.stopall(None) if not ret: return self.failure("Test setup failed") failed = [] self.set_timer() for node in self.Env["nodes"]: if not self.start(node): failed.append(node) if len(failed) > 0: return self.failure("Some node failed to start: " + repr(failed)) return self.success() # Register StartOnebyOne as a good test to run AllTestClasses.append(StartOnebyOne) class SimulStart(CTSTest): '''Start all the nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SimulStart" self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) def __call__(self, dummy): '''Perform the 'SimulStart' test. ''' self.incr("calls") # We ignore the "node" parameter... # Shut down all the nodes... ret = self.stopall(None) if not ret: return self.failure("Setup failed") if not self.startall(None): return self.failure("Startall failed") return self.success() # Register SimulStart as a good test to run AllTestClasses.append(SimulStart) class SimulStop(CTSTest): '''Stop all the nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SimulStop" self.startall = SimulStartLite(cm) self.stopall = SimulStopLite(cm) def __call__(self, dummy): '''Perform the 'SimulStop' test. ''' self.incr("calls") # We ignore the "node" parameter... # Start up all the nodes... ret = self.startall(None) if not ret: return self.failure("Setup failed") if not self.stopall(None): return self.failure("Stopall failed") return self.success() # Register SimulStop as a good test to run AllTestClasses.append(SimulStop) class StopOnebyOne(CTSTest): '''Stop all the nodes in order''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "StopOnebyOne" self.startall = SimulStartLite(cm) self.stop = StopTest(cm) def __call__(self, dummy): '''Perform the 'StopOnebyOne' test. ''' self.incr("calls") # We ignore the "node" parameter... # Start up all the nodes... ret = self.startall(None) if not ret: return self.failure("Setup failed") failed = [] self.set_timer() for node in self.Env["nodes"]: if not self.stop(node): failed.append(node) if len(failed) > 0: return self.failure("Some node failed to stop: " + repr(failed)) return self.success() # Register StopOnebyOne as a good test to run AllTestClasses.append(StopOnebyOne) class RestartOnebyOne(CTSTest): '''Restart all the nodes in order''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "RestartOnebyOne" self.startall = SimulStartLite(cm) def __call__(self, dummy): '''Perform the 'RestartOnebyOne' test. ''' self.incr("calls") # We ignore the "node" parameter... # Start up all the nodes... ret = self.startall(None) if not ret: return self.failure("Setup failed") did_fail = [] self.set_timer() self.restart = RestartTest(self.CM) for node in self.Env["nodes"]: if not self.restart(node): did_fail.append(node) if did_fail: return self.failure("Could not restart %d nodes: %s" % (len(did_fail), repr(did_fail))) return self.success() # Register StopOnebyOne as a good test to run AllTestClasses.append(RestartOnebyOne) class PartialStart(CTSTest): '''Start a node - but tell it to stop before it finishes starting up''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "PartialStart" self.startall = SimulStartLite(cm) self.stopall = SimulStopLite(cm) self.stop = StopTest(cm) #self.is_unsafe = 1 def __call__(self, node): '''Perform the 'PartialStart' test. ''' self.incr("calls") ret = self.stopall(None) if not ret: return self.failure("Setup failed") watchpats = [] watchpats.append("pacemaker-controld.*Connecting to .* cluster infrastructure") watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() self.CM.StartaCMnoBlock(node) ret = watch.lookforall() if not ret: self.logger.log("Patterns not found: " + repr(watch.unmatched)) return self.failure("Setup of %s failed" % node) ret = self.stop(node) if not ret: return self.failure("%s did not stop in time" % node) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' # We might do some fencing in the 2-node case if we make it up far enough return [ r"Executing reboot fencing operation", r"Requesting fencing \([^)]+\) of node ", ] # Register StopOnebyOne as a good test to run AllTestClasses.append(PartialStart) class StandbyTest(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "Standby" self.benchmark = 1 self.start = StartTest(cm) self.startall = SimulStartLite(cm) # make sure the node is active # set the node to standby mode # check resources, none resource should be running on the node # set the node to active mode # check resouces, resources should have been migrated back (SHOULD THEY?) def __call__(self, node): self.incr("calls") ret = self.startall(None) if not ret: return self.failure("Start all nodes failed") self.debug("Make sure node %s is active" % node) if self.CM.StandbyStatus(node) != "off": if not self.CM.SetStandbyMode(node, "off"): return self.failure("can't set node %s to active mode" % node) self.CM.cluster_stable() status = self.CM.StandbyStatus(node) if status != "off": return self.failure("standby status of %s is [%s] but we expect [off]" % (node, status)) self.debug("Getting resources running on node %s" % node) rsc_on_node = self.CM.active_resources(node) watchpats = [] watchpats.append(r"State transition .* -> S_POLICY_ENGINE") watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() self.debug("Setting node %s to standby mode" % node) if not self.CM.SetStandbyMode(node, "on"): return self.failure("can't set node %s to standby mode" % node) self.set_timer("on") ret = watch.lookforall() if not ret: self.logger.log("Patterns not found: " + repr(watch.unmatched)) self.CM.SetStandbyMode(node, "off") return self.failure("cluster didn't react to standby change on %s" % node) self.CM.cluster_stable() status = self.CM.StandbyStatus(node) if status != "on": return self.failure("standby status of %s is [%s] but we expect [on]" % (node, status)) self.log_timer("on") self.debug("Checking resources") bad_run = self.CM.active_resources(node) if len(bad_run) > 0: rc = self.failure("%s set to standby, %s is still running on it" % (node, repr(bad_run))) self.debug("Setting node %s to active mode" % node) self.CM.SetStandbyMode(node, "off") return rc self.debug("Setting node %s to active mode" % node) if not self.CM.SetStandbyMode(node, "off"): return self.failure("can't set node %s to active mode" % node) self.set_timer("off") self.CM.cluster_stable() status = self.CM.StandbyStatus(node) if status != "off": return self.failure("standby status of %s is [%s] but we expect [off]" % (node, status)) self.log_timer("off") return self.success() AllTestClasses.append(StandbyTest) class ValgrindTest(CTSTest): '''Check for memory leaks''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "Valgrind" self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) self.is_valgrind = 1 self.is_loop = 1 def setup(self, node): self.incr("calls") ret = self.stopall(None) if not ret: return self.failure("Stop all nodes failed") # @TODO Edit /etc/sysconfig/pacemaker on all nodes to enable valgrind, # and clear any valgrind logs from previous runs. For now, we rely on # the user to do this manually. ret = self.startall(None) if not ret: return self.failure("Start all nodes failed") return self.success() def teardown(self, node): # Return all nodes to normal # @TODO Edit /etc/sysconfig/pacemaker on all nodes to disable valgrind ret = self.stopall(None) if not ret: return self.failure("Stop all nodes failed") return self.success() def find_leaks(self): # Check for leaks # (no longer used but kept in case feature is restored) leaked = [] self.stop = StopTest(self.CM) for node in self.Env["nodes"]: rc = self.stop(node) if not rc: self.failure("Couldn't shut down %s" % node) rc = self.rsh(node, "grep -e indirectly.*lost:.*[1-9] -e definitely.*lost:.*[1-9] -e (ERROR|error).*SUMMARY:.*[1-9].*errors %s" % self.logger.logPat, 0) if rc != 1: leaked.append(node) self.failure("Valgrind errors detected on %s" % node) (rc, output) = self.rsh(node, "grep -e lost: -e SUMMARY: %s" % self.logger.logPat, None) for line in output: self.logger.log(line) (rc, output) = self.rsh(node, "cat %s" % self.logger.logPat, None) for line in output: self.debug(line) self.rsh(node, "rm -f %s" % self.logger.logPat, None) return leaked def __call__(self, node): #leaked = self.find_leaks() #if len(leaked) > 0: # return self.failure("Nodes %s leaked" % repr(leaked)) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"pacemaker-based.*: \*\*\*\*\*\*\*\*\*\*\*\*\*", r"pacemaker-based.*: .* avoid confusing Valgrind", r"HA_VALGRIND_ENABLED", ] class StandbyLoopTest(ValgrindTest): '''Check for memory leaks by putting a node in and out of standby for an hour''' # @TODO This is not a useful test for memory leaks def __init__(self, cm): ValgrindTest.__init__(self,cm) self.name = "StandbyLoop" def __call__(self, node): lpc = 0 delay = 2 failed = 0 done = time.time() + self.Env["loop-minutes"] * 60 while time.time() <= done and not failed: lpc = lpc + 1 time.sleep(delay) if not self.CM.SetStandbyMode(node, "on"): self.failure("can't set node %s to standby mode" % node) failed = lpc time.sleep(delay) if not self.CM.SetStandbyMode(node, "off"): self.failure("can't set node %s to active mode" % node) failed = lpc leaked = self.find_leaks() if failed: return self.failure("Iteration %d failed" % failed) elif len(leaked) > 0: return self.failure("Nodes %s leaked" % repr(leaked)) return self.success() #AllTestClasses.append(StandbyLoopTest) class BandwidthTest(CTSTest): # Tests should not be cluster-manager-specific # If you need to find out cluster manager configuration to do this, then # it should be added to the generic cluster manager API. '''Test the bandwidth which the cluster uses''' def __init__(self, cm): CTSTest.__init__(self, cm) self.name = "Bandwidth" self.start = StartTest(cm) self.__setitem__("min",0) self.__setitem__("max",0) self.__setitem__("totalbandwidth",0) (handle, self.tempfile) = tempfile.mkstemp(".cts") os.close(handle) self.startall = SimulStartLite(cm) def __call__(self, node): '''Perform the Bandwidth test''' self.incr("calls") if self.CM.upcount() < 1: return self.skipped() Path = self.CM.InternalCommConfig() if "ip" not in Path["mediatype"]: return self.skipped() port = Path["port"][0] port = int(port) ret = self.startall(None) if not ret: return self.failure("Test setup failed") time.sleep(5) # We get extra messages right after startup. fstmpfile = "/var/run/band_estimate" dumpcmd = "tcpdump -p -n -c 102 -i any udp port %d > %s 2>&1" \ % (port, fstmpfile) rc = self.rsh(node, dumpcmd) if rc == 0: farfile = "root@%s:%s" % (node, fstmpfile) self.rsh.cp(farfile, self.tempfile) Bandwidth = self.countbandwidth(self.tempfile) if not Bandwidth: self.logger.log("Could not compute bandwidth.") return self.success() intband = int(Bandwidth + 0.5) self.logger.log("...bandwidth: %d bits/sec" % intband) self.Stats["totalbandwidth"] = self.Stats["totalbandwidth"] + Bandwidth if self.Stats["min"] == 0: self.Stats["min"] = Bandwidth if Bandwidth > self.Stats["max"]: self.Stats["max"] = Bandwidth if Bandwidth < self.Stats["min"]: self.Stats["min"] = Bandwidth self.rsh(node, "rm -f %s" % fstmpfile) os.unlink(self.tempfile) return self.success() else: return self.failure("no response from tcpdump command [%d]!" % rc) def countbandwidth(self, file): fp = open(file, "r") fp.seek(0) count = 0 sum = 0 while 1: line = fp.readline() if not line: return None if re.search("udp",line) or re.search("UDP,", line): count = count + 1 linesplit = line.split(" ") for j in range(len(linesplit)-1): if linesplit[j] == "udp": break if linesplit[j] == "length:": break try: sum = sum + int(linesplit[j+1]) except ValueError: self.logger.log("Invalid tcpdump line: %s" % line) return None T1 = linesplit[0] timesplit = T1.split(":") time2split = timesplit[2].split(".") time1 = (int(timesplit[0])*60+int(timesplit[1]))*60+int(time2split[0])+int(time2split[1])*0.000001 break while count < 100: line = fp.readline() if not line: return None if re.search("udp",line) or re.search("UDP,", line): count = count+1 linessplit = line.split(" ") for j in range(len(linessplit)-1): if linessplit[j] == "udp": break if linessplit[j] == "length:": break try: sum = int(linessplit[j+1]) + sum except ValueError: self.logger.log("Invalid tcpdump line: %s" % line) return None T2 = linessplit[0] timesplit = T2.split(":") time2split = timesplit[2].split(".") time2 = (int(timesplit[0])*60+int(timesplit[1]))*60+int(time2split[0])+int(time2split[1])*0.000001 time = time2-time1 if (time <= 0): return 0 return int((sum*8)/time) def is_applicable(self): '''BandwidthTest never applicable''' return 0 AllTestClasses.append(BandwidthTest) ################################################################### class MaintenanceMode(CTSTest): ################################################################### def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "MaintenanceMode" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.max = 30 #self.is_unsafe = 1 self.benchmark = 1 self.action = "asyncmon" self.interval = 0 self.rid = "maintenanceDummy" def toggleMaintenanceMode(self, node, action): pats = [] pats.append(self.templates["Pat:DC_IDLE"]) # fail the resource right after turning Maintenance mode on # verify it is not recovered until maintenance mode is turned off if action == "On": pats.append(self.templates["Pat:RscOpFail"] % (self.action, self.rid)) else: pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.rid)) pats.append(self.templates["Pat:RscOpOK"] % ("start", self.rid)) watch = self.create_watch(pats, 60) watch.setwatch() self.debug("Turning maintenance mode %s" % action) self.rsh(node, self.templates["MaintenanceMode%s" % (action)]) if (action == "On"): self.rsh(node, "crm_resource -V -F -r %s -H %s &>/dev/null" % (self.rid, node)) self.set_timer("recover%s" % (action)) watch.lookforall() self.log_timer("recover%s" % (action)) if watch.unmatched: self.debug("Failed to find patterns when turning maintenance mode %s" % action) return repr(watch.unmatched) return "" def insertMaintenanceDummy(self, node): pats = [] pats.append(("%s.*" % node) + (self.templates["Pat:RscOpOK"] % ("start", self.rid))) watch = self.create_watch(pats, 60) watch.setwatch() self.CM.AddDummyRsc(node, self.rid) self.set_timer("addDummy") watch.lookforall() self.log_timer("addDummy") if watch.unmatched: self.debug("Failed to find patterns when adding maintenance dummy resource") return repr(watch.unmatched) return "" def removeMaintenanceDummy(self, node): pats = [] pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.rid)) watch = self.create_watch(pats, 60) watch.setwatch() self.CM.RemoveDummyRsc(node, self.rid) self.set_timer("removeDummy") watch.lookforall() self.log_timer("removeDummy") if watch.unmatched: self.debug("Failed to find patterns when removing maintenance dummy resource") return repr(watch.unmatched) return "" def managedRscList(self, node): rscList = [] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): tmp = AuditResource(self.CM, line) if tmp.managed(): rscList.append(tmp.id) return rscList def verifyResources(self, node, rscList, managed): managedList = list(rscList) managed_str = "managed" if not managed: managed_str = "unmanaged" (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): tmp = AuditResource(self.CM, line) if managed and not tmp.managed(): continue elif not managed and tmp.managed(): continue elif managedList.count(tmp.id): managedList.remove(tmp.id) if len(managedList) == 0: self.debug("Found all %s resources on %s" % (managed_str, node)) return True self.logger.log("Could not find all %s resources on %s. %s" % (managed_str, node, managedList)) return False def __call__(self, node): '''Perform the 'MaintenanceMode' test. ''' self.incr("calls") verify_managed = False verify_unmanaged = False failPat = "" ret = self.startall(None) if not ret: return self.failure("Setup failed") # get a list of all the managed resources. We use this list # after enabling maintenance mode to verify all managed resources # become un-managed. After maintenance mode is turned off, we use # this list to verify all the resources become managed again. managedResources = self.managedRscList(node) if len(managedResources) == 0: self.logger.log("No managed resources on %s" % node) return self.skipped() # insert a fake resource we can fail during maintenance mode # so we can verify recovery does not take place until after maintenance # mode is disabled. failPat = failPat + self.insertMaintenanceDummy(node) # toggle maintenance mode ON, then fail dummy resource. failPat = failPat + self.toggleMaintenanceMode(node, "On") # verify all the resources are now unmanaged if self.verifyResources(node, managedResources, False): verify_unmanaged = True # Toggle maintenance mode OFF, verify dummy is recovered. failPat = failPat + self.toggleMaintenanceMode(node, "Off") # verify all the resources are now managed again if self.verifyResources(node, managedResources, True): verify_managed = True # Remove our maintenance dummy resource. failPat = failPat + self.removeMaintenanceDummy(node) self.CM.cluster_stable() if failPat != "": return self.failure("Unmatched patterns: %s" % (failPat)) elif verify_unmanaged is False: return self.failure("Failed to verify resources became unmanaged during maintenance mode") elif verify_managed is False: return self.failure("Failed to verify resources switched back to managed after disabling maintenance mode") return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"Updating failcount for %s" % self.rid, r"schedulerd.*: Recover %s\s*\(.*\)" % self.rid, r"Unknown operation: fail", self.templates["Pat:RscOpOK"] % (self.action, self.rid), r"(ERROR|error).*: Action %s_%s_%d .* initiated outside of a transition" % (self.rid, self.action, self.interval), ] AllTestClasses.append(MaintenanceMode) class ResourceRecover(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "ResourceRecover" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.max = 30 self.rid = None self.rid_alt = None #self.is_unsafe = 1 self.benchmark = 1 # these are the values used for the new LRM API call self.action = "asyncmon" self.interval = 0 def __call__(self, node): '''Perform the 'ResourceRecover' test. ''' self.incr("calls") ret = self.startall(None) if not ret: return self.failure("Setup failed") resourcelist = self.CM.active_resources(node) # if there are no resourcelist, return directly if len(resourcelist) == 0: self.logger.log("No active resources on %s" % node) return self.skipped() self.rid = self.Env.RandomGen.choice(resourcelist) self.rid_alt = self.rid rsc = None (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): tmp = AuditResource(self.CM, line) if tmp.id == self.rid: rsc = tmp # Handle anonymous clones that get renamed self.rid = rsc.clone_id break if not rsc: return self.failure("Could not find %s in the resource list" % self.rid) self.debug("Shooting %s aka. %s" % (rsc.clone_id, rsc.id)) pats = [] pats.append(self.templates["Pat:CloneOpFail"] % (self.action, rsc.id, rsc.clone_id)) if rsc.managed(): pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.rid)) if rsc.unique(): pats.append(self.templates["Pat:RscOpOK"] % ("start", self.rid)) else: # Anonymous clones may get restarted with a different clone number pats.append(self.templates["Pat:RscOpOK"] % ("start", ".*")) watch = self.create_watch(pats, 60) watch.setwatch() self.rsh(node, "crm_resource -V -F -r %s -H %s &>/dev/null" % (self.rid, node)) self.set_timer("recover") watch.lookforall() self.log_timer("recover") self.CM.cluster_stable() recovered = self.CM.ResourceLocation(self.rid) if watch.unmatched: return self.failure("Patterns not found: %s" % repr(watch.unmatched)) elif rsc.unique() and len(recovered) > 1: return self.failure("%s is now active on more than one node: %s"%(self.rid, repr(recovered))) elif len(recovered) > 0: self.debug("%s is running on: %s" % (self.rid, repr(recovered))) elif rsc.managed(): return self.failure("%s was not recovered and is inactive" % self.rid) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"Updating failcount for %s" % self.rid, r"schedulerd.*: Recover (%s|%s)\s*\(.*\)" % (self.rid, self.rid_alt), r"Unknown operation: fail", self.templates["Pat:RscOpOK"] % (self.action, self.rid), r"(ERROR|error).*: Action %s_%s_%d .* initiated outside of a transition" % (self.rid, self.action, self.interval), ] AllTestClasses.append(ResourceRecover) class ComponentFail(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "ComponentFail" # TODO make this work correctly in docker. self.is_docker_unsafe = 1 self.startall = SimulStartLite(cm) self.complist = cm.Components() self.patterns = [] self.okerrpatterns = [] self.is_unsafe = 1 def __call__(self, node): '''Perform the 'ComponentFail' test. ''' self.incr("calls") self.patterns = [] self.okerrpatterns = [] # start all nodes ret = self.startall(None) if not ret: return self.failure("Setup failed") if not self.CM.cluster_stable(self.Env["StableTime"]): return self.failure("Setup failed - unstable") node_is_dc = self.CM.is_node_dc(node, None) # select a component to kill chosen = self.Env.RandomGen.choice(self.complist) while chosen.dc_only == 1 and node_is_dc == 0: chosen = self.Env.RandomGen.choice(self.complist) self.debug("...component %s (dc=%d,boot=%d)" % (chosen.name, node_is_dc,chosen.triggersreboot)) self.incr(chosen.name) if chosen.name != "corosync": self.patterns.append(self.templates["Pat:ChildKilled"] %(node, chosen.name)) self.patterns.append(self.templates["Pat:ChildRespawn"] %(node, chosen.name)) self.patterns.extend(chosen.pats) if node_is_dc: self.patterns.extend(chosen.dc_pats) # @TODO this should be a flag in the Component if chosen.name in [ "corosync", "pacemaker-based", "pacemaker-fenced" ]: # Ignore actions for fence devices if fencer will respawn # (their registration will be lost, and probes will fail) self.okerrpatterns = [ self.templates["Pat:Fencing_active"] ] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rclass == "stonith": self.okerrpatterns.append(self.templates["Pat:Fencing_recover"] % r.id) self.okerrpatterns.append(self.templates["Pat:Fencing_probe"] % r.id) # supply a copy so self.patterns doesn't end up empty tmpPats = [] tmpPats.extend(self.patterns) self.patterns.extend(chosen.badnews_ignore) # Look for STONITH ops, depending on Env["at-boot"] we might need to change the nodes status stonithPats = [] stonithPats.append(self.templates["Pat:Fencing_ok"] % node) stonith = self.create_watch(stonithPats, 0) stonith.setwatch() # set the watch for stable watch = self.create_watch( tmpPats, self.Env["DeadTime"] + self.Env["StableTime"] + self.Env["StartTime"]) watch.setwatch() # kill the component chosen.kill(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.debug("Waiting for any fenced node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.Env["nodes"], 600) self.debug("Waiting for the cluster to re-stabilize with all nodes") self.CM.cluster_stable(self.Env["StartTime"]) self.debug("Checking if %s was shot" % node) shot = stonith.look(60) if shot: self.debug("Found: " + repr(shot)) self.okerrpatterns.append(self.templates["Pat:Fencing_start"] % node) if self.Env["at-boot"] == 0: self.CM.ShouldBeStatus[node] = "down" # If fencing occurred, chances are many (if not all) the expected logs # will not be sent - or will be lost when the node reboots return self.success() # check for logs indicating a graceful recovery matched = watch.lookforall(allow_multiple_matches=1) if watch.unmatched: self.logger.log("Patterns not found: " + repr(watch.unmatched)) self.debug("Waiting for the cluster to re-stabilize with all nodes") is_stable = self.CM.cluster_stable(self.Env["StartTime"]) if not matched: return self.failure("Didn't find all expected %s patterns" % chosen.name) elif not is_stable: return self.failure("Cluster did not become stable after killing %s" % chosen.name) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' # Note that okerrpatterns refers to the last time we ran this test # The good news is that this works fine for us... self.okerrpatterns.extend(self.patterns) return self.okerrpatterns AllTestClasses.append(ComponentFail) class SplitBrainTest(CTSTest): '''It is used to test split-brain. when the path between the two nodes break check the two nodes both take over the resource''' def __init__(self,cm): CTSTest.__init__(self,cm) self.name = "SplitBrain" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.is_experimental = 1 def isolate_partition(self, partition): other_nodes = [] other_nodes.extend(self.Env["nodes"]) for node in partition: try: other_nodes.remove(node) except ValueError: self.logger.log("Node "+node+" not in " + repr(self.Env["nodes"]) + " from " +repr(partition)) if len(other_nodes) == 0: return 1 self.debug("Creating partition: " + repr(partition)) self.debug("Everyone else: " + repr(other_nodes)) for node in partition: if not self.CM.isolate_node(node, other_nodes): self.logger.log("Could not isolate %s" % node) return 0 return 1 def heal_partition(self, partition): other_nodes = [] other_nodes.extend(self.Env["nodes"]) for node in partition: try: other_nodes.remove(node) except ValueError: self.logger.log("Node "+node+" not in " + repr(self.Env["nodes"])) if len(other_nodes) == 0: return 1 self.debug("Healing partition: " + repr(partition)) self.debug("Everyone else: " + repr(other_nodes)) for node in partition: self.CM.unisolate_node(node, other_nodes) def __call__(self, node): '''Perform split-brain test''' self.incr("calls") self.passed = 1 partitions = {} ret = self.startall(None) if not ret: return self.failure("Setup failed") while 1: # Retry until we get multiple partitions partitions = {} p_max = len(self.Env["nodes"]) for node in self.Env["nodes"]: p = self.Env.RandomGen.randint(1, p_max) if not p in partitions: partitions[p] = [] partitions[p].append(node) p_max = len(list(partitions.keys())) if p_max > 1: break # else, try again self.debug("Created %d partitions" % p_max) for key in list(partitions.keys()): self.debug("Partition["+str(key)+"]:\t"+repr(partitions[key])) # Disabling STONITH to reduce test complexity for now self.rsh(node, "crm_attribute -V -n stonith-enabled -v false") for key in list(partitions.keys()): self.isolate_partition(partitions[key]) count = 30 while count > 0: if len(self.CM.find_partitions()) != p_max: time.sleep(10) else: break else: self.failure("Expected partitions were not created") # Target number of partitions formed - wait for stability if not self.CM.cluster_stable(): self.failure("Partitioned cluster not stable") # Now audit the cluster state self.CM.partitions_expected = p_max if not self.audit(): self.failure("Audits failed") self.CM.partitions_expected = 1 # And heal them again for key in list(partitions.keys()): self.heal_partition(partitions[key]) # Wait for a single partition to form count = 30 while count > 0: if len(self.CM.find_partitions()) != 1: time.sleep(10) count -= 1 else: break else: self.failure("Cluster did not reform") # Wait for it to have the right number of members count = 30 while count > 0: members = [] partitions = self.CM.find_partitions() if len(partitions) > 0: members = partitions[0].split() if len(members) != len(self.Env["nodes"]): time.sleep(10) count -= 1 else: break else: self.failure("Cluster did not completely reform") # Wait up to 20 minutes - the delay is more preferable than # trying to continue with in a messed up state if not self.CM.cluster_stable(1200): self.failure("Reformed cluster not stable") if self.Env["continue"] == 1: answer = "Y" else: try: answer = input_wrapper('Continue? [nY]') except EOFError as e: answer = "n" if answer and answer == "n": raise ValueError("Reformed cluster not stable") # Turn fencing back on if self.Env["DoFencing"]: self.rsh(node, "crm_attribute -V -D -n stonith-enabled") self.CM.cluster_stable() if self.passed: return self.success() return self.failure("See previous errors") def errorstoignore(self): '''Return list of errors which are 'normal' and should be ignored''' return [ r"Another DC detected:", r"(ERROR|error).*: .*Application of an update diff failed", r"pacemaker-controld.*:.*not in our membership list", r"CRIT:.*node.*returning after partition", ] def is_applicable(self): if not self.is_applicable_common(): return 0 return len(self.Env["nodes"]) > 2 AllTestClasses.append(SplitBrainTest) class Reattach(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "Reattach" self.startall = SimulStartLite(cm) self.restart1 = RestartTest(cm) self.stopall = SimulStopLite(cm) self.is_unsafe = 0 # Handled by canrunnow() def _is_managed(self, node): is_managed = self.rsh(node, "crm_attribute -t rsc_defaults -n is-managed -q -G -d true", 1) is_managed = is_managed[:-1] # Strip off the newline return is_managed == "true" def _set_unmanaged(self, node): self.debug("Disable resource management") self.rsh(node, "crm_attribute -t rsc_defaults -n is-managed -v false") def _set_managed(self, node): self.debug("Re-enable resource management") self.rsh(node, "crm_attribute -t rsc_defaults -n is-managed -D") def setup(self, node): attempt = 0 if not self.startall(None): return None # Make sure we are really _really_ stable and that all # resources, including those that depend on transient node # attributes, are started while not self.CM.cluster_stable(double_check=True): if attempt < 5: attempt += 1 self.debug("Not stable yet, re-testing") else: self.logger.log("Cluster is not stable") return None return 1 def teardown(self, node): # Make sure 'node' is up start = StartTest(self.CM) start(node) if not self._is_managed(node): self.logger.log("Attempting to re-enable resource management on %s" % node) self._set_managed(node) self.CM.cluster_stable() if not self._is_managed(node): self.logger.log("Could not re-enable resource management") return 0 return 1 def canrunnow(self, node): '''Return TRUE if we can meaningfully run right now''' if self.find_ocfs2_resources(node): self.logger.log("Detach/Reattach scenarios are not possible with OCFS2 services present") return 0 return 1 def __call__(self, node): self.incr("calls") pats = [] # Conveniently, the scheduler will display this message when disabling # management, even if fencing is not enabled, so we can rely on it. managed = self.create_watch(["Delaying fencing operations"], 60) managed.setwatch() self._set_unmanaged(node) if not managed.lookforall(): self.logger.log("Patterns not found: " + repr(managed.unmatched)) return self.failure("Resource management not disabled") pats = [] pats.append(self.templates["Pat:RscOpOK"] % ("start", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("stop", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("promote", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("demote", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("migrate", ".*")) watch = self.create_watch(pats, 60, "ShutdownActivity") watch.setwatch() self.debug("Shutting down the cluster") ret = self.stopall(None) if not ret: self._set_managed(node) return self.failure("Couldn't shut down the cluster") self.debug("Bringing the cluster back up") ret = self.startall(None) time.sleep(5) # allow ping to update the CIB if not ret: self._set_managed(node) return self.failure("Couldn't restart the cluster") if self.local_badnews("ResourceActivity:", watch): self._set_managed(node) return self.failure("Resources stopped or started during cluster restart") watch = self.create_watch(pats, 60, "StartupActivity") watch.setwatch() # Re-enable resource management (and verify it happened). self._set_managed(node) self.CM.cluster_stable() if not self._is_managed(node): return self.failure("Could not re-enable resource management") # Ignore actions for STONITH resources ignore = [] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rclass == "stonith": self.debug("Ignoring start actions for %s" % r.id) ignore.append(self.templates["Pat:RscOpOK"] % ("start", r.id)) if self.local_badnews("ResourceActivity:", watch, ignore): return self.failure("Resources stopped or started after resource management was re-enabled") return ret def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"resource( was|s were) active at shutdown", ] def is_applicable(self): return 1 AllTestClasses.append(Reattach) class SpecialTest1(CTSTest): '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SpecialTest1" self.startall = SimulStartLite(cm) self.restart1 = RestartTest(cm) self.stopall = SimulStopLite(cm) def __call__(self, node): '''Perform the 'SpecialTest1' test for Andrew. ''' self.incr("calls") # Shut down all the nodes... ret = self.stopall(None) if not ret: return self.failure("Could not stop all nodes") # Test config recovery when the other nodes come up self.rsh(node, "rm -f "+CTSvars.CRM_CONFIG_DIR+"/cib*") # Start the selected node ret = self.restart1(node) if not ret: return self.failure("Could not start "+node) # Start all remaining nodes ret = self.startall(None) if not ret: return self.failure("Could not start the remaining nodes") return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' # Errors that occur as a result of the CIB being wiped return [ r"error.*: v1 patchset error, patch failed to apply: Application of an update diff failed", r"error.*: Resource start-up disabled since no STONITH resources have been defined", r"error.*: Either configure some or disable STONITH with the stonith-enabled option", r"error.*: NOTE: Clusters with shared data need STONITH to ensure data integrity", ] AllTestClasses.append(SpecialTest1) class HAETest(CTSTest): '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "HAETest" self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) self.is_loop = 1 def setup(self, node): # Start all remaining nodes ret = self.startall(None) if not ret: return self.failure("Couldn't start all nodes") return self.success() def teardown(self, node): # Stop everything ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") return self.success() def wait_on_state(self, node, resource, expected_clones, attempts=240): while attempts > 0: active = 0 (rc, lines) = self.rsh(node, "crm_resource -r %s -W -Q" % resource, stdout=None) # Hack until crm_resource does the right thing if rc == 0 and lines: active = len(lines) if len(lines) == expected_clones: return 1 elif rc == 1: self.debug("Resource %s is still inactive" % resource) elif rc == 234: self.logger.log("Unknown resource %s" % resource) return 0 elif rc == 246: self.logger.log("Cluster is inactive") return 0 elif rc != 0: self.logger.log("Call to crm_resource failed, rc=%d" % rc) return 0 else: self.debug("Resource %s is active on %d times instead of %d" % (resource, active, expected_clones)) attempts -= 1 time.sleep(1) return 0 def find_dlm(self, node): self.r_dlm = None (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rtype == "controld" and r.parent != "NA": self.debug("Found dlm: %s" % self.r_dlm) self.r_dlm = r.parent return 1 return 0 def find_hae_resources(self, node): self.r_dlm = None self.r_o2cb = None self.r_ocfs2 = [] if self.find_dlm(node): self.find_ocfs2_resources(node) def is_applicable(self): if not self.is_applicable_common(): return 0 if self.Env["Schema"] == "hae": return 1 return None class HAERoleTest(HAETest): def __init__(self, cm): '''Lars' mount/unmount test for the HA extension. ''' HAETest.__init__(self,cm) self.name = "HAERoleTest" def change_state(self, node, resource, target): rc = self.rsh(node, "crm_resource -V -r %s -p target-role -v %s --meta" % (resource, target)) return rc def __call__(self, node): self.incr("calls") lpc = 0 failed = 0 delay = 2 done = time.time() + self.Env["loop-minutes"]*60 self.find_hae_resources(node) clone_max = len(self.Env["nodes"]) while time.time() <= done and not failed: lpc = lpc + 1 self.change_state(node, self.r_dlm, "Stopped") if not self.wait_on_state(node, self.r_dlm, 0): self.failure("%s did not go down correctly" % self.r_dlm) failed = lpc self.change_state(node, self.r_dlm, "Started") if not self.wait_on_state(node, self.r_dlm, clone_max): self.failure("%s did not come up correctly" % self.r_dlm) failed = lpc if not self.wait_on_state(node, self.r_o2cb, clone_max): self.failure("%s did not come up correctly" % self.r_o2cb) failed = lpc for fs in self.r_ocfs2: if not self.wait_on_state(node, fs, clone_max): self.failure("%s did not come up correctly" % fs) failed = lpc if failed: return self.failure("iteration %d failed" % failed) return self.success() AllTestClasses.append(HAERoleTest) class HAEStandbyTest(HAETest): '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): HAETest.__init__(self,cm) self.name = "HAEStandbyTest" def change_state(self, node, resource, target): rc = self.rsh(node, "crm_standby -V -l reboot -v %s" % (target)) return rc def __call__(self, node): self.incr("calls") lpc = 0 failed = 0 done = time.time() + self.Env["loop-minutes"]*60 self.find_hae_resources(node) clone_max = len(self.Env["nodes"]) while time.time() <= done and not failed: lpc = lpc + 1 self.change_state(node, self.r_dlm, "true") if not self.wait_on_state(node, self.r_dlm, clone_max-1): self.failure("%s did not go down correctly" % self.r_dlm) failed = lpc self.change_state(node, self.r_dlm, "false") if not self.wait_on_state(node, self.r_dlm, clone_max): self.failure("%s did not come up correctly" % self.r_dlm) failed = lpc if not self.wait_on_state(node, self.r_o2cb, clone_max): self.failure("%s did not come up correctly" % self.r_o2cb) failed = lpc for fs in self.r_ocfs2: if not self.wait_on_state(node, fs, clone_max): self.failure("%s did not come up correctly" % fs) failed = lpc if failed: return self.failure("iteration %d failed" % failed) return self.success() AllTestClasses.append(HAEStandbyTest) class NearQuorumPointTest(CTSTest): ''' This test brings larger clusters near the quorum point (50%). In addition, it will test doing starts and stops at the same time. Here is how I think it should work: - loop over the nodes and decide randomly which will be up and which will be down Use a 50% probability for each of up/down. - figure out what to do to get into that state from the current state - in parallel, bring up those going up and bring those going down. ''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "NearQuorumPoint" def __call__(self, dummy): '''Perform the 'NearQuorumPoint' test. ''' self.incr("calls") startset = [] stopset = [] stonith = self.CM.prepare_fencing_watcher("NearQuorumPoint") #decide what to do with each node for node in self.Env["nodes"]: action = self.Env.RandomGen.choice(["start","stop"]) #action = self.Env.RandomGen.choice(["start","stop","no change"]) if action == "start" : startset.append(node) elif action == "stop" : stopset.append(node) self.debug("start nodes:" + repr(startset)) self.debug("stop nodes:" + repr(stopset)) #add search patterns watchpats = [ ] for node in stopset: if self.CM.ShouldBeStatus[node] == "up": watchpats.append(self.templates["Pat:We_stopped"] % node) for node in startset: if self.CM.ShouldBeStatus[node] == "down": #watchpats.append(self.templates["Pat:NonDC_started"] % node) watchpats.append(self.templates["Pat:Local_started"] % node) else: for stopping in stopset: if self.CM.ShouldBeStatus[stopping] == "up": watchpats.append(self.templates["Pat:They_stopped"] % (node, self.CM.key_for_node(stopping))) if len(watchpats) == 0: return self.skipped() if len(startset) != 0: watchpats.append(self.templates["Pat:DC_IDLE"]) watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() #begin actions for node in stopset: if self.CM.ShouldBeStatus[node] == "up": self.CM.StopaCMnoBlock(node) for node in startset: if self.CM.ShouldBeStatus[node] == "down": self.CM.StartaCMnoBlock(node) #get the result if watch.lookforall(): self.CM.cluster_stable() self.CM.fencing_cleanup("NearQuorumPoint", stonith) return self.success() self.logger.log("Warn: Patterns not found: " + repr(watch.unmatched)) #get the "bad" nodes upnodes = [] for node in stopset: if self.CM.StataCM(node) == 1: upnodes.append(node) downnodes = [] for node in startset: if self.CM.StataCM(node) == 0: downnodes.append(node) self.CM.fencing_cleanup("NearQuorumPoint", stonith) if upnodes == [] and downnodes == []: self.CM.cluster_stable() # Make sure they're completely down with no residule for node in stopset: self.rsh(node, self.templates["StopCmd"]) return self.success() if len(upnodes) > 0: self.logger.log("Warn: Unstoppable nodes: " + repr(upnodes)) if len(downnodes) > 0: self.logger.log("Warn: Unstartable nodes: " + repr(downnodes)) return self.failure() def is_applicable(self): return 1 AllTestClasses.append(NearQuorumPointTest) class RollingUpgradeTest(CTSTest): '''Perform a rolling upgrade of the cluster''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "RollingUpgrade" self.start = StartTest(cm) self.stop = StopTest(cm) self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) def setup(self, node): # Start all remaining nodes ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") for node in self.Env["nodes"]: if not self.downgrade(node, None): return self.failure("Couldn't downgrade %s" % node) ret = self.startall(None) if not ret: return self.failure("Couldn't start all nodes") return self.success() def teardown(self, node): # Stop everything ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") for node in self.Env["nodes"]: if not self.upgrade(node, None): return self.failure("Couldn't upgrade %s" % node) return self.success() def install(self, node, version, start=1, flags="--force"): target_dir = "/tmp/rpm-%s" % version src_dir = "%s/%s" % (self.Env["rpm-dir"], version) self.logger.log("Installing %s on %s with %s" % (version, node, flags)) if not self.stop(node): return self.failure("stop failure: "+node) rc = self.rsh(node, "mkdir -p %s" % target_dir) rc = self.rsh(node, "rm -f %s/*.rpm" % target_dir) (rc, lines) = self.rsh(node, "ls -1 %s/*.rpm" % src_dir, None) for line in lines: line = line[:-1] rc = self.rsh.cp("%s" % (line), "%s:%s/" % (node, target_dir)) rc = self.rsh(node, "rpm -Uvh %s %s/*.rpm" % (flags, target_dir)) if start and not self.start(node): return self.failure("start failure: "+node) return self.success() def upgrade(self, node, start=1): return self.install(node, self.Env["current-version"], start) def downgrade(self, node, start=1): return self.install(node, self.Env["previous-version"], start, "--force --nodeps") def __call__(self, node): '''Perform the 'Rolling Upgrade' test. ''' self.incr("calls") for node in self.Env["nodes"]: if self.upgrade(node): return self.failure("Couldn't upgrade %s" % node) self.CM.cluster_stable() return self.success() def is_applicable(self): if not self.is_applicable_common(): return None if not "rpm-dir" in list(self.Env.keys()): return None if not "current-version" in list(self.Env.keys()): return None if not "previous-version" in list(self.Env.keys()): return None return 1 # Register RestartTest as a good test to run AllTestClasses.append(RollingUpgradeTest) class BSC_AddResource(CTSTest): '''Add a resource to the cluster''' def __init__(self, cm): CTSTest.__init__(self, cm) self.name = "AddResource" self.resource_offset = 0 self.cib_cmd = """cibadmin -C -o %s -X '%s' """ def __call__(self, node): self.incr("calls") self.resource_offset = self.resource_offset + 1 r_id = "bsc-rsc-%s-%d" % (node, self.resource_offset) start_pat = "pacemaker-controld.*%s_start_0.*confirmed.*ok" patterns = [] patterns.append(start_pat % r_id) watch = self.create_watch(patterns, self.Env["DeadTime"]) watch.setwatch() ip = self.NextIP() if not self.make_ip_resource(node, r_id, "ocf", "IPaddr", ip): return self.failure("Make resource %s failed" % r_id) failed = 0 watch_result = watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.logger.log ("Warn: Pattern not found: %s" % (regex)) failed = 1 if failed: return self.failure("Resource pattern(s) not found") if not self.CM.cluster_stable(self.Env["DeadTime"]): return self.failure("Unstable cluster") return self.success() def NextIP(self): ip = self.Env["IPBase"] if ":" in ip: fields = ip.rpartition(":") fields[2] = str(hex(int(fields[2], 16)+1)) print(str(hex(int(f[2], 16)+1))) else: fields = ip.rpartition('.') fields[2] = str(int(fields[2])+1) ip = fields[0] + fields[1] + fields[3]; self.Env["IPBase"] = ip return ip.strip() def make_ip_resource(self, node, id, rclass, type, ip): self.logger.log("Creating %s:%s:%s (%s) on %s" % (rclass,type,id,ip,node)) rsc_xml=""" """ % (id, rclass, type, id, id, ip) node_constraint = """ """ % (id, id, id, id, node) rc = 0 (rc, lines) = self.rsh(node, self.cib_cmd % ("constraints", node_constraint), None) if rc != 0: self.logger.log("Constraint creation failed: %d" % rc) return None (rc, lines) = self.rsh(node, self.cib_cmd % ("resources", rsc_xml), None) if rc != 0: self.logger.log("Resource creation failed: %d" % rc) return None return 1 def is_applicable(self): if self.Env["DoBSC"]: return 1 return None AllTestClasses.append(BSC_AddResource) class SimulStopLite(CTSTest): '''Stop any active nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SimulStopLite" def __call__(self, dummy): '''Perform the 'SimulStopLite' setup work. ''' self.incr("calls") self.debug("Setup: " + self.name) # We ignore the "node" parameter... watchpats = [ ] for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.incr("WasStarted") watchpats.append(self.templates["Pat:We_stopped"] % node) if len(watchpats) == 0: return self.success() # Stop all the nodes - at about the same time... watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() self.set_timer() for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.CM.StopaCMnoBlock(node) if watch.lookforall(): # Make sure they're completely down with no residule for node in self.Env["nodes"]: self.rsh(node, self.templates["StopCmd"]) return self.success() did_fail = 0 up_nodes = [] for node in self.Env["nodes"]: if self.CM.StataCM(node) == 1: did_fail = 1 up_nodes.append(node) if did_fail: return self.failure("Active nodes exist: " + repr(up_nodes)) - self.logger.log("Warn: All nodes stopped but CTS didnt detect: " + self.logger.log("Warn: All nodes stopped but CTS didn't detect: " + repr(watch.unmatched)) return self.failure("Missing log message: "+repr(watch.unmatched)) def is_applicable(self): '''SimulStopLite is a setup test and never applicable''' return 0 class SimulStartLite(CTSTest): '''Start any stopped nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SimulStartLite" def __call__(self, dummy): '''Perform the 'SimulStartList' setup work. ''' self.incr("calls") self.debug("Setup: " + self.name) # We ignore the "node" parameter... node_list = [] for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "down": self.incr("WasStopped") node_list.append(node) self.set_timer() while len(node_list) > 0: # Repeat until all nodes come up watchpats = [ ] uppat = self.templates["Pat:NonDC_started"] if self.CM.upcount() == 0: uppat = self.templates["Pat:Local_started"] watchpats.append(self.templates["Pat:DC_IDLE"]) for node in node_list: watchpats.append(uppat % node) watchpats.append(self.templates["Pat:InfraUp"] % node) watchpats.append(self.templates["Pat:PacemakerUp"] % node) # Start all the nodes - at about the same time... watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() stonith = self.CM.prepare_fencing_watcher(self.name) for node in node_list: self.CM.StartaCMnoBlock(node) watch.lookforall() node_list = self.CM.fencing_cleanup(self.name, stonith) if node_list == None: return self.failure("Cluster did not stabilize") # Remove node_list messages from watch.unmatched for node in node_list: self.logger.debug("Dealing with stonith operations for %s" % repr(node_list)) if watch.unmatched: try: watch.unmatched.remove(uppat % node) except: self.debug("Already matched: %s" % (uppat % node)) try: watch.unmatched.remove(self.templates["Pat:InfraUp"] % node) except: self.debug("Already matched: %s" % (self.templates["Pat:InfraUp"] % node)) try: watch.unmatched.remove(self.templates["Pat:PacemakerUp"] % node) except: self.debug("Already matched: %s" % (self.templates["Pat:PacemakerUp"] % node)) if watch.unmatched: for regex in watch.unmatched: self.logger.log ("Warn: Startup pattern not found: %s" %(regex)) if not self.CM.cluster_stable(): return self.failure("Cluster did not stabilize") did_fail = 0 unstable = [] for node in self.Env["nodes"]: if self.CM.StataCM(node) == 0: did_fail = 1 unstable.append(node) if did_fail: return self.failure("Unstarted nodes exist: " + repr(unstable)) unstable = [] for node in self.Env["nodes"]: if not self.CM.node_stable(node): did_fail = 1 unstable.append(node) if did_fail: return self.failure("Unstable cluster nodes exist: " + repr(unstable)) return self.success() def is_applicable(self): '''SimulStartLite is a setup test and never applicable''' return 0 def TestList(cm, audits): result = [] for testclass in AllTestClasses: bound_test = testclass(cm) if bound_test.is_applicable(): bound_test.Audits = audits result.append(bound_test) return result class RemoteLXC(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "RemoteLXC" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.num_containers = 2 self.is_container = 1 self.is_docker_unsafe = 1 self.failed = 0 self.fail_string = "" def start_lxc_simple(self, node): # restore any artifacts laying around from a previous test. self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -s -R &>/dev/null") # generate the containers, put them in the config, add some resources to them pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("start", "lxc1")) pats.append(self.templates["Pat:RscOpOK"] % ("start", "lxc2")) pats.append(self.templates["Pat:RscOpOK"] % ("start", "lxc-ms")) pats.append(self.templates["Pat:RscOpOK"] % ("promote", "lxc-ms")) self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -g -a -m -s -c %d &>/dev/null" % self.num_containers) self.set_timer("remoteSimpleInit") watch.lookforall() self.log_timer("remoteSimpleInit") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 def cleanup_lxc_simple(self, node): pats = [ ] # if the test failed, attempt to clean up the cib and libvirt environment # as best as possible if self.failed == 1: # restore libvirt and cib self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -s -R &>/dev/null") return watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("stop", "container1")) pats.append(self.templates["Pat:RscOpOK"] % ("stop", "container2")) self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -p &>/dev/null") self.set_timer("remoteSimpleCleanup") watch.lookforall() self.log_timer("remoteSimpleCleanup") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 # cleanup libvirt self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -s -R &>/dev/null") def __call__(self, node): '''Perform the 'RemoteLXC' test. ''' self.incr("calls") ret = self.startall(None) if not ret: return self.failure("Setup failed, start all nodes failed.") rc = self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -v &>/dev/null") if rc == 1: self.log("Environment test for lxc support failed.") return self.skipped() self.start_lxc_simple(node) self.cleanup_lxc_simple(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed == 1: return self.failure(self.fail_string) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"Updating failcount for ping", r"schedulerd.*: Recover (ping|lxc-ms|container)\s*\(.*\)", # The orphaned lxc-ms resource causes an expected transition error # that is a result of the scheduler not having knowledge that the # promotable resource used to be a clone. As a result, it looks like that # resource is running in multiple locations when it shouldn't... But in # this instance we know why this error is occurring and that it is expected. r"Calculated [Tt]ransition .*pe-error", r"Resource lxc-ms .* is active on 2 nodes attempting recovery", r"Unknown operation: fail", r"VirtualDomain.*ERROR: Unable to determine emulator", ] AllTestClasses.append(RemoteLXC) class RemoteDriver(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = self.__class__.__name__ self.is_docker_unsafe = 1 self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.stop = StopTest(cm) self.remote_rsc = "remote-rsc" self.cib_cmd = """cibadmin -C -o %s -X '%s' """ self.reset() def reset(self): self.pcmk_started = 0 self.failed = False self.fail_string = "" self.remote_node_added = 0 self.remote_rsc_added = 0 self.remote_use_reconnect_interval = self.Env.RandomGen.choice([True,False]) def fail(self, msg): """ Mark test as failed. """ self.failed = True # Always log the failure. self.logger.log(msg) # Use first failure as test status, as it's likely to be most useful. if not self.fail_string: self.fail_string = msg def get_othernode(self, node): for othernode in self.Env["nodes"]: if othernode == node: # we don't want to try and use the cib that we just shutdown. # find a cluster node that is not our soon to be remote-node. continue else: return othernode def del_rsc(self, node, rsc): othernode = self.get_othernode(node) rc = self.rsh(othernode, "crm_resource -D -r %s -t primitive" % (rsc)) if rc != 0: self.fail("Removal of resource '%s' failed" % rsc) def add_rsc(self, node, rsc_xml): othernode = self.get_othernode(node) rc = self.rsh(othernode, self.cib_cmd % ("resources", rsc_xml)) if rc != 0: self.fail("resource creation failed") def add_primitive_rsc(self, node): rsc_xml = """ """ % { "node": self.remote_rsc } self.add_rsc(node, rsc_xml) if not self.failed: self.remote_rsc_added = 1 def add_connection_rsc(self, node): rsc_xml = """ """ % { "node": self.remote_node, "server": node } if self.remote_use_reconnect_interval: # Set reconnect interval on resource rsc_xml = rsc_xml + """ """ % (self.remote_node) rsc_xml = rsc_xml + """ """ % { "node": self.remote_node } self.add_rsc(node, rsc_xml) if not self.failed: self.remote_node_added = 1 def disable_services(self, node): self.corosync_enabled = self.Env.service_is_enabled(node, "corosync") if self.corosync_enabled: self.Env.disable_service(node, "corosync") self.pacemaker_enabled = self.Env.service_is_enabled(node, "pacemaker") if self.pacemaker_enabled: self.Env.disable_service(node, "pacemaker") def restore_services(self, node): if self.corosync_enabled: self.Env.enable_service(node, "corosync") if self.pacemaker_enabled: self.Env.enable_service(node, "pacemaker") def stop_pcmk_remote(self, node): # disable pcmk remote for i in range(10): rc = self.rsh(node, "service pacemaker_remote stop") if rc != 0: time.sleep(6) else: break def start_pcmk_remote(self, node): for i in range(10): rc = self.rsh(node, "service pacemaker_remote start") if rc != 0: time.sleep(6) else: self.pcmk_started = 1 break def freeze_pcmk_remote(self, node): """ Simulate a Pacemaker Remote daemon failure. """ # We freeze the process. self.rsh(node, "killall -STOP pacemaker-remoted") def resume_pcmk_remote(self, node): # We resume the process. self.rsh(node, "killall -CONT pacemaker-remoted") def start_metal(self, node): # Cluster nodes are reused as remote nodes in remote tests. If cluster # services were enabled at boot, in case the remote node got fenced, the # cluster node would join instead of the expected remote one. Meanwhile # pacemaker_remote would not be able to start. Depending on the chances, # the situations might not be able to be orchestrated gracefully any more. # # Temporarily disable any enabled cluster serivces. self.disable_services(node) pcmk_started = 0 # make sure the resource doesn't already exist for some reason self.rsh(node, "crm_resource -D -r %s -t primitive" % (self.remote_rsc)) self.rsh(node, "crm_resource -D -r %s -t primitive" % (self.remote_node)) if not self.stop(node): self.fail("Failed to shutdown cluster node %s" % node) return self.start_pcmk_remote(node) if self.pcmk_started == 0: self.fail("Failed to start pacemaker_remote on node %s" % node) return # Convert node to baremetal now that it has shutdown the cluster stack pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("start", self.remote_node)) pats.append(self.templates["Pat:DC_IDLE"]) self.add_connection_rsc(node) self.set_timer("remoteMetalInit") watch.lookforall() self.log_timer("remoteMetalInit") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) def migrate_connection(self, node): if self.failed: return pats = [ ] pats.append(self.templates["Pat:RscOpOK"] % ("migrate_to", self.remote_node)) pats.append(self.templates["Pat:RscOpOK"] % ("migrate_from", self.remote_node)) pats.append(self.templates["Pat:DC_IDLE"]) watch = self.create_watch(pats, 120) watch.setwatch() (rc, lines) = self.rsh(node, "crm_resource -M -r %s" % (self.remote_node), None) if rc != 0: self.fail("failed to move remote node connection resource") return self.set_timer("remoteMetalMigrate") watch.lookforall() self.log_timer("remoteMetalMigrate") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) return def fail_rsc(self, node): if self.failed: return watchpats = [ ] watchpats.append(self.templates["Pat:RscRemoteOpOK"] % ("stop", self.remote_rsc, self.remote_node)) watchpats.append(self.templates["Pat:RscRemoteOpOK"] % ("start", self.remote_rsc, self.remote_node)) watchpats.append(self.templates["Pat:DC_IDLE"]) watch = self.create_watch(watchpats, 120) watch.setwatch() self.debug("causing dummy rsc to fail.") rc = self.rsh(node, "rm -f /var/run/resource-agents/Dummy*") self.set_timer("remoteRscFail") watch.lookforall() self.log_timer("remoteRscFail") if watch.unmatched: self.fail("Unmatched patterns during rsc fail: %s" % watch.unmatched) def fail_connection(self, node): if self.failed: return watchpats = [ ] watchpats.append(self.templates["Pat:FenceOpOK"] % self.remote_node) watchpats.append(self.templates["Pat:NodeFenced"] % self.remote_node) watch = self.create_watch(watchpats, 120) watch.setwatch() # freeze the pcmk remote daemon. this will result in fencing self.debug("Force stopped active remote node") self.freeze_pcmk_remote(node) self.debug("Waiting for remote node to be fenced.") self.set_timer("remoteMetalFence") watch.lookforall() self.log_timer("remoteMetalFence") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) return self.debug("Waiting for the remote node to come back up") self.CM.ns.WaitForNodeToComeUp(node, 120); pats = [ ] watch = self.create_watch(pats, 240) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("start", self.remote_node)) if self.remote_rsc_added == 1: pats.append(self.templates["Pat:RscRemoteOpOK"] % ("start", self.remote_rsc, self.remote_node)) # start the remote node again watch it integrate back into cluster. self.start_pcmk_remote(node) if self.pcmk_started == 0: self.fail("Failed to start pacemaker_remote on node %s" % node) return self.debug("Waiting for remote node to rejoin cluster after being fenced.") self.set_timer("remoteMetalRestart") watch.lookforall() self.log_timer("remoteMetalRestart") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) return def add_dummy_rsc(self, node): if self.failed: return # verify we can put a resource on the remote node pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscRemoteOpOK"] % ("start", self.remote_rsc, self.remote_node)) pats.append(self.templates["Pat:DC_IDLE"]) # Add a resource that must live on remote-node self.add_primitive_rsc(node) # force that rsc to prefer the remote node. (rc, line) = self.CM.rsh(node, "crm_resource -M -r %s -N %s -f" % (self.remote_rsc, self.remote_node), None) if rc != 0: self.fail("Failed to place remote resource on remote node.") return self.set_timer("remoteMetalRsc") watch.lookforall() self.log_timer("remoteMetalRsc") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) def test_attributes(self, node): if self.failed: return # This verifies permanent attributes can be set on a remote-node. It also # verifies the remote-node can edit its own cib node section remotely. (rc, line) = self.CM.rsh(node, "crm_attribute -l forever -n testattr -v testval -N %s" % (self.remote_node), None) if rc != 0: self.fail("Failed to set remote-node attribute. rc:%s output:%s" % (rc, line)) return (rc, line) = self.CM.rsh(node, "crm_attribute -l forever -n testattr -q -N %s" % (self.remote_node), None) if rc != 0: self.fail("Failed to get remote-node attribute") return (rc, line) = self.CM.rsh(node, "crm_attribute -l forever -n testattr -D -N %s" % (self.remote_node), None) if rc != 0: self.fail("Failed to delete remote-node attribute") return def cleanup_metal(self, node): self.restore_services(node) if self.pcmk_started == 0: return pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() if self.remote_rsc_added == 1: pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.remote_rsc)) if self.remote_node_added == 1: pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.remote_node)) self.set_timer("remoteMetalCleanup") self.resume_pcmk_remote(node) if self.remote_rsc_added == 1: # Remove dummy resource added for remote node tests self.debug("Cleaning up dummy rsc put on remote node") self.rsh(self.get_othernode(node), "crm_resource -U -r %s" % self.remote_rsc) self.del_rsc(node, self.remote_rsc) if self.remote_node_added == 1: # Remove remote node's connection resource self.debug("Cleaning up remote node connection resource") self.rsh(self.get_othernode(node), "crm_resource -U -r %s" % (self.remote_node)) self.del_rsc(node, self.remote_node) watch.lookforall() self.log_timer("remoteMetalCleanup") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) self.stop_pcmk_remote(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.remote_node_added == 1: # Remove remote node itself self.debug("Cleaning up node entry for remote node") self.rsh(self.get_othernode(node), "crm_node --force --remove %s" % self.remote_node) def setup_env(self, node): self.remote_node = "remote-%s" % (node) # we are assuming if all nodes have a key, that it is # the right key... If any node doesn't have a remote # key, we regenerate it everywhere. if self.rsh.exists_on_all("/etc/pacemaker/authkey", self.Env["nodes"]): return # create key locally (handle, keyfile) = tempfile.mkstemp(".cts") os.close(handle) subprocess.check_call(["dd", "if=/dev/urandom", "of=%s" % keyfile, "bs=4096", "count=1"], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL) # sync key throughout the cluster for node in self.Env["nodes"]: self.rsh(node, "mkdir -p --mode=0750 /etc/pacemaker") self.rsh.cp(keyfile, "root@%s:/etc/pacemaker/authkey" % node) self.rsh(node, "chgrp haclient /etc/pacemaker /etc/pacemaker/authkey") self.rsh(node, "chmod 0640 /etc/pacemaker/authkey") os.unlink(keyfile) def is_applicable(self): if not self.is_applicable_common(): return False for node in self.Env["nodes"]: rc = self.rsh(node, "which pacemaker-remoted >/dev/null 2>&1") if rc != 0: return False return True def start_new_test(self, node): self.incr("calls") self.reset() ret = self.startall(None) if not ret: return self.failure("setup failed: could not start all nodes") self.setup_env(node) self.start_metal(node) self.add_dummy_rsc(node) return True def __call__(self, node): return self.failure("This base class is not meant to be called directly.") def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"""is running on remote.*which isn't allowed""", r"""Connection terminated""", r"""Could not send remote""", ] # RemoteDriver is just a base class for other tests, so it is not added to AllTestClasses class RemoteBasic(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteBaremetal' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) self.test_attributes(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() AllTestClasses.append(RemoteBasic) class RemoteStonithd(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteStonithd' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) self.fail_connection(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() def is_applicable(self): if not RemoteDriver.is_applicable(self): return False if "DoFencing" in list(self.Env.keys()): return self.Env["DoFencing"] return True def errorstoignore(self): ignore_pats = [ r"Lost connection to Pacemaker Remote node", r"Software caused connection abort", r"pacemaker-controld.*:\s+error.*: Operation remote-.*_monitor", r"pacemaker-controld.*:\s+error.*: Result of monitor operation for remote-.*", r"schedulerd.*:\s+Recover remote-.*\s*\(.*\)", r"error: Result of monitor operation for .* on remote-.*: No executor connection", ] ignore_pats.extend(RemoteDriver.errorstoignore(self)) return ignore_pats AllTestClasses.append(RemoteStonithd) class RemoteMigrate(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteMigrate' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) self.migrate_connection(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() def is_applicable(self): if not RemoteDriver.is_applicable(self): return 0 # This test requires at least three nodes: one to convert to a # remote node, one to host the connection originally, and one # to migrate the connection to. if len(self.Env["nodes"]) < 3: return 0 return 1 AllTestClasses.append(RemoteMigrate) class RemoteRscFailure(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteRscFailure' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) # This is an important step. We are migrating the connection # before failing the resource. This verifies that the migration # has properly maintained control over the remote-node. self.migrate_connection(node) self.fail_rsc(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() def errorstoignore(self): ignore_pats = [ r"schedulerd.*: Recover remote-rsc\s*\(.*\)", r"Dummy.*: No process state file found", ] ignore_pats.extend(RemoteDriver.errorstoignore(self)) return ignore_pats def is_applicable(self): if not RemoteDriver.is_applicable(self): return 0 # This test requires at least three nodes: one to convert to a # remote node, one to host the connection originally, and one # to migrate the connection to. if len(self.Env["nodes"]) < 3: return 0 return 1 AllTestClasses.append(RemoteRscFailure) # vim:ts=4:sw=4:et: diff --git a/cts/scheduler/summary/stonith-4.summary b/cts/scheduler/summary/stonith-4.summary index ef1449c0ab..6aa0f4d6d6 100644 --- a/cts/scheduler/summary/stonith-4.summary +++ b/cts/scheduler/summary/stonith-4.summary @@ -1,40 +1,40 @@ Current cluster status: * Node List: * Node pcmk-2: pending * Node pcmk-3: pending * Node pcmk-5: UNCLEAN (offline) * Node pcmk-7: UNCLEAN (online) * Node pcmk-8: UNCLEAN (offline) * Node pcmk-9: pending * Node pcmk-10: UNCLEAN (online) * Node pcmk-11: pending * Online: [ pcmk-1 ] * OFFLINE: [ pcmk-4 pcmk-6 ] * Full List of Resources: * Fencing (stonith:fence_xvm): Stopped Transition Summary: * Fence (reboot) pcmk-10 'peer process is no longer available' * Fence (reboot) pcmk-8 'peer has not been seen by the cluster' - * Fence (reboot) pcmk-7 'peer failed the pacemaker membership criteria' + * Fence (reboot) pcmk-7 'peer failed Pacemaker membership criteria' * Fence (reboot) pcmk-5 'peer has not been seen by the cluster' * Start Fencing ( pcmk-1 ) blocked Executing Cluster Transition: * Fencing pcmk-5 (reboot) * Fencing pcmk-7 (reboot) * Fencing pcmk-8 (reboot) * Fencing pcmk-10 (reboot) Revised Cluster Status: * Node List: * Node pcmk-2: pending * Node pcmk-3: pending * Node pcmk-9: pending * Node pcmk-11: pending * Online: [ pcmk-1 ] * OFFLINE: [ pcmk-4 pcmk-5 pcmk-6 pcmk-7 pcmk-8 pcmk-10 ] * Full List of Resources: * Fencing (stonith:fence_xvm): Stopped diff --git a/daemons/controld/controld_callbacks.c b/daemons/controld/controld_callbacks.c index 1a42983135..7dba754019 100644 --- a/daemons/controld/controld_callbacks.c +++ b/daemons/controld/controld_callbacks.c @@ -1,361 +1,361 @@ /* - * Copyright 2004-2020 the Pacemaker project contributors + * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include /* From join_dc... */ extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); void crmd_ha_msg_filter(xmlNode * msg) { if (AM_I_DC) { const char *sys_from = crm_element_value(msg, F_CRM_SYS_FROM); if (pcmk__str_eq(sys_from, CRM_SYSTEM_DC, pcmk__str_casei)) { const char *from = crm_element_value(msg, F_ORIG); if (!pcmk__str_eq(from, fsa_our_uname, pcmk__str_casei)) { int level = LOG_INFO; const char *op = crm_element_value(msg, F_CRM_TASK); /* make sure the election happens NOW */ if (fsa_state != S_ELECTION) { ha_msg_input_t new_input; level = LOG_WARNING; new_input.msg = msg; register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, &new_input, __func__); } do_crm_log(level, "Another DC detected: %s (op=%s)", from, op); goto done; } } } else { const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO); if (pcmk__str_eq(sys_to, CRM_SYSTEM_DC, pcmk__str_casei)) { return; } } /* crm_log_xml_trace("HA[inbound]", msg); */ route_message(C_HA_MESSAGE, msg); done: trigger_fsa(); } /*! * \internal * \brief Check whether a node is online * * \param[in] node Node to check * * \retval -1 if completely dead * \retval 0 if partially alive * \retval 1 if completely alive */ static int node_alive(const crm_node_t *node) { if (pcmk_is_set(node->flags, crm_remote_node)) { // Pacemaker Remote nodes can't be partially alive return pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei) ? 1: -1; } else if (crm_is_peer_active(node)) { // Completely up cluster node: both cluster member and peer return 1; } else if (!pcmk_is_set(node->processes, crm_get_cluster_proc()) && !pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei)) { // Completely down cluster node: neither cluster member nor peer return -1; } // Partially up cluster node: only cluster member or only peer return 0; } #define state_text(state) ((state)? (const char *)(state) : "in unknown state") bool controld_dc_left = false; void peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data) { uint32_t old = 0; bool appeared = FALSE; bool is_remote = pcmk_is_set(node->flags, crm_remote_node); /* The controller waits to receive some information from the membership * layer before declaring itself operational. If this is being called for a * cluster node, indicate that we have it. */ if (!is_remote) { controld_set_fsa_input_flags(R_PEER_DATA); } if (type == crm_status_processes && pcmk_is_set(node->processes, crm_get_cluster_proc()) && !AM_I_DC && !is_remote) { /* * This is a hack until we can send to a nodeid and/or we fix node name lookups * These messages are ignored in crmd_ha_msg_filter() */ xmlNode *query = create_request(CRM_OP_HELLO, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); crm_debug("Sending hello to node %u so that it learns our node name", node->id); send_cluster_message(node, crm_msg_crmd, query, FALSE); free_xml(query); } if (node->uname == NULL) { return; } switch (type) { case crm_status_uname: /* If we've never seen the node, then it also won't be in the status section */ crm_info("%s node %s is now %s", (is_remote? "Remote" : "Cluster"), node->uname, state_text(node->state)); return; case crm_status_nstate: /* This callback should not be called unless the state actually * changed, but here's a failsafe just in case. */ CRM_CHECK(!pcmk__str_eq(data, node->state, pcmk__str_casei), return); crm_info("%s node %s is now %s (was %s)", (is_remote? "Remote" : "Cluster"), node->uname, state_text(node->state), state_text(data)); if (pcmk__str_eq(CRM_NODE_MEMBER, node->state, pcmk__str_casei)) { appeared = TRUE; if (!is_remote) { remove_stonith_cleanup(node->uname); } } else { controld_remove_voter(node->uname); } crmd_alert_node_event(node); break; case crm_status_processes: CRM_CHECK(data != NULL, return); old = *(const uint32_t *)data; appeared = pcmk_is_set(node->processes, crm_get_cluster_proc()); crm_info("Node %s is %s a peer " CRM_XS " DC=%s old=0x%07x new=0x%07x", node->uname, (appeared? "now" : "no longer"), (AM_I_DC? "true" : (fsa_our_dc? fsa_our_dc : "")), old, node->processes); if (!pcmk_is_set((node->processes ^ old), crm_get_cluster_proc())) { /* Peer status did not change. This should not be possible, * since we don't track process flags other than peer status. */ crm_trace("Process flag 0x%7x did not change from 0x%7x to 0x%7x", crm_get_cluster_proc(), old, node->processes); return; } if (!appeared) { controld_remove_voter(node->uname); } if (!pcmk_is_set(fsa_input_register, R_CIB_CONNECTED)) { crm_trace("Ignoring peer status change because not connected to CIB"); return; } else if (fsa_state == S_STOPPING) { crm_trace("Ignoring peer status change because stopping"); return; } if (pcmk__str_eq(node->uname, fsa_our_uname, pcmk__str_casei) && !appeared) { /* Did we get evicted? */ crm_notice("Our peer connection failed"); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL); } else if (pcmk__str_eq(node->uname, fsa_our_dc, pcmk__str_casei) && crm_is_peer_active(node) == FALSE) { /* Did the DC leave us? */ crm_notice("Our peer on the DC (%s) is dead", fsa_our_dc); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL); /* @COMPAT DC < 1.1.13: If a DC shuts down normally, we don't * want to fence it. Newer DCs will send their shutdown request * to all peers, who will update the DC's expected state to * down, thus avoiding fencing. We can safely erase the DC's * transient attributes when it leaves in that case. However, * the only way to avoid fencing older DCs is to leave the * transient attributes intact until it rejoins. */ if (compare_version(fsa_our_dc_version, "3.0.9") > 0) { controld_delete_node_state(node->uname, controld_section_attrs, cib_scope_local); } } else if (AM_I_DC || controld_dc_left || (fsa_our_dc == NULL)) { /* This only needs to be done once, so normally the DC should do * it. However if there is no DC, every node must do it, since * there is no other way to ensure some one node does it. */ if (appeared) { te_trigger_stonith_history_sync(FALSE); } else { controld_delete_node_state(node->uname, controld_section_attrs, cib_scope_local); } } break; } if (AM_I_DC) { xmlNode *update = NULL; int flags = node_update_peer; int alive = node_alive(node); crm_action_t *down = match_down_event(node->uuid); crm_trace("Alive=%d, appeared=%d, down=%d", alive, appeared, (down? down->id : -1)); if (appeared && (alive > 0) && !is_remote) { register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } if (down) { const char *task = crm_element_value(down->xml, XML_LRM_ATTR_TASK); if (pcmk__str_eq(task, CRM_OP_FENCE, pcmk__str_casei)) { /* tengine_stonith_callback() confirms fence actions */ crm_trace("Updating CIB %s fencer reported fencing of %s complete", (down->confirmed? "after" : "before"), node->uname); } else if (!appeared && pcmk__str_eq(task, CRM_OP_SHUTDOWN, pcmk__str_casei)) { // Shutdown actions are immediately confirmed (i.e. no_wait) if (!is_remote) { flags |= node_update_join | node_update_expected; crmd_peer_down(node, FALSE); check_join_state(fsa_state, __func__); } if (alive >= 0) { crm_info("%s of peer %s is in progress " CRM_XS " action=%d", task, node->uname, down->id); } else { crm_notice("%s of peer %s is complete " CRM_XS " action=%d", task, node->uname, down->id); pcmk__update_graph(transition_graph, down); trigger_graph(); } } else { crm_trace("Node %s is %s, was expected to %s (op %d)", node->uname, ((alive > 0)? "alive" : ((alive < 0)? "dead" : "partially alive")), task, down->id); } } else if (appeared == FALSE) { crm_warn("Stonith/shutdown of node %s was not expected", node->uname); if (!is_remote) { crm_update_peer_join(__func__, node, crm_join_none); check_join_state(fsa_state, __func__); } abort_transition(INFINITY, tg_restart, "Node failure", NULL); fail_incompletable_actions(transition_graph, node->uuid); } else { crm_trace("Node %s came up, was not expected to be down", node->uname); } if (is_remote) { /* A pacemaker_remote node won't have its cluster status updated * in the CIB by membership-layer callbacks, so do it here. */ flags |= node_update_cluster; /* Trigger resource placement on newly integrated nodes */ if (appeared) { abort_transition(INFINITY, tg_restart, - "pacemaker_remote node integrated", NULL); + "Pacemaker Remote node integrated", NULL); } } /* Update the CIB node state */ update = create_node_state_update(node, flags, NULL, __func__); if (update == NULL) { crm_debug("Node state update not yet possible for %s", node->uname); } else { fsa_cib_anon_update(XML_CIB_TAG_STATUS, update); } free_xml(update); } trigger_fsa(); } void crmd_cib_connection_destroy(gpointer user_data) { CRM_CHECK(user_data == fsa_cib_conn,;); crm_trace("Invoked"); trigger_fsa(); fsa_cib_conn->state = cib_disconnected; if (!pcmk_is_set(fsa_input_register, R_CIB_CONNECTED)) { crm_info("Connection to the CIB manager terminated"); return; } // @TODO This should trigger a reconnect, not a shutdown crm_crit("Lost connection to the CIB manager, shutting down"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_CIB_CONNECTED); return; } gboolean crm_fsa_trigger(gpointer user_data) { crm_trace("Invoked (queue len: %d)", g_list_length(fsa_message_queue)); s_crmd_fsa(C_FSA_INTERNAL); crm_trace("Exited (queue len: %d)", g_list_length(fsa_message_queue)); return TRUE; } diff --git a/daemons/controld/controld_control.c b/daemons/controld/controld_control.c index b5da6a46c7..6f25799b4e 100644 --- a/daemons/controld/controld_control.c +++ b/daemons/controld/controld_control.c @@ -1,838 +1,838 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include qb_ipcs_service_t *ipcs = NULL; #if SUPPORT_COROSYNC extern gboolean crm_connect_corosync(crm_cluster_t * cluster); #endif void crm_shutdown(int nsig); gboolean crm_read_options(gpointer user_data); gboolean fsa_has_quorum = FALSE; crm_trigger_t *fsa_source = NULL; crm_trigger_t *config_read = NULL; bool no_quorum_suicide_escalation = FALSE; bool controld_shutdown_lock_enabled = false; /* A_HA_CONNECT */ void do_ha_control(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { gboolean registered = FALSE; static crm_cluster_t *cluster = NULL; if (cluster == NULL) { cluster = calloc(1, sizeof(crm_cluster_t)); } if (action & A_HA_DISCONNECT) { crm_cluster_disconnect(cluster); crm_info("Disconnected from the cluster"); controld_set_fsa_input_flags(R_HA_DISCONNECTED); } if (action & A_HA_CONNECT) { crm_set_status_callback(&peer_update_callback); crm_set_autoreap(FALSE); if (is_corosync_cluster()) { #if SUPPORT_COROSYNC registered = crm_connect_corosync(cluster); #endif } if (registered == TRUE) { controld_election_init(cluster->uname); fsa_our_uname = cluster->uname; fsa_our_uuid = cluster->uuid; if(cluster->uuid == NULL) { crm_err("Could not obtain local uuid"); registered = FALSE; } } if (registered == FALSE) { controld_set_fsa_input_flags(R_HA_DISCONNECTED); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } populate_cib_nodes(node_update_none, __func__); controld_clear_fsa_input_flags(R_HA_DISCONNECTED); crm_info("Connected to the cluster"); } if (action & ~(A_HA_CONNECT | A_HA_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __func__); } } /* A_SHUTDOWN */ void do_shutdown(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { /* just in case */ controld_set_fsa_input_flags(R_SHUTDOWN); controld_disconnect_fencer(FALSE); } /* A_SHUTDOWN_REQ */ void do_shutdown_req(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { xmlNode *msg = NULL; controld_set_fsa_input_flags(R_SHUTDOWN); //controld_set_fsa_input_flags(R_STAYDOWN); crm_info("Sending shutdown request to all peers (DC is %s)", (fsa_our_dc? fsa_our_dc : "not set")); msg = create_request(CRM_OP_SHUTDOWN_REQ, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } free_xml(msg); } extern char *max_generation_from; extern xmlNode *max_generation_xml; extern GHashTable *resource_history; extern GHashTable *voted; void crmd_fast_exit(crm_exit_t exit_code) { if (pcmk_is_set(fsa_input_register, R_STAYDOWN)) { crm_warn("Inhibiting respawn "CRM_XS" remapping exit code %d to %d", exit_code, CRM_EX_FATAL); exit_code = CRM_EX_FATAL; } else if ((exit_code == CRM_EX_OK) && pcmk_is_set(fsa_input_register, R_IN_RECOVERY)) { crm_err("Could not recover from internal error"); exit_code = CRM_EX_ERROR; } crm_exit(exit_code); } crm_exit_t crmd_exit(crm_exit_t exit_code) { GList *gIter = NULL; GMainLoop *mloop = crmd_mainloop; static bool in_progress = FALSE; if (in_progress && (exit_code == CRM_EX_OK)) { crm_debug("Exit is already in progress"); return exit_code; } else if(in_progress) { crm_notice("Error during shutdown process, exiting now with status %d (%s)", exit_code, crm_exit_str(exit_code)); crm_write_blackbox(SIGTRAP, NULL); crmd_fast_exit(exit_code); } in_progress = TRUE; crm_trace("Preparing to exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); /* Suppress secondary errors resulting from us disconnecting everything */ controld_set_fsa_input_flags(R_HA_DISCONNECTED); /* Close all IPC servers and clients to ensure any and all shared memory files are cleaned up */ if(ipcs) { crm_trace("Closing IPC server"); mainloop_del_ipc_server(ipcs); ipcs = NULL; } controld_close_attrd_ipc(); pe_subsystem_free(); controld_disconnect_fencer(TRUE); if ((exit_code == CRM_EX_OK) && (crmd_mainloop == NULL)) { crm_debug("No mainloop detected"); exit_code = CRM_EX_ERROR; } /* On an error, just get out. * * Otherwise, make the effort to have mainloop exit gracefully so * that it (mostly) cleans up after itself and valgrind has less * to report on - allowing real errors stand out */ if (exit_code != CRM_EX_OK) { crm_notice("Forcing immediate exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); crm_write_blackbox(SIGTRAP, NULL); crmd_fast_exit(exit_code); } /* Clean up as much memory as possible for valgrind */ for (gIter = fsa_message_queue; gIter != NULL; gIter = gIter->next) { fsa_data_t *fsa_data = gIter->data; crm_info("Dropping %s: [ state=%s cause=%s origin=%s ]", fsa_input2string(fsa_data->fsa_input), fsa_state2string(fsa_state), fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin); delete_fsa_input(fsa_data); } controld_clear_fsa_input_flags(R_MEMBERSHIP); g_list_free(fsa_message_queue); fsa_message_queue = NULL; metadata_cache_fini(); controld_election_fini(); /* Tear down the CIB manager connection, but don't free it yet -- it could * be used when we drain the mainloop later. */ fsa_cib_conn->cmds->del_notify_callback(fsa_cib_conn, T_CIB_REPLACE_NOTIFY, do_cib_replaced); fsa_cib_conn->cmds->del_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, do_cib_updated); cib_free_callbacks(fsa_cib_conn); fsa_cib_conn->cmds->signoff(fsa_cib_conn); verify_stopped(fsa_state, LOG_WARNING); controld_clear_fsa_input_flags(R_LRM_CONNECTED); lrm_state_destroy_all(); /* This basically will not work, since mainloop has a reference to it */ mainloop_destroy_trigger(fsa_source); fsa_source = NULL; mainloop_destroy_trigger(config_read); config_read = NULL; mainloop_destroy_trigger(transition_trigger); transition_trigger = NULL; pcmk__client_cleanup(); crm_peer_destroy(); controld_free_fsa_timers(); te_cleanup_stonith_history_sync(NULL, TRUE); controld_free_sched_timer(); free(fsa_our_dc_version); fsa_our_dc_version = NULL; free(fsa_our_uname); fsa_our_uname = NULL; free(fsa_our_uuid); fsa_our_uuid = NULL; free(fsa_our_dc); fsa_our_dc = NULL; free(fsa_cluster_name); fsa_cluster_name = NULL; free(te_uuid); te_uuid = NULL; free(failed_stop_offset); failed_stop_offset = NULL; free(failed_start_offset); failed_start_offset = NULL; free(max_generation_from); max_generation_from = NULL; free_xml(max_generation_xml); max_generation_xml = NULL; mainloop_destroy_signal(SIGPIPE); mainloop_destroy_signal(SIGUSR1); mainloop_destroy_signal(SIGTERM); mainloop_destroy_signal(SIGTRAP); /* leave SIGCHLD engaged as we might still want to drain some service-actions */ if (mloop) { GMainContext *ctx = g_main_loop_get_context(crmd_mainloop); /* Don't re-enter this block */ crmd_mainloop = NULL; /* no signals on final draining anymore */ mainloop_destroy_signal(SIGCHLD); crm_trace("Draining mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx)); { int lpc = 0; while((g_main_context_pending(ctx) && lpc < 10)) { lpc++; crm_trace("Iteration %d", lpc); g_main_context_dispatch(ctx); } } crm_trace("Closing mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx)); g_main_loop_quit(mloop); /* Won't do anything yet, since we're inside it now */ g_main_loop_unref(mloop); } else { mainloop_destroy_signal(SIGCHLD); } cib_delete(fsa_cib_conn); fsa_cib_conn = NULL; throttle_fini(); /* Graceful */ crm_trace("Done preparing for exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); return exit_code; } /* A_EXIT_0, A_EXIT_1 */ void do_exit(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_exit_t exit_code = CRM_EX_OK; int log_level = LOG_INFO; const char *exit_type = "gracefully"; if (action & A_EXIT_1) { log_level = LOG_ERR; exit_type = "forcefully"; exit_code = CRM_EX_ERROR; } verify_stopped(cur_state, LOG_ERR); do_crm_log(log_level, "Performing %s - %s exiting the controller", fsa_action2string(action), exit_type); crm_info("[%s] stopped (%d)", crm_system_name, exit_code); crmd_exit(exit_code); } static void sigpipe_ignore(int nsig) { return; } /* A_STARTUP */ void do_startup(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_debug("Registering Signal Handlers"); mainloop_add_signal(SIGTERM, crm_shutdown); mainloop_add_signal(SIGPIPE, sigpipe_ignore); fsa_source = mainloop_add_trigger(G_PRIORITY_HIGH, crm_fsa_trigger, NULL); config_read = mainloop_add_trigger(G_PRIORITY_HIGH, crm_read_options, NULL); transition_trigger = mainloop_add_trigger(G_PRIORITY_LOW, te_graph_trigger, NULL); crm_debug("Creating CIB manager and executor objects"); fsa_cib_conn = cib_new(); lrm_state_init_local(); if (controld_init_fsa_timers() == FALSE) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } // \return libqb error code (0 on success, -errno on error) static int32_t accept_controller_client(qb_ipcs_connection_t *c, uid_t uid, gid_t gid) { crm_trace("Accepting new IPC client connection"); if (pcmk__new_client(c, uid, gid) == NULL) { return -EIO; } return 0; } // \return libqb error code (0 on success, -errno on error) static int32_t dispatch_controller_ipc(qb_ipcs_connection_t * c, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; pcmk__client_t *client = pcmk__find_client(c); xmlNode *msg = pcmk__client_data2xml(client, data, &id, &flags); if (msg == NULL) { pcmk__ipc_send_ack(client, id, flags, "ack", CRM_EX_PROTOCOL); return 0; } pcmk__ipc_send_ack(client, id, flags, "ack", CRM_EX_INDETERMINATE); CRM_ASSERT(client->user != NULL); pcmk__update_acl_user(msg, F_CRM_USER, client->user); crm_xml_add(msg, F_CRM_SYS_FROM, client->id); if (controld_authorize_ipc_message(msg, client, NULL)) { crm_trace("Processing IPC message from client %s", pcmk__client_name(client)); route_message(C_IPC_MESSAGE, msg); } trigger_fsa(); free_xml(msg); return 0; } static int32_t crmd_ipc_closed(qb_ipcs_connection_t * c) { pcmk__client_t *client = pcmk__find_client(c); if (client) { crm_trace("Disconnecting %sregistered client %s (%p/%p)", (client->userdata? "" : "un"), pcmk__client_name(client), c, client); free(client->userdata); pcmk__free_client(client); trigger_fsa(); } return 0; } static void crmd_ipc_destroy(qb_ipcs_connection_t * c) { crm_trace("Connection %p", c); crmd_ipc_closed(c); } /* A_STOP */ void do_stop(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_trace("Closing IPC server"); mainloop_del_ipc_server(ipcs); ipcs = NULL; register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL); } /* A_STARTED */ void do_started(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { static struct qb_ipcs_service_handlers crmd_callbacks = { .connection_accept = accept_controller_client, .connection_created = NULL, .msg_process = dispatch_controller_ipc, .connection_closed = crmd_ipc_closed, .connection_destroyed = crmd_ipc_destroy }; if (cur_state != S_STARTING) { crm_err("Start cancelled... %s", fsa_state2string(cur_state)); return; } else if (!pcmk_is_set(fsa_input_register, R_MEMBERSHIP)) { crm_info("Delaying start, no membership data (%.16llx)", R_MEMBERSHIP); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(fsa_input_register, R_LRM_CONNECTED)) { crm_info("Delaying start, not connected to executor (%.16llx)", R_LRM_CONNECTED); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(fsa_input_register, R_CIB_CONNECTED)) { crm_info("Delaying start, CIB not connected (%.16llx)", R_CIB_CONNECTED); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(fsa_input_register, R_READ_CONFIG)) { crm_info("Delaying start, Config not read (%.16llx)", R_READ_CONFIG); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(fsa_input_register, R_PEER_DATA)) { crm_info("Delaying start, No peer data (%.16llx)", R_PEER_DATA); crmd_fsa_stall(TRUE); return; } crm_debug("Init server comms"); ipcs = pcmk__serve_controld_ipc(&crmd_callbacks); if (ipcs == NULL) { crm_err("Failed to create IPC server: shutting down and inhibiting respawn"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else { crm_notice("Pacemaker controller successfully started and accepting connections"); } controld_trigger_fencer_connect(); controld_clear_fsa_input_flags(R_STARTING); register_fsa_input(msg_data->fsa_cause, I_PENDING, NULL); } /* A_RECOVER */ void do_recover(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { controld_set_fsa_input_flags(R_IN_RECOVERY); crm_warn("Fast-tracking shutdown in response to errors"); register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL); } static pcmk__cluster_option_t crmd_opts[] = { /* name, old name, type, allowed values, * default value, validator, * short description, * long description */ { "dc-version", NULL, "string", NULL, "none", NULL, "Pacemaker version on cluster node elected Designated Controller (DC)", "Includes a hash which identifies the exact changeset the code was " "built from. Used for diagnostic purposes." }, { "cluster-infrastructure", NULL, "string", NULL, "corosync", NULL, "The messaging stack on which Pacemaker is currently running", "Used for informational and diagnostic purposes." }, { "cluster-name", NULL, "string", NULL, NULL, NULL, "An arbitrary name for the cluster", "This optional value is mostly for users' convenience as desired " "in administration, but may also be used in Pacemaker " "configuration rules via the #cluster-name node attribute, and " "by higher-level tools and resource agents." }, { XML_CONFIG_ATTR_DC_DEADTIME, NULL, "time", NULL, "20s", pcmk__valid_interval_spec, "How long to wait for a response from other nodes during start-up", "The optimal value will depend on the speed and load of your network " "and the type of switches used." }, { XML_CONFIG_ATTR_RECHECK, NULL, "time", "Zero disables polling, while positive values are an interval in seconds" "(unless other units are specified, for example \"5min\")", "15min", pcmk__valid_interval_spec, "Polling interval to recheck cluster state and evaluate rules " "with date specifications", "Pacemaker is primarily event-driven, and looks ahead to know when to " "recheck cluster state for failure timeouts and most time-based " "rules. However, it will also recheck the cluster after this " "amount of inactivity, to evaluate rules with date specifications " "and serve as a fail-safe for certain types of scheduler bugs." }, { "load-threshold", NULL, "percentage", NULL, "80%", pcmk__valid_utilization, "Maximum amount of system load that should be used by cluster nodes", "The cluster will slow down its recovery process when the amount of " "system resources used (currently CPU) approaches this limit", }, { "node-action-limit", NULL, "integer", NULL, "0", pcmk__valid_number, "Maximum number of jobs that can be scheduled per node " "(defaults to 2x cores)" }, { XML_CONFIG_ATTR_FENCE_REACTION, NULL, "string", NULL, "stop", NULL, "How a cluster node should react if notified of its own fencing", "A cluster node may receive notification of its own fencing if fencing " "is misconfigured, or if fabric fencing is in use that doesn't cut " "cluster communication. Allowed values are \"stop\" to attempt to " - "immediately stop pacemaker and stay stopped, or \"panic\" to attempt " + "immediately stop Pacemaker and stay stopped, or \"panic\" to attempt " "to immediately reboot the local node, falling back to stop on failure." }, { XML_CONFIG_ATTR_ELECTION_FAIL, NULL, "time", NULL, "2min", pcmk__valid_interval_spec, "*** Advanced Use Only ***", "Declare an election failed if it is not decided within this much " "time. If you need to adjust this value, it probably indicates " "the presence of a bug." }, { XML_CONFIG_ATTR_FORCE_QUIT, NULL, "time", NULL, "20min", pcmk__valid_interval_spec, "*** Advanced Use Only ***", "Exit immediately if shutdown does not complete within this much " "time. If you need to adjust this value, it probably indicates " "the presence of a bug." }, { "join-integration-timeout", "crmd-integration-timeout", "time", NULL, "3min", pcmk__valid_interval_spec, "*** Advanced Use Only ***", "If you need to adjust this value, it probably indicates " "the presence of a bug." }, { "join-finalization-timeout", "crmd-finalization-timeout", "time", NULL, "30min", pcmk__valid_interval_spec, "*** Advanced Use Only ***", "If you need to adjust this value, it probably indicates " "the presence of a bug." }, { "transition-delay", "crmd-transition-delay", "time", NULL, "0s", pcmk__valid_interval_spec, "*** Advanced Use Only *** Enabling this option will slow down " "cluster recovery under all conditions", "Delay cluster recovery for this much time to allow for additional " "events to occur. Useful if your configuration is sensitive to " "the order in which ping updates arrive." }, { "stonith-watchdog-timeout", NULL, "time", NULL, "0", controld_verify_stonith_watchdog_timeout, "How long to wait before we can assume nodes are safely down " "when watchdog-based self-fencing via SBD is in use", "If nonzero, along with `have-watchdog=true` automatically set by the " "cluster, when fencing is required, watchdog-based self-fencing " "will be performed via SBD without requiring a fencing resource " "explicitly configured. " "If `stonith-watchdog-timeout` is set to a positive value, unseen " "nodes are assumed to self-fence within this much time. +WARNING:+ " "It must be ensured that this value is larger than the " "`SBD_WATCHDOG_TIMEOUT` environment variable on all nodes. " "Pacemaker verifies the settings individually on all nodes and " "prevents startup or shuts down if configured wrongly on the fly. " "It's strongly recommended that `SBD_WATCHDOG_TIMEOUT` is set to " "the same value on all nodes. " "If `stonith-watchdog-timeout` is set to a negative value, and " "`SBD_WATCHDOG_TIMEOUT` is set, twice that value will be used. " "+WARNING:+ In this case, it's essential (currently not verified by " - "pacemaker) that `SBD_WATCHDOG_TIMEOUT` is set to the same value on " + "Pacemaker) that `SBD_WATCHDOG_TIMEOUT` is set to the same value on " "all nodes." }, { "stonith-max-attempts", NULL, "integer", NULL, "10", pcmk__valid_positive_number, "How many times fencing can fail before it will no longer be " "immediately re-attempted on a target" }, // Already documented in libpe_status (other values must be kept identical) { "no-quorum-policy", NULL, "enum", "stop, freeze, ignore, demote, suicide", "stop", pcmk__valid_quorum, NULL, NULL }, { XML_CONFIG_ATTR_SHUTDOWN_LOCK, NULL, "boolean", NULL, "false", pcmk__valid_boolean, NULL, NULL }, }; void crmd_metadata(void) { pcmk__print_option_metadata("pacemaker-controld", "1.0", "Pacemaker controller options", "Cluster options used by Pacemaker's " "controller (formerly called crmd)", crmd_opts, PCMK__NELEM(crmd_opts)); } static void verify_crmd_options(GHashTable * options) { pcmk__validate_cluster_options(options, crmd_opts, PCMK__NELEM(crmd_opts)); } static const char * crmd_pref(GHashTable * options, const char *name) { return pcmk__cluster_option(options, crmd_opts, PCMK__NELEM(crmd_opts), name); } static void config_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { const char *value = NULL; GHashTable *config_hash = NULL; crm_time_t *now = crm_time_new(NULL); xmlNode *crmconfig = NULL; xmlNode *alerts = NULL; if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Local CIB query resulted in an error: %s", pcmk_strerror(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); if (rc == -EACCES || rc == -pcmk_err_schema_validation) { crm_err("The cluster is mis-configured - shutting down and staying down"); controld_set_fsa_input_flags(R_STAYDOWN); } goto bail; } crmconfig = output; if ((crmconfig) && (crm_element_name(crmconfig)) && (strcmp(crm_element_name(crmconfig), XML_CIB_TAG_CRMCONFIG) != 0)) { crmconfig = first_named_child(crmconfig, XML_CIB_TAG_CRMCONFIG); } if (!crmconfig) { fsa_data_t *msg_data = NULL; crm_err("Local CIB query for " XML_CIB_TAG_CRMCONFIG " section failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); goto bail; } crm_debug("Call %d : Parsing CIB options", call_id); config_hash = pcmk__strkey_table(free, free); pe_unpack_nvpairs(crmconfig, crmconfig, XML_CIB_TAG_PROPSET, NULL, config_hash, CIB_OPTIONS_FIRST, FALSE, now, NULL); verify_crmd_options(config_hash); value = crmd_pref(config_hash, XML_CONFIG_ATTR_DC_DEADTIME); election_trigger->period_ms = crm_parse_interval_spec(value); value = crmd_pref(config_hash, "node-action-limit"); /* Also checks migration-limit */ throttle_update_job_max(value); value = crmd_pref(config_hash, "load-threshold"); if(value) { throttle_set_load_target(strtof(value, NULL) / 100.0); } value = crmd_pref(config_hash, "no-quorum-policy"); if (pcmk__str_eq(value, "suicide", pcmk__str_casei) && pcmk__locate_sbd()) { no_quorum_suicide_escalation = TRUE; } set_fence_reaction(crmd_pref(config_hash, XML_CONFIG_ATTR_FENCE_REACTION)); value = crmd_pref(config_hash,"stonith-max-attempts"); update_stonith_max_attempts(value); value = crmd_pref(config_hash, XML_CONFIG_ATTR_FORCE_QUIT); shutdown_escalation_timer->period_ms = crm_parse_interval_spec(value); crm_debug("Shutdown escalation occurs if DC has not responded to request in %ums", shutdown_escalation_timer->period_ms); value = crmd_pref(config_hash, XML_CONFIG_ATTR_ELECTION_FAIL); controld_set_election_period(value); value = crmd_pref(config_hash, XML_CONFIG_ATTR_RECHECK); recheck_interval_ms = crm_parse_interval_spec(value); crm_debug("Re-run scheduler after %dms of inactivity", recheck_interval_ms); value = crmd_pref(config_hash, "transition-delay"); transition_timer->period_ms = crm_parse_interval_spec(value); value = crmd_pref(config_hash, "join-integration-timeout"); integration_timer->period_ms = crm_parse_interval_spec(value); value = crmd_pref(config_hash, "join-finalization-timeout"); finalization_timer->period_ms = crm_parse_interval_spec(value); value = crmd_pref(config_hash, XML_CONFIG_ATTR_SHUTDOWN_LOCK); controld_shutdown_lock_enabled = crm_is_true(value); free(fsa_cluster_name); fsa_cluster_name = NULL; value = g_hash_table_lookup(config_hash, "cluster-name"); if (value) { fsa_cluster_name = strdup(value); } alerts = first_named_child(output, XML_CIB_TAG_ALERTS); crmd_unpack_alerts(alerts); controld_set_fsa_input_flags(R_READ_CONFIG); crm_trace("Triggering FSA: %s", __func__); mainloop_set_trigger(fsa_source); g_hash_table_destroy(config_hash); bail: crm_time_free(now); } gboolean crm_read_options(gpointer user_data) { int call_id = fsa_cib_conn->cmds->query(fsa_cib_conn, "//" XML_CIB_TAG_CRMCONFIG " | //" XML_CIB_TAG_ALERTS, NULL, cib_xpath | cib_scope_local); fsa_register_cib_callback(call_id, FALSE, NULL, config_query_callback); crm_trace("Querying the CIB... call %d", call_id); return TRUE; } /* A_READCONFIG */ void do_read_config(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { throttle_init(); mainloop_set_trigger(config_read); } void crm_shutdown(int nsig) { if ((crmd_mainloop == NULL) || !g_main_loop_is_running(crmd_mainloop)) { crmd_exit(CRM_EX_OK); return; } if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)) { crm_err("Escalating shutdown"); register_fsa_input_before(C_SHUTDOWN, I_ERROR, NULL); return; } controld_set_fsa_input_flags(R_SHUTDOWN); register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); if (shutdown_escalation_timer->period_ms == 0) { const char *value = crmd_pref(NULL, XML_CONFIG_ATTR_FORCE_QUIT); shutdown_escalation_timer->period_ms = crm_parse_interval_spec(value); } crm_notice("Initiating controller shutdown sequence " CRM_XS " limit=%ums", shutdown_escalation_timer->period_ms); controld_start_timer(shutdown_escalation_timer); } diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c index 90261a36dd..aef32d3b72 100644 --- a/daemons/controld/controld_execd.c +++ b/daemons/controld/controld_execd.c @@ -1,2886 +1,2894 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include // lrmd_event_data_t, lrmd_rsc_info_t, etc. #include #include #include #include #include #include #define START_DELAY_THRESHOLD 5 * 60 * 1000 #define MAX_LRM_REG_FAILS 30 struct delete_event_s { int rc; const char *rsc; lrm_state_t *lrm_state; }; static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id); static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list); static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data); static lrmd_event_data_t *construct_op(lrm_state_t * lrm_state, xmlNode * rsc_op, const char *rsc_id, const char *operation); static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *operation, xmlNode *msg); static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level); static int do_update_resource(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, time_t lock_time); static void lrm_connection_destroy(void) { if (pcmk_is_set(fsa_input_register, R_LRM_CONNECTED)) { crm_crit("Connection to executor failed"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_LRM_CONNECTED); } else { crm_info("Disconnected from executor"); } } static char * make_stop_id(const char *rsc, int call_id) { return crm_strdup_printf("%s:%d", rsc, call_id); } static void copy_instance_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") == NULL) { g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value)); } } static void copy_meta_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") != NULL) { g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value)); } } /*! * \internal * \brief Remove a recurring operation from a resource's history * * \param[in,out] history Resource history to modify * \param[in] op Operation to remove * * \return TRUE if the operation was found and removed, FALSE otherwise */ static gboolean history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_event_data_t *existing = iter->data; if ((op->interval_ms == existing->interval_ms) && pcmk__str_eq(op->rsc_id, existing->rsc_id, pcmk__str_none) && pcmk__str_eq(op->op_type, existing->op_type, pcmk__str_casei)) { history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter); lrmd_free_event(existing); return TRUE; } } return FALSE; } /*! * \internal * \brief Free all recurring operations in resource history * * \param[in,out] history Resource history to modify */ static void history_free_recurring_ops(rsc_history_t *history) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_free_event(iter->data); } g_list_free(history->recurring_op_list); history->recurring_op_list = NULL; } /*! * \internal * \brief Free resource history * * \param[in,out] history Resource history to free */ void history_free(gpointer data) { rsc_history_t *history = (rsc_history_t*)data; if (history->stop_params) { g_hash_table_destroy(history->stop_params); } /* Don't need to free history->rsc.id because it's set to history->id */ free(history->rsc.type); free(history->rsc.standard); free(history->rsc.provider); lrmd_free_event(history->failed); lrmd_free_event(history->last); free(history->id); history_free_recurring_ops(history); free(history); } static void update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op) { int target_rc = 0; rsc_history_t *entry = NULL; if (op->rsc_deleted) { crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type); controld_delete_resource_history(op->rsc_id, lrm_state->node_name, NULL, crmd_cib_smart_opt()); return; } if (pcmk__str_eq(op->op_type, RSC_NOTIFY, pcmk__str_casei)) { return; } crm_debug("Updating history for '%s' with %s op", op->rsc_id, op->op_type); entry = g_hash_table_lookup(lrm_state->resource_history, op->rsc_id); if (entry == NULL && rsc) { entry = calloc(1, sizeof(rsc_history_t)); entry->id = strdup(op->rsc_id); g_hash_table_insert(lrm_state->resource_history, entry->id, entry); entry->rsc.id = entry->id; entry->rsc.type = strdup(rsc->type); entry->rsc.standard = strdup(rsc->standard); if (rsc->provider) { entry->rsc.provider = strdup(rsc->provider); } else { entry->rsc.provider = NULL; } } else if (entry == NULL) { crm_info("Resource %s no longer exists, not updating cache", op->rsc_id); return; } entry->last_callid = op->call_id; target_rc = rsc_op_expected_rc(op); if (op->op_status == PCMK_LRM_OP_CANCELLED) { if (op->interval_ms > 0) { crm_trace("Removing cancelled recurring op: " PCMK__OP_FMT, op->rsc_id, op->op_type, op->interval_ms); history_remove_recurring_op(entry, op); return; } else { crm_trace("Skipping " PCMK__OP_FMT " rc=%d, status=%d", op->rsc_id, op->op_type, op->interval_ms, op->rc, op->op_status); } } else if (did_rsc_op_fail(op, target_rc)) { /* Store failed monitors here, otherwise the block below will cause them * to be forgotten when a stop happens. */ if (entry->failed) { lrmd_free_event(entry->failed); } entry->failed = lrmd_copy_event(op); } else if (op->interval_ms == 0) { if (entry->last) { lrmd_free_event(entry->last); } entry->last = lrmd_copy_event(op); if (op->params && pcmk__strcase_any_of(op->op_type, CRMD_ACTION_START, CRMD_ACTION_RELOAD, CRMD_ACTION_RELOAD_AGENT, CRMD_ACTION_STATUS, NULL)) { if (entry->stop_params) { g_hash_table_destroy(entry->stop_params); } entry->stop_params = pcmk__strkey_table(free, free); g_hash_table_foreach(op->params, copy_instance_keys, entry->stop_params); } } if (op->interval_ms > 0) { /* Ensure there are no duplicates */ history_remove_recurring_op(entry, op); crm_trace("Adding recurring op: " PCMK__OP_FMT, op->rsc_id, op->op_type, op->interval_ms); entry->recurring_op_list = g_list_prepend(entry->recurring_op_list, lrmd_copy_event(op)); } else if (entry->recurring_op_list && !pcmk__str_eq(op->op_type, RSC_STATUS, pcmk__str_casei)) { crm_trace("Dropping %d recurring ops because of: " PCMK__OP_FMT, g_list_length(entry->recurring_op_list), op->rsc_id, op->op_type, op->interval_ms); history_free_recurring_ops(entry); } } /*! * \internal * \brief Send a direct OK ack for a resource task * * \param[in] lrm_state LRM connection * \param[in] input Input message being ack'ed * \param[in] rsc_id ID of affected resource * \param[in] rsc Affected resource (if available) * \param[in] task Operation task being ack'ed * \param[in] ack_host Name of host to send ack to * \param[in] ack_sys IPC system name to ack */ static void send_task_ok_ack(lrm_state_t *lrm_state, ha_msg_input_t *input, const char *rsc_id, lrmd_rsc_info_t *rsc, const char *task, const char *ack_host, const char *ack_sys) { lrmd_event_data_t *op = construct_op(lrm_state, input->xml, rsc_id, task); op->rc = PCMK_OCF_OK; op->op_status = PCMK_LRM_OP_DONE; controld_ack_event_directly(ack_host, ack_sys, rsc, op, rsc_id); lrmd_free_event(op); } static inline const char * op_node_name(lrmd_event_data_t *op) { return op->remote_nodename? op->remote_nodename : fsa_our_uname; } void lrm_op_callback(lrmd_event_data_t * op) { CRM_CHECK(op != NULL, return); switch (op->type) { case lrmd_event_disconnect: if (op->remote_nodename == NULL) { /* If this is the local executor IPC connection, set the right * bits in the controller when the connection goes down. */ lrm_connection_destroy(); } break; case lrmd_event_exec_complete: { lrm_state_t *lrm_state = lrm_state_find(op_node_name(op)); CRM_ASSERT(lrm_state != NULL); process_lrm_event(lrm_state, op, NULL, NULL); } break; default: break; } } +static void +try_local_executor_connect(long long action, fsa_data_t *msg_data, + lrm_state_t *lrm_state) +{ + int rc = pcmk_rc_ok; + + crm_debug("Connecting to the local executor"); + + // If we can connect, great + rc = controld_connect_local_executor(lrm_state); + if (rc == pcmk_rc_ok) { + controld_set_fsa_input_flags(R_LRM_CONNECTED); + crm_info("Connection to the local executor established"); + return; + } + + // Otherwise, if we can try again, set a timer to do so + if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) { + crm_warn("Failed to connect to the local executor %d time%s " + "(%d max): %s", lrm_state->num_lrm_register_fails, + pcmk__plural_s(lrm_state->num_lrm_register_fails), + MAX_LRM_REG_FAILS, pcmk_rc_str(rc)); + controld_start_timer(wait_timer); + crmd_fsa_stall(FALSE); + return; + } + + // Otherwise give up + crm_err("Failed to connect to the executor the max allowed " + "%d time%s: %s", lrm_state->num_lrm_register_fails, + pcmk__plural_s(lrm_state->num_lrm_register_fails), + pcmk_rc_str(rc)); + register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); +} + /* A_LRM_CONNECT */ void do_lrm_control(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { /* This only pertains to local executor connections. Remote connections are * handled as resources within the scheduler. Connecting and disconnecting * from remote executor instances is handled differently. */ lrm_state_t *lrm_state = NULL; if(fsa_our_uname == NULL) { return; /* Nothing to do */ } lrm_state = lrm_state_find_or_create(fsa_our_uname); if (lrm_state == NULL) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } if (action & A_LRM_DISCONNECT) { if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) { if (action == A_LRM_DISCONNECT) { crmd_fsa_stall(FALSE); return; } } controld_clear_fsa_input_flags(R_LRM_CONNECTED); crm_info("Disconnecting from the executor"); lrm_state_disconnect(lrm_state); lrm_state_reset_tables(lrm_state, FALSE); crm_notice("Disconnected from the executor"); } if (action & A_LRM_CONNECT) { - int ret = pcmk_ok; - - crm_debug("Connecting to the executor"); - ret = lrm_state_ipc_connect(lrm_state); - - if (ret != pcmk_ok) { - if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) { - crm_warn("Failed to connect to the executor %d time%s (%d max)", - lrm_state->num_lrm_register_fails, - pcmk__plural_s(lrm_state->num_lrm_register_fails), - MAX_LRM_REG_FAILS); - - controld_start_timer(wait_timer); - crmd_fsa_stall(FALSE); - return; - } - } - - if (ret != pcmk_ok) { - crm_err("Failed to connect to the executor the max allowed %d time%s", - lrm_state->num_lrm_register_fails, - pcmk__plural_s(lrm_state->num_lrm_register_fails)); - register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); - return; - } - - controld_set_fsa_input_flags(R_LRM_CONNECTED); - crm_info("Connection to the executor established"); + try_local_executor_connect(action, msg_data, lrm_state); } if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __func__); } } static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level) { int counter = 0; gboolean rc = TRUE; const char *when = "lrm disconnect"; GHashTableIter gIter; const char *key = NULL; rsc_history_t *entry = NULL; active_op_t *pending = NULL; crm_debug("Checking for active resources before exit"); if (cur_state == S_TERMINATE) { log_level = LOG_ERR; when = "shutdown"; } else if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)) { when = "shutdown... waiting"; } if (lrm_state->pending_ops && lrm_state_is_connected(lrm_state) == TRUE) { guint removed = g_hash_table_foreach_remove( lrm_state->pending_ops, stop_recurring_actions, lrm_state); guint nremaining = g_hash_table_size(lrm_state->pending_ops); if (removed || nremaining) { crm_notice("Stopped %u recurring operation%s at %s (%u remaining)", removed, pcmk__plural_s(removed), when, nremaining); } } if (lrm_state->pending_ops) { g_hash_table_iter_init(&gIter, lrm_state->pending_ops); while (g_hash_table_iter_next(&gIter, NULL, (void **)&pending)) { /* Ignore recurring actions in the shutdown calculations */ if (pending->interval_ms == 0) { counter++; } } } if (counter > 0) { do_crm_log(log_level, "%d pending executor operation%s at %s", counter, pcmk__plural_s(counter), when); if ((cur_state == S_TERMINATE) || !pcmk_is_set(fsa_input_register, R_SENT_RSC_STOP)) { g_hash_table_iter_init(&gIter, lrm_state->pending_ops); while (g_hash_table_iter_next(&gIter, (gpointer*)&key, (gpointer*)&pending)) { do_crm_log(log_level, "Pending action: %s (%s)", key, pending->op_key); } } else { rc = FALSE; } return rc; } if (lrm_state->resource_history == NULL) { return rc; } if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)) { /* At this point we're not waiting, we're just shutting down */ when = "shutdown"; } counter = 0; g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (gpointer*)&entry)) { if (is_rsc_active(lrm_state, entry->id) == FALSE) { continue; } counter++; if (log_level == LOG_ERR) { crm_info("Found %s active at %s", entry->id, when); } else { crm_trace("Found %s active at %s", entry->id, when); } if (lrm_state->pending_ops) { GHashTableIter hIter; g_hash_table_iter_init(&hIter, lrm_state->pending_ops); while (g_hash_table_iter_next(&hIter, (gpointer*)&key, (gpointer*)&pending)) { if (pcmk__str_eq(entry->id, pending->rsc_id, pcmk__str_none)) { crm_notice("%sction %s (%s) incomplete at %s", pending->interval_ms == 0 ? "A" : "Recurring a", key, pending->op_key, when); } } } } if (counter) { crm_err("%d resource%s active at %s", counter, (counter == 1)? " was" : "s were", when); } return rc; } /*! * \internal * \brief Build XML and string of parameters meeting some criteria, for digest * * \param[in] op Executor event with parameter table to use * \param[in] metadata Parsed meta-data for executed resource agent * \param[in] param_type Flag used for selection criteria * \param[out] result Will be set to newly created XML with selected * parameters as attributes * * \return Newly allocated space-separated string of parameter names * \note Selection criteria varies by param_type: for the restart digest, we * want parameters that are *not* marked reloadable (OCF 1.1) or that * *are* marked unique (pre-1.1), for both string and XML results; for the * secure digest, we want parameters that *are* marked private for the * string, but parameters that are *not* marked private for the XML. * \note It is the caller's responsibility to free the string return value with * free() and the XML result with free_xml(). */ static char * build_parameter_list(const lrmd_event_data_t *op, const struct ra_metadata_s *metadata, enum ra_param_flags_e param_type, xmlNode **result) { char *list = NULL; size_t len = 0; *result = create_xml_node(NULL, XML_TAG_PARAMS); for (GList *iter = metadata->ra_params; iter != NULL; iter = iter->next) { struct ra_param_s *param = (struct ra_param_s *) iter->data; bool accept_for_list = false; bool accept_for_xml = false; switch (param_type) { case ra_param_reloadable: accept_for_list = !pcmk_is_set(param->rap_flags, param_type); accept_for_xml = accept_for_list; break; case ra_param_unique: accept_for_list = pcmk_is_set(param->rap_flags, param_type); accept_for_xml = accept_for_list; break; case ra_param_private: accept_for_list = pcmk_is_set(param->rap_flags, param_type); accept_for_xml = !accept_for_list; break; } if (accept_for_list) { crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type)); if (list == NULL) { // We will later search for " WORD ", so start list with a space pcmk__add_word(&list, &len, " "); } pcmk__add_word(&list, &len, param->rap_name); } else { crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type)); } if (accept_for_xml) { const char *v = g_hash_table_lookup(op->params, param->rap_name); if (v != NULL) { crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v); crm_xml_add(*result, param->rap_name, v); } } } if (list != NULL) { // We will later search for " WORD ", so end list with a space pcmk__add_word(&list, &len, " "); } return list; } static void append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { char *list = NULL; char *digest = NULL; xmlNode *restart = NULL; CRM_LOG_ASSERT(op->params != NULL); if (op->interval_ms > 0) { /* monitors are not reloadable */ return; } if (pcmk_is_set(metadata->ra_flags, ra_supports_reload_agent)) { // Add parameters not marked reloadable to the "op-force-restart" list list = build_parameter_list(op, metadata, ra_param_reloadable, &restart); } else if (pcmk_is_set(metadata->ra_flags, ra_supports_legacy_reload)) { /* @COMPAT pre-OCF-1.1 resource agents * * Before OCF 1.1, Pacemaker abused "unique=0" to indicate * reloadability. Add any parameters with unique="1" to the * "op-force-restart" list. */ list = build_parameter_list(op, metadata, ra_param_unique, &restart); } else { // Resource does not support agent reloads return; } digest = calculate_operation_digest(restart, version); /* Add "op-force-restart" and "op-restart-digest" to indicate the resource supports reload, * no matter if it actually supports any parameters with unique="1"). */ crm_xml_add(update, XML_LRM_ATTR_OP_RESTART, list? list: ""); crm_xml_add(update, XML_LRM_ATTR_RESTART_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, list); crm_log_xml_trace(restart, "restart digest source"); free_xml(restart); free(digest); free(list); } static void append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { char *list = NULL; char *digest = NULL; xmlNode *secure = NULL; CRM_LOG_ASSERT(op->params != NULL); /* * To keep XML_LRM_ATTR_OP_SECURE short, we want it to contain the * secure parameters but XML_LRM_ATTR_SECURE_DIGEST to be based on * the insecure ones */ list = build_parameter_list(op, metadata, ra_param_private, &secure); if (list != NULL) { digest = calculate_operation_digest(secure, version); crm_xml_add(update, XML_LRM_ATTR_OP_SECURE, list); crm_xml_add(update, XML_LRM_ATTR_SECURE_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, list); crm_log_xml_trace(secure, "secure digest source"); } else { crm_trace("%s: no secure parameters", op->rsc_id); } free_xml(secure); free(digest); free(list); } static gboolean build_operation_update(xmlNode * parent, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op, const char *node_name, const char *src) { int target_rc = 0; xmlNode *xml_op = NULL; struct ra_metadata_s *metadata = NULL; const char *caller_version = NULL; lrm_state_t *lrm_state = NULL; uint32_t metadata_source = controld_metadata_from_agent; if (op == NULL) { return FALSE; } target_rc = rsc_op_expected_rc(op); /* there is a small risk in formerly mixed clusters that it will * be sub-optimal. * * however with our upgrade policy, the update we send should * still be completely supported anyway */ caller_version = g_hash_table_lookup(op->params, XML_ATTR_CRM_VERSION); CRM_LOG_ASSERT(caller_version != NULL); if(caller_version == NULL) { caller_version = CRM_FEATURE_SET; } crm_trace("Building %s operation update with originator version: %s", op->rsc_id, caller_version); xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc, fsa_our_uname, src, LOG_DEBUG); if (xml_op == NULL) { return TRUE; } if ((rsc == NULL) || (op->params == NULL) || !crm_op_needs_metadata(rsc->standard, op->op_type)) { crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)", op->op_type, op->rsc_id, op->params, rsc); return TRUE; } lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { crm_warn("Cannot calculate digests for operation " PCMK__OP_FMT " because we have no connection to executor for %s", op->rsc_id, op->op_type, op->interval_ms, node_name); return TRUE; } /* Getting meta-data from cache is OK unless this is a successful start * action -- always refresh from the agent for those, in case the * resource agent was updated. * * @TODO Only refresh the meta-data after starts if the agent actually * changed (using something like inotify, or a hash or modification time of * the agent executable). */ if ((op->op_status != PCMK_LRM_OP_DONE) || (op->rc != target_rc) || !pcmk__str_eq(op->op_type, CRMD_ACTION_START, pcmk__str_none)) { metadata_source |= controld_metadata_from_cache; } metadata = controld_get_rsc_metadata(lrm_state, rsc, metadata_source); if (metadata == NULL) { return TRUE; } #if ENABLE_VERSIONED_ATTRS crm_xml_add(xml_op, XML_ATTR_RA_VERSION, metadata->ra_version); #endif crm_trace("Including additional digests for %s:%s:%s", rsc->standard, rsc->provider, rsc->type); append_restart_list(op, metadata, xml_op, caller_version); append_secure_list(op, metadata, xml_op, caller_version); return TRUE; } static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id) { rsc_history_t *entry = NULL; entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->last == NULL) { return FALSE; } crm_trace("Processing %s: %s.%d=%d", rsc_id, entry->last->op_type, entry->last->interval_ms, entry->last->rc); if (entry->last->rc == PCMK_OCF_OK && pcmk__str_eq(entry->last->op_type, CRMD_ACTION_STOP, pcmk__str_casei)) { return FALSE; } else if (entry->last->rc == PCMK_OCF_OK && pcmk__str_eq(entry->last->op_type, CRMD_ACTION_MIGRATE, pcmk__str_casei)) { // A stricter check is too complex ... leave that to the scheduler return FALSE; } else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) { return FALSE; } else if ((entry->last->interval_ms == 0) && (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) { /* Badly configured resources can't be reliably stopped */ return FALSE; } return TRUE; } static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list) { GHashTableIter iter; rsc_history_t *entry = NULL; g_hash_table_iter_init(&iter, lrm_state->resource_history); while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) { GList *gIter = NULL; xmlNode *xml_rsc = create_xml_node(rsc_list, XML_LRM_TAG_RESOURCE); crm_xml_add(xml_rsc, XML_ATTR_ID, entry->id); crm_xml_add(xml_rsc, XML_ATTR_TYPE, entry->rsc.type); crm_xml_add(xml_rsc, XML_AGENT_ATTR_CLASS, entry->rsc.standard); crm_xml_add(xml_rsc, XML_AGENT_ATTR_PROVIDER, entry->rsc.provider); if (entry->last && entry->last->params) { const char *container = g_hash_table_lookup(entry->last->params, CRM_META"_"XML_RSC_ATTR_CONTAINER); if (container) { crm_trace("Resource %s is a part of container resource %s", entry->id, container); crm_xml_add(xml_rsc, XML_RSC_ATTR_CONTAINER, container); } } build_operation_update(xml_rsc, &(entry->rsc), entry->failed, lrm_state->node_name, __func__); build_operation_update(xml_rsc, &(entry->rsc), entry->last, lrm_state->node_name, __func__); for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) { build_operation_update(xml_rsc, &(entry->rsc), gIter->data, lrm_state->node_name, __func__); } } return FALSE; } static xmlNode * do_lrm_query_internal(lrm_state_t *lrm_state, int update_flags) { xmlNode *xml_state = NULL; xmlNode *xml_data = NULL; xmlNode *rsc_list = NULL; crm_node_t *peer = NULL; peer = crm_get_peer_full(0, lrm_state->node_name, CRM_GET_PEER_ANY); CRM_CHECK(peer != NULL, return NULL); xml_state = create_node_state_update(peer, update_flags, NULL, __func__); if (xml_state == NULL) { return NULL; } xml_data = create_xml_node(xml_state, XML_CIB_TAG_LRM); crm_xml_add(xml_data, XML_ATTR_ID, peer->uuid); rsc_list = create_xml_node(xml_data, XML_LRM_TAG_RESOURCES); /* Build a list of active (not always running) resources */ build_active_RAs(lrm_state, rsc_list); crm_log_xml_trace(xml_state, "Current executor state"); return xml_state; } xmlNode * controld_query_executor_state(const char *node_name) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (!lrm_state) { crm_err("Could not find executor state for node %s", node_name); return NULL; } return do_lrm_query_internal(lrm_state, node_update_cluster|node_update_peer); } /*! * \internal * \brief Map standard Pacemaker return code to operation status and OCF code * * \param[out] event Executor event whose status and return code should be set * \param[in] rc Standard Pacemaker return code */ void controld_rc2event(lrmd_event_data_t *event, int rc) { switch (rc) { case pcmk_rc_ok: event->rc = PCMK_OCF_OK; event->op_status = PCMK_LRM_OP_DONE; break; case EACCES: event->rc = PCMK_OCF_INSUFFICIENT_PRIV; event->op_status = PCMK_LRM_OP_ERROR; break; default: event->rc = PCMK_OCF_UNKNOWN_ERROR; event->op_status = PCMK_LRM_OP_ERROR; break; } } /*! * \internal * \brief Trigger a new transition after CIB status was deleted * * If a CIB status delete was not expected (as part of the transition graph), * trigger a new transition by updating the (arbitrary) "last-lrm-refresh" * cluster property. * * \param[in] from_sys IPC name that requested the delete * \param[in] rsc_id Resource whose status was deleted (for logging only) */ void controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id) { if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) { char *now_s = crm_strdup_printf("%lld", (long long) time(NULL)); crm_debug("Triggering a refresh after %s cleaned %s", from_sys, rsc_id); update_attr_delegate(fsa_cib_conn, cib_none, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, "last-lrm-refresh", now_s, FALSE, NULL, NULL); free(now_s); } } static void notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc) { lrmd_event_data_t *op = NULL; const char *from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM); const char *from_host = crm_element_value(input->msg, F_CRM_HOST_FROM); crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "localhost"), rsc_id, ((rc == pcmk_ok)? "" : " not")); op = construct_op(lrm_state, input->xml, rsc_id, CRMD_ACTION_DELETE); controld_rc2event(op, pcmk_legacy2rc(rc)); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); controld_trigger_delete_refresh(from_sys, rsc_id); } static gboolean lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data) { struct delete_event_s *event = user_data; struct pending_deletion_op_s *op = value; if (pcmk__str_eq(event->rsc, op->rsc, pcmk__str_none)) { notify_deleted(event->lrm_state, op->input, event->rsc, event->rc); return TRUE; } return FALSE; } static gboolean lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data) { const char *rsc = user_data; active_op_t *pending = value; if (pcmk__str_eq(rsc, pending->rsc_id, pcmk__str_none)) { crm_info("Removing op %s:%d for deleted resource %s", pending->op_key, pending->call_id, rsc); return TRUE; } return FALSE; } static void delete_rsc_entry(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, GHashTableIter * rsc_gIter, int rc, const char *user_name) { struct delete_event_s event; CRM_CHECK(rsc_id != NULL, return); if (rc == pcmk_ok) { char *rsc_id_copy = strdup(rsc_id); if (rsc_gIter) { g_hash_table_iter_remove(rsc_gIter); } else { g_hash_table_remove(lrm_state->resource_history, rsc_id_copy); } controld_delete_resource_history(rsc_id_copy, lrm_state->node_name, user_name, crmd_cib_smart_opt()); g_hash_table_foreach_remove(lrm_state->pending_ops, lrm_remove_deleted_op, rsc_id_copy); free(rsc_id_copy); } if (input) { notify_deleted(lrm_state, input, rsc_id, rc); } event.rc = rc; event.rsc = rsc_id; event.lrm_state = lrm_state; g_hash_table_foreach_remove(lrm_state->deletion_ops, lrm_remove_deleted_rsc, &event); } /*! * \internal * \brief Erase an LRM history entry from the CIB, given the operation data * * \param[in] lrm_state LRM state of the desired node * \param[in] op Operation whose history should be deleted */ static void erase_lrm_history_by_op(lrm_state_t *lrm_state, lrmd_event_data_t *op) { xmlNode *xml_top = NULL; CRM_CHECK(op != NULL, return); xml_top = create_xml_node(NULL, XML_LRM_TAG_RSC_OP); crm_xml_add_int(xml_top, XML_LRM_ATTR_CALLID, op->call_id); crm_xml_add(xml_top, XML_ATTR_TRANSITION_KEY, op->user_data); if (op->interval_ms > 0) { char *op_id = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); /* Avoid deleting last_failure too (if it was a result of this recurring op failing) */ crm_xml_add(xml_top, XML_ATTR_ID, op_id); free(op_id); } crm_debug("Erasing resource operation history for " PCMK__OP_FMT " (call=%d)", op->rsc_id, op->op_type, op->interval_ms, op->call_id); fsa_cib_conn->cmds->remove(fsa_cib_conn, XML_CIB_TAG_STATUS, xml_top, cib_quorum_override); crm_log_xml_trace(xml_top, "op:cancel"); free_xml(xml_top); } /* Define xpath to find LRM resource history entry by node and resource */ #define XPATH_HISTORY \ "/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \ "/" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='%s']" \ "/" XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES \ "/" XML_LRM_TAG_RESOURCE "[@" XML_ATTR_ID "='%s']" \ "/" XML_LRM_TAG_RSC_OP /* ... and also by operation key */ #define XPATH_HISTORY_ID XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s']" /* ... and also by operation key and operation call ID */ #define XPATH_HISTORY_CALL XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_CALLID "='%d']" /* ... and also by operation key and original operation key */ #define XPATH_HISTORY_ORIG XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_TASK_KEY "='%s']" /*! * \internal * \brief Erase an LRM history entry from the CIB, given operation identifiers * * \param[in] lrm_state LRM state of the node to clear history for * \param[in] rsc_id Name of resource to clear history for * \param[in] key Operation key of operation to clear history for * \param[in] orig_op If specified, delete only if it has this original op * \param[in] call_id If specified, delete entry only if it has this call ID */ static void erase_lrm_history_by_id(lrm_state_t *lrm_state, const char *rsc_id, const char *key, const char *orig_op, int call_id) { char *op_xpath = NULL; CRM_CHECK((rsc_id != NULL) && (key != NULL), return); if (call_id > 0) { op_xpath = crm_strdup_printf(XPATH_HISTORY_CALL, lrm_state->node_name, rsc_id, key, call_id); } else if (orig_op) { op_xpath = crm_strdup_printf(XPATH_HISTORY_ORIG, lrm_state->node_name, rsc_id, key, orig_op); } else { op_xpath = crm_strdup_printf(XPATH_HISTORY_ID, lrm_state->node_name, rsc_id, key); } crm_debug("Erasing resource operation history for %s on %s (call=%d)", key, rsc_id, call_id); fsa_cib_conn->cmds->remove(fsa_cib_conn, op_xpath, NULL, cib_quorum_override | cib_xpath); free(op_xpath); } static inline gboolean last_failed_matches_op(rsc_history_t *entry, const char *op, guint interval_ms) { if (entry == NULL) { return FALSE; } if (op == NULL) { return TRUE; } return (pcmk__str_eq(op, entry->failed->op_type, pcmk__str_casei) && (interval_ms == entry->failed->interval_ms)); } /*! * \internal * \brief Clear a resource's last failure * * Erase a resource's last failure on a particular node from both the * LRM resource history in the CIB, and the resource history remembered * for the LRM state. * * \param[in] rsc_id Resource name * \param[in] node_name Node name * \param[in] operation If specified, only clear if matching this operation * \param[in] interval_ms If operation is specified, it has this interval */ void lrm_clear_last_failure(const char *rsc_id, const char *node_name, const char *operation, guint interval_ms) { char *op_key = NULL; char *orig_op_key = NULL; lrm_state_t *lrm_state = NULL; lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { return; } /* Erase from CIB */ op_key = pcmk__op_key(rsc_id, "last_failure", 0); if (operation) { orig_op_key = pcmk__op_key(rsc_id, operation, interval_ms); } erase_lrm_history_by_id(lrm_state, rsc_id, op_key, orig_op_key, 0); free(op_key); free(orig_op_key); /* Remove from memory */ if (lrm_state->resource_history) { rsc_history_t *entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (last_failed_matches_op(entry, operation, interval_ms)) { lrmd_free_event(entry->failed); entry->failed = NULL; } } } /* Returns: gboolean - cancellation is in progress */ static gboolean cancel_op(lrm_state_t * lrm_state, const char *rsc_id, const char *key, int op, gboolean remove) { int rc = pcmk_ok; char *local_key = NULL; active_op_t *pending = NULL; CRM_CHECK(op != 0, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); if (key == NULL) { local_key = make_stop_id(rsc_id, op); key = local_key; } pending = g_hash_table_lookup(lrm_state->pending_ops, key); if (pending) { if (remove && !pcmk_is_set(pending->flags, active_op_remove)) { controld_set_active_op_flags(pending, active_op_remove); crm_debug("Scheduling %s for removal", key); } if (pcmk_is_set(pending->flags, active_op_cancelled)) { crm_debug("Operation %s already cancelled", key); free(local_key); return FALSE; } controld_set_active_op_flags(pending, active_op_cancelled); } else { crm_info("No pending op found for %s", key); free(local_key); return FALSE; } crm_debug("Cancelling op %d for %s (%s)", op, rsc_id, key); rc = lrm_state_cancel(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms); if (rc == pcmk_ok) { crm_debug("Op %d for %s (%s): cancelled", op, rsc_id, key); free(local_key); return TRUE; } crm_debug("Op %d for %s (%s): Nothing to cancel", op, rsc_id, key); /* The caller needs to make sure the entry is * removed from the pending_ops list * * Usually by returning TRUE inside the worker function * supplied to g_hash_table_foreach_remove() * * Not removing the entry from pending_ops will block * the node from shutting down */ free(local_key); return FALSE; } struct cancel_data { gboolean done; gboolean remove; const char *key; lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean cancel_action_by_key(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct cancel_data *data = user_data; active_op_t *op = value; if (pcmk__str_eq(op->op_key, data->key, pcmk__str_none)) { data->done = TRUE; remove = !cancel_op(data->lrm_state, data->rsc->id, key, op->call_id, data->remove); } return remove; } static gboolean cancel_op_key(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *key, gboolean remove) { guint removed = 0; struct cancel_data data; CRM_CHECK(rsc != NULL, return FALSE); CRM_CHECK(key != NULL, return FALSE); data.key = key; data.rsc = rsc; data.done = FALSE; data.remove = remove; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove(lrm_state->pending_ops, cancel_action_by_key, &data); crm_trace("Removed %u op cache entries, new size: %u", removed, g_hash_table_size(lrm_state->pending_ops)); return data.done; } /*! * \internal * \brief Retrieve resource information from LRM * * \param[in] lrm_state LRM connection to use * \param[in] rsc_xml XML containing resource configuration * \param[in] do_create If true, register resource with LRM if not already * \param[out] rsc_info Where to store resource information obtained from LRM * * \retval pcmk_ok Success (and rsc_info holds newly allocated result) * \retval -EINVAL Required information is missing from arguments * \retval -ENOTCONN No active connection to LRM * \retval -ENODEV Resource not found * \retval -errno Error communicating with executor when registering resource * * \note Caller is responsible for freeing result on success. */ static int get_lrm_resource(lrm_state_t *lrm_state, xmlNode *rsc_xml, gboolean do_create, lrmd_rsc_info_t **rsc_info) { const char *id = ID(rsc_xml); CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL); CRM_CHECK(id, return -EINVAL); if (lrm_state_is_connected(lrm_state) == FALSE) { return -ENOTCONN; } crm_trace("Retrieving resource information for %s from the executor", id); *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); // If resource isn't known by ID, try clone name, if provided if (!*rsc_info) { const char *long_id = crm_element_value(rsc_xml, XML_ATTR_ID_LONG); if (long_id) { *rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0); } } if ((*rsc_info == NULL) && do_create) { const char *class = crm_element_value(rsc_xml, XML_AGENT_ATTR_CLASS); const char *provider = crm_element_value(rsc_xml, XML_AGENT_ATTR_PROVIDER); const char *type = crm_element_value(rsc_xml, XML_ATTR_TYPE); int rc; crm_trace("Registering resource %s with the executor", id); rc = lrm_state_register_rsc(lrm_state, id, class, provider, type, lrmd_opt_drop_recurring); if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Could not register resource %s with the executor on %s: %s " CRM_XS " rc=%d", id, lrm_state->node_name, pcmk_strerror(rc), rc); /* Register this as an internal error if this involves the local * executor. Otherwise, we're likely dealing with an unresponsive * remote node, which is not an FSA failure. */ if (lrm_state_is_local(lrm_state) == TRUE) { register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } return rc; } *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); } return *rsc_info? pcmk_ok : -ENODEV; } static void delete_resource(lrm_state_t * lrm_state, const char *id, lrmd_rsc_info_t * rsc, GHashTableIter * gIter, const char *sys, const char *user, ha_msg_input_t * request, gboolean unregister) { int rc = pcmk_ok; crm_info("Removing resource %s from executor for %s%s%s", id, sys, (user? " as " : ""), (user? user : "")); if (rsc && unregister) { rc = lrm_state_unregister_rsc(lrm_state, id, 0); } if (rc == pcmk_ok) { crm_trace("Resource %s deleted from executor", id); } else if (rc == -EINPROGRESS) { crm_info("Deletion of resource '%s' from executor is pending", id); if (request) { struct pending_deletion_op_s *op = NULL; char *ref = crm_element_value_copy(request->msg, XML_ATTR_REFERENCE); op = calloc(1, sizeof(struct pending_deletion_op_s)); op->rsc = strdup(rsc->id); op->input = copy_ha_msg_input(request); g_hash_table_insert(lrm_state->deletion_ops, ref, op); } return; } else { crm_warn("Could not delete '%s' from executor for %s%s%s: %s " CRM_XS " rc=%d", id, sys, (user? " as " : ""), (user? user : ""), pcmk_strerror(rc), rc); } delete_rsc_entry(lrm_state, request, id, gIter, rc, user); } static int get_fake_call_id(lrm_state_t *lrm_state, const char *rsc_id) { int call_id = 999999999; rsc_history_t *entry = NULL; if(lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* Make sure the call id is greater than the last successful operation, * otherwise the failure will not result in a possible recovery of the resource * as it could appear the failure occurred before the successful start */ if (entry) { call_id = entry->last_callid + 1; } if (call_id < 0) { call_id = 1; } return call_id; } static void fake_op_status(lrm_state_t *lrm_state, lrmd_event_data_t *op, int op_status, enum ocf_exitcode op_exitcode) { op->call_id = get_fake_call_id(lrm_state, op->rsc_id); op->t_run = time(NULL); op->t_rcchange = op->t_run; op->op_status = op_status; op->rc = op_exitcode; } static void force_reprobe(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node) { GHashTableIter gIter; rsc_history_t *entry = NULL; crm_info("Clearing resource history on node %s", lrm_state->node_name); g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { /* only unregister the resource during a reprobe if it is not a remote connection * resource. otherwise unregistering the connection will terminate remote-node * membership */ gboolean unregister = TRUE; if (is_remote_lrmd_ra(NULL, NULL, entry->id)) { lrm_state_t *remote_lrm_state = lrm_state_find(entry->id); if (remote_lrm_state) { /* when forcing a reprobe, make sure to clear remote node before * clearing the remote node's connection resource */ force_reprobe(remote_lrm_state, from_sys, from_host, user_name, TRUE); } unregister = FALSE; } delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys, user_name, NULL, unregister); } /* Now delete the copy in the CIB */ controld_delete_node_state(lrm_state->node_name, controld_section_lrm, cib_scope_local); /* Finally, _delete_ the value in pacemaker-attrd -- setting it to FALSE * would result in the scheduler sending us back here again */ update_attrd(lrm_state->node_name, CRM_OP_PROBED, NULL, user_name, is_remote_node); } /*! * \internal * \brief Fail a requested action without actually executing it * * For an action that can't be executed, process it similarly to an actual * execution result, with specified error status (except for notify actions, * which will always be treated as successful). * * \param[in] lrm_state Executor connection that action is for * \param[in] action Action XML from request * \param[in] rc Desired return code to use * \param[in] op_status Desired operation status to use */ static void synthesize_lrmd_failure(lrm_state_t *lrm_state, xmlNode *action, int op_status, enum ocf_exitcode rc) { lrmd_event_data_t *op = NULL; const char *operation = crm_element_value(action, XML_LRM_ATTR_TASK); const char *target_node = crm_element_value(action, XML_LRM_ATTR_TARGET); xmlNode *xml_rsc = find_xml_node(action, XML_CIB_TAG_RESOURCE, TRUE); if ((xml_rsc == NULL) || (ID(xml_rsc) == NULL)) { /* @TODO Should we do something else, like direct ack? */ crm_info("Can't fake %s failure (%d) on %s without resource configuration", crm_element_value(action, XML_LRM_ATTR_TASK_KEY), rc, target_node); return; } else if(operation == NULL) { /* This probably came from crm_resource -C, nothing to do */ crm_info("Can't fake %s failure (%d) on %s without operation", ID(xml_rsc), rc, target_node); return; } op = construct_op(lrm_state, action, ID(xml_rsc), operation); if (pcmk__str_eq(operation, RSC_NOTIFY, pcmk__str_casei)) { // Notifications can't fail fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_OK); } else { fake_op_status(lrm_state, op, op_status, rc); } crm_info("Faking " PCMK__OP_FMT " result (%d) on %s", op->rsc_id, op->op_type, op->interval_ms, op->rc, target_node); // Process the result as if it came from the LRM process_lrm_event(lrm_state, op, NULL, action); lrmd_free_event(op); } /*! * \internal * \brief Get target of an LRM operation * * \param[in] xml LRM operation data XML * * \return LRM operation target node name (local node or Pacemaker Remote node) */ static const char * lrm_op_target(xmlNode *xml) { const char *target = NULL; if (xml) { target = crm_element_value(xml, XML_LRM_ATTR_TARGET); } if (target == NULL) { target = fsa_our_uname; } return target; } static void fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { lrmd_event_data_t *op = NULL; lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = find_xml_node(xml, XML_CIB_TAG_RESOURCE, TRUE); CRM_CHECK(xml_rsc != NULL, return); /* The executor simply executes operations and reports the results, without * any concept of success or failure, so to fail a resource, we must fake * what a failure looks like. * * To do this, we create a fake executor operation event for the resource, * and pass that event to the executor client callback so it will be * processed as if it came from the executor. */ op = construct_op(lrm_state, xml, ID(xml_rsc), "asyncmon"); fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_UNKNOWN_ERROR); free((char*) op->user_data); op->user_data = NULL; op->interval_ms = 0; if (user_name && !pcmk__is_privileged(user_name)) { crm_err("%s does not have permission to fail %s", user_name, ID(xml_rsc)); controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc)); lrmd_free_event(op); return; } if (get_lrm_resource(lrm_state, xml_rsc, TRUE, &rsc) == pcmk_ok) { crm_info("Failing resource %s...", rsc->id); op->exit_reason = strdup("Simulated failure"); process_lrm_event(lrm_state, op, NULL, xml); op->op_status = PCMK_LRM_OP_DONE; op->rc = PCMK_OCF_OK; lrmd_free_rsc_info(rsc); } else { crm_info("Cannot find/create resource in order to fail it..."); crm_log_xml_warn(xml, "bad input"); } controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc)); lrmd_free_event(op); } static void handle_refresh_op(lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { int rc = pcmk_ok; xmlNode *fragment = do_lrm_query_internal(lrm_state, node_update_all); fsa_cib_update(XML_CIB_TAG_STATUS, fragment, cib_quorum_override, rc, user_name); crm_info("Forced a local resource history refresh: call=%d", rc); if (!pcmk__str_eq(CRM_SYSTEM_CRMD, from_sys, pcmk__str_casei)) { xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, fragment, from_host, from_sys, CRM_SYSTEM_LRMD, fsa_our_uuid); crm_debug("ACK'ing refresh from %s (%s)", from_sys, from_host); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(reply); } free_xml(fragment); } static void handle_query_op(xmlNode *msg, lrm_state_t *lrm_state) { xmlNode *data = do_lrm_query_internal(lrm_state, node_update_all); xmlNode *reply = create_reply(msg, data); if (relay_message(reply, TRUE) == FALSE) { crm_err("Unable to route reply"); crm_log_xml_err(reply, "reply"); } free_xml(reply); free_xml(data); } static void handle_reprobe_op(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node) { crm_notice("Forcing the status of all resources to be redetected"); force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node); if (!pcmk__strcase_any_of(from_sys, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) { xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, NULL, from_host, from_sys, CRM_SYSTEM_LRMD, fsa_our_uuid); crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(reply); } } static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys) { char *op_key = NULL; char *meta_key = NULL; int call = 0; const char *call_id = NULL; const char *op_task = NULL; guint interval_ms = 0; gboolean in_progress = FALSE; xmlNode *params = find_xml_node(input->xml, XML_TAG_ATTRS, TRUE); CRM_CHECK(params != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_TASK); op_task = crm_element_value(params, meta_key); free(meta_key); CRM_CHECK(op_task != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_INTERVAL_MS); if (crm_element_value_ms(params, meta_key, &interval_ms) != pcmk_ok) { free(meta_key); return FALSE; } free(meta_key); op_key = pcmk__op_key(rsc->id, op_task, interval_ms); meta_key = crm_meta_name(XML_LRM_ATTR_CALLID); call_id = crm_element_value(params, meta_key); free(meta_key); crm_debug("Scheduler requested op %s (call=%s) be cancelled", op_key, (call_id? call_id : "NA")); pcmk__scan_min_int(call_id, &call, 0); if (call == 0) { // Normal case when the scheduler cancels a recurring op in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE); } else { // Normal case when the scheduler cancels an orphan op in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE); } // Acknowledge cancellation operation if for a remote connection resource if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) { char *op_id = make_stop_id(rsc->id, call); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) { crm_info("Nothing known about operation %d for %s", call, op_key); } erase_lrm_history_by_id(lrm_state, rsc->id, op_key, NULL, call); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); /* needed at least for cancellation of a remote operation */ g_hash_table_remove(lrm_state->pending_ops, op_id); free(op_id); } else { /* No ack is needed since abcdaa8, but peers with older versions * in a rolling upgrade need one. We didn't bump the feature set * at that commit, so we can only compare against the previous * CRM version (3.0.8). If any peers have feature set 3.0.9 but * not abcdaa8, they will time out waiting for the ack (no * released versions of Pacemaker are affected). */ const char *peer_version = crm_element_value(params, XML_ATTR_CRM_VERSION); if (compare_version(peer_version, "3.0.8") <= 0) { crm_info("Sending compatibility ack for %s cancellation to %s (CRM version %s)", op_key, from_host, peer_version); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); } } free(op_key); return TRUE; } static void do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host, bool crm_rsc_delete, const char *user_name) { gboolean unregister = TRUE; int cib_rc = controld_delete_resource_history(rsc->id, lrm_state->node_name, user_name, cib_dryrun|cib_sync_call); if (cib_rc != pcmk_rc_ok) { lrmd_event_data_t *op = NULL; op = construct_op(lrm_state, input->xml, rsc->id, CRMD_ACTION_DELETE); op->op_status = PCMK_LRM_OP_ERROR; if (cib_rc == EACCES) { op->rc = PCMK_OCF_INSUFFICIENT_PRIV; } else { op->rc = PCMK_OCF_UNKNOWN_ERROR; } controld_ack_event_directly(from_host, from_sys, NULL, op, rsc->id); lrmd_free_event(op); return; } if (crm_rsc_delete && is_remote_lrmd_ra(NULL, NULL, rsc->id)) { unregister = FALSE; } delete_resource(lrm_state, rsc->id, rsc, NULL, from_sys, user_name, input, unregister); } /* A_LRM_INVOKE */ void do_lrm_invoke(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { lrm_state_t *lrm_state = NULL; const char *crm_op = NULL; const char *from_sys = NULL; const char *from_host = NULL; const char *operation = NULL; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *user_name = NULL; const char *target_node = NULL; gboolean is_remote_node = FALSE; bool crm_rsc_delete = FALSE; target_node = lrm_op_target(input->xml); is_remote_node = !pcmk__str_eq(target_node, fsa_our_uname, pcmk__str_casei); lrm_state = lrm_state_find(target_node); if ((lrm_state == NULL) && is_remote_node) { crm_err("Failing action because local node has never had connection to remote node %s", target_node); synthesize_lrmd_failure(NULL, input->xml, PCMK_LRM_OP_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR); return; } CRM_ASSERT(lrm_state != NULL); user_name = pcmk__update_acl_user(input->msg, F_CRM_USER, NULL); crm_op = crm_element_value(input->msg, F_CRM_TASK); from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM); if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) { from_host = crm_element_value(input->msg, F_CRM_HOST_FROM); } crm_trace("Executor %s command from %s as user %s", crm_op, from_sys, user_name); if (pcmk__str_eq(crm_op, CRM_OP_LRM_DELETE, pcmk__str_casei)) { if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) { crm_rsc_delete = TRUE; // from crm_resource } operation = CRMD_ACTION_DELETE; } else if (pcmk__str_eq(crm_op, CRM_OP_LRM_FAIL, pcmk__str_casei)) { fail_lrm_resource(input->xml, lrm_state, user_name, from_host, from_sys); return; } else if (input->xml != NULL) { operation = crm_element_value(input->xml, XML_LRM_ATTR_TASK); } if (pcmk__str_eq(crm_op, CRM_OP_LRM_REFRESH, pcmk__str_casei)) { handle_refresh_op(lrm_state, user_name, from_host, from_sys); } else if (pcmk__str_eq(crm_op, CRM_OP_LRM_QUERY, pcmk__str_casei)) { handle_query_op(input->msg, lrm_state); } else if (pcmk__str_eq(operation, CRM_OP_PROBED, pcmk__str_casei)) { update_attrd(lrm_state->node_name, CRM_OP_PROBED, XML_BOOLEAN_TRUE, user_name, is_remote_node); } else if (pcmk__str_eq(crm_op, CRM_OP_REPROBE, pcmk__str_casei) || pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_casei)) { handle_reprobe_op(lrm_state, from_sys, from_host, user_name, is_remote_node); } else if (operation != NULL) { lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = find_xml_node(input->xml, XML_CIB_TAG_RESOURCE, TRUE); gboolean create_rsc = !pcmk__str_eq(operation, CRMD_ACTION_DELETE, pcmk__str_casei); int rc; // We can't return anything meaningful without a resource ID CRM_CHECK(xml_rsc && ID(xml_rsc), return); rc = get_lrm_resource(lrm_state, xml_rsc, create_rsc, &rsc); if (rc == -ENOTCONN) { synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR); return; } else if ((rc < 0) && !create_rsc) { /* Delete of malformed or nonexistent resource * (deleting something that does not exist is a success) */ crm_notice("Not registering resource '%s' for a %s event " CRM_XS " get-rc=%d (%s) transition-key=%s", ID(xml_rsc), operation, rc, pcmk_strerror(rc), ID(input->xml)); delete_rsc_entry(lrm_state, input, ID(xml_rsc), NULL, pcmk_ok, user_name); return; } else if (rc == -EINVAL) { // Resource operation on malformed resource crm_err("Invalid resource definition for %s", ID(xml_rsc)); crm_log_xml_warn(input->msg, "invalid resource"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_ERROR, PCMK_OCF_NOT_CONFIGURED); // fatal error return; } else if (rc < 0) { // Error communicating with the executor crm_err("Could not register resource '%s' with executor: %s " CRM_XS " rc=%d", ID(xml_rsc), pcmk_strerror(rc), rc); crm_log_xml_warn(input->msg, "failed registration"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_ERROR, PCMK_OCF_INVALID_PARAM); // hard error return; } if (pcmk__str_eq(operation, CRMD_ACTION_CANCEL, pcmk__str_casei)) { if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) { crm_log_xml_warn(input->xml, "Bad command"); } } else if (pcmk__str_eq(operation, CRMD_ACTION_DELETE, pcmk__str_casei)) { do_lrm_delete(input, lrm_state, rsc, from_sys, from_host, crm_rsc_delete, user_name); } else if (pcmk__str_any_of(operation, CRMD_ACTION_RELOAD, CRMD_ACTION_RELOAD_AGENT, NULL)) { /* Pre-2.1.0 DCs will schedule reload actions only, and 2.1.0+ DCs * will schedule reload-agent actions only. In either case, we need * to map that to whatever the resource agent actually supports. * Default to the OCF 1.1 name. */ struct ra_metadata_s *md = NULL; const char *reload_name = CRMD_ACTION_RELOAD_AGENT; md = controld_get_rsc_metadata(lrm_state, rsc, controld_metadata_from_cache); if ((md != NULL) && pcmk_is_set(md->ra_flags, ra_supports_legacy_reload)) { reload_name = CRMD_ACTION_RELOAD; } do_lrm_rsc_op(lrm_state, rsc, reload_name, input->xml); } else { do_lrm_rsc_op(lrm_state, rsc, operation, input->xml); } lrmd_free_rsc_info(rsc); } else { crm_err("Cannot perform operation %s of unknown type", crm_str(crm_op)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } #if ENABLE_VERSIONED_ATTRS static void resolve_versioned_parameters(lrm_state_t *lrm_state, const char *rsc_id, const xmlNode *rsc_op, GHashTable *params) { /* Resource info *should* already be cached, so we don't get * executor call */ lrmd_rsc_info_t *rsc = lrm_state_get_rsc_info(lrm_state, rsc_id, 0); struct ra_metadata_s *metadata; metadata = controld_get_rsc_metadata(lrm_state, rsc, controld_metadata_from_cache); if (metadata) { xmlNode *versioned_attrs = NULL; GHashTable *hash = NULL; char *key = NULL; char *value = NULL; GHashTableIter iter; versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_ATTRS); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_iter_steal(&iter); g_hash_table_replace(params, key, value); } g_hash_table_destroy(hash); versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_META); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_replace(params, crm_meta_name(key), strdup(value)); if (pcmk__str_eq(key, XML_ATTR_TIMEOUT, pcmk__str_casei)) { pcmk__scan_min_int(value, &op->timeout, 0); } else if (pcmk__str_eq(key, XML_OP_ATTR_START_DELAY, pcmk__str_casei)) { pcmk__scan_min_int(value, &op->start_delay, 0); } } g_hash_table_destroy(hash); versioned_attrs = first_named_child(rsc_op, XML_TAG_RSC_VER_ATTRS); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_iter_steal(&iter); g_hash_table_replace(params, key, value); } g_hash_table_destroy(hash); } lrmd_free_rsc_info(rsc); } #endif static lrmd_event_data_t * construct_op(lrm_state_t *lrm_state, xmlNode *rsc_op, const char *rsc_id, const char *operation) { lrmd_event_data_t *op = NULL; const char *op_delay = NULL; const char *op_timeout = NULL; GHashTable *params = NULL; xmlNode *primitive = NULL; const char *class = NULL; const char *transition = NULL; CRM_ASSERT(rsc_id && operation); op = lrmd_new_event(rsc_id, operation, 0); op->type = lrmd_event_exec_complete; op->op_status = PCMK_LRM_OP_PENDING; op->rc = -1; op->timeout = 0; op->start_delay = 0; if (rsc_op == NULL) { CRM_LOG_ASSERT(pcmk__str_eq(CRMD_ACTION_STOP, operation, pcmk__str_casei)); op->user_data = NULL; /* the stop_all_resources() case * by definition there is no DC (or they'd be shutting * us down). * So we should put our version here. */ op->params = pcmk__strkey_table(free, free); g_hash_table_insert(op->params, strdup(XML_ATTR_CRM_VERSION), strdup(CRM_FEATURE_SET)); crm_trace("Constructed %s op for %s", operation, rsc_id); return op; } params = xml2list(rsc_op); g_hash_table_remove(params, CRM_META "_op_target_rc"); op_delay = crm_meta_value(params, XML_OP_ATTR_START_DELAY); pcmk__scan_min_int(op_delay, &op->start_delay, 0); op_timeout = crm_meta_value(params, XML_ATTR_TIMEOUT); pcmk__scan_min_int(op_timeout, &op->timeout, 0); if (pcmk__guint_from_hash(params, CRM_META "_" XML_LRM_ATTR_INTERVAL_MS, 0, &(op->interval_ms)) != pcmk_rc_ok) { op->interval_ms = 0; } /* Use pcmk_monitor_timeout instead of meta timeout for stonith recurring monitor, if set */ primitive = find_xml_node(rsc_op, XML_CIB_TAG_RESOURCE, FALSE); class = crm_element_value(primitive, XML_AGENT_ATTR_CLASS); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_fence_params) && pcmk__str_eq(operation, CRMD_ACTION_STATUS, pcmk__str_casei) && (op->interval_ms > 0)) { op_timeout = g_hash_table_lookup(params, "pcmk_monitor_timeout"); if (op_timeout != NULL) { op->timeout = crm_get_msec(op_timeout); } } #if ENABLE_VERSIONED_ATTRS if (lrm_state && !is_remote_lrmd_ra(NULL, NULL, rsc_id) && !pcmk__strcase_any_of(op_type, CRMD_ACTION_METADATA, CRMD_ACTION_DELETE, NULL)) { resolve_versioned_parameters(lrm_state, rsc_id, rsc_op, params); } #endif if (!pcmk__str_eq(operation, RSC_STOP, pcmk__str_casei)) { op->params = params; } else { rsc_history_t *entry = NULL; if (lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* If we do not have stop parameters cached, use * whatever we are given */ if (!entry || !entry->stop_params) { op->params = params; } else { /* Copy the cached parameter list so that we stop the resource * with the old attributes, not the new ones */ op->params = pcmk__strkey_table(free, free); g_hash_table_foreach(params, copy_meta_keys, op->params); g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params); g_hash_table_destroy(params); params = NULL; } } /* sanity */ if (op->timeout <= 0) { op->timeout = op->interval_ms; } if (op->start_delay < 0) { op->start_delay = 0; } transition = crm_element_value(rsc_op, XML_ATTR_TRANSITION_KEY); CRM_CHECK(transition != NULL, return op); op->user_data = strdup(transition); if (op->interval_ms != 0) { if (pcmk__strcase_any_of(operation, CRMD_ACTION_START, CRMD_ACTION_STOP, NULL)) { crm_err("Start and Stop actions cannot have an interval: %u", op->interval_ms); op->interval_ms = 0; } } crm_trace("Constructed %s op for %s: interval=%u", operation, rsc_id, op->interval_ms); return op; } /*! * \internal * \brief Send a (synthesized) event result * * Reply with a synthesized event result directly, as opposed to going through * the executor. * * \param[in] to_host Host to send result to * \param[in] to_sys IPC name to send result to (NULL for transition engine) * \param[in] rsc Type information about resource the result is for * \param[in] op Event with result to send * \param[in] rsc_id ID of resource the result is for */ void controld_ack_event_directly(const char *to_host, const char *to_sys, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *rsc_id) { xmlNode *reply = NULL; xmlNode *update, *iter; crm_node_t *peer = NULL; CRM_CHECK(op != NULL, return); if (op->rsc_id == NULL) { CRM_ASSERT(rsc_id != NULL); op->rsc_id = strdup(rsc_id); } if (to_sys == NULL) { to_sys = CRM_SYSTEM_TENGINE; } peer = crm_get_peer(0, fsa_our_uname); update = create_node_state_update(peer, node_update_none, NULL, __func__); iter = create_xml_node(update, XML_CIB_TAG_LRM); crm_xml_add(iter, XML_ATTR_ID, fsa_our_uuid); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE); crm_xml_add(iter, XML_ATTR_ID, op->rsc_id); build_operation_update(iter, rsc, op, fsa_our_uname, __func__); reply = create_request(CRM_OP_INVOKE_LRM, update, to_host, to_sys, CRM_SYSTEM_LRMD, NULL); crm_log_xml_trace(update, "[direct ACK]"); crm_debug("ACK'ing resource op " PCMK__OP_FMT " from %s: %s", op->rsc_id, op->op_type, op->interval_ms, op->user_data, crm_element_value(reply, XML_ATTR_REFERENCE)); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(update); free_xml(reply); } gboolean verify_stopped(enum crmd_fsa_state cur_state, int log_level) { gboolean res = TRUE; GList *lrm_state_list = lrm_state_get_list(); GList *state_entry; for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) { lrm_state_t *lrm_state = state_entry->data; if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) { /* keep iterating through all even when false is returned */ res = FALSE; } } controld_set_fsa_input_flags(R_SENT_RSC_STOP); g_list_free(lrm_state_list); lrm_state_list = NULL; return res; } struct stop_recurring_action_s { lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct stop_recurring_action_s *event = user_data; active_op_t *op = value; if ((op->interval_ms != 0) && pcmk__str_eq(op->rsc_id, event->rsc->id, pcmk__str_none)) { crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key); remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE); } return remove; } static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; lrm_state_t *lrm_state = user_data; active_op_t *op = value; if (op->interval_ms != 0) { crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (const char *) key); remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE); } return remove; } static void record_pending_op(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op) { const char *record_pending = NULL; CRM_CHECK(node_name != NULL, return); CRM_CHECK(rsc != NULL, return); CRM_CHECK(op != NULL, return); // Never record certain operation types as pending if ((op->op_type == NULL) || (op->params == NULL) || !controld_action_is_recordable(op->op_type)) { return; } // defaults to true record_pending = crm_meta_value(op->params, XML_OP_ATTR_PENDING); if (record_pending && !crm_is_true(record_pending)) { return; } op->call_id = -1; op->op_status = PCMK_LRM_OP_PENDING; op->rc = PCMK_OCF_UNKNOWN; op->t_run = time(NULL); op->t_rcchange = op->t_run; /* write a "pending" entry to the CIB, inhibit notification */ crm_debug("Recording pending op " PCMK__OP_FMT " on %s in the CIB", op->rsc_id, op->op_type, op->interval_ms, node_name); do_update_resource(node_name, rsc, op, 0); } static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *operation, xmlNode *msg) { int call_id = 0; char *op_id = NULL; lrmd_event_data_t *op = NULL; lrmd_key_value_t *params = NULL; fsa_data_t *msg_data = NULL; const char *transition = NULL; gboolean stop_recurring = FALSE; bool send_nack = FALSE; CRM_CHECK(rsc != NULL, return); CRM_CHECK(operation != NULL, return); if (msg != NULL) { transition = crm_element_value(msg, XML_ATTR_TRANSITION_KEY); if (transition == NULL) { crm_log_xml_err(msg, "Missing transition number"); } } op = construct_op(lrm_state, msg, rsc->id, operation); CRM_CHECK(op != NULL, return); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) && (op->interval_ms == 0) && strcmp(operation, CRMD_ACTION_MIGRATE) == 0) { /* pcmk remote connections are a special use case. * We never ever want to stop monitoring a connection resource until * the entire migration has completed. If the connection is unexpectedly * severed, even during a migration, this is an event we must detect.*/ stop_recurring = FALSE; } else if ((op->interval_ms == 0) && strcmp(operation, CRMD_ACTION_STATUS) != 0 && strcmp(operation, CRMD_ACTION_NOTIFY) != 0) { /* stop any previous monitor operations before changing the resource state */ stop_recurring = TRUE; } if (stop_recurring == TRUE) { guint removed = 0; struct stop_recurring_action_s data; data.rsc = rsc; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove( lrm_state->pending_ops, stop_recurring_action_by_rsc, &data); if (removed) { crm_debug("Stopped %u recurring operation%s in preparation for " PCMK__OP_FMT, removed, pcmk__plural_s(removed), rsc->id, operation, op->interval_ms); } } /* now do the op */ crm_notice("Requesting local execution of %s operation for %s on %s " CRM_XS " transition_key=%s op_key=" PCMK__OP_FMT, crm_action_str(op->op_type, op->interval_ms), rsc->id, lrm_state->node_name, transition, rsc->id, operation, op->interval_ms); if (pcmk_is_set(fsa_input_register, R_SHUTDOWN) && pcmk__str_eq(operation, RSC_START, pcmk__str_casei)) { register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); send_nack = TRUE; } else if (fsa_state != S_NOT_DC && fsa_state != S_POLICY_ENGINE /* Recalculating */ && fsa_state != S_TRANSITION_ENGINE && !pcmk__str_eq(operation, CRMD_ACTION_STOP, pcmk__str_casei)) { send_nack = TRUE; } if(send_nack) { crm_notice("Discarding attempt to perform action %s on %s in state %s (shutdown=%s)", operation, rsc->id, fsa_state2string(fsa_state), pcmk__btoa(pcmk_is_set(fsa_input_register, R_SHUTDOWN))); op->rc = PCMK_OCF_UNKNOWN_ERROR; op->op_status = PCMK_LRM_OP_INVALID; controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id); lrmd_free_event(op); free(op_id); return; } record_pending_op(lrm_state->node_name, rsc, op); op_id = pcmk__op_key(rsc->id, op->op_type, op->interval_ms); if (op->interval_ms > 0) { /* cancel it so we can then restart it without conflict */ cancel_op_key(lrm_state, rsc, op_id, FALSE); } if (op->params) { char *key = NULL; char *value = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, op->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { params = lrmd_key_value_add(params, key, value); } } call_id = lrm_state_exec(lrm_state, rsc->id, op->op_type, op->user_data, op->interval_ms, op->timeout, op->start_delay, params); if (call_id <= 0 && lrm_state_is_local(lrm_state)) { crm_err("Operation %s on %s failed: %d", operation, rsc->id, call_id); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } else if (call_id <= 0) { crm_err("Operation %s on resource %s failed to execute on remote node %s: %d", operation, rsc->id, lrm_state->node_name, call_id); fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_UNKNOWN_ERROR); process_lrm_event(lrm_state, op, NULL, NULL); } else { /* record all operations so we can wait * for them to complete during shutdown */ char *call_id_s = make_stop_id(rsc->id, call_id); active_op_t *pending = NULL; pending = calloc(1, sizeof(active_op_t)); crm_trace("Recording pending op: %d - %s %s", call_id, op_id, call_id_s); pending->call_id = call_id; pending->interval_ms = op->interval_ms; pending->op_type = strdup(operation); pending->op_key = strdup(op_id); pending->rsc_id = strdup(rsc->id); pending->start_time = time(NULL); pending->user_data = op->user_data? strdup(op->user_data) : NULL; if (crm_element_value_epoch(msg, XML_CONFIG_ATTR_SHUTDOWN_LOCK, &(pending->lock_time)) != pcmk_ok) { pending->lock_time = 0; } g_hash_table_replace(lrm_state->pending_ops, call_id_s, pending); if ((op->interval_ms > 0) && (op->start_delay > START_DELAY_THRESHOLD)) { int target_rc = 0; crm_info("Faking confirmation of %s: execution postponed for over 5 minutes", op_id); decode_transition_key(op->user_data, NULL, NULL, NULL, &target_rc); op->rc = target_rc; op->op_status = PCMK_LRM_OP_DONE; controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id); } pending->params = op->params; op->params = NULL; } free(op_id); lrmd_free_event(op); return; } int last_resource_update = 0; static void cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { switch (rc) { case pcmk_ok: case -pcmk_err_diff_failed: case -pcmk_err_diff_resync: crm_trace("Resource update %d complete: rc=%d", call_id, rc); break; default: crm_warn("Resource update %d failed: (rc=%d) %s", call_id, rc, pcmk_strerror(rc)); } if (call_id == last_resource_update) { last_resource_update = 0; trigger_fsa(); } } /* Only successful stops, and probes that found the resource inactive, get locks * recorded in the history. This ensures the resource stays locked to the node * until it is active there again after the node comes back up. */ static bool should_preserve_lock(lrmd_event_data_t *op) { if (!controld_shutdown_lock_enabled) { return false; } if (!strcmp(op->op_type, RSC_STOP) && (op->rc == PCMK_OCF_OK)) { return true; } if (!strcmp(op->op_type, RSC_STATUS) && (op->rc == PCMK_OCF_NOT_RUNNING)) { return true; } return false; } static int do_update_resource(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, time_t lock_time) { /* */ int rc = pcmk_ok; xmlNode *update, *iter = NULL; int call_opt = crmd_cib_smart_opt(); const char *uuid = NULL; CRM_CHECK(op != NULL, return 0); iter = create_xml_node(iter, XML_CIB_TAG_STATUS); update = iter; iter = create_xml_node(iter, XML_CIB_TAG_STATE); if (pcmk__str_eq(node_name, fsa_our_uname, pcmk__str_casei)) { uuid = fsa_our_uuid; } else { /* remote nodes uuid and uname are equal */ uuid = node_name; crm_xml_add(iter, XML_NODE_IS_REMOTE, "true"); } CRM_LOG_ASSERT(uuid != NULL); if(uuid == NULL) { rc = -EINVAL; goto done; } crm_xml_add(iter, XML_ATTR_UUID, uuid); crm_xml_add(iter, XML_ATTR_UNAME, node_name); crm_xml_add(iter, XML_ATTR_ORIGIN, __func__); iter = create_xml_node(iter, XML_CIB_TAG_LRM); crm_xml_add(iter, XML_ATTR_ID, uuid); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE); crm_xml_add(iter, XML_ATTR_ID, op->rsc_id); build_operation_update(iter, rsc, op, node_name, __func__); if (rsc) { const char *container = NULL; crm_xml_add(iter, XML_ATTR_TYPE, rsc->type); crm_xml_add(iter, XML_AGENT_ATTR_CLASS, rsc->standard); crm_xml_add(iter, XML_AGENT_ATTR_PROVIDER, rsc->provider); if (lock_time != 0) { /* Actions on a locked resource should either preserve the lock by * recording it with the action result, or clear it. */ if (!should_preserve_lock(op)) { lock_time = 0; } crm_xml_add_ll(iter, XML_CONFIG_ATTR_SHUTDOWN_LOCK, (long long) lock_time); } if (op->params) { container = g_hash_table_lookup(op->params, CRM_META"_"XML_RSC_ATTR_CONTAINER); } if (container) { crm_trace("Resource %s is a part of container resource %s", op->rsc_id, container); crm_xml_add(iter, XML_RSC_ATTR_CONTAINER, container); } } else { crm_warn("Resource %s no longer exists in the executor", op->rsc_id); controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id); goto cleanup; } crm_log_xml_trace(update, __func__); /* make it an asynchronous call and be done with it * * Best case: * the resource state will be discovered during * the next signup or election. * * Bad case: * we are shutting down and there is no DC at the time, * but then why were we shutting down then anyway? * (probably because of an internal error) * * Worst case: * we get shot for having resources "running" that really weren't * * the alternative however means blocking here for too long, which * isn't acceptable */ fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, rc, NULL); if (rc > 0) { last_resource_update = rc; } done: /* the return code is a call number, not an error code */ crm_trace("Sent resource state update message: %d for %s=%u on %s", rc, op->op_type, op->interval_ms, op->rsc_id); fsa_register_cib_callback(rc, FALSE, NULL, cib_rsc_callback); cleanup: free_xml(update); return rc; } void do_lrm_event(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input cur_input, fsa_data_t * msg_data) { CRM_CHECK(FALSE, return); } static char * unescape_newlines(const char *string) { char *pch = NULL; char *ret = NULL; static const char *escaped_newline = "\\n"; if (!string) { return NULL; } ret = strdup(string); pch = strstr(ret, escaped_newline); while (pch != NULL) { /* Replace newline escape pattern with actual newline (and a space so we * don't have to shuffle the rest of the buffer) */ pch[0] = '\n'; pch[1] = ' '; pch = strstr(pch, escaped_newline); } return ret; } static bool did_lrm_rsc_op_fail(lrm_state_t *lrm_state, const char * rsc_id, const char * op_type, guint interval_ms) { rsc_history_t *entry = NULL; CRM_CHECK(lrm_state != NULL, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); CRM_CHECK(op_type != NULL, return FALSE); entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->failed == NULL) { return FALSE; } if (pcmk__str_eq(entry->failed->rsc_id, rsc_id, pcmk__str_none) && pcmk__str_eq(entry->failed->op_type, op_type, pcmk__str_casei) && entry->failed->interval_ms == interval_ms) { return TRUE; } return FALSE; } void process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op, active_op_t *pending, xmlNode *action_xml) { char *op_id = NULL; char *op_key = NULL; int update_id = 0; gboolean remove = FALSE; gboolean removed = FALSE; bool need_direct_ack = FALSE; lrmd_rsc_info_t *rsc = NULL; const char *node_name = NULL; CRM_CHECK(op != NULL, return); CRM_CHECK(op->rsc_id != NULL, return); // Remap new status codes for older DCs if (compare_version(fsa_our_dc_version, "3.2.0") < 0) { switch (op->op_status) { case PCMK_LRM_OP_NOT_CONNECTED: op->op_status = PCMK_LRM_OP_ERROR; op->rc = PCMK_OCF_CONNECTION_DIED; break; case PCMK_LRM_OP_INVALID: op->op_status = PCMK_LRM_OP_ERROR; op->rc = CRM_DIRECT_NACK_RC; break; default: break; } } op_id = make_stop_id(op->rsc_id, op->call_id); op_key = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); // Get resource info if available (from executor state or action XML) if (lrm_state) { rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0); } if ((rsc == NULL) && action_xml) { xmlNode *xml = find_xml_node(action_xml, XML_CIB_TAG_RESOURCE, TRUE); const char *standard = crm_element_value(xml, XML_AGENT_ATTR_CLASS); const char *provider = crm_element_value(xml, XML_AGENT_ATTR_PROVIDER); const char *type = crm_element_value(xml, XML_ATTR_TYPE); if (standard && type) { crm_info("%s agent information not cached, using %s%s%s:%s from action XML", op->rsc_id, standard, (provider? ":" : ""), (provider? provider : ""), type); rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type); } else { crm_err("Can't process %s result because %s agent information not cached or in XML", op_key, op->rsc_id); } } // Get node name if available (from executor state or action XML) if (lrm_state) { node_name = lrm_state->node_name; } else if (action_xml) { node_name = crm_element_value(action_xml, XML_LRM_ATTR_TARGET); } if(pending == NULL) { remove = TRUE; if (lrm_state) { pending = g_hash_table_lookup(lrm_state->pending_ops, op_id); } } if (op->op_status == PCMK_LRM_OP_ERROR) { switch(op->rc) { case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_PROMOTED: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: // Leave it to the TE/scheduler to decide if this is an error op->op_status = PCMK_LRM_OP_DONE; break; default: /* Nothing to do */ break; } } if (op->op_status != PCMK_LRM_OP_CANCELLED) { /* We might not record the result, so directly acknowledge it to the * originator instead, so it doesn't time out waiting for the result * (especially important if part of a transition). */ need_direct_ack = TRUE; if (controld_action_is_recordable(op->op_type)) { if (node_name && rsc) { // We should record the result, and happily, we can update_id = do_update_resource(node_name, rsc, op, pending? pending->lock_time : 0); need_direct_ack = FALSE; } else if (op->rsc_deleted) { /* We shouldn't record the result (likely the resource was * refreshed, cleaned, or removed while this operation was * in flight). */ crm_notice("Not recording %s result in CIB because " "resource information was removed since it was initiated", op_key); } else { /* This shouldn't be possible; the executor didn't consider the * resource deleted, but we couldn't find resource or node * information. */ crm_err("Unable to record %s result in CIB: %s", op_key, (node_name? "No resource information" : "No node name")); } } } else if (op->interval_ms == 0) { /* A non-recurring operation was cancelled. Most likely, the * never-initiated action was removed from the executor's pending * operations list upon resource removal. */ need_direct_ack = TRUE; } else if (pending == NULL) { /* This recurring operation was cancelled, but was not pending. No * transition actions are waiting on it, nothing needs to be done. */ } else if (op->user_data == NULL) { /* This recurring operation was cancelled and pending, but we don't * have a transition key. This should never happen. */ crm_err("Recurring operation %s was cancelled without transition information", op_key); } else if (pcmk_is_set(pending->flags, active_op_remove)) { /* This recurring operation was cancelled (by us) and pending, and we * have been waiting for it to finish. */ if (lrm_state) { erase_lrm_history_by_op(lrm_state, op); } /* If the recurring operation had failed, the lrm_rsc_op is recorded as * "last_failure" which won't get erased from the cib given the logic on * purpose in erase_lrm_history_by_op(). So that the cancel action won't * have a chance to get confirmed by DC with process_op_deletion(). * Cluster transition would get stuck waiting for the remaining action * timer to time out. * * Directly acknowledge the cancel operation in this case. */ if (did_lrm_rsc_op_fail(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms)) { need_direct_ack = TRUE; } } else if (op->rsc_deleted) { /* This recurring operation was cancelled (but not by us, and the * executor does not have resource information, likely due to resource * cleanup, refresh, or removal) and pending. */ crm_debug("Recurring op %s was cancelled due to resource deletion", op_key); need_direct_ack = TRUE; } else { /* This recurring operation was cancelled (but not by us, likely by the * executor before stopping the resource) and pending. We don't need to * do anything special. */ } if (need_direct_ack) { controld_ack_event_directly(NULL, NULL, NULL, op, op->rsc_id); } if(remove == FALSE) { /* The caller will do this afterwards, but keep the logging consistent */ removed = TRUE; } else if (lrm_state && ((op->interval_ms == 0) || (op->op_status == PCMK_LRM_OP_CANCELLED))) { gboolean found = g_hash_table_remove(lrm_state->pending_ops, op_id); if (op->interval_ms != 0) { removed = TRUE; } else if (found) { removed = TRUE; crm_trace("Op %s (call=%d, stop-id=%s, remaining=%u): Confirmed", op_key, op->call_id, op_id, g_hash_table_size(lrm_state->pending_ops)); } } if (node_name == NULL) { node_name = "unknown node"; // for logging } switch (op->op_status) { case PCMK_LRM_OP_CANCELLED: crm_info("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s confirmed=%s", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_lrm_status_str(op->op_status), op->call_id, op_key, pcmk__btoa(removed)); break; case PCMK_LRM_OP_DONE: crm_notice("Result of %s operation for %s on %s: %s " CRM_XS " rc=%d call=%d key=%s confirmed=%s cib-update=%d", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_ocf_exitcode_str(op->rc), op->rc, op->call_id, op_key, pcmk__btoa(removed), update_id); break; case PCMK_LRM_OP_TIMEOUT: crm_err("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s timeout=%dms", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_lrm_status_str(op->op_status), op->call_id, op_key, op->timeout); break; default: crm_err("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s confirmed=%s status=%d cib-update=%d", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_lrm_status_str(op->op_status), op->call_id, op_key, pcmk__btoa(removed), op->op_status, update_id); } if (op->output) { char *prefix = crm_strdup_printf("%s-" PCMK__OP_FMT ":%d", node_name, op->rsc_id, op->op_type, op->interval_ms, op->call_id); if (op->rc) { crm_log_output(LOG_NOTICE, prefix, op->output); } else { crm_log_output(LOG_DEBUG, prefix, op->output); } free(prefix); } if (lrm_state) { if (!pcmk__str_eq(op->op_type, RSC_METADATA, pcmk__str_casei)) { crmd_alert_resource_op(lrm_state->node_name, op); } else if (rsc && (op->rc == PCMK_OCF_OK)) { char *metadata = unescape_newlines(op->output); metadata_cache_update(lrm_state->metadata_cache, rsc, metadata); free(metadata); } } if (op->rsc_deleted) { crm_info("Deletion of resource '%s' complete after %s", op->rsc_id, op_key); if (lrm_state) { delete_rsc_entry(lrm_state, NULL, op->rsc_id, NULL, pcmk_ok, NULL); } } /* If a shutdown was escalated while operations were pending, * then the FSA will be stalled right now... allow it to continue */ mainloop_set_trigger(fsa_source); if (lrm_state && rsc) { update_history_cache(lrm_state, rsc, op); } lrmd_free_rsc_info(rsc); free(op_key); free(op_id); } diff --git a/daemons/controld/controld_execd_state.c b/daemons/controld/controld_execd_state.c index e1e34c3574..2b07476e60 100644 --- a/daemons/controld/controld_execd_state.c +++ b/daemons/controld/controld_execd_state.c @@ -1,811 +1,828 @@ /* * Copyright 2012-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include + +#include + #include #include #include #include #include #include #include #include GHashTable *lrm_state_table = NULL; extern GHashTable *proxy_table; int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg); void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg)); static void free_rsc_info(gpointer value) { lrmd_rsc_info_t *rsc_info = value; lrmd_free_rsc_info(rsc_info); } static void free_deletion_op(gpointer value) { struct pending_deletion_op_s *op = value; free(op->rsc); delete_ha_msg_input(op->input); free(op); } static void free_recurring_op(gpointer value) { active_op_t *op = value; free(op->user_data); free(op->rsc_id); free(op->op_type); free(op->op_key); if (op->params) { g_hash_table_destroy(op->params); } free(op); } static gboolean fail_pending_op(gpointer key, gpointer value, gpointer user_data) { lrmd_event_data_t event = { 0, }; lrm_state_t *lrm_state = user_data; active_op_t *op = value; crm_trace("Pre-emptively failing " PCMK__OP_FMT " on %s (call=%s, %s)", op->rsc_id, op->op_type, op->interval_ms, lrm_state->node_name, (char*)key, op->user_data); event.type = lrmd_event_exec_complete; event.rsc_id = op->rsc_id; event.op_type = op->op_type; event.user_data = op->user_data; event.timeout = 0; event.interval_ms = op->interval_ms; event.rc = PCMK_OCF_UNKNOWN_ERROR; event.op_status = PCMK_LRM_OP_NOT_CONNECTED; event.t_run = (unsigned int) op->start_time; event.t_rcchange = (unsigned int) op->start_time; event.call_id = op->call_id; event.remote_nodename = lrm_state->node_name; event.params = op->params; process_lrm_event(lrm_state, &event, op, NULL); return TRUE; } gboolean lrm_state_is_local(lrm_state_t *lrm_state) { if (lrm_state == NULL || fsa_our_uname == NULL) { return FALSE; } if (strcmp(lrm_state->node_name, fsa_our_uname) != 0) { return FALSE; } return TRUE; } lrm_state_t * lrm_state_create(const char *node_name) { lrm_state_t *state = NULL; if (!node_name) { crm_err("No node name given for lrm state object"); return NULL; } state = calloc(1, sizeof(lrm_state_t)); if (!state) { return NULL; } state->node_name = strdup(node_name); state->rsc_info_cache = pcmk__strkey_table(NULL, free_rsc_info); state->deletion_ops = pcmk__strkey_table(free, free_deletion_op); state->pending_ops = pcmk__strkey_table(free, free_recurring_op); state->resource_history = pcmk__strkey_table(NULL, history_free); state->metadata_cache = metadata_cache_new(); g_hash_table_insert(lrm_state_table, (char *)state->node_name, state); return state; } void lrm_state_destroy(const char *node_name) { g_hash_table_remove(lrm_state_table, node_name); } static gboolean remote_proxy_remove_by_node(gpointer key, gpointer value, gpointer user_data) { remote_proxy_t *proxy = value; const char *node_name = user_data; if (pcmk__str_eq(node_name, proxy->node_name, pcmk__str_casei)) { return TRUE; } return FALSE; } static void internal_lrm_state_destroy(gpointer data) { lrm_state_t *lrm_state = data; if (!lrm_state) { return; } crm_trace("Destroying proxy table %s with %d members", lrm_state->node_name, g_hash_table_size(proxy_table)); g_hash_table_foreach_remove(proxy_table, remote_proxy_remove_by_node, (char *) lrm_state->node_name); remote_ra_cleanup(lrm_state); lrmd_api_delete(lrm_state->conn); if (lrm_state->rsc_info_cache) { crm_trace("Destroying rsc info cache with %d members", g_hash_table_size(lrm_state->rsc_info_cache)); g_hash_table_destroy(lrm_state->rsc_info_cache); } if (lrm_state->resource_history) { crm_trace("Destroying history op cache with %d members", g_hash_table_size(lrm_state->resource_history)); g_hash_table_destroy(lrm_state->resource_history); } if (lrm_state->deletion_ops) { crm_trace("Destroying deletion op cache with %d members", g_hash_table_size(lrm_state->deletion_ops)); g_hash_table_destroy(lrm_state->deletion_ops); } if (lrm_state->pending_ops) { crm_trace("Destroying pending op cache with %d members", g_hash_table_size(lrm_state->pending_ops)); g_hash_table_destroy(lrm_state->pending_ops); } metadata_cache_free(lrm_state->metadata_cache); free((char *)lrm_state->node_name); free(lrm_state); } void lrm_state_reset_tables(lrm_state_t * lrm_state, gboolean reset_metadata) { if (lrm_state->resource_history) { crm_trace("Re-setting history op cache with %d members", g_hash_table_size(lrm_state->resource_history)); g_hash_table_remove_all(lrm_state->resource_history); } if (lrm_state->deletion_ops) { crm_trace("Re-setting deletion op cache with %d members", g_hash_table_size(lrm_state->deletion_ops)); g_hash_table_remove_all(lrm_state->deletion_ops); } if (lrm_state->pending_ops) { crm_trace("Re-setting pending op cache with %d members", g_hash_table_size(lrm_state->pending_ops)); g_hash_table_remove_all(lrm_state->pending_ops); } if (lrm_state->rsc_info_cache) { crm_trace("Re-setting rsc info cache with %d members", g_hash_table_size(lrm_state->rsc_info_cache)); g_hash_table_remove_all(lrm_state->rsc_info_cache); } if (reset_metadata) { metadata_cache_reset(lrm_state->metadata_cache); } } gboolean lrm_state_init_local(void) { if (lrm_state_table) { return TRUE; } lrm_state_table = pcmk__strikey_table(NULL, internal_lrm_state_destroy); if (!lrm_state_table) { return FALSE; } proxy_table = pcmk__strikey_table(NULL, remote_proxy_free); if (!proxy_table) { g_hash_table_destroy(lrm_state_table); lrm_state_table = NULL; return FALSE; } return TRUE; } void lrm_state_destroy_all(void) { if (lrm_state_table) { crm_trace("Destroying state table with %d members", g_hash_table_size(lrm_state_table)); g_hash_table_destroy(lrm_state_table); lrm_state_table = NULL; } if(proxy_table) { crm_trace("Destroying proxy table with %d members", g_hash_table_size(proxy_table)); g_hash_table_destroy(proxy_table); proxy_table = NULL; } } lrm_state_t * lrm_state_find(const char *node_name) { if (!node_name) { return NULL; } return g_hash_table_lookup(lrm_state_table, node_name); } lrm_state_t * lrm_state_find_or_create(const char *node_name) { lrm_state_t *lrm_state; lrm_state = g_hash_table_lookup(lrm_state_table, node_name); if (!lrm_state) { lrm_state = lrm_state_create(node_name); } return lrm_state; } GList * lrm_state_get_list(void) { return g_hash_table_get_values(lrm_state_table); } static remote_proxy_t * find_connected_proxy_by_node(const char * node_name) { GHashTableIter gIter; remote_proxy_t *proxy = NULL; CRM_CHECK(proxy_table != NULL, return NULL); g_hash_table_iter_init(&gIter, proxy_table); while (g_hash_table_iter_next(&gIter, NULL, (gpointer *) &proxy)) { if (proxy->source && pcmk__str_eq(node_name, proxy->node_name, pcmk__str_casei)) { return proxy; } } return NULL; } static void remote_proxy_disconnect_by_node(const char * node_name) { remote_proxy_t *proxy = NULL; CRM_CHECK(proxy_table != NULL, return); while ((proxy = find_connected_proxy_by_node(node_name)) != NULL) { /* mainloop_del_ipc_client() eventually calls remote_proxy_disconnected() * , which removes the entry from proxy_table. * Do not do this in a g_hash_table_iter_next() loop. */ if (proxy->source) { mainloop_del_ipc_client(proxy->source); } } return; } void lrm_state_disconnect_only(lrm_state_t * lrm_state) { int removed = 0; if (!lrm_state->conn) { return; } crm_trace("Disconnecting %s", lrm_state->node_name); remote_proxy_disconnect_by_node(lrm_state->node_name); ((lrmd_t *) lrm_state->conn)->cmds->disconnect(lrm_state->conn); if (!pcmk_is_set(fsa_input_register, R_SHUTDOWN)) { removed = g_hash_table_foreach_remove(lrm_state->pending_ops, fail_pending_op, lrm_state); crm_trace("Synthesized %d operation failures for %s", removed, lrm_state->node_name); } } void lrm_state_disconnect(lrm_state_t * lrm_state) { if (!lrm_state->conn) { return; } lrm_state_disconnect_only(lrm_state); lrmd_api_delete(lrm_state->conn); lrm_state->conn = NULL; } int lrm_state_is_connected(lrm_state_t * lrm_state) { if (!lrm_state->conn) { return FALSE; } return ((lrmd_t *) lrm_state->conn)->cmds->is_connected(lrm_state->conn); } int lrm_state_poke_connection(lrm_state_t * lrm_state) { if (!lrm_state->conn) { - return -1; + return -ENOTCONN; } return ((lrmd_t *) lrm_state->conn)->cmds->poke_connection(lrm_state->conn); } +// \return Standard Pacemaker return code int -lrm_state_ipc_connect(lrm_state_t * lrm_state) +controld_connect_local_executor(lrm_state_t *lrm_state) { - int ret; + int rc = pcmk_rc_ok; - if (!lrm_state->conn) { - lrm_state->conn = lrmd_api_new(); - ((lrmd_t *) lrm_state->conn)->cmds->set_callback(lrm_state->conn, lrm_op_callback); + if (lrm_state->conn == NULL) { + lrmd_t *api = NULL; + + rc = lrmd__new(&api, NULL, NULL, 0); + if (rc != pcmk_rc_ok) { + return rc; + } + api->cmds->set_callback(api, lrm_op_callback); + lrm_state->conn = api; } - ret = ((lrmd_t *) lrm_state->conn)->cmds->connect(lrm_state->conn, CRM_SYSTEM_CRMD, NULL); + rc = ((lrmd_t *) lrm_state->conn)->cmds->connect(lrm_state->conn, + CRM_SYSTEM_CRMD, NULL); + rc = pcmk_legacy2rc(rc); - if (ret != pcmk_ok) { - lrm_state->num_lrm_register_fails++; - } else { + if (rc == pcmk_rc_ok) { lrm_state->num_lrm_register_fails = 0; + } else { + lrm_state->num_lrm_register_fails++; } - - return ret; + return rc; } static remote_proxy_t * crmd_remote_proxy_new(lrmd_t *lrmd, const char *node_name, const char *session_id, const char *channel) { struct ipc_client_callbacks proxy_callbacks = { .dispatch = remote_proxy_dispatch, .destroy = remote_proxy_disconnected }; remote_proxy_t *proxy = remote_proxy_new(lrmd, &proxy_callbacks, node_name, session_id, channel); return proxy; } gboolean crmd_is_proxy_session(const char *session) { return g_hash_table_lookup(proxy_table, session) ? TRUE : FALSE; } void crmd_proxy_send(const char *session, xmlNode *msg) { remote_proxy_t *proxy = g_hash_table_lookup(proxy_table, session); lrm_state_t *lrm_state = NULL; if (!proxy) { return; } crm_log_xml_trace(msg, "to-proxy"); lrm_state = lrm_state_find(proxy->node_name); if (lrm_state) { crm_trace("Sending event to %.8s on %s", proxy->session_id, proxy->node_name); remote_proxy_relay_event(proxy, msg); } } static void crmd_proxy_dispatch(const char *session, xmlNode *msg) { crm_trace("Processing proxied IPC message from session %s", session); crm_log_xml_trace(msg, "controller[inbound]"); crm_xml_add(msg, F_CRM_SYS_FROM, session); if (controld_authorize_ipc_message(msg, NULL, session)) { route_message(C_IPC_MESSAGE, msg); } trigger_fsa(); } static void remote_config_check(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { if (rc != pcmk_ok) { crm_err("Query resulted in an error: %s", pcmk_strerror(rc)); if (rc == -EACCES || rc == -pcmk_err_schema_validation) { crm_err("The cluster is mis-configured - shutting down and staying down"); } } else { lrmd_t * lrmd = (lrmd_t *)user_data; crm_time_t *now = crm_time_new(NULL); GHashTable *config_hash = pcmk__strkey_table(free, free); crm_debug("Call %d : Parsing CIB options", call_id); pe_unpack_nvpairs(output, output, XML_CIB_TAG_PROPSET, NULL, config_hash, CIB_OPTIONS_FIRST, FALSE, now, NULL); /* Now send it to the remote peer */ lrmd__validate_remote_settings(lrmd, config_hash); g_hash_table_destroy(config_hash); crm_time_free(now); } } static void crmd_remote_proxy_cb(lrmd_t *lrmd, void *userdata, xmlNode *msg) { lrm_state_t *lrm_state = userdata; const char *session = crm_element_value(msg, F_LRMD_IPC_SESSION); remote_proxy_t *proxy = g_hash_table_lookup(proxy_table, session); const char *op = crm_element_value(msg, F_LRMD_IPC_OP); if (pcmk__str_eq(op, LRMD_IPC_OP_NEW, pcmk__str_casei)) { const char *channel = crm_element_value(msg, F_LRMD_IPC_IPC_SERVER); proxy = crmd_remote_proxy_new(lrmd, lrm_state->node_name, session, channel); if (!remote_ra_controlling_guest(lrm_state)) { if (proxy != NULL) { /* Look up stonith-watchdog-timeout and send to the remote peer for validation */ int rc = fsa_cib_conn->cmds->query(fsa_cib_conn, XML_CIB_TAG_CRMCONFIG, NULL, cib_scope_local); fsa_cib_conn->cmds->register_callback_full(fsa_cib_conn, rc, 10, FALSE, lrmd, "remote_config_check", remote_config_check, NULL); } } else { crm_debug("Skipping remote_config_check for guest-nodes"); } } else if (pcmk__str_eq(op, LRMD_IPC_OP_SHUTDOWN_REQ, pcmk__str_casei)) { char *now_s = NULL; crm_notice("%s requested shutdown of its remote connection", lrm_state->node_name); if (!remote_ra_is_in_maintenance(lrm_state)) { now_s = pcmk__ttoa(time(NULL)); update_attrd(lrm_state->node_name, XML_CIB_ATTR_SHUTDOWN, now_s, NULL, TRUE); free(now_s); remote_proxy_ack_shutdown(lrmd); crm_warn("Reconnection attempts to %s may result in failures that must be cleared", lrm_state->node_name); } else { remote_proxy_nack_shutdown(lrmd); crm_notice("Remote resource for %s is not managed so no ordered shutdown happening", lrm_state->node_name); } return; } else if (pcmk__str_eq(op, LRMD_IPC_OP_REQUEST, pcmk__str_casei) && proxy && proxy->is_local) { /* This is for the controller, which we are, so don't try * to send to ourselves over IPC -- do it directly. */ int flags = 0; xmlNode *request = get_message_xml(msg, F_LRMD_IPC_MSG); CRM_CHECK(request != NULL, return); CRM_CHECK(lrm_state->node_name, return); crm_xml_add(request, XML_ACL_TAG_ROLE, "pacemaker-remote"); pcmk__update_acl_user(request, F_LRMD_IPC_USER, lrm_state->node_name); /* Pacemaker Remote nodes don't know their own names (as known to the * cluster). When getting a node info request with no name or ID, add * the name, so we don't return info for ourselves instead of the * Pacemaker Remote node. */ if (pcmk__str_eq(crm_element_value(request, F_CRM_TASK), CRM_OP_NODE_INFO, pcmk__str_casei)) { int node_id = 0; crm_element_value_int(request, XML_ATTR_ID, &node_id); if ((node_id <= 0) && (crm_element_value(request, XML_ATTR_UNAME) == NULL)) { crm_xml_add(request, XML_ATTR_UNAME, lrm_state->node_name); } } crmd_proxy_dispatch(session, request); crm_element_value_int(msg, F_LRMD_IPC_MSG_FLAGS, &flags); if (flags & crm_ipc_client_response) { int msg_id = 0; xmlNode *op_reply = create_xml_node(NULL, "ack"); crm_xml_add(op_reply, "function", __func__); crm_xml_add_int(op_reply, "line", __LINE__); crm_element_value_int(msg, F_LRMD_IPC_MSG_ID, &msg_id); remote_proxy_relay_response(proxy, op_reply, msg_id); free_xml(op_reply); } } else { remote_proxy_cb(lrmd, lrm_state->node_name, msg); } } +// \return Standard Pacemaker return code int -lrm_state_remote_connect_async(lrm_state_t * lrm_state, const char *server, int port, - int timeout_ms) +controld_connect_remote_executor(lrm_state_t *lrm_state, const char *server, + int port, int timeout_ms) { - int ret; + int rc = pcmk_rc_ok; - if (!lrm_state->conn) { - lrm_state->conn = lrmd_remote_api_new(lrm_state->node_name, server, port); - if (!lrm_state->conn) { - return -1; + if (lrm_state->conn == NULL) { + lrmd_t *api = NULL; + + rc = lrmd__new(&api, lrm_state->node_name, server, port); + if (rc != pcmk_rc_ok) { + crm_warn("Pacemaker Remote connection to %s:%s failed: %s " + CRM_XS " rc=%d", server, port, pcmk_rc_str(rc), rc); + + return rc; } - ((lrmd_t *) lrm_state->conn)->cmds->set_callback(lrm_state->conn, remote_lrm_op_callback); - lrmd_internal_set_proxy_callback(lrm_state->conn, lrm_state, crmd_remote_proxy_cb); + lrm_state->conn = api; + api->cmds->set_callback(api, remote_lrm_op_callback); + lrmd_internal_set_proxy_callback(api, lrm_state, crmd_remote_proxy_cb); } - crm_trace("initiating remote connection to %s at %d with timeout %d", server, port, timeout_ms); - ret = - ((lrmd_t *) lrm_state->conn)->cmds->connect_async(lrm_state->conn, lrm_state->node_name, - timeout_ms); - - if (ret != pcmk_ok) { - lrm_state->num_lrm_register_fails++; - } else { + crm_trace("Initiating remote connection to %s:%d with timeout %dms", + server, port, timeout_ms); + rc = ((lrmd_t *) lrm_state->conn)->cmds->connect_async(lrm_state->conn, + lrm_state->node_name, + timeout_ms); + if (rc == pcmk_ok) { lrm_state->num_lrm_register_fails = 0; + } else { + lrm_state->num_lrm_register_fails++; // Ignored for remote connections } - - return ret; + return pcmk_legacy2rc(rc); } int lrm_state_get_metadata(lrm_state_t * lrm_state, const char *class, const char *provider, const char *agent, char **output, enum lrmd_call_options options) { lrmd_key_value_t *params = NULL; if (!lrm_state->conn) { return -ENOTCONN; } /* Add the node name to the environment, as is done with normal resource * action calls. Meta-data calls shouldn't need it, but some agents are * written with an ocf_local_nodename call at the beginning regardless of * action. Without the environment variable, the agent would try to contact * the controller to get the node name -- but the controller would be * blocking on the synchronous meta-data call. * * At this point, we have to assume that agents are unlikely to make other * calls that require the controller, such as crm_node --quorum or * --cluster-id. * * @TODO Make meta-data calls asynchronous. (This will be part of a larger * project to make meta-data calls via the executor rather than directly.) */ params = lrmd_key_value_add(params, CRM_META "_" XML_LRM_ATTR_TARGET, lrm_state->node_name); return ((lrmd_t *) lrm_state->conn)->cmds->get_metadata_params(lrm_state->conn, class, provider, agent, output, options, params); } int lrm_state_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms) { if (!lrm_state->conn) { return -ENOTCONN; } /* Figure out a way to make this async? * NOTICE: Currently it's synced and directly acknowledged in do_lrm_invoke(). */ if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) { return remote_ra_cancel(lrm_state, rsc_id, action, interval_ms); } return ((lrmd_t *) lrm_state->conn)->cmds->cancel(lrm_state->conn, rsc_id, action, interval_ms); } lrmd_rsc_info_t * lrm_state_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id, enum lrmd_call_options options) { lrmd_rsc_info_t *rsc = NULL; if (!lrm_state->conn) { return NULL; } if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) { return remote_ra_get_rsc_info(lrm_state, rsc_id); } rsc = g_hash_table_lookup(lrm_state->rsc_info_cache, rsc_id); if (rsc == NULL) { /* only contact the lrmd if we don't already have a cached rsc info */ rsc = ((lrmd_t *) lrm_state->conn)->cmds->get_rsc_info(lrm_state->conn, rsc_id, options); if (rsc == NULL) { return NULL; } /* cache the result */ g_hash_table_insert(lrm_state->rsc_info_cache, rsc->id, rsc); } return lrmd_copy_rsc_info(rsc); } int lrm_state_exec(lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout, /* ms */ int start_delay, /* ms */ lrmd_key_value_t * params) { if (!lrm_state->conn) { lrmd_key_value_freeall(params); return -ENOTCONN; } if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) { return remote_ra_exec(lrm_state, rsc_id, action, userdata, interval_ms, timeout, start_delay, params); } return ((lrmd_t *) lrm_state->conn)->cmds->exec(lrm_state->conn, rsc_id, action, userdata, interval_ms, timeout, start_delay, lrmd_opt_notify_changes_only, params); } int lrm_state_register_rsc(lrm_state_t * lrm_state, const char *rsc_id, const char *class, const char *provider, const char *agent, enum lrmd_call_options options) { lrmd_t *conn = (lrmd_t *) lrm_state->conn; if (conn == NULL) { return -ENOTCONN; } if (is_remote_lrmd_ra(agent, provider, NULL)) { return lrm_state_find_or_create(rsc_id)? pcmk_ok : -EINVAL; } /* @TODO Implement an asynchronous version of this (currently a blocking * call to the lrmd). */ return conn->cmds->register_rsc(lrm_state->conn, rsc_id, class, provider, agent, options); } int lrm_state_unregister_rsc(lrm_state_t * lrm_state, const char *rsc_id, enum lrmd_call_options options) { if (!lrm_state->conn) { return -ENOTCONN; } if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) { lrm_state_destroy(rsc_id); return pcmk_ok; } g_hash_table_remove(lrm_state->rsc_info_cache, rsc_id); /* @TODO Optimize this ... this function is a blocking round trip from * client to daemon. The controld_execd_state.c code path that uses this * function should always treat it as an async operation. The executor API * should make an async version available. */ return ((lrmd_t *) lrm_state->conn)->cmds->unregister_rsc(lrm_state->conn, rsc_id, options); } /* * Functions for sending alerts via local executor connection */ static GList *crmd_alert_list = NULL; void crmd_unpack_alerts(xmlNode *alerts) { pe_free_alert_list(crmd_alert_list); crmd_alert_list = pe_unpack_alerts(alerts); } void crmd_alert_node_event(crm_node_t *node) { lrm_state_t *lrm_state; if (crmd_alert_list == NULL) { return; } lrm_state = lrm_state_find(fsa_our_uname); if (lrm_state == NULL) { return; } lrmd_send_node_alert((lrmd_t *) lrm_state->conn, crmd_alert_list, node->uname, node->id, node->state); } void crmd_alert_fencing_op(stonith_event_t * e) { char *desc; lrm_state_t *lrm_state; if (crmd_alert_list == NULL) { return; } lrm_state = lrm_state_find(fsa_our_uname); if (lrm_state == NULL) { return; } desc = crm_strdup_printf("Operation %s of %s by %s for %s@%s: %s (ref=%s)", e->action, e->target, (e->executioner? e->executioner : ""), e->client_origin, e->origin, pcmk_strerror(e->result), e->id); lrmd_send_fencing_alert((lrmd_t *) lrm_state->conn, crmd_alert_list, e->target, e->operation, desc, e->result); free(desc); } void crmd_alert_resource_op(const char *node, lrmd_event_data_t * op) { lrm_state_t *lrm_state; if (crmd_alert_list == NULL) { return; } lrm_state = lrm_state_find(fsa_our_uname); if (lrm_state == NULL) { return; } lrmd_send_resource_alert((lrmd_t *) lrm_state->conn, crmd_alert_list, node, op); } diff --git a/daemons/controld/controld_lrm.h b/daemons/controld/controld_lrm.h index 8bc16107b4..340205778f 100644 --- a/daemons/controld/controld_lrm.h +++ b/daemons/controld/controld_lrm.h @@ -1,192 +1,192 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef CONTROLD_LRM__H # define CONTROLD_LRM__H #include extern gboolean verify_stopped(enum crmd_fsa_state cur_state, int log_level); void lrm_clear_last_failure(const char *rsc_id, const char *node_name, const char *operation, guint interval_ms); void lrm_op_callback(lrmd_event_data_t * op); lrmd_t *crmd_local_lrmd_conn(void); typedef struct resource_history_s { char *id; uint32_t last_callid; lrmd_rsc_info_t rsc; lrmd_event_data_t *last; lrmd_event_data_t *failed; GList *recurring_op_list; /* Resources must be stopped using the same * parameters they were started with. This hashtable * holds the parameters that should be used for the next stop * cmd on this resource. */ GHashTable *stop_params; } rsc_history_t; void history_free(gpointer data); enum active_op_e { active_op_remove = (1 << 0), active_op_cancelled = (1 << 1), }; // In-flight action (recurring or pending) typedef struct active_op_s { guint interval_ms; int call_id; uint32_t flags; // bitmask of active_op_e time_t start_time; time_t lock_time; char *rsc_id; char *op_type; char *op_key; char *user_data; GHashTable *params; } active_op_t; #define controld_set_active_op_flags(active_op, flags_to_set) do { \ (active_op)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Active operation", (active_op)->op_key, \ (active_op)->flags, (flags_to_set), #flags_to_set); \ } while (0) #define controld_clear_active_op_flags(active_op, flags_to_clear) do { \ (active_op)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Active operation", (active_op)->op_key, \ (active_op)->flags, (flags_to_clear), #flags_to_clear); \ } while (0) typedef struct lrm_state_s { const char *node_name; void *conn; // Reserved for controld_execd_state.c usage void *remote_ra_data; // Reserved for controld_remote_ra.c usage GHashTable *resource_history; GHashTable *pending_ops; GHashTable *deletion_ops; GHashTable *rsc_info_cache; GHashTable *metadata_cache; // key = class[:provider]:agent, value = ra_metadata_s int num_lrm_register_fails; } lrm_state_t; struct pending_deletion_op_s { char *rsc; ha_msg_input_t *input; }; /*! * \brief Check whether this the local IPC connection to the executor */ gboolean lrm_state_is_local(lrm_state_t *lrm_state); /*! * \brief Clear all state information from a single state entry. * \note It sometimes useful to save metadata cache when it won't go stale. * \note This does not close the executor connection */ void lrm_state_reset_tables(lrm_state_t * lrm_state, gboolean reset_metadata); GList *lrm_state_get_list(void); /*! * \brief Initiate internal state tables */ gboolean lrm_state_init_local(void); /*! * \brief Destroy all state entries and internal state tables */ void lrm_state_destroy_all(void); /*! * \brief Create executor connection entry */ lrm_state_t *lrm_state_create(const char *node_name); /*! * \brief Destroy executor connection by node name */ void lrm_state_destroy(const char *node_name); /*! * \brief Find lrm_state data by node name */ lrm_state_t *lrm_state_find(const char *node_name); /*! * \brief Either find or create a new entry */ lrm_state_t *lrm_state_find_or_create(const char *node_name); /*! * The functions below are wrappers for the executor API the the controller * uses. These wrapper functions allow us to treat the controller's remote * executor connection resources the same as regular resources. Internally, * regular resources go to the executor, and remote connection resources are * handled locally in the controller. */ void lrm_state_disconnect_only(lrm_state_t * lrm_state); void lrm_state_disconnect(lrm_state_t * lrm_state); -int lrm_state_ipc_connect(lrm_state_t * lrm_state); -int lrm_state_remote_connect_async(lrm_state_t * lrm_state, const char *server, int port, - int timeout); +int controld_connect_local_executor(lrm_state_t *lrm_state); +int controld_connect_remote_executor(lrm_state_t *lrm_state, const char *server, + int port, int timeout); int lrm_state_is_connected(lrm_state_t * lrm_state); int lrm_state_poke_connection(lrm_state_t * lrm_state); int lrm_state_get_metadata(lrm_state_t * lrm_state, const char *class, const char *provider, const char *agent, char **output, enum lrmd_call_options options); int lrm_state_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms); int lrm_state_exec(lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout, /* ms */ int start_delay, /* ms */ lrmd_key_value_t * params); lrmd_rsc_info_t *lrm_state_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id, enum lrmd_call_options options); int lrm_state_register_rsc(lrm_state_t * lrm_state, const char *rsc_id, const char *class, const char *provider, const char *agent, enum lrmd_call_options options); int lrm_state_unregister_rsc(lrm_state_t * lrm_state, const char *rsc_id, enum lrmd_call_options options); // Functions used to manage remote executor connection resources void remote_lrm_op_callback(lrmd_event_data_t * op); gboolean is_remote_lrmd_ra(const char *agent, const char *provider, const char *id); lrmd_rsc_info_t *remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id); int remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms); int remote_ra_exec(lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout, /* ms */ int start_delay, /* ms */ lrmd_key_value_t * params); void remote_ra_cleanup(lrm_state_t * lrm_state); void remote_ra_fail(const char *node_name); void remote_ra_process_pseudo(xmlNode *xml); gboolean remote_ra_is_in_maintenance(lrm_state_t * lrm_state); void remote_ra_process_maintenance_nodes(xmlNode *xml); gboolean remote_ra_controlling_guest(lrm_state_t * lrm_state); void process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op, active_op_t *pending, xmlNode *action_xml); void controld_ack_event_directly(const char *to_host, const char *to_sys, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *rsc_id); void controld_rc2event(lrmd_event_data_t *event, int rc); void controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id); #endif diff --git a/daemons/controld/controld_messages.c b/daemons/controld/controld_messages.c index 7e3a036329..19c0032c02 100644 --- a/daemons/controld/controld_messages.c +++ b/daemons/controld/controld_messages.c @@ -1,1283 +1,1283 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include GList *fsa_message_queue = NULL; extern void crm_shutdown(int nsig); static enum crmd_fsa_input handle_message(xmlNode *msg, enum crmd_fsa_cause cause); static void handle_response(xmlNode *stored_msg); static enum crmd_fsa_input handle_request(xmlNode *stored_msg, enum crmd_fsa_cause cause); static enum crmd_fsa_input handle_shutdown_request(xmlNode *stored_msg); static void send_msg_via_ipc(xmlNode * msg, const char *sys); /* debug only, can wrap all it likes */ int last_data_id = 0; void register_fsa_error_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input, fsa_data_t * cur_data, void *new_data, const char *raised_from) { /* save the current actions if any */ if (fsa_actions != A_NOTHING) { register_fsa_input_adv(cur_data ? cur_data->fsa_cause : C_FSA_INTERNAL, I_NULL, cur_data ? cur_data->data : NULL, fsa_actions, TRUE, __func__); } /* reset the action list */ crm_info("Resetting the current action list"); fsa_dump_actions(fsa_actions, "Drop"); fsa_actions = A_NOTHING; /* register the error */ register_fsa_input_adv(cause, input, new_data, A_NOTHING, TRUE, raised_from); } int register_fsa_input_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input, void *data, uint64_t with_actions, gboolean prepend, const char *raised_from) { unsigned old_len = g_list_length(fsa_message_queue); fsa_data_t *fsa_data = NULL; if (raised_from == NULL) { raised_from = ""; } if (input == I_NULL && with_actions == A_NOTHING /* && data == NULL */ ) { /* no point doing anything */ crm_err("Cannot add entry to queue: no input and no action"); return 0; } if (input == I_WAIT_FOR_EVENT) { do_fsa_stall = TRUE; crm_debug("Stalling the FSA pending further input: source=%s cause=%s data=%p queue=%d", raised_from, fsa_cause2string(cause), data, old_len); if (old_len > 0) { fsa_dump_queue(LOG_TRACE); prepend = FALSE; } if (data == NULL) { controld_set_fsa_action_flags(with_actions); fsa_dump_actions(with_actions, "Restored"); return 0; } /* Store everything in the new event and reset fsa_actions */ with_actions |= fsa_actions; fsa_actions = A_NOTHING; } last_data_id++; crm_trace("%s %s FSA input %d (%s) due to %s, %s data", raised_from, (prepend? "prepended" : "appended"), last_data_id, fsa_input2string(input), fsa_cause2string(cause), (data? "with" : "without")); fsa_data = calloc(1, sizeof(fsa_data_t)); fsa_data->id = last_data_id; fsa_data->fsa_input = input; fsa_data->fsa_cause = cause; fsa_data->origin = raised_from; fsa_data->data = NULL; fsa_data->data_type = fsa_dt_none; fsa_data->actions = with_actions; if (with_actions != A_NOTHING) { crm_trace("Adding actions %.16llx to input", (unsigned long long) with_actions); } if (data != NULL) { switch (cause) { case C_FSA_INTERNAL: case C_CRMD_STATUS_CALLBACK: case C_IPC_MESSAGE: case C_HA_MESSAGE: CRM_CHECK(((ha_msg_input_t *) data)->msg != NULL, crm_err("Bogus data from %s", raised_from)); crm_trace("Copying %s data from %s as cluster message data", fsa_cause2string(cause), raised_from); fsa_data->data = copy_ha_msg_input(data); fsa_data->data_type = fsa_dt_ha_msg; break; case C_LRM_OP_CALLBACK: crm_trace("Copying %s data from %s as lrmd_event_data_t", fsa_cause2string(cause), raised_from); fsa_data->data = lrmd_copy_event((lrmd_event_data_t *) data); fsa_data->data_type = fsa_dt_lrm; break; case C_TIMER_POPPED: case C_SHUTDOWN: case C_UNKNOWN: case C_STARTUP: crm_crit("Copying %s data (from %s) is not yet implemented", fsa_cause2string(cause), raised_from); crmd_exit(CRM_EX_SOFTWARE); break; } } /* make sure to free it properly later */ if (prepend) { fsa_message_queue = g_list_prepend(fsa_message_queue, fsa_data); } else { fsa_message_queue = g_list_append(fsa_message_queue, fsa_data); } crm_trace("FSA message queue length is %d", g_list_length(fsa_message_queue)); /* fsa_dump_queue(LOG_TRACE); */ if (old_len == g_list_length(fsa_message_queue)) { crm_err("Couldn't add message to the queue"); } if (fsa_source && input != I_WAIT_FOR_EVENT) { crm_trace("Triggering FSA"); mainloop_set_trigger(fsa_source); } return last_data_id; } void fsa_dump_queue(int log_level) { int offset = 0; GList *lpc = NULL; for (lpc = fsa_message_queue; lpc != NULL; lpc = lpc->next) { fsa_data_t *data = (fsa_data_t *) lpc->data; do_crm_log_unlikely(log_level, "queue[%d.%d]: input %s raised by %s(%p.%d)\t(cause=%s)", offset++, data->id, fsa_input2string(data->fsa_input), data->origin, data->data, data->data_type, fsa_cause2string(data->fsa_cause)); } } ha_msg_input_t * copy_ha_msg_input(ha_msg_input_t * orig) { ha_msg_input_t *copy = calloc(1, sizeof(ha_msg_input_t)); CRM_ASSERT(copy != NULL); copy->msg = (orig && orig->msg)? copy_xml(orig->msg) : NULL; copy->xml = get_message_xml(copy->msg, F_CRM_DATA); return copy; } void delete_fsa_input(fsa_data_t * fsa_data) { lrmd_event_data_t *op = NULL; xmlNode *foo = NULL; if (fsa_data == NULL) { return; } crm_trace("About to free %s data", fsa_cause2string(fsa_data->fsa_cause)); if (fsa_data->data != NULL) { switch (fsa_data->data_type) { case fsa_dt_ha_msg: delete_ha_msg_input(fsa_data->data); break; case fsa_dt_xml: foo = fsa_data->data; free_xml(foo); break; case fsa_dt_lrm: op = (lrmd_event_data_t *) fsa_data->data; lrmd_free_event(op); break; case fsa_dt_none: if (fsa_data->data != NULL) { crm_err("Don't know how to free %s data from %s", fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin); crmd_exit(CRM_EX_SOFTWARE); } break; } crm_trace("%s data freed", fsa_cause2string(fsa_data->fsa_cause)); } free(fsa_data); } /* returns the next message */ fsa_data_t * get_message(void) { fsa_data_t *message = g_list_nth_data(fsa_message_queue, 0); fsa_message_queue = g_list_remove(fsa_message_queue, message); crm_trace("Processing input %d", message->id); return message; } void * fsa_typed_data_adv(fsa_data_t * fsa_data, enum fsa_data_type a_type, const char *caller) { void *ret_val = NULL; if (fsa_data == NULL) { crm_err("%s: No FSA data available", caller); } else if (fsa_data->data == NULL) { crm_err("%s: No message data available. Origin: %s", caller, fsa_data->origin); } else if (fsa_data->data_type != a_type) { crm_crit("%s: Message data was the wrong type! %d vs. requested=%d. Origin: %s", caller, fsa_data->data_type, a_type, fsa_data->origin); CRM_ASSERT(fsa_data->data_type == a_type); } else { ret_val = fsa_data->data; } return ret_val; } /* A_MSG_ROUTE */ void do_msg_route(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); route_message(msg_data->fsa_cause, input->msg); } void route_message(enum crmd_fsa_cause cause, xmlNode * input) { ha_msg_input_t fsa_input; enum crmd_fsa_input result = I_NULL; fsa_input.msg = input; CRM_CHECK(cause == C_IPC_MESSAGE || cause == C_HA_MESSAGE, return); /* try passing the buck first */ if (relay_message(input, cause == C_IPC_MESSAGE)) { return; } /* handle locally */ result = handle_message(input, cause); /* done or process later? */ switch (result) { case I_NULL: case I_CIB_OP: case I_ROUTER: case I_NODE_JOIN: case I_JOIN_REQUEST: case I_JOIN_RESULT: break; default: /* Defering local processing of message */ register_fsa_input_later(cause, result, &fsa_input); return; } if (result != I_NULL) { /* add to the front of the queue */ register_fsa_input(cause, result, &fsa_input); } } gboolean relay_message(xmlNode * msg, gboolean originated_locally) { int dest = 1; int is_for_dc = 0; int is_for_dcib = 0; int is_for_te = 0; int is_for_crm = 0; int is_for_cib = 0; int is_local = 0; const char *host_to = crm_element_value(msg, F_CRM_HOST_TO); const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO); const char *sys_from = crm_element_value(msg, F_CRM_SYS_FROM); const char *type = crm_element_value(msg, F_TYPE); const char *task = crm_element_value(msg, F_CRM_TASK); const char *ref = crm_element_value(msg, XML_ATTR_REFERENCE); if (ref == NULL) { ref = "without reference ID"; } if (msg == NULL) { crm_warn("Cannot route empty message"); return TRUE; } else if (pcmk__str_eq(task, CRM_OP_HELLO, pcmk__str_casei)) { /* quietly ignore */ crm_trace("No routing needed for hello message %s", ref); return TRUE; } else if (!pcmk__str_eq(type, T_CRM, pcmk__str_casei)) { crm_warn("Cannot route message %s: Type is '%s' not '" T_CRM "'", ref, (type? type : "missing")); crm_log_xml_warn(msg, "[bad message type]"); return TRUE; } else if (sys_to == NULL) { crm_warn("Cannot route message %s: No subsystem specified", ref); crm_log_xml_warn(msg, "[no subsystem]"); return TRUE; } is_for_dc = (strcasecmp(CRM_SYSTEM_DC, sys_to) == 0); is_for_dcib = (strcasecmp(CRM_SYSTEM_DCIB, sys_to) == 0); is_for_te = (strcasecmp(CRM_SYSTEM_TENGINE, sys_to) == 0); is_for_cib = (strcasecmp(CRM_SYSTEM_CIB, sys_to) == 0); is_for_crm = (strcasecmp(CRM_SYSTEM_CRMD, sys_to) == 0); is_local = 0; if (pcmk__str_empty(host_to)) { if (is_for_dc || is_for_te) { is_local = 0; } else if (is_for_crm) { if (pcmk__strcase_any_of(task, CRM_OP_NODE_INFO, PCMK__CONTROLD_CMD_NODES, NULL)) { /* Node info requests do not specify a host, which is normally * treated as "all hosts", because the whole point is that the * client may not know the local node name. Always handle these * requests locally. */ is_local = 1; } else { is_local = !originated_locally; } } else { is_local = 1; } } else if (pcmk__str_eq(fsa_our_uname, host_to, pcmk__str_casei)) { is_local = 1; } else if (is_for_crm && pcmk__str_eq(task, CRM_OP_LRM_DELETE, pcmk__str_casei)) { xmlNode *msg_data = get_message_xml(msg, F_CRM_DATA); const char *mode = crm_element_value(msg_data, PCMK__XA_MODE); if (pcmk__str_eq(mode, XML_TAG_CIB, pcmk__str_casei)) { // Local delete of an offline node's resource history is_local = 1; } } if (is_for_dc || is_for_dcib || is_for_te) { if (AM_I_DC && is_for_te) { crm_trace("Route message %s locally as transition request", ref); send_msg_via_ipc(msg, sys_to); } else if (AM_I_DC) { crm_trace("Route message %s locally as DC request", ref); return FALSE; // More to be done by caller } else if (originated_locally && !pcmk__strcase_any_of(sys_from, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) { #if SUPPORT_COROSYNC if (is_corosync_cluster()) { dest = text2msg_type(sys_to); } #endif crm_trace("Relay message %s to DC", ref); send_cluster_message(host_to ? crm_get_peer(0, host_to) : NULL, dest, msg, TRUE); } else { /* Neither the TE nor the scheduler should be sending messages * to DCs on other nodes. By definition, if we are no longer the DC, * then the scheduler's or TE's data should be discarded. */ crm_trace("Discard message %s because we are not DC", ref); } } else if (is_local && (is_for_crm || is_for_cib)) { crm_trace("Route message %s locally as controller request", ref); return FALSE; // More to be done by caller } else if (is_local) { crm_trace("Relay message %s locally to %s", ref, (sys_to? sys_to : "unknown client")); crm_log_xml_trace(msg, "[IPC relay]"); send_msg_via_ipc(msg, sys_to); } else { crm_node_t *node_to = NULL; #if SUPPORT_COROSYNC if (is_corosync_cluster()) { dest = text2msg_type(sys_to); if (dest == crm_msg_none || dest > crm_msg_stonith_ng) { dest = crm_msg_crmd; } } #endif if (host_to) { node_to = pcmk__search_cluster_node_cache(0, host_to); if (node_to == NULL) { crm_warn("Cannot route message %s: Unknown node %s", ref, host_to); return TRUE; } crm_trace("Relay message %s to %s", ref, (node_to->uname? node_to->uname : "peer")); } else { crm_trace("Broadcast message %s to all peers", ref); } send_cluster_message(host_to ? node_to : NULL, dest, msg, TRUE); } return TRUE; // No further processing of message is needed } // Return true if field contains a positive integer static bool authorize_version(xmlNode *message_data, const char *field, const char *client_name, const char *ref, const char *uuid) { const char *version = crm_element_value(message_data, field); long long version_num; if ((pcmk__scan_ll(version, &version_num, -1LL) != pcmk_rc_ok) || (version_num < 0LL)) { crm_warn("Rejected IPC hello from %s: '%s' is not a valid protocol %s " CRM_XS " ref=%s uuid=%s", client_name, ((version == NULL)? "" : version), field, (ref? ref : "none"), uuid); return false; } return true; } /*! * \internal * \brief Check whether a client IPC message is acceptable * * If a given client IPC message is a hello, "authorize" it by ensuring it has * valid information such as a protocol version, and return false indicating * that nothing further needs to be done with the message. If the message is not * a hello, just return true to indicate it needs further processing. * * \param[in] client_msg XML of IPC message * \param[in] curr_client If IPC is not proxied, client that sent message * \param[in] proxy_session If IPC is proxied, the session ID * * \return true if message needs further processing, false if it doesn't */ bool controld_authorize_ipc_message(xmlNode *client_msg, pcmk__client_t *curr_client, const char *proxy_session) { xmlNode *message_data = NULL; const char *client_name = NULL; const char *op = crm_element_value(client_msg, F_CRM_TASK); const char *ref = crm_element_value(client_msg, XML_ATTR_REFERENCE); const char *uuid = (curr_client? curr_client->id : proxy_session); if (uuid == NULL) { crm_warn("IPC message from client rejected: No client identifier " CRM_XS " ref=%s", (ref? ref : "none")); goto rejected; } if (!pcmk__str_eq(CRM_OP_HELLO, op, pcmk__str_casei)) { // Only hello messages need to be authorized return true; } message_data = get_message_xml(client_msg, F_CRM_DATA); client_name = crm_element_value(message_data, "client_name"); if (pcmk__str_empty(client_name)) { crm_warn("IPC hello from client rejected: No client name", CRM_XS " ref=%s uuid=%s", (ref? ref : "none"), uuid); goto rejected; } if (!authorize_version(message_data, "major_version", client_name, ref, uuid)) { goto rejected; } if (!authorize_version(message_data, "minor_version", client_name, ref, uuid)) { goto rejected; } crm_trace("Validated IPC hello from client %s", client_name); if (curr_client) { curr_client->userdata = strdup(client_name); } mainloop_set_trigger(fsa_source); return false; rejected: if (curr_client) { qb_ipcs_disconnect(curr_client->ipcs); } return false; } static enum crmd_fsa_input handle_message(xmlNode *msg, enum crmd_fsa_cause cause) { const char *type = NULL; CRM_CHECK(msg != NULL, return I_NULL); type = crm_element_value(msg, F_CRM_MSG_TYPE); if (pcmk__str_eq(type, XML_ATTR_REQUEST, pcmk__str_none)) { return handle_request(msg, cause); } else if (pcmk__str_eq(type, XML_ATTR_RESPONSE, pcmk__str_none)) { handle_response(msg); return I_NULL; } crm_err("Unknown message type: %s", type); return I_NULL; } static enum crmd_fsa_input handle_failcount_op(xmlNode * stored_msg) { const char *rsc = NULL; const char *uname = NULL; const char *op = NULL; char *interval_spec = NULL; guint interval_ms = 0; gboolean is_remote_node = FALSE; xmlNode *xml_op = get_message_xml(stored_msg, F_CRM_DATA); if (xml_op) { xmlNode *xml_rsc = first_named_child(xml_op, XML_CIB_TAG_RESOURCE); xmlNode *xml_attrs = first_named_child(xml_op, XML_TAG_ATTRS); if (xml_rsc) { rsc = ID(xml_rsc); } if (xml_attrs) { op = crm_element_value(xml_attrs, CRM_META "_" XML_RSC_ATTR_CLEAR_OP); crm_element_value_ms(xml_attrs, CRM_META "_" XML_RSC_ATTR_CLEAR_INTERVAL, &interval_ms); } } uname = crm_element_value(xml_op, XML_LRM_ATTR_TARGET); if ((rsc == NULL) || (uname == NULL)) { crm_log_xml_warn(stored_msg, "invalid failcount op"); return I_NULL; } if (crm_element_value(xml_op, XML_LRM_ATTR_ROUTER_NODE)) { is_remote_node = TRUE; } if (interval_ms) { interval_spec = crm_strdup_printf("%ums", interval_ms); } update_attrd_clear_failures(uname, rsc, op, interval_spec, is_remote_node); free(interval_spec); lrm_clear_last_failure(rsc, uname, op, interval_ms); return I_NULL; } static enum crmd_fsa_input handle_lrm_delete(xmlNode *stored_msg) { const char *mode = NULL; xmlNode *msg_data = get_message_xml(stored_msg, F_CRM_DATA); CRM_CHECK(msg_data != NULL, return I_NULL); /* CRM_OP_LRM_DELETE has two distinct modes. The default behavior is to * relay the operation to the affected node, which will unregister the * resource from the local executor, clear the resource's history from the * CIB, and do some bookkeeping in the controller. * * However, if the affected node is offline, the client will specify * mode="cib" which means the controller receiving the operation should * clear the resource's history from the CIB and nothing else. This is used * to clear shutdown locks. */ mode = crm_element_value(msg_data, PCMK__XA_MODE); if ((mode == NULL) || strcmp(mode, XML_TAG_CIB)) { // Relay to affected node crm_xml_add(stored_msg, F_CRM_SYS_TO, CRM_SYSTEM_LRMD); return I_ROUTER; } else { // Delete CIB history locally (compare with do_lrm_delete()) const char *from_sys = NULL; const char *user_name = NULL; const char *rsc_id = NULL; const char *node = NULL; xmlNode *rsc_xml = NULL; int rc = pcmk_rc_ok; rsc_xml = first_named_child(msg_data, XML_CIB_TAG_RESOURCE); CRM_CHECK(rsc_xml != NULL, return I_NULL); rsc_id = ID(rsc_xml); from_sys = crm_element_value(stored_msg, F_CRM_SYS_FROM); node = crm_element_value(msg_data, XML_LRM_ATTR_TARGET); user_name = pcmk__update_acl_user(stored_msg, F_CRM_USER, NULL); crm_debug("Handling " CRM_OP_LRM_DELETE " for %s on %s locally%s%s " "(clearing CIB resource history only)", rsc_id, node, (user_name? " for user " : ""), (user_name? user_name : "")); rc = controld_delete_resource_history(rsc_id, node, user_name, cib_dryrun|cib_sync_call); if (rc == pcmk_rc_ok) { rc = controld_delete_resource_history(rsc_id, node, user_name, crmd_cib_smart_opt()); } //Notify client and tengine.(Only notify tengine if mode = "cib" and CRM_OP_LRM_DELETE.) if (from_sys) { lrmd_event_data_t *op = NULL; const char *from_host = crm_element_value(stored_msg, F_CRM_HOST_FROM); const char *transition; if (strcmp(from_sys, CRM_SYSTEM_TENGINE)) { transition = crm_element_value(msg_data, XML_ATTR_TRANSITION_KEY); } else { transition = crm_element_value(stored_msg, XML_ATTR_TRANSITION_KEY); } crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "local node"), rsc_id, ((rc == pcmk_rc_ok)? "" : " not")); op = lrmd_new_event(rsc_id, CRMD_ACTION_DELETE, 0); op->type = lrmd_event_exec_complete; op->user_data = strdup(transition? transition : FAKE_TE_ID); op->params = pcmk__strkey_table(free, free); g_hash_table_insert(op->params, strdup(XML_ATTR_CRM_VERSION), strdup(CRM_FEATURE_SET)); controld_rc2event(op, rc); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); controld_trigger_delete_refresh(from_sys, rsc_id); } return I_NULL; } } /*! * \brief Handle a CRM_OP_REMOTE_STATE message by updating remote peer cache * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_remote_state(xmlNode *msg) { const char *remote_uname = ID(msg); const char *remote_is_up = crm_element_value(msg, XML_NODE_IN_CLUSTER); crm_node_t *remote_peer; CRM_CHECK(remote_uname && remote_is_up, return I_NULL); remote_peer = crm_remote_peer_get(remote_uname); CRM_CHECK(remote_peer, return I_NULL); pcmk__update_peer_state(__func__, remote_peer, crm_is_true(remote_is_up)? CRM_NODE_MEMBER : CRM_NODE_LOST, 0); return I_NULL; } /*! * \brief Handle a CRM_OP_PING message * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_ping(xmlNode *msg) { const char *value = NULL; xmlNode *ping = NULL; // Build reply ping = create_xml_node(NULL, XML_CRM_TAG_PING); value = crm_element_value(msg, F_CRM_SYS_TO); crm_xml_add(ping, XML_PING_ATTR_SYSFROM, value); // Add controller state value = fsa_state2string(fsa_state); crm_xml_add(ping, XML_PING_ATTR_CRMDSTATE, value); crm_notice("Current ping state: %s", value); // CTS needs this // Add controller health // @TODO maybe do some checks to determine meaningful status crm_xml_add(ping, XML_PING_ATTR_STATUS, "ok"); // Send reply msg = create_reply(msg, ping); free_xml(ping); if (msg) { (void) relay_message(msg, TRUE); free_xml(msg); } // Nothing further to do return I_NULL; } /*! * \brief Handle a PCMK__CONTROLD_CMD_NODES message * * \return Next FSA input */ static enum crmd_fsa_input handle_node_list(xmlNode *request) { GHashTableIter iter; crm_node_t *node = NULL; xmlNode *reply = NULL; xmlNode *reply_data = NULL; // Create message data for reply reply_data = create_xml_node(NULL, XML_CIB_TAG_NODES); g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { xmlNode *xml = create_xml_node(reply_data, XML_CIB_TAG_NODE); crm_xml_add_ll(xml, XML_ATTR_ID, (long long) node->id); // uint32_t crm_xml_add(xml, XML_ATTR_UNAME, node->uname); crm_xml_add(xml, XML_NODE_IN_CLUSTER, node->state); } // Create and send reply reply = create_reply(request, reply_data); free_xml(reply_data); if (reply) { (void) relay_message(reply, TRUE); free_xml(reply); } // Nothing further to do return I_NULL; } /*! * \brief Handle a CRM_OP_NODE_INFO request * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_node_info_request(xmlNode *msg) { const char *value = NULL; crm_node_t *node = NULL; int node_id = 0; xmlNode *reply = NULL; // Build reply reply = create_xml_node(NULL, XML_CIB_TAG_NODE); crm_xml_add(reply, XML_PING_ATTR_SYSFROM, CRM_SYSTEM_CRMD); // Add whether current partition has quorum crm_xml_add_boolean(reply, XML_ATTR_HAVE_QUORUM, fsa_has_quorum); // Check whether client requested node info by ID and/or name crm_element_value_int(msg, XML_ATTR_ID, &node_id); if (node_id < 0) { node_id = 0; } value = crm_element_value(msg, XML_ATTR_UNAME); // Default to local node if none given if ((node_id == 0) && (value == NULL)) { value = fsa_our_uname; } node = pcmk__search_node_caches(node_id, value, CRM_GET_PEER_ANY); if (node) { crm_xml_add_int(reply, XML_ATTR_ID, node->id); crm_xml_add(reply, XML_ATTR_UUID, node->uuid); crm_xml_add(reply, XML_ATTR_UNAME, node->uname); crm_xml_add(reply, XML_NODE_IS_PEER, node->state); crm_xml_add_boolean(reply, XML_NODE_IS_REMOTE, node->flags & crm_remote_node); } // Send reply msg = create_reply(msg, reply); free_xml(reply); if (msg) { (void) relay_message(msg, TRUE); free_xml(msg); } // Nothing further to do return I_NULL; } static void verify_feature_set(xmlNode *msg) { const char *dc_version = crm_element_value(msg, XML_ATTR_CRM_VERSION); if (dc_version == NULL) { /* All we really know is that the DC feature set is older than 3.1.0, * but that's also all that really matters. */ dc_version = "3.0.14"; } if (feature_set_compatible(dc_version, CRM_FEATURE_SET)) { crm_trace("Local feature set (%s) is compatible with DC's (%s)", CRM_FEATURE_SET, dc_version); } else { crm_err("Local feature set (%s) is incompatible with DC's (%s)", CRM_FEATURE_SET, dc_version); // Nothing is likely to improve without administrator involvement controld_set_fsa_input_flags(R_STAYDOWN); crmd_exit(CRM_EX_FATAL); } } // DC gets own shutdown all-clear static enum crmd_fsa_input handle_shutdown_self_ack(xmlNode *stored_msg) { const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM); if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)) { // The expected case -- we initiated own shutdown sequence crm_info("Shutting down controller"); return I_STOP; } if (pcmk__str_eq(host_from, fsa_our_dc, pcmk__str_casei)) { // Must be logic error -- DC confirming its own unrequested shutdown crm_err("Shutting down controller immediately due to " "unexpected shutdown confirmation"); return I_TERMINATE; } if (fsa_state != S_STOPPING) { // Shouldn't happen -- non-DC confirming unrequested shutdown crm_err("Starting new DC election because %s is " "confirming shutdown we did not request", (host_from? host_from : "another node")); return I_ELECTION; } // Shouldn't happen, but we are already stopping anyway crm_debug("Ignoring unexpected shutdown confirmation from %s", (host_from? host_from : "another node")); return I_NULL; } // Non-DC gets shutdown all-clear from DC static enum crmd_fsa_input handle_shutdown_ack(xmlNode *stored_msg) { const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM); if (host_from == NULL) { crm_warn("Ignoring shutdown request without origin specified"); return I_NULL; } if ((fsa_our_dc == NULL) || (strcmp(host_from, fsa_our_dc) == 0)) { if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)) { crm_info("Shutting down controller after confirmation from %s", host_from); } else { crm_err("Shutting down controller after unexpected " "shutdown request from %s", host_from); controld_set_fsa_input_flags(R_STAYDOWN); } return I_STOP; } crm_warn("Ignoring shutdown request from %s because DC is %s", host_from, fsa_our_dc); return I_NULL; } static enum crmd_fsa_input handle_request(xmlNode *stored_msg, enum crmd_fsa_cause cause) { xmlNode *msg = NULL; const char *op = crm_element_value(stored_msg, F_CRM_TASK); /* Optimize this for the DC - it has the most to do */ if (op == NULL) { crm_log_xml_warn(stored_msg, "[request without " F_CRM_TASK "]"); return I_NULL; } if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) { const char *from = crm_element_value(stored_msg, F_CRM_HOST_FROM); crm_node_t *node = pcmk__search_cluster_node_cache(0, from); pcmk__update_peer_expected(__func__, node, CRMD_JOINSTATE_DOWN); if(AM_I_DC == FALSE) { return I_NULL; /* Done */ } } /*========== DC-Only Actions ==========*/ if (AM_I_DC) { if (strcmp(op, CRM_OP_JOIN_ANNOUNCE) == 0) { return I_NODE_JOIN; } else if (strcmp(op, CRM_OP_JOIN_REQUEST) == 0) { return I_JOIN_REQUEST; } else if (strcmp(op, CRM_OP_JOIN_CONFIRM) == 0) { return I_JOIN_RESULT; } else if (strcmp(op, CRM_OP_SHUTDOWN) == 0) { return handle_shutdown_self_ack(stored_msg); } else if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) { // Another controller wants to shut down its node return handle_shutdown_request(stored_msg); } else if (strcmp(op, CRM_OP_REMOTE_STATE) == 0) { /* a remote connection host is letting us know the node state */ return handle_remote_state(stored_msg); } } /*========== common actions ==========*/ if (strcmp(op, CRM_OP_NOVOTE) == 0) { ha_msg_input_t fsa_input; fsa_input.msg = stored_msg; register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input, A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __func__); } else if (strcmp(op, CRM_OP_THROTTLE) == 0) { throttle_update(stored_msg); if (AM_I_DC && transition_graph != NULL) { if (transition_graph->complete == FALSE) { crm_debug("The throttle changed. Trigger a graph."); trigger_graph(); } } return I_NULL; } else if (strcmp(op, CRM_OP_CLEAR_FAILCOUNT) == 0) { return handle_failcount_op(stored_msg); } else if (strcmp(op, CRM_OP_VOTE) == 0) { /* count the vote and decide what to do after that */ ha_msg_input_t fsa_input; fsa_input.msg = stored_msg; register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input, A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __func__); /* Sometimes we _must_ go into S_ELECTION */ if (fsa_state == S_HALT) { crm_debug("Forcing an election from S_HALT"); return I_ELECTION; #if 0 } else if (AM_I_DC) { /* This is the old way of doing things but what is gained? */ return I_ELECTION; #endif } } else if (strcmp(op, CRM_OP_JOIN_OFFER) == 0) { verify_feature_set(stored_msg); crm_debug("Raising I_JOIN_OFFER: join-%s", crm_element_value(stored_msg, F_CRM_JOIN_ID)); return I_JOIN_OFFER; } else if (strcmp(op, CRM_OP_JOIN_ACKNAK) == 0) { crm_debug("Raising I_JOIN_RESULT: join-%s", crm_element_value(stored_msg, F_CRM_JOIN_ID)); return I_JOIN_RESULT; } else if (strcmp(op, CRM_OP_LRM_DELETE) == 0) { return handle_lrm_delete(stored_msg); } else if ((strcmp(op, CRM_OP_LRM_FAIL) == 0) || (strcmp(op, CRM_OP_LRM_REFRESH) == 0) || (strcmp(op, CRM_OP_REPROBE) == 0)) { crm_xml_add(stored_msg, F_CRM_SYS_TO, CRM_SYSTEM_LRMD); return I_ROUTER; } else if (strcmp(op, CRM_OP_NOOP) == 0) { return I_NULL; } else if (strcmp(op, CRM_OP_LOCAL_SHUTDOWN) == 0) { crm_shutdown(SIGTERM); /*return I_SHUTDOWN; */ return I_NULL; } else if (strcmp(op, CRM_OP_PING) == 0) { return handle_ping(stored_msg); } else if (strcmp(op, CRM_OP_NODE_INFO) == 0) { return handle_node_info_request(stored_msg); } else if (strcmp(op, CRM_OP_RM_NODE_CACHE) == 0) { int id = 0; const char *name = NULL; crm_element_value_int(stored_msg, XML_ATTR_ID, &id); name = crm_element_value(stored_msg, XML_ATTR_UNAME); if(cause == C_IPC_MESSAGE) { msg = create_request(CRM_OP_RM_NODE_CACHE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) { crm_err("Could not instruct peers to remove references to node %s/%u", name, id); } else { crm_notice("Instructing peers to remove references to node %s/%u", name, id); } free_xml(msg); } else { reap_crm_member(id, name); /* If we're forgetting this node, also forget any failures to fence * it, so we don't carry that over to any node added later with the * same name. */ st_fail_count_reset(name); } } else if (strcmp(op, CRM_OP_MAINTENANCE_NODES) == 0) { xmlNode *xml = get_message_xml(stored_msg, F_CRM_DATA); remote_ra_process_maintenance_nodes(xml); } else if (strcmp(op, PCMK__CONTROLD_CMD_NODES) == 0) { return handle_node_list(stored_msg); /*========== (NOT_DC)-Only Actions ==========*/ } else if (!AM_I_DC) { if (strcmp(op, CRM_OP_SHUTDOWN) == 0) { return handle_shutdown_ack(stored_msg); } } else { crm_err("Unexpected request (%s) sent to %s", op, AM_I_DC ? "the DC" : "non-DC node"); crm_log_xml_err(stored_msg, "Unexpected"); } return I_NULL; } static void handle_response(xmlNode *stored_msg) { const char *op = crm_element_value(stored_msg, F_CRM_TASK); if (op == NULL) { crm_log_xml_err(stored_msg, "Bad message"); } else if (AM_I_DC && strcmp(op, CRM_OP_PECALC) == 0) { // Check whether scheduler answer been superseded by subsequent request const char *msg_ref = crm_element_value(stored_msg, XML_ATTR_REFERENCE); if (msg_ref == NULL) { crm_err("%s - Ignoring calculation with no reference", op); } else if (pcmk__str_eq(msg_ref, fsa_pe_ref, pcmk__str_casei)) { ha_msg_input_t fsa_input; controld_stop_sched_timer(); fsa_input.msg = stored_msg; register_fsa_input_later(C_IPC_MESSAGE, I_PE_SUCCESS, &fsa_input); } else { crm_info("%s calculation %s is obsolete", op, msg_ref); } } else if (strcmp(op, CRM_OP_VOTE) == 0 || strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0 || strcmp(op, CRM_OP_SHUTDOWN) == 0) { } else { const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM); crm_err("Unexpected response (op=%s, src=%s) sent to the %s", op, host_from, AM_I_DC ? "DC" : "controller"); } } static enum crmd_fsa_input handle_shutdown_request(xmlNode * stored_msg) { /* handle here to avoid potential version issues * where the shutdown message/procedure may have * been changed in later versions. * * This way the DC is always in control of the shutdown */ char *now_s = NULL; const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM); if (host_from == NULL) { /* we're shutting down and the DC */ host_from = fsa_our_uname; } crm_info("Creating shutdown request for %s (state=%s)", host_from, fsa_state2string(fsa_state)); crm_log_xml_trace(stored_msg, "message"); now_s = pcmk__ttoa(time(NULL)); update_attrd(host_from, XML_CIB_ATTR_SHUTDOWN, now_s, NULL, FALSE); free(now_s); /* will be picked up by the TE as long as its running */ return I_NULL; } /* msg is deleted by the time this returns */ extern gboolean process_te_message(xmlNode * msg, xmlNode * xml_data); static void send_msg_via_ipc(xmlNode * msg, const char *sys) { pcmk__client_t *client_channel = pcmk__find_client_by_id(sys); if (crm_element_value(msg, F_CRM_HOST_FROM) == NULL) { crm_xml_add(msg, F_CRM_HOST_FROM, fsa_our_uname); } if (client_channel != NULL) { /* Transient clients such as crmadmin */ pcmk__ipc_send_xml(client_channel, 0, msg, crm_ipc_server_event); } else if (sys != NULL && strcmp(sys, CRM_SYSTEM_TENGINE) == 0) { xmlNode *data = get_message_xml(msg, F_CRM_DATA); process_te_message(msg, data); } else if (sys != NULL && strcmp(sys, CRM_SYSTEM_LRMD) == 0) { fsa_data_t fsa_data; ha_msg_input_t fsa_input; fsa_input.msg = msg; fsa_input.xml = get_message_xml(msg, F_CRM_DATA); fsa_data.id = 0; fsa_data.actions = 0; fsa_data.data = &fsa_input; fsa_data.fsa_input = I_MESSAGE; fsa_data.fsa_cause = C_IPC_MESSAGE; fsa_data.origin = __func__; fsa_data.data_type = fsa_dt_ha_msg; do_lrm_invoke(A_LRM_INVOKE, C_IPC_MESSAGE, fsa_state, I_MESSAGE, &fsa_data); } else if (sys != NULL && crmd_is_proxy_session(sys)) { crmd_proxy_send(sys, msg); } else { crm_debug("Unknown Sub-system (%s)... discarding message.", crm_str(sys)); } } void delete_ha_msg_input(ha_msg_input_t * orig) { if (orig == NULL) { return; } free_xml(orig->msg); free(orig); } /*! * \internal * \brief Notify the DC of a remote node state change * * \param[in] node_name Node's name * \param[in] node_up TRUE if node is up, FALSE if down */ void send_remote_state_message(const char *node_name, gboolean node_up) { /* If we don't have a DC, or the message fails, we have a failsafe: * the DC will eventually pick up the change via the CIB node state. * The message allows it to happen sooner if possible. */ if (fsa_our_dc) { xmlNode *msg = create_request(CRM_OP_REMOTE_STATE, NULL, fsa_our_dc, CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL); - crm_info("Notifying DC %s of pacemaker_remote node %s %s", + crm_info("Notifying DC %s of Pacemaker Remote node %s %s", fsa_our_dc, node_name, (node_up? "coming up" : "going down")); crm_xml_add(msg, XML_ATTR_ID, node_name); crm_xml_add_boolean(msg, XML_NODE_IN_CLUSTER, node_up); send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, msg, TRUE); free_xml(msg); } else { - crm_debug("No DC to notify of pacemaker_remote node %s %s", + crm_debug("No DC to notify of Pacemaker Remote node %s %s", node_name, (node_up? "coming up" : "going down")); } } diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c index 8faeead095..2b0920f7cd 100644 --- a/daemons/controld/controld_remote_ra.c +++ b/daemons/controld/controld_remote_ra.c @@ -1,1309 +1,1310 @@ /* * Copyright 2013-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #define REMOTE_LRMD_RA "remote" /* The max start timeout before cmd retry */ #define MAX_START_TIMEOUT_MS 10000 typedef struct remote_ra_cmd_s { /*! the local node the cmd is issued from */ char *owner; /*! the remote node the cmd is executed on */ char *rsc_id; /*! the action to execute */ char *action; /*! some string the client wants us to give it back */ char *userdata; char *exit_reason; // descriptive text on error /*! start delay in ms */ int start_delay; /*! timer id used for start delay. */ int delay_id; /*! timeout in ms for cmd */ int timeout; int remaining_timeout; /*! recurring interval in ms */ guint interval_ms; /*! interval timer id */ int interval_id; int reported_success; int monitor_timeout_id; int takeover_timeout_id; /*! action parameters */ lrmd_key_value_t *params; /*! executed rc */ int rc; int op_status; int call_id; time_t start_time; gboolean cancel; } remote_ra_cmd_t; enum remote_migration_status { expect_takeover = 1, takeover_complete, }; typedef struct remote_ra_data_s { crm_trigger_t *work; remote_ra_cmd_t *cur_cmd; GList *cmds; GList *recurring_cmds; enum remote_migration_status migrate_status; gboolean active; /* Maintenance mode is difficult to determine from the controller's context, * so we have it signalled back with the transition from the scheduler. */ gboolean is_maintenance; /* Similar for whether we are controlling a guest node or remote node. * Fortunately there is a meta-attribute in the transition already and * as the situation doesn't change over time we can use the * resource start for noting down the information for later use when * the attributes aren't at hand. */ gboolean controlling_guest; } remote_ra_data_t; static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms); static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd); static GList *fail_all_monitor_cmds(GList * list); static void free_cmd(gpointer user_data) { remote_ra_cmd_t *cmd = user_data; if (!cmd) { return; } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->interval_id) { g_source_remove(cmd->interval_id); } if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); } if (cmd->takeover_timeout_id) { g_source_remove(cmd->takeover_timeout_id); } free(cmd->owner); free(cmd->rsc_id); free(cmd->action); free(cmd->userdata); free(cmd->exit_reason); lrmd_key_value_freeall(cmd->params); free(cmd); } static int generate_callid(void) { static int remote_ra_callid = 0; remote_ra_callid++; if (remote_ra_callid <= 0) { remote_ra_callid = 1; } return remote_ra_callid; } static gboolean recurring_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->interval_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; ra_data->recurring_cmds = g_list_remove(ra_data->recurring_cmds, cmd); ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); } return FALSE; } static gboolean start_delay_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->delay_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; mainloop_set_trigger(ra_data->work); } return FALSE; } /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node joining * * \param[in] node_name Name of newly integrated pacemaker_remote node */ static void remote_node_up(const char *node_name) { int call_opt, call_id = 0; xmlNode *update, *state; crm_node_t *node; enum controld_section_e section = controld_section_all; CRM_CHECK(node_name != NULL, return); - crm_info("Announcing pacemaker_remote node %s", node_name); + crm_info("Announcing Pacemaker Remote node %s", node_name); /* Clear node's entire state (resource history and transient attributes) * other than shutdown locks. The transient attributes should and normally * will be cleared when the node leaves, but since remote node state has a * number of corner cases, clear them here as well, to be sure. */ call_opt = crmd_cib_smart_opt(); if (controld_shutdown_lock_enabled) { section = controld_section_all_unlocked; } /* Purge node from attrd's memory */ update_attrd_remote_node_removed(node_name, NULL); controld_delete_node_state(node_name, section, call_opt); /* Clear node's probed attribute */ update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE); /* Ensure node is in the remote peer cache with member status */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0); /* pacemaker_remote nodes don't participate in the membership layer, * so cluster nodes don't automatically get notified when they come and go. * We send a cluster message to the DC, and update the CIB node state entry, * so the DC will get it sooner (via message) or later (via CIB refresh), * and any other interested parties can query the CIB. */ send_remote_state_message(node_name, TRUE); update = create_xml_node(NULL, XML_CIB_TAG_STATUS); state = create_node_state_update(node, node_update_cluster, update, __func__); /* Clear the XML_NODE_IS_FENCED flag in the node state. If the node ever * needs to be fenced, this flag will allow various actions to determine * whether the fencing has happened yet. */ crm_xml_add(state, XML_NODE_IS_FENCED, "0"); /* TODO: If the remote connection drops, and this (async) CIB update either * failed or has not yet completed, later actions could mistakenly think the * node has already been fenced (if the XML_NODE_IS_FENCED attribute was * previously set, because it won't have been cleared). This could prevent * actual fencing or allow recurring monitor failures to be cleared too * soon. Ideally, we wouldn't rely on the CIB for the fenced status. */ fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL); if (call_id < 0) { crm_perror(LOG_WARNING, "%s CIB node state setup", node_name); } free_xml(update); } enum down_opts { DOWN_KEEP_LRM, DOWN_ERASE_LRM }; /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node leaving * * \param[in] node_name Name of lost node * \param[in] opts Whether to keep or erase LRM history */ static void remote_node_down(const char *node_name, const enum down_opts opts) { xmlNode *update; int call_id = 0; int call_opt = crmd_cib_smart_opt(); crm_node_t *node; /* Purge node from attrd's memory */ update_attrd_remote_node_removed(node_name, NULL); /* Normally, only node attributes should be erased, and the resource history * should be kept until the node comes back up. However, after a successful * fence, we want to clear the history as well, so we don't think resources * are still running on the node. */ if (opts == DOWN_ERASE_LRM) { controld_delete_node_state(node_name, controld_section_all, call_opt); } else { controld_delete_node_state(node_name, controld_section_attrs, call_opt); } /* Ensure node is in the remote peer cache with lost state */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_LOST, 0); /* Notify DC */ send_remote_state_message(node_name, FALSE); /* Update CIB node state */ update = create_xml_node(NULL, XML_CIB_TAG_STATUS); create_node_state_update(node, node_update_cluster, update, __func__); fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL); if (call_id < 0) { crm_perror(LOG_ERR, "%s CIB node state update", node_name); } free_xml(update); } /*! * \internal * \brief Handle effects of a remote RA command on node state * * \param[in] cmd Completed remote RA command */ static void check_remote_node_state(remote_ra_cmd_t *cmd) { /* Only successful actions can change node state */ if (cmd->rc != PCMK_OCF_OK) { return; } if (pcmk__str_eq(cmd->action, "start", pcmk__str_casei)) { remote_node_up(cmd->rsc_id); } else if (pcmk__str_eq(cmd->action, "migrate_from", pcmk__str_casei)) { /* After a successful migration, we don't need to do remote_node_up() * because the DC already knows the node is up, and we don't want to * clear LRM history etc. We do need to add the remote node to this * host's remote peer cache, because (unless it happens to be DC) * it hasn't been tracking the remote node, and other code relies on * the cache to distinguish remote nodes from unseen cluster nodes. */ crm_node_t *node = crm_remote_peer_get(cmd->rsc_id); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0); } else if (pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { lrm_state_t *lrm_state = lrm_state_find(cmd->rsc_id); remote_ra_data_t *ra_data = lrm_state? lrm_state->remote_ra_data : NULL; if (ra_data) { if (ra_data->migrate_status != takeover_complete) { /* Stop means down if we didn't successfully migrate elsewhere */ remote_node_down(cmd->rsc_id, DOWN_KEEP_LRM); } else if (AM_I_DC == FALSE) { /* Only the connection host and DC track node state, * so if the connection migrated elsewhere and we aren't DC, * un-cache the node, so we don't have stale info */ crm_remote_peer_cache_remove(cmd->rsc_id); } } } /* We don't do anything for successful monitors, which is correct for * routine recurring monitors, and for monitors on nodes where the * connection isn't supposed to be (the cluster will stop the connection in * that case). However, if the initial probe finds the connection already * active on the node where we want it, we probably should do * remote_node_up(). Unfortunately, we can't distinguish that case here. * Given that connections have to be initiated by the cluster, the chance of * that should be close to zero. */ } static void report_remote_ra_result(remote_ra_cmd_t * cmd) { lrmd_event_data_t op = { 0, }; check_remote_node_state(cmd); op.type = lrmd_event_exec_complete; op.rsc_id = cmd->rsc_id; op.op_type = cmd->action; op.user_data = cmd->userdata; op.exit_reason = cmd->exit_reason; op.timeout = cmd->timeout; op.interval_ms = cmd->interval_ms; op.rc = cmd->rc; op.op_status = cmd->op_status; op.t_run = (unsigned int) cmd->start_time; op.t_rcchange = (unsigned int) cmd->start_time; if (cmd->reported_success && cmd->rc != PCMK_OCF_OK) { op.t_rcchange = (unsigned int) time(NULL); /* This edge case will likely never ever occur, but if it does the * result is that a failure will not be processed correctly. This is only * remotely possible because we are able to detect a connection resource's tcp * connection has failed at any moment after start has completed. The actual * recurring operation is just a connectivity ping. * * basically, we are not guaranteed that the first successful monitor op and * a subsequent failed monitor op will not occur in the same timestamp. We have to * make it look like the operations occurred at separate times though. */ if (op.t_rcchange == op.t_run) { op.t_rcchange++; } } if (cmd->params) { lrmd_key_value_t *tmp; op.params = pcmk__strkey_table(free, free); for (tmp = cmd->params; tmp; tmp = tmp->next) { g_hash_table_insert(op.params, strdup(tmp->key), strdup(tmp->value)); } } op.call_id = cmd->call_id; op.remote_nodename = cmd->owner; lrm_op_callback(&op); if (op.params) { g_hash_table_destroy(op.params); } } static void update_remaining_timeout(remote_ra_cmd_t * cmd) { cmd->remaining_timeout = ((cmd->timeout / 1000) - (time(NULL) - cmd->start_time)) * 1000; } static gboolean retry_start_cmd_cb(gpointer data) { lrm_state_t *lrm_state = data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd = NULL; - int rc = -1; + int rc = pcmk_rc_error; if (!ra_data || !ra_data->cur_cmd) { return FALSE; } cmd = ra_data->cur_cmd; if (!pcmk__strcase_any_of(cmd->action, "start", "migrate_from", NULL)) { return FALSE; } update_remaining_timeout(cmd); if (cmd->remaining_timeout > 0) { rc = handle_remote_ra_start(lrm_state, cmd, cmd->remaining_timeout); } - if (rc != 0) { + if (rc != pcmk_rc_ok) { cmd->rc = PCMK_OCF_UNKNOWN_ERROR; cmd->op_status = PCMK_LRM_OP_ERROR; report_remote_ra_result(cmd); if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } ra_data->cur_cmd = NULL; free_cmd(cmd); } else { /* wait for connection event */ } return FALSE; } static gboolean connection_takeover_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; crm_info("takeover event timed out for node %s", cmd->rsc_id); cmd->takeover_timeout_id = 0; lrm_state = lrm_state_find(cmd->rsc_id); handle_remote_ra_stop(lrm_state, cmd); free_cmd(cmd); return FALSE; } static gboolean monitor_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; lrm_state = lrm_state_find(cmd->rsc_id); crm_info("Timed out waiting for remote poke response from %s%s", cmd->rsc_id, (lrm_state? "" : " (no LRM state)")); cmd->monitor_timeout_id = 0; cmd->op_status = PCMK_LRM_OP_TIMEOUT; cmd->rc = PCMK_OCF_UNKNOWN_ERROR; if (lrm_state && lrm_state->remote_ra_data) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (ra_data->cur_cmd == cmd) { ra_data->cur_cmd = NULL; } if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } } report_remote_ra_result(cmd); free_cmd(cmd); if(lrm_state) { lrm_state_disconnect(lrm_state); } return FALSE; } static void synthesize_lrmd_success(lrm_state_t *lrm_state, const char *rsc_id, const char *op_type) { lrmd_event_data_t op = { 0, }; if (lrm_state == NULL) { /* if lrm_state not given assume local */ lrm_state = lrm_state_find(fsa_our_uname); } CRM_ASSERT(lrm_state != NULL); op.type = lrmd_event_exec_complete; op.rsc_id = rsc_id; op.op_type = op_type; op.rc = PCMK_OCF_OK; op.op_status = PCMK_LRM_OP_DONE; op.t_run = (unsigned int) time(NULL); op.t_rcchange = op.t_run; op.call_id = generate_callid(); process_lrm_event(lrm_state, &op, NULL, NULL); } void remote_lrm_op_callback(lrmd_event_data_t * op) { gboolean cmd_handled = FALSE; lrm_state_t *lrm_state = NULL; remote_ra_data_t *ra_data = NULL; remote_ra_cmd_t *cmd = NULL; crm_debug("Processing '%s%s%s' event on remote connection to %s: %s " "(%d) status=%s (%d)", (op->op_type? op->op_type : ""), (op->op_type? " " : ""), lrmd_event_type2str(op->type), op->remote_nodename, services_ocf_exitcode_str(op->rc), op->rc, services_lrm_status_str(op->op_status), op->op_status); lrm_state = lrm_state_find(op->remote_nodename); if (!lrm_state || !lrm_state->remote_ra_data) { crm_debug("No state information found for remote connection event"); return; } ra_data = lrm_state->remote_ra_data; if (op->type == lrmd_event_new_client) { // Another client has connected to the remote daemon if (ra_data->migrate_status == expect_takeover) { // Great, we knew this was coming ra_data->migrate_status = takeover_complete; } else { - crm_err("Unexpected pacemaker_remote client takeover for %s. Disconnecting", op->remote_nodename); + crm_err("Disconnecting from Pacemaker Remote node %s due to " + "unexpected client takeover", op->remote_nodename); /* In this case, lrmd_tls_connection_destroy() will be called under the control of mainloop. */ /* Do not free lrm_state->conn yet. */ /* It'll be freed in the following stop action. */ lrm_state_disconnect_only(lrm_state); } return; } /* filter all EXEC events up */ if (op->type == lrmd_event_exec_complete) { if (ra_data->migrate_status == takeover_complete) { crm_debug("ignoring event, this connection is taken over by another node"); } else { lrm_op_callback(op); } return; } if ((op->type == lrmd_event_disconnect) && (ra_data->cur_cmd == NULL)) { if (ra_data->active == FALSE) { crm_debug("Disconnection from Pacemaker Remote node %s complete", lrm_state->node_name); } else if (!remote_ra_is_in_maintenance(lrm_state)) { crm_err("Lost connection to Pacemaker Remote node %s", lrm_state->node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } else { crm_notice("Unmanaged Pacemaker Remote node %s disconnected", lrm_state->node_name); /* Do roughly what a 'stop' on the remote-resource would do */ handle_remote_ra_stop(lrm_state, NULL); remote_node_down(lrm_state->node_name, DOWN_KEEP_LRM); /* now fake the reply of a successful 'stop' */ synthesize_lrmd_success(NULL, lrm_state->node_name, "stop"); } return; } if (!ra_data->cur_cmd) { crm_debug("no event to match"); return; } cmd = ra_data->cur_cmd; /* Start actions and migrate from actions complete after connection * comes back to us. */ if (op->type == lrmd_event_connect && pcmk__strcase_any_of(cmd->action, "start", "migrate_from", NULL)) { if (op->connection_rc < 0) { update_remaining_timeout(cmd); if (op->connection_rc == -ENOKEY) { // Hard error, don't retry cmd->op_status = PCMK_LRM_OP_ERROR; cmd->rc = PCMK_OCF_INVALID_PARAM; cmd->exit_reason = strdup("Authentication key not readable"); } else if (cmd->remaining_timeout > 3000) { crm_trace("rescheduling start, remaining timeout %d", cmd->remaining_timeout); g_timeout_add(1000, retry_start_cmd_cb, lrm_state); return; } else { crm_trace("can't reschedule start, remaining timeout too small %d", cmd->remaining_timeout); cmd->op_status = PCMK_LRM_OP_TIMEOUT; cmd->rc = PCMK_OCF_UNKNOWN_ERROR; } } else { lrm_state_reset_tables(lrm_state, TRUE); cmd->rc = PCMK_OCF_OK; cmd->op_status = PCMK_LRM_OP_DONE; ra_data->active = TRUE; } crm_debug("Remote connection event matched %s action", cmd->action); report_remote_ra_result(cmd); cmd_handled = TRUE; } else if (op->type == lrmd_event_poke && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); cmd->monitor_timeout_id = 0; } /* Only report success the first time, after that only worry about failures. * For this function, if we get the poke pack, it is always a success. Pokes * only fail if the send fails, or the response times out. */ if (!cmd->reported_success) { cmd->rc = PCMK_OCF_OK; cmd->op_status = PCMK_LRM_OP_DONE; report_remote_ra_result(cmd); cmd->reported_success = 1; } crm_debug("Remote poke event matched %s action", cmd->action); /* success, keep rescheduling if interval is present. */ if (cmd->interval_ms && (cmd->cancel == FALSE)) { ra_data->recurring_cmds = g_list_append(ra_data->recurring_cmds, cmd); cmd->interval_id = g_timeout_add(cmd->interval_ms, recurring_helper, cmd); cmd = NULL; /* prevent free */ } cmd_handled = TRUE; } else if (op->type == lrmd_event_disconnect && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { if (ra_data->active == TRUE && (cmd->cancel == FALSE)) { cmd->rc = PCMK_OCF_UNKNOWN_ERROR; cmd->op_status = PCMK_LRM_OP_ERROR; report_remote_ra_result(cmd); crm_err("Remote connection to %s unexpectedly dropped during monitor", lrm_state->node_name); } cmd_handled = TRUE; } else if (op->type == lrmd_event_new_client && pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { handle_remote_ra_stop(lrm_state, cmd); cmd_handled = TRUE; } else { crm_debug("Event did not match %s action", ra_data->cur_cmd->action); } if (cmd_handled) { ra_data->cur_cmd = NULL; if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } free_cmd(cmd); } } static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd) { remote_ra_data_t *ra_data = NULL; CRM_ASSERT(lrm_state); ra_data = lrm_state->remote_ra_data; if (ra_data->migrate_status != takeover_complete) { /* delete pending ops when ever the remote connection is intentionally stopped */ g_hash_table_remove_all(lrm_state->pending_ops); } else { /* we no longer hold the history if this connection has been migrated, * however, we keep metadata cache for future use */ lrm_state_reset_tables(lrm_state, FALSE); } ra_data->active = FALSE; lrm_state_disconnect(lrm_state); if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } ra_data->cmds = NULL; ra_data->recurring_cmds = NULL; ra_data->cur_cmd = NULL; if (cmd) { cmd->rc = PCMK_OCF_OK; cmd->op_status = PCMK_LRM_OP_DONE; report_remote_ra_result(cmd); } } +// \return Standard Pacemaker return code static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms) { const char *server = NULL; lrmd_key_value_t *tmp = NULL; int port = 0; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; int timeout_used = timeout_ms > MAX_START_TIMEOUT_MS ? MAX_START_TIMEOUT_MS : timeout_ms; for (tmp = cmd->params; tmp; tmp = tmp->next) { if (pcmk__strcase_any_of(tmp->key, XML_RSC_ATTR_REMOTE_RA_ADDR, XML_RSC_ATTR_REMOTE_RA_SERVER, NULL)) { server = tmp->value; } else if (pcmk__str_eq(tmp->key, XML_RSC_ATTR_REMOTE_RA_PORT, pcmk__str_casei)) { port = atoi(tmp->value); } else if (pcmk__str_eq(tmp->key, CRM_META "_" XML_RSC_ATTR_CONTAINER, pcmk__str_casei)) { ra_data->controlling_guest = TRUE; } } - return lrm_state_remote_connect_async(lrm_state, server, port, timeout_used); + return controld_connect_remote_executor(lrm_state, server, port, + timeout_used); } static gboolean handle_remote_ra_exec(gpointer user_data) { int rc = 0; lrm_state_t *lrm_state = user_data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd; GList *first = NULL; if (ra_data->cur_cmd) { /* still waiting on previous cmd */ return TRUE; } while (ra_data->cmds) { first = ra_data->cmds; cmd = first->data; if (cmd->delay_id) { /* still waiting for start delay timer to trip */ return TRUE; } ra_data->cmds = g_list_remove_link(ra_data->cmds, first); g_list_free_1(first); if (!strcmp(cmd->action, "start") || !strcmp(cmd->action, "migrate_from")) { ra_data->migrate_status = 0; - rc = handle_remote_ra_start(lrm_state, cmd, cmd->timeout); - if (rc == 0) { + if (handle_remote_ra_start(lrm_state, cmd, + cmd->timeout) == pcmk_rc_ok) { /* take care of this later when we get async connection result */ crm_debug("Initiated async remote connection, %s action will complete after connect event", cmd->action); ra_data->cur_cmd = cmd; return TRUE; } else { - crm_debug("Could not initiate remote connection for %s action", - cmd->action); cmd->rc = PCMK_OCF_UNKNOWN_ERROR; cmd->op_status = PCMK_LRM_OP_ERROR; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, "monitor")) { if (lrm_state_is_connected(lrm_state) == TRUE) { rc = lrm_state_poke_connection(lrm_state); if (rc < 0) { cmd->rc = PCMK_OCF_UNKNOWN_ERROR; cmd->op_status = PCMK_LRM_OP_ERROR; } } else { rc = -1; cmd->op_status = PCMK_LRM_OP_DONE; cmd->rc = PCMK_OCF_NOT_RUNNING; } if (rc == 0) { crm_debug("Poked Pacemaker Remote at node %s, waiting for async response", cmd->rsc_id); ra_data->cur_cmd = cmd; cmd->monitor_timeout_id = g_timeout_add(cmd->timeout, monitor_timeout_cb, cmd); return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, "stop")) { if (ra_data->migrate_status == expect_takeover) { /* briefly wait on stop for the takeover event to occur. If the * takeover event does not occur during the wait period, that's fine. * It just means that the remote-node's lrm_status section is going to get * cleared which will require all the resources running in the remote-node * to be explicitly re-detected via probe actions. If the takeover does occur * successfully, then we can leave the status section intact. */ cmd->takeover_timeout_id = g_timeout_add((cmd->timeout/2), connection_takeover_timeout_cb, cmd); ra_data->cur_cmd = cmd; return TRUE; } handle_remote_ra_stop(lrm_state, cmd); } else if (!strcmp(cmd->action, "migrate_to")) { ra_data->migrate_status = expect_takeover; cmd->rc = PCMK_OCF_OK; cmd->op_status = PCMK_LRM_OP_DONE; report_remote_ra_result(cmd); } else if (pcmk__str_any_of(cmd->action, CRMD_ACTION_RELOAD, CRMD_ACTION_RELOAD_AGENT, NULL)) { /* Currently the only reloadable parameter is reconnect_interval, * which is only used by the scheduler via the CIB, so reloads are a * no-op. * * @COMPAT DC <2.1.0: We only need to check for "reload" in case * we're in a rolling upgrade with a DC scheduling "reload" instead * of "reload-agent". An OCF 1.1 "reload" would be a no-op anyway, * so this would work for that purpose as well. */ cmd->rc = PCMK_OCF_OK; cmd->op_status = PCMK_LRM_OP_DONE; report_remote_ra_result(cmd); } free_cmd(cmd); } return TRUE; } static void remote_ra_data_init(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = NULL; if (lrm_state->remote_ra_data) { return; } ra_data = calloc(1, sizeof(remote_ra_data_t)); ra_data->work = mainloop_add_trigger(G_PRIORITY_HIGH, handle_remote_ra_exec, lrm_state); lrm_state->remote_ra_data = ra_data; } void remote_ra_cleanup(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (!ra_data) { return; } if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } mainloop_destroy_trigger(ra_data->work); free(ra_data); lrm_state->remote_ra_data = NULL; } gboolean is_remote_lrmd_ra(const char *agent, const char *provider, const char *id) { if (agent && provider && !strcmp(agent, REMOTE_LRMD_RA) && !strcmp(provider, "pacemaker")) { return TRUE; } if (id && lrm_state_find(id) && !pcmk__str_eq(id, fsa_our_uname, pcmk__str_casei)) { return TRUE; } return FALSE; } lrmd_rsc_info_t * remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id) { lrmd_rsc_info_t *info = NULL; if ((lrm_state_find(rsc_id))) { info = calloc(1, sizeof(lrmd_rsc_info_t)); info->id = strdup(rsc_id); info->type = strdup(REMOTE_LRMD_RA); info->standard = strdup(PCMK_RESOURCE_CLASS_OCF); info->provider = strdup("pacemaker"); } return info; } static gboolean is_remote_ra_supported_action(const char *action) { return pcmk__str_any_of(action, CRMD_ACTION_START, CRMD_ACTION_STOP, CRMD_ACTION_STATUS, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, CRMD_ACTION_RELOAD_AGENT, CRMD_ACTION_RELOAD, NULL); } static GList * fail_all_monitor_cmds(GList * list) { GList *rm_list = NULL; remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms > 0) && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { rm_list = g_list_append(rm_list, cmd); } } for (gIter = rm_list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; cmd->rc = PCMK_OCF_UNKNOWN_ERROR; cmd->op_status = PCMK_LRM_OP_ERROR; crm_trace("Pre-emptively failing %s %s (interval=%u, %s)", cmd->action, cmd->rsc_id, cmd->interval_ms, cmd->userdata); report_remote_ra_result(cmd); list = g_list_remove(list, cmd); free_cmd(cmd); } /* frees only the list data, not the cmds */ g_list_free(rm_list); return list; } static GList * remove_cmd(GList * list, const char *action, guint interval_ms) { remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, action, pcmk__str_casei)) { break; } cmd = NULL; } if (cmd) { list = g_list_remove(list, cmd); free_cmd(cmd); } return list; } int remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms) { lrm_state_t *connection_rsc = NULL; remote_ra_data_t *ra_data = NULL; connection_rsc = lrm_state_find(rsc_id); if (!connection_rsc || !connection_rsc->remote_ra_data) { return -EINVAL; } ra_data = connection_rsc->remote_ra_data; ra_data->cmds = remove_cmd(ra_data->cmds, action, interval_ms); ra_data->recurring_cmds = remove_cmd(ra_data->recurring_cmds, action, interval_ms); if (ra_data->cur_cmd && (ra_data->cur_cmd->interval_ms == interval_ms) && (pcmk__str_eq(ra_data->cur_cmd->action, action, pcmk__str_casei))) { ra_data->cur_cmd->cancel = TRUE; } return 0; } static remote_ra_cmd_t * handle_dup_monitor(remote_ra_data_t *ra_data, guint interval_ms, const char *userdata) { GList *gIter = NULL; remote_ra_cmd_t *cmd = NULL; /* there are 3 places a potential duplicate monitor operation * could exist. * 1. recurring_cmds list. where the op is waiting for its next interval * 2. cmds list, where the op is queued to get executed immediately * 3. cur_cmd, which means the monitor op is in flight right now. */ if (interval_ms == 0) { return NULL; } if (ra_data->cur_cmd && ra_data->cur_cmd->cancel == FALSE && (ra_data->cur_cmd->interval_ms == interval_ms) && pcmk__str_eq(ra_data->cur_cmd->action, "monitor", pcmk__str_casei)) { cmd = ra_data->cur_cmd; goto handle_dup; } for (gIter = ra_data->recurring_cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { goto handle_dup; } } for (gIter = ra_data->cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { goto handle_dup; } } return NULL; handle_dup: crm_trace("merging duplicate monitor cmd " PCMK__OP_FMT, cmd->rsc_id, "monitor", interval_ms); /* update the userdata */ if (userdata) { free(cmd->userdata); cmd->userdata = strdup(userdata); } /* if we've already reported success, generate a new call id */ if (cmd->reported_success) { cmd->start_time = time(NULL); cmd->call_id = generate_callid(); cmd->reported_success = 0; } /* if we have an interval_id set, that means we are in the process of * waiting for this cmd's next interval. instead of waiting, cancel * the timer and execute the action immediately */ if (cmd->interval_id) { g_source_remove(cmd->interval_id); cmd->interval_id = 0; recurring_helper(cmd); } return cmd; } int remote_ra_exec(lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout, /* ms */ int start_delay, /* ms */ lrmd_key_value_t * params) { int rc = 0; lrm_state_t *connection_rsc = NULL; remote_ra_cmd_t *cmd = NULL; remote_ra_data_t *ra_data = NULL; if (is_remote_ra_supported_action(action) == FALSE) { rc = -EINVAL; goto exec_done; } connection_rsc = lrm_state_find(rsc_id); if (!connection_rsc) { rc = -EINVAL; goto exec_done; } remote_ra_data_init(connection_rsc); ra_data = connection_rsc->remote_ra_data; cmd = handle_dup_monitor(ra_data, interval_ms, userdata); if (cmd) { rc = cmd->call_id; goto exec_done; } cmd = calloc(1, sizeof(remote_ra_cmd_t)); cmd->owner = strdup(lrm_state->node_name); cmd->rsc_id = strdup(rsc_id); cmd->action = strdup(action); cmd->userdata = strdup(userdata); cmd->interval_ms = interval_ms; cmd->timeout = timeout; cmd->start_delay = start_delay; cmd->params = params; cmd->start_time = time(NULL); cmd->call_id = generate_callid(); if (cmd->start_delay) { cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd); } ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); return cmd->call_id; exec_done: lrmd_key_value_freeall(params); return rc; } /*! * \internal * \brief Immediately fail all monitors of a remote node, if proxied here * * \param[in] node_name Name of pacemaker_remote node */ void remote_ra_fail(const char *node_name) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (lrm_state && lrm_state_is_connected(lrm_state)) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; - crm_info("Failing monitors on pacemaker_remote node %s", node_name); + crm_info("Failing monitors on Pacemaker Remote node %s", node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } } /* A guest node fencing implied by host fencing looks like: * * * * * * * */ #define XPATH_PSEUDO_FENCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \ "[@" XML_LRM_ATTR_TASK "='stonith']/" XML_GRAPH_TAG_DOWNED \ "/" XML_CIB_TAG_NODE /*! * \internal * \brief Check a pseudo-action for Pacemaker Remote node side effects * * \param[in] xml XML of pseudo-action to check */ void remote_ra_process_pseudo(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_FENCE); if (numXpathResults(search) == 1) { xmlNode *result = getXpathResult(search, 0); /* Normally, we handle the necessary side effects of a guest node stop * action when reporting the remote agent's result. However, if the stop * is implied due to fencing, it will be a fencing pseudo-event, and * there won't be a result to report. Handle that case here. * * This will result in a duplicate call to remote_node_down() if the * guest stop was real instead of implied, but that shouldn't hurt. * * There is still one corner case that isn't handled: if a guest node * isn't running any resources when its host is fenced, it will appear * to be cleanly stopped, so there will be no pseudo-fence, and our * peer cache state will be incorrect unless and until the guest is * recovered. */ if (result) { const char *remote = ID(result); if (remote) { remote_node_down(remote, DOWN_ERASE_LRM); } } } freeXpathObject(search); } static void remote_ra_maintenance(lrm_state_t * lrm_state, gboolean maintenance) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; xmlNode *update, *state; int call_opt, call_id = 0; crm_node_t *node; call_opt = crmd_cib_smart_opt(); node = crm_remote_peer_get(lrm_state->node_name); CRM_CHECK(node != NULL, return); update = create_xml_node(NULL, XML_CIB_TAG_STATUS); state = create_node_state_update(node, node_update_none, update, __func__); crm_xml_add(state, XML_NODE_IS_MAINTENANCE, maintenance?"1":"0"); fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL); if (call_id < 0) { crm_perror(LOG_WARNING, "%s CIB node state update failed", lrm_state->node_name); } else { /* TODO: still not 100% sure that async update will succeed ... */ ra_data->is_maintenance = maintenance; } free_xml(update); } #define XPATH_PSEUDO_MAINTENANCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \ "[@" XML_LRM_ATTR_TASK "='" CRM_OP_MAINTENANCE_NODES "']/" \ XML_GRAPH_TAG_MAINTENANCE /*! * \internal * \brief Check a pseudo-action holding updates for maintenance state * * \param[in] xml XML of pseudo-action to check */ void remote_ra_process_maintenance_nodes(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_MAINTENANCE); if (numXpathResults(search) == 1) { xmlNode *node; int cnt = 0, cnt_remote = 0; for (node = first_named_child(getXpathResult(search, 0), XML_CIB_TAG_NODE); node != NULL; node = pcmk__xml_next(node)) { lrm_state_t *lrm_state = lrm_state_find(ID(node)); cnt++; if (lrm_state && lrm_state->remote_ra_data && ((remote_ra_data_t *) lrm_state->remote_ra_data)->active) { int is_maint; cnt_remote++; pcmk__scan_min_int(crm_element_value(node, XML_NODE_IS_MAINTENANCE), &is_maint, 0); remote_ra_maintenance(lrm_state, is_maint); } } crm_trace("Action holds %d nodes (%d remotes found) " "adjusting maintenance-mode", cnt, cnt_remote); } freeXpathObject(search); } gboolean remote_ra_is_in_maintenance(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return ra_data->is_maintenance; } gboolean remote_ra_controlling_guest(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return ra_data->controlling_guest; } diff --git a/daemons/controld/controld_transition.c b/daemons/controld/controld_transition.c index 05d2eae987..6a49377589 100644 --- a/daemons/controld/controld_transition.c +++ b/daemons/controld/controld_transition.c @@ -1,217 +1,216 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include extern crm_graph_functions_t te_graph_fns; static void global_cib_callback(const xmlNode * msg, int callid, int rc, xmlNode * output) { } static crm_graph_t * create_blank_graph(void) { crm_graph_t *a_graph = pcmk__unpack_graph(NULL, NULL); a_graph->complete = TRUE; a_graph->abort_reason = "DC Takeover"; a_graph->completion_action = tg_restart; return a_graph; } /* A_TE_START, A_TE_STOP, O_TE_RESTART */ void do_te_control(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { gboolean init_ok = TRUE; if (action & A_TE_STOP) { if (transition_graph) { pcmk__free_graph(transition_graph); transition_graph = NULL; } if (fsa_cib_conn) { fsa_cib_conn->cmds->del_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, te_update_diff); } controld_clear_fsa_input_flags(R_TE_CONNECTED); crm_info("Transitioner is now inactive"); } if ((action & A_TE_START) == 0) { return; } else if (pcmk_is_set(fsa_input_register, R_TE_CONNECTED)) { crm_debug("The transitioner is already active"); return; } else if ((action & A_TE_START) && cur_state == S_STOPPING) { crm_info("Ignoring request to start the transitioner while shutting down"); return; } if (te_uuid == NULL) { te_uuid = crm_generate_uuid(); crm_info("Registering TE UUID: %s", te_uuid); } if (fsa_cib_conn == NULL) { crm_err("Could not set CIB callbacks"); init_ok = FALSE; } else { if (fsa_cib_conn->cmds->add_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, te_update_diff) != pcmk_ok) { crm_err("Could not set CIB notification callback"); init_ok = FALSE; } if (fsa_cib_conn->cmds->set_op_callback(fsa_cib_conn, global_cib_callback) != pcmk_ok) { crm_err("Could not set CIB global callback"); init_ok = FALSE; } } if (init_ok) { pcmk__set_graph_functions(&te_graph_fns); if (transition_graph) { pcmk__free_graph(transition_graph); } /* create a blank one */ crm_debug("Transitioner is now active"); transition_graph = create_blank_graph(); controld_set_fsa_input_flags(R_TE_CONNECTED); } } /* A_TE_INVOKE, A_TE_CANCEL */ void do_te_invoke(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { if (AM_I_DC == FALSE || (fsa_state != S_TRANSITION_ENGINE && (action & A_TE_INVOKE))) { crm_notice("No need to invoke the TE (%s) in state %s", fsa_action2string(action), fsa_state2string(fsa_state)); return; } if (action & A_TE_CANCEL) { crm_debug("Cancelling the transition: %s", transition_graph->complete ? "inactive" : "active"); abort_transition(INFINITY, tg_restart, "Peer Cancelled", NULL); if (transition_graph->complete == FALSE) { crmd_fsa_stall(FALSE); } } else if (action & A_TE_HALT) { - crm_debug("Halting the transition: %s", transition_graph->complete ? "inactive" : "active"); abort_transition(INFINITY, tg_stop, "Peer Halt", NULL); if (transition_graph->complete == FALSE) { crmd_fsa_stall(FALSE); } } else if (action & A_TE_INVOKE) { const char *value = NULL; xmlNode *graph_data = NULL; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *ref = crm_element_value(input->msg, XML_ATTR_REFERENCE); const char *graph_file = crm_element_value(input->msg, F_CRM_TGRAPH); const char *graph_input = crm_element_value(input->msg, F_CRM_TGRAPH_INPUT); if (graph_file == NULL && input->xml == NULL) { crm_log_xml_err(input->msg, "Bad command"); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); return; } if (transition_graph->complete == FALSE) { crm_info("Another transition is already active"); abort_transition(INFINITY, tg_restart, "Transition Active", NULL); return; } if (fsa_pe_ref == NULL || !pcmk__str_eq(fsa_pe_ref, ref, pcmk__str_casei)) { crm_info("Transition is redundant: %s vs. %s", crm_str(fsa_pe_ref), crm_str(ref)); abort_transition(INFINITY, tg_restart, "Transition Redundant", NULL); } graph_data = input->xml; if (graph_data == NULL && graph_file != NULL) { graph_data = filename2xml(graph_file); } if (is_timer_started(transition_timer)) { crm_debug("The transitioner wait for a transition timer"); return; } CRM_CHECK(graph_data != NULL, crm_err("Input raised by %s is invalid", msg_data->origin); crm_log_xml_err(input->msg, "Bad command"); return); pcmk__free_graph(transition_graph); transition_graph = pcmk__unpack_graph(graph_data, graph_input); if (transition_graph == NULL) { CRM_CHECK(transition_graph != NULL,); transition_graph = create_blank_graph(); return; } crm_info("Processing graph %d (ref=%s) derived from %s", transition_graph->id, ref, graph_input); te_reset_job_counts(); value = crm_element_value(graph_data, "failed-stop-offset"); if (value) { free(failed_stop_offset); failed_stop_offset = strdup(value); } value = crm_element_value(graph_data, "failed-start-offset"); if (value) { free(failed_start_offset); failed_start_offset = strdup(value); } if ((crm_element_value_epoch(graph_data, "recheck-by", &recheck_by) != pcmk_ok) || (recheck_by < 0)) { recheck_by = 0; } trigger_graph(); pcmk__log_graph(LOG_TRACE, transition_graph); if (graph_data != input->xml) { free_xml(graph_data); } } } diff --git a/daemons/fenced/fenced_remote.c b/daemons/fenced/fenced_remote.c index 224f2baba4..48763717ea 100644 --- a/daemons/fenced/fenced_remote.c +++ b/daemons/fenced/fenced_remote.c @@ -1,2199 +1,2212 @@ /* * Copyright 2009-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TIMEOUT_MULTIPLY_FACTOR 1.2 /* When one fencer queries its peers for devices able to handle a fencing * request, each peer will reply with a list of such devices available to it. * Each reply will be parsed into a st_query_result_t, with each device's * information kept in a device_properties_t. */ typedef struct device_properties_s { /* Whether access to this device has been verified */ gboolean verified; /* The remaining members are indexed by the operation's "phase" */ /* Whether this device has been executed in each phase */ gboolean executed[st_phase_max]; /* Whether this device is disallowed from executing in each phase */ gboolean disallowed[st_phase_max]; /* Action-specific timeout for each phase */ int custom_action_timeout[st_phase_max]; /* Action-specific maximum random delay for each phase */ int delay_max[st_phase_max]; /* Action-specific base delay for each phase */ int delay_base[st_phase_max]; } device_properties_t; typedef struct st_query_result_s { /* Name of peer that sent this result */ char *host; /* Only try peers for non-topology based operations once */ gboolean tried; /* Number of entries in the devices table */ int ndevices; /* Devices available to this host that are capable of fencing the target */ GHashTable *devices; } st_query_result_t; GHashTable *stonith_remote_op_list = NULL; void call_remote_stonith(remote_fencing_op_t * op, st_query_result_t * peer, int rc); static void remote_op_done(remote_fencing_op_t * op, xmlNode * data, int rc, int dup); extern xmlNode *stonith_create_op(int call_id, const char *token, const char *op, xmlNode * data, int call_options); static void report_timeout_period(remote_fencing_op_t * op, int op_timeout); static int get_op_total_timeout(const remote_fencing_op_t *op, const st_query_result_t *chosen_peer); static gint sort_strings(gconstpointer a, gconstpointer b) { return strcmp(a, b); } static void free_remote_query(gpointer data) { if (data) { st_query_result_t *query = data; crm_trace("Free'ing query result from %s", query->host); g_hash_table_destroy(query->devices); free(query->host); free(query); } } void free_stonith_remote_op_list() { if (stonith_remote_op_list != NULL) { g_hash_table_destroy(stonith_remote_op_list); stonith_remote_op_list = NULL; } } struct peer_count_data { const remote_fencing_op_t *op; gboolean verified_only; int count; }; /*! * \internal * \brief Increment a counter if a device has not been executed yet * * \param[in] key Device ID (ignored) * \param[in] value Device properties * \param[in] user_data Peer count data */ static void count_peer_device(gpointer key, gpointer value, gpointer user_data) { device_properties_t *props = (device_properties_t*)value; struct peer_count_data *data = user_data; if (!props->executed[data->op->phase] && (!data->verified_only || props->verified)) { ++(data->count); } } /*! * \internal * \brief Check the number of available devices in a peer's query results * * \param[in] op Operation that results are for * \param[in] peer Peer to count * \param[in] verified_only Whether to count only verified devices * * \return Number of devices available to peer that were not already executed */ static int count_peer_devices(const remote_fencing_op_t *op, const st_query_result_t *peer, gboolean verified_only) { struct peer_count_data data; data.op = op; data.verified_only = verified_only; data.count = 0; if (peer) { g_hash_table_foreach(peer->devices, count_peer_device, &data); } return data.count; } /*! * \internal * \brief Search for a device in a query result * * \param[in] op Operation that result is for * \param[in] peer Query result for a peer * \param[in] device Device ID to search for * * \return Device properties if found, NULL otherwise */ static device_properties_t * find_peer_device(const remote_fencing_op_t *op, const st_query_result_t *peer, const char *device) { device_properties_t *props = g_hash_table_lookup(peer->devices, device); return (props && !props->executed[op->phase] && !props->disallowed[op->phase])? props : NULL; } /*! * \internal * \brief Find a device in a peer's device list and mark it as executed * * \param[in] op Operation that peer result is for * \param[in,out] peer Peer with results to search * \param[in] device ID of device to mark as done * \param[in] verified_devices_only Only consider verified devices * * \return TRUE if device was found and marked, FALSE otherwise */ static gboolean grab_peer_device(const remote_fencing_op_t *op, st_query_result_t *peer, const char *device, gboolean verified_devices_only) { device_properties_t *props = find_peer_device(op, peer, device); if ((props == NULL) || (verified_devices_only && !props->verified)) { return FALSE; } crm_trace("Removing %s from %s (%d remaining)", device, peer->host, count_peer_devices(op, peer, FALSE)); props->executed[op->phase] = TRUE; return TRUE; } static void clear_remote_op_timers(remote_fencing_op_t * op) { if (op->query_timer) { g_source_remove(op->query_timer); op->query_timer = 0; } if (op->op_timer_total) { g_source_remove(op->op_timer_total); op->op_timer_total = 0; } if (op->op_timer_one) { g_source_remove(op->op_timer_one); op->op_timer_one = 0; } } static void free_remote_op(gpointer data) { remote_fencing_op_t *op = data; crm_log_xml_debug(op->request, "Destroying"); clear_remote_op_timers(op); free(op->id); free(op->action); free(op->delegate); free(op->target); free(op->client_id); free(op->client_name); free(op->originator); if (op->query_results) { g_list_free_full(op->query_results, free_remote_query); } if (op->request) { free_xml(op->request); op->request = NULL; } if (op->devices_list) { g_list_free_full(op->devices_list, free); op->devices_list = NULL; } g_list_free_full(op->automatic_list, free); g_list_free(op->duplicates); free(op); } void init_stonith_remote_op_hash_table(GHashTable **table) { if (*table == NULL) { *table = pcmk__strkey_table(NULL, free_remote_op); } } /*! * \internal * \brief Return an operation's originally requested action (before any remap) * * \param[in] op Operation to check * * \return Operation's original action */ static const char * op_requested_action(const remote_fencing_op_t *op) { return ((op->phase > st_phase_requested)? "reboot" : op->action); } /*! * \internal * \brief Remap a "reboot" operation to the "off" phase * * \param[in,out] op Operation to remap */ static void op_phase_off(remote_fencing_op_t *op) { crm_info("Remapping multiple-device reboot targeting %s to 'off' " CRM_XS " id=%.8s", op->target, op->id); op->phase = st_phase_off; /* Happily, "off" and "on" are shorter than "reboot", so we can reuse the * memory allocation at each phase. */ strcpy(op->action, "off"); } /*! * \internal * \brief Advance a remapped reboot operation to the "on" phase * * \param[in,out] op Operation to remap */ static void op_phase_on(remote_fencing_op_t *op) { GList *iter = NULL; crm_info("Remapped 'off' targeting %s complete, " "remapping to 'on' for %s " CRM_XS " id=%.8s", op->target, op->client_name, op->id); op->phase = st_phase_on; strcpy(op->action, "on"); /* Skip devices with automatic unfencing, because the cluster will handle it * when the node rejoins. */ for (iter = op->automatic_list; iter != NULL; iter = iter->next) { GList *match = g_list_find_custom(op->devices_list, iter->data, sort_strings); if (match) { op->devices_list = g_list_remove(op->devices_list, match->data); } } g_list_free_full(op->automatic_list, free); op->automatic_list = NULL; /* Rewind device list pointer */ op->devices = op->devices_list; } /*! * \internal * \brief Reset a remapped reboot operation * * \param[in,out] op Operation to reset */ static void undo_op_remap(remote_fencing_op_t *op) { if (op->phase > 0) { crm_info("Undoing remap of reboot targeting %s for %s " CRM_XS " id=%.8s", op->target, op->client_name, op->id); op->phase = st_phase_requested; strcpy(op->action, "reboot"); } } static xmlNode * create_op_done_notify(remote_fencing_op_t * op, int rc) { xmlNode *notify_data = create_xml_node(NULL, T_STONITH_NOTIFY_FENCE); crm_xml_add_int(notify_data, "state", op->state); crm_xml_add_int(notify_data, F_STONITH_RC, rc); crm_xml_add(notify_data, F_STONITH_TARGET, op->target); crm_xml_add(notify_data, F_STONITH_ACTION, op->action); crm_xml_add(notify_data, F_STONITH_DELEGATE, op->delegate); crm_xml_add(notify_data, F_STONITH_REMOTE_OP_ID, op->id); crm_xml_add(notify_data, F_STONITH_ORIGIN, op->originator); crm_xml_add(notify_data, F_STONITH_CLIENTID, op->client_id); crm_xml_add(notify_data, F_STONITH_CLIENTNAME, op->client_name); return notify_data; } void stonith_bcast_result_to_peers(remote_fencing_op_t * op, int rc, gboolean op_merged) { static int count = 0; xmlNode *bcast = create_xml_node(NULL, T_STONITH_REPLY); xmlNode *notify_data = create_op_done_notify(op, rc); count++; crm_trace("Broadcasting result to peers"); crm_xml_add(bcast, F_TYPE, T_STONITH_NOTIFY); crm_xml_add(bcast, F_SUBTYPE, "broadcast"); crm_xml_add(bcast, F_STONITH_OPERATION, T_STONITH_NOTIFY); crm_xml_add_int(bcast, "count", count); if (op_merged) { crm_xml_add(bcast, F_STONITH_MERGED, "true"); } add_message_xml(bcast, F_STONITH_CALLDATA, notify_data); send_cluster_message(NULL, crm_msg_stonith_ng, bcast, FALSE); free_xml(notify_data); free_xml(bcast); return; } static void handle_local_reply_and_notify(remote_fencing_op_t * op, xmlNode * data, int rc) { xmlNode *notify_data = NULL; xmlNode *reply = NULL; if (op->notify_sent == TRUE) { /* nothing to do */ return; } /* Do notification with a clean data object */ notify_data = create_op_done_notify(op, rc); crm_xml_add_int(data, "state", op->state); crm_xml_add(data, F_STONITH_TARGET, op->target); crm_xml_add(data, F_STONITH_OPERATION, op->action); reply = stonith_construct_reply(op->request, NULL, data, rc); crm_xml_add(reply, F_STONITH_DELEGATE, op->delegate); /* Send fencing OP reply to local client that initiated fencing */ do_local_reply(reply, op->client_id, op->call_options & st_opt_sync_call, FALSE); /* bcast to all local clients that the fencing operation happend */ do_stonith_notify(0, T_STONITH_NOTIFY_FENCE, rc, notify_data); do_stonith_notify(0, T_STONITH_NOTIFY_HISTORY, 0, NULL); /* mark this op as having notify's already sent */ op->notify_sent = TRUE; free_xml(reply); free_xml(notify_data); } static void handle_duplicates(remote_fencing_op_t * op, xmlNode * data, int rc) { GList *iter = NULL; for (iter = op->duplicates; iter != NULL; iter = iter->next) { remote_fencing_op_t *other = iter->data; if (other->state == st_duplicate) { other->state = op->state; crm_debug("Performing duplicate notification for %s@%s: %s " CRM_XS " id=%.8s", other->client_name, other->originator, pcmk_strerror(rc), other->id); remote_op_done(other, data, rc, TRUE); } else { // Possible if (for example) it timed out already crm_err("Skipping duplicate notification for %s@%s " CRM_XS " state=%s id=%.8s", other->client_name, other->originator, stonith_op_state_str(other->state), other->id); } } } +static char * +delegate_from_xml(xmlNode *xml) +{ + xmlNode *match = get_xpath_object("//@" F_STONITH_DELEGATE, xml, LOG_NEVER); + + if (match == NULL) { + return crm_element_value_copy(xml, F_ORIG); + } else { + return crm_element_value_copy(match, F_STONITH_DELEGATE); + } +} + /*! * \internal * \brief Finalize a remote operation. * * \description This function has two code paths. * * Path 1. This node is the owner of the operation and needs * to notify the cpg group via a broadcast as to the operation's * results. * * Path 2. The cpg broadcast is received. All nodes notify their local * stonith clients the operation results. * * So, The owner of the operation first notifies the cluster of the result, * and once that cpg notify is received back it notifies all the local clients. * * Nodes that are passive watchers of the operation will receive the * broadcast and only need to notify their local clients the operation finished. * * \param op, The fencing operation to finalize * \param data, The xml msg reply (if present) of the last delegated fencing * operation. * \param dup, Is this operation a duplicate, if so treat it a little differently * making sure the broadcast is not sent out. */ static void remote_op_done(remote_fencing_op_t * op, xmlNode * data, int rc, int dup) { int level = LOG_ERR; const char *subt = NULL; xmlNode *local_data = NULL; gboolean op_merged = FALSE; op->completed = time(NULL); clear_remote_op_timers(op); undo_op_remap(op); if (op->notify_sent == TRUE) { crm_err("Already sent notifications for '%s' targeting %s by %s for " "client %s@%s: %s " CRM_XS " rc=%d state=%s id=%.8s", op->action, op->target, (op->delegate? op->delegate : "unknown node"), op->client_name, op->originator, pcmk_strerror(rc), rc, stonith_op_state_str(op->state), op->id); goto remote_op_done_cleanup; } - if (!op->delegate && data && rc != -ENODEV && rc != -EHOSTUNREACH) { - xmlNode *ndata = get_xpath_object("//@" F_STONITH_DELEGATE, data, - LOG_NEVER); - if(ndata) { - op->delegate = crm_element_value_copy(ndata, F_STONITH_DELEGATE); - } else { - op->delegate = crm_element_value_copy(data, F_ORIG); + if (op->delegate == NULL) { + switch (rc) { + case -ENODEV: + case -EHOSTUNREACH: + break; + default: + op->delegate = delegate_from_xml(data); + break; } } if (data == NULL) { data = create_xml_node(NULL, "remote-op"); local_data = data; } if(dup) { op_merged = TRUE; } else if (crm_element_value(data, F_STONITH_MERGED)) { op_merged = TRUE; } /* Tell everyone the operation is done, we will continue * with doing the local notifications once we receive * the broadcast back. */ subt = crm_element_value(data, F_SUBTYPE); if (dup == FALSE && !pcmk__str_eq(subt, "broadcast", pcmk__str_casei)) { /* Defer notification until the bcast message arrives */ stonith_bcast_result_to_peers(op, rc, (op_merged? TRUE: FALSE)); goto remote_op_done_cleanup; } if (rc == pcmk_ok || dup) { level = LOG_NOTICE; } else if (!pcmk__str_eq(op->originator, stonith_our_uname, pcmk__str_casei)) { level = LOG_NOTICE; } do_crm_log(level, "Operation '%s'%s%s by %s for %s@%s%s: %s " CRM_XS " id=%.8s", op->action, (op->target? " targeting " : ""), (op->target? op->target : ""), (op->delegate? op->delegate : "unknown node"), op->client_name, op->originator, (op_merged? " (merged)" : ""), pcmk_strerror(rc), op->id); handle_local_reply_and_notify(op, data, rc); if (dup == FALSE) { handle_duplicates(op, data, rc); } /* Free non-essential parts of the record * Keep the record around so we can query the history */ if (op->query_results) { g_list_free_full(op->query_results, free_remote_query); op->query_results = NULL; } if (op->request) { free_xml(op->request); op->request = NULL; } remote_op_done_cleanup: free_xml(local_data); } static gboolean remote_op_watchdog_done(gpointer userdata) { remote_fencing_op_t *op = userdata; op->op_timer_one = 0; crm_notice("Self-fencing (%s) by %s for %s assumed complete " CRM_XS " id=%.8s", op->action, op->target, op->client_name, op->id); op->state = st_done; remote_op_done(op, NULL, pcmk_ok, FALSE); return FALSE; } static gboolean remote_op_timeout_one(gpointer userdata) { remote_fencing_op_t *op = userdata; op->op_timer_one = 0; crm_notice("Peer's '%s' action targeting %s for client %s timed out " CRM_XS " id=%.8s", op->action, op->target, op->client_name, op->id); call_remote_stonith(op, NULL, pcmk_ok); return FALSE; } static gboolean remote_op_timeout(gpointer userdata) { remote_fencing_op_t *op = userdata; op->op_timer_total = 0; if (op->state == st_done) { crm_debug("Action '%s' targeting %s for client %s already completed " CRM_XS " id=%.8s", op->action, op->target, op->client_name, op->id); return FALSE; } crm_debug("Action '%s' targeting %s for client %s timed out " CRM_XS " id=%.8s", op->action, op->target, op->client_name, op->id); if (op->phase == st_phase_on) { /* A remapped reboot operation timed out in the "on" phase, but the * "off" phase completed successfully, so quit trying any further * devices, and return success. */ remote_op_done(op, NULL, pcmk_ok, FALSE); return FALSE; } op->state = st_failed; remote_op_done(op, NULL, -ETIME, FALSE); return FALSE; } static gboolean remote_op_query_timeout(gpointer data) { remote_fencing_op_t *op = data; op->query_timer = 0; if (op->state == st_done) { crm_debug("Operation %.8s targeting %s already completed", op->id, op->target); } else if (op->state == st_exec) { crm_debug("Operation %.8s targeting %s already in progress", op->id, op->target); } else if (op->query_results) { crm_debug("Query %.8s targeting %s complete (state=%s)", op->id, op->target, stonith_op_state_str(op->state)); call_remote_stonith(op, NULL, pcmk_ok); } else { crm_debug("Query %.8s targeting %s timed out (state=%s)", op->id, op->target, stonith_op_state_str(op->state)); if (op->op_timer_total) { g_source_remove(op->op_timer_total); op->op_timer_total = 0; } remote_op_timeout(op); } return FALSE; } static gboolean topology_is_empty(stonith_topology_t *tp) { int i; if (tp == NULL) { return TRUE; } for (i = 0; i < ST_LEVEL_MAX; i++) { if (tp->levels[i] != NULL) { return FALSE; } } return TRUE; } /*! * \internal * \brief Add a device to an operation's automatic unfencing list * * \param[in,out] op Operation to modify * \param[in] device Device ID to add */ static void add_required_device(remote_fencing_op_t *op, const char *device) { GList *match = g_list_find_custom(op->automatic_list, device, sort_strings); if (!match) { op->automatic_list = g_list_prepend(op->automatic_list, strdup(device)); } } /*! * \internal * \brief Remove a device from the automatic unfencing list * * \param[in,out] op Operation to modify * \param[in] device Device ID to remove */ static void remove_required_device(remote_fencing_op_t *op, const char *device) { GList *match = g_list_find_custom(op->automatic_list, device, sort_strings); if (match) { op->automatic_list = g_list_remove(op->automatic_list, match->data); } } /* deep copy the device list */ static void set_op_device_list(remote_fencing_op_t * op, GList *devices) { GList *lpc = NULL; if (op->devices_list) { g_list_free_full(op->devices_list, free); op->devices_list = NULL; } for (lpc = devices; lpc != NULL; lpc = lpc->next) { op->devices_list = g_list_append(op->devices_list, strdup(lpc->data)); } op->devices = op->devices_list; } /*! * \internal * \brief Check whether a node matches a topology target * * \param[in] tp Topology table entry to check * \param[in] node Name of node to check * * \return TRUE if node matches topology target */ static gboolean topology_matches(const stonith_topology_t *tp, const char *node) { regex_t r_patt; CRM_CHECK(node && tp && tp->target, return FALSE); switch(tp->kind) { case 2: /* This level targets by attribute, so tp->target is a NAME=VALUE pair * of a permanent attribute applied to targeted nodes. The test below * relies on the locally cached copy of the CIB, so if fencing needs to * be done before the initial CIB is received or after a malformed CIB * is received, then the topology will be unable to be used. */ if (node_has_attr(node, tp->target_attribute, tp->target_value)) { crm_notice("Matched %s with %s by attribute", node, tp->target); return TRUE; } break; case 1: /* This level targets by name, so tp->target is a regular expression * matching names of nodes to be targeted. */ if (regcomp(&r_patt, tp->target_pattern, REG_EXTENDED|REG_NOSUB)) { crm_info("Bad regex '%s' for fencing level", tp->target); } else { int status = regexec(&r_patt, node, 0, NULL, 0); regfree(&r_patt); if (status == 0) { crm_notice("Matched %s with %s by name", node, tp->target); return TRUE; } } break; case 0: crm_trace("Testing %s against %s", node, tp->target); return pcmk__str_eq(tp->target, node, pcmk__str_casei); } crm_trace("No match for %s with %s", node, tp->target); return FALSE; } stonith_topology_t * find_topology_for_host(const char *host) { GHashTableIter tIter; stonith_topology_t *tp = g_hash_table_lookup(topology, host); if(tp != NULL) { crm_trace("Found %s for %s in %d entries", tp->target, host, g_hash_table_size(topology)); return tp; } g_hash_table_iter_init(&tIter, topology); while (g_hash_table_iter_next(&tIter, NULL, (gpointer *) & tp)) { if (topology_matches(tp, host)) { crm_trace("Found %s for %s in %d entries", tp->target, host, g_hash_table_size(topology)); return tp; } } crm_trace("No matches for %s in %d topology entries", host, g_hash_table_size(topology)); return NULL; } /*! * \internal * \brief Set fencing operation's device list to target's next topology level * * \param[in,out] op Remote fencing operation to modify * \param[in] empty_ok If true, an operation without a target (i.e. * queries) or a target without a topology will get a * pcmk_rc_ok return value instead of ENODEV * * \return Standard Pacemaker return value */ static int advance_topology_level(remote_fencing_op_t *op, bool empty_ok) { stonith_topology_t *tp = NULL; if (op->target) { tp = find_topology_for_host(op->target); } if (topology_is_empty(tp)) { return empty_ok? pcmk_rc_ok : ENODEV; } CRM_ASSERT(tp->levels != NULL); stonith__set_call_options(op->call_options, op->id, st_opt_topology); /* This is a new level, so undo any remapping left over from previous */ undo_op_remap(op); do { op->level++; } while (op->level < ST_LEVEL_MAX && tp->levels[op->level] == NULL); if (op->level < ST_LEVEL_MAX) { crm_trace("Attempting fencing level %d targeting %s (%d devices) " "for client %s@%s (id=%.8s)", op->level, op->target, g_list_length(tp->levels[op->level]), op->client_name, op->originator, op->id); set_op_device_list(op, tp->levels[op->level]); // The requested delay has been applied for the first fencing level if (op->level > 1 && op->delay > 0) { op->delay = 0; } if (g_list_next(op->devices_list) && pcmk__str_eq(op->action, "reboot", pcmk__str_casei)) { /* A reboot has been requested for a topology level with multiple * devices. Instead of rebooting the devices sequentially, we will * turn them all off, then turn them all on again. (Think about * switched power outlets for redundant power supplies.) */ op_phase_off(op); } return pcmk_rc_ok; } crm_notice("All fencing options targeting %s for client %s@%s failed " CRM_XS " id=%.8s", op->target, op->client_name, op->originator, op->id); return ENODEV; } /*! * \brief Check to see if this operation is a duplicate of another in flight * operation. If so merge this operation into the inflight operation, and mark * it as a duplicate. */ static void merge_duplicates(remote_fencing_op_t * op) { GHashTableIter iter; remote_fencing_op_t *other = NULL; time_t now = time(NULL); g_hash_table_iter_init(&iter, stonith_remote_op_list); while (g_hash_table_iter_next(&iter, NULL, (void **)&other)) { const char *other_action = op_requested_action(other); if (!strcmp(op->id, other->id)) { continue; // Don't compare against self } if (other->state > st_exec) { crm_trace("%.8s not duplicate of %.8s: not in progress", op->id, other->id); continue; } if (!pcmk__str_eq(op->target, other->target, pcmk__str_casei)) { crm_trace("%.8s not duplicate of %.8s: node %s vs. %s", op->id, other->id, op->target, other->target); continue; } if (!pcmk__str_eq(op->action, other_action, pcmk__str_casei)) { crm_trace("%.8s not duplicate of %.8s: action %s vs. %s", op->id, other->id, op->action, other_action); continue; } if (pcmk__str_eq(op->client_name, other->client_name, pcmk__str_casei)) { crm_trace("%.8s not duplicate of %.8s: same client %s", op->id, other->id, op->client_name); continue; } if (pcmk__str_eq(other->target, other->originator, pcmk__str_casei)) { crm_trace("%.8s not duplicate of %.8s: suicide for %s", op->id, other->id, other->target); continue; } if (!fencing_peer_active(crm_get_peer(0, other->originator))) { crm_notice("Failing action '%s' targeting %s originating from " "client %s@%s: Originator is dead " CRM_XS " id=%.8s", other->action, other->target, other->client_name, other->originator, other->id); crm_trace("%.8s not duplicate of %.8s: originator dead", op->id, other->id); other->state = st_failed; continue; } if ((other->total_timeout > 0) && (now > (other->total_timeout + other->created))) { crm_trace("%.8s not duplicate of %.8s: old (%ld vs. %ld + %d)", op->id, other->id, now, other->created, other->total_timeout); continue; } /* There is another in-flight request to fence the same host * Piggyback on that instead. If it fails, so do we. */ other->duplicates = g_list_append(other->duplicates, op); if (other->total_timeout == 0) { other->total_timeout = op->total_timeout = TIMEOUT_MULTIPLY_FACTOR * get_op_total_timeout(op, NULL); crm_trace("Best guess as to timeout used for %.8s: %d", other->id, other->total_timeout); } crm_notice("Merging fencing action '%s' targeting %s originating from " "client %s with identical request from %s@%s " CRM_XS " original=%.8s duplicate=%.8s total_timeout=%ds", op->action, op->target, op->client_name, other->client_name, other->originator, op->id, other->id, other->total_timeout); report_timeout_period(op, other->total_timeout); op->state = st_duplicate; } } static uint32_t fencing_active_peers(void) { uint32_t count = 0; crm_node_t *entry; GHashTableIter gIter; g_hash_table_iter_init(&gIter, crm_peer_cache); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { if(fencing_peer_active(entry)) { count++; } } return count; } int stonith_manual_ack(xmlNode * msg, remote_fencing_op_t * op) { xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, msg, LOG_ERR); op->state = st_done; op->completed = time(NULL); op->delegate = strdup("a human"); crm_notice("Injecting manual confirmation that %s is safely off/down", crm_element_value(dev, F_STONITH_TARGET)); remote_op_done(op, msg, pcmk_ok, FALSE); /* Replies are sent via done_cb->stonith_send_async_reply()->do_local_reply() */ return -EINPROGRESS; } /*! * \internal * \brief Create a new remote stonith operation * * \param[in] client ID of local stonith client that initiated the operation * \param[in] request The request from the client that started the operation * \param[in] peer TRUE if this operation is owned by another stonith peer * (an operation owned by one peer is stored on all peers, * but only the owner executes it; all nodes get the results * once the owner finishes execution) */ void * create_remote_stonith_op(const char *client, xmlNode * request, gboolean peer) { remote_fencing_op_t *op = NULL; xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, request, LOG_NEVER); int call_options = 0; const char *operation = NULL; init_stonith_remote_op_hash_table(&stonith_remote_op_list); /* If this operation is owned by another node, check to make * sure we haven't already created this operation. */ if (peer && dev) { const char *op_id = crm_element_value(dev, F_STONITH_REMOTE_OP_ID); CRM_CHECK(op_id != NULL, return NULL); op = g_hash_table_lookup(stonith_remote_op_list, op_id); if (op) { crm_debug("Reusing existing remote fencing op %.8s for %s", op_id, ((client == NULL)? "unknown client" : client)); return op; } } op = calloc(1, sizeof(remote_fencing_op_t)); CRM_ASSERT(op != NULL); crm_element_value_int(request, F_STONITH_TIMEOUT, &(op->base_timeout)); // Value -1 means disable any static/random fencing delays crm_element_value_int(request, F_STONITH_DELAY, &(op->delay)); if (peer && dev) { op->id = crm_element_value_copy(dev, F_STONITH_REMOTE_OP_ID); } else { op->id = crm_generate_uuid(); } g_hash_table_replace(stonith_remote_op_list, op->id, op); op->state = st_query; op->replies_expected = fencing_active_peers(); op->action = crm_element_value_copy(dev, F_STONITH_ACTION); op->originator = crm_element_value_copy(dev, F_STONITH_ORIGIN); op->delegate = crm_element_value_copy(dev, F_STONITH_DELEGATE); /* May not be set */ op->created = time(NULL); if (op->originator == NULL) { /* Local or relayed request */ op->originator = strdup(stonith_our_uname); } CRM_LOG_ASSERT(client != NULL); if (client) { op->client_id = strdup(client); } /* For a RELAY operation, set fenced on the client. */ operation = crm_element_value(request, F_STONITH_OPERATION); if (pcmk__str_eq(operation, STONITH_OP_RELAY, pcmk__str_none)) { op->client_name = crm_strdup_printf("%s.%lu", crm_system_name, (unsigned long) getpid()); } else { op->client_name = crm_element_value_copy(request, F_STONITH_CLIENTNAME); } op->target = crm_element_value_copy(dev, F_STONITH_TARGET); op->request = copy_xml(request); /* TODO: Figure out how to avoid this */ crm_element_value_int(request, F_STONITH_CALLOPTS, &call_options); op->call_options = call_options; crm_element_value_int(request, F_STONITH_CALLID, &(op->client_callid)); crm_trace("%s new fencing op %s ('%s' targeting %s for client %s, " "base timeout %d, %u %s expected)", (peer && dev)? "Recorded" : "Generated", op->id, op->action, op->target, op->client_name, op->base_timeout, op->replies_expected, pcmk__plural_alt(op->replies_expected, "reply", "replies")); if (op->call_options & st_opt_cs_nodeid) { int nodeid; crm_node_t *node; pcmk__scan_min_int(op->target, &nodeid, 0); node = pcmk__search_known_node_cache(nodeid, NULL, CRM_GET_PEER_ANY); /* Ensure the conversion only happens once */ stonith__clear_call_options(op->call_options, op->id, st_opt_cs_nodeid); if (node && node->uname) { free(op->target); op->target = strdup(node->uname); } else { crm_warn("Could not expand nodeid '%s' into a host name", op->target); } } /* check to see if this is a duplicate operation of another in-flight operation */ merge_duplicates(op); if (op->state != st_duplicate) { /* kick history readers */ do_stonith_notify(0, T_STONITH_NOTIFY_HISTORY, 0, NULL); } /* safe to trim as long as that doesn't touch pending ops */ stonith_fence_history_trim(); return op; } remote_fencing_op_t * initiate_remote_stonith_op(pcmk__client_t *client, xmlNode *request, gboolean manual_ack) { int query_timeout = 0; xmlNode *query = NULL; const char *client_id = NULL; remote_fencing_op_t *op = NULL; const char *relay_op_id = NULL; const char *operation = NULL; if (client) { client_id = client->id; } else { client_id = crm_element_value(request, F_STONITH_CLIENTID); } CRM_LOG_ASSERT(client_id != NULL); op = create_remote_stonith_op(client_id, request, FALSE); op->owner = TRUE; if (manual_ack) { crm_notice("Processing manual confirmation of fencing targeting %s " CRM_XS " id=%.8s", op->target, op->id); return op; } CRM_CHECK(op->action, return NULL); if (advance_topology_level(op, true) != pcmk_rc_ok) { op->state = st_failed; } switch (op->state) { case st_failed: crm_warn("Could not request peer fencing (%s) targeting %s " CRM_XS " id=%.8s", op->action, op->target, op->id); remote_op_done(op, NULL, -EINVAL, FALSE); return op; case st_duplicate: crm_info("Requesting peer fencing (%s) targeting %s (duplicate) " CRM_XS " id=%.8s", op->action, op->target, op->id); return op; default: crm_notice("Requesting peer fencing (%s) targeting %s " CRM_XS " id=%.8s state=%s base_timeout=%d", op->action, op->target, op->id, stonith_op_state_str(op->state), op->base_timeout); } query = stonith_create_op(op->client_callid, op->id, STONITH_OP_QUERY, NULL, op->call_options); crm_xml_add(query, F_STONITH_REMOTE_OP_ID, op->id); crm_xml_add(query, F_STONITH_TARGET, op->target); crm_xml_add(query, F_STONITH_ACTION, op_requested_action(op)); crm_xml_add(query, F_STONITH_ORIGIN, op->originator); crm_xml_add(query, F_STONITH_CLIENTID, op->client_id); crm_xml_add(query, F_STONITH_CLIENTNAME, op->client_name); crm_xml_add_int(query, F_STONITH_TIMEOUT, op->base_timeout); /* In case of RELAY operation, RELAY information is added to the query to delete the original operation of RELAY. */ operation = crm_element_value(request, F_STONITH_OPERATION); if (pcmk__str_eq(operation, STONITH_OP_RELAY, pcmk__str_none)) { relay_op_id = crm_element_value(request, F_STONITH_REMOTE_OP_ID); if (relay_op_id) { crm_xml_add(query, F_STONITH_REMOTE_OP_ID_RELAY, relay_op_id); } } send_cluster_message(NULL, crm_msg_stonith_ng, query, FALSE); free_xml(query); query_timeout = op->base_timeout * TIMEOUT_MULTIPLY_FACTOR; op->query_timer = g_timeout_add((1000 * query_timeout), remote_op_query_timeout, op); return op; } enum find_best_peer_options { /*! Skip checking the target peer for capable fencing devices */ FIND_PEER_SKIP_TARGET = 0x0001, /*! Only check the target peer for capable fencing devices */ FIND_PEER_TARGET_ONLY = 0x0002, /*! Skip peers and devices that are not verified */ FIND_PEER_VERIFIED_ONLY = 0x0004, }; static st_query_result_t * find_best_peer(const char *device, remote_fencing_op_t * op, enum find_best_peer_options options) { GList *iter = NULL; gboolean verified_devices_only = (options & FIND_PEER_VERIFIED_ONLY) ? TRUE : FALSE; if (!device && pcmk_is_set(op->call_options, st_opt_topology)) { return NULL; } for (iter = op->query_results; iter != NULL; iter = iter->next) { st_query_result_t *peer = iter->data; crm_trace("Testing result from %s targeting %s with %d device%s: %d %x", peer->host, op->target, peer->ndevices, pcmk__plural_s(peer->ndevices), peer->tried, options); if ((options & FIND_PEER_SKIP_TARGET) && pcmk__str_eq(peer->host, op->target, pcmk__str_casei)) { continue; } if ((options & FIND_PEER_TARGET_ONLY) && !pcmk__str_eq(peer->host, op->target, pcmk__str_casei)) { continue; } if (pcmk_is_set(op->call_options, st_opt_topology)) { if (grab_peer_device(op, peer, device, verified_devices_only)) { return peer; } } else if ((peer->tried == FALSE) && count_peer_devices(op, peer, verified_devices_only)) { /* No topology: Use the current best peer */ crm_trace("Simple fencing"); return peer; } } return NULL; } static st_query_result_t * stonith_choose_peer(remote_fencing_op_t * op) { const char *device = NULL; st_query_result_t *peer = NULL; uint32_t active = fencing_active_peers(); do { if (op->devices) { device = op->devices->data; crm_trace("Checking for someone to fence (%s) %s using %s", op->action, op->target, device); } else { crm_trace("Checking for someone to fence (%s) %s", op->action, op->target); } /* Best choice is a peer other than the target with verified access */ peer = find_best_peer(device, op, FIND_PEER_SKIP_TARGET|FIND_PEER_VERIFIED_ONLY); if (peer) { crm_trace("Found verified peer %s for %s", peer->host, device?device:""); return peer; } if(op->query_timer != 0 && op->replies < QB_MIN(op->replies_expected, active)) { crm_trace("Waiting before looking for unverified devices to fence %s", op->target); return NULL; } /* If no other peer has verified access, next best is unverified access */ peer = find_best_peer(device, op, FIND_PEER_SKIP_TARGET); if (peer) { crm_trace("Found best unverified peer %s", peer->host); return peer; } /* If no other peer can do it, last option is self-fencing * (which is never allowed for the "on" phase of a remapped reboot) */ if (op->phase != st_phase_on) { peer = find_best_peer(device, op, FIND_PEER_TARGET_ONLY); if (peer) { crm_trace("%s will fence itself", peer->host); return peer; } } /* Try the next fencing level if there is one (unless we're in the "on" * phase of a remapped "reboot", because we ignore errors in that case) */ } while ((op->phase != st_phase_on) && pcmk_is_set(op->call_options, st_opt_topology) && (advance_topology_level(op, false) == pcmk_rc_ok)); crm_notice("Couldn't find anyone to fence (%s) %s using %s", op->action, op->target, (device? device : "any device")); return NULL; } static int get_device_timeout(const remote_fencing_op_t *op, const st_query_result_t *peer, const char *device) { device_properties_t *props; if (!peer || !device) { return op->base_timeout; } props = g_hash_table_lookup(peer->devices, device); if (!props) { return op->base_timeout; } return (props->custom_action_timeout[op->phase]? props->custom_action_timeout[op->phase] : op->base_timeout) + props->delay_max[op->phase]; } struct timeout_data { const remote_fencing_op_t *op; const st_query_result_t *peer; int total_timeout; }; /*! * \internal * \brief Add timeout to a total if device has not been executed yet * * \param[in] key GHashTable key (device ID) * \param[in] value GHashTable value (device properties) * \param[in] user_data Timeout data */ static void add_device_timeout(gpointer key, gpointer value, gpointer user_data) { const char *device_id = key; device_properties_t *props = value; struct timeout_data *timeout = user_data; if (!props->executed[timeout->op->phase] && !props->disallowed[timeout->op->phase]) { timeout->total_timeout += get_device_timeout(timeout->op, timeout->peer, device_id); } } static int get_peer_timeout(const remote_fencing_op_t *op, const st_query_result_t *peer) { struct timeout_data timeout; timeout.op = op; timeout.peer = peer; timeout.total_timeout = 0; g_hash_table_foreach(peer->devices, add_device_timeout, &timeout); return (timeout.total_timeout? timeout.total_timeout : op->base_timeout); } static int get_op_total_timeout(const remote_fencing_op_t *op, const st_query_result_t *chosen_peer) { int total_timeout = 0; stonith_topology_t *tp = find_topology_for_host(op->target); if (pcmk_is_set(op->call_options, st_opt_topology) && tp) { int i; GList *device_list = NULL; GList *iter = NULL; /* Yep, this looks scary, nested loops all over the place. * Here is what is going on. * Loop1: Iterate through fencing levels. * Loop2: If a fencing level has devices, loop through each device * Loop3: For each device in a fencing level, see what peer owns it * and what that peer has reported the timeout is for the device. */ for (i = 0; i < ST_LEVEL_MAX; i++) { if (!tp->levels[i]) { continue; } for (device_list = tp->levels[i]; device_list; device_list = device_list->next) { for (iter = op->query_results; iter != NULL; iter = iter->next) { const st_query_result_t *peer = iter->data; if (find_peer_device(op, peer, device_list->data)) { total_timeout += get_device_timeout(op, peer, device_list->data); break; } } /* End Loop3: match device with peer that owns device, find device's timeout period */ } /* End Loop2: iterate through devices at a specific level */ } /*End Loop1: iterate through fencing levels */ } else if (chosen_peer) { total_timeout = get_peer_timeout(op, chosen_peer); } else { total_timeout = op->base_timeout; } return total_timeout ? total_timeout : op->base_timeout; } static void report_timeout_period(remote_fencing_op_t * op, int op_timeout) { GList *iter = NULL; xmlNode *update = NULL; const char *client_node = NULL; const char *client_id = NULL; const char *call_id = NULL; if (op->call_options & st_opt_sync_call) { /* There is no reason to report the timeout for a synchronous call. It * is impossible to use the reported timeout to do anything when the client * is blocking for the response. This update is only important for * async calls that require a callback to report the results in. */ return; } else if (!op->request) { return; } crm_trace("Reporting timeout for %s (id=%.8s)", op->client_name, op->id); client_node = crm_element_value(op->request, F_STONITH_CLIENTNODE); call_id = crm_element_value(op->request, F_STONITH_CALLID); client_id = crm_element_value(op->request, F_STONITH_CLIENTID); if (!client_node || !call_id || !client_id) { return; } if (pcmk__str_eq(client_node, stonith_our_uname, pcmk__str_casei)) { /* The client is connected to this node, send the update direclty to them */ do_stonith_async_timeout_update(client_id, call_id, op_timeout); return; } /* The client is connected to another node, relay this update to them */ update = stonith_create_op(op->client_callid, op->id, STONITH_OP_TIMEOUT_UPDATE, NULL, 0); crm_xml_add(update, F_STONITH_REMOTE_OP_ID, op->id); crm_xml_add(update, F_STONITH_CLIENTID, client_id); crm_xml_add(update, F_STONITH_CALLID, call_id); crm_xml_add_int(update, F_STONITH_TIMEOUT, op_timeout); send_cluster_message(crm_get_peer(0, client_node), crm_msg_stonith_ng, update, FALSE); free_xml(update); for (iter = op->duplicates; iter != NULL; iter = iter->next) { remote_fencing_op_t *dup = iter->data; crm_trace("Reporting timeout for duplicate %.8s to client %s", dup->id, dup->client_name); report_timeout_period(iter->data, op_timeout); } } /*! * \internal * \brief Advance an operation to the next device in its topology * * \param[in,out] op Operation to advance * \param[in] device ID of device just completed * \param[in] msg XML reply that contained device result (if available) * \param[in] rc Return code of device's execution */ static void advance_topology_device_in_level(remote_fencing_op_t *op, const char *device, xmlNode *msg, int rc) { /* Advance to the next device at this topology level, if any */ if (op->devices) { op->devices = op->devices->next; } /* Handle automatic unfencing if an "on" action was requested */ if ((op->phase == st_phase_requested) && pcmk__str_eq(op->action, "on", pcmk__str_casei)) { /* If the device we just executed was required, it's not anymore */ remove_required_device(op, device); /* If there are no more devices at this topology level, run through any * remaining devices with automatic unfencing */ if (op->devices == NULL) { op->devices = op->automatic_list; } } if ((op->devices == NULL) && (op->phase == st_phase_off)) { /* We're done with this level and with required devices, but we had * remapped "reboot" to "off", so start over with "on". If any devices * need to be turned back on, op->devices will be non-NULL after this. */ op_phase_on(op); } if (op->devices) { /* Necessary devices remain, so execute the next one */ crm_trace("Next targeting %s on behalf of %s@%s (rc was %d)", op->target, op->client_name, op->originator, rc); // The requested delay has been applied for the first device if (op->delay > 0) { op->delay = 0; } call_remote_stonith(op, NULL, pcmk_ok); } else { /* We're done with all devices and phases, so finalize operation */ crm_trace("Marking complex fencing op targeting %s as complete", op->target); op->state = st_done; remote_op_done(op, msg, rc, FALSE); } } static gboolean check_watchdog_fencing_and_wait(remote_fencing_op_t * op) { if (node_does_watchdog_fencing(op->target)) { crm_notice("Waiting %lds for %s to self-fence (%s) for " "client %s " CRM_XS " id=%.8s", (stonith_watchdog_timeout_ms / 1000), op->target, op->action, op->client_name, op->id); op->op_timer_one = g_timeout_add(stonith_watchdog_timeout_ms, remote_op_watchdog_done, op); return TRUE; } else { crm_debug("Skipping fallback to watchdog-fencing as %s is " "not in host-list", op->target); } return FALSE; } void call_remote_stonith(remote_fencing_op_t * op, st_query_result_t * peer, int rc) { const char *device = NULL; int timeout = op->base_timeout; crm_trace("Action %.8s targeting %s for %s is %s", op->id, op->target, op->client_name, stonith_op_state_str(op->state)); if ((peer == NULL) && !pcmk_is_set(op->call_options, st_opt_topology)) { peer = stonith_choose_peer(op); } if (!op->op_timer_total) { int total_timeout = get_op_total_timeout(op, peer); op->total_timeout = TIMEOUT_MULTIPLY_FACTOR * total_timeout; op->op_timer_total = g_timeout_add(1000 * op->total_timeout, remote_op_timeout, op); report_timeout_period(op, op->total_timeout); crm_info("Total timeout set to %d for peer's fencing targeting %s for %s" CRM_XS "id=%.8s", total_timeout, op->target, op->client_name, op->id); } if (pcmk_is_set(op->call_options, st_opt_topology) && op->devices) { /* Ignore any peer preference, they might not have the device we need */ /* When using topology, stonith_choose_peer() removes the device from * further consideration, so be sure to calculate timeout beforehand */ peer = stonith_choose_peer(op); device = op->devices->data; timeout = get_device_timeout(op, peer, device); } if (peer) { int timeout_one = 0; xmlNode *remote_op = stonith_create_op(op->client_callid, op->id, STONITH_OP_FENCE, NULL, 0); crm_xml_add(remote_op, F_STONITH_REMOTE_OP_ID, op->id); crm_xml_add(remote_op, F_STONITH_TARGET, op->target); crm_xml_add(remote_op, F_STONITH_ACTION, op->action); crm_xml_add(remote_op, F_STONITH_ORIGIN, op->originator); crm_xml_add(remote_op, F_STONITH_CLIENTID, op->client_id); crm_xml_add(remote_op, F_STONITH_CLIENTNAME, op->client_name); crm_xml_add_int(remote_op, F_STONITH_TIMEOUT, timeout); crm_xml_add_int(remote_op, F_STONITH_CALLOPTS, op->call_options); crm_xml_add_int(remote_op, F_STONITH_DELAY, op->delay); if (device) { timeout_one = TIMEOUT_MULTIPLY_FACTOR * get_device_timeout(op, peer, device); crm_notice("Requesting that %s perform '%s' action targeting %s " "using %s " CRM_XS " for client %s (%ds)", peer->host, op->action, op->target, device, op->client_name, timeout_one); crm_xml_add(remote_op, F_STONITH_DEVICE, device); } else { timeout_one = TIMEOUT_MULTIPLY_FACTOR * get_peer_timeout(op, peer); crm_notice("Requesting that %s perform '%s' action targeting %s " CRM_XS " for client %s (%ds, %lds)", peer->host, op->action, op->target, op->client_name, timeout_one, stonith_watchdog_timeout_ms); } op->state = st_exec; if (op->op_timer_one) { g_source_remove(op->op_timer_one); } if (!(stonith_watchdog_timeout_ms > 0 && ( (pcmk__str_eq(device, STONITH_WATCHDOG_ID, pcmk__str_none)) || (pcmk__str_eq(peer->host, op->target, pcmk__str_casei) && !pcmk__str_eq(op->action, "on", pcmk__str_casei))) && check_watchdog_fencing_and_wait(op))) { /* Some thoughts about self-fencing cases reaching this point: - Actually check in check_watchdog_fencing_and_wait shouldn't fail if STONITH_WATCHDOG_ID is chosen as fencing-device and it being present implies watchdog-fencing is enabled anyway - If watchdog-fencing is disabled either in general or for a specific target - detected in check_watchdog_fencing_and_wait - for some other kind of self-fencing we can't expect a success answer but timeout is fine if the node doesn't come back in between - Delicate might be the case where we have watchdog-fencing enabled for a node but the watchdog-fencing-device isn't explicitly chosen for suicide. Local pe-execution in sbd may detect the node as unclean and lead to timely suicide. Otherwise the selection of stonith-watchdog-timeout at least is questionable. */ op->op_timer_one = g_timeout_add((1000 * timeout_one), remote_op_timeout_one, op); } send_cluster_message(crm_get_peer(0, peer->host), crm_msg_stonith_ng, remote_op, FALSE); peer->tried = TRUE; free_xml(remote_op); return; } else if (op->phase == st_phase_on) { /* A remapped "on" cannot be executed, but the node was already * turned off successfully, so ignore the error and continue. */ crm_warn("Ignoring %s 'on' failure (no capable peers) targeting %s " "after successful 'off'", device, op->target); advance_topology_device_in_level(op, device, NULL, pcmk_ok); return; } else if (op->owner == FALSE) { crm_err("Fencing (%s) targeting %s for client %s is not ours to control", op->action, op->target, op->client_name); } else if (op->query_timer == 0) { /* We've exhausted all available peers */ crm_info("No remaining peers capable of fencing (%s) %s for client %s " CRM_XS " state=%s", op->action, op->target, op->client_name, stonith_op_state_str(op->state)); CRM_LOG_ASSERT(op->state < st_done); remote_op_timeout(op); } else if(op->replies >= op->replies_expected || op->replies >= fencing_active_peers()) { // int rc = -EHOSTUNREACH; /* if the operation never left the query state, * but we have all the expected replies, then no devices * are available to execute the fencing operation. */ if(stonith_watchdog_timeout_ms > 0 && pcmk__str_eq(device, STONITH_WATCHDOG_ID, pcmk__str_null_matches)) { if (check_watchdog_fencing_and_wait(op)) { return; } } if (op->state == st_query) { crm_info("No peers (out of %d) have devices capable of fencing " "(%s) %s for client %s " CRM_XS " state=%s", op->replies, op->action, op->target, op->client_name, stonith_op_state_str(op->state)); rc = -ENODEV; } else { if (pcmk_is_set(op->call_options, st_opt_topology)) { rc = -EHOSTUNREACH; } crm_info("No peers (out of %d) are capable of fencing (%s) %s " "for client %s " CRM_XS " state=%s", op->replies, op->action, op->target, op->client_name, stonith_op_state_str(op->state)); } op->state = st_failed; remote_op_done(op, NULL, rc, FALSE); } else { crm_info("Waiting for additional peers capable of fencing (%s) %s%s%s " "for client %s " CRM_XS " id=%.8s", op->action, op->target, (device? " using " : ""), (device? device : ""), op->client_name, op->id); } } /*! * \internal * \brief Comparison function for sorting query results * * \param[in] a GList item to compare * \param[in] b GList item to compare * * \return Per the glib documentation, "a negative integer if the first value * comes before the second, 0 if they are equal, or a positive integer * if the first value comes after the second." */ static gint sort_peers(gconstpointer a, gconstpointer b) { const st_query_result_t *peer_a = a; const st_query_result_t *peer_b = b; return (peer_b->ndevices - peer_a->ndevices); } /*! * \internal * \brief Determine if all the devices in the topology are found or not */ static gboolean all_topology_devices_found(remote_fencing_op_t * op) { GList *device = NULL; GList *iter = NULL; device_properties_t *match = NULL; stonith_topology_t *tp = NULL; gboolean skip_target = FALSE; int i; tp = find_topology_for_host(op->target); if (!tp) { return FALSE; } if (pcmk__strcase_any_of(op->action, "off", "reboot", NULL)) { /* Don't count the devices on the target node if we are killing * the target node. */ skip_target = TRUE; } for (i = 0; i < ST_LEVEL_MAX; i++) { for (device = tp->levels[i]; device; device = device->next) { match = NULL; for (iter = op->query_results; iter && !match; iter = iter->next) { st_query_result_t *peer = iter->data; if (skip_target && pcmk__str_eq(peer->host, op->target, pcmk__str_casei)) { continue; } match = find_peer_device(op, peer, device->data); } if (!match) { return FALSE; } } } return TRUE; } /*! * \internal * \brief Parse action-specific device properties from XML * * \param[in] msg XML element containing the properties * \param[in] peer Name of peer that sent XML (for logs) * \param[in] device Device ID (for logs) * \param[in] action Action the properties relate to (for logs) * \param[in] phase Phase the properties relate to * \param[in,out] props Device properties to update */ static void parse_action_specific(xmlNode *xml, const char *peer, const char *device, const char *action, remote_fencing_op_t *op, enum st_remap_phase phase, device_properties_t *props) { props->custom_action_timeout[phase] = 0; crm_element_value_int(xml, F_STONITH_ACTION_TIMEOUT, &props->custom_action_timeout[phase]); if (props->custom_action_timeout[phase]) { crm_trace("Peer %s with device %s returned %s action timeout %d", peer, device, action, props->custom_action_timeout[phase]); } props->delay_max[phase] = 0; crm_element_value_int(xml, F_STONITH_DELAY_MAX, &props->delay_max[phase]); if (props->delay_max[phase]) { crm_trace("Peer %s with device %s returned maximum of random delay %d for %s", peer, device, props->delay_max[phase], action); } props->delay_base[phase] = 0; crm_element_value_int(xml, F_STONITH_DELAY_BASE, &props->delay_base[phase]); if (props->delay_base[phase]) { crm_trace("Peer %s with device %s returned base delay %d for %s", peer, device, props->delay_base[phase], action); } /* Handle devices with automatic unfencing */ if (pcmk__str_eq(action, "on", pcmk__str_casei)) { int required = 0; crm_element_value_int(xml, F_STONITH_DEVICE_REQUIRED, &required); if (required) { crm_trace("Peer %s requires device %s to execute for action %s", peer, device, action); add_required_device(op, device); } } /* If a reboot is remapped to off+on, it's possible that a node is allowed * to perform one action but not another. */ if (crm_is_true(crm_element_value(xml, F_STONITH_ACTION_DISALLOWED))) { props->disallowed[phase] = TRUE; crm_trace("Peer %s is disallowed from executing %s for device %s", peer, action, device); } } /*! * \internal * \brief Parse one device's properties from peer's XML query reply * * \param[in] xml XML node containing device properties * \param[in,out] op Operation that query and reply relate to * \param[in,out] result Peer's results * \param[in] device ID of device being parsed */ static void add_device_properties(xmlNode *xml, remote_fencing_op_t *op, st_query_result_t *result, const char *device) { xmlNode *child; int verified = 0; device_properties_t *props = calloc(1, sizeof(device_properties_t)); /* Add a new entry to this result's devices list */ CRM_ASSERT(props != NULL); g_hash_table_insert(result->devices, strdup(device), props); /* Peers with verified (monitored) access will be preferred */ crm_element_value_int(xml, F_STONITH_DEVICE_VERIFIED, &verified); if (verified) { crm_trace("Peer %s has confirmed a verified device %s", result->host, device); props->verified = TRUE; } /* Parse action-specific device properties */ parse_action_specific(xml, result->host, device, op_requested_action(op), op, st_phase_requested, props); for (child = pcmk__xml_first_child(xml); child != NULL; child = pcmk__xml_next(child)) { /* Replies for "reboot" operations will include the action-specific * values for "off" and "on" in child elements, just in case the reboot * winds up getting remapped. */ if (pcmk__str_eq(ID(child), "off", pcmk__str_casei)) { parse_action_specific(child, result->host, device, "off", op, st_phase_off, props); } else if (pcmk__str_eq(ID(child), "on", pcmk__str_casei)) { parse_action_specific(child, result->host, device, "on", op, st_phase_on, props); } } } /*! * \internal * \brief Parse a peer's XML query reply and add it to operation's results * * \param[in,out] op Operation that query and reply relate to * \param[in] host Name of peer that sent this reply * \param[in] ndevices Number of devices expected in reply * \param[in] xml XML node containing device list * * \return Newly allocated result structure with parsed reply */ static st_query_result_t * add_result(remote_fencing_op_t *op, const char *host, int ndevices, xmlNode *xml) { st_query_result_t *result = calloc(1, sizeof(st_query_result_t)); xmlNode *child; // cppcheck seems not to understand the abort logic in CRM_CHECK // cppcheck-suppress memleak CRM_CHECK(result != NULL, return NULL); result->host = strdup(host); result->devices = pcmk__strkey_table(free, free); /* Each child element describes one capable device available to the peer */ for (child = pcmk__xml_first_child(xml); child != NULL; child = pcmk__xml_next(child)) { const char *device = ID(child); if (device) { add_device_properties(child, op, result, device); } } result->ndevices = g_hash_table_size(result->devices); CRM_CHECK(ndevices == result->ndevices, crm_err("Query claimed to have %d device%s but %d found", ndevices, pcmk__plural_s(ndevices), result->ndevices)); op->query_results = g_list_insert_sorted(op->query_results, result, sort_peers); return result; } /*! * \internal * \brief Handle a peer's reply to our fencing query * * Parse a query result from XML and store it in the remote operation * table, and when enough replies have been received, issue a fencing request. * * \param[in] msg XML reply received * * \return pcmk_ok on success, -errno on error * * \note See initiate_remote_stonith_op() for how the XML query was initially * formed, and stonith_query() for how the peer formed its XML reply. */ int process_remote_stonith_query(xmlNode * msg) { int ndevices = 0; gboolean host_is_target = FALSE; gboolean have_all_replies = FALSE; const char *id = NULL; const char *host = NULL; remote_fencing_op_t *op = NULL; st_query_result_t *result = NULL; uint32_t replies_expected; xmlNode *dev = get_xpath_object("//@" F_STONITH_REMOTE_OP_ID, msg, LOG_ERR); CRM_CHECK(dev != NULL, return -EPROTO); id = crm_element_value(dev, F_STONITH_REMOTE_OP_ID); CRM_CHECK(id != NULL, return -EPROTO); dev = get_xpath_object("//@" F_STONITH_AVAILABLE_DEVICES, msg, LOG_ERR); CRM_CHECK(dev != NULL, return -EPROTO); crm_element_value_int(dev, F_STONITH_AVAILABLE_DEVICES, &ndevices); op = g_hash_table_lookup(stonith_remote_op_list, id); if (op == NULL) { crm_debug("Received query reply for unknown or expired operation %s", id); return -EOPNOTSUPP; } replies_expected = fencing_active_peers(); if (op->replies_expected < replies_expected) { replies_expected = op->replies_expected; } if ((++op->replies >= replies_expected) && (op->state == st_query)) { have_all_replies = TRUE; } host = crm_element_value(msg, F_ORIG); host_is_target = pcmk__str_eq(host, op->target, pcmk__str_casei); crm_info("Query result %d of %d from %s for %s/%s (%d device%s) %s", op->replies, replies_expected, host, op->target, op->action, ndevices, pcmk__plural_s(ndevices), id); if (ndevices > 0) { result = add_result(op, host, ndevices, dev); } if (pcmk_is_set(op->call_options, st_opt_topology)) { /* If we start the fencing before all the topology results are in, * it is possible fencing levels will be skipped because of the missing * query results. */ if (op->state == st_query && all_topology_devices_found(op)) { /* All the query results are in for the topology, start the fencing ops. */ crm_trace("All topology devices found"); call_remote_stonith(op, result, pcmk_ok); } else if (have_all_replies) { crm_info("All topology query replies have arrived, continuing (%d expected/%d received) ", replies_expected, op->replies); call_remote_stonith(op, NULL, pcmk_ok); } } else if (op->state == st_query) { int nverified = count_peer_devices(op, result, TRUE); /* We have a result for a non-topology fencing op that looks promising, * go ahead and start fencing before query timeout */ if (result && (host_is_target == FALSE) && nverified) { /* we have a verified device living on a peer that is not the target */ crm_trace("Found %d verified device%s", nverified, pcmk__plural_s(nverified)); call_remote_stonith(op, result, pcmk_ok); } else if (have_all_replies) { crm_info("All query replies have arrived, continuing (%d expected/%d received) ", replies_expected, op->replies); call_remote_stonith(op, NULL, pcmk_ok); } else { crm_trace("Waiting for more peer results before launching fencing operation"); } } else if (result && (op->state == st_done)) { crm_info("Discarding query result from %s (%d device%s): " "Operation is %s", result->host, result->ndevices, pcmk__plural_s(result->ndevices), stonith_op_state_str(op->state)); } return pcmk_ok; } /*! * \internal * \brief Handle a peer's reply to a fencing request * * Parse a fencing reply from XML, and either finalize the operation * or attempt another device as appropriate. * * \param[in] msg XML reply received * * \return pcmk_ok on success, -errno on error */ int process_remote_stonith_exec(xmlNode * msg) { int rc = 0; const char *id = NULL; const char *device = NULL; remote_fencing_op_t *op = NULL; xmlNode *dev = get_xpath_object("//@" F_STONITH_REMOTE_OP_ID, msg, LOG_ERR); CRM_CHECK(dev != NULL, return -EPROTO); id = crm_element_value(dev, F_STONITH_REMOTE_OP_ID); CRM_CHECK(id != NULL, return -EPROTO); dev = get_xpath_object("//@" F_STONITH_RC, msg, LOG_ERR); CRM_CHECK(dev != NULL, return -EPROTO); crm_element_value_int(dev, F_STONITH_RC, &rc); device = crm_element_value(dev, F_STONITH_DEVICE); if (stonith_remote_op_list) { op = g_hash_table_lookup(stonith_remote_op_list, id); } if (op == NULL && rc == pcmk_ok) { /* Record successful fencing operations */ const char *client_id = crm_element_value(dev, F_STONITH_CLIENTID); op = create_remote_stonith_op(client_id, dev, TRUE); } if (op == NULL) { /* Could be for an event that began before we started */ /* TODO: Record the op for later querying */ crm_info("Received peer result of unknown or expired operation %s", id); return -EOPNOTSUPP; } if (op->devices && device && !pcmk__str_eq(op->devices->data, device, pcmk__str_casei)) { crm_err("Received outdated reply for device %s (instead of %s) to " "fence (%s) %s. Operation already timed out at peer level.", device, (const char *) op->devices->data, op->action, op->target); return rc; } if (pcmk__str_eq(crm_element_value(msg, F_SUBTYPE), "broadcast", pcmk__str_casei)) { crm_debug("Finalizing action '%s' targeting %s on behalf of %s@%s: %s " CRM_XS " rc=%d id=%.8s", op->action, op->target, op->client_name, op->originator, pcmk_strerror(rc), rc, op->id); if (rc == pcmk_ok) { op->state = st_done; } else { op->state = st_failed; } remote_op_done(op, msg, rc, FALSE); return pcmk_ok; } else if (!pcmk__str_eq(op->originator, stonith_our_uname, pcmk__str_casei)) { /* If this isn't a remote level broadcast, and we are not the * originator of the operation, we should not be receiving this msg. */ crm_err("Received non-broadcast fencing result for operation %.8s " "we do not own (device %s targeting %s)", op->id, device, op->target); return rc; } if (pcmk_is_set(op->call_options, st_opt_topology)) { const char *device = crm_element_value(msg, F_STONITH_DEVICE); crm_notice("Action '%s' targeting %s using %s on behalf of %s@%s: %s " CRM_XS " rc=%d", op->action, op->target, device, op->client_name, op->originator, pcmk_strerror(rc), rc); /* We own the op, and it is complete. broadcast the result to all nodes * and notify our local clients. */ if (op->state == st_done) { remote_op_done(op, msg, rc, FALSE); return rc; } if ((op->phase == 2) && (rc != pcmk_ok)) { /* A remapped "on" failed, but the node was already turned off * successfully, so ignore the error and continue. */ crm_warn("Ignoring %s 'on' failure (exit code %d) targeting %s " "after successful 'off'", device, rc, op->target); rc = pcmk_ok; } if (rc == pcmk_ok) { /* An operation completed successfully. Try another device if * necessary, otherwise mark the operation as done. */ advance_topology_device_in_level(op, device, msg, rc); return rc; } else { /* This device failed, time to try another topology level. If no other * levels are available, mark this operation as failed and report results. */ if (advance_topology_level(op, false) != pcmk_rc_ok) { op->state = st_failed; remote_op_done(op, msg, rc, FALSE); return rc; } } } else if (rc == pcmk_ok && op->devices == NULL) { crm_trace("All done for %s", op->target); op->state = st_done; remote_op_done(op, msg, rc, FALSE); return rc; } else if (rc == -ETIME && op->devices == NULL) { /* If the operation timed out don't bother retrying other peers. */ op->state = st_failed; remote_op_done(op, msg, rc, FALSE); return rc; } else { /* fall-through and attempt other fencing action using another peer */ } /* Retry on failure */ crm_trace("Next for %s on behalf of %s@%s (rc was %d)", op->target, op->originator, op->client_name, rc); call_remote_stonith(op, NULL, rc); return rc; } gboolean stonith_check_fence_tolerance(int tolerance, const char *target, const char *action) { GHashTableIter iter; time_t now = time(NULL); remote_fencing_op_t *rop = NULL; if (tolerance <= 0 || !stonith_remote_op_list || target == NULL || action == NULL) { return FALSE; } g_hash_table_iter_init(&iter, stonith_remote_op_list); while (g_hash_table_iter_next(&iter, NULL, (void **)&rop)) { if (strcmp(rop->target, target) != 0) { continue; } else if (rop->state != st_done) { continue; /* We don't have to worry about remapped reboots here * because if state is done, any remapping has been undone */ } else if (strcmp(rop->action, action) != 0) { continue; } else if ((rop->completed + tolerance) < now) { continue; } crm_notice("Target %s was fenced (%s) less than %ds ago by %s on behalf of %s", target, action, tolerance, rop->delegate, rop->originator); return TRUE; } return FALSE; } diff --git a/daemons/pacemakerd/pcmkd_corosync.c b/daemons/pacemakerd/pcmkd_corosync.c index 4aec8706aa..d98593d22c 100644 --- a/daemons/pacemakerd/pcmkd_corosync.c +++ b/daemons/pacemakerd/pcmkd_corosync.c @@ -1,352 +1,352 @@ /* * Copyright 2010-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include "pacemakerd.h" #include #include /* for calls to stat() */ #include /* For basename() and dirname() */ #include #include /* For getpwname() */ #include #include #include #include #include #include /* for crm_ipc_is_authentic_process */ #include #include /* PCMK__SPECIAL_PID* */ static corosync_cfg_handle_t cfg_handle = 0; static mainloop_timer_t *reconnect_timer = NULL; /* =::=::=::= CFG - Shutdown stuff =::=::=::= */ static void cfg_shutdown_callback(corosync_cfg_handle_t h, corosync_cfg_shutdown_flags_t flags) { crm_info("Corosync wants to shut down: %s", (flags == COROSYNC_CFG_SHUTDOWN_FLAG_IMMEDIATE) ? "immediate" : (flags == COROSYNC_CFG_SHUTDOWN_FLAG_REGARDLESS) ? "forced" : "optional"); /* Never allow corosync to shut down while we're running */ corosync_cfg_replyto_shutdown(h, COROSYNC_CFG_SHUTDOWN_FLAG_NO); } static corosync_cfg_callbacks_t cfg_callbacks = { .corosync_cfg_shutdown_callback = cfg_shutdown_callback, }; static int pcmk_cfg_dispatch(gpointer user_data) { corosync_cfg_handle_t *handle = (corosync_cfg_handle_t *) user_data; cs_error_t rc = corosync_cfg_dispatch(*handle, CS_DISPATCH_ALL); if (rc != CS_OK) { return -1; } return 0; } static void close_cfg(void) { if (cfg_handle != 0) { #ifdef HAVE_COROSYNC_CFG_TRACKSTART /* Ideally, we would call corosync_cfg_trackstop(cfg_handle) here, but a * bug in corosync 3.1.1 and 3.1.2 makes it hang forever. Thankfully, * it's not necessary since we exit immediately after this. */ #endif corosync_cfg_finalize(cfg_handle); cfg_handle = 0; } } static gboolean cluster_reconnect_cb(gpointer data) { if (cluster_connect_cfg()) { mainloop_timer_del(reconnect_timer); reconnect_timer = NULL; crm_notice("Cluster reconnect succeeded"); } else { crm_info("Cluster reconnect failed" "(connection will be reattempted once per second)"); } /* * In theory this will continue forever. In practice the CIB connection from * attrd will timeout and shut down Pacemaker when it gets bored. */ return TRUE; } static void cfg_connection_destroy(gpointer user_data) { crm_warn("Lost connection to cluster layer " "(connection will be reattempted once per second)"); corosync_cfg_finalize(cfg_handle); cfg_handle = 0; reconnect_timer = mainloop_timer_add("corosync reconnect", 1000, TRUE, cluster_reconnect_cb, NULL); mainloop_timer_start(reconnect_timer); } void cluster_disconnect_cfg(void) { close_cfg(); if (reconnect_timer != NULL) { /* The mainloop should be gone by this point, so this isn't necessary, * but cleaning up memory should make valgrind happier. */ mainloop_timer_del(reconnect_timer); reconnect_timer = NULL; } } #define cs_repeat(counter, max, code) do { \ code; \ if(rc == CS_ERR_TRY_AGAIN || rc == CS_ERR_QUEUE_FULL) { \ counter++; \ crm_debug("Retrying Corosync operation after %ds", counter); \ sleep(counter); \ } else { \ break; \ } \ } while(counter < max) gboolean cluster_connect_cfg(void) { cs_error_t rc; int fd = -1, retries = 0, rv; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; uint32_t nodeid; static struct mainloop_fd_callbacks cfg_fd_callbacks = { .dispatch = pcmk_cfg_dispatch, .destroy = cfg_connection_destroy, }; cs_repeat(retries, 30, rc = corosync_cfg_initialize(&cfg_handle, &cfg_callbacks)); if (rc != CS_OK) { crm_crit("Could not connect to Corosync CFG: %s " CRM_XS " rc=%d", cs_strerror(rc), rc); return FALSE; } rc = corosync_cfg_fd_get(cfg_handle, &fd); if (rc != CS_OK) { crm_crit("Could not get Corosync CFG descriptor: %s " CRM_XS " rc=%d", cs_strerror(rc), rc); goto bail; } /* CFG provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_crit("Rejecting Corosync CFG provider because process %lld " "is running as uid %lld gid %lld, not root", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rv < 0) { crm_crit("Could not authenticate Corosync CFG provider: %s " CRM_XS " rc=%d", strerror(-rv), -rv); goto bail; } retries = 0; cs_repeat(retries, 30, rc = corosync_cfg_local_get(cfg_handle, &nodeid)); if (rc != CS_OK) { crm_crit("Could not get local node ID from Corosync: %s " CRM_XS " rc=%d", cs_strerror(rc), rc); goto bail; } crm_debug("Corosync reports local node ID is %lu", (unsigned long) nodeid); #ifdef HAVE_COROSYNC_CFG_TRACKSTART rc = corosync_cfg_trackstart(cfg_handle, 0); if (rc != CS_OK) { crm_crit("Could not enable Corosync CFG shutdown tracker: %s " CRM_XS " rc=%d", cs_strerror(rc), rc); goto bail; } #endif mainloop_add_fd("corosync-cfg", G_PRIORITY_DEFAULT, fd, &cfg_handle, &cfg_fd_callbacks); return TRUE; bail: corosync_cfg_finalize(cfg_handle); return FALSE; } void pcmkd_shutdown_corosync(void) { cs_error_t rc; if (cfg_handle == 0) { crm_warn("Unable to shut down Corosync: No connection"); return; } crm_info("Asking Corosync to shut down"); rc = corosync_cfg_try_shutdown(cfg_handle, COROSYNC_CFG_SHUTDOWN_FLAG_IMMEDIATE); if (rc == CS_OK) { close_cfg(); } else { crm_warn("Corosync shutdown failed: %s " CRM_XS " rc=%d", cs_strerror(rc), rc); } } /* =::=::=::= Configuration =::=::=::= */ static int get_config_opt(uint64_t unused, cmap_handle_t object_handle, const char *key, char **value, const char *fallback) { int rc = 0, retries = 0; cs_repeat(retries, 5, rc = cmap_get_string(object_handle, key, value)); if (rc != CS_OK) { crm_trace("Search for %s failed %d, defaulting to %s", key, rc, fallback); if (fallback) { *value = strdup(fallback); } else { *value = NULL; } } crm_trace("%s: %s", key, *value); return rc; } gboolean mcp_read_config(void) { cs_error_t rc = CS_OK; int retries = 0; cmap_handle_t local_handle; uint64_t config = 0; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; enum cluster_type_e stack; // There can be only one possibility do { rc = pcmk__init_cmap(&local_handle); if (rc != CS_OK) { retries++; crm_info("Could not connect to Corosync CMAP: %s (retrying in %ds) " CRM_XS " rc=%d", cs_strerror(rc), retries, rc); sleep(retries); } else { break; } } while (retries < 5); if (rc != CS_OK) { crm_crit("Could not connect to Corosync CMAP: %s " CRM_XS " rc=%d", cs_strerror(rc), rc); return FALSE; } rc = cmap_fd_get(local_handle, &fd); if (rc != CS_OK) { crm_crit("Could not get Corosync CMAP descriptor: %s " CRM_XS " rc=%d", cs_strerror(rc), rc); cmap_finalize(local_handle); return FALSE; } /* CMAP provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_crit("Rejecting Corosync CMAP provider because process %lld " "is running as uid %lld gid %lld, not root", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); cmap_finalize(local_handle); return FALSE; } else if (rv < 0) { crm_crit("Could not authenticate Corosync CMAP provider: %s " CRM_XS " rc=%d", strerror(-rv), -rv); cmap_finalize(local_handle); return FALSE; } stack = get_cluster_type(); if (stack != pcmk_cluster_corosync) { - crm_crit("Expected corosync stack but detected %s " CRM_XS " stack=%d", - name_for_cluster_type(stack), stack); + crm_crit("Expected Corosync cluster layer but detected %s " + CRM_XS " stack=%d", name_for_cluster_type(stack), stack); return FALSE; } crm_info("Reading configuration for %s stack", name_for_cluster_type(stack)); pcmk__set_env_option("cluster_type", "corosync"); pcmk__set_env_option("quorum_type", "corosync"); // If debug logging is not configured, check whether corosync has it if (pcmk__env_option("debug") == NULL) { char *debug_enabled = NULL; get_config_opt(config, local_handle, "logging.debug", &debug_enabled, "off"); if (crm_is_true(debug_enabled)) { pcmk__set_env_option("debug", "1"); if (get_crm_log_level() < LOG_DEBUG) { set_crm_log_level(LOG_DEBUG); } } else { pcmk__set_env_option("debug", "0"); } free(debug_enabled); } if(local_handle){ gid_t gid = 0; if (pcmk_daemon_user(NULL, &gid) < 0) { crm_warn("Could not authorize group with Corosync " CRM_XS " No group found for user %s", CRM_DAEMON_USER); } else { char key[PATH_MAX]; snprintf(key, PATH_MAX, "uidgid.gid.%u", gid); rc = cmap_set_uint8(local_handle, key, 1); if (rc != CS_OK) { crm_warn("Could not authorize group with Corosync: %s " CRM_XS " group=%u rc=%d", pcmk__cs_err_str(rc), gid, rc); } } } cmap_finalize(local_handle); return TRUE; } diff --git a/include/crm/lrmd_internal.h b/include/crm/lrmd_internal.h index de279a9d60..4480285589 100644 --- a/include/crm/lrmd_internal.h +++ b/include/crm/lrmd_internal.h @@ -1,76 +1,78 @@ /* * Copyright 2015-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef LRMD_INTERNAL__H #define LRMD_INTERNAL__H #include // uint32_t #include // GList, GHashTable, gpointer #include // xmlNode #include // crm_ipc_t #include // mainloop_io_t, ipc_client_callbacks #include // pcmk__output_t #include // pcmk__remote_t #include // lrmd_t, lrmd_event_data_t +int lrmd__new(lrmd_t **api, const char *nodename, const char *server, int port); + int lrmd_send_attribute_alert(lrmd_t *lrmd, GList *alert_list, const char *node, uint32_t nodeid, const char *attr_name, const char *attr_value); int lrmd_send_node_alert(lrmd_t *lrmd, GList *alert_list, const char *node, uint32_t nodeid, const char *state); int lrmd_send_fencing_alert(lrmd_t *lrmd, GList *alert_list, const char *target, const char *task, const char *desc, int op_rc); int lrmd_send_resource_alert(lrmd_t *lrmd, GList *alert_list, const char *node, lrmd_event_data_t *op); int lrmd__remote_send_xml(pcmk__remote_t *session, xmlNode *msg, uint32_t id, const char *msg_type); /* Shared functions for IPC proxy back end */ typedef struct remote_proxy_s { char *node_name; char *session_id; gboolean is_local; crm_ipc_t *ipc; mainloop_io_t *source; uint32_t last_request_id; lrmd_t *lrm; } remote_proxy_t; remote_proxy_t *remote_proxy_new(lrmd_t *lrmd, struct ipc_client_callbacks *proxy_callbacks, const char *node_name, const char *session_id, const char *channel); int lrmd__validate_remote_settings(lrmd_t *lrmd, GHashTable *hash); void remote_proxy_cb(lrmd_t *lrmd, const char *node_name, xmlNode *msg); void remote_proxy_ack_shutdown(lrmd_t *lrmd); void remote_proxy_nack_shutdown(lrmd_t *lrmd); int remote_proxy_dispatch(const char *buffer, ssize_t length, gpointer userdata); void remote_proxy_disconnected(gpointer data); void remote_proxy_free(gpointer data); void remote_proxy_relay_event(remote_proxy_t *proxy, xmlNode *msg); void remote_proxy_relay_response(remote_proxy_t *proxy, xmlNode *msg, int msg_id); void lrmd__register_messages(pcmk__output_t *out); #ifdef HAVE_GNUTLS_GNUTLS_H int lrmd__init_remote_key(gnutls_datum_t *key); #endif #endif diff --git a/include/pcmki/pcmki_sched_utils.h b/include/pcmki/pcmki_sched_utils.h index 1de341f20e..a80b470478 100644 --- a/include/pcmki/pcmki_sched_utils.h +++ b/include/pcmki/pcmki_sched_utils.h @@ -1,103 +1,107 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PENGINE_AUTILS__H # define PENGINE_AUTILS__H #include // bool #include // GList, GHashTable, gboolean, guint #include // lrmd_event_data_t #include // cib_t #include #include #include /* Constraint helper functions */ pcmk__colocation_t *invert_constraint(pcmk__colocation_t *constraint); pe__location_t *copy_constraint(pe__location_t *constraint); pe__location_t *rsc2node_new(const char *id, pe_resource_t *rsc, int weight, const char *discovery_mode, pe_node_t *node, pe_working_set_t *data_set); void pcmk__new_colocation(const char *id, const char *node_attr, int score, pe_resource_t *rsc_lh, pe_resource_t *rsc_rh, const char *state_lh, const char *state_rh, bool influence, pe_working_set_t *data_set); extern gboolean rsc_ticket_new(const char *id, pe_resource_t * rsc_lh, pe_ticket_t * ticket, const char *state_lh, const char *loss_policy, pe_working_set_t * data_set); GHashTable *pcmk__copy_node_table(GHashTable *nodes); GList *pcmk__copy_node_list(const GList *list, bool reset); GList *sort_nodes_by_weight(GList *nodes, pe_node_t *active_node, pe_working_set_t *data_set); extern gboolean can_run_resources(const pe_node_t * node); extern gboolean native_assign_node(pe_resource_t *rsc, pe_node_t *chosen, gboolean force); void native_deallocate(pe_resource_t * rsc); extern void log_action(unsigned int log_level, const char *pre_text, pe_action_t * action, gboolean details); gboolean can_run_any(GHashTable * nodes); pe_resource_t *find_compatible_child(pe_resource_t *local_child, pe_resource_t *rsc, enum rsc_role_e filter, gboolean current, pe_working_set_t *data_set); pe_resource_t *find_compatible_child_by_node(pe_resource_t * local_child, pe_node_t * local_node, pe_resource_t * rsc, enum rsc_role_e filter, gboolean current); gboolean is_child_compatible(pe_resource_t *child_rsc, pe_node_t * local_node, enum rsc_role_e filter, gboolean current); bool assign_node(pe_resource_t * rsc, pe_node_t * node, gboolean force); enum pe_action_flags summary_action_flags(pe_action_t * action, GList *children, pe_node_t * node); enum action_tasks clone_child_action(pe_action_t * action); int copies_per_node(pe_resource_t * rsc); enum filter_colocation_res { influence_nothing = 0, influence_rsc_location, influence_rsc_priority, }; extern enum filter_colocation_res filter_colocation_constraint(pe_resource_t * rsc_lh, pe_resource_t * rsc_rh, pcmk__colocation_t *constraint, gboolean preview); extern int compare_capacity(const pe_node_t * node1, const pe_node_t * node2); extern void calculate_utilization(GHashTable * current_utilization, GHashTable * utilization, gboolean plus); extern void process_utilization(pe_resource_t * rsc, pe_node_t ** prefer, pe_working_set_t * data_set); pe_action_t *create_pseudo_resource_op(pe_resource_t * rsc, const char *task, bool optional, bool runnable, pe_working_set_t *data_set); pe_action_t *pe_cancel_op(pe_resource_t *rsc, const char *name, guint interval_ms, pe_node_t *node, pe_working_set_t *data_set); pe_action_t *sched_shutdown_op(pe_node_t *node, pe_working_set_t *data_set); xmlNode *pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *event, const char *caller_version, int target_rc, const char *node, const char *origin, int level); # define LOAD_STOPPED "load_stopped" void modify_configuration( pe_working_set_t * data_set, cib_t *cib, const char *quorum, const char *watchdog, GList *node_up, GList *node_down, GList *node_fail, GList *op_inject, GList *ticket_grant, GList *ticket_revoke, GList *ticket_standby, GList *ticket_activate); int run_simulation(pe_working_set_t * data_set, cib_t *cib, GList *op_fail_list); pcmk__output_t *pcmk__new_logger(void); +bool pcmk__threshold_reached(pe_resource_t *rsc, pe_node_t *node, + pe_working_set_t *data_set, + pe_resource_t **failed); + #endif diff --git a/lib/cluster/corosync.c b/lib/cluster/corosync.c index d7bd453d9e..6e3f7420a0 100644 --- a/lib/cluster/corosync.c +++ b/lib/cluster/corosync.c @@ -1,814 +1,814 @@ /* - * Copyright 2004-2020 the Pacemaker project contributors + * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include /* U64T ~ PRIu64 */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* PCMK__SPECIAL_PID* */ #include "crmcluster_private.h" static quorum_handle_t pcmk_quorum_handle = 0; static gboolean (*quorum_app_callback)(unsigned long long seq, gboolean quorate) = NULL; /*! * \internal * \brief Get the Corosync UUID associated with a Pacemaker node * * \param[in] node Pacemaker node * * \return Newly allocated string with node's Corosync UUID, or NULL if unknown * \note It is the caller's responsibility to free the result with free(). */ char * pcmk__corosync_uuid(crm_node_t *node) { if ((node != NULL) && is_corosync_cluster()) { if (node->id > 0) { return crm_strdup_printf("%u", node->id); } else { - crm_info("Node %s is not yet known by corosync", node->uname); + crm_info("Node %s is not yet known by Corosync", node->uname); } } return NULL; } static bool node_name_is_valid(const char *key, const char *name) { int octet; if (name == NULL) { crm_trace("%s is empty", key); return false; } else if (sscanf(name, "%d.%d.%d.%d", &octet, &octet, &octet, &octet) == 4) { crm_trace("%s contains an IPv4 address (%s), ignoring", key, name); return false; } else if (strstr(name, ":") != NULL) { crm_trace("%s contains an IPv6 address (%s), ignoring", key, name); return false; } crm_trace("'%s: %s' is valid", key, name); return true; } /* * \internal * \brief Get Corosync node name corresponding to a node ID * * \param[in] cmap_handle Connection to Corosync CMAP * \param[in] nodeid Node ID to check * * \return Newly allocated string with name or (if no name) IP address * associated with first address assigned to a Corosync node ID (or NULL * if unknown) * \note It is the caller's responsibility to free the result with free(). */ char * pcmk__corosync_name(uint64_t /*cmap_handle_t */ cmap_handle, uint32_t nodeid) { // Originally based on corosync-quorumtool.c:node_name() int lpc = 0; cs_error_t rc = CS_OK; int retries = 0; char *name = NULL; cmap_handle_t local_handle = 0; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; if (nodeid == 0) { nodeid = get_local_nodeid(0); } if (cmap_handle == 0 && local_handle == 0) { retries = 0; crm_trace("Initializing CMAP connection"); do { rc = pcmk__init_cmap(&local_handle); if (rc != CS_OK) { retries++; crm_debug("API connection setup failed: %s. Retrying in %ds", cs_strerror(rc), retries); sleep(retries); } } while (retries < 5 && rc != CS_OK); if (rc != CS_OK) { crm_warn("Could not connect to Cluster Configuration Database API, error %s", cs_strerror(rc)); local_handle = 0; } } if (cmap_handle == 0) { cmap_handle = local_handle; rc = cmap_fd_get(cmap_handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CMAP API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CMAP provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_err("CMAP provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CMAP provider: %s (%d)", strerror(-rv), -rv); goto bail; } } while (name == NULL && cmap_handle != 0) { uint32_t id = 0; char *key = NULL; key = crm_strdup_printf("nodelist.node.%d.nodeid", lpc); rc = cmap_get_uint32(cmap_handle, key, &id); crm_trace("Checking %u vs %u from %s", nodeid, id, key); free(key); if (rc != CS_OK) { break; } if (nodeid == id) { crm_trace("Searching for node name for %u in nodelist.node.%d %s", nodeid, lpc, crm_str(name)); if (name == NULL) { key = crm_strdup_printf("nodelist.node.%d.name", lpc); cmap_get_string(cmap_handle, key, &name); crm_trace("%s = %s", key, crm_str(name)); free(key); } if (name == NULL) { key = crm_strdup_printf("nodelist.node.%d.ring0_addr", lpc); cmap_get_string(cmap_handle, key, &name); crm_trace("%s = %s", key, crm_str(name)); if (!node_name_is_valid(key, name)) { free(name); name = NULL; } free(key); } break; } lpc++; } bail: if(local_handle) { cmap_finalize(local_handle); } if (name == NULL) { crm_info("Unable to get node name for nodeid %u", nodeid); } return name; } /*! * \internal * \brief Disconnect from Corosync cluster * * \param[in] cluster Cluster connection to disconnect */ void pcmk__corosync_disconnect(crm_cluster_t *cluster) { cluster_disconnect_cpg(cluster); if (pcmk_quorum_handle) { quorum_finalize(pcmk_quorum_handle); pcmk_quorum_handle = 0; } crm_notice("Disconnected from Corosync"); } /*! * \internal * \brief Dispatch function for quorum connection file descriptor * * \param[in] user_data Ignored * * \return 0 on success, -1 on error (per mainloop_io_t interface) */ static int quorum_dispatch_cb(gpointer user_data) { int rc = quorum_dispatch(pcmk_quorum_handle, CS_DISPATCH_ALL); if (rc < 0) { crm_err("Connection to the Quorum API failed: %d", rc); quorum_finalize(pcmk_quorum_handle); pcmk_quorum_handle = 0; return -1; } return 0; } /*! * \internal * \brief Notification callback for Corosync quorum connection * * \param[in] handle Corosync quorum connection * \param[in] quorate Whether cluster is quorate * \param[in] ring_id Corosync ring ID * \param[in] view_list_entries Number of entries in \p view_list * \param[in] view_list Corosync node IDs in membership */ static void quorum_notification_cb(quorum_handle_t handle, uint32_t quorate, uint64_t ring_id, uint32_t view_list_entries, uint32_t *view_list) { int i; GHashTableIter iter; crm_node_t *node = NULL; static gboolean init_phase = TRUE; if (quorate != crm_have_quorum) { if (quorate) { crm_notice("Quorum acquired " CRM_XS " membership=%" U64T " members=%lu", ring_id, (long unsigned int)view_list_entries); } else { crm_warn("Quorum lost " CRM_XS " membership=%" U64T " members=%lu", ring_id, (long unsigned int)view_list_entries); } crm_have_quorum = quorate; } else { crm_info("Quorum %s " CRM_XS " membership=%" U64T " members=%lu", (quorate? "retained" : "still lost"), ring_id, (long unsigned int)view_list_entries); } if (view_list_entries == 0 && init_phase) { crm_info("Corosync membership is still forming, ignoring"); return; } init_phase = FALSE; /* Reset last_seen for all cached nodes so we can tell which ones aren't * in the view list */ g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { node->last_seen = 0; } /* Update the peer cache for each node in view list */ for (i = 0; i < view_list_entries; i++) { uint32_t id = view_list[i]; crm_debug("Member[%d] %u ", i, id); /* Get this node's peer cache entry (adding one if not already there) */ node = crm_get_peer(id, NULL); if (node->uname == NULL) { char *name = pcmk__corosync_name(0, id); crm_info("Obtaining name for new node %u", id); node = crm_get_peer(id, name); free(name); } /* Update the node state (including updating last_seen to ring_id) */ pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, ring_id); } /* Remove any peer cache entries we didn't update */ pcmk__reap_unseen_nodes(ring_id); if (quorum_app_callback) { quorum_app_callback(ring_id, quorate); } } /*! * \internal * \brief Connect to Corosync quorum service * * \param[in] dispatch Connection dispatch callback * \param[in] destroy Connection destroy callback */ void pcmk__corosync_quorum_connect(gboolean (*dispatch)(unsigned long long, gboolean), void (*destroy)(gpointer)) { cs_error_t rc; int fd = 0; int quorate = 0; uint32_t quorum_type = 0; struct mainloop_fd_callbacks quorum_fd_callbacks; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; quorum_fd_callbacks.dispatch = quorum_dispatch_cb; quorum_fd_callbacks.destroy = destroy; crm_debug("Configuring Pacemaker to obtain quorum from Corosync"); { #if 0 // New way but not supported by all Corosync 2 versions quorum_model_v0_data_t quorum_model_data = { .model = QUORUM_MODEL_V0, .quorum_notify_fn = quorum_notification_cb, }; rc = quorum_model_initialize(&pcmk_quorum_handle, QUORUM_MODEL_V0, (quorum_model_data_t *) &quorum_model_data, &quorum_type, NULL); #else quorum_callbacks_t quorum_callbacks = { .quorum_notify_fn = quorum_notification_cb, }; rc = quorum_initialize(&pcmk_quorum_handle, &quorum_callbacks, &quorum_type); #endif } if (rc != CS_OK) { crm_err("Could not connect to the Quorum API: %s (%d)", cs_strerror(rc), rc); goto bail; } else if (quorum_type != QUORUM_SET) { crm_err("Corosync quorum is not configured"); goto bail; } rc = quorum_fd_get(pcmk_quorum_handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the Quorum API connection: %s (%d)", strerror(rc), rc); goto bail; } /* Quorum provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_err("Quorum provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); rc = CS_ERR_ACCESS; goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of Quorum provider: %s (%d)", strerror(-rv), -rv); rc = CS_ERR_ACCESS; goto bail; } rc = quorum_getquorate(pcmk_quorum_handle, &quorate); if (rc != CS_OK) { crm_err("Could not obtain the current Quorum API state: %d", rc); goto bail; } if (quorate) { crm_notice("Quorum acquired"); } else { - crm_warn("Quorum lost"); + crm_warn("No quorum"); } quorum_app_callback = dispatch; crm_have_quorum = quorate; rc = quorum_trackstart(pcmk_quorum_handle, CS_TRACK_CHANGES | CS_TRACK_CURRENT); if (rc != CS_OK) { crm_err("Could not setup Quorum API notifications: %d", rc); goto bail; } mainloop_add_fd("quorum", G_PRIORITY_HIGH, fd, dispatch, &quorum_fd_callbacks); pcmk__corosync_add_nodes(NULL); bail: if (rc != CS_OK) { quorum_finalize(pcmk_quorum_handle); } } /*! * \internal * \brief Connect to Corosync cluster layer * * \param[in] cluster Initialized cluster object to connect */ gboolean pcmk__corosync_connect(crm_cluster_t *cluster) { crm_node_t *peer = NULL; enum cluster_type_e stack = get_cluster_type(); crm_peer_init(); if (stack != pcmk_cluster_corosync) { crm_err("Invalid cluster type: %s " CRM_XS " stack=%d", name_for_cluster_type(stack), stack); return FALSE; } if (!cluster_connect_cpg(cluster)) { // Error message was logged by cluster_connect_cpg() return FALSE; } crm_info("Connection to %s established", name_for_cluster_type(stack)); cluster->nodeid = get_local_nodeid(0); if (cluster->nodeid == 0) { crm_err("Could not determine local node ID"); return FALSE; } cluster->uname = get_node_name(0); if (cluster->uname == NULL) { crm_err("Could not determine local node name"); return FALSE; } // Ensure local node always exists in peer cache peer = crm_get_peer(cluster->nodeid, cluster->uname); cluster->uuid = pcmk__corosync_uuid(peer); return TRUE; } /*! * \internal * \brief Check whether a Corosync cluster is active * * \return pcmk_cluster_corosync if Corosync is found, else pcmk_cluster_unknown */ enum cluster_type_e pcmk__corosync_detect(void) { int rc = CS_OK; cmap_handle_t handle; rc = pcmk__init_cmap(&handle); switch(rc) { case CS_OK: break; case CS_ERR_SECURITY: crm_debug("Failed to initialize the cmap API: Permission denied (%d)", rc); /* It's there, we just can't talk to it. * Good enough for us to identify as 'corosync' */ return pcmk_cluster_corosync; default: crm_info("Failed to initialize the cmap API: %s (%d)", pcmk__cs_err_str(rc), rc); return pcmk_cluster_unknown; } cmap_finalize(handle); return pcmk_cluster_corosync; } /*! * \brief Check whether a Corosync cluster peer is active * * \param[in] node Node to check * * \return TRUE if \p node is an active Corosync peer, otherwise FALSE */ gboolean crm_is_corosync_peer_active(const crm_node_t *node) { if (node == NULL) { crm_trace("Corosync peer inactive: NULL"); return FALSE; } else if (!pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei)) { crm_trace("Corosync peer %s inactive: state=%s", node->uname, node->state); return FALSE; } else if (!pcmk_is_set(node->processes, crm_proc_cpg)) { crm_trace("Corosync peer %s inactive: processes=%.16x", node->uname, node->processes); return FALSE; } return TRUE; } /*! * \internal * \brief Load Corosync node list (via CMAP) into peer cache and optionally XML * * \param[in] xml_parent If not NULL, add a entry to this for each node * * \return true if any nodes were found, false otherwise */ bool pcmk__corosync_add_nodes(xmlNode *xml_parent) { int lpc = 0; cs_error_t rc = CS_OK; int retries = 0; bool any = false; cmap_handle_t cmap_handle; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; do { rc = pcmk__init_cmap(&cmap_handle); if (rc != CS_OK) { retries++; crm_debug("API connection setup failed: %s. Retrying in %ds", cs_strerror(rc), retries); sleep(retries); } } while (retries < 5 && rc != CS_OK); if (rc != CS_OK) { crm_warn("Could not connect to Cluster Configuration Database API, error %d", rc); return false; } rc = cmap_fd_get(cmap_handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CMAP API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CMAP provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_err("CMAP provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CMAP provider: %s (%d)", strerror(-rv), -rv); goto bail; } crm_peer_init(); - crm_trace("Initializing corosync nodelist"); + crm_trace("Initializing Corosync node list"); for (lpc = 0; TRUE; lpc++) { uint32_t nodeid = 0; char *name = NULL; char *key = NULL; key = crm_strdup_printf("nodelist.node.%d.nodeid", lpc); rc = cmap_get_uint32(cmap_handle, key, &nodeid); free(key); if (rc != CS_OK) { break; } name = pcmk__corosync_name(cmap_handle, nodeid); if (name != NULL) { GHashTableIter iter; crm_node_t *node = NULL; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if(node && node->uname && strcasecmp(node->uname, name) == 0) { if (node->id && node->id != nodeid) { crm_crit("Nodes %u and %u share the same name '%s': shutting down", node->id, nodeid, name); crm_exit(CRM_EX_FATAL); } } } } if (nodeid > 0 || name != NULL) { crm_trace("Initializing node[%d] %u = %s", lpc, nodeid, name); crm_get_peer(nodeid, name); } if (nodeid > 0 && name != NULL) { any = true; if (xml_parent) { xmlNode *node = create_xml_node(xml_parent, XML_CIB_TAG_NODE); crm_xml_set_id(node, "%u", nodeid); crm_xml_add(node, XML_ATTR_UNAME, name); } } free(name); } bail: cmap_finalize(cmap_handle); return any; } /*! * \internal * \brief Get cluster name from Corosync configuration (via CMAP) * * \return Newly allocated string with cluster name if configured, or NULL */ char * pcmk__corosync_cluster_name(void) { cmap_handle_t handle; char *cluster_name = NULL; cs_error_t rc = CS_OK; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; rc = pcmk__init_cmap(&handle); if (rc != CS_OK) { crm_info("Failed to initialize the cmap API: %s (%d)", cs_strerror(rc), rc); return NULL; } rc = cmap_fd_get(handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CMAP API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CMAP provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_err("CMAP provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CMAP provider: %s (%d)", strerror(-rv), -rv); goto bail; } rc = cmap_get_string(handle, "totem.cluster_name", &cluster_name); if (rc != CS_OK) { crm_info("Cannot get totem.cluster_name: %s (%d)", cs_strerror(rc), rc); } else { crm_debug("cmap totem.cluster_name = '%s'", cluster_name); } bail: cmap_finalize(handle); return cluster_name; } /*! * \internal * \brief Check (via CMAP) whether Corosync configuration has a node list * * \return true if Corosync has node list, otherwise false */ bool pcmk__corosync_has_nodelist(void) { cs_error_t cs_rc = CS_OK; int retries = 0; cmap_handle_t cmap_handle; cmap_iter_handle_t iter_handle; char key_name[CMAP_KEYNAME_MAXLEN + 1]; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rc = pcmk_ok; static bool got_result = false; static bool result = false; if (got_result) { return result; } // Connect to CMAP do { cs_rc = pcmk__init_cmap(&cmap_handle); if (cs_rc != CS_OK) { retries++; crm_debug("CMAP connection failed: %s (rc=%d, retrying in %ds)", cs_strerror(cs_rc), cs_rc, retries); sleep(retries); } } while ((retries < 5) && (cs_rc != CS_OK)); if (cs_rc != CS_OK) { crm_warn("Assuming Corosync does not have node list: " "CMAP connection failed (%s) " CRM_XS " rc=%d", cs_strerror(cs_rc), cs_rc); return false; } // Get CMAP connection file descriptor cs_rc = cmap_fd_get(cmap_handle, &fd); if (cs_rc != CS_OK) { crm_warn("Assuming Corosync does not have node list: " "CMAP unusable (%s) " CRM_XS " rc=%d", cs_strerror(cs_rc), cs_rc); goto bail; } // Check whether CMAP connection is authentic (i.e. provided by root) rc = crm_ipc_is_authentic_process(fd, (uid_t) 0, (gid_t) 0, &found_pid, &found_uid, &found_gid); if (rc == 0) { crm_warn("Assuming Corosync does not have node list: " "CMAP provider is inauthentic " CRM_XS " pid=%lld uid=%lld gid=%lld", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rc < 0) { crm_warn("Assuming Corosync does not have node list: " "Could not verify CMAP authenticity (%s) " CRM_XS " rc=%d", pcmk_strerror(rc), rc); goto bail; } // Check whether nodelist section is presetn cs_rc = cmap_iter_init(cmap_handle, "nodelist", &iter_handle); if (cs_rc != CS_OK) { crm_warn("Assuming Corosync does not have node list: " "CMAP not readable (%s) " CRM_XS " rc=%d", cs_strerror(cs_rc), cs_rc); goto bail; } cs_rc = cmap_iter_next(cmap_handle, iter_handle, key_name, NULL, NULL); if (cs_rc == CS_OK) { result = true; } cmap_iter_finalize(cmap_handle, iter_handle); got_result = true; crm_debug("Corosync %s node list", (result? "has" : "does not have")); bail: cmap_finalize(cmap_handle); return result; } diff --git a/lib/cluster/cpg.c b/lib/cluster/cpg.c index 67d26ad790..d3d137eb86 100644 --- a/lib/cluster/cpg.c +++ b/lib/cluster/cpg.c @@ -1,1076 +1,1073 @@ /* - * Copyright 2004-2020 the Pacemaker project contributors + * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* PCMK__SPECIAL_PID* */ #include "crmcluster_private.h" /* @TODO Once we can update the public API to require crm_cluster_t* in more * functions, we can ditch this in favor of cluster->cpg_handle. */ static cpg_handle_t pcmk_cpg_handle = 0; // @TODO These could be moved to crm_cluster_t* at that time as well static bool cpg_evicted = false; static GList *cs_message_queue = NULL; static int cs_message_timer = 0; struct pcmk__cpg_host_s { uint32_t id; uint32_t pid; gboolean local; enum crm_ais_msg_types type; uint32_t size; char uname[MAX_NAME]; } __attribute__ ((packed)); typedef struct pcmk__cpg_host_s pcmk__cpg_host_t; struct pcmk__cpg_msg_s { struct qb_ipc_response_header header __attribute__ ((aligned(8))); uint32_t id; gboolean is_compressed; pcmk__cpg_host_t host; pcmk__cpg_host_t sender; uint32_t size; uint32_t compressed_size; /* 584 bytes */ char data[0]; } __attribute__ ((packed)); typedef struct pcmk__cpg_msg_s pcmk__cpg_msg_t; static void crm_cs_flush(gpointer data); #define msg_data_len(msg) (msg->is_compressed?msg->compressed_size:msg->size) #define cs_repeat(rc, counter, max, code) do { \ rc = code; \ if ((rc == CS_ERR_TRY_AGAIN) || (rc == CS_ERR_QUEUE_FULL)) { \ counter++; \ crm_debug("Retrying operation after %ds", counter); \ sleep(counter); \ } else { \ break; \ } \ } while (counter < max) /*! * \brief Disconnect from Corosync CPG * * \param[in] Cluster to disconnect */ void cluster_disconnect_cpg(crm_cluster_t *cluster) { pcmk_cpg_handle = 0; if (cluster->cpg_handle) { crm_trace("Disconnecting CPG"); cpg_leave(cluster->cpg_handle, &cluster->group); cpg_finalize(cluster->cpg_handle); cluster->cpg_handle = 0; } else { crm_info("No CPG connection"); } } /*! * \brief Get the local Corosync node ID (via CPG) * * \param[in] handle CPG connection to use (or 0 to use new connection) * * \return Corosync ID of local node (or 0 if not known) */ uint32_t get_local_nodeid(cpg_handle_t handle) { cs_error_t rc = CS_OK; int retries = 0; static uint32_t local_nodeid = 0; cpg_handle_t local_handle = handle; cpg_model_v1_data_t cpg_model_info = {CPG_MODEL_V1, NULL, NULL, NULL, 0}; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; if(local_nodeid != 0) { return local_nodeid; } if(handle == 0) { crm_trace("Creating connection"); cs_repeat(rc, retries, 5, cpg_model_initialize(&local_handle, CPG_MODEL_V1, (cpg_model_data_t *)&cpg_model_info, NULL)); if (rc != CS_OK) { crm_err("Could not connect to the CPG API: %s (%d)", cs_strerror(rc), rc); return 0; } rc = cpg_fd_get(local_handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CPG API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CPG provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_err("CPG provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CPG provider: %s (%d)", strerror(-rv), -rv); goto bail; } } if (rc == CS_OK) { retries = 0; crm_trace("Performing lookup"); cs_repeat(rc, retries, 5, cpg_local_get(local_handle, &local_nodeid)); } if (rc != CS_OK) { crm_err("Could not get local node id from the CPG API: %s (%d)", pcmk__cs_err_str(rc), rc); } bail: if(handle == 0) { crm_trace("Closing connection"); cpg_finalize(local_handle); } crm_debug("Local nodeid is %u", local_nodeid); return local_nodeid; } /*! * \internal * \brief Callback function for Corosync message queue timer * * \param[in] data CPG handle * * \return FALSE (to indicate to glib that timer should not be removed) */ static gboolean crm_cs_flush_cb(gpointer data) { cs_message_timer = 0; crm_cs_flush(data); return FALSE; } // Send no more than this many CPG messages in one flush #define CS_SEND_MAX 200 /*! * \internal * \brief Send messages in Corosync CPG message queue * * \param[in] data CPG handle */ static void crm_cs_flush(gpointer data) { unsigned int sent = 0; guint queue_len = 0; cs_error_t rc = 0; cpg_handle_t *handle = (cpg_handle_t *) data; if (*handle == 0) { crm_trace("Connection is dead"); return; } queue_len = g_list_length(cs_message_queue); if (((queue_len % 1000) == 0) && (queue_len > 1)) { crm_err("CPG queue has grown to %d", queue_len); } else if (queue_len == CS_SEND_MAX) { crm_warn("CPG queue has grown to %d", queue_len); } if (cs_message_timer != 0) { /* There is already a timer, wait until it goes off */ crm_trace("Timer active %d", cs_message_timer); return; } while ((cs_message_queue != NULL) && (sent < CS_SEND_MAX)) { struct iovec *iov = cs_message_queue->data; rc = cpg_mcast_joined(*handle, CPG_TYPE_AGREED, iov, 1); if (rc != CS_OK) { break; } sent++; crm_trace("CPG message sent, size=%llu", (unsigned long long) iov->iov_len); cs_message_queue = g_list_remove(cs_message_queue, iov); free(iov->iov_base); free(iov); } queue_len -= sent; - if ((sent > 1) || (cs_message_queue != NULL)) { - crm_info("Sent %u CPG messages (%d remaining): %s (%d)", - sent, queue_len, pcmk__cs_err_str(rc), (int) rc); - } else { - crm_trace("Sent %u CPG messages (%d remaining): %s (%d)", - sent, queue_len, pcmk__cs_err_str(rc), (int) rc); - } + do_crm_log((queue_len > 5)? LOG_INFO : LOG_TRACE, + "Sent %u CPG message%s (%d still queued): %s (rc=%d)", + sent, pcmk__plural_s(sent), queue_len, pcmk__cs_err_str(rc), + (int) rc); if (cs_message_queue) { uint32_t delay_ms = 100; if (rc != CS_OK) { /* Proportionally more if sending failed but cap at 1s */ delay_ms = QB_MIN(1000, CS_SEND_MAX + (10 * queue_len)); } cs_message_timer = g_timeout_add(delay_ms, crm_cs_flush_cb, data); } } /*! * \internal * \brief Dispatch function for CPG handle * * \param[in] user_data Cluster object * * \return 0 on success, -1 on error (per mainloop_io_t interface) */ static int pcmk_cpg_dispatch(gpointer user_data) { cs_error_t rc = CS_OK; crm_cluster_t *cluster = (crm_cluster_t *) user_data; rc = cpg_dispatch(cluster->cpg_handle, CS_DISPATCH_ONE); if (rc != CS_OK) { crm_err("Connection to the CPG API failed: %s (%d)", pcmk__cs_err_str(rc), rc); cpg_finalize(cluster->cpg_handle); cluster->cpg_handle = 0; return -1; } else if (cpg_evicted) { crm_err("Evicted from CPG membership"); return -1; } return 0; } static inline const char * ais_dest(const pcmk__cpg_host_t *host) { if (host->local) { return "local"; } else if (host->size > 0) { return host->uname; } else { return ""; } } static inline const char * msg_type2text(enum crm_ais_msg_types type) { const char *text = "unknown"; switch (type) { case crm_msg_none: text = "unknown"; break; case crm_msg_ais: text = "ais"; break; case crm_msg_cib: text = "cib"; break; case crm_msg_crmd: text = "crmd"; break; case crm_msg_pe: text = "pengine"; break; case crm_msg_te: text = "tengine"; break; case crm_msg_lrmd: text = "lrmd"; break; case crm_msg_attrd: text = "attrd"; break; case crm_msg_stonithd: text = "stonithd"; break; case crm_msg_stonith_ng: text = "stonith-ng"; break; } return text; } /*! * \internal * \brief Check whether a Corosync CPG message is valid * * \param[in] msg Corosync CPG message to check * * \return true if \p msg is valid, otherwise false */ static bool check_message_sanity(const pcmk__cpg_msg_t *msg) { int32_t payload_size = msg->header.size - sizeof(pcmk__cpg_msg_t); if (payload_size < 1) { crm_err("%sCPG message %d from %s invalid: " "Claimed size of %d bytes is too small " CRM_XS " from %s[%u] to %s@%s", (msg->is_compressed? "Compressed " : ""), msg->id, ais_dest(&(msg->sender)), (int) msg->header.size, msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } if (msg->header.error != CS_OK) { crm_err("%sCPG message %d from %s invalid: " "Sender indicated error %d " CRM_XS " from %s[%u] to %s@%s", (msg->is_compressed? "Compressed " : ""), msg->id, ais_dest(&(msg->sender)), msg->header.error, msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } if (msg_data_len(msg) != payload_size) { crm_err("%sCPG message %d from %s invalid: " "Total size %d inconsistent with payload size %d " CRM_XS " from %s[%u] to %s@%s", (msg->is_compressed? "Compressed " : ""), msg->id, ais_dest(&(msg->sender)), (int) msg->header.size, (int) msg_data_len(msg), msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } if (!msg->is_compressed && /* msg->size != (strlen(msg->data) + 1) would be a stronger check, * but checking the last byte or two should be quick */ (((msg->size > 1) && (msg->data[msg->size - 2] == '\0')) || (msg->data[msg->size - 1] != '\0'))) { crm_err("CPG message %d from %s invalid: " "Payload does not end at byte %llu " CRM_XS " from %s[%u] to %s@%s", msg->id, ais_dest(&(msg->sender)), (unsigned long long) msg->size, msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } crm_trace("Verified %d-byte %sCPG message %d from %s[%u]@%s to %s@%s", (int) msg->header.size, (msg->is_compressed? "compressed " : ""), msg->id, msg_type2text(msg->sender.type), msg->sender.pid, ais_dest(&(msg->sender)), msg_type2text(msg->host.type), ais_dest(&(msg->host))); return true; } /*! * \brief Extract text data from a Corosync CPG message * * \param[in] handle CPG connection (to get local node ID if not yet known) * \param[in] nodeid Corosync ID of node that sent message * \param[in] pid Process ID of message sender (for logging only) * \param[in] content CPG message * \param[out] kind If not NULL, will be set to CPG header ID * (which should be an enum crm_ais_msg_class value, * currently always crm_class_cluster) * \param[out] from If not NULL, will be set to sender uname * (valid for the lifetime of \p content) * * \return Newly allocated string with message data * \note It is the caller's responsibility to free the return value with free(). */ char * pcmk_message_common_cs(cpg_handle_t handle, uint32_t nodeid, uint32_t pid, void *content, uint32_t *kind, const char **from) { char *data = NULL; pcmk__cpg_msg_t *msg = (pcmk__cpg_msg_t *) content; if(handle) { // Do filtering and field massaging uint32_t local_nodeid = get_local_nodeid(handle); const char *local_name = get_local_node_name(); if (msg->sender.id > 0 && msg->sender.id != nodeid) { crm_err("Nodeid mismatch from %d.%d: claimed nodeid=%u", nodeid, pid, msg->sender.id); return NULL; } else if (msg->host.id != 0 && (local_nodeid != msg->host.id)) { /* Not for us */ crm_trace("Not for us: %u != %u", msg->host.id, local_nodeid); return NULL; } else if (msg->host.size != 0 && !pcmk__str_eq(msg->host.uname, local_name, pcmk__str_casei)) { /* Not for us */ crm_trace("Not for us: %s != %s", msg->host.uname, local_name); return NULL; } msg->sender.id = nodeid; if (msg->sender.size == 0) { crm_node_t *peer = crm_get_peer(nodeid, NULL); if (peer == NULL) { crm_err("Peer with nodeid=%u is unknown", nodeid); } else if (peer->uname == NULL) { crm_err("No uname for peer with nodeid=%u", nodeid); } else { crm_notice("Fixing uname for peer with nodeid=%u", nodeid); msg->sender.size = strlen(peer->uname); memset(msg->sender.uname, 0, MAX_NAME); memcpy(msg->sender.uname, peer->uname, msg->sender.size); } } } crm_trace("Got new%s message (size=%d, %d, %d)", msg->is_compressed ? " compressed" : "", msg_data_len(msg), msg->size, msg->compressed_size); if (kind != NULL) { *kind = msg->header.id; } if (from != NULL) { *from = msg->sender.uname; } if (msg->is_compressed && msg->size > 0) { int rc = BZ_OK; char *uncompressed = NULL; unsigned int new_size = msg->size + 1; if (!check_message_sanity(msg)) { goto badmsg; } crm_trace("Decompressing message data"); uncompressed = calloc(1, new_size); rc = BZ2_bzBuffToBuffDecompress(uncompressed, &new_size, msg->data, msg->compressed_size, 1, 0); if (rc != BZ_OK) { crm_err("Decompression failed: %s " CRM_XS " bzerror=%d", bz2_strerror(rc), rc); free(uncompressed); goto badmsg; } CRM_ASSERT(rc == BZ_OK); CRM_ASSERT(new_size == msg->size); data = uncompressed; } else if (!check_message_sanity(msg)) { goto badmsg; } else { data = strdup(msg->data); } // Is this necessary? crm_get_peer(msg->sender.id, msg->sender.uname); crm_trace("Payload: %.200s", data); return data; badmsg: crm_err("Invalid message (id=%d, dest=%s:%s, from=%s:%s.%d):" " min=%d, total=%d, size=%d, bz2_size=%d", msg->id, ais_dest(&(msg->host)), msg_type2text(msg->host.type), ais_dest(&(msg->sender)), msg_type2text(msg->sender.type), msg->sender.pid, (int)sizeof(pcmk__cpg_msg_t), msg->header.size, msg->size, msg->compressed_size); free(data); return NULL; } /*! * \internal * \brief Compare cpg_address objects by node ID * * \param[in] first First cpg_address structure to compare * \param[in] second Second cpg_address structure to compare * * \return Negative number if first's node ID is lower, * positive number if first's node ID is greater, * or 0 if both node IDs are equal */ static int cmp_member_list_nodeid(const void *first, const void *second) { const struct cpg_address *const a = *((const struct cpg_address **) first), *const b = *((const struct cpg_address **) second); if (a->nodeid < b->nodeid) { return -1; } else if (a->nodeid > b->nodeid) { return 1; } /* don't bother with "reason" nor "pid" */ return 0; } /*! * \internal * \brief Get a readable string equivalent of a cpg_reason_t value * * \param[in] reason CPG reason value * * \return Readable string suitable for logging */ static const char * cpgreason2str(cpg_reason_t reason) { switch (reason) { case CPG_REASON_JOIN: return " via cpg_join"; case CPG_REASON_LEAVE: return " via cpg_leave"; case CPG_REASON_NODEDOWN: return " via cluster exit"; case CPG_REASON_NODEUP: return " via cluster join"; case CPG_REASON_PROCDOWN: return " for unknown reason"; default: break; } return ""; } /*! * \internal * \brief Get a log-friendly node name * * \param[in] peer Node to check * * \return Node's uname, or readable string if not known */ static inline const char * peer_name(crm_node_t *peer) { if (peer == NULL) { return "unknown node"; } else if (peer->uname == NULL) { return "peer node"; } else { return peer->uname; } } /*! * \brief Handle a CPG configuration change event * * \param[in] handle CPG connection * \param[in] cpg_name CPG group name * \param[in] member_list List of current CPG members * \param[in] member_list_entries Number of entries in \p member_list * \param[in] left_list List of CPG members that left * \param[in] left_list_entries Number of entries in \p left_list * \param[in] joined_list List of CPG members that joined * \param[in] joined_list_entries Number of entries in \p joined_list */ void pcmk_cpg_membership(cpg_handle_t handle, const struct cpg_name *groupName, const struct cpg_address *member_list, size_t member_list_entries, const struct cpg_address *left_list, size_t left_list_entries, const struct cpg_address *joined_list, size_t joined_list_entries) { int i; gboolean found = FALSE; static int counter = 0; uint32_t local_nodeid = get_local_nodeid(handle); const struct cpg_address *key, **sorted; sorted = malloc(member_list_entries * sizeof(const struct cpg_address *)); CRM_ASSERT(sorted != NULL); for (size_t iter = 0; iter < member_list_entries; iter++) { sorted[iter] = member_list + iter; } /* so that the cross-matching multiply-subscribed nodes is then cheap */ qsort(sorted, member_list_entries, sizeof(const struct cpg_address *), cmp_member_list_nodeid); for (i = 0; i < left_list_entries; i++) { crm_node_t *peer = pcmk__search_cluster_node_cache(left_list[i].nodeid, NULL); const struct cpg_address **rival = NULL; /* in CPG world, NODE:PROCESS-IN-MEMBERSHIP-OF-G is an 1:N relation and not playing by this rule may go wild in case of multiple residual instances of the same pacemaker daemon at the same node -- we must ensure that the possible local rival(s) won't make us cry out and bail (e.g. when they quit themselves), since all the surrounding logic denies this simple fact that the full membership is discriminated also per the PID of the process beside mere node ID (and implicitly, group ID); practically, this will be sound in terms of not preventing progress, since all the CPG joiners are also API end-point carriers, and that's what matters locally (who's the winner); remotely, we will just compare leave_list and member_list and if the left process has its node retained in member_list (under some other PID, anyway) we will just ignore it as well XXX: long-term fix is to establish in-out PID-aware tracking? */ if (peer) { key = &left_list[i]; rival = bsearch(&key, sorted, member_list_entries, sizeof(const struct cpg_address *), cmp_member_list_nodeid); } if (rival == NULL) { crm_info("Group %s event %d: %s (node %u pid %u) left%s", groupName->value, counter, peer_name(peer), left_list[i].nodeid, left_list[i].pid, cpgreason2str(left_list[i].reason)); if (peer) { crm_update_peer_proc(__func__, peer, crm_proc_cpg, OFFLINESTATUS); } } else if (left_list[i].nodeid == local_nodeid) { crm_warn("Group %s event %d: duplicate local pid %u left%s", groupName->value, counter, left_list[i].pid, cpgreason2str(left_list[i].reason)); } else { crm_warn("Group %s event %d: " "%s (node %u) duplicate pid %u left%s (%u remains)", groupName->value, counter, peer_name(peer), left_list[i].nodeid, left_list[i].pid, cpgreason2str(left_list[i].reason), (*rival)->pid); } } free(sorted); sorted = NULL; for (i = 0; i < joined_list_entries; i++) { crm_info("Group %s event %d: node %u pid %u joined%s", groupName->value, counter, joined_list[i].nodeid, joined_list[i].pid, cpgreason2str(joined_list[i].reason)); } for (i = 0; i < member_list_entries; i++) { crm_node_t *peer = crm_get_peer(member_list[i].nodeid, NULL); if (member_list[i].nodeid == local_nodeid && member_list[i].pid != getpid()) { /* see the note above */ crm_warn("Group %s event %d: detected duplicate local pid %u", groupName->value, counter, member_list[i].pid); continue; } crm_info("Group %s event %d: %s (node %u pid %u) is member", groupName->value, counter, peer_name(peer), member_list[i].nodeid, member_list[i].pid); /* If the caller left auto-reaping enabled, this will also update the * state to member. */ peer = crm_update_peer_proc(__func__, peer, crm_proc_cpg, ONLINESTATUS); if (peer && peer->state && strcmp(peer->state, CRM_NODE_MEMBER)) { /* The node is a CPG member, but we currently think it's not a * cluster member. This is possible only if auto-reaping was * disabled. The node may be joining, and we happened to get the CPG * notification before the quorum notification; or the node may have * just died, and we are processing its final messages; or a bug * has affected the peer cache. */ time_t now = time(NULL); if (peer->when_lost == 0) { // Track when we first got into this contradictory state peer->when_lost = now; } else if (now > (peer->when_lost + 60)) { // If it persists for more than a minute, update the state crm_warn("Node %u is member of group %s but was believed offline", member_list[i].nodeid, groupName->value); pcmk__update_peer_state(__func__, peer, CRM_NODE_MEMBER, 0); } } if (local_nodeid == member_list[i].nodeid) { found = TRUE; } } if (!found) { crm_err("Local node was evicted from group %s", groupName->value); cpg_evicted = true; } counter++; } /*! * \brief Connect to Corosync CPG * * \param[in] cluster Cluster object * * \return TRUE on success, otherwise FALSE */ gboolean cluster_connect_cpg(crm_cluster_t *cluster) { cs_error_t rc; int fd = -1; int retries = 0; uint32_t id = 0; crm_node_t *peer = NULL; cpg_handle_t handle = 0; const char *message_name = pcmk__message_name(crm_system_name); uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; struct mainloop_fd_callbacks cpg_fd_callbacks = { .dispatch = pcmk_cpg_dispatch, .destroy = cluster->destroy, }; cpg_model_v1_data_t cpg_model_info = { .model = CPG_MODEL_V1, .cpg_deliver_fn = cluster->cpg.cpg_deliver_fn, .cpg_confchg_fn = cluster->cpg.cpg_confchg_fn, .cpg_totem_confchg_fn = NULL, .flags = 0, }; cpg_evicted = false; cluster->group.length = 0; cluster->group.value[0] = 0; /* group.value is char[128] */ strncpy(cluster->group.value, message_name, 127); cluster->group.value[127] = 0; cluster->group.length = 1 + QB_MIN(127, strlen(cluster->group.value)); cs_repeat(rc, retries, 30, cpg_model_initialize(&handle, CPG_MODEL_V1, (cpg_model_data_t *)&cpg_model_info, NULL)); if (rc != CS_OK) { crm_err("Could not connect to the CPG API: %s (%d)", cs_strerror(rc), rc); goto bail; } rc = cpg_fd_get(handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CPG API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CPG provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_err("CPG provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); rc = CS_ERR_ACCESS; goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CPG provider: %s (%d)", strerror(-rv), -rv); rc = CS_ERR_ACCESS; goto bail; } id = get_local_nodeid(handle); if (id == 0) { crm_err("Could not get local node id from the CPG API"); goto bail; } cluster->nodeid = id; retries = 0; cs_repeat(rc, retries, 30, cpg_join(handle, &cluster->group)); if (rc != CS_OK) { crm_err("Could not join the CPG group '%s': %d", message_name, rc); goto bail; } pcmk_cpg_handle = handle; cluster->cpg_handle = handle; mainloop_add_fd("corosync-cpg", G_PRIORITY_MEDIUM, fd, cluster, &cpg_fd_callbacks); bail: if (rc != CS_OK) { cpg_finalize(handle); return FALSE; } peer = crm_get_peer(id, NULL); crm_update_peer_proc(__func__, peer, crm_proc_cpg, ONLINESTATUS); return TRUE; } /*! * \internal * \brief Send an XML message via Corosync CPG * * \param[in] msg XML message to send * \param[in] node Cluster node to send message to * \param[in] dest Type of message to send * * \return TRUE on success, otherwise FALSE */ gboolean pcmk__cpg_send_xml(xmlNode *msg, crm_node_t *node, enum crm_ais_msg_types dest) { gboolean rc = TRUE; char *data = NULL; data = dump_xml_unformatted(msg); rc = send_cluster_text(crm_class_cluster, data, FALSE, node, dest); free(data); return rc; } /*! * \internal * \brief Send string data via Corosync CPG * * \param[in] msg_class Message class (to set as CPG header ID) * \param[in] data Data to send * \param[in] local What to set as host "local" value (which is never used) * \param[in] node Cluster node to send message to * \param[in] dest Type of message to send * * \return TRUE on success, otherwise FALSE */ gboolean send_cluster_text(enum crm_ais_msg_class msg_class, const char *data, gboolean local, crm_node_t *node, enum crm_ais_msg_types dest) { static int msg_id = 0; static int local_pid = 0; static int local_name_len = 0; static const char *local_name = NULL; char *target = NULL; struct iovec *iov; pcmk__cpg_msg_t *msg = NULL; enum crm_ais_msg_types sender = text2msg_type(crm_system_name); switch (msg_class) { case crm_class_cluster: break; default: crm_err("Invalid message class: %d", msg_class); return FALSE; } CRM_CHECK(dest != crm_msg_ais, return FALSE); if (local_name == NULL) { local_name = get_local_node_name(); } if ((local_name_len == 0) && (local_name != NULL)) { local_name_len = strlen(local_name); } if (data == NULL) { data = ""; } if (local_pid == 0) { local_pid = getpid(); } if (sender == crm_msg_none) { sender = local_pid; } msg = calloc(1, sizeof(pcmk__cpg_msg_t)); msg_id++; msg->id = msg_id; msg->header.id = msg_class; msg->header.error = CS_OK; msg->host.type = dest; msg->host.local = local; if (node) { if (node->uname) { target = strdup(node->uname); msg->host.size = strlen(node->uname); memset(msg->host.uname, 0, MAX_NAME); memcpy(msg->host.uname, node->uname, msg->host.size); } else { target = crm_strdup_printf("%u", node->id); } msg->host.id = node->id; } else { target = strdup("all"); } msg->sender.id = 0; msg->sender.type = sender; msg->sender.pid = local_pid; msg->sender.size = local_name_len; memset(msg->sender.uname, 0, MAX_NAME); if ((local_name != NULL) && (msg->sender.size != 0)) { memcpy(msg->sender.uname, local_name, msg->sender.size); } msg->size = 1 + strlen(data); msg->header.size = sizeof(pcmk__cpg_msg_t) + msg->size; if (msg->size < CRM_BZ2_THRESHOLD) { msg = pcmk__realloc(msg, msg->header.size); memcpy(msg->data, data, msg->size); } else { char *compressed = NULL; unsigned int new_size = 0; char *uncompressed = strdup(data); if (pcmk__compress(uncompressed, (unsigned int) msg->size, 0, &compressed, &new_size) == pcmk_rc_ok) { msg->header.size = sizeof(pcmk__cpg_msg_t) + new_size; msg = pcmk__realloc(msg, msg->header.size); memcpy(msg->data, compressed, new_size); msg->is_compressed = TRUE; msg->compressed_size = new_size; } else { // cppcheck seems not to understand the abort logic in pcmk__realloc // cppcheck-suppress memleak msg = pcmk__realloc(msg, msg->header.size); memcpy(msg->data, data, msg->size); } free(uncompressed); free(compressed); } iov = calloc(1, sizeof(struct iovec)); iov->iov_base = msg; iov->iov_len = msg->header.size; if (msg->compressed_size) { crm_trace("Queueing CPG message %u to %s (%llu bytes, %d bytes compressed payload): %.200s", msg->id, target, (unsigned long long) iov->iov_len, msg->compressed_size, data); } else { crm_trace("Queueing CPG message %u to %s (%llu bytes, %d bytes payload): %.200s", msg->id, target, (unsigned long long) iov->iov_len, msg->size, data); } free(target); cs_message_queue = g_list_append(cs_message_queue, iov); crm_cs_flush(&pcmk_cpg_handle); return TRUE; } /*! * \brief Get the message type equivalent of a string * * \param[in] text String of message type * * \return Message type equivalent of \p text */ enum crm_ais_msg_types text2msg_type(const char *text) { int type = crm_msg_none; CRM_CHECK(text != NULL, return type); text = pcmk__message_name(text); if (pcmk__str_eq(text, "ais", pcmk__str_casei)) { type = crm_msg_ais; } else if (pcmk__str_eq(text, CRM_SYSTEM_CIB, pcmk__str_casei)) { type = crm_msg_cib; } else if (pcmk__strcase_any_of(text, CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL)) { type = crm_msg_crmd; } else if (pcmk__str_eq(text, CRM_SYSTEM_TENGINE, pcmk__str_casei)) { type = crm_msg_te; } else if (pcmk__str_eq(text, CRM_SYSTEM_PENGINE, pcmk__str_casei)) { type = crm_msg_pe; } else if (pcmk__str_eq(text, CRM_SYSTEM_LRMD, pcmk__str_casei)) { type = crm_msg_lrmd; } else if (pcmk__str_eq(text, CRM_SYSTEM_STONITHD, pcmk__str_casei)) { type = crm_msg_stonithd; } else if (pcmk__str_eq(text, "stonith-ng", pcmk__str_casei)) { type = crm_msg_stonith_ng; } else if (pcmk__str_eq(text, "attrd", pcmk__str_casei)) { type = crm_msg_attrd; } else { /* This will normally be a transient client rather than * a cluster daemon. Set the type to the pid of the client */ int scan_rc = sscanf(text, "%d", &type); if (scan_rc != 1 || type <= crm_msg_stonith_ng) { /* Ensure it's sane */ type = crm_msg_none; } } return type; } diff --git a/lib/lrmd/lrmd_client.c b/lib/lrmd/lrmd_client.c index bf4bceb423..c0cc50c83a 100644 --- a/lib/lrmd/lrmd_client.c +++ b/lib/lrmd/lrmd_client.c @@ -1,2282 +1,2344 @@ /* * Copyright 2012-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include // uint32_t, uint64_t #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef HAVE_GNUTLS_GNUTLS_H # undef KEYFILE # include #endif #include #include #include #include #include #define MAX_TLS_RECV_WAIT 10000 CRM_TRACE_INIT_DATA(lrmd); static int lrmd_api_disconnect(lrmd_t * lrmd); static int lrmd_api_is_connected(lrmd_t * lrmd); /* IPC proxy functions */ int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg); static void lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg); void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg)); #ifdef HAVE_GNUTLS_GNUTLS_H # define LRMD_CLIENT_HANDSHAKE_TIMEOUT 5000 /* 5 seconds */ gnutls_psk_client_credentials_t psk_cred_s; static void lrmd_tls_disconnect(lrmd_t * lrmd); static int global_remote_msg_id = 0; static void lrmd_tls_connection_destroy(gpointer userdata); #endif typedef struct lrmd_private_s { uint64_t type; char *token; mainloop_io_t *source; /* IPC parameters */ crm_ipc_t *ipc; pcmk__remote_t *remote; /* Extra TLS parameters */ char *remote_nodename; #ifdef HAVE_GNUTLS_GNUTLS_H char *server; int port; gnutls_psk_client_credentials_t psk_cred_c; /* while the async connection is occurring, this is the id * of the connection timeout timer. */ int async_timer; int sock; /* since tls requires a round trip across the network for a * request/reply, there are times where we just want to be able * to send a request from the client and not wait around (or even care * about) what the reply is. */ int expected_late_replies; GList *pending_notify; crm_trigger_t *process_notify; #endif lrmd_event_callback callback; /* Internal IPC proxy msg passing for remote guests */ void (*proxy_callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg); void *proxy_callback_userdata; char *peer_version; } lrmd_private_t; static lrmd_list_t * lrmd_list_add(lrmd_list_t * head, const char *value) { lrmd_list_t *p, *end; p = calloc(1, sizeof(lrmd_list_t)); p->val = strdup(value); end = head; while (end && end->next) { end = end->next; } if (end) { end->next = p; } else { head = p; } return head; } void lrmd_list_freeall(lrmd_list_t * head) { lrmd_list_t *p; while (head) { char *val = (char *)head->val; p = head->next; free(val); free(head); head = p; } } lrmd_key_value_t * lrmd_key_value_add(lrmd_key_value_t * head, const char *key, const char *value) { lrmd_key_value_t *p, *end; p = calloc(1, sizeof(lrmd_key_value_t)); p->key = strdup(key); p->value = strdup(value); end = head; while (end && end->next) { end = end->next; } if (end) { end->next = p; } else { head = p; } return head; } void lrmd_key_value_freeall(lrmd_key_value_t * head) { lrmd_key_value_t *p; while (head) { p = head->next; free(head->key); free(head->value); free(head); head = p; } } /*! * Create a new lrmd_event_data_t object * * \param[in] rsc_id ID of resource involved in event * \param[in] task Action name * \param[in] interval_ms Action interval * * \return Newly allocated and initialized lrmd_event_data_t * \note This functions asserts on memory errors, so the return value is * guaranteed to be non-NULL. The caller is responsible for freeing the * result with lrmd_free_event(). */ lrmd_event_data_t * lrmd_new_event(const char *rsc_id, const char *task, guint interval_ms) { lrmd_event_data_t *event = calloc(1, sizeof(lrmd_event_data_t)); CRM_ASSERT(event != NULL); if (rsc_id != NULL) { event->rsc_id = strdup(rsc_id); CRM_ASSERT(event->rsc_id != NULL); } if (task != NULL) { event->op_type = strdup(task); CRM_ASSERT(event->op_type != NULL); } event->interval_ms = interval_ms; return event; } lrmd_event_data_t * lrmd_copy_event(lrmd_event_data_t * event) { lrmd_event_data_t *copy = NULL; copy = calloc(1, sizeof(lrmd_event_data_t)); /* This will get all the int values. * we just have to be careful not to leave any * dangling pointers to strings. */ memcpy(copy, event, sizeof(lrmd_event_data_t)); copy->rsc_id = event->rsc_id ? strdup(event->rsc_id) : NULL; copy->op_type = event->op_type ? strdup(event->op_type) : NULL; copy->user_data = event->user_data ? strdup(event->user_data) : NULL; copy->output = event->output ? strdup(event->output) : NULL; copy->exit_reason = event->exit_reason ? strdup(event->exit_reason) : NULL; copy->remote_nodename = event->remote_nodename ? strdup(event->remote_nodename) : NULL; copy->params = pcmk__str_table_dup(event->params); return copy; } /*! * \brief Free an executor event * * \param[in] Executor event object to free */ void lrmd_free_event(lrmd_event_data_t *event) { if (event == NULL) { return; } // @TODO Why are these const char *? free((void *) event->rsc_id); free((void *) event->op_type); free((void *) event->user_data); free((void *) event->output); free((void *) event->exit_reason); free((void *) event->remote_nodename); if (event->params != NULL) { g_hash_table_destroy(event->params); } free(event); } static void lrmd_dispatch_internal(lrmd_t * lrmd, xmlNode * msg) { const char *type; const char *proxy_session = crm_element_value(msg, F_LRMD_IPC_SESSION); lrmd_private_t *native = lrmd->lrmd_private; lrmd_event_data_t event = { 0, }; if (proxy_session != NULL) { /* this is proxy business */ lrmd_internal_proxy_dispatch(lrmd, msg); return; } else if (!native->callback) { /* no callback set */ crm_trace("notify event received but client has not set callback"); return; } event.remote_nodename = native->remote_nodename; type = crm_element_value(msg, F_LRMD_OPERATION); crm_element_value_int(msg, F_LRMD_CALLID, &event.call_id); event.rsc_id = crm_element_value(msg, F_LRMD_RSC_ID); if (pcmk__str_eq(type, LRMD_OP_RSC_REG, pcmk__str_none)) { event.type = lrmd_event_register; } else if (pcmk__str_eq(type, LRMD_OP_RSC_UNREG, pcmk__str_none)) { event.type = lrmd_event_unregister; } else if (pcmk__str_eq(type, LRMD_OP_RSC_EXEC, pcmk__str_none)) { time_t epoch = 0; crm_element_value_int(msg, F_LRMD_TIMEOUT, &event.timeout); crm_element_value_ms(msg, F_LRMD_RSC_INTERVAL, &event.interval_ms); crm_element_value_int(msg, F_LRMD_RSC_START_DELAY, &event.start_delay); crm_element_value_int(msg, F_LRMD_EXEC_RC, (int *)&event.rc); crm_element_value_int(msg, F_LRMD_OP_STATUS, &event.op_status); crm_element_value_int(msg, F_LRMD_RSC_DELETED, &event.rsc_deleted); crm_element_value_epoch(msg, F_LRMD_RSC_RUN_TIME, &epoch); event.t_run = (unsigned int) epoch; crm_element_value_epoch(msg, F_LRMD_RSC_RCCHANGE_TIME, &epoch); event.t_rcchange = (unsigned int) epoch; crm_element_value_int(msg, F_LRMD_RSC_EXEC_TIME, (int *)&event.exec_time); crm_element_value_int(msg, F_LRMD_RSC_QUEUE_TIME, (int *)&event.queue_time); event.op_type = crm_element_value(msg, F_LRMD_RSC_ACTION); event.user_data = crm_element_value(msg, F_LRMD_RSC_USERDATA_STR); event.output = crm_element_value(msg, F_LRMD_RSC_OUTPUT); event.exit_reason = crm_element_value(msg, F_LRMD_RSC_EXIT_REASON); event.type = lrmd_event_exec_complete; event.params = xml2list(msg); } else if (pcmk__str_eq(type, LRMD_OP_NEW_CLIENT, pcmk__str_none)) { event.type = lrmd_event_new_client; } else if (pcmk__str_eq(type, LRMD_OP_POKE, pcmk__str_none)) { event.type = lrmd_event_poke; } else { return; } crm_trace("op %s notify event received", type); native->callback(&event); if (event.params) { g_hash_table_destroy(event.params); } } // \return Always 0, to indicate that IPC mainloop source should be kept static int lrmd_ipc_dispatch(const char *buffer, ssize_t length, gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; if (native->callback != NULL) { xmlNode *msg = string2xml(buffer); lrmd_dispatch_internal(lrmd, msg); free_xml(msg); } return 0; } #ifdef HAVE_GNUTLS_GNUTLS_H static void lrmd_free_xml(gpointer userdata) { free_xml((xmlNode *) userdata); } static bool remote_executor_connected(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; return (native->remote->tls_session != NULL); } /*! * \internal * \brief TLS dispatch function (for both trigger and file descriptor sources) * * \param[in] userdata API connection * * \return Always return a nonnegative value, which as a file descriptor * dispatch function means keep the mainloop source, and as a * trigger dispatch function, 0 means remove the trigger from the * mainloop while 1 means keep it (and job completed) */ static int lrmd_tls_dispatch(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *xml = NULL; int rc = pcmk_rc_ok; if (!remote_executor_connected(lrmd)) { crm_trace("TLS dispatch triggered after disconnect"); return 0; } crm_trace("TLS dispatch triggered"); /* First check if there are any pending notifies to process that came * while we were waiting for replies earlier. */ if (native->pending_notify) { GList *iter = NULL; crm_trace("Processing pending notifies"); for (iter = native->pending_notify; iter; iter = iter->next) { lrmd_dispatch_internal(lrmd, iter->data); } g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; } /* Next read the current buffer and see if there are any messages to handle. */ switch (pcmk__remote_ready(native->remote, 0)) { case pcmk_rc_ok: rc = pcmk__read_remote_message(native->remote, -1); xml = pcmk__remote_message_xml(native->remote); break; case ETIME: // Nothing to read, check if a full message is already in buffer xml = pcmk__remote_message_xml(native->remote); break; default: rc = ENOTCONN; break; } while (xml) { const char *msg_type = crm_element_value(xml, F_LRMD_REMOTE_MSG_TYPE); if (pcmk__str_eq(msg_type, "notify", pcmk__str_casei)) { lrmd_dispatch_internal(lrmd, xml); } else if (pcmk__str_eq(msg_type, "reply", pcmk__str_casei)) { if (native->expected_late_replies > 0) { native->expected_late_replies--; } else { int reply_id = 0; crm_element_value_int(xml, F_LRMD_CALLID, &reply_id); /* if this happens, we want to know about it */ crm_err("Got outdated Pacemaker Remote reply %d", reply_id); } } free_xml(xml); xml = pcmk__remote_message_xml(native->remote); } if (rc == ENOTCONN) { crm_info("Lost %s executor connection while reading data", (native->remote_nodename? native->remote_nodename : "local")); lrmd_tls_disconnect(lrmd); return 0; } return 1; } #endif /* Not used with mainloop */ int lrmd_poll(lrmd_t * lrmd, int timeout) { lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: return crm_ipc_ready(native->ipc); #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: if (native->pending_notify) { return 1; } else { int rc = pcmk__remote_ready(native->remote, 0); switch (rc) { case pcmk_rc_ok: return 1; case ETIME: return 0; default: return pcmk_rc2legacy(rc); } } #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); return -EPROTONOSUPPORT; } } /* Not used with mainloop */ bool lrmd_dispatch(lrmd_t * lrmd) { lrmd_private_t *private = NULL; CRM_ASSERT(lrmd != NULL); private = lrmd->lrmd_private; switch (private->type) { case pcmk__client_ipc: while (crm_ipc_ready(private->ipc)) { if (crm_ipc_read(private->ipc) > 0) { const char *msg = crm_ipc_buffer(private->ipc); lrmd_ipc_dispatch(msg, strlen(msg), lrmd); } } break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: lrmd_tls_dispatch(lrmd); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", private->type); } if (lrmd_api_is_connected(lrmd) == FALSE) { crm_err("Connection closed"); return FALSE; } return TRUE; } static xmlNode * lrmd_create_op(const char *token, const char *op, xmlNode *data, int timeout, enum lrmd_call_options options) { xmlNode *op_msg = create_xml_node(NULL, "lrmd_command"); CRM_CHECK(op_msg != NULL, return NULL); CRM_CHECK(token != NULL, return NULL); crm_xml_add(op_msg, F_XML_TAGNAME, "lrmd_command"); crm_xml_add(op_msg, F_TYPE, T_LRMD); crm_xml_add(op_msg, F_LRMD_CALLBACK_TOKEN, token); crm_xml_add(op_msg, F_LRMD_OPERATION, op); crm_xml_add_int(op_msg, F_LRMD_TIMEOUT, timeout); crm_xml_add_int(op_msg, F_LRMD_CALLOPTS, options); if (data != NULL) { add_message_xml(op_msg, F_LRMD_CALLDATA, data); } crm_trace("Created executor %s command with call options %.8lx (%d)", op, (long)options, options); return op_msg; } static void lrmd_ipc_connection_destroy(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; crm_info("IPC connection destroyed"); /* Prevent these from being cleaned up in lrmd_api_disconnect() */ native->ipc = NULL; native->source = NULL; if (native->callback) { lrmd_event_data_t event = { 0, }; event.type = lrmd_event_disconnect; event.remote_nodename = native->remote_nodename; native->callback(&event); } } #ifdef HAVE_GNUTLS_GNUTLS_H static void lrmd_tls_connection_destroy(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; crm_info("TLS connection destroyed"); if (native->remote->tls_session) { gnutls_bye(*native->remote->tls_session, GNUTLS_SHUT_RDWR); gnutls_deinit(*native->remote->tls_session); gnutls_free(native->remote->tls_session); } if (native->psk_cred_c) { gnutls_psk_free_client_credentials(native->psk_cred_c); } if (native->sock) { close(native->sock); } if (native->process_notify) { mainloop_destroy_trigger(native->process_notify); native->process_notify = NULL; } if (native->pending_notify) { g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; } free(native->remote->buffer); native->remote->buffer = NULL; native->source = 0; native->sock = 0; native->psk_cred_c = NULL; native->remote->tls_session = NULL; native->sock = 0; if (native->callback) { lrmd_event_data_t event = { 0, }; event.remote_nodename = native->remote_nodename; event.type = lrmd_event_disconnect; native->callback(&event); } return; } // \return Standard Pacemaker return code int lrmd__remote_send_xml(pcmk__remote_t *session, xmlNode *msg, uint32_t id, const char *msg_type) { crm_xml_add_int(msg, F_LRMD_REMOTE_MSG_ID, id); crm_xml_add(msg, F_LRMD_REMOTE_MSG_TYPE, msg_type); return pcmk__remote_send_xml(session, msg); } // \return Standard Pacemaker return code static int read_remote_reply(lrmd_t *lrmd, int total_timeout, int expected_reply_id, xmlNode **reply) { lrmd_private_t *native = lrmd->lrmd_private; time_t start = time(NULL); const char *msg_type = NULL; int reply_id = 0; int remaining_timeout = 0; int rc = pcmk_rc_ok; /* A timeout of 0 here makes no sense. We have to wait a period of time * for the response to come back. If -1 or 0, default to 10 seconds. */ if (total_timeout <= 0 || total_timeout > MAX_TLS_RECV_WAIT) { total_timeout = MAX_TLS_RECV_WAIT; } for (*reply = NULL; *reply == NULL; ) { *reply = pcmk__remote_message_xml(native->remote); if (*reply == NULL) { /* read some more off the tls buffer if we still have time left. */ if (remaining_timeout) { remaining_timeout = total_timeout - ((time(NULL) - start) * 1000); } else { remaining_timeout = total_timeout; } if (remaining_timeout <= 0) { return ETIME; } rc = pcmk__read_remote_message(native->remote, remaining_timeout); if (rc != pcmk_rc_ok) { return rc; } *reply = pcmk__remote_message_xml(native->remote); if (*reply == NULL) { return ENOMSG; } } crm_element_value_int(*reply, F_LRMD_REMOTE_MSG_ID, &reply_id); msg_type = crm_element_value(*reply, F_LRMD_REMOTE_MSG_TYPE); if (!msg_type) { crm_err("Empty msg type received while waiting for reply"); free_xml(*reply); *reply = NULL; } else if (pcmk__str_eq(msg_type, "notify", pcmk__str_casei)) { /* got a notify while waiting for reply, trigger the notify to be processed later */ crm_info("queueing notify"); native->pending_notify = g_list_append(native->pending_notify, *reply); if (native->process_notify) { crm_info("notify trigger set."); mainloop_set_trigger(native->process_notify); } *reply = NULL; } else if (!pcmk__str_eq(msg_type, "reply", pcmk__str_casei)) { /* msg isn't a reply, make some noise */ crm_err("Expected a reply, got %s", msg_type); free_xml(*reply); *reply = NULL; } else if (reply_id != expected_reply_id) { if (native->expected_late_replies > 0) { native->expected_late_replies--; } else { crm_err("Got outdated reply, expected id %d got id %d", expected_reply_id, reply_id); } free_xml(*reply); *reply = NULL; } } if (native->remote->buffer && native->process_notify) { mainloop_set_trigger(native->process_notify); } return rc; } // \return Standard Pacemaker return code static int send_remote_message(lrmd_t *lrmd, xmlNode *msg) { int rc = pcmk_rc_ok; lrmd_private_t *native = lrmd->lrmd_private; global_remote_msg_id++; if (global_remote_msg_id <= 0) { global_remote_msg_id = 1; } rc = lrmd__remote_send_xml(native->remote, msg, global_remote_msg_id, "request"); if (rc != pcmk_rc_ok) { crm_err("Disconnecting because TLS message could not be sent to " "Pacemaker Remote: %s", pcmk_rc_str(rc)); lrmd_tls_disconnect(lrmd); } return rc; } static int lrmd_tls_send_recv(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply) { int rc = 0; xmlNode *xml = NULL; if (!remote_executor_connected(lrmd)) { return -ENOTCONN; } rc = send_remote_message(lrmd, msg); if (rc != pcmk_rc_ok) { return pcmk_rc2legacy(rc); } rc = read_remote_reply(lrmd, timeout, global_remote_msg_id, &xml); if (rc != pcmk_rc_ok) { crm_err("Disconnecting remote after request %d reply not received: %s " CRM_XS " rc=%d timeout=%dms", global_remote_msg_id, pcmk_rc_str(rc), rc, timeout); lrmd_tls_disconnect(lrmd); } if (reply) { *reply = xml; } else { free_xml(xml); } return pcmk_rc2legacy(rc); } #endif static int lrmd_send_xml(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = crm_ipc_send(native->ipc, msg, crm_ipc_client_response, timeout, reply); break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: rc = lrmd_tls_send_recv(lrmd, msg, timeout, reply); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static int lrmd_send_xml_no_reply(lrmd_t * lrmd, xmlNode * msg) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = crm_ipc_send(native->ipc, msg, crm_ipc_flags_none, 0, NULL); break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: rc = send_remote_message(lrmd, msg); if (rc == pcmk_rc_ok) { /* we don't want to wait around for the reply, but * since the request/reply protocol needs to behave the same * as libqb, a reply will eventually come later anyway. */ native->expected_late_replies++; } rc = pcmk_rc2legacy(rc); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static int lrmd_api_is_connected(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: return crm_ipc_connected(native->ipc); #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: return remote_executor_connected(lrmd); #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); return 0; } } /*! * \internal * \brief Send a prepared API command to the executor * * \param[in] lrmd Existing connection to the executor * \param[in] op Name of API command to send * \param[in] data Command data XML to add to the sent command * \param[out] output_data If expecting a reply, it will be stored here * \param[in] timeout Timeout in milliseconds (if 0, defaults to * a sensible value per the type of connection, * standard vs. pacemaker remote); * also propagated to the command XML * \param[in] call_options Call options to pass to server when sending * \param[in] expect_reply If TRUE, wait for a reply from the server; * must be TRUE for IPC (as opposed to TLS) clients * * \return pcmk_ok on success, -errno on error */ static int lrmd_send_command(lrmd_t *lrmd, const char *op, xmlNode *data, xmlNode **output_data, int timeout, enum lrmd_call_options options, gboolean expect_reply) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *op_msg = NULL; xmlNode *op_reply = NULL; if (!lrmd_api_is_connected(lrmd)) { return -ENOTCONN; } if (op == NULL) { crm_err("No operation specified"); return -EINVAL; } CRM_CHECK(native->token != NULL,; ); crm_trace("Sending %s op to executor", op); op_msg = lrmd_create_op(native->token, op, data, timeout, options); if (op_msg == NULL) { return -EINVAL; } if (expect_reply) { rc = lrmd_send_xml(lrmd, op_msg, timeout, &op_reply); } else { rc = lrmd_send_xml_no_reply(lrmd, op_msg); goto done; } if (rc < 0) { crm_perror(LOG_ERR, "Couldn't perform %s operation (timeout=%d): %d", op, timeout, rc); rc = -ECOMM; goto done; } else if(op_reply == NULL) { rc = -ENOMSG; goto done; } rc = pcmk_ok; crm_trace("%s op reply received", op); if (crm_element_value_int(op_reply, F_LRMD_RC, &rc) != 0) { rc = -ENOMSG; goto done; } crm_log_xml_trace(op_reply, "Reply"); if (output_data) { *output_data = op_reply; op_reply = NULL; /* Prevent subsequent free */ } done: if (lrmd_api_is_connected(lrmd) == FALSE) { crm_err("Executor disconnected"); } free_xml(op_msg); free_xml(op_reply); return rc; } static int lrmd_api_poke_connection(lrmd_t * lrmd) { int rc; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *data = create_xml_node(NULL, F_LRMD_RSC); crm_xml_add(data, F_LRMD_ORIGIN, __func__); rc = lrmd_send_command(lrmd, LRMD_OP_POKE, data, NULL, 0, 0, (native->type == pcmk__client_ipc)); free_xml(data); return rc < 0 ? rc : pcmk_ok; } // \return Standard Pacemaker return code int lrmd__validate_remote_settings(lrmd_t *lrmd, GHashTable *hash) { int rc = pcmk_rc_ok; const char *value; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *data = create_xml_node(NULL, F_LRMD_OPERATION); crm_xml_add(data, F_LRMD_ORIGIN, __func__); value = g_hash_table_lookup(hash, "stonith-watchdog-timeout"); if ((value) && (stonith__watchdog_fencing_enabled_for_node(native->remote_nodename))) { crm_xml_add(data, F_LRMD_WATCHDOG, value); } rc = lrmd_send_command(lrmd, LRMD_OP_CHECK, data, NULL, 0, 0, (native->type == pcmk__client_ipc)); free_xml(data); return (rc < 0)? pcmk_legacy2rc(rc) : pcmk_rc_ok; } static int lrmd_handshake(lrmd_t * lrmd, const char *name) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *reply = NULL; xmlNode *hello = create_xml_node(NULL, "lrmd_command"); crm_xml_add(hello, F_TYPE, T_LRMD); crm_xml_add(hello, F_LRMD_OPERATION, CRM_OP_REGISTER); crm_xml_add(hello, F_LRMD_CLIENTNAME, name); crm_xml_add(hello, F_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION); /* advertise that we are a proxy provider */ if (native->proxy_callback) { crm_xml_add(hello, F_LRMD_IS_IPC_PROVIDER, "true"); } rc = lrmd_send_xml(lrmd, hello, -1, &reply); if (rc < 0) { crm_perror(LOG_DEBUG, "Couldn't complete registration with the executor API: %d", rc); rc = -ECOMM; } else if (reply == NULL) { crm_err("Did not receive registration reply"); rc = -EPROTO; } else { const char *version = crm_element_value(reply, F_LRMD_PROTOCOL_VERSION); const char *msg_type = crm_element_value(reply, F_LRMD_OPERATION); const char *tmp_ticket = crm_element_value(reply, F_LRMD_CLIENTID); crm_element_value_int(reply, F_LRMD_RC, &rc); if (rc == -EPROTO) { crm_err("Executor protocol version mismatch between client (%s) and server (%s)", LRMD_PROTOCOL_VERSION, version); crm_log_xml_err(reply, "Protocol Error"); } else if (!pcmk__str_eq(msg_type, CRM_OP_REGISTER, pcmk__str_casei)) { crm_err("Invalid registration message: %s", msg_type); crm_log_xml_err(reply, "Bad reply"); rc = -EPROTO; } else if (tmp_ticket == NULL) { crm_err("No registration token provided"); crm_log_xml_err(reply, "Bad reply"); rc = -EPROTO; } else { crm_trace("Obtained registration token: %s", tmp_ticket); native->token = strdup(tmp_ticket); native->peer_version = strdup(version?version:"1.0"); /* Included since 1.1 */ rc = pcmk_ok; } } free_xml(reply); free_xml(hello); if (rc != pcmk_ok) { lrmd_api_disconnect(lrmd); } return rc; } static int lrmd_ipc_connect(lrmd_t * lrmd, int *fd) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; struct ipc_client_callbacks lrmd_callbacks = { .dispatch = lrmd_ipc_dispatch, .destroy = lrmd_ipc_connection_destroy }; crm_info("Connecting to executor"); if (fd) { /* No mainloop */ native->ipc = crm_ipc_new(CRM_SYSTEM_LRMD, 0); if (native->ipc && crm_ipc_connect(native->ipc)) { *fd = crm_ipc_get_fd(native->ipc); } else if (native->ipc) { crm_perror(LOG_ERR, "Connection to executor failed"); rc = -ENOTCONN; } } else { native->source = mainloop_add_ipc_client(CRM_SYSTEM_LRMD, G_PRIORITY_HIGH, 0, lrmd, &lrmd_callbacks); native->ipc = mainloop_get_ipc_client(native->source); } if (native->ipc == NULL) { crm_debug("Could not connect to the executor API"); rc = -ENOTCONN; } return rc; } #ifdef HAVE_GNUTLS_GNUTLS_H static void copy_gnutls_datum(gnutls_datum_t *dest, gnutls_datum_t *source) { dest->data = gnutls_malloc(source->size); CRM_ASSERT(dest->data); memcpy(dest->data, source->data, source->size); dest->size = source->size; } static void clear_gnutls_datum(gnutls_datum_t *datum) { gnutls_free(datum->data); datum->data = NULL; datum->size = 0; } #define KEY_READ_LEN 256 // Chunk size for reading key from file // \return Standard Pacemaker return code static int read_gnutls_key(const char *location, gnutls_datum_t *key) { FILE *stream = NULL; size_t buf_len = KEY_READ_LEN; if ((location == NULL) || (key == NULL)) { return EINVAL; } stream = fopen(location, "r"); if (stream == NULL) { return errno; } key->data = gnutls_malloc(buf_len); key->size = 0; while (!feof(stream)) { int next = fgetc(stream); if (next == EOF) { if (!feof(stream)) { crm_warn("Pacemaker Remote key read was partially successful " "(copy in memory may be corrupted)"); } break; } if (key->size == buf_len) { buf_len = key->size + KEY_READ_LEN; key->data = gnutls_realloc(key->data, buf_len); CRM_ASSERT(key->data); } key->data[key->size++] = (unsigned char) next; } fclose(stream); if (key->size == 0) { clear_gnutls_datum(key); return ENOKEY; } return pcmk_rc_ok; } // Cache the most recently used Pacemaker Remote authentication key struct key_cache_s { time_t updated; // When cached key was read (valid for 1 minute) const char *location; // Where cached key was read from gnutls_datum_t key; // Cached key }; static bool key_is_cached(struct key_cache_s *key_cache) { return key_cache->updated != 0; } static bool key_cache_expired(struct key_cache_s *key_cache) { return (time(NULL) - key_cache->updated) >= 60; } static void clear_key_cache(struct key_cache_s *key_cache) { clear_gnutls_datum(&(key_cache->key)); if ((key_cache->updated != 0) || (key_cache->location != NULL)) { key_cache->updated = 0; key_cache->location = NULL; crm_debug("Cleared Pacemaker Remote key cache"); } } static void get_cached_key(struct key_cache_s *key_cache, gnutls_datum_t *key) { copy_gnutls_datum(key, &(key_cache->key)); crm_debug("Using cached Pacemaker Remote key from %s", crm_str(key_cache->location)); } static void cache_key(struct key_cache_s *key_cache, gnutls_datum_t *key, const char *location) { key_cache->updated = time(NULL); key_cache->location = location; copy_gnutls_datum(&(key_cache->key), key); crm_debug("Using (and cacheing) Pacemaker Remote key from %s", crm_str(location)); } /*! * \internal * \brief Get Pacemaker Remote authentication key from file or cache * * \param[in] location Path to key file to try (this memory must * persist across all calls of this function) * \param[out] key Key from location or cache * * \return Standard Pacemaker return code */ static int get_remote_key(const char *location, gnutls_datum_t *key) { static struct key_cache_s key_cache = { 0, }; int rc = pcmk_rc_ok; if ((location == NULL) || (key == NULL)) { return EINVAL; } if (key_is_cached(&key_cache)) { if (key_cache_expired(&key_cache)) { clear_key_cache(&key_cache); } else { get_cached_key(&key_cache, key); return pcmk_rc_ok; } } rc = read_gnutls_key(location, key); if (rc != pcmk_rc_ok) { return rc; } cache_key(&key_cache, key, location); return pcmk_rc_ok; } /*! * \internal * \brief Initialize the Pacemaker Remote authentication key * * Try loading the Pacemaker Remote authentication key from cache if available, * otherwise from these locations, in order of preference: the value of the * PCMK_authkey_location environment variable, if set; the Pacemaker default key * file location; or (for historical reasons) /etc/corosync/authkey. * * \param[out] key Where to store key * * \return Standard Pacemaker return code */ int lrmd__init_remote_key(gnutls_datum_t *key) { static const char *env_location = NULL; static bool need_env = true; int env_rc = pcmk_rc_ok; int default_rc = pcmk_rc_ok; int alt_rc = pcmk_rc_ok; bool env_is_default = false; bool env_is_fallback = false; if (need_env) { env_location = getenv("PCMK_authkey_location"); need_env = false; } // Try location in environment variable, if set if (env_location != NULL) { env_rc = get_remote_key(env_location, key); if (env_rc == pcmk_rc_ok) { return pcmk_rc_ok; } env_is_default = !strcmp(env_location, DEFAULT_REMOTE_KEY_LOCATION); env_is_fallback = !strcmp(env_location, ALT_REMOTE_KEY_LOCATION); /* @TODO It would be more secure to fail, rather than fall back to the * default, if an explicitly set key location is not readable, and it * would be better to never use the Corosync location as a fallback. * However, that would break any deployments currently working with the * fallbacks. */ } // Try default location, if environment wasn't explicitly set to it if (env_is_default) { default_rc = env_rc; } else { default_rc = get_remote_key(DEFAULT_REMOTE_KEY_LOCATION, key); } // Try fallback location, if environment wasn't set to it and default failed if (env_is_fallback) { alt_rc = env_rc; } else if (default_rc != pcmk_rc_ok) { alt_rc = get_remote_key(ALT_REMOTE_KEY_LOCATION, key); } // We have all results, so log and return if ((env_rc != pcmk_rc_ok) && (default_rc != pcmk_rc_ok) && (alt_rc != pcmk_rc_ok)) { // Environment set, everything failed crm_warn("Could not read Pacemaker Remote key from %s (%s%s%s%s%s): %s", env_location, env_is_default? "" : "or default location ", env_is_default? "" : DEFAULT_REMOTE_KEY_LOCATION, !env_is_default && !env_is_fallback? " " : "", env_is_fallback? "" : "or fallback location ", env_is_fallback? "" : ALT_REMOTE_KEY_LOCATION, pcmk_rc_str(env_rc)); return ENOKEY; } if (env_rc != pcmk_rc_ok) { // Environment set but failed, using a default crm_warn("Could not read Pacemaker Remote key from %s " "(using %s location %s instead): %s", env_location, (default_rc == pcmk_rc_ok)? "default" : "fallback", (default_rc == pcmk_rc_ok)? DEFAULT_REMOTE_KEY_LOCATION : ALT_REMOTE_KEY_LOCATION, pcmk_rc_str(env_rc)); return pcmk_rc_ok; } if ((default_rc != pcmk_rc_ok) && (alt_rc != pcmk_rc_ok)) { // Environment unset, defaults failed crm_warn("Could not read Pacemaker Remote key from default location %s" " (or fallback location %s): %s", DEFAULT_REMOTE_KEY_LOCATION, ALT_REMOTE_KEY_LOCATION, pcmk_rc_str(default_rc)); return ENOKEY; } return pcmk_rc_ok; // Environment variable unset, a default worked } static void lrmd_gnutls_global_init(void) { static int gnutls_init = 0; if (!gnutls_init) { crm_gnutls_global_init(); } gnutls_init = 1; } #endif static void report_async_connection_result(lrmd_t * lrmd, int rc) { lrmd_private_t *native = lrmd->lrmd_private; if (native->callback) { lrmd_event_data_t event = { 0, }; event.type = lrmd_event_connect; event.remote_nodename = native->remote_nodename; event.connection_rc = rc; native->callback(&event); } } #ifdef HAVE_GNUTLS_GNUTLS_H static inline int lrmd__tls_client_handshake(pcmk__remote_t *remote) { return pcmk__tls_client_handshake(remote, LRMD_CLIENT_HANDSHAKE_TIMEOUT); } /*! * \internal * \brief Add trigger and file descriptor mainloop sources for TLS * * \param[in] lrmd API connection with established TLS session * \param[in] do_handshake Whether to perform executor handshake * * \return Standard Pacemaker return code */ static int add_tls_to_mainloop(lrmd_t *lrmd, bool do_handshake) { lrmd_private_t *native = lrmd->lrmd_private; int rc = pcmk_rc_ok; char *name = crm_strdup_printf("pacemaker-remote-%s:%d", native->server, native->port); struct mainloop_fd_callbacks tls_fd_callbacks = { .dispatch = lrmd_tls_dispatch, .destroy = lrmd_tls_connection_destroy, }; native->process_notify = mainloop_add_trigger(G_PRIORITY_HIGH, lrmd_tls_dispatch, lrmd); native->source = mainloop_add_fd(name, G_PRIORITY_HIGH, native->sock, lrmd, &tls_fd_callbacks); /* Async connections lose the client name provided by the API caller, so we * have to use our generated name here to perform the executor handshake. * * @TODO Keep track of the caller-provided name. Perhaps we should be using * that name in this function instead of generating one anyway. */ if (do_handshake) { rc = lrmd_handshake(lrmd, name); rc = pcmk_legacy2rc(rc); } free(name); return rc; } static void lrmd_tcp_connect_cb(void *userdata, int rc, int sock) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; gnutls_datum_t psk_key = { NULL, 0 }; native->async_timer = 0; if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); crm_info("Could not connect to Pacemaker Remote at %s:%d: %s " CRM_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); return; } /* The TCP connection was successful, so establish the TLS connection. * @TODO make this async to avoid blocking code in client */ native->sock = sock; rc = lrmd__init_remote_key(&psk_key); if (rc != pcmk_rc_ok) { crm_info("Could not connect to Pacemaker Remote at %s:%d: %s " CRM_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); return; } gnutls_psk_allocate_client_credentials(&native->psk_cred_c); gnutls_psk_set_client_credentials(native->psk_cred_c, DEFAULT_REMOTE_USERNAME, &psk_key, GNUTLS_PSK_KEY_RAW); gnutls_free(psk_key.data); native->remote->tls_session = pcmk__new_tls_session(sock, GNUTLS_CLIENT, GNUTLS_CRD_PSK, native->psk_cred_c); if (native->remote->tls_session == NULL) { lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, -EPROTO); return; } if (lrmd__tls_client_handshake(native->remote) != pcmk_rc_ok) { crm_warn("Disconnecting after TLS handshake with Pacemaker Remote server %s:%d failed", native->server, native->port); gnutls_deinit(*native->remote->tls_session); gnutls_free(native->remote->tls_session); native->remote->tls_session = NULL; lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, -EKEYREJECTED); return; } crm_info("TLS connection to Pacemaker Remote server %s:%d succeeded", native->server, native->port); rc = add_tls_to_mainloop(lrmd, true); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); } static int lrmd_tls_connect_async(lrmd_t * lrmd, int timeout /*ms */ ) { int rc; int timer_id = 0; lrmd_private_t *native = lrmd->lrmd_private; lrmd_gnutls_global_init(); native->sock = -1; rc = pcmk__connect_remote(native->server, native->port, timeout, &timer_id, &(native->sock), lrmd, lrmd_tcp_connect_cb); if (rc != pcmk_rc_ok) { crm_warn("Pacemaker Remote connection to %s:%s failed: %s " CRM_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); return pcmk_rc2legacy(rc); } native->async_timer = timer_id; return pcmk_ok; } static int lrmd_tls_connect(lrmd_t * lrmd, int *fd) { int rc; lrmd_private_t *native = lrmd->lrmd_private; gnutls_datum_t psk_key = { NULL, 0 }; lrmd_gnutls_global_init(); native->sock = -1; rc = pcmk__connect_remote(native->server, native->port, 0, NULL, &(native->sock), NULL, NULL); if (rc != pcmk_rc_ok) { crm_warn("Pacemaker Remote connection to %s:%s failed: %s " CRM_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); lrmd_tls_connection_destroy(lrmd); return -ENOTCONN; } rc = lrmd__init_remote_key(&psk_key); if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); return pcmk_rc2legacy(rc); } gnutls_psk_allocate_client_credentials(&native->psk_cred_c); gnutls_psk_set_client_credentials(native->psk_cred_c, DEFAULT_REMOTE_USERNAME, &psk_key, GNUTLS_PSK_KEY_RAW); gnutls_free(psk_key.data); native->remote->tls_session = pcmk__new_tls_session(native->sock, GNUTLS_CLIENT, GNUTLS_CRD_PSK, native->psk_cred_c); if (native->remote->tls_session == NULL) { lrmd_tls_connection_destroy(lrmd); return -EPROTO; } if (lrmd__tls_client_handshake(native->remote) != pcmk_rc_ok) { crm_err("Session creation for %s:%d failed", native->server, native->port); gnutls_deinit(*native->remote->tls_session); gnutls_free(native->remote->tls_session); native->remote->tls_session = NULL; lrmd_tls_connection_destroy(lrmd); return -EKEYREJECTED; } crm_info("Client TLS connection established with Pacemaker Remote server %s:%d", native->server, native->port); if (fd) { *fd = native->sock; } else { add_tls_to_mainloop(lrmd, false); } return pcmk_ok; } #endif static int lrmd_api_connect(lrmd_t * lrmd, const char *name, int *fd) { int rc = -ENOTCONN; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = lrmd_ipc_connect(lrmd, fd); break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: rc = lrmd_tls_connect(lrmd, fd); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } if (rc == pcmk_ok) { rc = lrmd_handshake(lrmd, name); } return rc; } static int lrmd_api_connect_async(lrmd_t * lrmd, const char *name, int timeout) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; CRM_CHECK(native && native->callback, return -EINVAL); switch (native->type) { case pcmk__client_ipc: /* fake async connection with ipc. it should be fast * enough that we gain very little from async */ rc = lrmd_api_connect(lrmd, name, NULL); if (!rc) { report_async_connection_result(lrmd, rc); } break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: rc = lrmd_tls_connect_async(lrmd, timeout); if (rc) { /* connection failed, report rc now */ report_async_connection_result(lrmd, rc); } break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static void lrmd_ipc_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; native->ipc = NULL; } else if (native->ipc) { /* Not attached to mainloop */ crm_ipc_t *ipc = native->ipc; native->ipc = NULL; crm_ipc_close(ipc); crm_ipc_destroy(ipc); } } #ifdef HAVE_GNUTLS_GNUTLS_H static void lrmd_tls_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->remote->tls_session) { gnutls_bye(*native->remote->tls_session, GNUTLS_SHUT_RDWR); gnutls_deinit(*native->remote->tls_session); gnutls_free(native->remote->tls_session); native->remote->tls_session = 0; } if (native->async_timer) { g_source_remove(native->async_timer); native->async_timer = 0; } if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; } else if (native->sock) { close(native->sock); native->sock = 0; } if (native->pending_notify) { g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; } } #endif static int lrmd_api_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; int rc = pcmk_ok; crm_info("Disconnecting %s %s executor connection", pcmk__client_type_str(native->type), (native->remote_nodename? native->remote_nodename : "local")); switch (native->type) { case pcmk__client_ipc: lrmd_ipc_disconnect(lrmd); break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: lrmd_tls_disconnect(lrmd); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } free(native->token); native->token = NULL; free(native->peer_version); native->peer_version = NULL; return rc; } static int lrmd_api_register_rsc(lrmd_t * lrmd, const char *rsc_id, const char *class, const char *provider, const char *type, enum lrmd_call_options options) { int rc = pcmk_ok; xmlNode *data = NULL; if (!class || !type || !rsc_id) { return -EINVAL; } if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider) && (provider == NULL)) { return -EINVAL; } data = create_xml_node(NULL, F_LRMD_RSC); crm_xml_add(data, F_LRMD_ORIGIN, __func__); crm_xml_add(data, F_LRMD_RSC_ID, rsc_id); crm_xml_add(data, F_LRMD_CLASS, class); crm_xml_add(data, F_LRMD_PROVIDER, provider); crm_xml_add(data, F_LRMD_TYPE, type); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_REG, data, NULL, 0, options, TRUE); free_xml(data); return rc; } static int lrmd_api_unregister_rsc(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options) { int rc = pcmk_ok; xmlNode *data = create_xml_node(NULL, F_LRMD_RSC); crm_xml_add(data, F_LRMD_ORIGIN, __func__); crm_xml_add(data, F_LRMD_RSC_ID, rsc_id); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_UNREG, data, NULL, 0, options, TRUE); free_xml(data); return rc; } lrmd_rsc_info_t * lrmd_new_rsc_info(const char *rsc_id, const char *standard, const char *provider, const char *type) { lrmd_rsc_info_t *rsc_info = calloc(1, sizeof(lrmd_rsc_info_t)); CRM_ASSERT(rsc_info); if (rsc_id) { rsc_info->id = strdup(rsc_id); CRM_ASSERT(rsc_info->id); } if (standard) { rsc_info->standard = strdup(standard); CRM_ASSERT(rsc_info->standard); } if (provider) { rsc_info->provider = strdup(provider); CRM_ASSERT(rsc_info->provider); } if (type) { rsc_info->type = strdup(type); CRM_ASSERT(rsc_info->type); } return rsc_info; } lrmd_rsc_info_t * lrmd_copy_rsc_info(lrmd_rsc_info_t * rsc_info) { return lrmd_new_rsc_info(rsc_info->id, rsc_info->standard, rsc_info->provider, rsc_info->type); } void lrmd_free_rsc_info(lrmd_rsc_info_t * rsc_info) { if (!rsc_info) { return; } free(rsc_info->id); free(rsc_info->type); free(rsc_info->standard); free(rsc_info->provider); free(rsc_info); } static lrmd_rsc_info_t * lrmd_api_get_rsc_info(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options) { lrmd_rsc_info_t *rsc_info = NULL; xmlNode *data = create_xml_node(NULL, F_LRMD_RSC); xmlNode *output = NULL; const char *class = NULL; const char *provider = NULL; const char *type = NULL; crm_xml_add(data, F_LRMD_ORIGIN, __func__); crm_xml_add(data, F_LRMD_RSC_ID, rsc_id); lrmd_send_command(lrmd, LRMD_OP_RSC_INFO, data, &output, 0, options, TRUE); free_xml(data); if (!output) { return NULL; } class = crm_element_value(output, F_LRMD_CLASS); provider = crm_element_value(output, F_LRMD_PROVIDER); type = crm_element_value(output, F_LRMD_TYPE); if (!class || !type) { free_xml(output); return NULL; } else if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider) && !provider) { free_xml(output); return NULL; } rsc_info = lrmd_new_rsc_info(rsc_id, class, provider, type); free_xml(output); return rsc_info; } void lrmd_free_op_info(lrmd_op_info_t *op_info) { if (op_info) { free(op_info->rsc_id); free(op_info->action); free(op_info->interval_ms_s); free(op_info->timeout_ms_s); free(op_info); } } static int lrmd_api_get_recurring_ops(lrmd_t *lrmd, const char *rsc_id, int timeout_ms, enum lrmd_call_options options, GList **output) { xmlNode *data = NULL; xmlNode *output_xml = NULL; int rc = pcmk_ok; if (output == NULL) { return -EINVAL; } *output = NULL; // Send request if (rsc_id) { data = create_xml_node(NULL, F_LRMD_RSC); crm_xml_add(data, F_LRMD_ORIGIN, __func__); crm_xml_add(data, F_LRMD_RSC_ID, rsc_id); } rc = lrmd_send_command(lrmd, LRMD_OP_GET_RECURRING, data, &output_xml, timeout_ms, options, TRUE); if (data) { free_xml(data); } // Process reply if ((rc != pcmk_ok) || (output_xml == NULL)) { return rc; } for (xmlNode *rsc_xml = first_named_child(output_xml, F_LRMD_RSC); (rsc_xml != NULL) && (rc == pcmk_ok); rsc_xml = crm_next_same_xml(rsc_xml)) { rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); if (rsc_id == NULL) { crm_err("Could not parse recurring operation information from executor"); continue; } for (xmlNode *op_xml = first_named_child(rsc_xml, T_LRMD_RSC_OP); op_xml != NULL; op_xml = crm_next_same_xml(op_xml)) { lrmd_op_info_t *op_info = calloc(1, sizeof(lrmd_op_info_t)); if (op_info == NULL) { rc = -ENOMEM; break; } op_info->rsc_id = strdup(rsc_id); op_info->action = crm_element_value_copy(op_xml, F_LRMD_RSC_ACTION); op_info->interval_ms_s = crm_element_value_copy(op_xml, F_LRMD_RSC_INTERVAL); op_info->timeout_ms_s = crm_element_value_copy(op_xml, F_LRMD_TIMEOUT); *output = g_list_prepend(*output, op_info); } } free_xml(output_xml); return rc; } static void lrmd_api_set_callback(lrmd_t * lrmd, lrmd_event_callback callback) { lrmd_private_t *native = lrmd->lrmd_private; native->callback = callback; } void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg)) { lrmd_private_t *native = lrmd->lrmd_private; native->proxy_callback = callback; native->proxy_callback_userdata = userdata; } void lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg) { lrmd_private_t *native = lrmd->lrmd_private; if (native->proxy_callback) { crm_log_xml_trace(msg, "PROXY_INBOUND"); native->proxy_callback(lrmd, native->proxy_callback_userdata, msg); } } int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg) { if (lrmd == NULL) { return -ENOTCONN; } crm_xml_add(msg, F_LRMD_OPERATION, CRM_OP_IPC_FWD); crm_log_xml_trace(msg, "PROXY_OUTBOUND"); return lrmd_send_xml_no_reply(lrmd, msg); } static int stonith_get_metadata(const char *provider, const char *type, char **output) { int rc = pcmk_ok; stonith_t *stonith_api = stonith_api_new(); if (stonith_api == NULL) { crm_err("Could not get fence agent meta-data: API memory allocation failed"); return -ENOMEM; } rc = stonith_api->cmds->metadata(stonith_api, st_opt_sync_call, type, provider, output, 0); if ((rc == pcmk_ok) && (*output == NULL)) { rc = -EIO; } stonith_api->cmds->free(stonith_api); return rc; } static int lrmd_api_get_metadata(lrmd_t *lrmd, const char *standard, const char *provider, const char *type, char **output, enum lrmd_call_options options) { return lrmd->cmds->get_metadata_params(lrmd, standard, provider, type, output, options, NULL); } static int lrmd_api_get_metadata_params(lrmd_t *lrmd, const char *standard, const char *provider, const char *type, char **output, enum lrmd_call_options options, lrmd_key_value_t *params) { svc_action_t *action = NULL; GHashTable *params_table = NULL; if (!standard || !type) { lrmd_key_value_freeall(params); return -EINVAL; } if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { lrmd_key_value_freeall(params); return stonith_get_metadata(provider, type, output); } params_table = pcmk__strkey_table(free, free); for (const lrmd_key_value_t *param = params; param; param = param->next) { g_hash_table_insert(params_table, strdup(param->key), strdup(param->value)); } action = resources_action_create(type, standard, provider, type, CRMD_ACTION_METADATA, 0, CRMD_METADATA_CALL_TIMEOUT, params_table, 0); lrmd_key_value_freeall(params); if (action == NULL) { crm_err("Unable to retrieve meta-data for %s:%s:%s", standard, provider, type); return -EINVAL; } if (!services_action_sync(action)) { crm_err("Failed to retrieve meta-data for %s:%s:%s", standard, provider, type); services_action_free(action); return -EIO; } if (!action->stdout_data) { crm_err("Failed to receive meta-data for %s:%s:%s", standard, provider, type); services_action_free(action); return -EIO; } *output = strdup(action->stdout_data); services_action_free(action); return pcmk_ok; } static int lrmd_api_exec(lrmd_t *lrmd, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout, /* ms */ int start_delay, /* ms */ enum lrmd_call_options options, lrmd_key_value_t * params) { int rc = pcmk_ok; xmlNode *data = create_xml_node(NULL, F_LRMD_RSC); xmlNode *args = create_xml_node(data, XML_TAG_ATTRS); lrmd_key_value_t *tmp = NULL; crm_xml_add(data, F_LRMD_ORIGIN, __func__); crm_xml_add(data, F_LRMD_RSC_ID, rsc_id); crm_xml_add(data, F_LRMD_RSC_ACTION, action); crm_xml_add(data, F_LRMD_RSC_USERDATA_STR, userdata); crm_xml_add_ms(data, F_LRMD_RSC_INTERVAL, interval_ms); crm_xml_add_int(data, F_LRMD_TIMEOUT, timeout); crm_xml_add_int(data, F_LRMD_RSC_START_DELAY, start_delay); for (tmp = params; tmp; tmp = tmp->next) { hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args); } rc = lrmd_send_command(lrmd, LRMD_OP_RSC_EXEC, data, NULL, timeout, options, TRUE); free_xml(data); lrmd_key_value_freeall(params); return rc; } /* timeout is in ms */ static int lrmd_api_exec_alert(lrmd_t *lrmd, const char *alert_id, const char *alert_path, int timeout, lrmd_key_value_t *params) { int rc = pcmk_ok; xmlNode *data = create_xml_node(NULL, F_LRMD_ALERT); xmlNode *args = create_xml_node(data, XML_TAG_ATTRS); lrmd_key_value_t *tmp = NULL; crm_xml_add(data, F_LRMD_ORIGIN, __func__); crm_xml_add(data, F_LRMD_ALERT_ID, alert_id); crm_xml_add(data, F_LRMD_ALERT_PATH, alert_path); crm_xml_add_int(data, F_LRMD_TIMEOUT, timeout); for (tmp = params; tmp; tmp = tmp->next) { hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args); } rc = lrmd_send_command(lrmd, LRMD_OP_ALERT_EXEC, data, NULL, timeout, lrmd_opt_notify_orig_only, TRUE); free_xml(data); lrmd_key_value_freeall(params); return rc; } static int lrmd_api_cancel(lrmd_t *lrmd, const char *rsc_id, const char *action, guint interval_ms) { int rc = pcmk_ok; xmlNode *data = create_xml_node(NULL, F_LRMD_RSC); crm_xml_add(data, F_LRMD_ORIGIN, __func__); crm_xml_add(data, F_LRMD_RSC_ACTION, action); crm_xml_add(data, F_LRMD_RSC_ID, rsc_id); crm_xml_add_ms(data, F_LRMD_RSC_INTERVAL, interval_ms); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_CANCEL, data, NULL, 0, 0, TRUE); free_xml(data); return rc; } static int list_stonith_agents(lrmd_list_t ** resources) { int rc = 0; stonith_t *stonith_api = stonith_api_new(); stonith_key_value_t *stonith_resources = NULL; stonith_key_value_t *dIter = NULL; if (stonith_api == NULL) { crm_err("Could not list fence agents: API memory allocation failed"); return -ENOMEM; } stonith_api->cmds->list_agents(stonith_api, st_opt_sync_call, NULL, &stonith_resources, 0); stonith_api->cmds->free(stonith_api); for (dIter = stonith_resources; dIter; dIter = dIter->next) { rc++; if (resources) { *resources = lrmd_list_add(*resources, dIter->value); } } stonith_key_value_freeall(stonith_resources, 1, 0); return rc; } static int lrmd_api_list_agents(lrmd_t * lrmd, lrmd_list_t ** resources, const char *class, const char *provider) { int rc = 0; int stonith_count = 0; // Initially, whether to include stonith devices if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { stonith_count = 1; } else { GList *gIter = NULL; GList *agents = resources_list_agents(class, provider); for (gIter = agents; gIter != NULL; gIter = gIter->next) { *resources = lrmd_list_add(*resources, (const char *)gIter->data); rc++; } g_list_free_full(agents, free); if (!class) { stonith_count = 1; } } if (stonith_count) { // Now, if stonith devices are included, how many there are stonith_count = list_stonith_agents(resources); if (stonith_count > 0) { rc += stonith_count; } } if (rc == 0) { crm_notice("No agents found for class %s", class); rc = -EPROTONOSUPPORT; } return rc; } static bool does_provider_have_agent(const char *agent, const char *provider, const char *class) { bool found = false; GList *agents = NULL; GList *gIter2 = NULL; agents = resources_list_agents(class, provider); for (gIter2 = agents; gIter2 != NULL; gIter2 = gIter2->next) { if (pcmk__str_eq(agent, gIter2->data, pcmk__str_casei)) { found = true; } } g_list_free_full(agents, free); return found; } static int lrmd_api_list_ocf_providers(lrmd_t * lrmd, const char *agent, lrmd_list_t ** providers) { int rc = pcmk_ok; char *provider = NULL; GList *ocf_providers = NULL; GList *gIter = NULL; ocf_providers = resources_list_providers(PCMK_RESOURCE_CLASS_OCF); for (gIter = ocf_providers; gIter != NULL; gIter = gIter->next) { provider = gIter->data; if (!agent || does_provider_have_agent(agent, provider, PCMK_RESOURCE_CLASS_OCF)) { *providers = lrmd_list_add(*providers, (const char *)gIter->data); rc++; } } g_list_free_full(ocf_providers, free); return rc; } static int lrmd_api_list_standards(lrmd_t * lrmd, lrmd_list_t ** supported) { int rc = 0; GList *standards = NULL; GList *gIter = NULL; standards = resources_list_standards(); for (gIter = standards; gIter != NULL; gIter = gIter->next) { *supported = lrmd_list_add(*supported, (const char *)gIter->data); rc++; } if (list_stonith_agents(NULL) > 0) { *supported = lrmd_list_add(*supported, PCMK_RESOURCE_CLASS_STONITH); rc++; } g_list_free_full(standards, free); return rc; } -lrmd_t * -lrmd_api_new(void) +/*! + * \internal + * \brief Create an executor API object + * + * \param[out] api Will be set to newly created API object (it is the + * caller's responsibility to free this value with + * lrmd_api_delete() if this function succeeds) + * \param[in] nodename If the object will be used for a remote connection, + * the node name to use in cluster for remote executor + * \param[in] server If the object will be used for a remote connection, + * the resolvable host name to connect to + * \param[in] port If the object will be used for a remote connection, + * port number on \p server to connect to + * + * \return Standard Pacemaker return code + * \note If the caller leaves one of \p nodename or \p server NULL, the other's + * value will be used for both. If the caller leaves both NULL, an API + * object will be created for a local executor connection. + */ +int +lrmd__new(lrmd_t **api, const char *nodename, const char *server, int port) { - lrmd_t *new_lrmd = NULL; lrmd_private_t *pvt = NULL; - new_lrmd = calloc(1, sizeof(lrmd_t)); + if (api == NULL) { + return EINVAL; + } + *api = NULL; + + // Allocate all memory needed + + *api = calloc(1, sizeof(lrmd_t)); + if (*api == NULL) { + return ENOMEM; + } + pvt = calloc(1, sizeof(lrmd_private_t)); + if (pvt == NULL) { + lrmd_api_delete(*api); + *api = NULL; + return ENOMEM; + } + (*api)->lrmd_private = pvt; + + // @TODO Do we need to do this for local connections? pvt->remote = calloc(1, sizeof(pcmk__remote_t)); - new_lrmd->cmds = calloc(1, sizeof(lrmd_api_operations_t)); - - pvt->type = pcmk__client_ipc; - new_lrmd->lrmd_private = pvt; - - new_lrmd->cmds->connect = lrmd_api_connect; - new_lrmd->cmds->connect_async = lrmd_api_connect_async; - new_lrmd->cmds->is_connected = lrmd_api_is_connected; - new_lrmd->cmds->poke_connection = lrmd_api_poke_connection; - new_lrmd->cmds->disconnect = lrmd_api_disconnect; - new_lrmd->cmds->register_rsc = lrmd_api_register_rsc; - new_lrmd->cmds->unregister_rsc = lrmd_api_unregister_rsc; - new_lrmd->cmds->get_rsc_info = lrmd_api_get_rsc_info; - new_lrmd->cmds->get_recurring_ops = lrmd_api_get_recurring_ops; - new_lrmd->cmds->set_callback = lrmd_api_set_callback; - new_lrmd->cmds->get_metadata = lrmd_api_get_metadata; - new_lrmd->cmds->exec = lrmd_api_exec; - new_lrmd->cmds->cancel = lrmd_api_cancel; - new_lrmd->cmds->list_agents = lrmd_api_list_agents; - new_lrmd->cmds->list_ocf_providers = lrmd_api_list_ocf_providers; - new_lrmd->cmds->list_standards = lrmd_api_list_standards; - new_lrmd->cmds->exec_alert = lrmd_api_exec_alert; - new_lrmd->cmds->get_metadata_params = lrmd_api_get_metadata_params; - - return new_lrmd; + + (*api)->cmds = calloc(1, sizeof(lrmd_api_operations_t)); + + if ((pvt->remote == NULL) || ((*api)->cmds == NULL)) { + lrmd_api_delete(*api); + *api = NULL; + return ENOMEM; + } + + // Set methods + (*api)->cmds->connect = lrmd_api_connect; + (*api)->cmds->connect_async = lrmd_api_connect_async; + (*api)->cmds->is_connected = lrmd_api_is_connected; + (*api)->cmds->poke_connection = lrmd_api_poke_connection; + (*api)->cmds->disconnect = lrmd_api_disconnect; + (*api)->cmds->register_rsc = lrmd_api_register_rsc; + (*api)->cmds->unregister_rsc = lrmd_api_unregister_rsc; + (*api)->cmds->get_rsc_info = lrmd_api_get_rsc_info; + (*api)->cmds->get_recurring_ops = lrmd_api_get_recurring_ops; + (*api)->cmds->set_callback = lrmd_api_set_callback; + (*api)->cmds->get_metadata = lrmd_api_get_metadata; + (*api)->cmds->exec = lrmd_api_exec; + (*api)->cmds->cancel = lrmd_api_cancel; + (*api)->cmds->list_agents = lrmd_api_list_agents; + (*api)->cmds->list_ocf_providers = lrmd_api_list_ocf_providers; + (*api)->cmds->list_standards = lrmd_api_list_standards; + (*api)->cmds->exec_alert = lrmd_api_exec_alert; + (*api)->cmds->get_metadata_params = lrmd_api_get_metadata_params; + + if ((nodename == NULL) && (server == NULL)) { + pvt->type = pcmk__client_ipc; + } else { +#ifdef HAVE_GNUTLS_GNUTLS_H + if (nodename == NULL) { + nodename = server; + } else if (server == NULL) { + server = nodename; + } + pvt->type = pcmk__client_tls; + pvt->remote_nodename = strdup(nodename); + pvt->server = strdup(server); + if ((pvt->remote_nodename == NULL) || (pvt->server == NULL)) { + lrmd_api_delete(*api); + *api = NULL; + return ENOMEM; + } + pvt->port = port; + if (pvt->port == 0) { + pvt->port = crm_default_remote_port(); + } +#else + crm_err("Cannot communicate with Pacemaker Remote " + "because GnuTLS is not enabled for this build"); + lrmd_api_delete(*api); + *api = NULL; + return EOPNOTSUPP; +#endif + } + return pcmk_rc_ok; } lrmd_t * -lrmd_remote_api_new(const char *nodename, const char *server, int port) +lrmd_api_new(void) { -#ifdef HAVE_GNUTLS_GNUTLS_H - lrmd_t *new_lrmd = lrmd_api_new(); - lrmd_private_t *native = new_lrmd->lrmd_private; + lrmd_t *api = NULL; - if (!nodename && !server) { - lrmd_api_delete(new_lrmd); - return NULL; - } + CRM_ASSERT(lrmd__new(&api, NULL, NULL, 0) == pcmk_rc_ok); + return api; +} - native->type = pcmk__client_tls; - native->remote_nodename = nodename ? strdup(nodename) : strdup(server); - native->server = server ? strdup(server) : strdup(nodename); - native->port = port; - if (native->port == 0) { - native->port = crm_default_remote_port(); - } +lrmd_t * +lrmd_remote_api_new(const char *nodename, const char *server, int port) +{ + lrmd_t *api = NULL; - return new_lrmd; -#else - crm_err("Cannot communicate with Pacemaker Remote because GnuTLS is not enabled for this build"); - return NULL; -#endif + CRM_ASSERT(lrmd__new(&api, nodename, server, port) == pcmk_rc_ok); + return api; } void lrmd_api_delete(lrmd_t * lrmd) { if (!lrmd) { return; } lrmd->cmds->disconnect(lrmd); /* no-op if already disconnected */ free(lrmd->cmds); if (lrmd->lrmd_private) { lrmd_private_t *native = lrmd->lrmd_private; #ifdef HAVE_GNUTLS_GNUTLS_H free(native->server); #endif free(native->remote_nodename); free(native->remote); free(native->token); free(native->peer_version); } free(lrmd->lrmd_private); free(lrmd); } diff --git a/lib/pacemaker/pcmk_sched_allocate.c b/lib/pacemaker/pcmk_sched_allocate.c index 0dbfd1893b..f46474b5ca 100644 --- a/lib/pacemaker/pcmk_sched_allocate.c +++ b/lib/pacemaker/pcmk_sched_allocate.c @@ -1,3035 +1,2987 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include CRM_TRACE_INIT_DATA(pacemaker); extern bool pcmk__is_daemon; void set_alloc_actions(pe_working_set_t * data_set); extern void ReloadRsc(pe_resource_t * rsc, pe_node_t *node, pe_working_set_t * data_set); extern gboolean DeleteRsc(pe_resource_t * rsc, pe_node_t * node, gboolean optional, pe_working_set_t * data_set); static void apply_remote_node_ordering(pe_working_set_t *data_set); static enum remote_connection_state get_remote_node_state(pe_node_t *node); enum remote_connection_state { remote_state_unknown = 0, remote_state_alive = 1, remote_state_resting = 2, remote_state_failed = 3, remote_state_stopped = 4 }; static const char * state2text(enum remote_connection_state state) { switch (state) { case remote_state_unknown: return "unknown"; case remote_state_alive: return "alive"; case remote_state_resting: return "resting"; case remote_state_failed: return "failed"; case remote_state_stopped: return "stopped"; } return "impossible"; } resource_alloc_functions_t resource_class_alloc_functions[] = { { pcmk__native_merge_weights, pcmk__native_allocate, native_create_actions, native_create_probe, native_internal_constraints, native_rsc_colocation_lh, native_rsc_colocation_rh, native_rsc_location, native_action_flags, native_update_actions, native_expand, native_append_meta, }, { pcmk__group_merge_weights, pcmk__group_allocate, group_create_actions, native_create_probe, group_internal_constraints, group_rsc_colocation_lh, group_rsc_colocation_rh, group_rsc_location, group_action_flags, group_update_actions, group_expand, group_append_meta, }, { pcmk__native_merge_weights, pcmk__clone_allocate, clone_create_actions, clone_create_probe, clone_internal_constraints, clone_rsc_colocation_lh, clone_rsc_colocation_rh, clone_rsc_location, clone_action_flags, pcmk__multi_update_actions, clone_expand, clone_append_meta, }, { pcmk__native_merge_weights, pcmk__bundle_allocate, pcmk__bundle_create_actions, pcmk__bundle_create_probe, pcmk__bundle_internal_constraints, pcmk__bundle_rsc_colocation_lh, pcmk__bundle_rsc_colocation_rh, pcmk__bundle_rsc_location, pcmk__bundle_action_flags, pcmk__multi_update_actions, pcmk__bundle_expand, pcmk__bundle_append_meta, } }; static gboolean check_rsc_parameters(pe_resource_t * rsc, pe_node_t * node, xmlNode * rsc_entry, gboolean active_here, pe_working_set_t * data_set) { int attr_lpc = 0; gboolean force_restart = FALSE; gboolean delete_resource = FALSE; gboolean changed = FALSE; const char *value = NULL; const char *old_value = NULL; const char *attr_list[] = { XML_ATTR_TYPE, XML_AGENT_ATTR_CLASS, XML_AGENT_ATTR_PROVIDER }; for (; attr_lpc < PCMK__NELEM(attr_list); attr_lpc++) { value = crm_element_value(rsc->xml, attr_list[attr_lpc]); old_value = crm_element_value(rsc_entry, attr_list[attr_lpc]); if (value == old_value /* i.e. NULL */ || pcmk__str_eq(value, old_value, pcmk__str_none)) { continue; } changed = TRUE; trigger_unfencing(rsc, node, "Device definition changed", NULL, data_set); if (active_here) { force_restart = TRUE; crm_notice("Forcing restart of %s on %s, %s changed: %s -> %s", rsc->id, node->details->uname, attr_list[attr_lpc], crm_str(old_value), crm_str(value)); } } if (force_restart) { /* make sure the restart happens */ stop_action(rsc, node, FALSE); pe__set_resource_flags(rsc, pe_rsc_start_pending); delete_resource = TRUE; } else if (changed) { delete_resource = TRUE; } return delete_resource; } static void CancelXmlOp(pe_resource_t * rsc, xmlNode * xml_op, pe_node_t * active_node, const char *reason, pe_working_set_t * data_set) { guint interval_ms = 0; pe_action_t *cancel = NULL; const char *task = NULL; const char *call_id = NULL; CRM_CHECK(xml_op != NULL, return); CRM_CHECK(active_node != NULL, return); task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); call_id = crm_element_value(xml_op, XML_LRM_ATTR_CALLID); crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); crm_info("Action " PCMK__OP_FMT " on %s will be stopped: %s", rsc->id, task, interval_ms, active_node->details->uname, (reason? reason : "unknown")); cancel = pe_cancel_op(rsc, task, interval_ms, active_node, data_set); add_hash_param(cancel->meta, XML_LRM_ATTR_CALLID, call_id); custom_action_order(rsc, stop_key(rsc), NULL, rsc, NULL, cancel, pe_order_optional, data_set); } static gboolean check_action_definition(pe_resource_t * rsc, pe_node_t * active_node, xmlNode * xml_op, pe_working_set_t * data_set) { char *key = NULL; guint interval_ms = 0; const op_digest_cache_t *digest_data = NULL; gboolean did_change = FALSE; const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *digest_secure = NULL; CRM_CHECK(active_node != NULL, return FALSE); crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if (interval_ms > 0) { xmlNode *op_match = NULL; /* we need to reconstruct the key because of the way we used to construct resource IDs */ key = pcmk__op_key(rsc->id, task, interval_ms); pe_rsc_trace(rsc, "Checking parameters for %s", key); op_match = find_rsc_op_entry(rsc, key); if ((op_match == NULL) && pcmk_is_set(data_set->flags, pe_flag_stop_action_orphans)) { CancelXmlOp(rsc, xml_op, active_node, "orphan", data_set); free(key); return TRUE; } else if (op_match == NULL) { pe_rsc_debug(rsc, "Orphan action detected: %s on %s", key, active_node->details->uname); free(key); return TRUE; } free(key); key = NULL; } crm_trace("Testing " PCMK__OP_FMT " on %s", rsc->id, task, interval_ms, active_node->details->uname); if ((interval_ms == 0) && pcmk__str_eq(task, RSC_STATUS, pcmk__str_casei)) { /* Reload based on the start action not a probe */ task = RSC_START; } else if ((interval_ms == 0) && pcmk__str_eq(task, RSC_MIGRATED, pcmk__str_casei)) { /* Reload based on the start action not a migrate */ task = RSC_START; } else if ((interval_ms == 0) && pcmk__str_eq(task, RSC_PROMOTE, pcmk__str_casei)) { /* Reload based on the start action not a promote */ task = RSC_START; } digest_data = rsc_action_digest_cmp(rsc, xml_op, active_node, data_set); if (pcmk_is_set(data_set->flags, pe_flag_sanitized)) { digest_secure = crm_element_value(xml_op, XML_LRM_ATTR_SECURE_DIGEST); } if(digest_data->rc != RSC_DIGEST_MATCH && digest_secure && digest_data->digest_secure_calc && strcmp(digest_data->digest_secure_calc, digest_secure) == 0) { if (!pcmk__is_daemon && data_set->priv != NULL) { pcmk__output_t *out = data_set->priv; out->info(out, "Only 'private' parameters to " PCMK__OP_FMT " on %s changed: %s", rsc->id, task, interval_ms, active_node->details->uname, crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC)); } } else if (digest_data->rc == RSC_DIGEST_RESTART) { /* Changes that force a restart */ pe_action_t *required = NULL; did_change = TRUE; key = pcmk__op_key(rsc->id, task, interval_ms); crm_log_xml_info(digest_data->params_restart, "params:restart"); required = custom_action(rsc, key, task, NULL, FALSE, TRUE, data_set); pe_action_set_reason(required, "resource definition change", true); trigger_unfencing(rsc, active_node, "Device parameters changed", NULL, data_set); } else if ((digest_data->rc == RSC_DIGEST_ALL) || (digest_data->rc == RSC_DIGEST_UNKNOWN)) { // Changes that can potentially be handled by an agent reload const char *digest_restart = crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST); did_change = TRUE; trigger_unfencing(rsc, active_node, "Device parameters changed (reload)", NULL, data_set); crm_log_xml_info(digest_data->params_all, "params:reload"); key = pcmk__op_key(rsc->id, task, interval_ms); if (interval_ms > 0) { pe_action_t *op = NULL; #if 0 /* Always reload/restart the entire resource */ ReloadRsc(rsc, active_node, data_set); #else /* Re-sending the recurring op is sufficient - the old one will be cancelled automatically */ op = custom_action(rsc, key, task, active_node, TRUE, TRUE, data_set); pe__set_action_flags(op, pe_action_reschedule); #endif } else if (digest_restart) { pe_rsc_trace(rsc, "Reloading '%s' action for resource %s", task, rsc->id); /* Reload this resource */ ReloadRsc(rsc, active_node, data_set); free(key); } else { pe_action_t *required = NULL; pe_rsc_trace(rsc, "Resource %s doesn't support agent reloads", rsc->id); /* Re-send the start/demote/promote op * Recurring ops will be detected independently */ required = custom_action(rsc, key, task, NULL, FALSE, TRUE, data_set); pe_action_set_reason(required, "resource definition change", true); } } return did_change; } /*! * \internal * \brief Do deferred action checks after allocation * * \param[in] data_set Working set for cluster */ static void check_params(pe_resource_t *rsc, pe_node_t *node, xmlNode *rsc_op, enum pe_check_parameters check, pe_working_set_t *data_set) { const char *reason = NULL; op_digest_cache_t *digest_data = NULL; switch (check) { case pe_check_active: if (check_action_definition(rsc, node, rsc_op, data_set) && pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL, data_set)) { reason = "action definition changed"; } break; case pe_check_last_failure: digest_data = rsc_action_digest_cmp(rsc, rsc_op, node, data_set); switch (digest_data->rc) { case RSC_DIGEST_UNKNOWN: crm_trace("Resource %s history entry %s on %s has no digest to compare", rsc->id, ID(rsc_op), node->details->id); break; case RSC_DIGEST_MATCH: break; default: reason = "resource parameters have changed"; break; } break; } if (reason) { pe__clear_failcount(rsc, node, reason, data_set); } } static void check_actions_for(xmlNode * rsc_entry, pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set) { GList *gIter = NULL; int offset = -1; int stop_index = 0; int start_index = 0; const char *task = NULL; xmlNode *rsc_op = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; CRM_CHECK(node != NULL, return); if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) { pe_resource_t *parent = uber_parent(rsc); if(parent == NULL || pe_rsc_is_clone(parent) == FALSE || pcmk_is_set(parent->flags, pe_rsc_unique)) { pe_rsc_trace(rsc, "Skipping param check for %s and deleting: orphan", rsc->id); DeleteRsc(rsc, node, FALSE, data_set); } else { pe_rsc_trace(rsc, "Skipping param check for %s (orphan clone)", rsc->id); } return; } else if (pe_find_node_id(rsc->running_on, node->details->id) == NULL) { if (check_rsc_parameters(rsc, node, rsc_entry, FALSE, data_set)) { DeleteRsc(rsc, node, FALSE, data_set); } pe_rsc_trace(rsc, "Skipping param check for %s: no longer active on %s", rsc->id, node->details->uname); return; } pe_rsc_trace(rsc, "Processing %s on %s", rsc->id, node->details->uname); if (check_rsc_parameters(rsc, node, rsc_entry, TRUE, data_set)) { DeleteRsc(rsc, node, FALSE, data_set); } for (rsc_op = pcmk__xe_first_child(rsc_entry); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) { if (pcmk__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, pcmk__str_none)) { op_list = g_list_prepend(op_list, rsc_op); } } sorted_op_list = g_list_sort(op_list, sort_op_by_callid); calculate_active_ops(sorted_op_list, &start_index, &stop_index); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; guint interval_ms = 0; offset++; if (start_index < stop_index) { /* stopped */ continue; } else if (offset < start_index) { /* action occurred prior to a start */ continue; } task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); crm_element_value_ms(rsc_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((interval_ms > 0) && (pcmk_is_set(rsc->flags, pe_rsc_maintenance) || node->details->maintenance)) { // Maintenance mode cancels recurring operations CancelXmlOp(rsc, rsc_op, node, "maintenance mode", data_set); } else if ((interval_ms > 0) || pcmk__strcase_any_of(task, RSC_STATUS, RSC_START, RSC_PROMOTE, RSC_MIGRATED, NULL)) { /* If a resource operation failed, and the operation's definition * has changed, clear any fail count so they can be retried fresh. */ if (pe__bundle_needs_remote_name(rsc, data_set)) { /* We haven't allocated resources to nodes yet, so if the * REMOTE_CONTAINER_HACK is used, we may calculate the digest * based on the literal "#uname" value rather than the properly * substituted value. That would mistakenly make the action * definition appear to have been changed. Defer the check until * later in this case. */ pe__add_param_check(rsc_op, rsc, node, pe_check_active, data_set); } else if (check_action_definition(rsc, node, rsc_op, data_set) && pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL, data_set)) { pe__clear_failcount(rsc, node, "action definition changed", data_set); } } } g_list_free(sorted_op_list); } static GList * find_rsc_list(GList *result, pe_resource_t * rsc, const char *id, gboolean renamed_clones, gboolean partial, pe_working_set_t * data_set) { GList *gIter = NULL; gboolean match = FALSE; if (id == NULL) { return NULL; } if (rsc == NULL) { if (data_set == NULL) { return NULL; } for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *child = (pe_resource_t *) gIter->data; result = find_rsc_list(result, child, id, renamed_clones, partial, NULL); } return result; } if (partial) { if (strstr(rsc->id, id)) { match = TRUE; } else if (renamed_clones && rsc->clone_name && strstr(rsc->clone_name, id)) { match = TRUE; } } else { if (strcmp(rsc->id, id) == 0) { match = TRUE; } else if (renamed_clones && rsc->clone_name && strcmp(rsc->clone_name, id) == 0) { match = TRUE; } } if (match) { result = g_list_prepend(result, rsc); } if (rsc->children) { gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child = (pe_resource_t *) gIter->data; result = find_rsc_list(result, child, id, renamed_clones, partial, NULL); } } return result; } static void check_actions(pe_working_set_t * data_set) { const char *id = NULL; pe_node_t *node = NULL; xmlNode *lrm_rscs = NULL; xmlNode *status = get_object_root(XML_CIB_TAG_STATUS, data_set->input); xmlNode *node_state = NULL; for (node_state = pcmk__xe_first_child(status); node_state != NULL; node_state = pcmk__xe_next(node_state)) { if (pcmk__str_eq((const char *)node_state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { id = crm_element_value(node_state, XML_ATTR_ID); lrm_rscs = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE); lrm_rscs = find_xml_node(lrm_rscs, XML_LRM_TAG_RESOURCES, FALSE); node = pe_find_node_id(data_set->nodes, id); if (node == NULL) { continue; /* Still need to check actions for a maintenance node to cancel existing monitor operations */ } else if (can_run_resources(node) == FALSE && node->details->maintenance == FALSE) { crm_trace("Skipping param check for %s: can't run resources", node->details->uname); continue; } crm_trace("Processing node %s", node->details->uname); if (node->details->online || pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { xmlNode *rsc_entry = NULL; for (rsc_entry = pcmk__xe_first_child(lrm_rscs); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) { if (pcmk__str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, pcmk__str_none)) { if (xml_has_children(rsc_entry)) { GList *gIter = NULL; GList *result = NULL; const char *rsc_id = ID(rsc_entry); CRM_CHECK(rsc_id != NULL, return); result = find_rsc_list(NULL, NULL, rsc_id, TRUE, FALSE, data_set); for (gIter = result; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; if (rsc->variant != pe_native) { continue; } check_actions_for(rsc_entry, rsc, node, data_set); } g_list_free(result); } } } } } } } static void apply_placement_constraints(pe_working_set_t * data_set) { for (GList *gIter = data_set->placement_constraints; gIter != NULL; gIter = gIter->next) { pe__location_t *cons = gIter->data; cons->rsc_lh->cmds->rsc_location(cons->rsc_lh, cons); } } static gboolean failcount_clear_action_exists(pe_node_t * node, pe_resource_t * rsc) { gboolean rc = FALSE; GList *list = pe__resource_actions(rsc, node, CRM_OP_CLEAR_FAILCOUNT, TRUE); if (list) { rc = TRUE; } g_list_free(list); return rc; } -/*! - * \internal - * \brief Force resource away if failures hit migration threshold - * - * \param[in,out] rsc Resource to check for failures - * \param[in,out] node Node to check for failures - * \param[in,out] data_set Cluster working set to update - */ -static void -check_migration_threshold(pe_resource_t *rsc, pe_node_t *node, - pe_working_set_t *data_set) -{ - int fail_count, countdown; - pe_resource_t *failed; - - /* Migration threshold of 0 means never force away */ - if (rsc->migration_threshold == 0) { - return; - } - - // If we're ignoring failures, also ignore the migration threshold - if (pcmk_is_set(rsc->flags, pe_rsc_failure_ignored)) { - return; - } - - /* If there are no failures, there's no need to force away */ - fail_count = pe_get_failcount(node, rsc, NULL, - pe_fc_effective|pe_fc_fillers, NULL, - data_set); - if (fail_count <= 0) { - return; - } - - /* How many more times recovery will be tried on this node */ - countdown = QB_MAX(rsc->migration_threshold - fail_count, 0); - - /* If failed resource has a parent, we'll force the parent away */ - failed = rsc; - if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) { - failed = uber_parent(rsc); - } - - if (countdown == 0) { - resource_location(failed, node, -INFINITY, "__fail_limit__", data_set); - crm_warn("Forcing %s away from %s after %d failures (max=%d)", - failed->id, node->details->uname, fail_count, - rsc->migration_threshold); - } else { - crm_info("%s can fail %d more times on %s before being forced off", - failed->id, countdown, node->details->uname); - } -} - static void common_apply_stickiness(pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set) { if (rsc->children) { GList *gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; common_apply_stickiness(child_rsc, node, data_set); } return; } if (pcmk_is_set(rsc->flags, pe_rsc_managed) && rsc->stickiness != 0 && pcmk__list_of_1(rsc->running_on)) { pe_node_t *current = pe_find_node_id(rsc->running_on, node->details->id); pe_node_t *match = pe_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (current == NULL) { } else if ((match != NULL) || pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster)) { pe_resource_t *sticky_rsc = rsc; resource_location(sticky_rsc, node, rsc->stickiness, "stickiness", data_set); pe_rsc_debug(sticky_rsc, "Resource %s: preferring current location" " (node=%s, weight=%d)", sticky_rsc->id, node->details->uname, rsc->stickiness); } else { GHashTableIter iter; pe_node_t *nIter = NULL; pe_rsc_debug(rsc, "Ignoring stickiness for %s: the cluster is asymmetric" " and node %s is not explicitly allowed", rsc->id, node->details->uname); g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&nIter)) { crm_err("%s[%s] = %d", rsc->id, nIter->details->uname, nIter->weight); } } } /* Check the migration threshold only if a failcount clear action * has not already been placed for this resource on the node. * There is no sense in potentially forcing the resource from this * node if the failcount is being reset anyway. * * @TODO A clear_failcount operation can be scheduled in stage4() via * check_actions_for(), or in stage5() via check_params(). This runs in * stage2(), so it cannot detect those, meaning we might check the migration * threshold when we shouldn't -- worst case, we stop or move the resource, * then move it back next transition. */ if (failcount_clear_action_exists(node, rsc) == FALSE) { - check_migration_threshold(rsc, node, data_set); + pe_resource_t *failed = NULL; + + if (pcmk__threshold_reached(rsc, node, data_set, &failed)) { + resource_location(failed, node, -INFINITY, "__fail_limit__", + data_set); + } } } void complex_set_cmds(pe_resource_t * rsc) { GList *gIter = rsc->children; rsc->cmds = &resource_class_alloc_functions[rsc->variant]; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; complex_set_cmds(child_rsc); } } void set_alloc_actions(pe_working_set_t * data_set) { GList *gIter = data_set->resources; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; complex_set_cmds(rsc); } } static void calculate_system_health(gpointer gKey, gpointer gValue, gpointer user_data) { const char *key = (const char *)gKey; const char *value = (const char *)gValue; int *system_health = (int *)user_data; if (!gKey || !gValue || !user_data) { return; } if (pcmk__starts_with(key, "#health")) { int score; /* Convert the value into an integer */ score = char2score(value); /* Add it to the running total */ *system_health = pe__add_scores(score, *system_health); } } static gboolean apply_system_health(pe_working_set_t * data_set) { GList *gIter = NULL; const char *health_strategy = pe_pref(data_set->config_hash, "node-health-strategy"); int base_health = 0; if (pcmk__str_eq(health_strategy, "none", pcmk__str_null_matches | pcmk__str_casei)) { /* Prevent any accidental health -> score translation */ pcmk__score_red = 0; pcmk__score_yellow = 0; pcmk__score_green = 0; return TRUE; } else if (pcmk__str_eq(health_strategy, "migrate-on-red", pcmk__str_casei)) { /* Resources on nodes which have health values of red are * weighted away from that node. */ pcmk__score_red = -INFINITY; pcmk__score_yellow = 0; pcmk__score_green = 0; } else if (pcmk__str_eq(health_strategy, "only-green", pcmk__str_casei)) { /* Resources on nodes which have health values of red or yellow * are forced away from that node. */ pcmk__score_red = -INFINITY; pcmk__score_yellow = -INFINITY; pcmk__score_green = 0; } else if (pcmk__str_eq(health_strategy, "progressive", pcmk__str_casei)) { /* Same as the above, but use the r/y/g scores provided by the user * Defaults are provided by the pe_prefs table * Also, custom health "base score" can be used */ base_health = char2score(pe_pref(data_set->config_hash, "node-health-base")); } else if (pcmk__str_eq(health_strategy, "custom", pcmk__str_casei)) { /* Requires the admin to configure the rsc_location constaints for * processing the stored health scores */ /* TODO: Check for the existence of appropriate node health constraints */ return TRUE; } else { crm_err("Unknown node health strategy: %s", health_strategy); return FALSE; } crm_info("Applying automated node health strategy: %s", health_strategy); for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { int system_health = base_health; pe_node_t *node = (pe_node_t *) gIter->data; /* Search through the node hash table for system health entries. */ g_hash_table_foreach(node->details->attrs, calculate_system_health, &system_health); crm_info(" Node %s has an combined system health of %d", node->details->uname, system_health); /* If the health is non-zero, then create a new rsc2node so that the * weight will be added later on. */ if (system_health != 0) { GList *gIter2 = data_set->resources; for (; gIter2 != NULL; gIter2 = gIter2->next) { pe_resource_t *rsc = (pe_resource_t *) gIter2->data; rsc2node_new(health_strategy, rsc, system_health, NULL, node, data_set); } } } return TRUE; } gboolean stage0(pe_working_set_t * data_set) { xmlNode *cib_constraints = get_object_root(XML_CIB_TAG_CONSTRAINTS, data_set->input); if (data_set->input == NULL) { return FALSE; } if (!pcmk_is_set(data_set->flags, pe_flag_have_status)) { crm_trace("Calculating status"); cluster_status(data_set); } set_alloc_actions(data_set); apply_system_health(data_set); unpack_constraints(cib_constraints, data_set); return TRUE; } /* * Check nodes for resources started outside of the LRM */ gboolean probe_resources(pe_working_set_t * data_set) { pe_action_t *probe_node_complete = NULL; for (GList *gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; const char *probed = pe_node_attribute_raw(node, CRM_OP_PROBED); if (node->details->online == FALSE) { if (pe__is_remote_node(node) && node->details->remote_rsc && (get_remote_node_state(node) == remote_state_failed)) { pe_fence_node(data_set, node, "the connection is unrecoverable", FALSE); } continue; } else if (node->details->unclean) { continue; } else if (node->details->rsc_discovery_enabled == FALSE) { /* resource discovery is disabled for this node */ continue; } if (probed != NULL && crm_is_true(probed) == FALSE) { pe_action_t *probe_op = custom_action(NULL, crm_strdup_printf("%s-%s", CRM_OP_REPROBE, node->details->uname), CRM_OP_REPROBE, node, FALSE, TRUE, data_set); add_hash_param(probe_op->meta, XML_ATTR_TE_NOWAIT, XML_BOOLEAN_TRUE); continue; } for (GList *gIter2 = data_set->resources; gIter2 != NULL; gIter2 = gIter2->next) { pe_resource_t *rsc = (pe_resource_t *) gIter2->data; rsc->cmds->create_probe(rsc, node, probe_node_complete, FALSE, data_set); } } return TRUE; } static void rsc_discover_filter(pe_resource_t *rsc, pe_node_t *node) { pe_resource_t *top = uber_parent(rsc); pe_node_t *match; if (rsc->exclusive_discover == FALSE && top->exclusive_discover == FALSE) { return; } g_list_foreach(rsc->children, (GFunc) rsc_discover_filter, node); match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (match && match->rsc_discover_mode != pe_discover_exclusive) { match->weight = -INFINITY; } } static time_t shutdown_time(pe_node_t *node, pe_working_set_t *data_set) { const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN); time_t result = 0; if (shutdown) { long long result_ll; if (pcmk__scan_ll(shutdown, &result_ll, 0LL) == pcmk_rc_ok) { result = (time_t) result_ll; } } return result? result : get_effective_time(data_set); } static void apply_shutdown_lock(pe_resource_t *rsc, pe_working_set_t *data_set) { const char *class; // Only primitives and (uncloned) groups may be locked if (rsc->variant == pe_group) { g_list_foreach(rsc->children, (GFunc) apply_shutdown_lock, data_set); } else if (rsc->variant != pe_native) { return; } // Fence devices and remote connections can't be locked class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches) || pe__resource_is_remote_conn(rsc, data_set)) { return; } if (rsc->lock_node != NULL) { // The lock was obtained from resource history if (rsc->running_on != NULL) { /* The resource was started elsewhere even though it is now * considered locked. This shouldn't be possible, but as a * failsafe, we don't want to disturb the resource now. */ pe_rsc_info(rsc, "Cancelling shutdown lock because %s is already active", rsc->id); pe__clear_resource_history(rsc, rsc->lock_node, data_set); rsc->lock_node = NULL; rsc->lock_time = 0; } // Only a resource active on exactly one node can be locked } else if (pcmk__list_of_1(rsc->running_on)) { pe_node_t *node = rsc->running_on->data; if (node->details->shutdown) { if (node->details->unclean) { pe_rsc_debug(rsc, "Not locking %s to unclean %s for shutdown", rsc->id, node->details->uname); } else { rsc->lock_node = node; rsc->lock_time = shutdown_time(node, data_set); } } } if (rsc->lock_node == NULL) { // No lock needed return; } if (data_set->shutdown_lock > 0) { time_t lock_expiration = rsc->lock_time + data_set->shutdown_lock; pe_rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)", rsc->id, rsc->lock_node->details->uname, (long long) lock_expiration); pe__update_recheck_time(++lock_expiration, data_set); } else { pe_rsc_info(rsc, "Locking %s to %s due to shutdown", rsc->id, rsc->lock_node->details->uname); } // If resource is locked to one node, ban it from all other nodes for (GList *item = data_set->nodes; item != NULL; item = item->next) { pe_node_t *node = item->data; if (strcmp(node->details->uname, rsc->lock_node->details->uname)) { resource_location(rsc, node, -CRM_SCORE_INFINITY, XML_CONFIG_ATTR_SHUTDOWN_LOCK, data_set); } } } /* * \internal * \brief Stage 2 of cluster status: apply node-specific criteria * * Count known nodes, and apply location constraints, stickiness, and exclusive * resource discovery. */ gboolean stage2(pe_working_set_t * data_set) { GList *gIter = NULL; if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { g_list_foreach(data_set->resources, (GFunc) apply_shutdown_lock, data_set); } if (!pcmk_is_set(data_set->flags, pe_flag_no_compat)) { // @COMPAT API backward compatibility for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; if (node && (node->weight >= 0) && node->details->online && (node->details->type != node_ping)) { data_set->max_valid_nodes++; } } } apply_placement_constraints(data_set); gIter = data_set->nodes; for (; gIter != NULL; gIter = gIter->next) { GList *gIter2 = NULL; pe_node_t *node = (pe_node_t *) gIter->data; gIter2 = data_set->resources; for (; gIter2 != NULL; gIter2 = gIter2->next) { pe_resource_t *rsc = (pe_resource_t *) gIter2->data; common_apply_stickiness(rsc, node, data_set); rsc_discover_filter(rsc, node); } } return TRUE; } /* * Create internal resource constraints before allocation */ gboolean stage3(pe_working_set_t * data_set) { GList *gIter = data_set->resources; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; rsc->cmds->internal_constraints(rsc, data_set); } return TRUE; } /* * Check for orphaned or redefined actions */ gboolean stage4(pe_working_set_t * data_set) { check_actions(data_set); return TRUE; } static void * convert_const_pointer(const void *ptr) { /* Worst function ever */ return (void *)ptr; } static gint sort_rsc_process_order(gconstpointer a, gconstpointer b, gpointer data) { int rc = 0; int r1_weight = -INFINITY; int r2_weight = -INFINITY; const char *reason = "existence"; GList *nodes = (GList *) data; const pe_resource_t *resource1 = a; const pe_resource_t *resource2 = b; pe_node_t *r1_node = NULL; pe_node_t *r2_node = NULL; GList *gIter = NULL; GHashTable *r1_nodes = NULL; GHashTable *r2_nodes = NULL; reason = "priority"; r1_weight = resource1->priority; r2_weight = resource2->priority; if (r1_weight > r2_weight) { rc = -1; goto done; } if (r1_weight < r2_weight) { rc = 1; goto done; } reason = "no node list"; if (nodes == NULL) { goto done; } r1_nodes = pcmk__native_merge_weights(convert_const_pointer(resource1), resource1->id, NULL, NULL, 1, pe_weights_forward | pe_weights_init); pe__show_node_weights(true, NULL, resource1->id, r1_nodes, resource1->cluster); r2_nodes = pcmk__native_merge_weights(convert_const_pointer(resource2), resource2->id, NULL, NULL, 1, pe_weights_forward | pe_weights_init); pe__show_node_weights(true, NULL, resource2->id, r2_nodes, resource2->cluster); /* Current location score */ reason = "current location"; r1_weight = -INFINITY; r2_weight = -INFINITY; if (resource1->running_on) { r1_node = pe__current_node(resource1); r1_node = g_hash_table_lookup(r1_nodes, r1_node->details->id); if (r1_node != NULL) { r1_weight = r1_node->weight; } } if (resource2->running_on) { r2_node = pe__current_node(resource2); r2_node = g_hash_table_lookup(r2_nodes, r2_node->details->id); if (r2_node != NULL) { r2_weight = r2_node->weight; } } if (r1_weight > r2_weight) { rc = -1; goto done; } if (r1_weight < r2_weight) { rc = 1; goto done; } reason = "score"; for (gIter = nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; r1_node = NULL; r2_node = NULL; r1_weight = -INFINITY; if (r1_nodes) { r1_node = g_hash_table_lookup(r1_nodes, node->details->id); } if (r1_node) { r1_weight = r1_node->weight; } r2_weight = -INFINITY; if (r2_nodes) { r2_node = g_hash_table_lookup(r2_nodes, node->details->id); } if (r2_node) { r2_weight = r2_node->weight; } if (r1_weight > r2_weight) { rc = -1; goto done; } if (r1_weight < r2_weight) { rc = 1; goto done; } } done: crm_trace("%s (%d) on %s %c %s (%d) on %s: %s", resource1->id, r1_weight, r1_node ? r1_node->details->id : "n/a", rc < 0 ? '>' : rc > 0 ? '<' : '=', resource2->id, r2_weight, r2_node ? r2_node->details->id : "n/a", reason); if (r1_nodes) { g_hash_table_destroy(r1_nodes); } if (r2_nodes) { g_hash_table_destroy(r2_nodes); } return rc; } static void allocate_resources(pe_working_set_t * data_set) { GList *gIter = NULL; if (pcmk_is_set(data_set->flags, pe_flag_have_remote_nodes)) { /* Allocate remote connection resources first (which will also allocate * any colocation dependencies). If the connection is migrating, always * prefer the partial migration target. */ for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; if (rsc->is_remote_node == FALSE) { continue; } pe_rsc_trace(rsc, "Allocating remote connection resource '%s'", rsc->id); rsc->cmds->allocate(rsc, rsc->partial_migration_target, data_set); } } /* now do the rest of the resources */ for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; if (rsc->is_remote_node == TRUE) { continue; } pe_rsc_trace(rsc, "Allocating %s resource '%s'", crm_element_name(rsc->xml), rsc->id); rsc->cmds->allocate(rsc, NULL, data_set); } } /* We always use pe_order_preserve with these convenience functions to exempt * internally generated constraints from the prohibition of user constraints * involving remote connection resources. * * The start ordering additionally uses pe_order_runnable_left so that the * specified action is not runnable if the start is not runnable. */ static inline void order_start_then_action(pe_resource_t *lh_rsc, pe_action_t *rh_action, enum pe_ordering extra, pe_working_set_t *data_set) { if (lh_rsc && rh_action && data_set) { custom_action_order(lh_rsc, start_key(lh_rsc), NULL, rh_action->rsc, NULL, rh_action, pe_order_preserve | pe_order_runnable_left | extra, data_set); } } static inline void order_action_then_stop(pe_action_t *lh_action, pe_resource_t *rh_rsc, enum pe_ordering extra, pe_working_set_t *data_set) { if (lh_action && rh_rsc && data_set) { custom_action_order(lh_action->rsc, NULL, lh_action, rh_rsc, stop_key(rh_rsc), NULL, pe_order_preserve | extra, data_set); } } // Clear fail counts for orphaned rsc on all online nodes static void cleanup_orphans(pe_resource_t * rsc, pe_working_set_t * data_set) { GList *gIter = NULL; for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; if (node->details->online && pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL, data_set)) { pe_action_t *clear_op = NULL; clear_op = pe__clear_failcount(rsc, node, "it is orphaned", data_set); /* We can't use order_action_then_stop() here because its * pe_order_preserve breaks things */ custom_action_order(clear_op->rsc, NULL, clear_op, rsc, stop_key(rsc), NULL, pe_order_optional, data_set); } } } gboolean stage5(pe_working_set_t * data_set) { pcmk__output_t *out = data_set->priv; GList *gIter = NULL; if (!pcmk__str_eq(data_set->placement_strategy, "default", pcmk__str_casei)) { GList *nodes = g_list_copy(data_set->nodes); nodes = sort_nodes_by_weight(nodes, NULL, data_set); data_set->resources = g_list_sort_with_data(data_set->resources, sort_rsc_process_order, nodes); g_list_free(nodes); } gIter = data_set->nodes; for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; if (pcmk_is_set(data_set->flags, pe_flag_show_utilization)) { out->message(out, "node-capacity", node, "Original"); } } crm_trace("Allocating services"); /* Take (next) highest resource, assign it and create its actions */ allocate_resources(data_set); gIter = data_set->nodes; for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; if (pcmk_is_set(data_set->flags, pe_flag_show_utilization)) { out->message(out, "node-capacity", node, "Remaining"); } } // Process deferred action checks pe__foreach_param_check(data_set, check_params); pe__free_param_checks(data_set); if (pcmk_is_set(data_set->flags, pe_flag_startup_probes)) { crm_trace("Calculating needed probes"); /* This code probably needs optimization * ptest -x with 100 nodes, 100 clones and clone-max=100: With probes: ptest[14781]: 2010/09/27_17:56:46 notice: TRACE: do_calculations: pengine.c:258 Calculate cluster status ptest[14781]: 2010/09/27_17:56:46 notice: TRACE: do_calculations: pengine.c:278 Applying placement constraints ptest[14781]: 2010/09/27_17:56:47 notice: TRACE: do_calculations: pengine.c:285 Create internal constraints ptest[14781]: 2010/09/27_17:56:47 notice: TRACE: do_calculations: pengine.c:292 Check actions ptest[14781]: 2010/09/27_17:56:48 notice: TRACE: do_calculations: pengine.c:299 Allocate resources ptest[14781]: 2010/09/27_17:56:48 notice: TRACE: stage5: allocate.c:881 Allocating services ptest[14781]: 2010/09/27_17:56:49 notice: TRACE: stage5: allocate.c:894 Calculating needed probes ptest[14781]: 2010/09/27_17:56:51 notice: TRACE: stage5: allocate.c:899 Creating actions ptest[14781]: 2010/09/27_17:56:52 notice: TRACE: stage5: allocate.c:905 Creating done ptest[14781]: 2010/09/27_17:56:52 notice: TRACE: do_calculations: pengine.c:306 Processing fencing and shutdown cases ptest[14781]: 2010/09/27_17:56:52 notice: TRACE: do_calculations: pengine.c:313 Applying ordering constraints 36s ptest[14781]: 2010/09/27_17:57:28 notice: TRACE: do_calculations: pengine.c:320 Create transition graph Without probes: ptest[14637]: 2010/09/27_17:56:21 notice: TRACE: do_calculations: pengine.c:258 Calculate cluster status ptest[14637]: 2010/09/27_17:56:22 notice: TRACE: do_calculations: pengine.c:278 Applying placement constraints ptest[14637]: 2010/09/27_17:56:22 notice: TRACE: do_calculations: pengine.c:285 Create internal constraints ptest[14637]: 2010/09/27_17:56:22 notice: TRACE: do_calculations: pengine.c:292 Check actions ptest[14637]: 2010/09/27_17:56:23 notice: TRACE: do_calculations: pengine.c:299 Allocate resources ptest[14637]: 2010/09/27_17:56:23 notice: TRACE: stage5: allocate.c:881 Allocating services ptest[14637]: 2010/09/27_17:56:24 notice: TRACE: stage5: allocate.c:899 Creating actions ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: stage5: allocate.c:905 Creating done ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: do_calculations: pengine.c:306 Processing fencing and shutdown cases ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: do_calculations: pengine.c:313 Applying ordering constraints ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: do_calculations: pengine.c:320 Create transition graph */ probe_resources(data_set); } crm_trace("Handle orphans"); if (pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)) { for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; /* There's no need to recurse into rsc->children because those * should just be unallocated clone instances. */ if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) { cleanup_orphans(rsc, data_set); } } } crm_trace("Creating actions"); for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; rsc->cmds->create_actions(rsc, data_set); } crm_trace("Creating done"); return TRUE; } static gboolean is_managed(const pe_resource_t * rsc) { GList *gIter = rsc->children; if (pcmk_is_set(rsc->flags, pe_rsc_managed)) { return TRUE; } for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; if (is_managed(child_rsc)) { return TRUE; } } return FALSE; } static gboolean any_managed_resources(pe_working_set_t * data_set) { GList *gIter = data_set->resources; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; if (is_managed(rsc)) { return TRUE; } } return FALSE; } /*! * \internal * \brief Create pseudo-op for guest node fence, and order relative to it * * \param[in] node Guest node to fence * \param[in] data_set Working set of CIB state */ static void fence_guest(pe_node_t *node, pe_working_set_t *data_set) { pe_resource_t *container = node->details->remote_rsc->container; pe_action_t *stop = NULL; pe_action_t *stonith_op = NULL; /* The fence action is just a label; we don't do anything differently for * off vs. reboot. We specify it explicitly, rather than let it default to * cluster's default action, because we are not _initiating_ fencing -- we * are creating a pseudo-event to describe fencing that is already occurring * by other means (container recovery). */ const char *fence_action = "off"; /* Check whether guest's container resource has any explicit stop or * start (the stop may be implied by fencing of the guest's host). */ if (container) { stop = find_first_action(container->actions, NULL, CRMD_ACTION_STOP, NULL); if (find_first_action(container->actions, NULL, CRMD_ACTION_START, NULL)) { fence_action = "reboot"; } } /* Create a fence pseudo-event, so we have an event to order actions * against, and the controller can always detect it. */ stonith_op = pe_fence_op(node, fence_action, FALSE, "guest is unclean", FALSE, data_set); pe__set_action_flags(stonith_op, pe_action_pseudo|pe_action_runnable); /* We want to imply stops/demotes after the guest is stopped, not wait until * it is restarted, so we always order pseudo-fencing after stop, not start * (even though start might be closer to what is done for a real reboot). */ if ((stop != NULL) && pcmk_is_set(stop->flags, pe_action_pseudo)) { pe_action_t *parent_stonith_op = pe_fence_op(stop->node, NULL, FALSE, NULL, FALSE, data_set); crm_info("Implying guest node %s is down (action %d) after %s fencing", node->details->uname, stonith_op->id, stop->node->details->uname); order_actions(parent_stonith_op, stonith_op, pe_order_runnable_left|pe_order_implies_then); } else if (stop) { order_actions(stop, stonith_op, pe_order_runnable_left|pe_order_implies_then); crm_info("Implying guest node %s is down (action %d) " "after container %s is stopped (action %d)", node->details->uname, stonith_op->id, container->id, stop->id); } else { /* If we're fencing the guest node but there's no stop for the guest * resource, we must think the guest is already stopped. However, we may * think so because its resource history was just cleaned. To avoid * unnecessarily considering the guest node down if it's really up, * order the pseudo-fencing after any stop of the connection resource, * which will be ordered after any container (re-)probe. */ stop = find_first_action(node->details->remote_rsc->actions, NULL, RSC_STOP, NULL); if (stop) { order_actions(stop, stonith_op, pe_order_optional); crm_info("Implying guest node %s is down (action %d) " "after connection is stopped (action %d)", node->details->uname, stonith_op->id, stop->id); } else { /* Not sure why we're fencing, but everything must already be * cleanly stopped. */ crm_info("Implying guest node %s is down (action %d) ", node->details->uname, stonith_op->id); } } /* Order/imply other actions relative to pseudo-fence as with real fence */ pcmk__order_vs_fence(stonith_op, data_set); } /* * Create dependencies for stonith and shutdown operations */ gboolean stage6(pe_working_set_t * data_set) { pe_action_t *dc_down = NULL; pe_action_t *stonith_op = NULL; gboolean integrity_lost = FALSE; gboolean need_stonith = TRUE; GList *gIter; GList *stonith_ops = NULL; GList *shutdown_ops = NULL; /* Remote ordering constraints need to happen prior to calculating fencing * because it is one more place we will mark the node as dirty. * * A nice side effect of doing them early is that apply_*_ordering() can be * simpler because pe_fence_node() has already done some of the work. */ crm_trace("Creating remote ordering constraints"); apply_remote_node_ordering(data_set); crm_trace("Processing fencing and shutdown cases"); if (any_managed_resources(data_set) == FALSE) { crm_notice("Delaying fencing operations until there are resources to manage"); need_stonith = FALSE; } /* Check each node for stonith/shutdown */ for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; /* Guest nodes are "fenced" by recovering their container resource, * so handle them separately. */ if (pe__is_guest_node(node)) { if (node->details->remote_requires_reset && need_stonith && pe_can_fence(data_set, node)) { fence_guest(node, data_set); } continue; } stonith_op = NULL; if (node->details->unclean && need_stonith && pe_can_fence(data_set, node)) { stonith_op = pe_fence_op(node, NULL, FALSE, "node is unclean", FALSE, data_set); pe_warn("Scheduling Node %s for STONITH", node->details->uname); pcmk__order_vs_fence(stonith_op, data_set); if (node->details->is_dc) { // Remember if the DC is being fenced dc_down = stonith_op; } else { if (!pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing) && (stonith_ops != NULL)) { /* Concurrent fencing is disabled, so order each non-DC * fencing in a chain. If there is any DC fencing or * shutdown, it will be ordered after the last action in the * chain later. */ order_actions((pe_action_t *) stonith_ops->data, stonith_op, pe_order_optional); } // Remember all non-DC fencing actions in a separate list stonith_ops = g_list_prepend(stonith_ops, stonith_op); } } else if (node->details->online && node->details->shutdown && /* TODO define what a shutdown op means for a remote node. * For now we do not send shutdown operations for remote nodes, but * if we can come up with a good use for this in the future, we will. */ pe__is_guest_or_remote_node(node) == FALSE) { pe_action_t *down_op = sched_shutdown_op(node, data_set); if (node->details->is_dc) { // Remember if the DC is being shut down dc_down = down_op; } else { // Remember non-DC shutdowns for later ordering shutdown_ops = g_list_prepend(shutdown_ops, down_op); } } if (node->details->unclean && stonith_op == NULL) { integrity_lost = TRUE; pe_warn("Node %s is unclean!", node->details->uname); } } if (integrity_lost) { if (!pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { pe_warn("YOUR RESOURCES ARE NOW LIKELY COMPROMISED"); pe_err("ENABLE STONITH TO KEEP YOUR RESOURCES SAFE"); } else if (!pcmk_is_set(data_set->flags, pe_flag_have_quorum)) { crm_notice("Cannot fence unclean nodes until quorum is" " attained (or no-quorum-policy is set to ignore)"); } } if (dc_down != NULL) { /* Order any non-DC shutdowns before any DC shutdown, to avoid repeated * DC elections. However, we don't want to order non-DC shutdowns before * a DC *fencing*, because even though we don't want a node that's * shutting down to become DC, the DC fencing could be ordered before a * clone stop that's also ordered before the shutdowns, thus leading to * a graph loop. */ if (pcmk__str_eq(dc_down->task, CRM_OP_SHUTDOWN, pcmk__str_casei)) { for (gIter = shutdown_ops; gIter != NULL; gIter = gIter->next) { pe_action_t *node_stop = (pe_action_t *) gIter->data; crm_debug("Ordering shutdown on %s before %s on DC %s", node_stop->node->details->uname, dc_down->task, dc_down->node->details->uname); order_actions(node_stop, dc_down, pe_order_optional); } } // Order any non-DC fencing before any DC fencing or shutdown if (pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)) { /* With concurrent fencing, order each non-DC fencing action * separately before any DC fencing or shutdown. */ for (gIter = stonith_ops; gIter != NULL; gIter = gIter->next) { order_actions((pe_action_t *) gIter->data, dc_down, pe_order_optional); } } else if (stonith_ops) { /* Without concurrent fencing, the non-DC fencing actions are * already ordered relative to each other, so we just need to order * the DC fencing after the last action in the chain (which is the * first item in the list). */ order_actions((pe_action_t *) stonith_ops->data, dc_down, pe_order_optional); } } g_list_free(stonith_ops); g_list_free(shutdown_ops); return TRUE; } /* * Determine the sets of independent actions and the correct order for the * actions in each set. * * Mark dependencies of un-runnable actions un-runnable * */ static GList * find_actions_by_task(GList *actions, pe_resource_t * rsc, const char *original_key) { GList *list = NULL; list = find_actions(actions, original_key, NULL); if (list == NULL) { /* we're potentially searching a child of the original resource */ char *key = NULL; char *task = NULL; guint interval_ms = 0; if (parse_op_key(original_key, NULL, &task, &interval_ms)) { key = pcmk__op_key(rsc->id, task, interval_ms); list = find_actions(actions, key, NULL); } else { crm_err("search key: %s", original_key); } free(key); free(task); } return list; } static void rsc_order_then(pe_action_t *lh_action, pe_resource_t *rsc, pe__ordering_t *order) { GList *gIter = NULL; GList *rh_actions = NULL; pe_action_t *rh_action = NULL; enum pe_ordering type; CRM_CHECK(rsc != NULL, return); CRM_CHECK(order != NULL, return); type = order->type; rh_action = order->rh_action; crm_trace("Applying ordering constraint %d (then: %s)", order->id, rsc->id); if (rh_action != NULL) { rh_actions = g_list_prepend(NULL, rh_action); } else if (rsc != NULL) { rh_actions = find_actions_by_task(rsc->actions, rsc, order->rh_action_task); } if (rh_actions == NULL) { pe_rsc_trace(rsc, "Ignoring constraint %d: then (%s for %s) not found", order->id, order->rh_action_task, rsc->id); return; } if ((lh_action != NULL) && (lh_action->rsc == rsc) && pcmk_is_set(lh_action->flags, pe_action_dangle)) { pe_rsc_trace(rsc, "Detected dangling operation %s -> %s", lh_action->uuid, order->rh_action_task); pe__clear_order_flags(type, pe_order_implies_then); } gIter = rh_actions; for (; gIter != NULL; gIter = gIter->next) { pe_action_t *rh_action_iter = (pe_action_t *) gIter->data; if (lh_action) { order_actions(lh_action, rh_action_iter, type); } else if (type & pe_order_implies_then) { pe__clear_action_flags(rh_action_iter, pe_action_runnable); crm_warn("Unrunnable %s 0x%.6x", rh_action_iter->uuid, type); } else { crm_warn("neither %s 0x%.6x", rh_action_iter->uuid, type); } } g_list_free(rh_actions); } static void rsc_order_first(pe_resource_t *lh_rsc, pe__ordering_t *order, pe_working_set_t *data_set) { GList *lh_actions = NULL; pe_action_t *lh_action = order->lh_action; pe_resource_t *rh_rsc = order->rh_rsc; CRM_ASSERT(lh_rsc != NULL); pe_rsc_trace(lh_rsc, "Applying ordering constraint %d (first: %s)", order->id, lh_rsc->id); if (lh_action != NULL) { lh_actions = g_list_prepend(NULL, lh_action); } else { lh_actions = find_actions_by_task(lh_rsc->actions, lh_rsc, order->lh_action_task); } if ((lh_actions == NULL) && (lh_rsc == rh_rsc)) { pe_rsc_trace(lh_rsc, "Ignoring constraint %d: first (%s for %s) not found", order->id, order->lh_action_task, lh_rsc->id); } else if (lh_actions == NULL) { char *key = NULL; char *op_type = NULL; guint interval_ms = 0; parse_op_key(order->lh_action_task, NULL, &op_type, &interval_ms); key = pcmk__op_key(lh_rsc->id, op_type, interval_ms); if (lh_rsc->fns->state(lh_rsc, TRUE) == RSC_ROLE_STOPPED && pcmk__str_eq(op_type, RSC_STOP, pcmk__str_casei)) { free(key); pe_rsc_trace(lh_rsc, "Ignoring constraint %d: first (%s for %s) not found", order->id, order->lh_action_task, lh_rsc->id); } else if ((lh_rsc->fns->state(lh_rsc, TRUE) == RSC_ROLE_UNPROMOTED) && pcmk__str_eq(op_type, RSC_DEMOTE, pcmk__str_casei)) { free(key); pe_rsc_trace(lh_rsc, "Ignoring constraint %d: first (%s for %s) not found", order->id, order->lh_action_task, lh_rsc->id); } else { pe_rsc_trace(lh_rsc, "Creating first (%s for %s) for constraint %d ", order->lh_action_task, lh_rsc->id, order->id); lh_action = custom_action(lh_rsc, key, op_type, NULL, TRUE, TRUE, data_set); lh_actions = g_list_prepend(NULL, lh_action); } free(op_type); } if (rh_rsc == NULL) { if (order->rh_action == NULL) { pe_rsc_trace(lh_rsc, "Ignoring constraint %d: then not found", order->id); return; } rh_rsc = order->rh_action->rsc; } for (GList *gIter = lh_actions; gIter != NULL; gIter = gIter->next) { lh_action = (pe_action_t *) gIter->data; if (rh_rsc == NULL) { order_actions(lh_action, order->rh_action, order->type); } else { rsc_order_then(lh_action, rh_rsc, order); } } g_list_free(lh_actions); } extern void update_colo_start_chain(pe_action_t *action, pe_working_set_t *data_set); static int is_recurring_action(pe_action_t *action) { guint interval_ms; if (pcmk__guint_from_hash(action->meta, XML_LRM_ATTR_INTERVAL_MS, 0, &interval_ms) != pcmk_rc_ok) { return 0; } return (interval_ms > 0); } static void apply_container_ordering(pe_action_t *action, pe_working_set_t *data_set) { /* VMs are also classified as containers for these purposes... in * that they both involve a 'thing' running on a real or remote * cluster node. * * This allows us to be smarter about the type and extent of * recovery actions required in various scenarios */ pe_resource_t *remote_rsc = NULL; pe_resource_t *container = NULL; enum action_tasks task = text2task(action->task); CRM_ASSERT(action->rsc); CRM_ASSERT(action->node); CRM_ASSERT(pe__is_guest_or_remote_node(action->node)); remote_rsc = action->node->details->remote_rsc; CRM_ASSERT(remote_rsc); container = remote_rsc->container; CRM_ASSERT(container); if (pcmk_is_set(container->flags, pe_rsc_failed)) { pe_fence_node(data_set, action->node, "container failed", FALSE); } crm_trace("Order %s action %s relative to %s%s for %s%s", action->task, action->uuid, pcmk_is_set(remote_rsc->flags, pe_rsc_failed)? "failed " : "", remote_rsc->id, pcmk_is_set(container->flags, pe_rsc_failed)? "failed " : "", container->id); if (pcmk__strcase_any_of(action->task, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, NULL)) { /* Migration ops map to "no_action", but we need to apply the same * ordering as for stop or demote (see get_router_node()). */ task = stop_rsc; } switch (task) { case start_rsc: case action_promote: /* Force resource recovery if the container is recovered */ order_start_then_action(container, action, pe_order_implies_then, data_set); /* Wait for the connection resource to be up too */ order_start_then_action(remote_rsc, action, pe_order_none, data_set); break; case stop_rsc: case action_demote: if (pcmk_is_set(container->flags, pe_rsc_failed)) { /* When the container representing a guest node fails, any stop * or demote actions for resources running on the guest node * are implied by the container stopping. This is similar to * how fencing operations work for cluster nodes and remote * nodes. */ } else { /* Ensure the operation happens before the connection is brought * down. * * If we really wanted to, we could order these after the * connection start, IFF the container's current role was * stopped (otherwise we re-introduce an ordering loop when the * connection is restarting). */ order_action_then_stop(action, remote_rsc, pe_order_none, data_set); } break; default: /* Wait for the connection resource to be up */ if (is_recurring_action(action)) { /* In case we ever get the recovery logic wrong, force * recurring monitors to be restarted, even if just * the connection was re-established */ if(task != no_action) { order_start_then_action(remote_rsc, action, pe_order_implies_then, data_set); } } else { order_start_then_action(remote_rsc, action, pe_order_none, data_set); } break; } } static enum remote_connection_state get_remote_node_state(pe_node_t *node) { pe_resource_t *remote_rsc = NULL; pe_node_t *cluster_node = NULL; CRM_ASSERT(node); remote_rsc = node->details->remote_rsc; CRM_ASSERT(remote_rsc); cluster_node = pe__current_node(remote_rsc); /* If the cluster node the remote connection resource resides on * is unclean or went offline, we can't process any operations * on that remote node until after it starts elsewhere. */ if(remote_rsc->next_role == RSC_ROLE_STOPPED || remote_rsc->allocated_to == NULL) { /* The connection resource is not going to run anywhere */ if (cluster_node && cluster_node->details->unclean) { /* The remote connection is failed because its resource is on a * failed node and can't be recovered elsewhere, so we must fence. */ return remote_state_failed; } if (!pcmk_is_set(remote_rsc->flags, pe_rsc_failed)) { /* Connection resource is cleanly stopped */ return remote_state_stopped; } /* Connection resource is failed */ if ((remote_rsc->next_role == RSC_ROLE_STOPPED) && remote_rsc->remote_reconnect_ms && node->details->remote_was_fenced && !pe__shutdown_requested(node)) { /* We won't know whether the connection is recoverable until the * reconnect interval expires and we reattempt connection. */ return remote_state_unknown; } /* The remote connection is in a failed state. If there are any * resources known to be active on it (stop) or in an unknown state * (probe), we must assume the worst and fence it. */ return remote_state_failed; } else if (cluster_node == NULL) { /* Connection is recoverable but not currently running anywhere, see if we can recover it first */ return remote_state_unknown; } else if(cluster_node->details->unclean == TRUE || cluster_node->details->online == FALSE) { /* Connection is running on a dead node, see if we can recover it first */ return remote_state_resting; } else if (pcmk__list_of_multiple(remote_rsc->running_on) && remote_rsc->partial_migration_source && remote_rsc->partial_migration_target) { /* We're in the middle of migrating a connection resource, * wait until after the resource migrates before performing * any actions. */ return remote_state_resting; } return remote_state_alive; } /*! * \internal * \brief Order actions on remote node relative to actions for the connection */ static void apply_remote_ordering(pe_action_t *action, pe_working_set_t *data_set) { pe_resource_t *remote_rsc = NULL; enum action_tasks task = text2task(action->task); enum remote_connection_state state = get_remote_node_state(action->node); enum pe_ordering order_opts = pe_order_none; if (action->rsc == NULL) { return; } CRM_ASSERT(action->node); CRM_ASSERT(pe__is_guest_or_remote_node(action->node)); remote_rsc = action->node->details->remote_rsc; CRM_ASSERT(remote_rsc); crm_trace("Order %s action %s relative to %s%s (state: %s)", action->task, action->uuid, pcmk_is_set(remote_rsc->flags, pe_rsc_failed)? "failed " : "", remote_rsc->id, state2text(state)); if (pcmk__strcase_any_of(action->task, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, NULL)) { /* Migration ops map to "no_action", but we need to apply the same * ordering as for stop or demote (see get_router_node()). */ task = stop_rsc; } switch (task) { case start_rsc: case action_promote: order_opts = pe_order_none; if (state == remote_state_failed) { /* Force recovery, by making this action required */ pe__set_order_flags(order_opts, pe_order_implies_then); } /* Ensure connection is up before running this action */ order_start_then_action(remote_rsc, action, order_opts, data_set); break; case stop_rsc: if(state == remote_state_alive) { order_action_then_stop(action, remote_rsc, pe_order_implies_first, data_set); } else if(state == remote_state_failed) { /* The resource is active on the node, but since we don't have a * valid connection, the only way to stop the resource is by * fencing the node. There is no need to order the stop relative * to the remote connection, since the stop will become implied * by the fencing. */ pe_fence_node(data_set, action->node, "resources are active and the connection is unrecoverable", FALSE); } else if(remote_rsc->next_role == RSC_ROLE_STOPPED) { /* State must be remote_state_unknown or remote_state_stopped. * Since the connection is not coming back up in this * transition, stop this resource first. */ order_action_then_stop(action, remote_rsc, pe_order_implies_first, data_set); } else { /* The connection is going to be started somewhere else, so * stop this resource after that completes. */ order_start_then_action(remote_rsc, action, pe_order_none, data_set); } break; case action_demote: /* Only order this demote relative to the connection start if the * connection isn't being torn down. Otherwise, the demote would be * blocked because the connection start would not be allowed. */ if(state == remote_state_resting || state == remote_state_unknown) { order_start_then_action(remote_rsc, action, pe_order_none, data_set); } /* Otherwise we can rely on the stop ordering */ break; default: /* Wait for the connection resource to be up */ if (is_recurring_action(action)) { /* In case we ever get the recovery logic wrong, force * recurring monitors to be restarted, even if just * the connection was re-established */ order_start_then_action(remote_rsc, action, pe_order_implies_then, data_set); } else { pe_node_t *cluster_node = pe__current_node(remote_rsc); if(task == monitor_rsc && state == remote_state_failed) { /* We would only be here if we do not know the * state of the resource on the remote node. * Since we have no way to find out, it is * necessary to fence the node. */ pe_fence_node(data_set, action->node, "resources are in an unknown state and the connection is unrecoverable", FALSE); } if(cluster_node && state == remote_state_stopped) { /* The connection is currently up, but is going * down permanently. * * Make sure we check services are actually * stopped _before_ we let the connection get * closed */ order_action_then_stop(action, remote_rsc, pe_order_runnable_left, data_set); } else { order_start_then_action(remote_rsc, action, pe_order_none, data_set); } } break; } } static void apply_remote_node_ordering(pe_working_set_t *data_set) { if (!pcmk_is_set(data_set->flags, pe_flag_have_remote_nodes)) { return; } for (GList *gIter = data_set->actions; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; pe_resource_t *remote = NULL; // We are only interested in resource actions if (action->rsc == NULL) { continue; } /* Special case: If we are clearing the failcount of an actual * remote connection resource, then make sure this happens before * any start of the resource in this transition. */ if (action->rsc->is_remote_node && pcmk__str_eq(action->task, CRM_OP_CLEAR_FAILCOUNT, pcmk__str_casei)) { custom_action_order(action->rsc, NULL, action, action->rsc, pcmk__op_key(action->rsc->id, RSC_START, 0), NULL, pe_order_optional, data_set); continue; } // We are only interested in actions allocated to a node if (action->node == NULL) { continue; } if (!pe__is_guest_or_remote_node(action->node)) { continue; } /* We are only interested in real actions. * * @TODO This is probably wrong; pseudo-actions might be converted to * real actions and vice versa later in update_actions() at the end of * stage7(). */ if (pcmk_is_set(action->flags, pe_action_pseudo)) { continue; } remote = action->node->details->remote_rsc; if (remote == NULL) { // Orphaned continue; } /* Another special case: if a resource is moving to a Pacemaker Remote * node, order the stop on the original node after any start of the * remote connection. This ensures that if the connection fails to * start, we leave the resource running on the original node. */ if (pcmk__str_eq(action->task, RSC_START, pcmk__str_casei)) { for (GList *item = action->rsc->actions; item != NULL; item = item->next) { pe_action_t *rsc_action = item->data; if ((rsc_action->node->details != action->node->details) && pcmk__str_eq(rsc_action->task, RSC_STOP, pcmk__str_casei)) { custom_action_order(remote, start_key(remote), NULL, action->rsc, NULL, rsc_action, pe_order_optional, data_set); } } } /* The action occurs across a remote connection, so create * ordering constraints that guarantee the action occurs while the node * is active (after start, before stop ... things like that). * * This is somewhat brittle in that we need to make sure the results of * this ordering are compatible with the result of get_router_node(). * It would probably be better to add XML_LRM_ATTR_ROUTER_NODE as part * of this logic rather than action2xml(). */ if (remote->container) { crm_trace("Container ordering for %s", action->uuid); apply_container_ordering(action, data_set); } else { crm_trace("Remote ordering for %s", action->uuid); apply_remote_ordering(action, data_set); } } } static gboolean order_first_probe_unneeded(pe_action_t * probe, pe_action_t * rh_action) { /* No need to probe the resource on the node that is being * unfenced. Otherwise it might introduce transition loop * since probe will be performed after the node is * unfenced. */ if (pcmk__str_eq(rh_action->task, CRM_OP_FENCE, pcmk__str_casei) && probe->node && rh_action->node && probe->node->details == rh_action->node->details) { const char *op = g_hash_table_lookup(rh_action->meta, "stonith_action"); if (pcmk__str_eq(op, "on", pcmk__str_casei)) { return TRUE; } } // Shutdown waits for probe to complete only if it's on the same node if ((pcmk__str_eq(rh_action->task, CRM_OP_SHUTDOWN, pcmk__str_casei)) && probe->node && rh_action->node && probe->node->details != rh_action->node->details) { return TRUE; } return FALSE; } static void order_first_probes_imply_stops(pe_working_set_t * data_set) { GList *gIter = NULL; for (gIter = data_set->ordering_constraints; gIter != NULL; gIter = gIter->next) { pe__ordering_t *order = gIter->data; enum pe_ordering order_type = pe_order_optional; pe_resource_t *lh_rsc = order->lh_rsc; pe_resource_t *rh_rsc = order->rh_rsc; pe_action_t *lh_action = order->lh_action; pe_action_t *rh_action = order->rh_action; const char *lh_action_task = order->lh_action_task; const char *rh_action_task = order->rh_action_task; GList *probes = NULL; GList *rh_actions = NULL; GList *pIter = NULL; if (lh_rsc == NULL) { continue; } else if (rh_rsc && lh_rsc == rh_rsc) { continue; } if (lh_action == NULL && lh_action_task == NULL) { continue; } if (rh_action == NULL && rh_action_task == NULL) { continue; } /* Technically probe is expected to return "not running", which could be * the alternative of stop action if the status of the resource is * unknown yet. */ if (lh_action && !pcmk__str_eq(lh_action->task, RSC_STOP, pcmk__str_casei)) { continue; } else if (lh_action == NULL && lh_action_task && !pcmk__ends_with(lh_action_task, "_" RSC_STOP "_0")) { continue; } /* Do not probe the resource inside of a stopping container. Otherwise * it might introduce transition loop since probe will be performed * after the container starts again. */ if (rh_rsc && lh_rsc->container == rh_rsc) { if (rh_action && pcmk__str_eq(rh_action->task, RSC_STOP, pcmk__str_casei)) { continue; } else if (rh_action == NULL && rh_action_task && pcmk__ends_with(rh_action_task,"_" RSC_STOP "_0")) { continue; } } if (order->type == pe_order_none) { continue; } // Preserve the order options for future filtering if (pcmk_is_set(order->type, pe_order_apply_first_non_migratable)) { pe__set_order_flags(order_type, pe_order_apply_first_non_migratable); } if (pcmk_is_set(order->type, pe_order_same_node)) { pe__set_order_flags(order_type, pe_order_same_node); } // Keep the order types for future filtering if (order->type == pe_order_anti_colocation || order->type == pe_order_load) { order_type = order->type; } probes = pe__resource_actions(lh_rsc, NULL, RSC_STATUS, FALSE); if (probes == NULL) { continue; } if (rh_action) { rh_actions = g_list_prepend(rh_actions, rh_action); } else if (rh_rsc && rh_action_task) { rh_actions = find_actions(rh_rsc->actions, rh_action_task, NULL); } if (rh_actions == NULL) { g_list_free(probes); continue; } crm_trace("Processing for LH probe based on ordering constraint %s -> %s" " (id=%d, type=%.6x)", lh_action ? lh_action->uuid : lh_action_task, rh_action ? rh_action->uuid : rh_action_task, order->id, order->type); for (pIter = probes; pIter != NULL; pIter = pIter->next) { pe_action_t *probe = (pe_action_t *) pIter->data; GList *rIter = NULL; for (rIter = rh_actions; rIter != NULL; rIter = rIter->next) { pe_action_t *rh_action_iter = (pe_action_t *) rIter->data; if (order_first_probe_unneeded(probe, rh_action_iter)) { continue; } order_actions(probe, rh_action_iter, order_type); } } g_list_free(rh_actions); g_list_free(probes); } } static void order_first_probe_then_restart_repromote(pe_action_t * probe, pe_action_t * after, pe_working_set_t * data_set) { GList *gIter = NULL; bool interleave = FALSE; pe_resource_t *compatible_rsc = NULL; if (probe == NULL || probe->rsc == NULL || probe->rsc->variant != pe_native) { return; } if (after == NULL // Avoid running into any possible loop || pcmk_is_set(after->flags, pe_action_tracking)) { return; } if (!pcmk__str_eq(probe->task, RSC_STATUS, pcmk__str_casei)) { return; } pe__set_action_flags(after, pe_action_tracking); crm_trace("Processing based on %s %s -> %s %s", probe->uuid, probe->node ? probe->node->details->uname: "", after->uuid, after->node ? after->node->details->uname : ""); if (after->rsc /* Better not build a dependency directly with a clone/group. * We are going to proceed through the ordering chain and build * dependencies with its children. */ && after->rsc->variant == pe_native && probe->rsc != after->rsc) { GList *then_actions = NULL; enum pe_ordering probe_order_type = pe_order_optional; if (pcmk__str_eq(after->task, RSC_START, pcmk__str_casei)) { then_actions = pe__resource_actions(after->rsc, NULL, RSC_STOP, FALSE); } else if (pcmk__str_eq(after->task, RSC_PROMOTE, pcmk__str_casei)) { then_actions = pe__resource_actions(after->rsc, NULL, RSC_DEMOTE, FALSE); } for (gIter = then_actions; gIter != NULL; gIter = gIter->next) { pe_action_t *then = (pe_action_t *) gIter->data; // Skip any pseudo action which for example is implied by fencing if (pcmk_is_set(then->flags, pe_action_pseudo)) { continue; } order_actions(probe, then, probe_order_type); } g_list_free(then_actions); } if (after->rsc && after->rsc->variant > pe_group) { const char *interleave_s = g_hash_table_lookup(after->rsc->meta, XML_RSC_ATTR_INTERLEAVE); interleave = crm_is_true(interleave_s); if (interleave) { /* For an interleaved clone, we should build a dependency only * with the relevant clone child. */ compatible_rsc = find_compatible_child(probe->rsc, after->rsc, RSC_ROLE_UNKNOWN, FALSE, data_set); } } for (gIter = after->actions_after; gIter != NULL; gIter = gIter->next) { pe_action_wrapper_t *after_wrapper = (pe_action_wrapper_t *) gIter->data; /* pe_order_implies_then is the reason why a required A.start * implies/enforces B.start to be required too, which is the cause of * B.restart/re-promote. * * Not sure about pe_order_implies_then_on_node though. It's now only * used for unfencing case, which tends to introduce transition * loops... */ if (!pcmk_is_set(after_wrapper->type, pe_order_implies_then)) { /* The order type between a group/clone and its child such as * B.start-> B_child.start is: * pe_order_implies_first_printed | pe_order_runnable_left * * Proceed through the ordering chain and build dependencies with * its children. */ if (after->rsc == NULL || after->rsc->variant < pe_group || probe->rsc->parent == after->rsc || after_wrapper->action->rsc == NULL || after_wrapper->action->rsc->variant > pe_group || after->rsc != after_wrapper->action->rsc->parent) { continue; } /* Proceed to the children of a group or a non-interleaved clone. * For an interleaved clone, proceed only to the relevant child. */ if (after->rsc->variant > pe_group && interleave == TRUE && (compatible_rsc == NULL || compatible_rsc != after_wrapper->action->rsc)) { continue; } } crm_trace("Proceeding through %s %s -> %s %s (type=0x%.6x)", after->uuid, after->node ? after->node->details->uname: "", after_wrapper->action->uuid, after_wrapper->action->node ? after_wrapper->action->node->details->uname : "", after_wrapper->type); order_first_probe_then_restart_repromote(probe, after_wrapper->action, data_set); } } static void clear_actions_tracking_flag(pe_working_set_t * data_set) { GList *gIter = NULL; for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (pcmk_is_set(action->flags, pe_action_tracking)) { pe__clear_action_flags(action, pe_action_tracking); } } } static void order_first_rsc_probes(pe_resource_t * rsc, pe_working_set_t * data_set) { GList *gIter = NULL; GList *probes = NULL; g_list_foreach(rsc->children, (GFunc) order_first_rsc_probes, data_set); if (rsc->variant != pe_native) { return; } probes = pe__resource_actions(rsc, NULL, RSC_STATUS, FALSE); for (gIter = probes; gIter != NULL; gIter= gIter->next) { pe_action_t *probe = (pe_action_t *) gIter->data; GList *aIter = NULL; for (aIter = probe->actions_after; aIter != NULL; aIter = aIter->next) { pe_action_wrapper_t *after_wrapper = (pe_action_wrapper_t *) aIter->data; order_first_probe_then_restart_repromote(probe, after_wrapper->action, data_set); clear_actions_tracking_flag(data_set); } } g_list_free(probes); } static void order_first_probes(pe_working_set_t * data_set) { GList *gIter = NULL; for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; order_first_rsc_probes(rsc, data_set); } order_first_probes_imply_stops(data_set); } static void order_then_probes(pe_working_set_t * data_set) { #if 0 GList *gIter = NULL; for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; /* Given "A then B", we would prefer to wait for A to be * started before probing B. * * If A was a filesystem on which the binaries and data for B * lived, it would have been useful if the author of B's agent * could assume that A is running before B.monitor will be * called. * * However we can't _only_ probe once A is running, otherwise * we'd not detect the state of B if A could not be started * for some reason. * * In practice however, we cannot even do an opportunistic * version of this because B may be moving: * * B.probe -> B.start * B.probe -> B.stop * B.stop -> B.start * A.stop -> A.start * A.start -> B.probe * * So far so good, but if we add the result of this code: * * B.stop -> A.stop * * Then we get a loop: * * B.probe -> B.stop -> A.stop -> A.start -> B.probe * * We could kill the 'B.probe -> B.stop' dependency, but that * could mean stopping B "too" soon, because B.start must wait * for the probes to complete. * * Another option is to allow it only if A is a non-unique * clone with clone-max == node-max (since we'll never be * moving it). However, we could still be stopping one * instance at the same time as starting another. * The complexity of checking for allowed conditions combined * with the ever narrowing usecase suggests that this code * should remain disabled until someone gets smarter. */ pe_action_t *start = NULL; GList *actions = NULL; GList *probes = NULL; actions = pe__resource_actions(rsc, NULL, RSC_START, FALSE); if (actions) { start = actions->data; g_list_free(actions); } if(start == NULL) { crm_err("No start action for %s", rsc->id); continue; } probes = pe__resource_actions(rsc, NULL, RSC_STATUS, FALSE); for (actions = start->actions_before; actions != NULL; actions = actions->next) { pe_action_wrapper_t *before = (pe_action_wrapper_t *) actions->data; GList *pIter = NULL; pe_action_t *first = before->action; pe_resource_t *first_rsc = first->rsc; if(first->required_runnable_before) { GList *clone_actions = NULL; for (clone_actions = first->actions_before; clone_actions != NULL; clone_actions = clone_actions->next) { before = (pe_action_wrapper_t *) clone_actions->data; crm_trace("Testing %s -> %s (%p) for %s", first->uuid, before->action->uuid, before->action->rsc, start->uuid); CRM_ASSERT(before->action->rsc); first_rsc = before->action->rsc; break; } } else if(!pcmk__str_eq(first->task, RSC_START, pcmk__str_casei)) { crm_trace("Not a start op %s for %s", first->uuid, start->uuid); } if(first_rsc == NULL) { continue; } else if(uber_parent(first_rsc) == uber_parent(start->rsc)) { crm_trace("Same parent %s for %s", first_rsc->id, start->uuid); continue; } else if(FALSE && pe_rsc_is_clone(uber_parent(first_rsc)) == FALSE) { crm_trace("Not a clone %s for %s", first_rsc->id, start->uuid); continue; } crm_err("Applying %s before %s %d", first->uuid, start->uuid, uber_parent(first_rsc)->variant); for (pIter = probes; pIter != NULL; pIter = pIter->next) { pe_action_t *probe = (pe_action_t *) pIter->data; crm_err("Ordering %s before %s", first->uuid, probe->uuid); order_actions(first, probe, pe_order_optional); } } } #endif } static void order_probes(pe_working_set_t * data_set) { order_first_probes(data_set); order_then_probes(data_set); } gboolean stage7(pe_working_set_t * data_set) { pcmk__output_t *prev_out = data_set->priv; pcmk__output_t *out = NULL; GList *gIter = NULL; crm_trace("Applying ordering constraints"); /* Don't ask me why, but apparently they need to be processed in * the order they were created in... go figure * * Also g_list_append() has horrendous performance characteristics * So we need to use g_list_prepend() and then reverse the list here */ data_set->ordering_constraints = g_list_reverse(data_set->ordering_constraints); for (gIter = data_set->ordering_constraints; gIter != NULL; gIter = gIter->next) { pe__ordering_t *order = gIter->data; pe_resource_t *rsc = order->lh_rsc; if (rsc != NULL) { rsc_order_first(rsc, order, data_set); continue; } rsc = order->rh_rsc; if (rsc != NULL) { rsc_order_then(order->lh_action, rsc, order); } else { crm_trace("Applying ordering constraint %d (non-resource actions)", order->id); order_actions(order->lh_action, order->rh_action, order->type); } } g_list_foreach(data_set->actions, (GFunc) update_colo_start_chain, data_set); crm_trace("Ordering probes"); order_probes(data_set); crm_trace("Updating %d actions", g_list_length(data_set->actions)); g_list_foreach(data_set->actions, (GFunc) update_action, data_set); // Check for invalid orderings for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; pe_action_wrapper_t *input = NULL; for (GList *input_iter = action->actions_before; input_iter != NULL; input_iter = input_iter->next) { input = (pe_action_wrapper_t *) input_iter->data; if (pcmk__ordering_is_invalid(action, input)) { input->type = pe_order_none; } } } /* stage7 only ever outputs to the log, so ignore whatever output object was * previously set and just log instead. */ out = pcmk__new_logger(); if (out == NULL) { return FALSE; } pcmk__output_set_log_level(out, LOG_NOTICE); data_set->priv = out; out->begin_list(out, NULL, NULL, "Actions"); LogNodeActions(data_set); g_list_foreach(data_set->resources, (GFunc) LogActions, data_set); out->end_list(out); out->finish(out, CRM_EX_OK, true, NULL); pcmk__output_free(out); data_set->priv = prev_out; return TRUE; } static int transition_id = -1; /*! * \internal * \brief Log a message after calculating a transition * * \param[in] filename Where transition input is stored */ void pcmk__log_transition_summary(const char *filename) { if (was_processing_error) { crm_err("Calculated transition %d (with errors)%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } else if (was_processing_warning) { crm_warn("Calculated transition %d (with warnings)%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } else { crm_notice("Calculated transition %d%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } if (crm_config_error) { crm_notice("Configuration errors found during scheduler processing," " please run \"crm_verify -L\" to identify issues"); } } /* * Create a dependency graph to send to the transitioner (via the controller) */ gboolean stage8(pe_working_set_t * data_set) { GList *gIter = NULL; const char *value = NULL; long long limit = 0LL; transition_id++; crm_trace("Creating transition graph %d.", transition_id); data_set->graph = create_xml_node(NULL, XML_TAG_GRAPH); value = pe_pref(data_set->config_hash, "cluster-delay"); crm_xml_add(data_set->graph, "cluster-delay", value); value = pe_pref(data_set->config_hash, "stonith-timeout"); crm_xml_add(data_set->graph, "stonith-timeout", value); crm_xml_add(data_set->graph, "failed-stop-offset", "INFINITY"); if (pcmk_is_set(data_set->flags, pe_flag_start_failure_fatal)) { crm_xml_add(data_set->graph, "failed-start-offset", "INFINITY"); } else { crm_xml_add(data_set->graph, "failed-start-offset", "1"); } value = pe_pref(data_set->config_hash, "batch-limit"); crm_xml_add(data_set->graph, "batch-limit", value); crm_xml_add_int(data_set->graph, "transition_id", transition_id); value = pe_pref(data_set->config_hash, "migration-limit"); if ((pcmk__scan_ll(value, &limit, 0LL) == pcmk_rc_ok) && (limit > 0)) { crm_xml_add(data_set->graph, "migration-limit", value); } if (data_set->recheck_by > 0) { char *recheck_epoch = NULL; recheck_epoch = crm_strdup_printf("%llu", (long long) data_set->recheck_by); crm_xml_add(data_set->graph, "recheck-by", recheck_epoch); free(recheck_epoch); } /* The following code will de-duplicate action inputs, so nothing past this * should rely on the action input type flags retaining their original * values. */ gIter = data_set->resources; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; pe_rsc_trace(rsc, "processing actions for rsc=%s", rsc->id); rsc->cmds->expand(rsc, data_set); } crm_log_xml_trace(data_set->graph, "created resource-driven action list"); /* pseudo action to distribute list of nodes with maintenance state update */ add_maintenance_update(data_set); /* catch any non-resource specific actions */ crm_trace("processing non-resource actions"); gIter = data_set->actions; for (; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (action->rsc && action->node && action->node->details->shutdown && !pcmk_is_set(action->rsc->flags, pe_rsc_maintenance) && !pcmk_any_flags_set(action->flags, pe_action_optional|pe_action_runnable) && pcmk__str_eq(action->task, RSC_STOP, pcmk__str_none) ) { /* Eventually we should just ignore the 'fence' case * But for now it's the best way to detect (in CTS) when * CIB resource updates are being lost */ if (pcmk_is_set(data_set->flags, pe_flag_have_quorum) || data_set->no_quorum_policy == no_quorum_ignore) { crm_crit("Cannot %s node '%s' because of %s:%s%s (%s)", action->node->details->unclean ? "fence" : "shut down", action->node->details->uname, action->rsc->id, pcmk_is_set(action->rsc->flags, pe_rsc_managed)? " blocked" : " unmanaged", pcmk_is_set(action->rsc->flags, pe_rsc_failed)? " failed" : "", action->uuid); } } graph_element_from_action(action, data_set); } crm_log_xml_trace(data_set->graph, "created generic action list"); crm_trace("Created transition graph %d.", transition_id); return TRUE; } void LogNodeActions(pe_working_set_t * data_set) { pcmk__output_t *out = data_set->priv; GList *gIter = NULL; for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) { char *node_name = NULL; char *task = NULL; pe_action_t *action = (pe_action_t *) gIter->data; if (action->rsc != NULL) { continue; } else if (pcmk_is_set(action->flags, pe_action_optional)) { continue; } if (pe__is_guest_node(action->node)) { node_name = crm_strdup_printf("%s (resource: %s)", action->node->details->uname, action->node->details->remote_rsc->container->id); } else if(action->node) { node_name = crm_strdup_printf("%s", action->node->details->uname); } if (pcmk__str_eq(action->task, CRM_OP_SHUTDOWN, pcmk__str_casei)) { task = strdup("Shutdown"); } else if (pcmk__str_eq(action->task, CRM_OP_FENCE, pcmk__str_casei)) { const char *op = g_hash_table_lookup(action->meta, "stonith_action"); task = crm_strdup_printf("Fence (%s)", op); } out->message(out, "node-action", task, node_name, action->reason); free(node_name); free(task); } } diff --git a/lib/pacemaker/pcmk_sched_bundle.c b/lib/pacemaker/pcmk_sched_bundle.c index 64c4094a87..660cc085f9 100644 --- a/lib/pacemaker/pcmk_sched_bundle.c +++ b/lib/pacemaker/pcmk_sched_bundle.c @@ -1,1103 +1,1064 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #define PE__VARIANT_BUNDLE 1 #include static bool is_bundle_node(pe__bundle_variant_data_t *data, pe_node_t *node) { for (GList *gIter = data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (node->details == replica->node->details) { return TRUE; } } return FALSE; } gint sort_clone_instance(gconstpointer a, gconstpointer b, gpointer data_set); void distribute_children(pe_resource_t *rsc, GList *children, GList *nodes, int max, int per_host_max, pe_working_set_t * data_set); static GList * get_container_list(pe_resource_t *rsc) { GList *containers = NULL; if (rsc->variant == pe_container) { pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, rsc); for (GList *gIter = data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; containers = g_list_append(containers, replica->container); } } return containers; } static inline GList * get_containers_or_children(pe_resource_t *rsc) { return (rsc->variant == pe_container)? get_container_list(rsc) : rsc->children; } -static bool -migration_threshold_reached(pe_resource_t *rsc, pe_node_t *node, - pe_working_set_t *data_set) -{ - int fail_count, countdown; - - /* Migration threshold of 0 means never force away */ - if (rsc->migration_threshold == 0) { - return FALSE; - } - - // If we're ignoring failures, also ignore the migration threshold - if (pcmk_is_set(rsc->flags, pe_rsc_failure_ignored)) { - return FALSE; - } - - /* If there are no failures, there's no need to force away */ - fail_count = pe_get_failcount(node, rsc, NULL, - pe_fc_effective|pe_fc_fillers, NULL, - data_set); - if (fail_count <= 0) { - return FALSE; - } - - /* How many more times recovery will be tried on this node */ - countdown = QB_MAX(rsc->migration_threshold - fail_count, 0); - - if (countdown == 0) { - crm_warn("Forcing %s away from %s after %d failures (max=%d)", - rsc->id, node->details->uname, fail_count, - rsc->migration_threshold); - return TRUE; - } - - crm_info("%s can fail %d more times on %s before being forced off", - rsc->id, countdown, node->details->uname); - return FALSE; -} - pe_node_t * pcmk__bundle_allocate(pe_resource_t *rsc, pe_node_t *prefer, pe_working_set_t *data_set) { GList *containers = NULL; GList *nodes = NULL; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return NULL); get_bundle_variant_data(bundle_data, rsc); pe__set_resource_flags(rsc, pe_rsc_allocating); containers = get_container_list(rsc); pe__show_node_weights(!pcmk_is_set(data_set->flags, pe_flag_show_scores), rsc, __func__, rsc->allowed_nodes, data_set); nodes = g_hash_table_get_values(rsc->allowed_nodes); nodes = sort_nodes_by_weight(nodes, NULL, data_set); containers = g_list_sort_with_data(containers, sort_clone_instance, data_set); distribute_children(rsc, containers, nodes, bundle_data->nreplicas, bundle_data->nreplicas_per_host, data_set); g_list_free(nodes); g_list_free(containers); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; pe_node_t *container_host = NULL; CRM_ASSERT(replica); if (replica->ip) { pe_rsc_trace(rsc, "Allocating bundle %s IP %s", rsc->id, replica->ip->id); replica->ip->cmds->allocate(replica->ip, prefer, data_set); } container_host = replica->container->allocated_to; if (replica->remote && pe__is_guest_or_remote_node(container_host)) { /* We need 'nested' connection resources to be on the same * host because pacemaker-remoted only supports a single * active connection */ pcmk__new_colocation("child-remote-with-docker-remote", NULL, INFINITY, replica->remote, container_host->details->remote_rsc, NULL, NULL, true, data_set); } if (replica->remote) { pe_rsc_trace(rsc, "Allocating bundle %s connection %s", rsc->id, replica->remote->id); replica->remote->cmds->allocate(replica->remote, prefer, data_set); } // Explicitly allocate replicas' children before bundle child if (replica->child) { pe_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, replica->child->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { if (node->details != replica->node->details) { node->weight = -INFINITY; - } else if (!migration_threshold_reached(replica->child, node, - data_set)) { + } else if (!pcmk__threshold_reached(replica->child, node, + data_set, NULL)) { node->weight = INFINITY; } } pe__set_resource_flags(replica->child->parent, pe_rsc_allocating); pe_rsc_trace(rsc, "Allocating bundle %s replica child %s", rsc->id, replica->child->id); replica->child->cmds->allocate(replica->child, replica->node, data_set); pe__clear_resource_flags(replica->child->parent, pe_rsc_allocating); } } if (bundle_data->child) { pe_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, bundle_data->child->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { if (is_bundle_node(bundle_data, node)) { node->weight = 0; } else { node->weight = -INFINITY; } } pe_rsc_trace(rsc, "Allocating bundle %s child %s", rsc->id, bundle_data->child->id); bundle_data->child->cmds->allocate(bundle_data->child, prefer, data_set); } pe__clear_resource_flags(rsc, pe_rsc_allocating|pe_rsc_provisional); return NULL; } void pcmk__bundle_create_actions(pe_resource_t *rsc, pe_working_set_t *data_set) { pe_action_t *action = NULL; GList *containers = NULL; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); containers = get_container_list(rsc); get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->ip) { replica->ip->cmds->create_actions(replica->ip, data_set); } if (replica->container) { replica->container->cmds->create_actions(replica->container, data_set); } if (replica->remote) { replica->remote->cmds->create_actions(replica->remote, data_set); } } clone_create_pseudo_actions(rsc, containers, NULL, NULL, data_set); if (bundle_data->child) { bundle_data->child->cmds->create_actions(bundle_data->child, data_set); if (pcmk_is_set(bundle_data->child->flags, pe_rsc_promotable)) { /* promote */ create_pseudo_resource_op(rsc, RSC_PROMOTE, TRUE, TRUE, data_set); action = create_pseudo_resource_op(rsc, RSC_PROMOTED, TRUE, TRUE, data_set); action->priority = INFINITY; /* demote */ create_pseudo_resource_op(rsc, RSC_DEMOTE, TRUE, TRUE, data_set); action = create_pseudo_resource_op(rsc, RSC_DEMOTED, TRUE, TRUE, data_set); action->priority = INFINITY; } } g_list_free(containers); } void pcmk__bundle_internal_constraints(pe_resource_t *rsc, pe_working_set_t *data_set) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); if (bundle_data->child) { new_rsc_order(rsc, RSC_START, bundle_data->child, RSC_START, pe_order_implies_first_printed, data_set); new_rsc_order(rsc, RSC_STOP, bundle_data->child, RSC_STOP, pe_order_implies_first_printed, data_set); if (bundle_data->child->children) { new_rsc_order(bundle_data->child, RSC_STARTED, rsc, RSC_STARTED, pe_order_implies_then_printed, data_set); new_rsc_order(bundle_data->child, RSC_STOPPED, rsc, RSC_STOPPED, pe_order_implies_then_printed, data_set); } else { new_rsc_order(bundle_data->child, RSC_START, rsc, RSC_STARTED, pe_order_implies_then_printed, data_set); new_rsc_order(bundle_data->child, RSC_STOP, rsc, RSC_STOPPED, pe_order_implies_then_printed, data_set); } } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); CRM_ASSERT(replica->container); replica->container->cmds->internal_constraints(replica->container, data_set); order_start_start(rsc, replica->container, pe_order_runnable_left|pe_order_implies_first_printed); if (replica->child) { order_stop_stop(rsc, replica->child, pe_order_implies_first_printed); } order_stop_stop(rsc, replica->container, pe_order_implies_first_printed); new_rsc_order(replica->container, RSC_START, rsc, RSC_STARTED, pe_order_implies_then_printed, data_set); new_rsc_order(replica->container, RSC_STOP, rsc, RSC_STOPPED, pe_order_implies_then_printed, data_set); if (replica->ip) { replica->ip->cmds->internal_constraints(replica->ip, data_set); // Start ip then container new_rsc_order(replica->ip, RSC_START, replica->container, RSC_START, pe_order_runnable_left|pe_order_preserve, data_set); new_rsc_order(replica->container, RSC_STOP, replica->ip, RSC_STOP, pe_order_implies_first|pe_order_preserve, data_set); pcmk__new_colocation("ip-with-docker", NULL, INFINITY, replica->ip, replica->container, NULL, NULL, true, data_set); } if (replica->remote) { /* This handles ordering and colocating remote relative to container * (via "resource-with-container"). Since IP is also ordered and * colocated relative to the container, we don't need to do anything * explicit here with IP. */ replica->remote->cmds->internal_constraints(replica->remote, data_set); } if (replica->child) { CRM_ASSERT(replica->remote); // "Start remote then child" is implicit in scheduler's remote logic } } if (bundle_data->child) { bundle_data->child->cmds->internal_constraints(bundle_data->child, data_set); if (pcmk_is_set(bundle_data->child->flags, pe_rsc_promotable)) { promote_demote_constraints(rsc, data_set); /* child demoted before global demoted */ new_rsc_order(bundle_data->child, RSC_DEMOTED, rsc, RSC_DEMOTED, pe_order_implies_then_printed, data_set); /* global demote before child demote */ new_rsc_order(rsc, RSC_DEMOTE, bundle_data->child, RSC_DEMOTE, pe_order_implies_first_printed, data_set); /* child promoted before global promoted */ new_rsc_order(bundle_data->child, RSC_PROMOTED, rsc, RSC_PROMOTED, pe_order_implies_then_printed, data_set); /* global promote before child promote */ new_rsc_order(rsc, RSC_PROMOTE, bundle_data->child, RSC_PROMOTE, pe_order_implies_first_printed, data_set); } } else { // int type = pe_order_optional | pe_order_implies_then | pe_order_restart; // custom_action_order(rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, // rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, pe_order_optional, data_set); } } static pe_resource_t * compatible_replica_for_node(pe_resource_t *rsc_lh, pe_node_t *candidate, pe_resource_t *rsc, enum rsc_role_e filter, gboolean current) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(candidate != NULL, return NULL); get_bundle_variant_data(bundle_data, rsc); crm_trace("Looking for compatible child from %s for %s on %s", rsc_lh->id, rsc->id, candidate->details->uname); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (is_child_compatible(replica->container, candidate, filter, current)) { crm_trace("Pairing %s with %s on %s", rsc_lh->id, replica->container->id, candidate->details->uname); return replica->container; } } crm_trace("Can't pair %s with %s", rsc_lh->id, rsc->id); return NULL; } static pe_resource_t * compatible_replica(pe_resource_t *rsc_lh, pe_resource_t *rsc, enum rsc_role_e filter, gboolean current, pe_working_set_t *data_set) { GList *scratch = NULL; pe_resource_t *pair = NULL; pe_node_t *active_node_lh = NULL; active_node_lh = rsc_lh->fns->location(rsc_lh, NULL, current); if (active_node_lh) { return compatible_replica_for_node(rsc_lh, active_node_lh, rsc, filter, current); } scratch = g_hash_table_get_values(rsc_lh->allowed_nodes); scratch = sort_nodes_by_weight(scratch, NULL, data_set); for (GList *gIter = scratch; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; pair = compatible_replica_for_node(rsc_lh, node, rsc, filter, current); if (pair) { goto done; } } pe_rsc_debug(rsc, "Can't pair %s with %s", rsc_lh->id, (rsc? rsc->id : "none")); done: g_list_free(scratch); return pair; } void pcmk__bundle_rsc_colocation_lh(pe_resource_t *rsc, pe_resource_t *rsc_rh, pcmk__colocation_t *constraint, pe_working_set_t *data_set) { /* -- Never called -- * * Instead we add the colocation constraints to the child and call from there */ CRM_ASSERT(FALSE); } int copies_per_node(pe_resource_t * rsc) { /* Strictly speaking, there should be a 'copies_per_node' addition * to the resource function table and each case would be a * function. However that would be serious overkill to return an * int. In fact, it seems to me that both function tables * could/should be replaced by resources.{c,h} full of * rsc_{some_operation} functions containing a switch as below * which calls out to functions named {variant}_{some_operation} * as needed. */ switch(rsc->variant) { case pe_unknown: return 0; case pe_native: case pe_group: return 1; case pe_clone: { const char *max_clones_node = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INCARNATION_NODEMAX); if (max_clones_node == NULL) { return 1; } else { int max_i; pcmk__scan_min_int(max_clones_node, &max_i, 0); return max_i; } } case pe_container: { pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, rsc); return data->nreplicas_per_host; } } return 0; } void pcmk__bundle_rsc_colocation_rh(pe_resource_t *rsc_lh, pe_resource_t *rsc, pcmk__colocation_t *constraint, pe_working_set_t *data_set) { GList *allocated_rhs = NULL; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(constraint != NULL, return); CRM_CHECK(rsc_lh != NULL, pe_err("rsc_lh was NULL for %s", constraint->id); return); CRM_CHECK(rsc != NULL, pe_err("rsc was NULL for %s", constraint->id); return); CRM_ASSERT(rsc_lh->variant == pe_native); if (pcmk_is_set(rsc->flags, pe_rsc_provisional)) { pe_rsc_trace(rsc, "%s is still provisional", rsc->id); return; } else if(constraint->rsc_lh->variant > pe_group) { pe_resource_t *rh_child = compatible_replica(rsc_lh, rsc, RSC_ROLE_UNKNOWN, FALSE, data_set); if (rh_child) { pe_rsc_debug(rsc, "Pairing %s with %s", rsc_lh->id, rh_child->id); rsc_lh->cmds->rsc_colocation_lh(rsc_lh, rh_child, constraint, data_set); } else if (constraint->score >= INFINITY) { crm_notice("Cannot pair %s with instance of %s", rsc_lh->id, rsc->id); assign_node(rsc_lh, NULL, TRUE); } else { pe_rsc_debug(rsc, "Cannot pair %s with instance of %s", rsc_lh->id, rsc->id); } return; } get_bundle_variant_data(bundle_data, rsc); pe_rsc_trace(rsc, "Processing constraint %s: %s -> %s %d", constraint->id, rsc_lh->id, rsc->id, constraint->score); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (constraint->score < INFINITY) { replica->container->cmds->rsc_colocation_rh(rsc_lh, replica->container, constraint, data_set); } else { pe_node_t *chosen = replica->container->fns->location(replica->container, NULL, FALSE); if ((chosen == NULL) || is_set_recursive(replica->container, pe_rsc_block, TRUE)) { continue; } if ((constraint->role_rh >= RSC_ROLE_PROMOTED) && (replica->child == NULL)) { continue; } if ((constraint->role_rh >= RSC_ROLE_PROMOTED) && (replica->child->next_role < RSC_ROLE_PROMOTED)) { continue; } pe_rsc_trace(rsc, "Allowing %s: %s %d", constraint->id, chosen->details->uname, chosen->weight); allocated_rhs = g_list_prepend(allocated_rhs, chosen); } } if (constraint->score >= INFINITY) { node_list_exclude(rsc_lh->allowed_nodes, allocated_rhs, FALSE); } g_list_free(allocated_rhs); } enum pe_action_flags pcmk__bundle_action_flags(pe_action_t *action, pe_node_t *node) { GList *containers = NULL; enum pe_action_flags flags = 0; pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, action->rsc); if(data->child) { enum action_tasks task = get_complex_task(data->child, action->task, TRUE); switch(task) { case no_action: case action_notify: case action_notified: case action_promote: case action_promoted: case action_demote: case action_demoted: return summary_action_flags(action, data->child->children, node); default: break; } } containers = get_container_list(action->rsc); flags = summary_action_flags(action, containers, node); g_list_free(containers); return flags; } pe_resource_t * find_compatible_child_by_node(pe_resource_t * local_child, pe_node_t * local_node, pe_resource_t * rsc, enum rsc_role_e filter, gboolean current) { GList *gIter = NULL; GList *children = NULL; if (local_node == NULL) { crm_err("Can't colocate unrunnable child %s with %s", local_child->id, rsc->id); return NULL; } crm_trace("Looking for compatible child from %s for %s on %s", local_child->id, rsc->id, local_node->details->uname); children = get_containers_or_children(rsc); for (gIter = children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; if(is_child_compatible(child_rsc, local_node, filter, current)) { crm_trace("Pairing %s with %s on %s", local_child->id, child_rsc->id, local_node->details->uname); return child_rsc; } } crm_trace("Can't pair %s with %s", local_child->id, rsc->id); if(children != rsc->children) { g_list_free(children); } return NULL; } static pe__bundle_replica_t * replica_for_container(pe_resource_t *rsc, pe_resource_t *container, pe_node_t *node) { if (rsc->variant == pe_container) { pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, rsc); for (GList *gIter = data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (replica->child && (container == replica->container) && (node->details == replica->node->details)) { return replica; } } } return NULL; } static enum pe_graph_flags multi_update_interleave_actions(pe_action_t *first, pe_action_t *then, pe_node_t *node, enum pe_action_flags flags, enum pe_action_flags filter, enum pe_ordering type, pe_working_set_t *data_set) { GList *gIter = NULL; GList *children = NULL; gboolean current = FALSE; enum pe_graph_flags changed = pe_graph_none; /* Fix this - lazy */ if (pcmk__ends_with(first->uuid, "_stopped_0") || pcmk__ends_with(first->uuid, "_demoted_0")) { current = TRUE; } children = get_containers_or_children(then->rsc); for (gIter = children; gIter != NULL; gIter = gIter->next) { pe_resource_t *then_child = gIter->data; pe_resource_t *first_child = find_compatible_child(then_child, first->rsc, RSC_ROLE_UNKNOWN, current, data_set); if (first_child == NULL && current) { crm_trace("Ignore"); } else if (first_child == NULL) { crm_debug("No match found for %s (%d / %s / %s)", then_child->id, current, first->uuid, then->uuid); /* Me no like this hack - but what else can we do? * * If there is no-one active or about to be active * on the same node as then_child, then they must * not be allowed to start */ if (type & (pe_order_runnable_left | pe_order_implies_then) /* Mandatory */ ) { pe_rsc_info(then->rsc, "Inhibiting %s from being active", then_child->id); if(assign_node(then_child, NULL, TRUE)) { pe__set_graph_flags(changed, first, pe_graph_updated_then); } } } else { pe_action_t *first_action = NULL; pe_action_t *then_action = NULL; enum action_tasks task = clone_child_action(first); const char *first_task = task2text(task); pe__bundle_replica_t *first_replica = NULL; pe__bundle_replica_t *then_replica = NULL; first_replica = replica_for_container(first->rsc, first_child, node); if (strstr(first->task, "stop") && first_replica && first_replica->child) { /* Except for 'stopped' we should be looking at the * in-container resource, actions for the child will * happen later and are therefor more likely to align * with the user's intent. */ first_action = find_first_action(first_replica->child->actions, NULL, task2text(task), node); } else { first_action = find_first_action(first_child->actions, NULL, task2text(task), node); } then_replica = replica_for_container(then->rsc, then_child, node); if (strstr(then->task, "mote") && then_replica && then_replica->child) { /* Promote/demote actions will never be found for the * container resource, look in the child instead * * Alternatively treat: * 'XXXX then promote YYYY' as 'XXXX then start container for YYYY', and * 'demote XXXX then stop YYYY' as 'stop container for XXXX then stop YYYY' */ then_action = find_first_action(then_replica->child->actions, NULL, then->task, node); } else { then_action = find_first_action(then_child->actions, NULL, then->task, node); } if (first_action == NULL) { if (!pcmk_is_set(first_child->flags, pe_rsc_orphan) && !pcmk__str_any_of(first_task, RSC_STOP, RSC_DEMOTE, NULL)) { crm_err("Internal error: No action found for %s in %s (first)", first_task, first_child->id); } else { crm_trace("No action found for %s in %s%s (first)", first_task, first_child->id, pcmk_is_set(first_child->flags, pe_rsc_orphan)? " (ORPHAN)" : ""); } continue; } /* We're only interested if 'then' is neither stopping nor being demoted */ if (then_action == NULL) { if (!pcmk_is_set(then_child->flags, pe_rsc_orphan) && !pcmk__str_any_of(then->task, RSC_STOP, RSC_DEMOTE, NULL)) { crm_err("Internal error: No action found for %s in %s (then)", then->task, then_child->id); } else { crm_trace("No action found for %s in %s%s (then)", then->task, then_child->id, pcmk_is_set(then_child->flags, pe_rsc_orphan)? " (ORPHAN)" : ""); } continue; } if (order_actions(first_action, then_action, type)) { crm_debug("Created constraint for %s (%d) -> %s (%d) %.6x", first_action->uuid, pcmk_is_set(first_action->flags, pe_action_optional), then_action->uuid, pcmk_is_set(then_action->flags, pe_action_optional), type); pe__set_graph_flags(changed, first, pe_graph_updated_first|pe_graph_updated_then); } if(first_action && then_action) { changed |= then_child->cmds->update_actions(first_action, then_action, node, first_child->cmds->action_flags(first_action, node), filter, type, data_set); } else { crm_err("Nothing found either for %s (%p) or %s (%p) %s", first_child->id, first_action, then_child->id, then_action, task2text(task)); } } } if(children != then->rsc->children) { g_list_free(children); } return changed; } static bool can_interleave_actions(pe_action_t *first, pe_action_t *then) { bool interleave = FALSE; pe_resource_t *rsc = NULL; const char *interleave_s = NULL; if(first->rsc == NULL || then->rsc == NULL) { crm_trace("Not interleaving %s with %s (both must be resources)", first->uuid, then->uuid); return FALSE; } else if(first->rsc == then->rsc) { crm_trace("Not interleaving %s with %s (must belong to different resources)", first->uuid, then->uuid); return FALSE; } else if(first->rsc->variant < pe_clone || then->rsc->variant < pe_clone) { crm_trace("Not interleaving %s with %s (both sides must be clones or bundles)", first->uuid, then->uuid); return FALSE; } if (pcmk__ends_with(then->uuid, "_stop_0") || pcmk__ends_with(then->uuid, "_demote_0")) { rsc = first->rsc; } else { rsc = then->rsc; } interleave_s = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERLEAVE); interleave = crm_is_true(interleave_s); crm_trace("Interleave %s -> %s: %s (based on %s)", first->uuid, then->uuid, interleave ? "yes" : "no", rsc->id); return interleave; } enum pe_graph_flags pcmk__multi_update_actions(pe_action_t *first, pe_action_t *then, pe_node_t *node, enum pe_action_flags flags, enum pe_action_flags filter, enum pe_ordering type, pe_working_set_t *data_set) { enum pe_graph_flags changed = pe_graph_none; crm_trace("%s -> %s", first->uuid, then->uuid); if(can_interleave_actions(first, then)) { changed = multi_update_interleave_actions(first, then, node, flags, filter, type, data_set); } else if(then->rsc) { GList *gIter = NULL; GList *children = NULL; // Handle the 'primitive' ordering case changed |= native_update_actions(first, then, node, flags, filter, type, data_set); // Now any children (or containers in the case of a bundle) children = get_containers_or_children(then->rsc); for (gIter = children; gIter != NULL; gIter = gIter->next) { pe_resource_t *then_child = (pe_resource_t *) gIter->data; enum pe_graph_flags then_child_changed = pe_graph_none; pe_action_t *then_child_action = find_first_action(then_child->actions, NULL, then->task, node); if (then_child_action) { enum pe_action_flags then_child_flags = then_child->cmds->action_flags(then_child_action, node); if (pcmk_is_set(then_child_flags, pe_action_runnable)) { then_child_changed |= then_child->cmds->update_actions(first, then_child_action, node, flags, filter, type, data_set); } changed |= then_child_changed; if (then_child_changed & pe_graph_updated_then) { for (GList *lpc = then_child_action->actions_after; lpc != NULL; lpc = lpc->next) { pe_action_wrapper_t *next = (pe_action_wrapper_t *) lpc->data; update_action(next->action, data_set); } } } } if(children != then->rsc->children) { g_list_free(children); } } return changed; } void pcmk__bundle_rsc_location(pe_resource_t *rsc, pe__location_t *constraint) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); native_rsc_location(rsc, constraint); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (replica->container) { replica->container->cmds->rsc_location(replica->container, constraint); } if (replica->ip) { replica->ip->cmds->rsc_location(replica->ip, constraint); } } if (bundle_data->child && ((constraint->role_filter == RSC_ROLE_UNPROMOTED) || (constraint->role_filter == RSC_ROLE_PROMOTED))) { bundle_data->child->cmds->rsc_location(bundle_data->child, constraint); bundle_data->child->rsc_location = g_list_prepend(bundle_data->child->rsc_location, constraint); } } void pcmk__bundle_expand(pe_resource_t *rsc, pe_working_set_t * data_set) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); if (bundle_data->child) { bundle_data->child->cmds->expand(bundle_data->child, data_set); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->remote && replica->container && pe__bundle_needs_remote_name(replica->remote, data_set)) { /* REMOTE_CONTAINER_HACK: Allow remote nodes to run containers that * run pacemaker-remoted inside, without needing a separate IP for * the container. This is done by configuring the inner remote's * connection host as the magic string "#uname", then * replacing it with the underlying host when needed. */ xmlNode *nvpair = get_xpath_object("//nvpair[@name='" XML_RSC_ATTR_REMOTE_RA_ADDR "']", replica->remote->xml, LOG_ERR); const char *calculated_addr = NULL; // Replace the value in replica->remote->xml (if appropriate) calculated_addr = pe__add_bundle_remote_name(replica->remote, data_set, nvpair, "value"); if (calculated_addr) { /* Since this is for the bundle as a resource, and not any * particular action, replace the value in the default * parameters (not evaluated for node). action2xml() will grab * it from there to replace it in node-evaluated parameters. */ GHashTable *params = pe_rsc_params(replica->remote, NULL, data_set); crm_trace("Set address for bundle connection %s to bundle host %s", replica->remote->id, calculated_addr); g_hash_table_replace(params, strdup(XML_RSC_ATTR_REMOTE_RA_ADDR), strdup(calculated_addr)); } else { /* The only way to get here is if the remote connection is * neither currently running nor scheduled to run. That means we * won't be doing any operations that require addr (only start * requires it; we additionally use it to compare digests when * unpacking status, promote, and migrate_from history, but * that's already happened by this point). */ crm_info("Unable to determine address for bundle %s remote connection", rsc->id); } } if (replica->ip) { replica->ip->cmds->expand(replica->ip, data_set); } if (replica->container) { replica->container->cmds->expand(replica->container, data_set); } if (replica->remote) { replica->remote->cmds->expand(replica->remote, data_set); } } } gboolean pcmk__bundle_create_probe(pe_resource_t *rsc, pe_node_t *node, pe_action_t *complete, gboolean force, pe_working_set_t * data_set) { bool any_created = FALSE; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return FALSE); get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->ip) { any_created |= replica->ip->cmds->create_probe(replica->ip, node, complete, force, data_set); } if (replica->child && (node->details == replica->node->details)) { any_created |= replica->child->cmds->create_probe(replica->child, node, complete, force, data_set); } if (replica->container) { bool created = replica->container->cmds->create_probe(replica->container, node, complete, force, data_set); if(created) { any_created = TRUE; /* If we're limited to one replica per host (due to * the lack of an IP range probably), then we don't * want any of our peer containers starting until * we've established that no other copies are already * running. * * Partly this is to ensure that nreplicas_per_host is * observed, but also to ensure that the containers * don't fail to start because the necessary port * mappings (which won't include an IP for uniqueness) * are already taken */ for (GList *tIter = bundle_data->replicas; tIter && (bundle_data->nreplicas_per_host == 1); tIter = tIter->next) { pe__bundle_replica_t *other = tIter->data; if ((other != replica) && (other != NULL) && (other->container != NULL)) { custom_action_order(replica->container, pcmk__op_key(replica->container->id, RSC_STATUS, 0), NULL, other->container, pcmk__op_key(other->container->id, RSC_START, 0), NULL, pe_order_optional|pe_order_same_node, data_set); } } } } if (replica->container && replica->remote && replica->remote->cmds->create_probe(replica->remote, node, complete, force, data_set)) { /* Do not probe the remote resource until we know where the * container is running. This is required for REMOTE_CONTAINER_HACK * to correctly probe remote resources. */ char *probe_uuid = pcmk__op_key(replica->remote->id, RSC_STATUS, 0); pe_action_t *probe = find_first_action(replica->remote->actions, probe_uuid, NULL, node); free(probe_uuid); if (probe) { any_created = TRUE; crm_trace("Ordering %s probe on %s", replica->remote->id, node->details->uname); custom_action_order(replica->container, pcmk__op_key(replica->container->id, RSC_START, 0), NULL, replica->remote, NULL, probe, pe_order_probe, data_set); } } } return any_created; } void pcmk__bundle_append_meta(pe_resource_t *rsc, xmlNode *xml) { } void pcmk__bundle_log_actions(pe_resource_t *rsc, pe_working_set_t *data_set) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->ip) { LogActions(replica->ip, data_set); } if (replica->container) { LogActions(replica->container, data_set); } if (replica->remote) { LogActions(replica->remote, data_set); } if (replica->child) { LogActions(replica->child, data_set); } } } diff --git a/lib/pacemaker/pcmk_sched_native.c b/lib/pacemaker/pcmk_sched_native.c index 8014977ac7..5121f2d0ff 100644 --- a/lib/pacemaker/pcmk_sched_native.c +++ b/lib/pacemaker/pcmk_sched_native.c @@ -1,3210 +1,3210 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include // The controller removes the resource from the CIB, making this redundant // #define DELETE_THEN_REFRESH 1 #define INFINITY_HACK (INFINITY * -100) #define VARIANT_NATIVE 1 #include extern bool pcmk__is_daemon; static void Recurring(pe_resource_t *rsc, pe_action_t *start, pe_node_t *node, pe_working_set_t *data_set); static void RecurringOp(pe_resource_t *rsc, pe_action_t *start, pe_node_t *node, xmlNode *operation, pe_working_set_t *data_set); static void Recurring_Stopped(pe_resource_t *rsc, pe_action_t *start, pe_node_t *node, pe_working_set_t *data_set); static void RecurringOp_Stopped(pe_resource_t *rsc, pe_action_t *start, pe_node_t *node, xmlNode *operation, pe_working_set_t *data_set); void ReloadRsc(pe_resource_t * rsc, pe_node_t *node, pe_working_set_t * data_set); gboolean DeleteRsc(pe_resource_t * rsc, pe_node_t * node, gboolean optional, pe_working_set_t * data_set); gboolean StopRsc(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set); gboolean StartRsc(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set); gboolean DemoteRsc(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set); gboolean PromoteRsc(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set); gboolean RoleError(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set); gboolean NullOp(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set); /* This array says what the *next* role should be when transitioning from one * role to another. For example going from Stopped to Promoted, the next role is * RSC_ROLE_UNPROMOTED, because the resource must be started before being promoted. * The current state then becomes Started, which is fed into this array again, * giving a next role of RSC_ROLE_PROMOTED. */ static enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = { /* Current state Next state*/ /* Unknown Stopped Started Unpromoted Promoted */ /* Unknown */ { RSC_ROLE_UNKNOWN, RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_STOPPED }, /* Stopped */ { RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_STARTED, RSC_ROLE_UNPROMOTED, RSC_ROLE_UNPROMOTED }, /* Started */ { RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_STARTED, RSC_ROLE_UNPROMOTED, RSC_ROLE_PROMOTED }, /* Unpromoted */ { RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_UNPROMOTED, RSC_ROLE_PROMOTED }, /* Promoted */ { RSC_ROLE_STOPPED, RSC_ROLE_UNPROMOTED, RSC_ROLE_UNPROMOTED, RSC_ROLE_UNPROMOTED, RSC_ROLE_PROMOTED }, }; typedef gboolean (*rsc_transition_fn)(pe_resource_t *rsc, pe_node_t *next, gboolean optional, pe_working_set_t *data_set); // This array picks the function needed to transition from one role to another static rsc_transition_fn rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = { /* Current state Next state */ /* Unknown Stopped Started Unpromoted Promoted */ /* Unknown */ { RoleError, StopRsc, RoleError, RoleError, RoleError, }, /* Stopped */ { RoleError, NullOp, StartRsc, StartRsc, RoleError, }, /* Started */ { RoleError, StopRsc, NullOp, NullOp, PromoteRsc, }, /* Unpromoted */ { RoleError, StopRsc, StopRsc, NullOp, PromoteRsc, }, /* Promoted */ { RoleError, DemoteRsc, DemoteRsc, DemoteRsc, NullOp, }, }; #define clear_node_weights_flags(nw_flags, nw_rsc, flags_to_clear) do { \ flags = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Node weight", (nw_rsc)->id, (flags), \ (flags_to_clear), #flags_to_clear); \ } while (0) static gboolean native_choose_node(pe_resource_t * rsc, pe_node_t * prefer, pe_working_set_t * data_set) { GList *nodes = NULL; pe_node_t *chosen = NULL; pe_node_t *best = NULL; int multiple = 1; int length = 0; gboolean result = FALSE; process_utilization(rsc, &prefer, data_set); if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return rsc->allocated_to ? TRUE : FALSE; } // Sort allowed nodes by weight if (rsc->allowed_nodes) { length = g_hash_table_size(rsc->allowed_nodes); } if (length > 0) { nodes = g_hash_table_get_values(rsc->allowed_nodes); nodes = sort_nodes_by_weight(nodes, pe__current_node(rsc), data_set); // First node in sorted list has the best score best = g_list_nth_data(nodes, 0); } if (prefer && nodes) { chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id); if (chosen == NULL) { pe_rsc_trace(rsc, "Preferred node %s for %s was unknown", prefer->details->uname, rsc->id); /* Favor the preferred node as long as its weight is at least as good as * the best allowed node's. * * An alternative would be to favor the preferred node even if the best * node is better, when the best node's weight is less than INFINITY. */ } else if ((chosen->weight < 0) || (chosen->weight < best->weight)) { pe_rsc_trace(rsc, "Preferred node %s for %s was unsuitable", chosen->details->uname, rsc->id); chosen = NULL; } else if (!can_run_resources(chosen)) { pe_rsc_trace(rsc, "Preferred node %s for %s was unavailable", chosen->details->uname, rsc->id); chosen = NULL; } else { pe_rsc_trace(rsc, "Chose preferred node %s for %s (ignoring %d candidates)", chosen->details->uname, rsc->id, length); } } if ((chosen == NULL) && nodes) { /* Either there is no preferred node, or the preferred node is not * available, but there are other nodes allowed to run the resource. */ chosen = best; pe_rsc_trace(rsc, "Chose node %s for %s from %d candidates", chosen ? chosen->details->uname : "", rsc->id, length); if (!pe_rsc_is_unique_clone(rsc->parent) && chosen && (chosen->weight > 0) && can_run_resources(chosen)) { /* If the resource is already running on a node, prefer that node if * it is just as good as the chosen node. * * We don't do this for unique clone instances, because * distribute_children() has already assigned instances to their * running nodes when appropriate, and if we get here, we don't want * remaining unallocated instances to prefer a node that's already * running another instance. */ pe_node_t *running = pe__current_node(rsc); if (running && (can_run_resources(running) == FALSE)) { pe_rsc_trace(rsc, "Current node for %s (%s) can't run resources", rsc->id, running->details->uname); } else if (running) { for (GList *iter = nodes->next; iter; iter = iter->next) { pe_node_t *tmp = (pe_node_t *) iter->data; if (tmp->weight != chosen->weight) { // The nodes are sorted by weight, so no more are equal break; } if (tmp->details == running->details) { // Scores are equal, so prefer the current node chosen = tmp; } multiple++; } } } } if (multiple > 1) { static char score[33]; int log_level = (chosen->weight >= INFINITY)? LOG_WARNING : LOG_INFO; score2char_stack(chosen->weight, score, sizeof(score)); do_crm_log(log_level, "Chose node %s for %s from %d nodes with score %s", chosen->details->uname, rsc->id, multiple, score); } result = native_assign_node(rsc, chosen, FALSE); g_list_free(nodes); return result; } /*! * \internal * \brief Find score of highest-scored node that matches colocation attribute * * \param[in] rsc Resource whose allowed nodes should be searched * \param[in] attr Colocation attribute name (must not be NULL) * \param[in] value Colocation attribute value to require */ static int best_node_score_matching_attr(const pe_resource_t *rsc, const char *attr, const char *value) { GHashTableIter iter; pe_node_t *node = NULL; int best_score = -INFINITY; const char *best_node = NULL; // Find best allowed node with matching attribute g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if ((node->weight > best_score) && can_run_resources(node) && pcmk__str_eq(value, pe_node_attribute_raw(node, attr), pcmk__str_casei)) { best_score = node->weight; best_node = node->details->uname; } } if (!pcmk__str_eq(attr, CRM_ATTR_UNAME, pcmk__str_casei)) { if (best_node == NULL) { crm_info("No allowed node for %s matches node attribute %s=%s", rsc->id, attr, value); } else { crm_info("Allowed node %s for %s had best score (%d) " "of those matching node attribute %s=%s", best_node, rsc->id, best_score, attr, value); } } return best_score; } /*! * \internal * \brief Add resource's colocation matches to current node allocation scores * * For each node in a given table, if any of a given resource's allowed nodes * have a matching value for the colocation attribute, add the highest of those * nodes' scores to the node's score. * * \param[in,out] nodes Hash table of nodes with allocation scores so far * \param[in] rsc Resource whose allowed nodes should be compared * \param[in] attr Colocation attribute that must match (NULL for default) * \param[in] factor Factor by which to multiply scores being added * \param[in] only_positive Whether to add only positive scores */ static void add_node_scores_matching_attr(GHashTable *nodes, const pe_resource_t *rsc, const char *attr, float factor, bool only_positive) { GHashTableIter iter; pe_node_t *node = NULL; if (attr == NULL) { attr = CRM_ATTR_UNAME; } // Iterate through each node g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { float weight_f = 0; int weight = 0; int score = 0; int new_score = 0; score = best_node_score_matching_attr(rsc, attr, pe_node_attribute_raw(node, attr)); if ((factor < 0) && (score < 0)) { /* Negative preference for a node with a negative score * should not become a positive preference. * * @TODO Consider filtering only if weight is -INFINITY */ crm_trace("%s: Filtering %d + %f * %d (double negative disallowed)", node->details->uname, node->weight, factor, score); continue; } if (node->weight == INFINITY_HACK) { crm_trace("%s: Filtering %d + %f * %d (node was marked unusable)", node->details->uname, node->weight, factor, score); continue; } weight_f = factor * score; // Round the number; see http://c-faq.com/fp/round.html weight = (int) ((weight_f < 0)? (weight_f - 0.5) : (weight_f + 0.5)); /* Small factors can obliterate the small scores that are often actually * used in configurations. If the score and factor are nonzero, ensure * that the result is nonzero as well. */ if ((weight == 0) && (score != 0)) { if (factor > 0.0) { weight = 1; } else if (factor < 0.0) { weight = -1; } } new_score = pe__add_scores(weight, node->weight); if (only_positive && (new_score < 0) && (node->weight > 0)) { crm_trace("%s: Filtering %d + %f * %d = %d " "(negative disallowed, marking node unusable)", node->details->uname, node->weight, factor, score, new_score); node->weight = INFINITY_HACK; continue; } if (only_positive && (new_score < 0) && (node->weight == 0)) { crm_trace("%s: Filtering %d + %f * %d = %d (negative disallowed)", node->details->uname, node->weight, factor, score, new_score); continue; } crm_trace("%s: %d + %f * %d = %d", node->details->uname, node->weight, factor, score, new_score); node->weight = new_score; } } static inline bool is_nonempty_group(pe_resource_t *rsc) { return rsc && (rsc->variant == pe_group) && (rsc->children != NULL); } /*! * \internal * \brief Incorporate colocation constraint scores into node weights * * \param[in,out] rsc Resource being placed * \param[in] rhs ID of 'with' resource * \param[in,out] nodes Nodes, with scores as of this point * \param[in] attr Colocation attribute (ID by default) * \param[in] factor Incorporate scores multiplied by this factor * \param[in] flags Bitmask of enum pe_weights values * * \return Nodes, with scores modified by this constraint * \note This function assumes ownership of the nodes argument. The caller * should free the returned copy rather than the original. */ GHashTable * pcmk__native_merge_weights(pe_resource_t *rsc, const char *rhs, GHashTable *nodes, const char *attr, float factor, uint32_t flags) { GHashTable *work = NULL; // Avoid infinite recursion if (pcmk_is_set(rsc->flags, pe_rsc_merging)) { pe_rsc_info(rsc, "%s: Breaking dependency loop at %s", rhs, rsc->id); return nodes; } pe__set_resource_flags(rsc, pe_rsc_merging); if (pcmk_is_set(flags, pe_weights_init)) { if (is_nonempty_group(rsc)) { GList *last = g_list_last(rsc->children); pe_resource_t *last_rsc = last->data; pe_rsc_trace(rsc, "%s: Merging scores from group %s " "using last member %s (at %.6f)", rhs, rsc->id, last_rsc->id, factor); work = pcmk__native_merge_weights(last_rsc, rhs, NULL, attr, factor, flags); } else { work = pcmk__copy_node_table(rsc->allowed_nodes); } clear_node_weights_flags(flags, rsc, pe_weights_init); } else if (is_nonempty_group(rsc)) { /* The first member of the group will recursively incorporate any * constraints involving other members (including the group internal * colocation). * * @TODO The indirect colocations from the dependent group's other * members will be incorporated at full strength rather than by * factor, so the group's combined stickiness will be treated as * (factor + (#members - 1)) * stickiness. It is questionable what * the right approach should be. */ pe_rsc_trace(rsc, "%s: Merging scores from first member of group %s " "(at %.6f)", rhs, rsc->id, factor); work = pcmk__copy_node_table(nodes); work = pcmk__native_merge_weights(rsc->children->data, rhs, work, attr, factor, flags); } else { pe_rsc_trace(rsc, "%s: Merging scores from %s (at %.6f)", rhs, rsc->id, factor); work = pcmk__copy_node_table(nodes); add_node_scores_matching_attr(work, rsc, attr, factor, pcmk_is_set(flags, pe_weights_positive)); } if (can_run_any(work)) { GList *gIter = NULL; int multiplier = (factor < 0)? -1 : 1; if (pcmk_is_set(flags, pe_weights_forward)) { gIter = rsc->rsc_cons; pe_rsc_trace(rsc, "Checking additional %d optional '%s with' constraints", g_list_length(gIter), rsc->id); } else if (is_nonempty_group(rsc)) { pe_resource_t *last_rsc = g_list_last(rsc->children)->data; gIter = last_rsc->rsc_cons_lhs; pe_rsc_trace(rsc, "Checking additional %d optional 'with group %s' " "constraints using last member %s", g_list_length(gIter), rsc->id, last_rsc->id); } else { gIter = rsc->rsc_cons_lhs; pe_rsc_trace(rsc, "Checking additional %d optional 'with %s' constraints", g_list_length(gIter), rsc->id); } for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *other = NULL; pcmk__colocation_t *constraint = (pcmk__colocation_t *) gIter->data; if (pcmk_is_set(flags, pe_weights_forward)) { other = constraint->rsc_rh; } else if (!pcmk__colocation_has_influence(constraint, NULL)) { continue; } else { other = constraint->rsc_lh; } pe_rsc_trace(rsc, "Optionally merging score of '%s' constraint (%s with %s)", constraint->id, constraint->rsc_lh->id, constraint->rsc_rh->id); work = pcmk__native_merge_weights(other, rhs, work, constraint->node_attribute, multiplier * constraint->score / (float) INFINITY, flags|pe_weights_rollback); pe__show_node_weights(true, NULL, rhs, work, rsc->cluster); } } else if (pcmk_is_set(flags, pe_weights_rollback)) { pe_rsc_info(rsc, "%s: Rolling back optional scores from %s", rhs, rsc->id); g_hash_table_destroy(work); pe__clear_resource_flags(rsc, pe_rsc_merging); return nodes; } if (pcmk_is_set(flags, pe_weights_positive)) { pe_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, work); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if (node->weight == INFINITY_HACK) { node->weight = 1; } } } if (nodes) { g_hash_table_destroy(nodes); } pe__clear_resource_flags(rsc, pe_rsc_merging); return work; } static inline bool node_has_been_unfenced(pe_node_t *node) { const char *unfenced = pe_node_attribute_raw(node, CRM_ATTR_UNFENCED); return !pcmk__str_eq(unfenced, "0", pcmk__str_null_matches); } static inline bool is_unfence_device(pe_resource_t *rsc, pe_working_set_t *data_set) { return pcmk_is_set(rsc->flags, pe_rsc_fence_device) && pcmk_is_set(data_set->flags, pe_flag_enable_unfencing); } pe_node_t * pcmk__native_allocate(pe_resource_t *rsc, pe_node_t *prefer, pe_working_set_t *data_set) { GList *gIter = NULL; if (rsc->parent && !pcmk_is_set(rsc->parent->flags, pe_rsc_allocating)) { /* never allocate children on their own */ pe_rsc_debug(rsc, "Escalating allocation of %s to its parent: %s", rsc->id, rsc->parent->id); rsc->parent->cmds->allocate(rsc->parent, prefer, data_set); } if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return rsc->allocated_to; } if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { pe_rsc_debug(rsc, "Dependency loop detected involving %s", rsc->id); return NULL; } pe__set_resource_flags(rsc, pe_rsc_allocating); pe__show_node_weights(true, rsc, "Pre-alloc", rsc->allowed_nodes, data_set); for (gIter = rsc->rsc_cons; gIter != NULL; gIter = gIter->next) { pcmk__colocation_t *constraint = (pcmk__colocation_t *) gIter->data; GHashTable *archive = NULL; pe_resource_t *rsc_rh = constraint->rsc_rh; if ((constraint->role_lh >= RSC_ROLE_PROMOTED) || (constraint->score < 0 && constraint->score > -INFINITY)) { archive = pcmk__copy_node_table(rsc->allowed_nodes); } pe_rsc_trace(rsc, "%s: Allocating %s first (constraint=%s score=%d role=%s)", rsc->id, rsc_rh->id, constraint->id, constraint->score, role2text(constraint->role_lh)); rsc_rh->cmds->allocate(rsc_rh, NULL, data_set); rsc->cmds->rsc_colocation_lh(rsc, rsc_rh, constraint, data_set); if (archive && can_run_any(rsc->allowed_nodes) == FALSE) { pe_rsc_info(rsc, "%s: Rolling back scores from %s", rsc->id, rsc_rh->id); g_hash_table_destroy(rsc->allowed_nodes); rsc->allowed_nodes = archive; archive = NULL; } if (archive) { g_hash_table_destroy(archive); } } pe__show_node_weights(true, rsc, "Post-coloc", rsc->allowed_nodes, data_set); for (gIter = rsc->rsc_cons_lhs; gIter != NULL; gIter = gIter->next) { pcmk__colocation_t *constraint = (pcmk__colocation_t *) gIter->data; if (!pcmk__colocation_has_influence(constraint, NULL)) { continue; } pe_rsc_trace(rsc, "Merging score of '%s' constraint (%s with %s)", constraint->id, constraint->rsc_lh->id, constraint->rsc_rh->id); rsc->allowed_nodes = constraint->rsc_lh->cmds->merge_weights(constraint->rsc_lh, rsc->id, rsc->allowed_nodes, constraint->node_attribute, (float)constraint->score / INFINITY, pe_weights_rollback); } if (rsc->next_role == RSC_ROLE_STOPPED) { pe_rsc_trace(rsc, "Making sure %s doesn't get allocated", rsc->id); /* make sure it doesn't come up again */ resource_location(rsc, NULL, -INFINITY, XML_RSC_ATTR_TARGET_ROLE, data_set); } else if(rsc->next_role > rsc->role && !pcmk_is_set(data_set->flags, pe_flag_have_quorum) && data_set->no_quorum_policy == no_quorum_freeze) { crm_notice("Resource %s cannot be elevated from %s to %s: no-quorum-policy=freeze", rsc->id, role2text(rsc->role), role2text(rsc->next_role)); pe__set_next_role(rsc, rsc->role, "no-quorum-policy=freeze"); } pe__show_node_weights(!pcmk_is_set(data_set->flags, pe_flag_show_scores), rsc, __func__, rsc->allowed_nodes, data_set); if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled) && !pcmk_is_set(data_set->flags, pe_flag_have_stonith_resource)) { pe__clear_resource_flags(rsc, pe_rsc_managed); } if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { const char *reason = NULL; pe_node_t *assign_to = NULL; pe__set_next_role(rsc, rsc->role, "unmanaged"); assign_to = pe__current_node(rsc); if (assign_to == NULL) { reason = "inactive"; } else if (rsc->role == RSC_ROLE_PROMOTED) { reason = "promoted"; } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { reason = "failed"; } else { reason = "active"; } pe_rsc_info(rsc, "Unmanaged resource %s allocated to %s: %s", rsc->id, (assign_to? assign_to->details->uname : "no node"), reason); native_assign_node(rsc, assign_to, TRUE); } else if (pcmk_is_set(data_set->flags, pe_flag_stop_everything)) { pe_rsc_debug(rsc, "Forcing %s to stop", rsc->id); native_assign_node(rsc, NULL, TRUE); } else if (pcmk_is_set(rsc->flags, pe_rsc_provisional) && native_choose_node(rsc, prefer, data_set)) { pe_rsc_trace(rsc, "Allocated resource %s to %s", rsc->id, rsc->allocated_to->details->uname); } else if (rsc->allocated_to == NULL) { if (!pcmk_is_set(rsc->flags, pe_rsc_orphan)) { pe_rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id); } else if (rsc->running_on != NULL) { pe_rsc_info(rsc, "Stopping orphan resource %s", rsc->id); } } else { pe_rsc_debug(rsc, "Pre-Allocated resource %s to %s", rsc->id, rsc->allocated_to->details->uname); } pe__clear_resource_flags(rsc, pe_rsc_allocating); if (rsc->is_remote_node) { pe_node_t *remote_node = pe_find_node(data_set->nodes, rsc->id); CRM_ASSERT(remote_node != NULL); if (rsc->allocated_to && rsc->next_role != RSC_ROLE_STOPPED) { crm_trace("Setting Pacemaker Remote node %s to ONLINE", remote_node->details->id); remote_node->details->online = TRUE; /* We shouldn't consider an unseen remote-node unclean if we are going * to try and connect to it. Otherwise we get an unnecessary fence */ if (remote_node->details->unseen == TRUE) { remote_node->details->unclean = FALSE; } } else { crm_trace("Setting Pacemaker Remote node %s to SHUTDOWN (next role %s, %sallocated)", remote_node->details->id, role2text(rsc->next_role), (rsc->allocated_to? "" : "un")); remote_node->details->shutdown = TRUE; } } return rsc->allocated_to; } static gboolean is_op_dup(pe_resource_t *rsc, const char *name, guint interval_ms) { gboolean dup = FALSE; const char *id = NULL; const char *value = NULL; xmlNode *operation = NULL; guint interval2_ms = 0; CRM_ASSERT(rsc); for (operation = pcmk__xe_first_child(rsc->ops_xml); operation != NULL; operation = pcmk__xe_next(operation)) { if (pcmk__str_eq((const char *)operation->name, "op", pcmk__str_none)) { value = crm_element_value(operation, "name"); if (!pcmk__str_eq(value, name, pcmk__str_casei)) { continue; } value = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); interval2_ms = crm_parse_interval_spec(value); if (interval_ms != interval2_ms) { continue; } if (id == NULL) { id = ID(operation); } else { pcmk__config_err("Operation %s is duplicate of %s (do not use " "same name and interval combination more " "than once per resource)", ID(operation), id); dup = TRUE; } } } return dup; } static bool op_cannot_recur(const char *name) { return pcmk__strcase_any_of(name, RSC_STOP, RSC_START, RSC_DEMOTE, RSC_PROMOTE, NULL); } static void RecurringOp(pe_resource_t * rsc, pe_action_t * start, pe_node_t * node, xmlNode * operation, pe_working_set_t * data_set) { char *key = NULL; const char *name = NULL; const char *role = NULL; const char *interval_spec = NULL; const char *node_uname = node? node->details->uname : "n/a"; guint interval_ms = 0; pe_action_t *mon = NULL; gboolean is_optional = TRUE; GList *possible_matches = NULL; CRM_ASSERT(rsc); /* Only process for the operations without role="Stopped" */ role = crm_element_value(operation, "role"); if (role && text2role(role) == RSC_ROLE_STOPPED) { return; } interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); interval_ms = crm_parse_interval_spec(interval_spec); if (interval_ms == 0) { return; } name = crm_element_value(operation, "name"); if (is_op_dup(rsc, name, interval_ms)) { crm_trace("Not creating duplicate recurring action %s for %dms %s", ID(operation), interval_ms, name); return; } if (op_cannot_recur(name)) { pcmk__config_err("Ignoring %s because action '%s' cannot be recurring", ID(operation), name); return; } key = pcmk__op_key(rsc->id, name, interval_ms); if (find_rsc_op_entry(rsc, key) == NULL) { crm_trace("Not creating recurring action %s for disabled resource %s", ID(operation), rsc->id); free(key); return; } pe_rsc_trace(rsc, "Creating recurring action %s for %s in role %s on %s", ID(operation), rsc->id, role2text(rsc->next_role), node_uname); if (start != NULL) { pe_rsc_trace(rsc, "Marking %s %s due to %s", key, pcmk_is_set(start->flags, pe_action_optional)? "optional" : "mandatory", start->uuid); is_optional = (rsc->cmds->action_flags(start, NULL) & pe_action_optional); } else { pe_rsc_trace(rsc, "Marking %s optional", key); is_optional = TRUE; } /* start a monitor for an already active resource */ possible_matches = find_actions_exact(rsc->actions, key, node); if (possible_matches == NULL) { is_optional = FALSE; pe_rsc_trace(rsc, "Marking %s mandatory: not active", key); } else { GList *gIter = NULL; for (gIter = possible_matches; gIter != NULL; gIter = gIter->next) { pe_action_t *op = (pe_action_t *) gIter->data; if (pcmk_is_set(op->flags, pe_action_reschedule)) { is_optional = FALSE; break; } } g_list_free(possible_matches); } if (((rsc->next_role == RSC_ROLE_PROMOTED) && (role == NULL)) || (role != NULL && text2role(role) != rsc->next_role)) { int log_level = LOG_TRACE; const char *result = "Ignoring"; if (is_optional) { char *after_key = NULL; pe_action_t *cancel_op = NULL; // It's running, so cancel it log_level = LOG_INFO; result = "Cancelling"; cancel_op = pe_cancel_op(rsc, name, interval_ms, node, data_set); switch (rsc->role) { case RSC_ROLE_UNPROMOTED: case RSC_ROLE_STARTED: if (rsc->next_role == RSC_ROLE_PROMOTED) { after_key = promote_key(rsc); } else if (rsc->next_role == RSC_ROLE_STOPPED) { after_key = stop_key(rsc); } break; case RSC_ROLE_PROMOTED: after_key = demote_key(rsc); break; default: break; } if (after_key) { custom_action_order(rsc, NULL, cancel_op, rsc, after_key, NULL, pe_order_runnable_left, data_set); } } do_crm_log(log_level, "%s action %s (%s vs. %s)", result, key, role ? role : role2text(RSC_ROLE_UNPROMOTED), role2text(rsc->next_role)); free(key); return; } mon = custom_action(rsc, key, name, node, is_optional, TRUE, data_set); key = mon->uuid; if (is_optional) { pe_rsc_trace(rsc, "%s\t %s (optional)", node_uname, mon->uuid); } if ((start == NULL) || !pcmk_is_set(start->flags, pe_action_runnable)) { pe_rsc_debug(rsc, "%s\t %s (cancelled : start un-runnable)", node_uname, mon->uuid); pe__clear_action_flags(mon, pe_action_runnable); } else if (node == NULL || node->details->online == FALSE || node->details->unclean) { pe_rsc_debug(rsc, "%s\t %s (cancelled : no node available)", node_uname, mon->uuid); pe__clear_action_flags(mon, pe_action_runnable); } else if (!pcmk_is_set(mon->flags, pe_action_optional)) { pe_rsc_info(rsc, " Start recurring %s (%us) for %s on %s", mon->task, interval_ms / 1000, rsc->id, node_uname); } if (rsc->next_role == RSC_ROLE_PROMOTED) { char *running_promoted = pcmk__itoa(PCMK_OCF_RUNNING_PROMOTED); add_hash_param(mon->meta, XML_ATTR_TE_TARGET_RC, running_promoted); free(running_promoted); } if ((node == NULL) || pcmk_is_set(rsc->flags, pe_rsc_managed)) { custom_action_order(rsc, start_key(rsc), NULL, NULL, strdup(key), mon, pe_order_implies_then | pe_order_runnable_left, data_set); custom_action_order(rsc, reload_key(rsc), NULL, NULL, strdup(key), mon, pe_order_implies_then | pe_order_runnable_left, data_set); if (rsc->next_role == RSC_ROLE_PROMOTED) { custom_action_order(rsc, promote_key(rsc), NULL, rsc, NULL, mon, pe_order_optional | pe_order_runnable_left, data_set); } else if (rsc->role == RSC_ROLE_PROMOTED) { custom_action_order(rsc, demote_key(rsc), NULL, rsc, NULL, mon, pe_order_optional | pe_order_runnable_left, data_set); } } } static void Recurring(pe_resource_t * rsc, pe_action_t * start, pe_node_t * node, pe_working_set_t * data_set) { if (!pcmk_is_set(rsc->flags, pe_rsc_maintenance) && (node == NULL || node->details->maintenance == FALSE)) { xmlNode *operation = NULL; for (operation = pcmk__xe_first_child(rsc->ops_xml); operation != NULL; operation = pcmk__xe_next(operation)) { if (pcmk__str_eq((const char *)operation->name, "op", pcmk__str_none)) { RecurringOp(rsc, start, node, operation, data_set); } } } } static void RecurringOp_Stopped(pe_resource_t * rsc, pe_action_t * start, pe_node_t * node, xmlNode * operation, pe_working_set_t * data_set) { char *key = NULL; const char *name = NULL; const char *role = NULL; const char *interval_spec = NULL; const char *node_uname = node? node->details->uname : "n/a"; guint interval_ms = 0; GList *possible_matches = NULL; GList *gIter = NULL; /* Only process for the operations with role="Stopped" */ role = crm_element_value(operation, "role"); if (role == NULL || text2role(role) != RSC_ROLE_STOPPED) { return; } interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); interval_ms = crm_parse_interval_spec(interval_spec); if (interval_ms == 0) { return; } name = crm_element_value(operation, "name"); if (is_op_dup(rsc, name, interval_ms)) { crm_trace("Not creating duplicate recurring action %s for %dms %s", ID(operation), interval_ms, name); return; } if (op_cannot_recur(name)) { pcmk__config_err("Ignoring %s because action '%s' cannot be recurring", ID(operation), name); return; } key = pcmk__op_key(rsc->id, name, interval_ms); if (find_rsc_op_entry(rsc, key) == NULL) { crm_trace("Not creating recurring action %s for disabled resource %s", ID(operation), rsc->id); free(key); return; } // @TODO add support if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) { crm_notice("Ignoring %s (recurring monitors for Stopped role are " "not supported for anonymous clones)", ID(operation)); return; } pe_rsc_trace(rsc, "Creating recurring action %s for %s in role %s on nodes where it should not be running", ID(operation), rsc->id, role2text(rsc->next_role)); /* if the monitor exists on the node where the resource will be running, cancel it */ if (node != NULL) { possible_matches = find_actions_exact(rsc->actions, key, node); if (possible_matches) { pe_action_t *cancel_op = NULL; g_list_free(possible_matches); cancel_op = pe_cancel_op(rsc, name, interval_ms, node, data_set); if ((rsc->next_role == RSC_ROLE_STARTED) || (rsc->next_role == RSC_ROLE_UNPROMOTED)) { /* rsc->role == RSC_ROLE_STOPPED: cancel the monitor before start */ /* rsc->role == RSC_ROLE_STARTED: for a migration, cancel the monitor on the target node before start */ custom_action_order(rsc, NULL, cancel_op, rsc, start_key(rsc), NULL, pe_order_runnable_left, data_set); } pe_rsc_info(rsc, "Cancel action %s (%s vs. %s) on %s", key, role, role2text(rsc->next_role), node_uname); } } for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *stop_node = (pe_node_t *) gIter->data; const char *stop_node_uname = stop_node->details->uname; gboolean is_optional = TRUE; gboolean probe_is_optional = TRUE; gboolean stop_is_optional = TRUE; pe_action_t *stopped_mon = NULL; char *rc_inactive = NULL; GList *probe_complete_ops = NULL; GList *stop_ops = NULL; GList *local_gIter = NULL; if (node && pcmk__str_eq(stop_node_uname, node_uname, pcmk__str_casei)) { continue; } pe_rsc_trace(rsc, "Creating recurring action %s for %s on %s", ID(operation), rsc->id, crm_str(stop_node_uname)); /* start a monitor for an already stopped resource */ possible_matches = find_actions_exact(rsc->actions, key, stop_node); if (possible_matches == NULL) { pe_rsc_trace(rsc, "Marking %s mandatory on %s: not active", key, crm_str(stop_node_uname)); is_optional = FALSE; } else { pe_rsc_trace(rsc, "Marking %s optional on %s: already active", key, crm_str(stop_node_uname)); is_optional = TRUE; g_list_free(possible_matches); } stopped_mon = custom_action(rsc, strdup(key), name, stop_node, is_optional, TRUE, data_set); rc_inactive = pcmk__itoa(PCMK_OCF_NOT_RUNNING); add_hash_param(stopped_mon->meta, XML_ATTR_TE_TARGET_RC, rc_inactive); free(rc_inactive); if (pcmk_is_set(rsc->flags, pe_rsc_managed)) { GList *probes = pe__resource_actions(rsc, stop_node, RSC_STATUS, FALSE); GList *pIter = NULL; for (pIter = probes; pIter != NULL; pIter = pIter->next) { pe_action_t *probe = (pe_action_t *) pIter->data; order_actions(probe, stopped_mon, pe_order_runnable_left); crm_trace("%s then %s on %s", probe->uuid, stopped_mon->uuid, stop_node->details->uname); } g_list_free(probes); } if (probe_complete_ops) { g_list_free(probe_complete_ops); } stop_ops = pe__resource_actions(rsc, stop_node, RSC_STOP, TRUE); for (local_gIter = stop_ops; local_gIter != NULL; local_gIter = local_gIter->next) { pe_action_t *stop = (pe_action_t *) local_gIter->data; if (!pcmk_is_set(stop->flags, pe_action_optional)) { stop_is_optional = FALSE; } if (!pcmk_is_set(stop->flags, pe_action_runnable)) { crm_debug("%s\t %s (cancelled : stop un-runnable)", crm_str(stop_node_uname), stopped_mon->uuid); pe__clear_action_flags(stopped_mon, pe_action_runnable); } if (pcmk_is_set(rsc->flags, pe_rsc_managed)) { custom_action_order(rsc, stop_key(rsc), stop, NULL, strdup(key), stopped_mon, pe_order_implies_then | pe_order_runnable_left, data_set); } } if (stop_ops) { g_list_free(stop_ops); } if (is_optional == FALSE && probe_is_optional && stop_is_optional && !pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe_rsc_trace(rsc, "Marking %s optional on %s due to unmanaged", key, crm_str(stop_node_uname)); pe__set_action_flags(stopped_mon, pe_action_optional); } if (pcmk_is_set(stopped_mon->flags, pe_action_optional)) { pe_rsc_trace(rsc, "%s\t %s (optional)", crm_str(stop_node_uname), stopped_mon->uuid); } if (stop_node->details->online == FALSE || stop_node->details->unclean) { pe_rsc_debug(rsc, "%s\t %s (cancelled : no node available)", crm_str(stop_node_uname), stopped_mon->uuid); pe__clear_action_flags(stopped_mon, pe_action_runnable); } if (pcmk_is_set(stopped_mon->flags, pe_action_runnable) && !pcmk_is_set(stopped_mon->flags, pe_action_optional)) { crm_notice(" Start recurring %s (%us) for %s on %s", stopped_mon->task, interval_ms / 1000, rsc->id, crm_str(stop_node_uname)); } } free(key); } static void Recurring_Stopped(pe_resource_t * rsc, pe_action_t * start, pe_node_t * node, pe_working_set_t * data_set) { if (!pcmk_is_set(rsc->flags, pe_rsc_maintenance) && (node == NULL || node->details->maintenance == FALSE)) { xmlNode *operation = NULL; for (operation = pcmk__xe_first_child(rsc->ops_xml); operation != NULL; operation = pcmk__xe_next(operation)) { if (pcmk__str_eq((const char *)operation->name, "op", pcmk__str_none)) { RecurringOp_Stopped(rsc, start, node, operation, data_set); } } } } static void handle_migration_actions(pe_resource_t * rsc, pe_node_t *current, pe_node_t *chosen, pe_working_set_t * data_set) { pe_action_t *migrate_to = NULL; pe_action_t *migrate_from = NULL; pe_action_t *start = NULL; pe_action_t *stop = NULL; gboolean partial = rsc->partial_migration_target ? TRUE : FALSE; pe_rsc_trace(rsc, "Processing migration actions %s moving from %s to %s . partial migration = %s", rsc->id, current->details->id, chosen->details->id, partial ? "TRUE" : "FALSE"); start = start_action(rsc, chosen, TRUE); stop = stop_action(rsc, current, TRUE); if (partial == FALSE) { migrate_to = custom_action(rsc, pcmk__op_key(rsc->id, RSC_MIGRATE, 0), RSC_MIGRATE, current, TRUE, TRUE, data_set); } migrate_from = custom_action(rsc, pcmk__op_key(rsc->id, RSC_MIGRATED, 0), RSC_MIGRATED, chosen, TRUE, TRUE, data_set); if ((migrate_to && migrate_from) || (migrate_from && partial)) { pe__set_action_flags(start, pe_action_migrate_runnable); pe__set_action_flags(stop, pe_action_migrate_runnable); // This is easier than trying to delete it from the graph pe__set_action_flags(start, pe_action_pseudo); /* order probes before migrations */ if (partial) { pe__set_action_flags(migrate_from, pe_action_migrate_runnable); migrate_from->needs = start->needs; custom_action_order(rsc, pcmk__op_key(rsc->id, RSC_STATUS, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_MIGRATED, 0), NULL, pe_order_optional, data_set); } else { pe__set_action_flags(migrate_from, pe_action_migrate_runnable); pe__set_action_flags(migrate_to, pe_action_migrate_runnable); migrate_to->needs = start->needs; custom_action_order(rsc, pcmk__op_key(rsc->id, RSC_STATUS, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_MIGRATE, 0), NULL, pe_order_optional, data_set); custom_action_order(rsc, pcmk__op_key(rsc->id, RSC_MIGRATE, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_MIGRATED, 0), NULL, pe_order_optional|pe_order_implies_first_migratable, data_set); } custom_action_order(rsc, pcmk__op_key(rsc->id, RSC_MIGRATED, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, pe_order_optional|pe_order_implies_first_migratable, data_set); custom_action_order(rsc, pcmk__op_key(rsc->id, RSC_MIGRATED, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, pe_order_optional|pe_order_implies_first_migratable|pe_order_pseudo_left, data_set); } if (migrate_to) { add_hash_param(migrate_to->meta, XML_LRM_ATTR_MIGRATE_SOURCE, current->details->uname); add_hash_param(migrate_to->meta, XML_LRM_ATTR_MIGRATE_TARGET, chosen->details->uname); /* Pacemaker Remote connections don't require pending to be recorded in * the CIB. We can reduce CIB writes by not setting PENDING for them. */ if (rsc->is_remote_node == FALSE) { /* migrate_to takes place on the source node, but can * have an effect on the target node depending on how * the agent is written. Because of this, we have to maintain * a record that the migrate_to occurred, in case the source node * loses membership while the migrate_to action is still in-flight. */ add_hash_param(migrate_to->meta, XML_OP_ATTR_PENDING, "true"); } } if (migrate_from) { add_hash_param(migrate_from->meta, XML_LRM_ATTR_MIGRATE_SOURCE, current->details->uname); add_hash_param(migrate_from->meta, XML_LRM_ATTR_MIGRATE_TARGET, chosen->details->uname); } } void native_create_actions(pe_resource_t * rsc, pe_working_set_t * data_set) { pe_action_t *start = NULL; pe_node_t *chosen = NULL; pe_node_t *current = NULL; gboolean need_stop = FALSE; bool need_promote = FALSE; gboolean is_moving = FALSE; gboolean allow_migrate = pcmk_is_set(rsc->flags, pe_rsc_allow_migrate)? TRUE : FALSE; GList *gIter = NULL; unsigned int num_all_active = 0; unsigned int num_clean_active = 0; bool multiply_active = FALSE; enum rsc_role_e role = RSC_ROLE_UNKNOWN; enum rsc_role_e next_role = RSC_ROLE_UNKNOWN; CRM_ASSERT(rsc); chosen = rsc->allocated_to; next_role = rsc->next_role; if (next_role == RSC_ROLE_UNKNOWN) { pe__set_next_role(rsc, (chosen == NULL)? RSC_ROLE_STOPPED : RSC_ROLE_STARTED, "allocation"); } pe_rsc_trace(rsc, "Creating all actions for %s transition from %s to %s (%s) on %s", rsc->id, role2text(rsc->role), role2text(rsc->next_role), ((next_role == RSC_ROLE_UNKNOWN)? "implicit" : "explicit"), ((chosen == NULL)? "no node" : chosen->details->uname)); current = pe__find_active_on(rsc, &num_all_active, &num_clean_active); for (gIter = rsc->dangling_migrations; gIter != NULL; gIter = gIter->next) { pe_node_t *dangling_source = (pe_node_t *) gIter->data; pe_action_t *stop = NULL; pe_rsc_trace(rsc, "Creating stop action %sfor %s on %s due to dangling migration", pcmk_is_set(data_set->flags, pe_flag_remove_after_stop)? "and cleanup " : "", rsc->id, dangling_source->details->uname); stop = stop_action(rsc, dangling_source, FALSE); pe__set_action_flags(stop, pe_action_dangle); if (pcmk_is_set(data_set->flags, pe_flag_remove_after_stop)) { DeleteRsc(rsc, dangling_source, FALSE, data_set); } } if ((num_all_active == 2) && (num_clean_active == 2) && chosen && rsc->partial_migration_source && rsc->partial_migration_target && (current->details == rsc->partial_migration_source->details) && (chosen->details == rsc->partial_migration_target->details)) { /* The chosen node is still the migration target from a partial * migration. Attempt to continue the migration instead of recovering * by stopping the resource everywhere and starting it on a single node. */ pe_rsc_trace(rsc, "Will attempt to continue with partial migration " "to target %s from %s", rsc->partial_migration_target->details->id, rsc->partial_migration_source->details->id); } else if (!pcmk_is_set(rsc->flags, pe_rsc_needs_fencing)) { /* If a resource has "requires" set to nothing or quorum, don't consider * it active on unclean nodes (similar to how all resources behave when * stonith-enabled is false). We can start such resources elsewhere * before fencing completes, and if we considered the resource active on * the failed node, we would attempt recovery for being active on * multiple nodes. */ multiply_active = (num_clean_active > 1); } else { multiply_active = (num_all_active > 1); } if (multiply_active) { if (rsc->partial_migration_target && rsc->partial_migration_source) { // Migration was in progress, but we've chosen a different target crm_notice("Resource %s can no longer migrate from %s to %s " "(will stop on both nodes)", rsc->id, rsc->partial_migration_source->details->uname, rsc->partial_migration_target->details->uname); } else { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); - // Resource was incorrectly multiply active - pe_proc_err("%s resource %s is active on %u nodes (%s)", + // Resource was (possibly) incorrectly multiply active + pe_proc_err("%s resource %s might be active on %u nodes (%s)", crm_str(class), rsc->id, num_all_active, recovery2text(rsc->recovery_type)); crm_notice("See https://wiki.clusterlabs.org/wiki/FAQ#Resource_is_Too_Active for more information"); } if (rsc->recovery_type == recovery_stop_start) { need_stop = TRUE; } /* If by chance a partial migration is in process, but the migration * target is not chosen still, clear all partial migration data. */ rsc->partial_migration_source = rsc->partial_migration_target = NULL; allow_migrate = FALSE; } if (pcmk_is_set(rsc->flags, pe_rsc_start_pending)) { pe_rsc_trace(rsc, "Creating start action for %s to represent already pending start", rsc->id); start = start_action(rsc, chosen, TRUE); pe__set_action_flags(start, pe_action_print_always); } if (current && chosen && current->details != chosen->details) { pe_rsc_trace(rsc, "Moving %s from %s to %s", rsc->id, crm_str(current->details->uname), crm_str(chosen->details->uname)); is_moving = TRUE; need_stop = TRUE; } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { if (pcmk_is_set(rsc->flags, pe_rsc_stop)) { need_stop = TRUE; pe_rsc_trace(rsc, "Recovering %s", rsc->id); } else { pe_rsc_trace(rsc, "Recovering %s by demotion", rsc->id); if (rsc->next_role == RSC_ROLE_PROMOTED) { need_promote = TRUE; } } } else if (pcmk_is_set(rsc->flags, pe_rsc_block)) { pe_rsc_trace(rsc, "Blocking further actions on %s", rsc->id); need_stop = TRUE; } else if (rsc->role > RSC_ROLE_STARTED && current != NULL && chosen != NULL) { pe_rsc_trace(rsc, "Creating start action for promoted resource %s", rsc->id); start = start_action(rsc, chosen, TRUE); if (!pcmk_is_set(start->flags, pe_action_optional)) { // Recovery of a promoted resource pe_rsc_trace(rsc, "%s restart is required for recovery", rsc->id); need_stop = TRUE; } } /* Create any additional actions required when bringing resource down and * back up to same level. */ role = rsc->role; while (role != RSC_ROLE_STOPPED) { next_role = rsc_state_matrix[role][RSC_ROLE_STOPPED]; pe_rsc_trace(rsc, "Creating %s action to take %s down from %s to %s", (need_stop? "required" : "optional"), rsc->id, role2text(role), role2text(next_role)); if (rsc_action_matrix[role][next_role] (rsc, current, !need_stop, data_set) == FALSE) { break; } role = next_role; } while ((rsc->role <= rsc->next_role) && (role != rsc->role) && !pcmk_is_set(rsc->flags, pe_rsc_block)) { bool required = need_stop; next_role = rsc_state_matrix[role][rsc->role]; if ((next_role == RSC_ROLE_PROMOTED) && need_promote) { required = true; } pe_rsc_trace(rsc, "Creating %s action to take %s up from %s to %s", (required? "required" : "optional"), rsc->id, role2text(role), role2text(next_role)); if (rsc_action_matrix[role][next_role](rsc, chosen, !required, data_set) == FALSE) { break; } role = next_role; } role = rsc->role; /* Required steps from this role to the next */ while (role != rsc->next_role) { next_role = rsc_state_matrix[role][rsc->next_role]; pe_rsc_trace(rsc, "Creating action to take %s from %s to %s (ending at %s)", rsc->id, role2text(role), role2text(next_role), role2text(rsc->next_role)); if (rsc_action_matrix[role][next_role] (rsc, chosen, FALSE, data_set) == FALSE) { break; } role = next_role; } if (pcmk_is_set(rsc->flags, pe_rsc_block)) { pe_rsc_trace(rsc, "Not creating recurring monitors for blocked resource %s", rsc->id); } else if ((rsc->next_role != RSC_ROLE_STOPPED) || !pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe_rsc_trace(rsc, "Creating recurring monitors for %s resource %s", ((rsc->next_role == RSC_ROLE_STOPPED)? "unmanaged" : "active"), rsc->id); start = start_action(rsc, chosen, TRUE); Recurring(rsc, start, chosen, data_set); Recurring_Stopped(rsc, start, chosen, data_set); } else { pe_rsc_trace(rsc, "Creating recurring monitors for inactive resource %s", rsc->id); Recurring_Stopped(rsc, NULL, NULL, data_set); } /* if we are stuck in a partial migration, where the target * of the partial migration no longer matches the chosen target. * A full stop/start is required */ if (rsc->partial_migration_target && (chosen == NULL || rsc->partial_migration_target->details != chosen->details)) { pe_rsc_trace(rsc, "Not allowing partial migration of %s to continue", rsc->id); allow_migrate = FALSE; } else if (!is_moving || !pcmk_is_set(rsc->flags, pe_rsc_managed) || pcmk_any_flags_set(rsc->flags, pe_rsc_failed|pe_rsc_start_pending) || (current && current->details->unclean) || rsc->next_role < RSC_ROLE_STARTED) { allow_migrate = FALSE; } if (allow_migrate) { handle_migration_actions(rsc, current, chosen, data_set); } } static void rsc_avoids_remote_nodes(pe_resource_t *rsc) { GHashTableIter iter; pe_node_t *node = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if (node->details->remote_rsc) { node->weight = -INFINITY; } } } /*! * \internal * \brief Return allowed nodes as (possibly sorted) list * * Convert a resource's hash table of allowed nodes to a list. If printing to * stdout, sort the list, to keep action ID numbers consistent for regression * test output (while avoiding the performance hit on a live cluster). * * \param[in] rsc Resource to check for allowed nodes * \param[in] data_set Cluster working set * * \return List of resource's allowed nodes * \note Callers should take care not to rely on the list being sorted. */ static GList * allowed_nodes_as_list(pe_resource_t *rsc, pe_working_set_t *data_set) { GList *allowed_nodes = NULL; if (rsc->allowed_nodes) { allowed_nodes = g_hash_table_get_values(rsc->allowed_nodes); } if (!pcmk__is_daemon) { allowed_nodes = g_list_sort(allowed_nodes, sort_node_uname); } return allowed_nodes; } void native_internal_constraints(pe_resource_t * rsc, pe_working_set_t * data_set) { /* This function is on the critical path and worth optimizing as much as possible */ pe_resource_t *top = NULL; GList *allowed_nodes = NULL; bool check_unfencing = FALSE; bool check_utilization = FALSE; if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe_rsc_trace(rsc, "Skipping native constraints for unmanaged resource: %s", rsc->id); return; } top = uber_parent(rsc); // Whether resource requires unfencing check_unfencing = !pcmk_is_set(rsc->flags, pe_rsc_fence_device) && pcmk_is_set(data_set->flags, pe_flag_enable_unfencing) && pcmk_is_set(rsc->flags, pe_rsc_needs_unfencing); // Whether a non-default placement strategy is used check_utilization = (g_hash_table_size(rsc->utilization) > 0) && !pcmk__str_eq(data_set->placement_strategy, "default", pcmk__str_casei); // Order stops before starts (i.e. restart) custom_action_order(rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, pe_order_optional|pe_order_implies_then|pe_order_restart, data_set); // Promotable ordering: demote before stop, start before promote if (pcmk_is_set(top->flags, pe_rsc_promotable) || (rsc->role > RSC_ROLE_UNPROMOTED)) { custom_action_order(rsc, pcmk__op_key(rsc->id, RSC_DEMOTE, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, pe_order_promoted_implies_first, data_set); custom_action_order(rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_PROMOTE, 0), NULL, pe_order_runnable_left, data_set); } // Don't clear resource history if probing on same node custom_action_order(rsc, pcmk__op_key(rsc->id, CRM_OP_LRM_DELETE, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_STATUS, 0), NULL, pe_order_same_node|pe_order_then_cancels_first, data_set); // Certain checks need allowed nodes if (check_unfencing || check_utilization || rsc->container) { allowed_nodes = allowed_nodes_as_list(rsc, data_set); } if (check_unfencing) { /* Check if the node needs to be unfenced first */ for (GList *item = allowed_nodes; item; item = item->next) { pe_node_t *node = item->data; pe_action_t *unfence = pe_fence_op(node, "on", TRUE, NULL, FALSE, data_set); crm_debug("Ordering any stops of %s before %s, and any starts after", rsc->id, unfence->uuid); /* * It would be more efficient to order clone resources once, * rather than order each instance, but ordering the instance * allows us to avoid unnecessary dependencies that might conflict * with user constraints. * * @TODO: This constraint can still produce a transition loop if the * resource has a stop scheduled on the node being unfenced, and * there is a user ordering constraint to start some other resource * (which will be ordered after the unfence) before stopping this * resource. An example is "start some slow-starting cloned service * before stopping an associated virtual IP that may be moving to * it": * stop this -> unfencing -> start that -> stop this */ custom_action_order(rsc, stop_key(rsc), NULL, NULL, strdup(unfence->uuid), unfence, pe_order_optional|pe_order_same_node, data_set); custom_action_order(NULL, strdup(unfence->uuid), unfence, rsc, start_key(rsc), NULL, pe_order_implies_then_on_node|pe_order_same_node, data_set); } } if (check_utilization) { GList *gIter = NULL; pe_rsc_trace(rsc, "Creating utilization constraints for %s - strategy: %s", rsc->id, data_set->placement_strategy); for (gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) { pe_node_t *current = (pe_node_t *) gIter->data; char *load_stopped_task = crm_strdup_printf(LOAD_STOPPED "_%s", current->details->uname); pe_action_t *load_stopped = get_pseudo_op(load_stopped_task, data_set); if (load_stopped->node == NULL) { load_stopped->node = pe__copy_node(current); pe__clear_action_flags(load_stopped, pe_action_optional); } custom_action_order(rsc, stop_key(rsc), NULL, NULL, load_stopped_task, load_stopped, pe_order_load, data_set); } for (GList *item = allowed_nodes; item; item = item->next) { pe_node_t *next = item->data; char *load_stopped_task = crm_strdup_printf(LOAD_STOPPED "_%s", next->details->uname); pe_action_t *load_stopped = get_pseudo_op(load_stopped_task, data_set); if (load_stopped->node == NULL) { load_stopped->node = pe__copy_node(next); pe__clear_action_flags(load_stopped, pe_action_optional); } custom_action_order(NULL, strdup(load_stopped_task), load_stopped, rsc, start_key(rsc), NULL, pe_order_load, data_set); custom_action_order(NULL, strdup(load_stopped_task), load_stopped, rsc, pcmk__op_key(rsc->id, RSC_MIGRATE, 0), NULL, pe_order_load, data_set); free(load_stopped_task); } } if (rsc->container) { pe_resource_t *remote_rsc = NULL; if (rsc->is_remote_node) { // rsc is the implicit remote connection for a guest or bundle node /* Do not allow a guest resource to live on a Pacemaker Remote node, * to avoid nesting remotes. However, allow bundles to run on remote * nodes. */ if (!pcmk_is_set(rsc->flags, pe_rsc_allow_remote_remotes)) { rsc_avoids_remote_nodes(rsc->container); } /* If someone cleans up a guest or bundle node's container, we will * likely schedule a (re-)probe of the container and recovery of the * connection. Order the connection stop after the container probe, * so that if we detect the container running, we will trigger a new * transition and avoid the unnecessary recovery. */ new_rsc_order(rsc->container, RSC_STATUS, rsc, RSC_STOP, pe_order_optional, data_set); /* A user can specify that a resource must start on a Pacemaker Remote * node by explicitly configuring it with the container=NODENAME * meta-attribute. This is of questionable merit, since location * constraints can accomplish the same thing. But we support it, so here * we check whether a resource (that is not itself a remote connection) * has container set to a remote node or guest node resource. */ } else if (rsc->container->is_remote_node) { remote_rsc = rsc->container; } else { remote_rsc = pe__resource_contains_guest_node(data_set, rsc->container); } if (remote_rsc) { /* Force the resource on the Pacemaker Remote node instead of * colocating the resource with the container resource. */ for (GList *item = allowed_nodes; item; item = item->next) { pe_node_t *node = item->data; if (node->details->remote_rsc != remote_rsc) { node->weight = -INFINITY; } } } else { /* This resource is either a filler for a container that does NOT * represent a Pacemaker Remote node, or a Pacemaker Remote * connection resource for a guest node or bundle. */ int score; crm_trace("Order and colocate %s relative to its container %s", rsc->id, rsc->container->id); custom_action_order(rsc->container, pcmk__op_key(rsc->container->id, RSC_START, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, pe_order_implies_then|pe_order_runnable_left, data_set); custom_action_order(rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, rsc->container, pcmk__op_key(rsc->container->id, RSC_STOP, 0), NULL, pe_order_implies_first, data_set); if (pcmk_is_set(rsc->flags, pe_rsc_allow_remote_remotes)) { score = 10000; /* Highly preferred but not essential */ } else { score = INFINITY; /* Force them to run on the same host */ } pcmk__new_colocation("resource-with-container", NULL, score, rsc, rsc->container, NULL, NULL, true, data_set); } } if (rsc->is_remote_node || pcmk_is_set(rsc->flags, pe_rsc_fence_device)) { /* don't allow remote nodes to run stonith devices * or remote connection resources.*/ rsc_avoids_remote_nodes(rsc); } g_list_free(allowed_nodes); } void native_rsc_colocation_lh(pe_resource_t *rsc_lh, pe_resource_t *rsc_rh, pcmk__colocation_t *constraint, pe_working_set_t *data_set) { if (rsc_lh == NULL) { pe_err("rsc_lh was NULL for %s", constraint->id); return; } else if (constraint->rsc_rh == NULL) { pe_err("rsc_rh was NULL for %s", constraint->id); return; } pe_rsc_trace(rsc_lh, "Processing colocation constraint between %s and %s", rsc_lh->id, rsc_rh->id); rsc_rh->cmds->rsc_colocation_rh(rsc_lh, rsc_rh, constraint, data_set); } enum filter_colocation_res filter_colocation_constraint(pe_resource_t * rsc_lh, pe_resource_t * rsc_rh, pcmk__colocation_t *constraint, gboolean preview) { /* rh side must be allocated before we can process constraint */ if (!preview && pcmk_is_set(rsc_rh->flags, pe_rsc_provisional)) { return influence_nothing; } if ((constraint->role_lh >= RSC_ROLE_UNPROMOTED) && rsc_lh->parent && pcmk_is_set(rsc_lh->parent->flags, pe_rsc_promotable) && !pcmk_is_set(rsc_lh->flags, pe_rsc_provisional)) { /* LH and RH resources have already been allocated, place the correct * priority on LH rsc for the given promotable clone resource role */ return influence_rsc_priority; } if (!preview && !pcmk_is_set(rsc_lh->flags, pe_rsc_provisional)) { // Log an error if we violated a mandatory colocation constraint const pe_node_t *rh_node = rsc_rh->allocated_to; if (rsc_lh->allocated_to == NULL) { // Dependent resource isn't allocated, so constraint doesn't matter return influence_nothing; } if (constraint->score >= INFINITY) { // Dependent resource must colocate with rh_node if ((rh_node == NULL) || (rh_node->details != rsc_lh->allocated_to->details)) { crm_err("%s must be colocated with %s but is not (%s vs. %s)", rsc_lh->id, rsc_rh->id, rsc_lh->allocated_to->details->uname, (rh_node? rh_node->details->uname : "unallocated")); } } else if (constraint->score <= -INFINITY) { // Dependent resource must anti-colocate with rh_node if ((rh_node != NULL) && (rsc_lh->allocated_to->details == rh_node->details)) { crm_err("%s and %s must be anti-colocated but are allocated " "to the same node (%s)", rsc_lh->id, rsc_rh->id, rh_node->details->uname); } } return influence_nothing; } if (constraint->score > 0 && constraint->role_lh != RSC_ROLE_UNKNOWN && constraint->role_lh != rsc_lh->next_role) { crm_trace("LH: Skipping constraint: \"%s\" state filter nextrole is %s", role2text(constraint->role_lh), role2text(rsc_lh->next_role)); return influence_nothing; } if (constraint->score > 0 && constraint->role_rh != RSC_ROLE_UNKNOWN && constraint->role_rh != rsc_rh->next_role) { crm_trace("RH: Skipping constraint: \"%s\" state filter", role2text(constraint->role_rh)); return influence_nothing; } if (constraint->score < 0 && constraint->role_lh != RSC_ROLE_UNKNOWN && constraint->role_lh == rsc_lh->next_role) { crm_trace("LH: Skipping negative constraint: \"%s\" state filter", role2text(constraint->role_lh)); return influence_nothing; } if (constraint->score < 0 && constraint->role_rh != RSC_ROLE_UNKNOWN && constraint->role_rh == rsc_rh->next_role) { crm_trace("RH: Skipping negative constraint: \"%s\" state filter", role2text(constraint->role_rh)); return influence_nothing; } return influence_rsc_location; } static void influence_priority(pe_resource_t *rsc_lh, pe_resource_t *rsc_rh, pcmk__colocation_t *constraint) { const char *rh_value = NULL; const char *lh_value = NULL; const char *attribute = CRM_ATTR_ID; int score_multiplier = 1; if (!rsc_rh->allocated_to || !rsc_lh->allocated_to) { return; } if (constraint->node_attribute != NULL) { attribute = constraint->node_attribute; } lh_value = pe_node_attribute_raw(rsc_lh->allocated_to, attribute); rh_value = pe_node_attribute_raw(rsc_rh->allocated_to, attribute); if (!pcmk__str_eq(lh_value, rh_value, pcmk__str_casei)) { if ((constraint->score == INFINITY) && (constraint->role_lh == RSC_ROLE_PROMOTED)) { rsc_lh->priority = -INFINITY; } return; } if (constraint->role_rh && (constraint->role_rh != rsc_rh->next_role)) { return; } if (constraint->role_lh == RSC_ROLE_UNPROMOTED) { score_multiplier = -1; } rsc_lh->priority = pe__add_scores(score_multiplier * constraint->score, rsc_lh->priority); } static void colocation_match(pe_resource_t *rsc_lh, pe_resource_t *rsc_rh, pcmk__colocation_t *constraint) { const char *attribute = CRM_ATTR_ID; const char *value = NULL; GHashTable *work = NULL; GHashTableIter iter; pe_node_t *node = NULL; if (constraint->node_attribute != NULL) { attribute = constraint->node_attribute; } if (rsc_rh->allocated_to) { value = pe_node_attribute_raw(rsc_rh->allocated_to, attribute); } else if (constraint->score < 0) { // Nothing to do (anti-colocation with something that is not running) return; } work = pcmk__copy_node_table(rsc_lh->allowed_nodes); g_hash_table_iter_init(&iter, work); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if (rsc_rh->allocated_to == NULL) { pe_rsc_trace(rsc_lh, "%s: %s@%s -= %d (%s inactive)", constraint->id, rsc_lh->id, node->details->uname, constraint->score, rsc_rh->id); node->weight = pe__add_scores(-constraint->score, node->weight); } else if (pcmk__str_eq(pe_node_attribute_raw(node, attribute), value, pcmk__str_casei)) { if (constraint->score < CRM_SCORE_INFINITY) { pe_rsc_trace(rsc_lh, "%s: %s@%s += %d", constraint->id, rsc_lh->id, node->details->uname, constraint->score); node->weight = pe__add_scores(constraint->score, node->weight); } } else if (constraint->score >= CRM_SCORE_INFINITY) { pe_rsc_trace(rsc_lh, "%s: %s@%s -= %d (%s mismatch)", constraint->id, rsc_lh->id, node->details->uname, constraint->score, attribute); node->weight = pe__add_scores(-constraint->score, node->weight); } } if (can_run_any(work) || constraint->score <= -INFINITY || constraint->score >= INFINITY) { g_hash_table_destroy(rsc_lh->allowed_nodes); rsc_lh->allowed_nodes = work; work = NULL; } else { pe_rsc_info(rsc_lh, "%s: Rolling back scores from %s (no available nodes)", rsc_lh->id, rsc_rh->id); } if (work) { g_hash_table_destroy(work); } } void native_rsc_colocation_rh(pe_resource_t *rsc_lh, pe_resource_t *rsc_rh, pcmk__colocation_t *constraint, pe_working_set_t *data_set) { enum filter_colocation_res filter_results; CRM_ASSERT(rsc_lh); CRM_ASSERT(rsc_rh); filter_results = filter_colocation_constraint(rsc_lh, rsc_rh, constraint, FALSE); pe_rsc_trace(rsc_lh, "%s %s with %s (%s, score=%d, filter=%d)", ((constraint->score > 0)? "Colocating" : "Anti-colocating"), rsc_lh->id, rsc_rh->id, constraint->id, constraint->score, filter_results); switch (filter_results) { case influence_rsc_priority: influence_priority(rsc_lh, rsc_rh, constraint); break; case influence_rsc_location: colocation_match(rsc_lh, rsc_rh, constraint); break; case influence_nothing: default: return; } } static gboolean filter_rsc_ticket(pe_resource_t * rsc_lh, rsc_ticket_t * rsc_ticket) { if (rsc_ticket->role_lh != RSC_ROLE_UNKNOWN && rsc_ticket->role_lh != rsc_lh->role) { pe_rsc_trace(rsc_lh, "LH: Skipping constraint: \"%s\" state filter", role2text(rsc_ticket->role_lh)); return FALSE; } return TRUE; } void rsc_ticket_constraint(pe_resource_t * rsc_lh, rsc_ticket_t * rsc_ticket, pe_working_set_t * data_set) { if (rsc_ticket == NULL) { pe_err("rsc_ticket was NULL"); return; } if (rsc_lh == NULL) { pe_err("rsc_lh was NULL for %s", rsc_ticket->id); return; } if (rsc_ticket->ticket->granted && rsc_ticket->ticket->standby == FALSE) { return; } if (rsc_lh->children) { GList *gIter = rsc_lh->children; pe_rsc_trace(rsc_lh, "Processing ticket dependencies from %s", rsc_lh->id); for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; rsc_ticket_constraint(child_rsc, rsc_ticket, data_set); } return; } pe_rsc_trace(rsc_lh, "%s: Processing ticket dependency on %s (%s, %s)", rsc_lh->id, rsc_ticket->ticket->id, rsc_ticket->id, role2text(rsc_ticket->role_lh)); if ((rsc_ticket->ticket->granted == FALSE) && (rsc_lh->running_on != NULL)) { GList *gIter = NULL; switch (rsc_ticket->loss_policy) { case loss_ticket_stop: resource_location(rsc_lh, NULL, -INFINITY, "__loss_of_ticket__", data_set); break; case loss_ticket_demote: // Promotion score will be set to -INFINITY in promotion_order() if (rsc_ticket->role_lh != RSC_ROLE_PROMOTED) { resource_location(rsc_lh, NULL, -INFINITY, "__loss_of_ticket__", data_set); } break; case loss_ticket_fence: if (filter_rsc_ticket(rsc_lh, rsc_ticket) == FALSE) { return; } resource_location(rsc_lh, NULL, -INFINITY, "__loss_of_ticket__", data_set); for (gIter = rsc_lh->running_on; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; pe_fence_node(data_set, node, "deadman ticket was lost", FALSE); } break; case loss_ticket_freeze: if (filter_rsc_ticket(rsc_lh, rsc_ticket) == FALSE) { return; } if (rsc_lh->running_on != NULL) { pe__clear_resource_flags(rsc_lh, pe_rsc_managed); pe__set_resource_flags(rsc_lh, pe_rsc_block); } break; } } else if (rsc_ticket->ticket->granted == FALSE) { if ((rsc_ticket->role_lh != RSC_ROLE_PROMOTED) || (rsc_ticket->loss_policy == loss_ticket_stop)) { resource_location(rsc_lh, NULL, -INFINITY, "__no_ticket__", data_set); } } else if (rsc_ticket->ticket->standby) { if ((rsc_ticket->role_lh != RSC_ROLE_PROMOTED) || (rsc_ticket->loss_policy == loss_ticket_stop)) { resource_location(rsc_lh, NULL, -INFINITY, "__ticket_standby__", data_set); } } } enum pe_action_flags native_action_flags(pe_action_t * action, pe_node_t * node) { return action->flags; } static inline bool is_primitive_action(pe_action_t *action) { return action && action->rsc && (action->rsc->variant == pe_native); } /*! * \internal * \brief Clear a single action flag and set reason text * * \param[in] action Action whose flag should be cleared * \param[in] flag Action flag that should be cleared * \param[in] reason Action that is the reason why flag is being cleared */ #define clear_action_flag_because(action, flag, reason) do { \ if (pcmk_is_set((action)->flags, (flag))) { \ pe__clear_action_flags(action, flag); \ if ((action)->rsc != (reason)->rsc) { \ char *reason_text = pe__action2reason((reason), (flag)); \ pe_action_set_reason((action), reason_text, \ ((flag) == pe_action_migrate_runnable)); \ free(reason_text); \ } \ } \ } while (0) /*! * \internal * \brief Set action bits appropriately when pe_restart_order is used * * \param[in] first 'First' action in an ordering with pe_restart_order * \param[in] then 'Then' action in an ordering with pe_restart_order * \param[in] filter What ordering flags to care about * * \note pe_restart_order is set for "stop resource before starting it" and * "stop later group member before stopping earlier group member" */ static void handle_restart_ordering(pe_action_t *first, pe_action_t *then, enum pe_action_flags filter) { const char *reason = NULL; CRM_ASSERT(is_primitive_action(first)); CRM_ASSERT(is_primitive_action(then)); // We need to update the action in two cases: // ... if 'then' is required if (pcmk_is_set(filter, pe_action_optional) && !pcmk_is_set(then->flags, pe_action_optional)) { reason = "restart"; } /* ... if 'then' is unrunnable action on same resource (if a resource * should restart but can't start, we still want to stop) */ if (pcmk_is_set(filter, pe_action_runnable) && !pcmk_is_set(then->flags, pe_action_runnable) && pcmk_is_set(then->rsc->flags, pe_rsc_managed) && (first->rsc == then->rsc)) { reason = "stop"; } if (reason == NULL) { return; } pe_rsc_trace(first->rsc, "Handling %s -> %s for %s", first->uuid, then->uuid, reason); // Make 'first' required if it is runnable if (pcmk_is_set(first->flags, pe_action_runnable)) { clear_action_flag_because(first, pe_action_optional, then); } // Make 'first' required if 'then' is required if (!pcmk_is_set(then->flags, pe_action_optional)) { clear_action_flag_because(first, pe_action_optional, then); } // Make 'first' unmigratable if 'then' is unmigratable if (!pcmk_is_set(then->flags, pe_action_migrate_runnable)) { clear_action_flag_because(first, pe_action_migrate_runnable, then); } // Make 'then' unrunnable if 'first' is required but unrunnable if (!pcmk_is_set(first->flags, pe_action_optional) && !pcmk_is_set(first->flags, pe_action_runnable)) { clear_action_flag_because(then, pe_action_runnable, first); } } enum pe_graph_flags native_update_actions(pe_action_t *first, pe_action_t *then, pe_node_t *node, enum pe_action_flags flags, enum pe_action_flags filter, enum pe_ordering type, pe_working_set_t *data_set) { /* flags == get_action_flags(first, then_node) called from update_action() */ enum pe_graph_flags changed = pe_graph_none; enum pe_action_flags then_flags = then->flags; enum pe_action_flags first_flags = first->flags; if (type & pe_order_asymmetrical) { pe_resource_t *then_rsc = then->rsc; enum rsc_role_e then_rsc_role = then_rsc ? then_rsc->fns->state(then_rsc, TRUE) : 0; if (!then_rsc) { /* ignore */ } else if ((then_rsc_role == RSC_ROLE_STOPPED) && pcmk__str_eq(then->task, RSC_STOP, pcmk__str_casei)) { /* ignore... if 'then' is supposed to be stopped after 'first', but * then is already stopped, there is nothing to be done when non-symmetrical. */ } else if ((then_rsc_role >= RSC_ROLE_STARTED) && pcmk__str_eq(then->task, RSC_START, pcmk__str_casei) && pcmk_is_set(then->flags, pe_action_optional) && then->node && pcmk__list_of_1(then_rsc->running_on) && then->node->details == ((pe_node_t *) then_rsc->running_on->data)->details) { /* Ignore. If 'then' is supposed to be started after 'first', but * 'then' is already started, there is nothing to be done when * asymmetrical -- unless the start is mandatory, which indicates * the resource is restarting, and the ordering is still needed. */ } else if (!(first->flags & pe_action_runnable)) { /* prevent 'then' action from happening if 'first' is not runnable and * 'then' has not yet occurred. */ clear_action_flag_because(then, pe_action_optional, first); clear_action_flag_because(then, pe_action_runnable, first); } else { /* ignore... then is allowed to start/stop if it wants to. */ } } if (pcmk_is_set(type, pe_order_implies_first) && !pcmk_is_set(then_flags, pe_action_optional)) { // Then is required, and implies first should be, too if (pcmk_is_set(filter, pe_action_optional) && !pcmk_is_set(flags, pe_action_optional) && pcmk_is_set(first_flags, pe_action_optional)) { clear_action_flag_because(first, pe_action_optional, then); } if (pcmk_is_set(flags, pe_action_migrate_runnable) && !pcmk_is_set(then->flags, pe_action_migrate_runnable)) { clear_action_flag_because(first, pe_action_migrate_runnable, then); } } if (type & pe_order_promoted_implies_first) { if ((filter & pe_action_optional) && ((then->flags & pe_action_optional) == FALSE) && (then->rsc != NULL) && (then->rsc->role == RSC_ROLE_PROMOTED)) { clear_action_flag_because(first, pe_action_optional, then); if (pcmk_is_set(first->flags, pe_action_migrate_runnable) && !pcmk_is_set(then->flags, pe_action_migrate_runnable)) { clear_action_flag_because(first, pe_action_migrate_runnable, then); } } } if ((type & pe_order_implies_first_migratable) && pcmk_is_set(filter, pe_action_optional)) { if (((then->flags & pe_action_migrate_runnable) == FALSE) || ((then->flags & pe_action_runnable) == FALSE)) { clear_action_flag_because(first, pe_action_runnable, then); } if ((then->flags & pe_action_optional) == 0) { clear_action_flag_because(first, pe_action_optional, then); } } if ((type & pe_order_pseudo_left) && pcmk_is_set(filter, pe_action_optional)) { if ((first->flags & pe_action_runnable) == FALSE) { clear_action_flag_because(then, pe_action_migrate_runnable, first); pe__clear_action_flags(then, pe_action_pseudo); } } if (pcmk_is_set(type, pe_order_runnable_left) && pcmk_is_set(filter, pe_action_runnable) && pcmk_is_set(then->flags, pe_action_runnable) && !pcmk_is_set(flags, pe_action_runnable)) { clear_action_flag_because(then, pe_action_runnable, first); clear_action_flag_because(then, pe_action_migrate_runnable, first); } if (pcmk_is_set(type, pe_order_implies_then) && pcmk_is_set(filter, pe_action_optional) && pcmk_is_set(then->flags, pe_action_optional) && !pcmk_is_set(flags, pe_action_optional) && !pcmk_is_set(first->flags, pe_action_migrate_runnable)) { clear_action_flag_because(then, pe_action_optional, first); } if (pcmk_is_set(type, pe_order_restart)) { handle_restart_ordering(first, then, filter); } if (then_flags != then->flags) { pe__set_graph_flags(changed, first, pe_graph_updated_then); pe_rsc_trace(then->rsc, "%s on %s: flags are now 0x%.6x (was 0x%.6x) " "because of 'first' %s (0x%.6x)", then->uuid, then->node? then->node->details->uname : "no node", then->flags, then_flags, first->uuid, first->flags); if(then->rsc && then->rsc->parent) { /* "X_stop then X_start" doesn't get handled for cloned groups unless we do this */ update_action(then, data_set); } } if (first_flags != first->flags) { pe__set_graph_flags(changed, first, pe_graph_updated_first); pe_rsc_trace(first->rsc, "%s on %s: flags are now 0x%.6x (was 0x%.6x) " "because of 'then' %s (0x%.6x)", first->uuid, first->node? first->node->details->uname : "no node", first->flags, first_flags, then->uuid, then->flags); } return changed; } void native_rsc_location(pe_resource_t *rsc, pe__location_t *constraint) { GList *gIter = NULL; bool need_role = false; CRM_CHECK((constraint != NULL) && (rsc != NULL), return); // If a role was specified, ensure constraint is applicable need_role = (constraint->role_filter > RSC_ROLE_UNKNOWN); if (need_role && (constraint->role_filter != rsc->next_role)) { pe_rsc_trace(rsc, "Not applying %s to %s because role will be %s not %s", constraint->id, rsc->id, role2text(rsc->next_role), role2text(constraint->role_filter)); return; } if (constraint->node_list_rh == NULL) { pe_rsc_trace(rsc, "Not applying %s to %s because no nodes match", constraint->id, rsc->id); return; } pe_rsc_trace(rsc, "Applying %s%s%s to %s", constraint->id, (need_role? " for role " : ""), (need_role? role2text(constraint->role_filter) : ""), rsc->id); for (gIter = constraint->node_list_rh; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; pe_node_t *other_node = NULL; other_node = (pe_node_t *) pe_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (other_node != NULL) { pe_rsc_trace(rsc, "* + %d on %s", node->weight, node->details->uname); other_node->weight = pe__add_scores(other_node->weight, node->weight); } else { pe_rsc_trace(rsc, "* = %d on %s", node->weight, node->details->uname); other_node = pe__copy_node(node); g_hash_table_insert(rsc->allowed_nodes, (gpointer) other_node->details->id, other_node); } if (other_node->rsc_discover_mode < constraint->discover_mode) { if (constraint->discover_mode == pe_discover_exclusive) { rsc->exclusive_discover = TRUE; } /* exclusive > never > always... always is default */ other_node->rsc_discover_mode = constraint->discover_mode; } } } void native_expand(pe_resource_t * rsc, pe_working_set_t * data_set) { GList *gIter = NULL; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "Processing actions from %s", rsc->id); for (gIter = rsc->actions; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; crm_trace("processing action %d for rsc=%s", action->id, rsc->id); graph_element_from_action(action, data_set); } for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; child_rsc->cmds->expand(child_rsc, data_set); } } #define STOP_SANITY_ASSERT(lineno) do { \ if(current && current->details->unclean) { \ /* It will be a pseudo op */ \ } else if(stop == NULL) { \ crm_err("%s:%d: No stop action exists for %s", \ __func__, lineno, rsc->id); \ CRM_ASSERT(stop != NULL); \ } else if (pcmk_is_set(stop->flags, pe_action_optional)) { \ crm_err("%s:%d: Action %s is still optional", \ __func__, lineno, stop->uuid); \ CRM_ASSERT(!pcmk_is_set(stop->flags, pe_action_optional)); \ } \ } while(0) void LogActions(pe_resource_t * rsc, pe_working_set_t * data_set) { pcmk__output_t *out = data_set->priv; pe_node_t *next = NULL; pe_node_t *current = NULL; gboolean moving = FALSE; if(rsc->variant == pe_container) { pcmk__bundle_log_actions(rsc, data_set); return; } if (rsc->children) { g_list_foreach(rsc->children, (GFunc) LogActions, data_set); return; } next = rsc->allocated_to; if (rsc->running_on) { current = pe__current_node(rsc); if (rsc->role == RSC_ROLE_STOPPED) { /* * This can occur when resources are being recovered * We fiddle with the current role in native_create_actions() */ rsc->role = RSC_ROLE_STARTED; } } if ((current == NULL) && pcmk_is_set(rsc->flags, pe_rsc_orphan)) { /* Don't log stopped orphans */ return; } out->message(out, "rsc-action", rsc, current, next, moving); } gboolean StopRsc(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set) { GList *gIter = NULL; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "%s", rsc->id); for (gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) { pe_node_t *current = (pe_node_t *) gIter->data; pe_action_t *stop; if (rsc->partial_migration_target) { if (rsc->partial_migration_target->details == current->details) { pe_rsc_trace(rsc, "Filtered %s -> %s %s", current->details->uname, next->details->uname, rsc->id); continue; } else { pe_rsc_trace(rsc, "Forced on %s %s", current->details->uname, rsc->id); optional = FALSE; } } pe_rsc_trace(rsc, "%s on %s", rsc->id, current->details->uname); stop = stop_action(rsc, current, optional); if(rsc->allocated_to == NULL) { pe_action_set_reason(stop, "node availability", TRUE); } if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe__clear_action_flags(stop, pe_action_runnable); } if (pcmk_is_set(data_set->flags, pe_flag_remove_after_stop)) { DeleteRsc(rsc, current, optional, data_set); } if (pcmk_is_set(rsc->flags, pe_rsc_needs_unfencing)) { pe_action_t *unfence = pe_fence_op(current, "on", TRUE, NULL, FALSE, data_set); order_actions(stop, unfence, pe_order_implies_first); if (!node_has_been_unfenced(current)) { pe_proc_err("Stopping %s until %s can be unfenced", rsc->id, current->details->uname); } } } return TRUE; } static void order_after_unfencing(pe_resource_t *rsc, pe_node_t *node, pe_action_t *action, enum pe_ordering order, pe_working_set_t *data_set) { /* When unfencing is in use, we order unfence actions before any probe or * start of resources that require unfencing, and also of fence devices. * * This might seem to violate the principle that fence devices require * only quorum. However, fence agents that unfence often don't have enough * information to even probe or start unless the node is first unfenced. */ if (is_unfence_device(rsc, data_set) || pcmk_is_set(rsc->flags, pe_rsc_needs_unfencing)) { /* Start with an optional ordering. Requiring unfencing would result in * the node being unfenced, and all its resources being stopped, * whenever a new resource is added -- which would be highly suboptimal. */ pe_action_t *unfence = pe_fence_op(node, "on", TRUE, NULL, FALSE, data_set); order_actions(unfence, action, order); if (!node_has_been_unfenced(node)) { // But unfencing is required if it has never been done char *reason = crm_strdup_printf("required by %s %s", rsc->id, action->task); trigger_unfencing(NULL, node, reason, NULL, data_set); free(reason); } } } gboolean StartRsc(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set) { pe_action_t *start = NULL; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "%s on %s %d %d", rsc->id, next ? next->details->uname : "N/A", optional, next ? next->weight : 0); start = start_action(rsc, next, TRUE); order_after_unfencing(rsc, next, start, pe_order_implies_then, data_set); if (pcmk_is_set(start->flags, pe_action_runnable) && !optional) { pe__clear_action_flags(start, pe_action_optional); } return TRUE; } gboolean PromoteRsc(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set) { GList *gIter = NULL; gboolean runnable = TRUE; GList *action_list = NULL; CRM_ASSERT(rsc); CRM_CHECK(next != NULL, return FALSE); pe_rsc_trace(rsc, "%s on %s", rsc->id, next->details->uname); action_list = pe__resource_actions(rsc, next, RSC_START, TRUE); for (gIter = action_list; gIter != NULL; gIter = gIter->next) { pe_action_t *start = (pe_action_t *) gIter->data; if (!pcmk_is_set(start->flags, pe_action_runnable)) { runnable = FALSE; } } g_list_free(action_list); if (runnable) { promote_action(rsc, next, optional); return TRUE; } pe_rsc_debug(rsc, "%s\tPromote %s (canceled)", next->details->uname, rsc->id); action_list = pe__resource_actions(rsc, next, RSC_PROMOTE, TRUE); for (gIter = action_list; gIter != NULL; gIter = gIter->next) { pe_action_t *promote = (pe_action_t *) gIter->data; pe__clear_action_flags(promote, pe_action_runnable); } g_list_free(action_list); return TRUE; } gboolean DemoteRsc(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set) { GList *gIter = NULL; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "%s", rsc->id); /* CRM_CHECK(rsc->next_role == RSC_ROLE_UNPROMOTED, return FALSE); */ for (gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) { pe_node_t *current = (pe_node_t *) gIter->data; pe_rsc_trace(rsc, "%s on %s", rsc->id, next ? next->details->uname : "N/A"); demote_action(rsc, current, optional); } return TRUE; } gboolean RoleError(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set) { CRM_ASSERT(rsc); crm_err("%s on %s", rsc->id, next ? next->details->uname : "N/A"); CRM_CHECK(FALSE, return FALSE); return FALSE; } gboolean NullOp(pe_resource_t * rsc, pe_node_t * next, gboolean optional, pe_working_set_t * data_set) { CRM_ASSERT(rsc); pe_rsc_trace(rsc, "%s", rsc->id); return FALSE; } gboolean DeleteRsc(pe_resource_t * rsc, pe_node_t * node, gboolean optional, pe_working_set_t * data_set) { if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { pe_rsc_trace(rsc, "Resource %s not deleted from %s: failed", rsc->id, node->details->uname); return FALSE; } else if (node == NULL) { pe_rsc_trace(rsc, "Resource %s not deleted: NULL node", rsc->id); return FALSE; } else if (node->details->unclean || node->details->online == FALSE) { pe_rsc_trace(rsc, "Resource %s not deleted from %s: unrunnable", rsc->id, node->details->uname); return FALSE; } crm_notice("Removing %s from %s", rsc->id, node->details->uname); delete_action(rsc, node, optional); new_rsc_order(rsc, RSC_STOP, rsc, RSC_DELETE, optional ? pe_order_implies_then : pe_order_optional, data_set); new_rsc_order(rsc, RSC_DELETE, rsc, RSC_START, optional ? pe_order_implies_then : pe_order_optional, data_set); return TRUE; } gboolean native_create_probe(pe_resource_t * rsc, pe_node_t * node, pe_action_t * complete, gboolean force, pe_working_set_t * data_set) { enum pe_ordering flags = pe_order_optional; char *key = NULL; pe_action_t *probe = NULL; pe_node_t *running = NULL; pe_node_t *allowed = NULL; pe_resource_t *top = uber_parent(rsc); static const char *rc_promoted = NULL; static const char *rc_inactive = NULL; if (rc_inactive == NULL) { rc_inactive = pcmk__itoa(PCMK_OCF_NOT_RUNNING); rc_promoted = pcmk__itoa(PCMK_OCF_RUNNING_PROMOTED); } CRM_CHECK(node != NULL, return FALSE); if (!force && !pcmk_is_set(data_set->flags, pe_flag_startup_probes)) { pe_rsc_trace(rsc, "Skipping active resource detection for %s", rsc->id); return FALSE; } if (pe__is_guest_or_remote_node(node)) { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { pe_rsc_trace(rsc, "Skipping probe for %s on %s because Pacemaker Remote nodes cannot run stonith agents", rsc->id, node->details->id); return FALSE; } else if (pe__is_guest_node(node) && pe__resource_contains_guest_node(data_set, rsc)) { pe_rsc_trace(rsc, "Skipping probe for %s on %s because guest nodes cannot run resources containing guest nodes", rsc->id, node->details->id); return FALSE; } else if (rsc->is_remote_node) { pe_rsc_trace(rsc, "Skipping probe for %s on %s because Pacemaker Remote nodes cannot host remote connections", rsc->id, node->details->id); return FALSE; } } if (rsc->children) { GList *gIter = NULL; gboolean any_created = FALSE; for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; any_created = child_rsc->cmds->create_probe(child_rsc, node, complete, force, data_set) || any_created; } return any_created; } else if ((rsc->container) && (!rsc->is_remote_node)) { pe_rsc_trace(rsc, "Skipping %s: it is within container %s", rsc->id, rsc->container->id); return FALSE; } if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) { pe_rsc_trace(rsc, "Skipping orphan: %s", rsc->id); return FALSE; } // Check whether resource is already known on node if (!force && g_hash_table_lookup(rsc->known_on, node->details->id)) { pe_rsc_trace(rsc, "Skipping known: %s on %s", rsc->id, node->details->uname); return FALSE; } allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (rsc->exclusive_discover || top->exclusive_discover) { if (allowed == NULL) { /* exclusive discover is enabled and this node is not in the allowed list. */ pe_rsc_trace(rsc, "Skipping probe for %s on node %s, A", rsc->id, node->details->id); return FALSE; } else if (allowed->rsc_discover_mode != pe_discover_exclusive) { /* exclusive discover is enabled and this node is not marked * as a node this resource should be discovered on */ pe_rsc_trace(rsc, "Skipping probe for %s on node %s, B", rsc->id, node->details->id); return FALSE; } } if(allowed == NULL && node->rsc_discover_mode == pe_discover_never) { /* If this node was allowed to host this resource it would * have been explicitly added to the 'allowed_nodes' list. * However it wasn't and the node has discovery disabled, so * no need to probe for this resource. */ pe_rsc_trace(rsc, "Skipping probe for %s on node %s, C", rsc->id, node->details->id); return FALSE; } if (allowed && allowed->rsc_discover_mode == pe_discover_never) { /* this resource is marked as not needing to be discovered on this node */ pe_rsc_trace(rsc, "Skipping probe for %s on node %s, discovery mode", rsc->id, node->details->id); return FALSE; } if (pe__is_guest_node(node)) { pe_resource_t *remote = node->details->remote_rsc->container; if(remote->role == RSC_ROLE_STOPPED) { /* If the container is stopped, then we know anything that * might have been inside it is also stopped and there is * no need to probe. * * If we don't know the container's state on the target * either: * * - the container is running, the transition will abort * and we'll end up in a different case next time, or * * - the container is stopped * * Either way there is no need to probe. * */ if(remote->allocated_to && g_hash_table_lookup(remote->known_on, remote->allocated_to->details->id) == NULL) { /* For safety, we order the 'rsc' start after 'remote' * has been probed. * * Using 'top' helps for groups, but we may need to * follow the start's ordering chain backwards. */ custom_action_order(remote, pcmk__op_key(remote->id, RSC_STATUS, 0), NULL, top, pcmk__op_key(top->id, RSC_START, 0), NULL, pe_order_optional, data_set); } pe_rsc_trace(rsc, "Skipping probe for %s on node %s, %s is stopped", rsc->id, node->details->id, remote->id); return FALSE; /* Here we really we want to check if remote->stop is required, * but that information doesn't exist yet */ } else if(node->details->remote_requires_reset || node->details->unclean || pcmk_is_set(remote->flags, pe_rsc_failed) || remote->next_role == RSC_ROLE_STOPPED || (remote->allocated_to && pe_find_node(remote->running_on, remote->allocated_to->details->uname) == NULL) ) { /* The container is stopping or restarting, don't start * 'rsc' until 'remote' stops as this also implies that * 'rsc' is stopped - avoiding the need to probe */ custom_action_order(remote, pcmk__op_key(remote->id, RSC_STOP, 0), NULL, top, pcmk__op_key(top->id, RSC_START, 0), NULL, pe_order_optional, data_set); pe_rsc_trace(rsc, "Skipping probe for %s on node %s, %s is stopping, restarting or moving", rsc->id, node->details->id, remote->id); return FALSE; /* } else { * The container is running so there is no problem probing it */ } } key = pcmk__op_key(rsc->id, RSC_STATUS, 0); probe = custom_action(rsc, key, RSC_STATUS, node, FALSE, TRUE, data_set); pe__clear_action_flags(probe, pe_action_optional); order_after_unfencing(rsc, node, probe, pe_order_optional, data_set); /* * We need to know if it's running_on (not just known_on) this node * to correctly determine the target rc. */ running = pe_find_node_id(rsc->running_on, node->details->id); if (running == NULL) { add_hash_param(probe->meta, XML_ATTR_TE_TARGET_RC, rc_inactive); } else if (rsc->role == RSC_ROLE_PROMOTED) { add_hash_param(probe->meta, XML_ATTR_TE_TARGET_RC, rc_promoted); } crm_debug("Probing %s on %s (%s) %d %p", rsc->id, node->details->uname, role2text(rsc->role), pcmk_is_set(probe->flags, pe_action_runnable), rsc->running_on); if (is_unfence_device(rsc, data_set) || !pe_rsc_is_clone(top)) { top = rsc; } else { crm_trace("Probing %s on %s (%s) as %s", rsc->id, node->details->uname, role2text(rsc->role), top->id); } if (!pcmk_is_set(probe->flags, pe_action_runnable) && (rsc->running_on == NULL)) { /* Prevent the start from occurring if rsc isn't active, but * don't cause it to stop if it was active already */ pe__set_order_flags(flags, pe_order_runnable_left); } custom_action_order(rsc, NULL, probe, top, pcmk__op_key(top->id, RSC_START, 0), NULL, flags, data_set); // Order the probe before any agent reload custom_action_order(rsc, NULL, probe, top, reload_key(rsc), NULL, pe_order_optional, data_set); #if 0 // complete is always null currently if (!is_unfence_device(rsc, data_set)) { /* Normally rsc.start depends on probe complete which depends * on rsc.probe. But this can't be the case for fence devices * with unfencing, as it would create graph loops. * * So instead we explicitly order 'rsc.probe then rsc.start' */ order_actions(probe, complete, pe_order_implies_then); } #endif return TRUE; } /*! * \internal * \brief Check whether a resource is known on a particular node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return TRUE if resource (or parent if an anonymous clone) is known */ static bool rsc_is_known_on(pe_resource_t *rsc, const pe_node_t *node) { if (pe_hash_table_lookup(rsc->known_on, node->details->id)) { return TRUE; } else if ((rsc->variant == pe_native) && pe_rsc_is_anon_clone(rsc->parent) && pe_hash_table_lookup(rsc->parent->known_on, node->details->id)) { /* We check only the parent, not the uber-parent, because we cannot * assume that the resource is known if it is in an anonymously cloned * group (which may be only partially known). */ return TRUE; } return FALSE; } /*! * \internal * \brief Order a resource's start and promote actions relative to fencing * * \param[in] rsc Resource to be ordered * \param[in] stonith_op Fence action * \param[in] data_set Cluster information */ static void native_start_constraints(pe_resource_t * rsc, pe_action_t * stonith_op, pe_working_set_t * data_set) { pe_node_t *target; GList *gIter = NULL; CRM_CHECK(stonith_op && stonith_op->node, return); target = stonith_op->node; for (gIter = rsc->actions; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; switch (action->needs) { case rsc_req_nothing: // Anything other than start or promote requires nothing break; case rsc_req_stonith: order_actions(stonith_op, action, pe_order_optional); break; case rsc_req_quorum: if (pcmk__str_eq(action->task, RSC_START, pcmk__str_casei) && pe_hash_table_lookup(rsc->allowed_nodes, target->details->id) && !rsc_is_known_on(rsc, target)) { /* If we don't know the status of the resource on the node * we're about to shoot, we have to assume it may be active * there. Order the resource start after the fencing. This * is analogous to waiting for all the probes for a resource * to complete before starting it. * * The most likely explanation is that the DC died and took * its status with it. */ pe_rsc_debug(rsc, "Ordering %s after %s recovery", action->uuid, target->details->uname); order_actions(stonith_op, action, pe_order_optional | pe_order_runnable_left); } break; } } } static void native_stop_constraints(pe_resource_t * rsc, pe_action_t * stonith_op, pe_working_set_t * data_set) { GList *gIter = NULL; GList *action_list = NULL; bool order_implicit = false; pe_resource_t *top = uber_parent(rsc); pe_action_t *parent_stop = NULL; pe_node_t *target; CRM_CHECK(stonith_op && stonith_op->node, return); target = stonith_op->node; /* Get a list of stop actions potentially implied by the fencing */ action_list = pe__resource_actions(rsc, target, RSC_STOP, FALSE); /* If resource requires fencing, implicit actions must occur after fencing. * * Implied stops and demotes of resources running on guest nodes are always * ordered after fencing, even if the resource does not require fencing, * because guest node "fencing" is actually just a resource stop. */ if (pcmk_is_set(rsc->flags, pe_rsc_needs_fencing) || pe__is_guest_node(target)) { order_implicit = true; } if (action_list && order_implicit) { parent_stop = find_first_action(top->actions, NULL, RSC_STOP, NULL); } for (gIter = action_list; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; // The stop would never complete, so convert it into a pseudo-action. pe__set_action_flags(action, pe_action_pseudo|pe_action_runnable); if (order_implicit) { pe__set_action_flags(action, pe_action_implied_by_stonith); /* Order the stonith before the parent stop (if any). * * Also order the stonith before the resource stop, unless the * resource is inside a bundle -- that would cause a graph loop. * We can rely on the parent stop's ordering instead. * * User constraints must not order a resource in a guest node * relative to the guest node container resource. The * pe_order_preserve flag marks constraints as generated by the * cluster and thus immune to that check (and is irrelevant if * target is not a guest). */ if (!pe_rsc_is_bundled(rsc)) { order_actions(stonith_op, action, pe_order_preserve); } order_actions(stonith_op, parent_stop, pe_order_preserve); } if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { crm_notice("Stop of failed resource %s is implicit %s %s is fenced", rsc->id, (order_implicit? "after" : "because"), target->details->uname); } else { crm_info("%s is implicit %s %s is fenced", action->uuid, (order_implicit? "after" : "because"), target->details->uname); } if (pcmk_is_set(rsc->flags, pe_rsc_notify)) { /* Create a second notification that will be delivered * immediately after the node is fenced * * Basic problem: * - C is a clone active on the node to be shot and stopping on another * - R is a resource that depends on C * * + C.stop depends on R.stop * + C.stopped depends on STONITH * + C.notify depends on C.stopped * + C.healthy depends on C.notify * + R.stop depends on C.healthy * * The extra notification here changes * + C.healthy depends on C.notify * into: * + C.healthy depends on C.notify' * + C.notify' depends on STONITH' * thus breaking the loop */ create_secondary_notification(action, rsc, stonith_op, data_set); } /* From Bug #1601, successful fencing must be an input to a failed resources stop action. However given group(rA, rB) running on nodeX and B.stop has failed, A := stop healthy resource (rA.stop) B := stop failed resource (pseudo operation B.stop) C := stonith nodeX A requires B, B requires C, C requires A This loop would prevent the cluster from making progress. This block creates the "C requires A" dependency and therefore must (at least for now) be disabled. Instead, run the block above and treat all resources on nodeX as B would be (marked as a pseudo op depending on the STONITH). TODO: Break the "A requires B" dependency in update_action() and re-enable this block } else if(is_stonith == FALSE) { crm_info("Moving healthy resource %s" " off %s before fencing", rsc->id, node->details->uname); * stop healthy resources before the * stonith op * custom_action_order( rsc, stop_key(rsc), NULL, NULL,strdup(CRM_OP_FENCE),stonith_op, pe_order_optional, data_set); */ } g_list_free(action_list); /* Get a list of demote actions potentially implied by the fencing */ action_list = pe__resource_actions(rsc, target, RSC_DEMOTE, FALSE); for (gIter = action_list; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (action->node->details->online == FALSE || action->node->details->unclean == TRUE || pcmk_is_set(rsc->flags, pe_rsc_failed)) { if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { pe_rsc_info(rsc, "Demote of failed resource %s is implicit after %s is fenced", rsc->id, target->details->uname); } else { pe_rsc_info(rsc, "%s is implicit after %s is fenced", action->uuid, target->details->uname); } /* The demote would never complete and is now implied by the * fencing, so convert it into a pseudo-action. */ pe__set_action_flags(action, pe_action_pseudo|pe_action_runnable); if (pe_rsc_is_bundled(rsc)) { /* Do nothing, let the recovery be ordered after the parent's implied stop */ } else if (order_implicit) { order_actions(stonith_op, action, pe_order_preserve|pe_order_optional); } } } g_list_free(action_list); } void rsc_stonith_ordering(pe_resource_t * rsc, pe_action_t * stonith_op, pe_working_set_t * data_set) { if (rsc->children) { GList *gIter = NULL; for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; rsc_stonith_ordering(child_rsc, stonith_op, data_set); } } else if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe_rsc_trace(rsc, "Skipping fencing constraints for unmanaged resource: %s", rsc->id); } else { native_start_constraints(rsc, stonith_op, data_set); native_stop_constraints(rsc, stonith_op, data_set); } } void ReloadRsc(pe_resource_t * rsc, pe_node_t *node, pe_working_set_t * data_set) { GList *gIter = NULL; pe_action_t *reload = NULL; if (rsc->children) { for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; ReloadRsc(child_rsc, node, data_set); } return; } else if (rsc->variant > pe_native) { /* Complex resource with no children */ return; } else if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe_rsc_trace(rsc, "%s: unmanaged", rsc->id); return; } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { /* We don't need to specify any particular actions here, normal failure * recovery will apply. */ pe_rsc_trace(rsc, "%s: preventing agent reload because failed", rsc->id); return; } else if (pcmk_is_set(rsc->flags, pe_rsc_start_pending)) { /* If a resource's configuration changed while a start was pending, * force a full restart. */ pe_rsc_trace(rsc, "%s: preventing agent reload because start pending", rsc->id); stop_action(rsc, node, FALSE); return; } else if (node == NULL) { pe_rsc_trace(rsc, "%s: not active", rsc->id); return; } pe_rsc_trace(rsc, "Processing %s", rsc->id); pe__set_resource_flags(rsc, pe_rsc_reload); reload = custom_action(rsc, reload_key(rsc), CRMD_ACTION_RELOAD_AGENT, node, FALSE, TRUE, data_set); pe_action_set_reason(reload, "resource definition change", FALSE); custom_action_order(NULL, NULL, reload, rsc, stop_key(rsc), NULL, pe_order_optional|pe_order_then_cancels_first, data_set); custom_action_order(NULL, NULL, reload, rsc, demote_key(rsc), NULL, pe_order_optional|pe_order_then_cancels_first, data_set); } void native_append_meta(pe_resource_t * rsc, xmlNode * xml) { char *value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INCARNATION); pe_resource_t *parent; if (value) { char *name = NULL; name = crm_meta_name(XML_RSC_ATTR_INCARNATION); crm_xml_add(xml, name, value); free(name); } value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_REMOTE_NODE); if (value) { char *name = NULL; name = crm_meta_name(XML_RSC_ATTR_REMOTE_NODE); crm_xml_add(xml, name, value); free(name); } for (parent = rsc; parent != NULL; parent = parent->parent) { if (parent->container) { crm_xml_add(xml, CRM_META"_"XML_RSC_ATTR_CONTAINER, parent->container->id); } } } diff --git a/lib/pacemaker/pcmk_sched_utils.c b/lib/pacemaker/pcmk_sched_utils.c index 85249f8608..5577a9c2e0 100644 --- a/lib/pacemaker/pcmk_sched_utils.c +++ b/lib/pacemaker/pcmk_sched_utils.c @@ -1,789 +1,852 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include // lrmd_event_data_t #include #include pe__location_t * rsc2node_new(const char *id, pe_resource_t *rsc, int node_weight, const char *discover_mode, pe_node_t *foo_node, pe_working_set_t *data_set) { pe__location_t *new_con = NULL; if (rsc == NULL || id == NULL) { pe_err("Invalid constraint %s for rsc=%p", crm_str(id), rsc); return NULL; } else if (foo_node == NULL) { CRM_CHECK(node_weight == 0, return NULL); } new_con = calloc(1, sizeof(pe__location_t)); if (new_con != NULL) { new_con->id = strdup(id); new_con->rsc_lh = rsc; new_con->node_list_rh = NULL; new_con->role_filter = RSC_ROLE_UNKNOWN; if (pcmk__str_eq(discover_mode, "always", pcmk__str_null_matches | pcmk__str_casei)) { new_con->discover_mode = pe_discover_always; } else if (pcmk__str_eq(discover_mode, "never", pcmk__str_casei)) { new_con->discover_mode = pe_discover_never; } else if (pcmk__str_eq(discover_mode, "exclusive", pcmk__str_casei)) { new_con->discover_mode = pe_discover_exclusive; rsc->exclusive_discover = TRUE; } else { pe_err("Invalid %s value %s in location constraint", XML_LOCATION_ATTR_DISCOVERY, discover_mode); } if (foo_node != NULL) { pe_node_t *copy = pe__copy_node(foo_node); copy->weight = node_weight; new_con->node_list_rh = g_list_prepend(NULL, copy); } data_set->placement_constraints = g_list_prepend(data_set->placement_constraints, new_con); rsc->rsc_location = g_list_prepend(rsc->rsc_location, new_con); } return new_con; } gboolean can_run_resources(const pe_node_t * node) { if (node == NULL) { return FALSE; } #if 0 if (node->weight < 0) { return FALSE; } #endif if (node->details->online == FALSE || node->details->shutdown || node->details->unclean || node->details->standby || node->details->maintenance) { crm_trace("%s: online=%d, unclean=%d, standby=%d, maintenance=%d", node->details->uname, node->details->online, node->details->unclean, node->details->standby, node->details->maintenance); return FALSE; } return TRUE; } /*! * \internal * \brief Copy a hash table of node objects * * \param[in] nodes Hash table to copy * * \return New copy of nodes (or NULL if nodes is NULL) */ GHashTable * pcmk__copy_node_table(GHashTable *nodes) { GHashTable *new_table = NULL; GHashTableIter iter; pe_node_t *node = NULL; if (nodes == NULL) { return NULL; } new_table = pcmk__strkey_table(NULL, free); g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { pe_node_t *new_node = pe__copy_node(node); g_hash_table_insert(new_table, (gpointer) new_node->details->id, new_node); } return new_table; } /*! * \internal * \brief Copy a list of node objects * * \param[in] list List to copy * \param[in] reset Set copies' scores to 0 * * \return New list of shallow copies of nodes in original list */ GList * pcmk__copy_node_list(const GList *list, bool reset) { GList *result = NULL; for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) { pe_node_t *new_node = NULL; pe_node_t *this_node = (pe_node_t *) gIter->data; new_node = pe__copy_node(this_node); if (reset) { new_node->weight = 0; } result = g_list_prepend(result, new_node); } return result; } struct node_weight_s { pe_node_t *active; pe_working_set_t *data_set; }; /* return -1 if 'a' is more preferred * return 1 if 'b' is more preferred */ static gint sort_node_weight(gconstpointer a, gconstpointer b, gpointer data) { const pe_node_t *node1 = (const pe_node_t *)a; const pe_node_t *node2 = (const pe_node_t *)b; struct node_weight_s *nw = data; int node1_weight = 0; int node2_weight = 0; int result = 0; if (a == NULL) { return 1; } if (b == NULL) { return -1; } node1_weight = node1->weight; node2_weight = node2->weight; if (can_run_resources(node1) == FALSE) { node1_weight = -INFINITY; } if (can_run_resources(node2) == FALSE) { node2_weight = -INFINITY; } if (node1_weight > node2_weight) { crm_trace("%s (%d) > %s (%d) : weight", node1->details->uname, node1_weight, node2->details->uname, node2_weight); return -1; } if (node1_weight < node2_weight) { crm_trace("%s (%d) < %s (%d) : weight", node1->details->uname, node1_weight, node2->details->uname, node2_weight); return 1; } crm_trace("%s (%d) == %s (%d) : weight", node1->details->uname, node1_weight, node2->details->uname, node2_weight); if (pcmk__str_eq(nw->data_set->placement_strategy, "minimal", pcmk__str_casei)) { goto equal; } if (pcmk__str_eq(nw->data_set->placement_strategy, "balanced", pcmk__str_casei)) { result = compare_capacity(node1, node2); if (result < 0) { crm_trace("%s > %s : capacity (%d)", node1->details->uname, node2->details->uname, result); return -1; } else if (result > 0) { crm_trace("%s < %s : capacity (%d)", node1->details->uname, node2->details->uname, result); return 1; } } /* now try to balance resources across the cluster */ if (node1->details->num_resources < node2->details->num_resources) { crm_trace("%s (%d) > %s (%d) : resources", node1->details->uname, node1->details->num_resources, node2->details->uname, node2->details->num_resources); return -1; } else if (node1->details->num_resources > node2->details->num_resources) { crm_trace("%s (%d) < %s (%d) : resources", node1->details->uname, node1->details->num_resources, node2->details->uname, node2->details->num_resources); return 1; } if (nw->active && nw->active->details == node1->details) { crm_trace("%s (%d) > %s (%d) : active", node1->details->uname, node1->details->num_resources, node2->details->uname, node2->details->num_resources); return -1; } else if (nw->active && nw->active->details == node2->details) { crm_trace("%s (%d) < %s (%d) : active", node1->details->uname, node1->details->num_resources, node2->details->uname, node2->details->num_resources); return 1; } equal: crm_trace("%s = %s", node1->details->uname, node2->details->uname); return strcmp(node1->details->uname, node2->details->uname); } GList * sort_nodes_by_weight(GList *nodes, pe_node_t *active_node, pe_working_set_t *data_set) { struct node_weight_s nw = { active_node, data_set }; return g_list_sort_with_data(nodes, sort_node_weight, &nw); } void native_deallocate(pe_resource_t * rsc) { if (rsc->allocated_to) { pe_node_t *old = rsc->allocated_to; crm_info("Deallocating %s from %s", rsc->id, old->details->uname); pe__set_resource_flags(rsc, pe_rsc_provisional); rsc->allocated_to = NULL; old->details->allocated_rsc = g_list_remove(old->details->allocated_rsc, rsc); old->details->num_resources--; /* old->count--; */ calculate_utilization(old->details->utilization, rsc->utilization, TRUE); free(old); } } gboolean native_assign_node(pe_resource_t *rsc, pe_node_t *chosen, gboolean force) { pcmk__output_t *out = rsc->cluster->priv; CRM_ASSERT(rsc->variant == pe_native); if (force == FALSE && chosen != NULL) { bool unset = FALSE; if(chosen->weight < 0) { unset = TRUE; // Allow the graph to assume that the remote resource will come up } else if (!can_run_resources(chosen) && !pe__is_guest_node(chosen)) { unset = TRUE; } if(unset) { crm_debug("All nodes for resource %s are unavailable" ", unclean or shutting down (%s: %d, %d)", rsc->id, chosen->details->uname, can_run_resources(chosen), chosen->weight); pe__set_next_role(rsc, RSC_ROLE_STOPPED, "node availability"); chosen = NULL; } } /* todo: update the old node for each resource to reflect its * new resource count */ native_deallocate(rsc); pe__clear_resource_flags(rsc, pe_rsc_provisional); if (chosen == NULL) { GList *gIter = NULL; char *rc_inactive = pcmk__itoa(PCMK_OCF_NOT_RUNNING); crm_debug("Could not allocate a node for %s", rsc->id); pe__set_next_role(rsc, RSC_ROLE_STOPPED, "unable to allocate"); for (gIter = rsc->actions; gIter != NULL; gIter = gIter->next) { pe_action_t *op = (pe_action_t *) gIter->data; const char *interval_ms_s = g_hash_table_lookup(op->meta, XML_LRM_ATTR_INTERVAL_MS); crm_debug("Processing %s", op->uuid); if(pcmk__str_eq(RSC_STOP, op->task, pcmk__str_casei)) { pe__clear_action_flags(op, pe_action_optional); } else if(pcmk__str_eq(RSC_START, op->task, pcmk__str_casei)) { pe__clear_action_flags(op, pe_action_runnable); //pe__set_resource_flags(rsc, pe_rsc_block); } else if (interval_ms_s && !pcmk__str_eq(interval_ms_s, "0", pcmk__str_casei)) { if(pcmk__str_eq(rc_inactive, g_hash_table_lookup(op->meta, XML_ATTR_TE_TARGET_RC), pcmk__str_casei)) { /* This is a recurring monitor for the stopped state, leave it alone */ } else { /* Normal monitor operation, cancel it */ pe__clear_action_flags(op, pe_action_runnable); } } } free(rc_inactive); return FALSE; } crm_debug("Assigning %s to %s", chosen->details->uname, rsc->id); rsc->allocated_to = pe__copy_node(chosen); chosen->details->allocated_rsc = g_list_prepend(chosen->details->allocated_rsc, rsc); chosen->details->num_resources++; chosen->count++; calculate_utilization(chosen->details->utilization, rsc->utilization, FALSE); if (pcmk_is_set(rsc->cluster->flags, pe_flag_show_utilization)) { out->message(out, "resource-util", rsc, chosen, __func__); } return TRUE; } void log_action(unsigned int log_level, const char *pre_text, pe_action_t * action, gboolean details) { const char *node_uname = NULL; const char *node_uuid = NULL; const char *desc = NULL; if (action == NULL) { crm_trace("%s%s: ", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": "); return; } if (pcmk_is_set(action->flags, pe_action_pseudo)) { node_uname = NULL; node_uuid = NULL; } else if (action->node != NULL) { node_uname = action->node->details->uname; node_uuid = action->node->details->id; } else { node_uname = ""; node_uuid = NULL; } switch (text2task(action->task)) { case stonith_node: case shutdown_crm: if (pcmk_is_set(action->flags, pe_action_pseudo)) { desc = "Pseudo "; } else if (pcmk_is_set(action->flags, pe_action_optional)) { desc = "Optional "; } else if (!pcmk_is_set(action->flags, pe_action_runnable)) { desc = "!!Non-Startable!! "; } else if (pcmk_is_set(action->flags, pe_action_processed)) { desc = ""; } else { desc = "(Provisional) "; } crm_trace("%s%s%sAction %d: %s%s%s%s%s%s", ((pre_text == NULL)? "" : pre_text), ((pre_text == NULL)? "" : ": "), desc, action->id, action->uuid, (node_uname? "\ton " : ""), (node_uname? node_uname : ""), (node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""), (node_uuid? ")" : "")); break; default: if (pcmk_is_set(action->flags, pe_action_optional)) { desc = "Optional "; } else if (pcmk_is_set(action->flags, pe_action_pseudo)) { desc = "Pseudo "; } else if (!pcmk_is_set(action->flags, pe_action_runnable)) { desc = "!!Non-Startable!! "; } else if (pcmk_is_set(action->flags, pe_action_processed)) { desc = ""; } else { desc = "(Provisional) "; } crm_trace("%s%s%sAction %d: %s %s%s%s%s%s%s", ((pre_text == NULL)? "" : pre_text), ((pre_text == NULL)? "" : ": "), desc, action->id, action->uuid, (action->rsc? action->rsc->id : ""), (node_uname? "\ton " : ""), (node_uname? node_uname : ""), (node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""), (node_uuid? ")" : "")); break; } if (details) { GList *gIter = NULL; crm_trace("\t\t====== Preceding Actions"); gIter = action->actions_before; for (; gIter != NULL; gIter = gIter->next) { pe_action_wrapper_t *other = (pe_action_wrapper_t *) gIter->data; log_action(log_level + 1, "\t\t", other->action, FALSE); } crm_trace("\t\t====== Subsequent Actions"); gIter = action->actions_after; for (; gIter != NULL; gIter = gIter->next) { pe_action_wrapper_t *other = (pe_action_wrapper_t *) gIter->data; log_action(log_level + 1, "\t\t", other->action, FALSE); } crm_trace("\t\t====== End"); } else { crm_trace("\t\t(before=%d, after=%d)", g_list_length(action->actions_before), g_list_length(action->actions_after)); } } gboolean can_run_any(GHashTable * nodes) { GHashTableIter iter; pe_node_t *node = NULL; if (nodes == NULL) { return FALSE; } g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if (can_run_resources(node) && node->weight >= 0) { return TRUE; } } return FALSE; } pe_action_t * create_pseudo_resource_op(pe_resource_t * rsc, const char *task, bool optional, bool runnable, pe_working_set_t *data_set) { pe_action_t *action = custom_action(rsc, pcmk__op_key(rsc->id, task, 0), task, NULL, optional, TRUE, data_set); pe__set_action_flags(action, pe_action_pseudo); if(runnable) { pe__set_action_flags(action, pe_action_runnable); } return action; } /*! * \internal * \brief Create an executor cancel op * * \param[in] rsc Resource of action to cancel * \param[in] task Name of action to cancel * \param[in] interval_ms Interval of action to cancel * \param[in] node Node of action to cancel * \param[in] data_set Working set of cluster * * \return Created op */ pe_action_t * pe_cancel_op(pe_resource_t *rsc, const char *task, guint interval_ms, pe_node_t *node, pe_working_set_t *data_set) { pe_action_t *cancel_op; char *interval_ms_s = crm_strdup_printf("%u", interval_ms); // @TODO dangerous if possible to schedule another action with this key char *key = pcmk__op_key(rsc->id, task, interval_ms); cancel_op = custom_action(rsc, key, RSC_CANCEL, node, FALSE, TRUE, data_set); free(cancel_op->task); cancel_op->task = strdup(RSC_CANCEL); free(cancel_op->cancel_task); cancel_op->cancel_task = strdup(task); add_hash_param(cancel_op->meta, XML_LRM_ATTR_TASK, task); add_hash_param(cancel_op->meta, XML_LRM_ATTR_INTERVAL_MS, interval_ms_s); free(interval_ms_s); return cancel_op; } /*! * \internal * \brief Create a shutdown op for a scheduler transition * * \param[in] node Node being shut down * \param[in] data_set Working set of cluster * * \return Created op */ pe_action_t * sched_shutdown_op(pe_node_t *node, pe_working_set_t *data_set) { char *shutdown_id = crm_strdup_printf("%s-%s", CRM_OP_SHUTDOWN, node->details->uname); pe_action_t *shutdown_op = custom_action(NULL, shutdown_id, CRM_OP_SHUTDOWN, node, FALSE, TRUE, data_set); - crm_notice("Scheduling shutdown of node %s", node->details->uname); shutdown_constraints(node, shutdown_op, data_set); add_hash_param(shutdown_op->meta, XML_ATTR_TE_NOWAIT, XML_BOOLEAN_TRUE); return shutdown_op; } static char * generate_transition_magic(const char *transition_key, int op_status, int op_rc) { CRM_CHECK(transition_key != NULL, return NULL); return crm_strdup_printf("%d:%d;%s", op_status, op_rc, transition_key); } static void append_digest(lrmd_event_data_t *op, xmlNode *update, const char *version, const char *magic, int level) { /* this will enable us to later determine that the * resource's parameters have changed and we should force * a restart */ char *digest = NULL; xmlNode *args_xml = NULL; if (op->params == NULL) { return; } args_xml = create_xml_node(NULL, XML_TAG_PARAMS); g_hash_table_foreach(op->params, hash2field, args_xml); pcmk__filter_op_for_digest(args_xml); digest = calculate_operation_digest(args_xml, version); #if 0 if (level < get_crm_log_level() && op->interval_ms == 0 && pcmk__str_eq(op->op_type, CRMD_ACTION_START, pcmk__str_none)) { char *digest_source = dump_xml_unformatted(args_xml); do_crm_log(level, "Calculated digest %s for %s (%s). Source: %s\n", digest, ID(update), magic, digest_source); free(digest_source); } #endif crm_xml_add(update, XML_LRM_ATTR_OP_DIGEST, digest); free_xml(args_xml); free(digest); } #define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" /*! * \internal * \brief Create XML for resource operation history update * * \param[in,out] parent Parent XML node to add to * \param[in,out] op Operation event data * \param[in] caller_version DC feature set * \param[in] target_rc Expected result of operation * \param[in] node Name of node on which operation was performed * \param[in] origin Arbitrary description of update source * \param[in] level A log message will be logged at this level * * \return Newly created XML node for history update */ xmlNode * pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *op, const char *caller_version, int target_rc, const char *node, const char *origin, int level) { char *key = NULL; char *magic = NULL; char *op_id = NULL; char *op_id_additional = NULL; char *local_user_data = NULL; const char *exit_reason = NULL; xmlNode *xml_op = NULL; const char *task = NULL; CRM_CHECK(op != NULL, return NULL); do_crm_log(level, "%s: Updating resource %s after %s op %s (interval=%u)", origin, op->rsc_id, op->op_type, services_lrm_status_str(op->op_status), op->interval_ms); crm_trace("DC version: %s", caller_version); task = op->op_type; /* Record a successful agent reload as a start, and a failed one as a * monitor, to make life easier for the scheduler when determining the * current state. * * @COMPAT We should check "reload" here only if the operation was for a * pre-OCF-1.1 resource agent, but we don't know that here, and we should * only ever get results for actions scheduled by us, so we can reasonably * assume any "reload" is actually a pre-1.1 agent reload. */ if (pcmk__str_any_of(task, CRMD_ACTION_RELOAD, CRMD_ACTION_RELOAD_AGENT, NULL)) { if (op->op_status == PCMK_LRM_OP_DONE) { task = CRMD_ACTION_START; } else { task = CRMD_ACTION_STATUS; } } key = pcmk__op_key(op->rsc_id, task, op->interval_ms); if (pcmk__str_eq(task, CRMD_ACTION_NOTIFY, pcmk__str_none)) { const char *n_type = crm_meta_value(op->params, "notify_type"); const char *n_task = crm_meta_value(op->params, "notify_operation"); CRM_LOG_ASSERT(n_type != NULL); CRM_LOG_ASSERT(n_task != NULL); op_id = pcmk__notify_key(op->rsc_id, n_type, n_task); if (op->op_status != PCMK_LRM_OP_PENDING) { /* Ignore notify errors. * * @TODO It might be better to keep the correct result here, and * ignore it in process_graph_event(). */ op->op_status = PCMK_LRM_OP_DONE; op->rc = 0; } } else if (did_rsc_op_fail(op, target_rc)) { op_id = pcmk__op_key(op->rsc_id, "last_failure", 0); if (op->interval_ms == 0) { // Ensure 'last' gets updated, in case record-pending is true op_id_additional = pcmk__op_key(op->rsc_id, "last", 0); } exit_reason = op->exit_reason; } else if (op->interval_ms > 0) { op_id = strdup(key); } else { op_id = pcmk__op_key(op->rsc_id, "last", 0); } again: xml_op = pcmk__xe_match(parent, XML_LRM_TAG_RSC_OP, XML_ATTR_ID, op_id); if (xml_op == NULL) { xml_op = create_xml_node(parent, XML_LRM_TAG_RSC_OP); } if (op->user_data == NULL) { crm_debug("Generating fake transition key for: " PCMK__OP_FMT " %d from %s", op->rsc_id, op->op_type, op->interval_ms, op->call_id, origin); local_user_data = pcmk__transition_key(-1, op->call_id, target_rc, FAKE_TE_ID); op->user_data = local_user_data; } if(magic == NULL) { magic = generate_transition_magic(op->user_data, op->op_status, op->rc); } crm_xml_add(xml_op, XML_ATTR_ID, op_id); crm_xml_add(xml_op, XML_LRM_ATTR_TASK_KEY, key); crm_xml_add(xml_op, XML_LRM_ATTR_TASK, task); crm_xml_add(xml_op, XML_ATTR_ORIGIN, origin); crm_xml_add(xml_op, XML_ATTR_CRM_VERSION, caller_version); crm_xml_add(xml_op, XML_ATTR_TRANSITION_KEY, op->user_data); crm_xml_add(xml_op, XML_ATTR_TRANSITION_MAGIC, magic); crm_xml_add(xml_op, XML_LRM_ATTR_EXIT_REASON, exit_reason == NULL ? "" : exit_reason); crm_xml_add(xml_op, XML_LRM_ATTR_TARGET, node); /* For context during triage */ crm_xml_add_int(xml_op, XML_LRM_ATTR_CALLID, op->call_id); crm_xml_add_int(xml_op, XML_LRM_ATTR_RC, op->rc); crm_xml_add_int(xml_op, XML_LRM_ATTR_OPSTATUS, op->op_status); crm_xml_add_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, op->interval_ms); if (compare_version("2.1", caller_version) <= 0) { if (op->t_run || op->t_rcchange || op->exec_time || op->queue_time) { crm_trace("Timing data (" PCMK__OP_FMT "): last=%u change=%u exec=%u queue=%u", op->rsc_id, op->op_type, op->interval_ms, op->t_run, op->t_rcchange, op->exec_time, op->queue_time); if ((op->interval_ms != 0) && (op->t_rcchange != 0)) { // Recurring ops may have changed rc after initial run crm_xml_add_ll(xml_op, XML_RSC_OP_LAST_CHANGE, (long long) op->t_rcchange); } else { crm_xml_add_ll(xml_op, XML_RSC_OP_LAST_CHANGE, (long long) op->t_run); } crm_xml_add_int(xml_op, XML_RSC_OP_T_EXEC, op->exec_time); crm_xml_add_int(xml_op, XML_RSC_OP_T_QUEUE, op->queue_time); } } if (pcmk__str_any_of(op->op_type, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, NULL)) { /* * Record migrate_source and migrate_target always for migrate ops. */ const char *name = XML_LRM_ATTR_MIGRATE_SOURCE; crm_xml_add(xml_op, name, crm_meta_value(op->params, name)); name = XML_LRM_ATTR_MIGRATE_TARGET; crm_xml_add(xml_op, name, crm_meta_value(op->params, name)); } append_digest(op, xml_op, caller_version, magic, LOG_DEBUG); if (op_id_additional) { free(op_id); op_id = op_id_additional; op_id_additional = NULL; goto again; } if (local_user_data) { free(local_user_data); op->user_data = NULL; } free(magic); free(op_id); free(key); return xml_op; } pcmk__output_t * pcmk__new_logger(void) { int rc = pcmk_rc_ok; pcmk__output_t *out = NULL; const char* argv[] = { "", NULL }; pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_LOG, { NULL, NULL, NULL } }; pcmk__register_formats(NULL, formats); rc = pcmk__output_new(&out, "log", NULL, (char**)argv); if ((rc != pcmk_rc_ok) || (out == NULL)) { crm_err("Can't log resource details due to internal error: %s\n", pcmk_rc_str(rc)); return NULL; } pe__register_messages(out); pcmk__register_lib_messages(out); return out; } + +/*! + * \internal + * \brief Check whether a resource has reached its migration threshold on a node + * + * \param[in] rsc Resource to check + * \param[in] node Node to check + * \param[in] data_set Cluster working set + * \param[out] failed If the threshold has been reached, this will be set to + * the resource that failed (possibly a parent of \p rsc) + * + * \return true if the migration threshold has been reached, false otherwise + */ +bool +pcmk__threshold_reached(pe_resource_t *rsc, pe_node_t *node, + pe_working_set_t *data_set, pe_resource_t **failed) +{ + int fail_count, remaining_tries; + pe_resource_t *rsc_to_ban = rsc; + + // Migration threshold of 0 means never force away + if (rsc->migration_threshold == 0) { + return false; + } + + // If we're ignoring failures, also ignore the migration threshold + if (pcmk_is_set(rsc->flags, pe_rsc_failure_ignored)) { + return false; + } + + // If there are no failures, there's no need to force away + fail_count = pe_get_failcount(node, rsc, NULL, + pe_fc_effective|pe_fc_fillers, NULL, + data_set); + if (fail_count <= 0) { + return false; + } + + // If failed resource is anonymous clone instance, we'll force clone away + if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) { + rsc_to_ban = uber_parent(rsc); + } + + // How many more times recovery will be tried on this node + remaining_tries = rsc->migration_threshold - fail_count; + + if (remaining_tries <= 0) { + crm_warn("%s cannot run on %s due to reaching migration threshold " + "(clean up resource to allow again)" + CRM_XS " failures=%d migration-threshold=%d", + rsc_to_ban->id, node->details->uname, fail_count, + rsc->migration_threshold); + if (failed != NULL) { + *failed = rsc_to_ban; + } + return true; + } + + crm_info("%s can fail %d more time%s on " + "%s before reaching migration threshold (%d)", + rsc_to_ban->id, remaining_tries, pcmk__plural_s(remaining_tries), + node->details->uname, rsc->migration_threshold); + return false; +} diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c index 5b9b03e51c..5b56c37c1b 100644 --- a/lib/pengine/unpack.c +++ b/lib/pengine/unpack.c @@ -1,4103 +1,4104 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include CRM_TRACE_INIT_DATA(pe_status); /* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than * use pe__set_working_set_flags()/pe__clear_working_set_flags() so that the * flag is stringified more readably in log messages. */ #define set_config_flag(data_set, option, flag) do { \ const char *scf_value = pe_pref((data_set)->config_hash, (option)); \ if (scf_value != NULL) { \ if (crm_is_true(scf_value)) { \ (data_set)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Working set", \ crm_system_name, (data_set)->flags, \ (flag), #flag); \ } else { \ (data_set)->flags = pcmk__clear_flags_as(__func__, __LINE__,\ LOG_TRACE, "Working set", \ crm_system_name, (data_set)->flags, \ (flag), #flag); \ } \ } \ } while(0) static void unpack_rsc_op(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *failed, pe_working_set_t *data_set); static void determine_remote_online_status(pe_working_set_t *data_set, pe_node_t *this_node); static void add_node_attrs(xmlNode *attrs, pe_node_t *node, bool overwrite, pe_working_set_t *data_set); static void determine_online_status(xmlNode *node_state, pe_node_t *this_node, pe_working_set_t *data_set); static void unpack_node_lrm(pe_node_t *node, xmlNode *xml, pe_working_set_t *data_set); // Bitmask for warnings we only want to print once uint32_t pe_wo = 0; static gboolean is_dangling_guest_node(pe_node_t *node) { /* we are looking for a remote-node that was supposed to be mapped to a * container resource, but all traces of that container have disappeared * from both the config and the status section. */ if (pe__is_guest_or_remote_node(node) && node->details->remote_rsc && node->details->remote_rsc->container == NULL && pcmk_is_set(node->details->remote_rsc->flags, pe_rsc_orphan_container_filler)) { return TRUE; } return FALSE; } /*! * \brief Schedule a fence action for a node * * \param[in,out] data_set Current working set of cluster * \param[in,out] node Node to fence * \param[in] reason Text description of why fencing is needed * \param[in] priority_delay Whether to consider `priority-fencing-delay` */ void pe_fence_node(pe_working_set_t * data_set, pe_node_t * node, const char *reason, bool priority_delay) { CRM_CHECK(node, return); /* A guest node is fenced by marking its container as failed */ if (pe__is_guest_node(node)) { pe_resource_t *rsc = node->details->remote_rsc->container; if (!pcmk_is_set(rsc->flags, pe_rsc_failed)) { if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { crm_notice("Not fencing guest node %s " "(otherwise would because %s): " "its guest resource %s is unmanaged", node->details->uname, reason, rsc->id); } else { crm_warn("Guest node %s will be fenced " "(by recovering its guest resource %s): %s", node->details->uname, rsc->id, reason); /* We don't mark the node as unclean because that would prevent the * node from running resources. We want to allow it to run resources * in this transition if the recovery succeeds. */ node->details->remote_requires_reset = TRUE; pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); } } } else if (is_dangling_guest_node(node)) { crm_info("Cleaning up dangling connection for guest node %s: " "fencing was already done because %s, " "and guest resource no longer exists", node->details->uname, reason); pe__set_resource_flags(node->details->remote_rsc, pe_rsc_failed|pe_rsc_stop); } else if (pe__is_remote_node(node)) { pe_resource_t *rsc = node->details->remote_rsc; if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_managed)) { crm_notice("Not fencing remote node %s " "(otherwise would because %s): connection is unmanaged", node->details->uname, reason); } else if(node->details->remote_requires_reset == FALSE) { node->details->remote_requires_reset = TRUE; crm_warn("Remote node %s %s: %s", node->details->uname, pe_can_fence(data_set, node)? "will be fenced" : "is unclean", reason); } node->details->unclean = TRUE; // No need to apply `priority-fencing-delay` for remote nodes pe_fence_op(node, NULL, TRUE, reason, FALSE, data_set); } else if (node->details->unclean) { crm_trace("Cluster node %s %s because %s", node->details->uname, pe_can_fence(data_set, node)? "would also be fenced" : "also is unclean", reason); } else { crm_warn("Cluster node %s %s: %s", node->details->uname, pe_can_fence(data_set, node)? "will be fenced" : "is unclean", reason); node->details->unclean = TRUE; pe_fence_op(node, NULL, TRUE, reason, priority_delay, data_set); } } // @TODO xpaths can't handle templates, rules, or id-refs // nvpair with provides or requires set to unfencing #define XPATH_UNFENCING_NVPAIR XML_CIB_TAG_NVPAIR \ "[(@" XML_NVPAIR_ATTR_NAME "='" PCMK_STONITH_PROVIDES "'" \ "or @" XML_NVPAIR_ATTR_NAME "='" XML_RSC_ATTR_REQUIRES "') " \ "and @" XML_NVPAIR_ATTR_VALUE "='unfencing']" // unfencing in rsc_defaults or any resource #define XPATH_ENABLE_UNFENCING \ "/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RESOURCES \ "//" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR \ "|/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RSCCONFIG \ "/" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR static void set_if_xpath(uint64_t flag, const char *xpath, pe_working_set_t *data_set) { xmlXPathObjectPtr result = NULL; if (!pcmk_is_set(data_set->flags, flag)) { result = xpath_search(data_set->input, xpath); if (result && (numXpathResults(result) > 0)) { pe__set_working_set_flags(data_set, flag); } freeXpathObject(result); } } gboolean unpack_config(xmlNode * config, pe_working_set_t * data_set) { const char *value = NULL; GHashTable *config_hash = pcmk__strkey_table(free, free); pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; data_set->config_hash = config_hash; pe__unpack_dataset_nvpairs(config, XML_CIB_TAG_PROPSET, &rule_data, config_hash, CIB_OPTIONS_FIRST, FALSE, data_set); verify_pe_options(data_set->config_hash); set_config_flag(data_set, "enable-startup-probes", pe_flag_startup_probes); if (!pcmk_is_set(data_set->flags, pe_flag_startup_probes)) { crm_info("Startup probes: disabled (dangerous)"); } value = pe_pref(data_set->config_hash, XML_ATTR_HAVE_WATCHDOG); if (value && crm_is_true(value)) { crm_info("Watchdog-based self-fencing will be performed via SBD if " "fencing is required and stonith-watchdog-timeout is nonzero"); pe__set_working_set_flags(data_set, pe_flag_have_stonith_resource); } /* Set certain flags via xpath here, so they can be used before the relevant * configuration sections are unpacked. */ set_if_xpath(pe_flag_enable_unfencing, XPATH_ENABLE_UNFENCING, data_set); value = pe_pref(data_set->config_hash, "stonith-timeout"); data_set->stonith_timeout = (int) crm_parse_interval_spec(value); crm_debug("STONITH timeout: %d", data_set->stonith_timeout); set_config_flag(data_set, "stonith-enabled", pe_flag_stonith_enabled); crm_debug("STONITH of failed nodes is %s", pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)? "enabled" : "disabled"); data_set->stonith_action = pe_pref(data_set->config_hash, "stonith-action"); if (!strcmp(data_set->stonith_action, "poweroff")) { pe_warn_once(pe_wo_poweroff, "Support for stonith-action of 'poweroff' is deprecated " "and will be removed in a future release (use 'off' instead)"); data_set->stonith_action = "off"; } crm_trace("STONITH will %s nodes", data_set->stonith_action); set_config_flag(data_set, "concurrent-fencing", pe_flag_concurrent_fencing); crm_debug("Concurrent fencing is %s", pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)? "enabled" : "disabled"); value = pe_pref(data_set->config_hash, XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY); if (value) { data_set->priority_fencing_delay = crm_parse_interval_spec(value) / 1000; crm_trace("Priority fencing delay is %ds", data_set->priority_fencing_delay); } set_config_flag(data_set, "stop-all-resources", pe_flag_stop_everything); crm_debug("Stop all active resources: %s", pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_stop_everything))); set_config_flag(data_set, "symmetric-cluster", pe_flag_symmetric_cluster); if (pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster)) { crm_debug("Cluster is symmetric" " - resources can run anywhere by default"); } value = pe_pref(data_set->config_hash, "no-quorum-policy"); if (pcmk__str_eq(value, "ignore", pcmk__str_casei)) { data_set->no_quorum_policy = no_quorum_ignore; } else if (pcmk__str_eq(value, "freeze", pcmk__str_casei)) { data_set->no_quorum_policy = no_quorum_freeze; } else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) { data_set->no_quorum_policy = no_quorum_demote; } else if (pcmk__str_eq(value, "suicide", pcmk__str_casei)) { if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { int do_panic = 0; crm_element_value_int(data_set->input, XML_ATTR_QUORUM_PANIC, &do_panic); if (do_panic || pcmk_is_set(data_set->flags, pe_flag_have_quorum)) { data_set->no_quorum_policy = no_quorum_suicide; } else { crm_notice("Resetting no-quorum-policy to 'stop': cluster has never had quorum"); data_set->no_quorum_policy = no_quorum_stop; } } else { pcmk__config_err("Resetting no-quorum-policy to 'stop' because " "fencing is disabled"); data_set->no_quorum_policy = no_quorum_stop; } } else { data_set->no_quorum_policy = no_quorum_stop; } switch (data_set->no_quorum_policy) { case no_quorum_freeze: crm_debug("On loss of quorum: Freeze resources"); break; case no_quorum_stop: crm_debug("On loss of quorum: Stop ALL resources"); break; case no_quorum_demote: crm_debug("On loss of quorum: " "Demote promotable resources and stop other resources"); break; case no_quorum_suicide: crm_notice("On loss of quorum: Fence all remaining nodes"); break; case no_quorum_ignore: crm_notice("On loss of quorum: Ignore"); break; } set_config_flag(data_set, "stop-orphan-resources", pe_flag_stop_rsc_orphans); crm_trace("Orphan resources are %s", pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)? "stopped" : "ignored"); set_config_flag(data_set, "stop-orphan-actions", pe_flag_stop_action_orphans); crm_trace("Orphan resource actions are %s", pcmk_is_set(data_set->flags, pe_flag_stop_action_orphans)? "stopped" : "ignored"); value = pe_pref(data_set->config_hash, "remove-after-stop"); if (value != NULL) { if (crm_is_true(value)) { pe__set_working_set_flags(data_set, pe_flag_remove_after_stop); #ifndef PCMK__COMPAT_2_0 pe_warn_once(pe_wo_remove_after, "Support for the remove-after-stop cluster property is" " deprecated and will be removed in a future release"); #endif } else { pe__clear_working_set_flags(data_set, pe_flag_remove_after_stop); } } set_config_flag(data_set, "maintenance-mode", pe_flag_maintenance_mode); crm_trace("Maintenance mode: %s", pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_maintenance_mode))); set_config_flag(data_set, "start-failure-is-fatal", pe_flag_start_failure_fatal); crm_trace("Start failures are %s", pcmk_is_set(data_set->flags, pe_flag_start_failure_fatal)? "always fatal" : "handled by failcount"); if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { set_config_flag(data_set, "startup-fencing", pe_flag_startup_fencing); } if (pcmk_is_set(data_set->flags, pe_flag_startup_fencing)) { crm_trace("Unseen nodes will be fenced"); } else { pe_warn_once(pe_wo_blind, "Blind faith: not fencing unseen nodes"); } pcmk__score_red = char2score(pe_pref(data_set->config_hash, "node-health-red")); pcmk__score_green = char2score(pe_pref(data_set->config_hash, "node-health-green")); pcmk__score_yellow = char2score(pe_pref(data_set->config_hash, "node-health-yellow")); crm_debug("Node scores: 'red' = %s, 'yellow' = %s, 'green' = %s", pe_pref(data_set->config_hash, "node-health-red"), pe_pref(data_set->config_hash, "node-health-yellow"), pe_pref(data_set->config_hash, "node-health-green")); data_set->placement_strategy = pe_pref(data_set->config_hash, "placement-strategy"); crm_trace("Placement strategy: %s", data_set->placement_strategy); set_config_flag(data_set, "shutdown-lock", pe_flag_shutdown_lock); crm_trace("Resources will%s be locked to cleanly shut down nodes", (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)? "" : " not")); if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { value = pe_pref(data_set->config_hash, XML_CONFIG_ATTR_SHUTDOWN_LOCK_LIMIT); data_set->shutdown_lock = crm_parse_interval_spec(value) / 1000; crm_trace("Shutdown locks expire after %us", data_set->shutdown_lock); } return TRUE; } pe_node_t * pe_create_node(const char *id, const char *uname, const char *type, const char *score, pe_working_set_t * data_set) { pe_node_t *new_node = NULL; if (pe_find_node(data_set->nodes, uname) != NULL) { pcmk__config_warn("More than one node entry has name '%s'", uname); } new_node = calloc(1, sizeof(pe_node_t)); if (new_node == NULL) { return NULL; } new_node->weight = char2score(score); new_node->fixed = FALSE; new_node->details = calloc(1, sizeof(struct pe_node_shared_s)); if (new_node->details == NULL) { free(new_node); return NULL; } crm_trace("Creating node for entry %s/%s", uname, id); new_node->details->id = id; new_node->details->uname = uname; new_node->details->online = FALSE; new_node->details->shutdown = FALSE; new_node->details->rsc_discovery_enabled = TRUE; new_node->details->running_rsc = NULL; new_node->details->type = node_ping; if (pcmk__str_eq(type, "remote", pcmk__str_casei)) { new_node->details->type = node_remote; pe__set_working_set_flags(data_set, pe_flag_have_remote_nodes); } else if (pcmk__str_eq(type, "member", pcmk__str_null_matches | pcmk__str_casei)) { new_node->details->type = node_member; } new_node->details->attrs = pcmk__strkey_table(free, free); if (pe__is_guest_or_remote_node(new_node)) { g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("remote")); } else { g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("cluster")); } new_node->details->utilization = pcmk__strkey_table(free, free); new_node->details->digest_cache = pcmk__strkey_table(free, pe__free_digests); data_set->nodes = g_list_insert_sorted(data_set->nodes, new_node, sort_node_uname); return new_node; } static const char * expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pe_working_set_t *data) { xmlNode *attr_set = NULL; xmlNode *attr = NULL; const char *container_id = ID(xml_obj); const char *remote_name = NULL; const char *remote_server = NULL; const char *remote_port = NULL; const char *connect_timeout = "60s"; const char *remote_allow_migrate=NULL; const char *is_managed = NULL; for (attr_set = pcmk__xe_first_child(xml_obj); attr_set != NULL; attr_set = pcmk__xe_next(attr_set)) { if (!pcmk__str_eq((const char *)attr_set->name, XML_TAG_META_SETS, pcmk__str_casei)) { continue; } for (attr = pcmk__xe_first_child(attr_set); attr != NULL; attr = pcmk__xe_next(attr)) { const char *value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE); const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME); if (pcmk__str_eq(name, XML_RSC_ATTR_REMOTE_NODE, pcmk__str_casei)) { remote_name = value; } else if (pcmk__str_eq(name, "remote-addr", pcmk__str_casei)) { remote_server = value; } else if (pcmk__str_eq(name, "remote-port", pcmk__str_casei)) { remote_port = value; } else if (pcmk__str_eq(name, "remote-connect-timeout", pcmk__str_casei)) { connect_timeout = value; } else if (pcmk__str_eq(name, "remote-allow-migrate", pcmk__str_casei)) { remote_allow_migrate=value; } else if (pcmk__str_eq(name, XML_RSC_ATTR_MANAGED, pcmk__str_casei)) { is_managed = value; } } } if (remote_name == NULL) { return NULL; } if (pe_find_resource(data->resources, remote_name) != NULL) { return NULL; } pe_create_remote_xml(parent, remote_name, container_id, remote_allow_migrate, is_managed, connect_timeout, remote_server, remote_port); return remote_name; } static void handle_startup_fencing(pe_working_set_t *data_set, pe_node_t *new_node) { if ((new_node->details->type == node_remote) && (new_node->details->remote_rsc == NULL)) { /* Ignore fencing for remote nodes that don't have a connection resource * associated with them. This happens when remote node entries get left * in the nodes section after the connection resource is removed. */ return; } if (pcmk_is_set(data_set->flags, pe_flag_startup_fencing)) { // All nodes are unclean until we've seen their status entry new_node->details->unclean = TRUE; } else { // Blind faith ... new_node->details->unclean = FALSE; } /* We need to be able to determine if a node's status section * exists or not separate from whether the node is unclean. */ new_node->details->unseen = TRUE; } gboolean unpack_nodes(xmlNode * xml_nodes, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; pe_node_t *new_node = NULL; const char *id = NULL; const char *uname = NULL; const char *type = NULL; const char *score = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; for (xml_obj = pcmk__xe_first_child(xml_nodes); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_NODE, pcmk__str_none)) { new_node = NULL; id = crm_element_value(xml_obj, XML_ATTR_ID); uname = crm_element_value(xml_obj, XML_ATTR_UNAME); type = crm_element_value(xml_obj, XML_ATTR_TYPE); score = crm_element_value(xml_obj, XML_RULE_ATTR_SCORE); crm_trace("Processing node %s/%s", uname, id); if (id == NULL) { pcmk__config_err("Ignoring <" XML_CIB_TAG_NODE "> entry in configuration without id"); continue; } new_node = pe_create_node(id, uname, type, score, data_set); if (new_node == NULL) { return FALSE; } /* if(data_set->have_quorum == FALSE */ /* && data_set->no_quorum_policy == no_quorum_stop) { */ /* /\* start shutting resources down *\/ */ /* new_node->weight = -INFINITY; */ /* } */ handle_startup_fencing(data_set, new_node); add_node_attrs(xml_obj, new_node, FALSE, data_set); pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_UTILIZATION, &rule_data, new_node->details->utilization, NULL, FALSE, data_set); crm_trace("Done with node %s", crm_element_value(xml_obj, XML_ATTR_UNAME)); } } if (data_set->localhost && pe_find_node(data_set->nodes, data_set->localhost) == NULL) { crm_info("Creating a fake local node"); pe_create_node(data_set->localhost, data_set->localhost, NULL, 0, data_set); } return TRUE; } static void setup_container(pe_resource_t * rsc, pe_working_set_t * data_set) { const char *container_id = NULL; if (rsc->children) { g_list_foreach(rsc->children, (GFunc) setup_container, data_set); return; } container_id = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_CONTAINER); if (container_id && !pcmk__str_eq(container_id, rsc->id, pcmk__str_casei)) { pe_resource_t *container = pe_find_resource(data_set->resources, container_id); if (container) { rsc->container = container; pe__set_resource_flags(container, pe_rsc_is_container); container->fillers = g_list_append(container->fillers, rsc); pe_rsc_trace(rsc, "Resource %s's container is %s", rsc->id, container_id); } else { pe_err("Resource %s: Unknown resource container (%s)", rsc->id, container_id); } } } gboolean unpack_remote_nodes(xmlNode * xml_resources, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; /* Create remote nodes and guest nodes from the resource configuration * before unpacking resources. */ for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { const char *new_node_id = NULL; /* Check for remote nodes, which are defined by ocf:pacemaker:remote * primitives. */ if (xml_contains_remote_node(xml_obj)) { new_node_id = ID(xml_obj); /* The "pe_find_node" check is here to make sure we don't iterate over * an expanded node that has already been added to the node list. */ if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found remote node %s defined by resource %s", new_node_id, ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } continue; } /* Check for guest nodes, which are defined by special meta-attributes * of a primitive of any type (for example, VirtualDomain or Xen). */ if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_RESOURCE, pcmk__str_none)) { /* This will add an ocf:pacemaker:remote primitive to the * configuration for the guest node's connection, to be unpacked * later. */ new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources, data_set); if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found guest node %s in resource %s", new_node_id, ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } continue; } /* Check for guest nodes inside a group. Clones are currently not * supported as guest nodes. */ if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_GROUP, pcmk__str_none)) { xmlNode *xml_obj2 = NULL; for (xml_obj2 = pcmk__xe_first_child(xml_obj); xml_obj2 != NULL; xml_obj2 = pcmk__xe_next(xml_obj2)) { new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources, data_set); if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found guest node %s in resource %s inside group %s", new_node_id, ID(xml_obj2), ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } } } } return TRUE; } /* Call this after all the nodes and resources have been * unpacked, but before the status section is read. * * A remote node's online status is reflected by the state * of the remote node's connection resource. We need to link * the remote node to this connection resource so we can have * easy access to the connection resource during the scheduler calculations. */ static void link_rsc2remotenode(pe_working_set_t *data_set, pe_resource_t *new_rsc) { pe_node_t *remote_node = NULL; if (new_rsc->is_remote_node == FALSE) { return; } if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) { /* remote_nodes and remote_resources are not linked in quick location calculations */ return; } remote_node = pe_find_node(data_set->nodes, new_rsc->id); CRM_CHECK(remote_node != NULL, return;); pe_rsc_trace(new_rsc, "Linking remote connection resource %s to node %s", new_rsc->id, remote_node->details->uname); remote_node->details->remote_rsc = new_rsc; if (new_rsc->container == NULL) { /* Handle start-up fencing for remote nodes (as opposed to guest nodes) * the same as is done for cluster nodes. */ handle_startup_fencing(data_set, remote_node); } else { /* pe_create_node() marks the new node as "remote" or "cluster"; now * that we know the node is a guest node, update it correctly. */ g_hash_table_replace(remote_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("container")); } } static void destroy_tag(gpointer data) { pe_tag_t *tag = data; if (tag) { free(tag->id); g_list_free_full(tag->refs, free); free(tag); } } /*! * \internal * \brief Parse configuration XML for resource information * * \param[in] xml_resources Top of resource configuration XML * \param[in,out] data_set Where to put resource information * * \return TRUE * * \note unpack_remote_nodes() MUST be called before this, so that the nodes can * be used when common_unpack() calls resource_location() */ gboolean unpack_resources(xmlNode * xml_resources, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; GList *gIter = NULL; data_set->template_rsc_sets = pcmk__strkey_table(free, destroy_tag); for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { pe_resource_t *new_rsc = NULL; if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_RSC_TEMPLATE, pcmk__str_none)) { const char *template_id = ID(xml_obj); if (template_id && g_hash_table_lookup_extended(data_set->template_rsc_sets, template_id, NULL, NULL) == FALSE) { /* Record the template's ID for the knowledge of its existence anyway. */ g_hash_table_insert(data_set->template_rsc_sets, strdup(template_id), NULL); } continue; } crm_trace("Beginning unpack... <%s id=%s... >", crm_element_name(xml_obj), ID(xml_obj)); if (common_unpack(xml_obj, &new_rsc, NULL, data_set) && (new_rsc != NULL)) { data_set->resources = g_list_append(data_set->resources, new_rsc); pe_rsc_trace(new_rsc, "Added resource %s", new_rsc->id); } else { pcmk__config_err("Ignoring <%s> resource '%s' " "because configuration is invalid", crm_element_name(xml_obj), crm_str(ID(xml_obj))); if (new_rsc != NULL && new_rsc->fns != NULL) { new_rsc->fns->free(new_rsc); } } } for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; setup_container(rsc, data_set); link_rsc2remotenode(data_set, rsc); } data_set->resources = g_list_sort(data_set->resources, sort_rsc_priority); if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) { /* Ignore */ } else if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled) && !pcmk_is_set(data_set->flags, pe_flag_have_stonith_resource)) { pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined"); pcmk__config_err("Either configure some or disable STONITH with the stonith-enabled option"); pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity"); } return TRUE; } gboolean unpack_tags(xmlNode * xml_tags, pe_working_set_t * data_set) { xmlNode *xml_tag = NULL; data_set->tags = pcmk__strkey_table(free, destroy_tag); for (xml_tag = pcmk__xe_first_child(xml_tags); xml_tag != NULL; xml_tag = pcmk__xe_next(xml_tag)) { xmlNode *xml_obj_ref = NULL; const char *tag_id = ID(xml_tag); if (!pcmk__str_eq((const char *)xml_tag->name, XML_CIB_TAG_TAG, pcmk__str_none)) { continue; } if (tag_id == NULL) { pcmk__config_err("Ignoring <%s> without " XML_ATTR_ID, crm_element_name(xml_tag)); continue; } for (xml_obj_ref = pcmk__xe_first_child(xml_tag); xml_obj_ref != NULL; xml_obj_ref = pcmk__xe_next(xml_obj_ref)) { const char *obj_ref = ID(xml_obj_ref); if (!pcmk__str_eq((const char *)xml_obj_ref->name, XML_CIB_TAG_OBJ_REF, pcmk__str_none)) { continue; } if (obj_ref == NULL) { pcmk__config_err("Ignoring <%s> for tag '%s' without " XML_ATTR_ID, crm_element_name(xml_obj_ref), tag_id); continue; } if (add_tag_ref(data_set->tags, tag_id, obj_ref) == FALSE) { return FALSE; } } } return TRUE; } /* The ticket state section: * "/cib/status/tickets/ticket_state" */ static gboolean unpack_ticket_state(xmlNode * xml_ticket, pe_working_set_t * data_set) { const char *ticket_id = NULL; const char *granted = NULL; const char *last_granted = NULL; const char *standby = NULL; xmlAttrPtr xIter = NULL; pe_ticket_t *ticket = NULL; ticket_id = ID(xml_ticket); if (pcmk__str_empty(ticket_id)) { return FALSE; } crm_trace("Processing ticket state for %s", ticket_id); ticket = g_hash_table_lookup(data_set->tickets, ticket_id); if (ticket == NULL) { ticket = ticket_new(ticket_id, data_set); if (ticket == NULL) { return FALSE; } } for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = crm_element_value(xml_ticket, prop_name); if (pcmk__str_eq(prop_name, XML_ATTR_ID, pcmk__str_none)) { continue; } g_hash_table_replace(ticket->state, strdup(prop_name), strdup(prop_value)); } granted = g_hash_table_lookup(ticket->state, "granted"); if (granted && crm_is_true(granted)) { ticket->granted = TRUE; crm_info("We have ticket '%s'", ticket->id); } else { ticket->granted = FALSE; crm_info("We do not have ticket '%s'", ticket->id); } last_granted = g_hash_table_lookup(ticket->state, "last-granted"); if (last_granted) { long long last_granted_ll; pcmk__scan_ll(last_granted, &last_granted_ll, 0LL); ticket->last_granted = (time_t) last_granted_ll; } standby = g_hash_table_lookup(ticket->state, "standby"); if (standby && crm_is_true(standby)) { ticket->standby = TRUE; if (ticket->granted) { crm_info("Granted ticket '%s' is in standby-mode", ticket->id); } } else { ticket->standby = FALSE; } crm_trace("Done with ticket state for %s", ticket_id); return TRUE; } static gboolean unpack_tickets_state(xmlNode * xml_tickets, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; for (xml_obj = pcmk__xe_first_child(xml_tickets); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (!pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_TICKET_STATE, pcmk__str_none)) { continue; } unpack_ticket_state(xml_obj, data_set); } return TRUE; } static void unpack_handle_remote_attrs(pe_node_t *this_node, xmlNode *state, pe_working_set_t * data_set) { const char *resource_discovery_enabled = NULL; xmlNode *attrs = NULL; pe_resource_t *rsc = NULL; if (!pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { return; } if ((this_node == NULL) || !pe__is_guest_or_remote_node(this_node)) { return; } crm_trace("Processing remote node id=%s, uname=%s", this_node->details->id, this_node->details->uname); pcmk__scan_min_int(crm_element_value(state, XML_NODE_IS_MAINTENANCE), &(this_node->details->remote_maintenance), 0); rsc = this_node->details->remote_rsc; if (this_node->details->remote_requires_reset == FALSE) { this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; } attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE); add_node_attrs(attrs, this_node, TRUE, data_set); if (pe__shutdown_requested(this_node)) { crm_info("Node %s is shutting down", this_node->details->uname); this_node->details->shutdown = TRUE; } if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) { crm_info("Node %s is in standby-mode", this_node->details->uname); this_node->details->standby = TRUE; } if (crm_is_true(pe_node_attribute_raw(this_node, "maintenance")) || ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_managed))) { crm_info("Node %s is in maintenance-mode", this_node->details->uname); this_node->details->maintenance = TRUE; } resource_discovery_enabled = pe_node_attribute_raw(this_node, XML_NODE_ATTR_RSC_DISCOVERY); if (resource_discovery_enabled && !crm_is_true(resource_discovery_enabled)) { if (pe__is_remote_node(this_node) && !pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { crm_warn("Ignoring %s attribute on remote node %s because stonith is disabled", XML_NODE_ATTR_RSC_DISCOVERY, this_node->details->uname); } else { /* This is either a remote node with fencing enabled, or a guest * node. We don't care whether fencing is enabled when fencing guest * nodes, because they are "fenced" by recovering their containing * resource. */ crm_info("Node %s has resource discovery disabled", this_node->details->uname); this_node->details->rsc_discovery_enabled = FALSE; } } } /*! * \internal * \brief Unpack a cluster node's transient attributes * * \param[in] state CIB node state XML * \param[in] node Cluster node whose attributes are being unpacked * \param[in] data_set Cluster working set */ static void unpack_transient_attributes(xmlNode *state, pe_node_t *node, pe_working_set_t *data_set) { const char *discovery = NULL; xmlNode *attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE); add_node_attrs(attrs, node, TRUE, data_set); if (crm_is_true(pe_node_attribute_raw(node, "standby"))) { crm_info("Node %s is in standby-mode", node->details->uname); node->details->standby = TRUE; } if (crm_is_true(pe_node_attribute_raw(node, "maintenance"))) { crm_info("Node %s is in maintenance-mode", node->details->uname); node->details->maintenance = TRUE; } discovery = pe_node_attribute_raw(node, XML_NODE_ATTR_RSC_DISCOVERY); if ((discovery != NULL) && !crm_is_true(discovery)) { crm_warn("Ignoring %s attribute for node %s because disabling " "resource discovery is not allowed for cluster nodes", XML_NODE_ATTR_RSC_DISCOVERY, node->details->uname); } } /*! * \internal * \brief Unpack a node state entry (first pass) * * Unpack one node state entry from status. This unpacks information from the * node_state element itself and node attributes inside it, but not the * resource history inside it. Multiple passes through the status are needed to * fully unpack everything. * * \param[in] state CIB node state XML * \param[in] data_set Cluster working set */ static void unpack_node_state(xmlNode *state, pe_working_set_t *data_set) { const char *id = NULL; const char *uname = NULL; pe_node_t *this_node = NULL; id = crm_element_value(state, XML_ATTR_ID); if (id == NULL) { crm_warn("Ignoring malformed " XML_CIB_TAG_STATE " entry without " XML_ATTR_ID); return; } uname = crm_element_value(state, XML_ATTR_UNAME); if (uname == NULL) { crm_warn("Ignoring malformed " XML_CIB_TAG_STATE " entry without " XML_ATTR_UNAME); return; } this_node = pe_find_node_any(data_set->nodes, id, uname); if (this_node == NULL) { pcmk__config_warn("Ignoring recorded node state for '%s' because " "it is no longer in the configuration", uname); return; } if (pe__is_guest_or_remote_node(this_node)) { /* We can't determine the online status of Pacemaker Remote nodes until * after all resource history has been unpacked. In this first pass, we * do need to mark whether the node has been fenced, as this plays a * role during unpacking cluster node resource state. */ pcmk__scan_min_int(crm_element_value(state, XML_NODE_IS_FENCED), &(this_node->details->remote_was_fenced), 0); return; } unpack_transient_attributes(state, this_node, data_set); /* Provisionally mark this cluster node as clean. We have at least seen it * in the current cluster's lifetime. */ this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; crm_trace("Determining online status of cluster node %s (id %s)", this_node->details->uname, id); determine_online_status(state, this_node, data_set); if (!pcmk_is_set(data_set->flags, pe_flag_have_quorum) && this_node->details->online && (data_set->no_quorum_policy == no_quorum_suicide)) { /* Everything else should flow from this automatically * (at least until the scheduler becomes able to migrate off * healthy resources) */ pe_fence_node(data_set, this_node, "cluster does not have quorum", FALSE); } } /*! * \internal * \brief Unpack nodes' resource history as much as possible * * Unpack as many nodes' resource history as possible in one pass through the * status. We need to process Pacemaker Remote nodes' connections/containers * before unpacking their history; the connection/container history will be * in another node's history, so it might take multiple passes to unpack * everything. * * \param[in] status CIB XML status section * \param[in] fence If true, treat any not-yet-unpacked nodes as unseen * \param[in] data_set Cluster working set * * \return Standard Pacemaker return code (specifically pcmk_rc_ok if done, * or EAGAIN if more unpacking remains to be done) */ static int unpack_node_history(xmlNode *status, bool fence, pe_working_set_t *data_set) { int rc = pcmk_rc_ok; // Loop through all node_state entries in CIB status for (xmlNode *state = first_named_child(status, XML_CIB_TAG_STATE); state != NULL; state = crm_next_same_xml(state)) { const char *id = ID(state); const char *uname = crm_element_value(state, XML_ATTR_UNAME); pe_node_t *this_node = NULL; if ((id == NULL) || (uname == NULL)) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history from malformed " XML_CIB_TAG_STATE " without id and/or uname"); continue; } this_node = pe_find_node_any(data_set->nodes, id, uname); if (this_node == NULL) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history for node %s because " "no longer in configuration", id); continue; } if (this_node->details->unpacked) { crm_trace("Not unpacking resource history for node %s because " "already unpacked", id); continue; } if (fence) { // We're processing all remaining nodes } else if (pe__is_guest_node(this_node)) { /* We can unpack a guest node's history only after we've unpacked * other resource history to the point that we know that the node's * connection and containing resource are both up. */ pe_resource_t *rsc = this_node->details->remote_rsc; if ((rsc == NULL) || (rsc->role != RSC_ROLE_STARTED) || (rsc->container->role != RSC_ROLE_STARTED)) { crm_trace("Not unpacking resource history for guest node %s " "because container and connection are not known to " "be up", id); continue; } } else if (pe__is_remote_node(this_node)) { /* We can unpack a remote node's history only after we've unpacked * other resource history to the point that we know that the node's * connection is up, with the exception of when shutdown locks are * in use. */ pe_resource_t *rsc = this_node->details->remote_rsc; if ((rsc == NULL) || (!pcmk_is_set(data_set->flags, pe_flag_shutdown_lock) && (rsc->role != RSC_ROLE_STARTED))) { crm_trace("Not unpacking resource history for remote node %s " "because connection is not known to be up", id); continue; } /* If fencing and shutdown locks are disabled and we're not processing * unseen nodes, then we don't want to unpack offline nodes until online * nodes have been unpacked. This allows us to number active clone * instances first. */ } else if (!pcmk_any_flags_set(data_set->flags, pe_flag_stonith_enabled |pe_flag_shutdown_lock) && !this_node->details->online) { crm_trace("Not unpacking resource history for offline " "cluster node %s", id); continue; } if (pe__is_guest_or_remote_node(this_node)) { determine_remote_online_status(data_set, this_node); unpack_handle_remote_attrs(this_node, state, data_set); } crm_trace("Unpacking resource history for %snode %s", (fence? "unseen " : ""), id); this_node->details->unpacked = TRUE; unpack_node_lrm(this_node, state, data_set); rc = EAGAIN; // Other node histories might depend on this one } return rc; } /* remove nodes that are down, stopping */ /* create positive rsc_to_node constraints between resources and the nodes they are running on */ /* anything else? */ gboolean unpack_status(xmlNode * status, pe_working_set_t * data_set) { xmlNode *state = NULL; crm_trace("Beginning unpack"); if (data_set->tickets == NULL) { data_set->tickets = pcmk__strkey_table(free, destroy_ticket); } for (state = pcmk__xe_first_child(status); state != NULL; state = pcmk__xe_next(state)) { if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_TICKETS, pcmk__str_none)) { unpack_tickets_state((xmlNode *) state, data_set); } else if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { unpack_node_state(state, data_set); } } while (unpack_node_history(status, FALSE, data_set) == EAGAIN) { crm_trace("Another pass through node resource histories is needed"); } // Now catch any nodes we didn't see unpack_node_history(status, pcmk_is_set(data_set->flags, pe_flag_stonith_enabled), data_set); /* Now that we know where resources are, we can schedule stops of containers * with failed bundle connections */ if (data_set->stop_needed != NULL) { for (GList *item = data_set->stop_needed; item; item = item->next) { pe_resource_t *container = item->data; pe_node_t *node = pe__current_node(container); if (node) { stop_action(container, node, FALSE); } } g_list_free(data_set->stop_needed); data_set->stop_needed = NULL; } /* Now that we know status of all Pacemaker Remote connections and nodes, * we can stop connections for node shutdowns, and check the online status * of remote/guest nodes that didn't have any node history to unpack. */ for (GList *gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *this_node = gIter->data; if (!pe__is_guest_or_remote_node(this_node)) { continue; } if (this_node->details->shutdown && (this_node->details->remote_rsc != NULL)) { pe__set_next_role(this_node->details->remote_rsc, RSC_ROLE_STOPPED, "remote shutdown"); } if (!this_node->details->unpacked) { determine_remote_online_status(data_set, this_node); } } return TRUE; } static gboolean determine_online_status_no_fencing(pe_working_set_t * data_set, xmlNode * node_state, pe_node_t * this_node) { gboolean online = FALSE; const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE); const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER); const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER); const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); if (!crm_is_true(in_cluster)) { crm_trace("Node is down: in_cluster=%s", crm_str(in_cluster)); } else if (pcmk__str_eq(is_peer, ONLINESTATUS, pcmk__str_casei)) { if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { online = TRUE; } else { crm_debug("Node is not ready to run resources: %s", join); } } else if (this_node->details->expected_up == FALSE) { crm_trace("Controller is down: in_cluster=%s", crm_str(in_cluster)); crm_trace("\tis_peer=%s, join=%s, expected=%s", crm_str(is_peer), crm_str(join), crm_str(exp_state)); } else { /* mark it unclean */ pe_fence_node(data_set, this_node, "peer is unexpectedly down", FALSE); crm_info("\tin_cluster=%s, is_peer=%s, join=%s, expected=%s", crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state)); } return online; } static gboolean determine_online_status_fencing(pe_working_set_t * data_set, xmlNode * node_state, pe_node_t * this_node) { gboolean online = FALSE; gboolean do_terminate = FALSE; bool crmd_online = FALSE; const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE); const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER); const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER); const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); const char *terminate = pe_node_attribute_raw(this_node, "terminate"); /* - XML_NODE_IN_CLUSTER ::= true|false - XML_NODE_IS_PEER ::= online|offline - XML_NODE_JOIN_STATE ::= member|down|pending|banned - XML_NODE_EXPECTED ::= member|down */ if (crm_is_true(terminate)) { do_terminate = TRUE; } else if (terminate != NULL && strlen(terminate) > 0) { /* could be a time() value */ char t = terminate[0]; if (t != '0' && isdigit(t)) { do_terminate = TRUE; } } crm_trace("%s: in_cluster=%s, is_peer=%s, join=%s, expected=%s, term=%d", this_node->details->uname, crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state), do_terminate); online = crm_is_true(in_cluster); crmd_online = pcmk__str_eq(is_peer, ONLINESTATUS, pcmk__str_casei); if (exp_state == NULL) { exp_state = CRMD_JOINSTATE_DOWN; } if (this_node->details->shutdown) { crm_debug("%s is shutting down", this_node->details->uname); /* Slightly different criteria since we can't shut down a dead peer */ online = crmd_online; } else if (in_cluster == NULL) { pe_fence_node(data_set, this_node, "peer has not been seen by the cluster", FALSE); } else if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_casei)) { - pe_fence_node(data_set, this_node, "peer failed the pacemaker membership criteria", FALSE); + pe_fence_node(data_set, this_node, + "peer failed Pacemaker membership criteria", FALSE); } else if (do_terminate == FALSE && pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN, pcmk__str_casei)) { if (crm_is_true(in_cluster) || crmd_online) { crm_info("- Node %s is not ready to run resources", this_node->details->uname); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { crm_trace("%s is down or still coming up", this_node->details->uname); } } else if (do_terminate && pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_casei) && crm_is_true(in_cluster) == FALSE && !crmd_online) { crm_info("Node %s was just shot", this_node->details->uname); online = FALSE; } else if (crm_is_true(in_cluster) == FALSE) { // Consider `priority-fencing-delay` for lost nodes pe_fence_node(data_set, this_node, "peer is no longer part of the cluster", TRUE); } else if (!crmd_online) { pe_fence_node(data_set, this_node, "peer process is no longer available", FALSE); /* Everything is running at this point, now check join state */ } else if (do_terminate) { pe_fence_node(data_set, this_node, "termination was requested", FALSE); } else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { crm_info("Node %s is active", this_node->details->uname); } else if (pcmk__strcase_any_of(join, CRMD_JOINSTATE_PENDING, CRMD_JOINSTATE_DOWN, NULL)) { crm_info("Node %s is not ready to run resources", this_node->details->uname); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { pe_fence_node(data_set, this_node, "peer was in an unknown state", FALSE); crm_warn("%s: in-cluster=%s, is-peer=%s, join=%s, expected=%s, term=%d, shutdown=%d", this_node->details->uname, crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state), do_terminate, this_node->details->shutdown); } return online; } static void determine_remote_online_status(pe_working_set_t * data_set, pe_node_t * this_node) { pe_resource_t *rsc = this_node->details->remote_rsc; pe_resource_t *container = NULL; pe_node_t *host = NULL; /* If there is a node state entry for a (former) Pacemaker Remote node * but no resource creating that node, the node's connection resource will * be NULL. Consider it an offline remote node in that case. */ if (rsc == NULL) { this_node->details->online = FALSE; goto remote_online_done; } container = rsc->container; if (container && pcmk__list_of_1(rsc->running_on)) { host = rsc->running_on->data; } /* If the resource is currently started, mark it online. */ if (rsc->role == RSC_ROLE_STARTED) { crm_trace("%s node %s presumed ONLINE because connection resource is started", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = TRUE; } /* consider this node shutting down if transitioning start->stop */ if (rsc->role == RSC_ROLE_STARTED && rsc->next_role == RSC_ROLE_STOPPED) { crm_trace("%s node %s shutting down because connection resource is stopping", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->shutdown = TRUE; } /* Now check all the failure conditions. */ if(container && pcmk_is_set(container->flags, pe_rsc_failed)) { crm_trace("Guest node %s UNCLEAN because guest resource failed", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { crm_trace("%s node %s OFFLINE because connection resource failed", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; } else if (rsc->role == RSC_ROLE_STOPPED || (container && container->role == RSC_ROLE_STOPPED)) { crm_trace("%s node %s OFFLINE because its resource is stopped", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = FALSE; } else if (host && (host->details->online == FALSE) && host->details->unclean) { crm_trace("Guest node %s UNCLEAN because host is unclean", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } remote_online_done: crm_trace("Remote node %s online=%s", this_node->details->id, this_node->details->online ? "TRUE" : "FALSE"); } static void determine_online_status(xmlNode * node_state, pe_node_t * this_node, pe_working_set_t * data_set) { gboolean online = FALSE; const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); CRM_CHECK(this_node != NULL, return); this_node->details->shutdown = FALSE; this_node->details->expected_up = FALSE; if (pe__shutdown_requested(this_node)) { this_node->details->shutdown = TRUE; } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { this_node->details->expected_up = TRUE; } if (this_node->details->type == node_ping) { this_node->details->unclean = FALSE; online = FALSE; /* As far as resource management is concerned, * the node is safely offline. * Anyone caught abusing this logic will be shot */ } else if (!pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { online = determine_online_status_no_fencing(data_set, node_state, this_node); } else { online = determine_online_status_fencing(data_set, node_state, this_node); } if (online) { this_node->details->online = TRUE; } else { /* remove node from contention */ this_node->fixed = TRUE; this_node->weight = -INFINITY; } if (online && this_node->details->shutdown) { /* don't run resources here */ this_node->fixed = TRUE; this_node->weight = -INFINITY; } if (this_node->details->type == node_ping) { - crm_info("Node %s is not a pacemaker node", this_node->details->uname); + crm_info("Node %s is not a Pacemaker node", this_node->details->uname); } else if (this_node->details->unclean) { pe_proc_warn("Node %s is unclean", this_node->details->uname); } else if (this_node->details->online) { crm_info("Node %s is %s", this_node->details->uname, this_node->details->shutdown ? "shutting down" : this_node->details->pending ? "pending" : this_node->details->standby ? "standby" : this_node->details->maintenance ? "maintenance" : "online"); } else { crm_trace("Node %s is offline", this_node->details->uname); } } /*! * \internal * \brief Find the end of a resource's name, excluding any clone suffix * * \param[in] id Resource ID to check * * \return Pointer to last character of resource's base name */ const char * pe_base_name_end(const char *id) { if (!pcmk__str_empty(id)) { const char *end = id + strlen(id) - 1; for (const char *s = end; s > id; --s) { switch (*s) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': break; case ':': return (s == end)? s : (s - 1); default: return end; } } return end; } return NULL; } /*! * \internal * \brief Get a resource name excluding any clone suffix * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_strip(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); char *basename = NULL; CRM_ASSERT(end); basename = strndup(last_rsc_id, end - last_rsc_id + 1); CRM_ASSERT(basename); return basename; } /*! * \internal * \brief Get the name of the first instance of a cloned resource * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name plus :0 * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_zero(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); size_t base_name_len = end - last_rsc_id + 1; char *zero = NULL; CRM_ASSERT(end); zero = calloc(base_name_len + 3, sizeof(char)); CRM_ASSERT(zero); memcpy(zero, last_rsc_id, base_name_len); zero[base_name_len] = ':'; zero[base_name_len + 1] = '0'; return zero; } static pe_resource_t * create_fake_resource(const char *rsc_id, xmlNode * rsc_entry, pe_working_set_t * data_set) { pe_resource_t *rsc = NULL; xmlNode *xml_rsc = create_xml_node(NULL, XML_CIB_TAG_RESOURCE); copy_in_properties(xml_rsc, rsc_entry); crm_xml_add(xml_rsc, XML_ATTR_ID, rsc_id); crm_log_xml_debug(xml_rsc, "Orphan resource"); if (!common_unpack(xml_rsc, &rsc, NULL, data_set)) { return NULL; } if (xml_contains_remote_node(xml_rsc)) { pe_node_t *node; crm_debug("Detected orphaned remote node %s", rsc_id); node = pe_find_node(data_set->nodes, rsc_id); if (node == NULL) { node = pe_create_node(rsc_id, rsc_id, "remote", NULL, data_set); } link_rsc2remotenode(data_set, rsc); if (node) { crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id); node->details->shutdown = TRUE; } } if (crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER)) { /* This orphaned rsc needs to be mapped to a container. */ crm_trace("Detected orphaned container filler %s", rsc_id); pe__set_resource_flags(rsc, pe_rsc_orphan_container_filler); } pe__set_resource_flags(rsc, pe_rsc_orphan); data_set->resources = g_list_append(data_set->resources, rsc); return rsc; } /*! * \internal * \brief Create orphan instance for anonymous clone resource history */ static pe_resource_t * create_anonymous_orphan(pe_resource_t *parent, const char *rsc_id, pe_node_t *node, pe_working_set_t *data_set) { pe_resource_t *top = pe__create_clone_child(parent, data_set); // find_rsc() because we might be a cloned group pe_resource_t *orphan = top->fns->find_rsc(top, rsc_id, NULL, pe_find_clone); pe_rsc_debug(parent, "Created orphan %s for %s: %s on %s", top->id, parent->id, rsc_id, node->details->uname); return orphan; } /*! * \internal * \brief Check a node for an instance of an anonymous clone * * Return a child instance of the specified anonymous clone, in order of * preference: (1) the instance running on the specified node, if any; * (2) an inactive instance (i.e. within the total of clone-max instances); * (3) a newly created orphan (i.e. clone-max instances are already active). * * \param[in] data_set Cluster information * \param[in] node Node on which to check for instance * \param[in] parent Clone to check * \param[in] rsc_id Name of cloned resource in history (without instance) */ static pe_resource_t * find_anonymous_clone(pe_working_set_t * data_set, pe_node_t * node, pe_resource_t * parent, const char *rsc_id) { GList *rIter = NULL; pe_resource_t *rsc = NULL; pe_resource_t *inactive_instance = NULL; gboolean skip_inactive = FALSE; CRM_ASSERT(parent != NULL); CRM_ASSERT(pe_rsc_is_clone(parent)); CRM_ASSERT(!pcmk_is_set(parent->flags, pe_rsc_unique)); // Check for active (or partially active, for cloned groups) instance pe_rsc_trace(parent, "Looking for %s on %s in %s", rsc_id, node->details->uname, parent->id); for (rIter = parent->children; rsc == NULL && rIter; rIter = rIter->next) { GList *locations = NULL; pe_resource_t *child = rIter->data; /* Check whether this instance is already known to be active or pending * anywhere, at this stage of unpacking. Because this function is called * for a resource before the resource's individual operation history * entries are unpacked, locations will generally not contain the * desired node. * * However, there are three exceptions: * (1) when child is a cloned group and we have already unpacked the * history of another member of the group on the same node; * (2) when we've already unpacked the history of another numbered * instance on the same node (which can happen if globally-unique * was flipped from true to false); and * (3) when we re-run calculations on the same data set as part of a * simulation. */ child->fns->location(child, &locations, 2); if (locations) { /* We should never associate the same numbered anonymous clone * instance with multiple nodes, and clone instances can't migrate, * so there must be only one location, regardless of history. */ CRM_LOG_ASSERT(locations->next == NULL); if (((pe_node_t *)locations->data)->details == node->details) { /* This child instance is active on the requested node, so check * for a corresponding configured resource. We use find_rsc() * instead of child because child may be a cloned group, and we * need the particular member corresponding to rsc_id. * * If the history entry is orphaned, rsc will be NULL. */ rsc = parent->fns->find_rsc(child, rsc_id, NULL, pe_find_clone); if (rsc) { /* If there are multiple instance history entries for an * anonymous clone in a single node's history (which can * happen if globally-unique is switched from true to * false), we want to consider the instances beyond the * first as orphans, even if there are inactive instance * numbers available. */ if (rsc->running_on) { crm_notice("Active (now-)anonymous clone %s has " "multiple (orphan) instance histories on %s", parent->id, node->details->uname); skip_inactive = TRUE; rsc = NULL; } else { pe_rsc_trace(parent, "Resource %s, active", rsc->id); } } } g_list_free(locations); } else { pe_rsc_trace(parent, "Resource %s, skip inactive", child->id); if (!skip_inactive && !inactive_instance && !pcmk_is_set(child->flags, pe_rsc_block)) { // Remember one inactive instance in case we don't find active inactive_instance = parent->fns->find_rsc(child, rsc_id, NULL, pe_find_clone); /* ... but don't use it if it was already associated with a * pending action on another node */ if (inactive_instance && inactive_instance->pending_node && (inactive_instance->pending_node->details != node->details)) { inactive_instance = NULL; } } } } if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) { pe_rsc_trace(parent, "Resource %s, empty slot", inactive_instance->id); rsc = inactive_instance; } /* If the resource has "requires" set to "quorum" or "nothing", and we don't * have a clone instance for every node, we don't want to consume a valid * instance number for unclean nodes. Such instances may appear to be active * according to the history, but should be considered inactive, so we can * start an instance elsewhere. Treat such instances as orphans. * * An exception is instances running on guest nodes -- since guest node * "fencing" is actually just a resource stop, requires shouldn't apply. * * @TODO Ideally, we'd use an inactive instance number if it is not needed * for any clean instances. However, we don't know that at this point. */ if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_needs_fencing) && (!node->details->online || node->details->unclean) && !pe__is_guest_node(node) && !pe__is_universal_clone(parent, data_set)) { rsc = NULL; } if (rsc == NULL) { rsc = create_anonymous_orphan(parent, rsc_id, node, data_set); pe_rsc_trace(parent, "Resource %s, orphan", rsc->id); } return rsc; } static pe_resource_t * unpack_find_resource(pe_working_set_t * data_set, pe_node_t * node, const char *rsc_id, xmlNode * rsc_entry) { pe_resource_t *rsc = NULL; pe_resource_t *parent = NULL; crm_trace("looking for %s", rsc_id); rsc = pe_find_resource(data_set->resources, rsc_id); if (rsc == NULL) { /* If we didn't find the resource by its name in the operation history, * check it again as a clone instance. Even when clone-max=0, we create * a single :0 orphan to match against here. */ char *clone0_id = clone_zero(rsc_id); pe_resource_t *clone0 = pe_find_resource(data_set->resources, clone0_id); if (clone0 && !pcmk_is_set(clone0->flags, pe_rsc_unique)) { rsc = clone0; parent = uber_parent(clone0); crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id); } else { crm_trace("%s is not known as %s either (orphan)", rsc_id, clone0_id); } free(clone0_id); } else if (rsc->variant > pe_native) { crm_trace("Resource history for %s is orphaned because it is no longer primitive", rsc_id); return NULL; } else { parent = uber_parent(rsc); } if (pe_rsc_is_anon_clone(parent)) { if (pe_rsc_is_bundled(parent)) { rsc = pe__find_bundle_replica(parent->parent, node); } else { char *base = clone_strip(rsc_id); rsc = find_anonymous_clone(data_set, node, parent, base); free(base); CRM_ASSERT(rsc != NULL); } } if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_casei) && !pcmk__str_eq(rsc_id, rsc->clone_name, pcmk__str_casei)) { free(rsc->clone_name); rsc->clone_name = strdup(rsc_id); pe_rsc_debug(rsc, "Internally renamed %s on %s to %s%s", rsc_id, node->details->uname, rsc->id, (pcmk_is_set(rsc->flags, pe_rsc_orphan)? " (ORPHAN)" : "")); } return rsc; } static pe_resource_t * process_orphan_resource(xmlNode * rsc_entry, pe_node_t * node, pe_working_set_t * data_set) { pe_resource_t *rsc = NULL; const char *rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID); crm_debug("Detected orphan resource %s on %s", rsc_id, node->details->uname); rsc = create_fake_resource(rsc_id, rsc_entry, data_set); if (!pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)) { pe__clear_resource_flags(rsc, pe_rsc_managed); } else { CRM_CHECK(rsc != NULL, return NULL); pe_rsc_trace(rsc, "Added orphan %s", rsc->id); resource_location(rsc, NULL, -INFINITY, "__orphan_do_not_run__", data_set); } return rsc; } static void process_rsc_state(pe_resource_t * rsc, pe_node_t * node, enum action_fail_response on_fail, xmlNode * migrate_op, pe_working_set_t * data_set) { pe_node_t *tmpnode = NULL; char *reason = NULL; enum action_fail_response save_on_fail = action_fail_ignore; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s", rsc->id, role2text(rsc->role), node->details->uname, fail2text(on_fail)); /* process current state */ if (rsc->role != RSC_ROLE_UNKNOWN) { pe_resource_t *iter = rsc; while (iter) { if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) { pe_node_t *n = pe__copy_node(node); pe_rsc_trace(rsc, "%s (aka. %s) known on %s", rsc->id, rsc->clone_name, n->details->uname); g_hash_table_insert(iter->known_on, (gpointer) n->details->id, n); } if (pcmk_is_set(iter->flags, pe_rsc_unique)) { break; } iter = iter->parent; } } /* If a managed resource is believed to be running, but node is down ... */ if (rsc->role > RSC_ROLE_STOPPED && node->details->online == FALSE && node->details->maintenance == FALSE && pcmk_is_set(rsc->flags, pe_rsc_managed)) { gboolean should_fence = FALSE; /* If this is a guest node, fence it (regardless of whether fencing is * enabled, because guest node fencing is done by recovery of the * container resource rather than by the fencer). Mark the resource * we're processing as failed. When the guest comes back up, its * operation history in the CIB will be cleared, freeing the affected * resource to run again once we are sure we know its state. */ if (pe__is_guest_node(node)) { pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); should_fence = TRUE; } else if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { if (pe__is_remote_node(node) && node->details->remote_rsc && !pcmk_is_set(node->details->remote_rsc->flags, pe_rsc_failed)) { /* Setting unseen means that fencing of the remote node will * occur only if the connection resource is not going to start * somewhere. This allows connection resources on a failed * cluster node to move to another node without requiring the * remote nodes to be fenced as well. */ node->details->unseen = TRUE; reason = crm_strdup_printf("%s is active there (fencing will be" " revoked if remote connection can " "be re-established elsewhere)", rsc->id); } should_fence = TRUE; } if (should_fence) { if (reason == NULL) { reason = crm_strdup_printf("%s is thought to be active there", rsc->id); } pe_fence_node(data_set, node, reason, FALSE); } free(reason); } /* In order to calculate priority_fencing_delay correctly, save the failure information and pass it to native_add_running(). */ save_on_fail = on_fail; if (node->details->unclean) { /* No extra processing needed * Also allows resources to be started again after a node is shot */ on_fail = action_fail_ignore; } switch (on_fail) { case action_fail_ignore: /* nothing to do */ break; case action_fail_demote: pe__set_resource_flags(rsc, pe_rsc_failed); demote_action(rsc, node, FALSE); break; case action_fail_fence: /* treat it as if it is still running * but also mark the node as unclean */ reason = crm_strdup_printf("%s failed there", rsc->id); pe_fence_node(data_set, node, reason, FALSE); free(reason); break; case action_fail_standby: node->details->standby = TRUE; node->details->standby_onfail = TRUE; break; case action_fail_block: /* is_managed == FALSE will prevent any * actions being sent for the resource */ pe__clear_resource_flags(rsc, pe_rsc_managed); pe__set_resource_flags(rsc, pe_rsc_block); break; case action_fail_migrate: /* make sure it comes up somewhere else * or not at all */ resource_location(rsc, node, -INFINITY, "__action_migration_auto__", data_set); break; case action_fail_stop: pe__set_next_role(rsc, RSC_ROLE_STOPPED, "on-fail=stop"); break; case action_fail_recover: if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); stop_action(rsc, node, FALSE); } break; case action_fail_restart_container: pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); if (rsc->container && pe_rsc_is_bundled(rsc)) { /* A bundle's remote connection can run on a different node than * the bundle's container. We don't necessarily know where the * container is running yet, so remember it and add a stop * action for it later. */ data_set->stop_needed = g_list_prepend(data_set->stop_needed, rsc->container); } else if (rsc->container) { stop_action(rsc->container, node, FALSE); } else if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { stop_action(rsc, node, FALSE); } break; case action_fail_reset_remote: pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { tmpnode = NULL; if (rsc->is_remote_node) { tmpnode = pe_find_node(data_set->nodes, rsc->id); } if (tmpnode && pe__is_remote_node(tmpnode) && tmpnode->details->remote_was_fenced == 0) { /* The remote connection resource failed in a way that * should result in fencing the remote node. */ pe_fence_node(data_set, tmpnode, "remote connection is unrecoverable", FALSE); } } /* require the stop action regardless if fencing is occurring or not. */ if (rsc->role > RSC_ROLE_STOPPED) { stop_action(rsc, node, FALSE); } /* if reconnect delay is in use, prevent the connection from exiting the * "STOPPED" role until the failure is cleared by the delay timeout. */ if (rsc->remote_reconnect_ms) { pe__set_next_role(rsc, RSC_ROLE_STOPPED, "remote reset"); } break; } /* ensure a remote-node connection failure forces an unclean remote-node * to be fenced. By setting unseen = FALSE, the remote-node failure will * result in a fencing operation regardless if we're going to attempt to * reconnect to the remote-node in this transition or not. */ if (pcmk_is_set(rsc->flags, pe_rsc_failed) && rsc->is_remote_node) { tmpnode = pe_find_node(data_set->nodes, rsc->id); if (tmpnode && tmpnode->details->unclean) { tmpnode->details->unseen = FALSE; } } if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) { if (pcmk_is_set(rsc->flags, pe_rsc_managed)) { pcmk__config_warn("Detected active orphan %s running on %s", rsc->id, node->details->uname); } else { pcmk__config_warn("Resource '%s' must be stopped manually on " "%s because cluster is configured not to " "stop active orphans", rsc->id, node->details->uname); } } native_add_running(rsc, node, data_set, (save_on_fail != action_fail_ignore)); switch (on_fail) { case action_fail_ignore: break; case action_fail_demote: case action_fail_block: pe__set_resource_flags(rsc, pe_rsc_failed); break; default: pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); break; } } else if (rsc->clone_name && strchr(rsc->clone_name, ':') != NULL) { /* Only do this for older status sections that included instance numbers * Otherwise stopped instances will appear as orphans */ pe_rsc_trace(rsc, "Resetting clone_name %s for %s (stopped)", rsc->clone_name, rsc->id); free(rsc->clone_name); rsc->clone_name = NULL; } else { GList *possible_matches = pe__resource_actions(rsc, node, RSC_STOP, FALSE); GList *gIter = possible_matches; for (; gIter != NULL; gIter = gIter->next) { pe_action_t *stop = (pe_action_t *) gIter->data; pe__set_action_flags(stop, pe_action_optional); } g_list_free(possible_matches); } } /* create active recurring operations as optional */ static void process_recurring(pe_node_t * node, pe_resource_t * rsc, int start_index, int stop_index, GList *sorted_op_list, pe_working_set_t * data_set) { int counter = -1; const char *task = NULL; const char *status = NULL; GList *gIter = sorted_op_list; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "%s: Start index %d, stop index = %d", rsc->id, start_index, stop_index); for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; guint interval_ms = 0; char *key = NULL; const char *id = ID(rsc_op); counter++; if (node->details->online == FALSE) { pe_rsc_trace(rsc, "Skipping %s/%s: node is offline", rsc->id, node->details->uname); break; /* Need to check if there's a monitor for role="Stopped" */ } else if (start_index < stop_index && counter <= stop_index) { pe_rsc_trace(rsc, "Skipping %s/%s: resource is not active", id, node->details->uname); continue; } else if (counter < start_index) { pe_rsc_trace(rsc, "Skipping %s/%s: old %d", id, node->details->uname, counter); continue; } crm_element_value_ms(rsc_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if (interval_ms == 0) { pe_rsc_trace(rsc, "Skipping %s/%s: non-recurring", id, node->details->uname); continue; } status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS); if (pcmk__str_eq(status, "-1", pcmk__str_casei)) { pe_rsc_trace(rsc, "Skipping %s/%s: status", id, node->details->uname); continue; } task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); /* create the action */ key = pcmk__op_key(rsc->id, task, interval_ms); pe_rsc_trace(rsc, "Creating %s/%s", key, node->details->uname); custom_action(rsc, key, task, node, TRUE, TRUE, data_set); } } void calculate_active_ops(GList *sorted_op_list, int *start_index, int *stop_index) { int counter = -1; int implied_monitor_start = -1; int implied_clone_start = -1; const char *task = NULL; const char *status = NULL; GList *gIter = sorted_op_list; *stop_index = -1; *start_index = -1; for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS); if (pcmk__str_eq(task, CRMD_ACTION_STOP, pcmk__str_casei) && pcmk__str_eq(status, "0", pcmk__str_casei)) { *stop_index = counter; } else if (pcmk__strcase_any_of(task, CRMD_ACTION_START, CRMD_ACTION_MIGRATED, NULL)) { *start_index = counter; } else if ((implied_monitor_start <= *stop_index) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { const char *rc = crm_element_value(rsc_op, XML_LRM_ATTR_RC); if (pcmk__strcase_any_of(rc, "0", "8", NULL)) { implied_monitor_start = counter; } } else if (pcmk__strcase_any_of(task, CRMD_ACTION_PROMOTE, CRMD_ACTION_DEMOTE, NULL)) { implied_clone_start = counter; } } if (*start_index == -1) { if (implied_clone_start != -1) { *start_index = implied_clone_start; } else if (implied_monitor_start != -1) { *start_index = implied_monitor_start; } } } // If resource history entry has shutdown lock, remember lock node and time static void unpack_shutdown_lock(xmlNode *rsc_entry, pe_resource_t *rsc, pe_node_t *node, pe_working_set_t *data_set) { time_t lock_time = 0; // When lock started (i.e. node shutdown time) if ((crm_element_value_epoch(rsc_entry, XML_CONFIG_ATTR_SHUTDOWN_LOCK, &lock_time) == pcmk_ok) && (lock_time != 0)) { if ((data_set->shutdown_lock > 0) && (get_effective_time(data_set) > (lock_time + data_set->shutdown_lock))) { pe_rsc_info(rsc, "Shutdown lock for %s on %s expired", rsc->id, node->details->uname); pe__clear_resource_history(rsc, node, data_set); } else { rsc->lock_node = node; rsc->lock_time = lock_time; } } } /*! * \internal * \brief Unpack one lrm_resource entry from a node's CIB status * * \param[in] node Node whose status is being unpacked * \param[in] rsc_entry lrm_resource XML being unpacked * \param[in] data_set Cluster working set * * \return Resource corresponding to the entry, or NULL if no operation history */ static pe_resource_t * unpack_lrm_resource(pe_node_t *node, xmlNode *lrm_resource, pe_working_set_t *data_set) { GList *gIter = NULL; int stop_index = -1; int start_index = -1; enum rsc_role_e req_role = RSC_ROLE_UNKNOWN; const char *task = NULL; const char *rsc_id = ID(lrm_resource); pe_resource_t *rsc = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; xmlNode *migrate_op = NULL; xmlNode *rsc_op = NULL; xmlNode *last_failure = NULL; enum action_fail_response on_fail = action_fail_ignore; enum rsc_role_e saved_role = RSC_ROLE_UNKNOWN; if (rsc_id == NULL) { crm_warn("Ignoring malformed " XML_LRM_TAG_RESOURCE " entry without id"); return NULL; } crm_trace("Unpacking " XML_LRM_TAG_RESOURCE " for %s on %s", rsc_id, node->details->uname); // Build a list of individual lrm_rsc_op entries, so we can sort them for (rsc_op = first_named_child(lrm_resource, XML_LRM_TAG_RSC_OP); rsc_op != NULL; rsc_op = crm_next_same_xml(rsc_op)) { op_list = g_list_prepend(op_list, rsc_op); } if (!pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } } /* find the resource */ rsc = unpack_find_resource(data_set, node, rsc_id, lrm_resource); if (rsc == NULL) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } else { rsc = process_orphan_resource(lrm_resource, node, data_set); } } CRM_ASSERT(rsc != NULL); // Check whether the resource is "shutdown-locked" to this node if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { unpack_shutdown_lock(lrm_resource, rsc, node, data_set); } /* process operations */ saved_role = rsc->role; rsc->role = RSC_ROLE_UNKNOWN; sorted_op_list = g_list_sort(op_list, sort_op_by_callid); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); if (pcmk__str_eq(task, CRMD_ACTION_MIGRATED, pcmk__str_casei)) { migrate_op = rsc_op; } unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail, data_set); } /* create active recurring operations as optional */ calculate_active_ops(sorted_op_list, &start_index, &stop_index); process_recurring(node, rsc, start_index, stop_index, sorted_op_list, data_set); /* no need to free the contents */ g_list_free(sorted_op_list); process_rsc_state(rsc, node, on_fail, migrate_op, data_set); if (get_target_role(rsc, &req_role)) { if (rsc->next_role == RSC_ROLE_UNKNOWN || req_role < rsc->next_role) { pe__set_next_role(rsc, req_role, XML_RSC_ATTR_TARGET_ROLE); } else if (req_role > rsc->next_role) { pe_rsc_info(rsc, "%s: Not overwriting calculated next role %s" " with requested next role %s", rsc->id, role2text(rsc->next_role), role2text(req_role)); } } if (saved_role > rsc->role) { rsc->role = saved_role; } return rsc; } static void handle_orphaned_container_fillers(xmlNode * lrm_rsc_list, pe_working_set_t * data_set) { xmlNode *rsc_entry = NULL; for (rsc_entry = pcmk__xe_first_child(lrm_rsc_list); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) { pe_resource_t *rsc; pe_resource_t *container; const char *rsc_id; const char *container_id; if (!pcmk__str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, pcmk__str_casei)) { continue; } container_id = crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER); rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID); if (container_id == NULL || rsc_id == NULL) { continue; } container = pe_find_resource(data_set->resources, container_id); if (container == NULL) { continue; } rsc = pe_find_resource(data_set->resources, rsc_id); if (rsc == NULL || !pcmk_is_set(rsc->flags, pe_rsc_orphan_container_filler) || rsc->container != NULL) { continue; } pe_rsc_trace(rsc, "Mapped container of orphaned resource %s to %s", rsc->id, container_id); rsc->container = container; container->fillers = g_list_append(container->fillers, rsc); } } /*! * \internal * \brief Unpack one node's lrm status section * * \param[in] node Node whose status is being unpacked * \param[in] xml CIB node state XML * \param[in] data_set Cluster working set */ static void unpack_node_lrm(pe_node_t *node, xmlNode *xml, pe_working_set_t *data_set) { bool found_orphaned_container_filler = false; // Drill down to lrm_resources section xml = find_xml_node(xml, XML_CIB_TAG_LRM, FALSE); if (xml == NULL) { return; } xml = find_xml_node(xml, XML_LRM_TAG_RESOURCES, FALSE); if (xml == NULL) { return; } // Unpack each lrm_resource entry for (xmlNode *rsc_entry = first_named_child(xml, XML_LRM_TAG_RESOURCE); rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) { pe_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, data_set); if ((rsc != NULL) && pcmk_is_set(rsc->flags, pe_rsc_orphan_container_filler)) { found_orphaned_container_filler = true; } } /* Now that all resource state has been unpacked for this node, map any * orphaned container fillers to their container resource. */ if (found_orphaned_container_filler) { handle_orphaned_container_fillers(xml, data_set); } } static void set_active(pe_resource_t * rsc) { pe_resource_t *top = uber_parent(rsc); if (top && pcmk_is_set(top->flags, pe_rsc_promotable)) { rsc->role = RSC_ROLE_UNPROMOTED; } else { rsc->role = RSC_ROLE_STARTED; } } static void set_node_score(gpointer key, gpointer value, gpointer user_data) { pe_node_t *node = value; int *score = user_data; node->weight = *score; } #define STATUS_PATH_MAX 1024 static xmlNode * find_lrm_op(const char *resource, const char *op, const char *node, const char *source, bool success_only, pe_working_set_t *data_set) { int offset = 0; char xpath[STATUS_PATH_MAX]; xmlNode *xml = NULL; offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "//node_state[@uname='%s']", node); offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "//" XML_LRM_TAG_RESOURCE "[@id='%s']", resource); /* Need to check against transition_magic too? */ if (source && pcmk__str_eq(op, CRMD_ACTION_MIGRATE, pcmk__str_casei)) { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s' and @migrate_target='%s']", op, source); } else if (source && pcmk__str_eq(op, CRMD_ACTION_MIGRATED, pcmk__str_casei)) { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s' and @migrate_source='%s']", op, source); } else { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s']", op); } CRM_LOG_ASSERT(offset > 0); xml = get_xpath_object(xpath, data_set->input, LOG_DEBUG); if (xml && success_only) { int rc = PCMK_OCF_UNKNOWN_ERROR; int status = PCMK_LRM_OP_ERROR; crm_element_value_int(xml, XML_LRM_ATTR_RC, &rc); crm_element_value_int(xml, XML_LRM_ATTR_OPSTATUS, &status); if ((rc != PCMK_OCF_OK) || (status != PCMK_LRM_OP_DONE)) { return NULL; } } return xml; } static int pe__call_id(xmlNode *op_xml) { int id = 0; if (op_xml) { crm_element_value_int(op_xml, XML_LRM_ATTR_CALLID, &id); } return id; } /*! * \brief Check whether a stop happened on the same node after some event * * \param[in] rsc Resource being checked * \param[in] node Node being checked * \param[in] xml_op Event that stop is being compared to * \param[in] data_set Cluster working set * * \return TRUE if stop happened after event, FALSE otherwise * * \note This is really unnecessary, but kept as a safety mechanism. We * currently don't save more than one successful event in history, so this * only matters when processing really old CIB files that we don't * technically support anymore, or as preparation for logging an extended * history in the future. */ static bool stop_happened_after(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { xmlNode *stop_op = find_lrm_op(rsc->id, CRMD_ACTION_STOP, node->details->uname, NULL, TRUE, data_set); return (stop_op && (pe__call_id(stop_op) > pe__call_id(xml_op))); } static void unpack_migrate_to_success(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { /* A successful migration sequence is: * migrate_to on source node * migrate_from on target node * stop on source node * * If a migrate_to is followed by a stop, the entire migration (successful * or failed) is complete, and we don't care what happened on the target. * * If no migrate_from has happened, the migration is considered to be * "partial". If the migrate_from failed, make sure the resource gets * stopped on both source and target (if up). * * If the migrate_to and migrate_from both succeeded (which also implies the * resource is no longer running on the source), but there is no stop, the * migration is considered to be "dangling". Schedule a stop on the source * in this case. */ int from_rc = 0; int from_status = 0; pe_node_t *target_node = NULL; pe_node_t *source_node = NULL; xmlNode *migrate_from = NULL; const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); // Sanity check CRM_CHECK(source && target && !strcmp(source, node->details->uname), return); if (stop_happened_after(rsc, node, xml_op, data_set)) { return; } // Clones are not allowed to migrate, so role can't be promoted rsc->role = RSC_ROLE_STARTED; target_node = pe_find_node(data_set->nodes, target); source_node = pe_find_node(data_set->nodes, source); // Check whether there was a migrate_from action on the target migrate_from = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATED, target, source, FALSE, data_set); if (migrate_from) { crm_element_value_int(migrate_from, XML_LRM_ATTR_RC, &from_rc); crm_element_value_int(migrate_from, XML_LRM_ATTR_OPSTATUS, &from_status); pe_rsc_trace(rsc, "%s op on %s exited with status=%d, rc=%d", ID(migrate_from), target, from_status, from_rc); } if (migrate_from && from_rc == PCMK_OCF_OK && from_status == PCMK_LRM_OP_DONE) { /* The migrate_to and migrate_from both succeeded, so mark the migration * as "dangling". This will be used to schedule a stop action on the * source without affecting the target. */ pe_rsc_trace(rsc, "Detected dangling migration op: %s on %s", ID(xml_op), source); rsc->role = RSC_ROLE_STOPPED; rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, node); } else if (migrate_from && (from_status != PCMK_LRM_OP_PENDING)) { // Failed if (target_node && target_node->details->online) { pe_rsc_trace(rsc, "Marking active on %s %p %d", target, target_node, target_node->details->online); native_add_running(rsc, target_node, data_set, TRUE); } } else { // Pending, or complete but erased if (target_node && target_node->details->online) { pe_rsc_trace(rsc, "Marking active on %s %p %d", target, target_node, target_node->details->online); native_add_running(rsc, target_node, data_set, FALSE); if (source_node && source_node->details->online) { /* This is a partial migration: the migrate_to completed * successfully on the source, but the migrate_from has not * completed. Remember the source and target; if the newly * chosen target remains the same when we schedule actions * later, we may continue with the migration. */ rsc->partial_migration_target = target_node; rsc->partial_migration_source = source_node; } } else { /* Consider it failed here - forces a restart, prevents migration */ pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); pe__clear_resource_flags(rsc, pe_rsc_allow_migrate); } } } // Is there an action_name in node_name's rsc history newer than call_id? static bool newer_op(pe_resource_t *rsc, const char *action_name, const char *node_name, int call_id, pe_working_set_t *data_set) { xmlNode *action = find_lrm_op(rsc->id, action_name, node_name, NULL, TRUE, data_set); return pe__call_id(action) > call_id; } static void unpack_migrate_to_failure(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { int target_stop_id = 0; int target_migrate_from_id = 0; xmlNode *target_stop = NULL; xmlNode *target_migrate_from = NULL; const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); // Sanity check CRM_CHECK(source && target && !strcmp(source, node->details->uname), return); /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ rsc->role = RSC_ROLE_STARTED; // Check for stop on the target target_stop = find_lrm_op(rsc->id, CRMD_ACTION_STOP, target, NULL, TRUE, data_set); target_stop_id = pe__call_id(target_stop); // Check for migrate_from on the target target_migrate_from = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATED, target, source, TRUE, data_set); target_migrate_from_id = pe__call_id(target_migrate_from); if ((target_stop == NULL) || (target_stop_id < target_migrate_from_id)) { /* There was no stop on the target, or a stop that happened before a * migrate_from, so assume the resource is still active on the target * (if it is up). */ pe_node_t *target_node = pe_find_node(data_set->nodes, target); pe_rsc_trace(rsc, "stop (%d) + migrate_from (%d)", target_stop_id, target_migrate_from_id); if (target_node && target_node->details->online) { native_add_running(rsc, target_node, data_set, FALSE); } } else if (target_migrate_from == NULL) { /* We know there was a stop on the target, but there may not have been a * migrate_from (the stop could have happened before migrate_from was * scheduled or attempted). * * That means this could be a "dangling" migration. But first, check * whether there is a newer successful stop, start, or migrate_from on * the source node -- it's possible the failed migration was followed by * a successful stop, full restart, or migration in the reverse * direction, in which case we don't want to force a stop. */ int source_migrate_to_id = pe__call_id(xml_op); if (newer_op(rsc, CRMD_ACTION_MIGRATED, source, source_migrate_to_id, data_set) || newer_op(rsc, CRMD_ACTION_START, source, source_migrate_to_id, data_set) || newer_op(rsc, CRMD_ACTION_STOP, source, source_migrate_to_id, data_set)) { return; } // Mark node as having dangling migration so we can force a stop later rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, node); } } static void unpack_migrate_from_failure(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { xmlNode *source_stop = NULL; xmlNode *source_migrate_to = NULL; const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); // Sanity check CRM_CHECK(source && target && !strcmp(target, node->details->uname), return); /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ rsc->role = RSC_ROLE_STARTED; // Check for a stop on the source source_stop = find_lrm_op(rsc->id, CRMD_ACTION_STOP, source, NULL, TRUE, data_set); // Check for a migrate_to on the source source_migrate_to = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATE, source, target, TRUE, data_set); if ((source_stop == NULL) || (pe__call_id(source_stop) < pe__call_id(source_migrate_to))) { /* There was no stop on the source, or a stop that happened before * migrate_to, so assume the resource is still active on the source (if * it is up). */ pe_node_t *source_node = pe_find_node(data_set->nodes, source); if (source_node && source_node->details->online) { native_add_running(rsc, source_node, data_set, TRUE); } } } static void record_failed_op(xmlNode *op, const pe_node_t *node, const pe_resource_t *rsc, pe_working_set_t *data_set) { xmlNode *xIter = NULL; const char *op_key = crm_element_value(op, XML_LRM_ATTR_TASK_KEY); if (node->details->online == FALSE) { return; } for (xIter = data_set->failed->children; xIter; xIter = xIter->next) { const char *key = crm_element_value(xIter, XML_LRM_ATTR_TASK_KEY); const char *uname = crm_element_value(xIter, XML_ATTR_UNAME); if(pcmk__str_eq(op_key, key, pcmk__str_casei) && pcmk__str_eq(uname, node->details->uname, pcmk__str_casei)) { crm_trace("Skipping duplicate entry %s on %s", op_key, node->details->uname); return; } } crm_trace("Adding entry %s on %s", op_key, node->details->uname); crm_xml_add(op, XML_ATTR_UNAME, node->details->uname); crm_xml_add(op, XML_LRM_ATTR_RSCID, rsc->id); add_node_copy(data_set->failed, op); } static const char *get_op_key(xmlNode *xml_op) { const char *key = crm_element_value(xml_op, XML_LRM_ATTR_TASK_KEY); if(key == NULL) { key = ID(xml_op); } return key; } static const char * last_change_str(xmlNode *xml_op) { time_t when; const char *when_s = NULL; if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE, &when) == pcmk_ok) { when_s = pcmk__epoch2str(&when); if (when_s) { // Skip day of week to make message shorter when_s = strchr(when_s, ' '); if (when_s) { ++when_s; } } } return ((when_s && *when_s)? when_s : "unknown time"); } /*! * \internal * \brief Compare two on-fail values * * \param[in] first One on-fail value to compare * \param[in] second The other on-fail value to compare * * \return A negative number if second is more severe than first, zero if they * are equal, or a positive number if first is more severe than second. * \note This is only needed until the action_fail_response values can be * renumbered at the next API compatibility break. */ static int cmp_on_fail(enum action_fail_response first, enum action_fail_response second) { switch (first) { case action_fail_demote: switch (second) { case action_fail_ignore: return 1; case action_fail_demote: return 0; default: return -1; } break; case action_fail_reset_remote: switch (second) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: return 1; case action_fail_reset_remote: return 0; default: return -1; } break; case action_fail_restart_container: switch (second) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: case action_fail_reset_remote: return 1; case action_fail_restart_container: return 0; default: return -1; } break; default: break; } switch (second) { case action_fail_demote: return (first == action_fail_ignore)? -1 : 1; case action_fail_reset_remote: switch (first) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: return -1; default: return 1; } break; case action_fail_restart_container: switch (first) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: case action_fail_reset_remote: return -1; default: return 1; } break; default: break; } return first - second; } static void unpack_rsc_op_failure(pe_resource_t * rsc, pe_node_t * node, int rc, xmlNode * xml_op, xmlNode ** last_failure, enum action_fail_response * on_fail, pe_working_set_t * data_set) { guint interval_ms = 0; bool is_probe = false; pe_action_t *action = NULL; const char *key = get_op_key(xml_op); const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON); CRM_ASSERT(rsc); CRM_CHECK(task != NULL, return); *last_failure = xml_op; crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((interval_ms == 0) && !strcmp(task, CRMD_ACTION_STATUS)) { is_probe = true; } if (exit_reason == NULL) { exit_reason = ""; } if (!pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster) && (rc == PCMK_OCF_NOT_INSTALLED)) { crm_trace("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " CRM_XS " rc=%d id=%s", services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, (is_probe? "probe" : task), rsc->id, node->details->uname, last_change_str(xml_op), rc, ID(xml_op)); } else { crm_warn("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " CRM_XS " rc=%d id=%s", services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, (is_probe? "probe" : task), rsc->id, node->details->uname, last_change_str(xml_op), rc, ID(xml_op)); if (is_probe && (rc != PCMK_OCF_OK) && (rc != PCMK_OCF_NOT_RUNNING) && (rc != PCMK_OCF_RUNNING_PROMOTED)) { /* A failed (not just unexpected) probe result could mean the user * didn't know resources will be probed even where they can't run. */ crm_notice("If it is not possible for %s to run on %s, see " "the resource-discovery option for location constraints", rsc->id, node->details->uname); } record_failed_op(xml_op, node, rsc, data_set); } action = custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set); if (cmp_on_fail(*on_fail, action->on_fail) < 0) { pe_rsc_trace(rsc, "on-fail %s -> %s for %s (%s)", fail2text(*on_fail), fail2text(action->on_fail), action->uuid, key); *on_fail = action->on_fail; } if (!strcmp(task, CRMD_ACTION_STOP)) { resource_location(rsc, node, -INFINITY, "__stop_fail__", data_set); } else if (!strcmp(task, CRMD_ACTION_MIGRATE)) { unpack_migrate_to_failure(rsc, node, xml_op, data_set); } else if (!strcmp(task, CRMD_ACTION_MIGRATED)) { unpack_migrate_from_failure(rsc, node, xml_op, data_set); } else if (!strcmp(task, CRMD_ACTION_PROMOTE)) { rsc->role = RSC_ROLE_PROMOTED; } else if (!strcmp(task, CRMD_ACTION_DEMOTE)) { if (action->on_fail == action_fail_block) { rsc->role = RSC_ROLE_PROMOTED; pe__set_next_role(rsc, RSC_ROLE_STOPPED, "demote with on-fail=block"); } else if(rc == PCMK_OCF_NOT_RUNNING) { rsc->role = RSC_ROLE_STOPPED; } else { /* Staying in the promoted role would put the scheduler and * controller into a loop. Setting the role to unpromoted is not * dangerous because the resource will be stopped as part of * recovery, and any promotion will be ordered after that stop. */ rsc->role = RSC_ROLE_UNPROMOTED; } } if(is_probe && rc == PCMK_OCF_NOT_INSTALLED) { /* leave stopped */ pe_rsc_trace(rsc, "Leaving %s stopped", rsc->id); rsc->role = RSC_ROLE_STOPPED; } else if (rsc->role < RSC_ROLE_STARTED) { pe_rsc_trace(rsc, "Setting %s active", rsc->id); set_active(rsc); } pe_rsc_trace(rsc, "Resource %s: role=%s, unclean=%s, on_fail=%s, fail_role=%s", rsc->id, role2text(rsc->role), pcmk__btoa(node->details->unclean), fail2text(action->on_fail), role2text(action->fail_role)); if (action->fail_role != RSC_ROLE_STARTED && rsc->next_role < action->fail_role) { pe__set_next_role(rsc, action->fail_role, "failure"); } if (action->fail_role == RSC_ROLE_STOPPED) { int score = -INFINITY; pe_resource_t *fail_rsc = rsc; if (fail_rsc->parent) { pe_resource_t *parent = uber_parent(fail_rsc); if (pe_rsc_is_clone(parent) && !pcmk_is_set(parent->flags, pe_rsc_unique)) { /* For clone resources, if a child fails on an operation * with on-fail = stop, all the resources fail. Do this by preventing * the parent from coming up again. */ fail_rsc = parent; } } crm_notice("%s will not be started under current conditions", fail_rsc->id); /* make sure it doesn't come up again */ if (fail_rsc->allowed_nodes != NULL) { g_hash_table_destroy(fail_rsc->allowed_nodes); } fail_rsc->allowed_nodes = pe__node_list2table(data_set->nodes); g_hash_table_foreach(fail_rsc->allowed_nodes, set_node_score, &score); } pe_free_action(action); } /*! * \internal * \brief Remap operation status based on action result * * Given an action result, determine an appropriate operation status for the * purposes of responding to the action (the status provided by the executor is * not directly usable since the executor does not know what was expected). * * \param[in,out] rsc Resource that operation history entry is for * \param[in] rc Actual return code of operation * \param[in] target_rc Expected return code of operation * \param[in] node Node where operation was executed * \param[in] xml_op Operation history entry XML from CIB status * \param[in,out] on_fail What should be done about the result * \param[in] data_set Current cluster working set * * \return Operation status based on return code and action info * \note This may update the resource's current and next role. */ static int determine_op_status( pe_resource_t *rsc, int rc, int target_rc, pe_node_t * node, xmlNode * xml_op, enum action_fail_response * on_fail, pe_working_set_t * data_set) { guint interval_ms = 0; bool is_probe = false; int result = PCMK_LRM_OP_DONE; const char *key = get_op_key(xml_op); const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON); CRM_ASSERT(rsc); CRM_CHECK(task != NULL, return PCMK_LRM_OP_ERROR); if (exit_reason == NULL) { exit_reason = ""; } crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((interval_ms == 0) && !strcmp(task, CRMD_ACTION_STATUS)) { is_probe = true; task = "probe"; } if (target_rc < 0) { /* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with * Heartbeat 2.0.7 or earlier as the cluster layer, did not include the * target_rc in the transition key, which (along with the similar case * of a corrupted transition key in the CIB) will be reported to this * function as -1. Pacemaker 2.0+ does not support rolling upgrades from * those versions or processing of saved CIB files from those versions, * so we do not need to care much about this case. */ result = PCMK_LRM_OP_ERROR; crm_warn("Expected result not found for %s on %s (corrupt or obsolete CIB?)", key, node->details->uname); } else if (target_rc != rc) { result = PCMK_LRM_OP_ERROR; pe_rsc_debug(rsc, "%s on %s: expected %d (%s), got %d (%s%s%s)", key, node->details->uname, target_rc, services_ocf_exitcode_str(target_rc), rc, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason); } switch (rc) { case PCMK_OCF_OK: if (is_probe && (target_rc == PCMK_OCF_NOT_RUNNING)) { result = PCMK_LRM_OP_DONE; pe_rsc_info(rsc, "Probe found %s active on %s at %s", rsc->id, node->details->uname, last_change_str(xml_op)); } break; case PCMK_OCF_NOT_RUNNING: if (is_probe || (target_rc == rc) || !pcmk_is_set(rsc->flags, pe_rsc_managed)) { result = PCMK_LRM_OP_DONE; rsc->role = RSC_ROLE_STOPPED; /* clear any previous failure actions */ *on_fail = action_fail_ignore; pe__set_next_role(rsc, RSC_ROLE_UNKNOWN, "not running"); } break; case PCMK_OCF_RUNNING_PROMOTED: if (is_probe && (rc != target_rc)) { result = PCMK_LRM_OP_DONE; pe_rsc_info(rsc, "Probe found %s active and promoted on %s at %s", rsc->id, node->details->uname, last_change_str(xml_op)); } rsc->role = RSC_ROLE_PROMOTED; break; case PCMK_OCF_DEGRADED_PROMOTED: case PCMK_OCF_FAILED_PROMOTED: rsc->role = RSC_ROLE_PROMOTED; result = PCMK_LRM_OP_ERROR; break; case PCMK_OCF_NOT_CONFIGURED: result = PCMK_LRM_OP_ERROR_FATAL; break; case PCMK_OCF_UNIMPLEMENT_FEATURE: if (interval_ms > 0) { result = PCMK_LRM_OP_NOTSUPPORTED; break; } // fall through case PCMK_OCF_NOT_INSTALLED: case PCMK_OCF_INVALID_PARAM: case PCMK_OCF_INSUFFICIENT_PRIV: if (!pe_can_fence(data_set, node) && !strcmp(task, CRMD_ACTION_STOP)) { /* If a stop fails and we can't fence, there's nothing else we can do */ pe_proc_err("No further recovery can be attempted for %s " "because %s on %s failed (%s%s%s) at %s " CRM_XS " rc=%d id=%s", rsc->id, task, node->details->uname, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, last_change_str(xml_op), rc, ID(xml_op)); pe__clear_resource_flags(rsc, pe_rsc_managed); pe__set_resource_flags(rsc, pe_rsc_block); } result = PCMK_LRM_OP_ERROR_HARD; break; default: if (result == PCMK_LRM_OP_DONE) { crm_info("Treating unknown exit status %d from %s of %s " "on %s at %s as failure", rc, task, rsc->id, node->details->uname, last_change_str(xml_op)); result = PCMK_LRM_OP_ERROR; } break; } return result; } // return TRUE if start or monitor last failure but parameters changed static bool should_clear_for_param_change(xmlNode *xml_op, const char *task, pe_resource_t *rsc, pe_node_t *node, pe_working_set_t *data_set) { if (!strcmp(task, "start") || !strcmp(task, "monitor")) { if (pe__bundle_needs_remote_name(rsc, data_set)) { /* We haven't allocated resources yet, so we can't reliably * substitute addr parameters for the REMOTE_CONTAINER_HACK. * When that's needed, defer the check until later. */ pe__add_param_check(xml_op, rsc, node, pe_check_last_failure, data_set); } else { op_digest_cache_t *digest_data = NULL; digest_data = rsc_action_digest_cmp(rsc, xml_op, node, data_set); switch (digest_data->rc) { case RSC_DIGEST_UNKNOWN: crm_trace("Resource %s history entry %s on %s" " has no digest to compare", rsc->id, get_op_key(xml_op), node->details->id); break; case RSC_DIGEST_MATCH: break; default: return TRUE; } } } return FALSE; } // Order action after fencing of remote node, given connection rsc static void order_after_remote_fencing(pe_action_t *action, pe_resource_t *remote_conn, pe_working_set_t *data_set) { pe_node_t *remote_node = pe_find_node(data_set->nodes, remote_conn->id); if (remote_node) { pe_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL, FALSE, data_set); order_actions(fence, action, pe_order_implies_then); } } static bool should_ignore_failure_timeout(pe_resource_t *rsc, xmlNode *xml_op, const char *task, guint interval_ms, bool is_last_failure, pe_working_set_t *data_set) { /* Clearing failures of recurring monitors has special concerns. The * executor reports only changes in the monitor result, so if the * monitor is still active and still getting the same failure result, * that will go undetected after the failure is cleared. * * Also, the operation history will have the time when the recurring * monitor result changed to the given code, not the time when the * result last happened. * * @TODO We probably should clear such failures only when the failure * timeout has passed since the last occurrence of the failed result. * However we don't record that information. We could maybe approximate * that by clearing only if there is a more recent successful monitor or * stop result, but we don't even have that information at this point * since we are still unpacking the resource's operation history. * * This is especially important for remote connection resources with a * reconnect interval, so in that case, we skip clearing failures * if the remote node hasn't been fenced. */ if (rsc->remote_reconnect_ms && pcmk_is_set(data_set->flags, pe_flag_stonith_enabled) && (interval_ms != 0) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { pe_node_t *remote_node = pe_find_node(data_set->nodes, rsc->id); if (remote_node && !remote_node->details->remote_was_fenced) { if (is_last_failure) { crm_info("Waiting to clear monitor failure for remote node %s" " until fencing has occurred", rsc->id); } return TRUE; } } return FALSE; } /*! * \internal * \brief Check operation age and schedule failure clearing when appropriate * * This function has two distinct purposes. The first is to check whether an * operation history entry is expired (i.e. the resource has a failure timeout, * the entry is older than the timeout, and the resource either has no fail * count or its fail count is entirely older than the timeout). The second is to * schedule fail count clearing when appropriate (i.e. the operation is expired * and either the resource has an expired fail count or the operation is a * last_failure for a remote connection resource with a reconnect interval, * or the operation is a last_failure for a start or monitor operation and the * resource's parameters have changed since the operation). * * \param[in] rsc Resource that operation happened to * \param[in] node Node that operation happened on * \param[in] rc Actual result of operation * \param[in] xml_op Operation history entry XML * \param[in] data_set Current working set * * \return TRUE if operation history entry is expired, FALSE otherwise */ static bool check_operation_expiry(pe_resource_t *rsc, pe_node_t *node, int rc, xmlNode *xml_op, pe_working_set_t *data_set) { bool expired = FALSE; bool is_last_failure = pcmk__ends_with(ID(xml_op), "_last_failure_0"); time_t last_run = 0; guint interval_ms = 0; int unexpired_fail_count = 0; const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *clear_reason = NULL; crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((rsc->failure_timeout > 0) && (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE, &last_run) == 0)) { // Resource has a failure-timeout, and history entry has a timestamp time_t now = get_effective_time(data_set); time_t last_failure = 0; // Is this particular operation history older than the failure timeout? if ((now >= (last_run + rsc->failure_timeout)) && !should_ignore_failure_timeout(rsc, xml_op, task, interval_ms, is_last_failure, data_set)) { expired = TRUE; } // Does the resource as a whole have an unexpired fail count? unexpired_fail_count = pe_get_failcount(node, rsc, &last_failure, pe_fc_effective, xml_op, data_set); // Update scheduler recheck time according to *last* failure crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d timeout=%ds" " last-failure@%lld", ID(xml_op), (long long) last_run, (expired? "" : "not "), (long long) now, unexpired_fail_count, rsc->failure_timeout, (long long) last_failure); last_failure += rsc->failure_timeout + 1; if (unexpired_fail_count && (now < last_failure)) { pe__update_recheck_time(last_failure, data_set); } } if (expired) { if (pe_get_failcount(node, rsc, NULL, pe_fc_default, xml_op, data_set)) { // There is a fail count ignoring timeout if (unexpired_fail_count == 0) { // There is no fail count considering timeout clear_reason = "it expired"; } else { /* This operation is old, but there is an unexpired fail count. * In a properly functioning cluster, this should only be * possible if this operation is not a failure (otherwise the * fail count should be expired too), so this is really just a * failsafe. */ expired = FALSE; } } else if (is_last_failure && rsc->remote_reconnect_ms) { /* Clear any expired last failure when reconnect interval is set, * even if there is no fail count. */ clear_reason = "reconnect interval is set"; } } if (!expired && is_last_failure && should_clear_for_param_change(xml_op, task, rsc, node, data_set)) { clear_reason = "resource parameters have changed"; } if (clear_reason != NULL) { // Schedule clearing of the fail count pe_action_t *clear_op = pe__clear_failcount(rsc, node, clear_reason, data_set); if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled) && rsc->remote_reconnect_ms) { /* If we're clearing a remote connection due to a reconnect * interval, we want to wait until any scheduled fencing * completes. * * We could limit this to remote_node->details->unclean, but at * this point, that's always true (it won't be reliable until * after unpack_node_history() is done). */ crm_info("Clearing %s failure will wait until any scheduled " "fencing of %s completes", task, rsc->id); order_after_remote_fencing(clear_op, rsc, data_set); } } if (expired && (interval_ms == 0) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { switch(rc) { case PCMK_OCF_OK: case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_PROMOTED: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: // Don't expire probes that return these values expired = FALSE; break; } } return expired; } int pe__target_rc_from_xml(xmlNode *xml_op) { int target_rc = 0; const char *key = crm_element_value(xml_op, XML_ATTR_TRANSITION_KEY); if (key == NULL) { return -1; } decode_transition_key(key, NULL, NULL, NULL, &target_rc); return target_rc; } static enum action_fail_response get_action_on_fail(pe_resource_t *rsc, const char *key, const char *task, pe_working_set_t * data_set) { enum action_fail_response result = action_fail_recover; pe_action_t *action = custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set); result = action->on_fail; pe_free_action(action); return result; } static void update_resource_state(pe_resource_t * rsc, pe_node_t * node, xmlNode * xml_op, const char * task, int rc, xmlNode * last_failure, enum action_fail_response * on_fail, pe_working_set_t * data_set) { gboolean clear_past_failure = FALSE; CRM_ASSERT(rsc); CRM_ASSERT(xml_op); if (rc == PCMK_OCF_NOT_RUNNING) { clear_past_failure = TRUE; } else if (rc == PCMK_OCF_NOT_INSTALLED) { rsc->role = RSC_ROLE_STOPPED; } else if (pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { if (last_failure) { const char *op_key = get_op_key(xml_op); const char *last_failure_key = get_op_key(last_failure); if (pcmk__str_eq(op_key, last_failure_key, pcmk__str_casei)) { clear_past_failure = TRUE; } } if (rsc->role < RSC_ROLE_STARTED) { set_active(rsc); } } else if (pcmk__str_eq(task, CRMD_ACTION_START, pcmk__str_casei)) { rsc->role = RSC_ROLE_STARTED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_STOP, pcmk__str_casei)) { rsc->role = RSC_ROLE_STOPPED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_PROMOTE, pcmk__str_casei)) { rsc->role = RSC_ROLE_PROMOTED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_DEMOTE, pcmk__str_casei)) { if (*on_fail == action_fail_demote) { // Demote clears an error only if on-fail=demote clear_past_failure = TRUE; } rsc->role = RSC_ROLE_UNPROMOTED; } else if (pcmk__str_eq(task, CRMD_ACTION_MIGRATED, pcmk__str_casei)) { rsc->role = RSC_ROLE_STARTED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_MIGRATE, pcmk__str_casei)) { unpack_migrate_to_success(rsc, node, xml_op, data_set); } else if (rsc->role < RSC_ROLE_STARTED) { pe_rsc_trace(rsc, "%s active on %s", rsc->id, node->details->uname); set_active(rsc); } /* clear any previous failure actions */ if (clear_past_failure) { switch (*on_fail) { case action_fail_stop: case action_fail_fence: case action_fail_migrate: case action_fail_standby: pe_rsc_trace(rsc, "%s.%s is not cleared by a completed stop", rsc->id, fail2text(*on_fail)); break; case action_fail_block: case action_fail_ignore: case action_fail_demote: case action_fail_recover: case action_fail_restart_container: *on_fail = action_fail_ignore; pe__set_next_role(rsc, RSC_ROLE_UNKNOWN, "clear past failures"); break; case action_fail_reset_remote: if (rsc->remote_reconnect_ms == 0) { /* With no reconnect interval, the connection is allowed to * start again after the remote node is fenced and * completely stopped. (With a reconnect interval, we wait * for the failure to be cleared entirely before attempting * to reconnect.) */ *on_fail = action_fail_ignore; pe__set_next_role(rsc, RSC_ROLE_UNKNOWN, "clear past failures and reset remote"); } break; } } } /*! * \internal * \brief Remap informational monitor results to usual values * * Certain OCF result codes are for providing extended information to the * user about services that aren't yet failed but not entirely healthy either. - * These must be treated as the "normal" result by pacemaker. + * These must be treated as the "normal" result by Pacemaker. * * \param[in] rc Actual result of a monitor action * \param[in] xml_op Operation history XML * \param[in] node Node that operation happened on * \param[in] rsc Resource that operation happened to * \param[in] data_set Cluster working set * * \return Result code that pacemaker should use * * \note If the result is remapped, and the node is not shutting down or failed, * the operation will be recorded in the data set's list of failed * operations, to highlight it for the user. */ static int remap_monitor_rc(int rc, xmlNode *xml_op, const pe_node_t *node, const pe_resource_t *rsc, pe_working_set_t *data_set) { int remapped_rc = pcmk__effective_rc(rc); if (rc != remapped_rc) { crm_trace("Remapping monitor result %d to %d", rc, remapped_rc); if (!node->details->shutdown || node->details->online) { record_failed_op(xml_op, node, rsc, data_set); } } return remapped_rc; } static void unpack_rsc_op(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *on_fail, pe_working_set_t *data_set) { int rc = 0; int task_id = 0; int target_rc = 0; int status = PCMK_LRM_OP_UNKNOWN; guint interval_ms = 0; const char *task = NULL; const char *task_key = NULL; const char *exit_reason = NULL; bool expired = FALSE; pe_resource_t *parent = rsc; enum action_fail_response failure_strategy = action_fail_recover; CRM_CHECK(rsc && node && xml_op, return); target_rc = pe__target_rc_from_xml(xml_op); task_key = get_op_key(xml_op); task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON); if (exit_reason == NULL) { exit_reason = ""; } crm_element_value_int(xml_op, XML_LRM_ATTR_RC, &rc); crm_element_value_int(xml_op, XML_LRM_ATTR_CALLID, &task_id); crm_element_value_int(xml_op, XML_LRM_ATTR_OPSTATUS, &status); crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); CRM_CHECK(task != NULL, return); CRM_CHECK(status <= PCMK_LRM_OP_INVALID, return); CRM_CHECK(status >= PCMK_LRM_OP_PENDING, return); if (!strcmp(task, CRMD_ACTION_NOTIFY) || !strcmp(task, CRMD_ACTION_METADATA)) { /* safe to ignore these */ return; } if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) { parent = uber_parent(rsc); } pe_rsc_trace(rsc, "Unpacking task %s/%s (call_id=%d, status=%d, rc=%d) on %s (role=%s)", task_key, task, task_id, status, rc, node->details->uname, role2text(rsc->role)); if (node->details->unclean) { pe_rsc_trace(rsc, "Node %s (where %s is running) is unclean." " Further action depends on the value of the stop's on-fail attribute", node->details->uname, rsc->id); } /* It should be possible to call remap_monitor_rc() first then call * check_operation_expiry() only if rc != target_rc, because there should * never be a fail count without at least one unexpected result in the * resource history. That would be more efficient by avoiding having to call * check_operation_expiry() for expected results. * * However, we do have such configurations in the scheduler regression * tests, even if it shouldn't be possible with the current code. It's * probably a good idea anyway, but that would require updating the test * inputs to something currently possible. */ if ((status != PCMK_LRM_OP_NOT_INSTALLED) && check_operation_expiry(rsc, node, rc, xml_op, data_set)) { expired = TRUE; } if (!strcmp(task, CRMD_ACTION_STATUS)) { rc = remap_monitor_rc(rc, xml_op, node, rsc, data_set); } if (expired && (rc != target_rc)) { const char *magic = crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC); if (interval_ms == 0) { crm_notice("Ignoring expired %s failure on %s " CRM_XS " actual=%d expected=%d magic=%s", task_key, node->details->uname, rc, target_rc, magic); goto done; } else if(node->details->online && node->details->unclean == FALSE) { /* Reschedule the recurring monitor. CancelXmlOp() won't work at * this stage, so as a hacky workaround, forcibly change the restart * digest so check_action_definition() does what we want later. * * @TODO We should skip this if there is a newer successful monitor. * Also, this causes rescheduling only if the history entry * has an op-digest (which the expire-non-blocked-failure * scheduler regression test doesn't, but that may not be a * realistic scenario in production). */ crm_notice("Rescheduling %s after failure expired on %s " CRM_XS " actual=%d expected=%d magic=%s", task_key, node->details->uname, rc, target_rc, magic); crm_xml_add(xml_op, XML_LRM_ATTR_RESTART_DIGEST, "calculated-failure-timeout"); goto done; } } /* If the executor reported an operation status of anything but done or * error, consider that final. But for done or error, we know better whether * it should be treated as a failure or not, because we know the expected * result. */ if(status == PCMK_LRM_OP_DONE || status == PCMK_LRM_OP_ERROR) { status = determine_op_status(rsc, rc, target_rc, node, xml_op, on_fail, data_set); pe_rsc_trace(rsc, "Remapped %s status to %d", task_key, status); } switch (status) { case PCMK_LRM_OP_CANCELLED: // Should never happen pe_err("Resource history contains cancellation '%s' " "(%s of %s on %s at %s)", ID(xml_op), task, rsc->id, node->details->uname, last_change_str(xml_op)); break; case PCMK_LRM_OP_PENDING: if (!strcmp(task, CRMD_ACTION_START)) { pe__set_resource_flags(rsc, pe_rsc_start_pending); set_active(rsc); } else if (!strcmp(task, CRMD_ACTION_PROMOTE)) { rsc->role = RSC_ROLE_PROMOTED; } else if (!strcmp(task, CRMD_ACTION_MIGRATE) && node->details->unclean) { /* If a pending migrate_to action is out on a unclean node, * we have to force the stop action on the target. */ const char *migrate_target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); pe_node_t *target = pe_find_node(data_set->nodes, migrate_target); if (target) { stop_action(rsc, target, FALSE); } } if (rsc->pending_task == NULL) { if ((interval_ms != 0) || strcmp(task, CRMD_ACTION_STATUS)) { rsc->pending_task = strdup(task); rsc->pending_node = node; } else { /* Pending probes are not printed, even if pending * operations are requested. If someone ever requests that * behavior, enable the below and the corresponding part of * native.c:native_pending_task(). */ #if 0 rsc->pending_task = strdup("probe"); rsc->pending_node = node; #endif } } break; case PCMK_LRM_OP_DONE: pe_rsc_trace(rsc, "%s of %s on %s completed at %s " CRM_XS " id=%s", task, rsc->id, node->details->uname, last_change_str(xml_op), ID(xml_op)); update_resource_state(rsc, node, xml_op, task, rc, *last_failure, on_fail, data_set); break; case PCMK_LRM_OP_NOT_INSTALLED: failure_strategy = get_action_on_fail(rsc, task_key, task, data_set); if (failure_strategy == action_fail_ignore) { crm_warn("Cannot ignore failed %s of %s on %s: " "Resource agent doesn't exist " CRM_XS " status=%d rc=%d id=%s", task, rsc->id, node->details->uname, status, rc, ID(xml_op)); /* Also for printing it as "FAILED" by marking it as pe_rsc_failed later */ *on_fail = action_fail_migrate; } resource_location(parent, node, -INFINITY, "hard-error", data_set); unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set); break; case PCMK_LRM_OP_NOT_CONNECTED: if (pe__is_guest_or_remote_node(node) && pcmk_is_set(node->details->remote_rsc->flags, pe_rsc_managed)) { /* We should never get into a situation where a managed remote * connection resource is considered OK but a resource action * behind the connection gets a "not connected" status. But as a * fail-safe in case a bug or unusual circumstances do lead to * that, ensure the remote connection is considered failed. */ pe__set_resource_flags(node->details->remote_rsc, pe_rsc_failed|pe_rsc_stop); } // fall through case PCMK_LRM_OP_ERROR: case PCMK_LRM_OP_ERROR_HARD: case PCMK_LRM_OP_ERROR_FATAL: case PCMK_LRM_OP_TIMEOUT: case PCMK_LRM_OP_NOTSUPPORTED: case PCMK_LRM_OP_INVALID: failure_strategy = get_action_on_fail(rsc, task_key, task, data_set); if ((failure_strategy == action_fail_ignore) || (failure_strategy == action_fail_restart_container && !strcmp(task, CRMD_ACTION_STOP))) { crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s " "succeeded " CRM_XS " rc=%d id=%s", task, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, rsc->id, node->details->uname, last_change_str(xml_op), rc, ID(xml_op)); update_resource_state(rsc, node, xml_op, task, target_rc, *last_failure, on_fail, data_set); crm_xml_add(xml_op, XML_ATTR_UNAME, node->details->uname); pe__set_resource_flags(rsc, pe_rsc_failure_ignored); record_failed_op(xml_op, node, rsc, data_set); if ((failure_strategy == action_fail_restart_container) && cmp_on_fail(*on_fail, action_fail_recover) <= 0) { *on_fail = failure_strategy; } } else { unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set); if(status == PCMK_LRM_OP_ERROR_HARD) { do_crm_log(rc != PCMK_OCF_NOT_INSTALLED?LOG_ERR:LOG_NOTICE, "Preventing %s from restarting on %s because " "of hard failure (%s%s%s)" CRM_XS " rc=%d id=%s", parent->id, node->details->uname, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, rc, ID(xml_op)); resource_location(parent, node, -INFINITY, "hard-error", data_set); } else if(status == PCMK_LRM_OP_ERROR_FATAL) { crm_err("Preventing %s from restarting anywhere because " "of fatal failure (%s%s%s) " CRM_XS " rc=%d id=%s", parent->id, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, rc, ID(xml_op)); resource_location(parent, NULL, -INFINITY, "fatal-error", data_set); } } break; } done: pe_rsc_trace(rsc, "Resource %s after %s: role=%s, next=%s", rsc->id, task, role2text(rsc->role), role2text(rsc->next_role)); } static void add_node_attrs(xmlNode *xml_obj, pe_node_t *node, bool overwrite, pe_working_set_t *data_set) { const char *cluster_name = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_UNAME), strdup(node->details->uname)); g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_ID), strdup(node->details->id)); if (pcmk__str_eq(node->details->id, data_set->dc_uuid, pcmk__str_casei)) { data_set->dc_node = node; node->details->is_dc = TRUE; g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_TRUE)); } else { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_FALSE)); } cluster_name = g_hash_table_lookup(data_set->config_hash, "cluster-name"); if (cluster_name) { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_CLUSTER_NAME), strdup(cluster_name)); } pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_ATTR_SETS, &rule_data, node->details->attrs, NULL, overwrite, data_set); if (pe_node_attribute_raw(node, CRM_ATTR_SITE_NAME) == NULL) { const char *site_name = pe_node_attribute_raw(node, "site-name"); if (site_name) { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_SITE_NAME), strdup(site_name)); } else if (cluster_name) { /* Default to cluster-name if unset */ g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_SITE_NAME), strdup(cluster_name)); } } } static GList * extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter) { int counter = -1; int stop_index = -1; int start_index = -1; xmlNode *rsc_op = NULL; GList *gIter = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; /* extract operations */ op_list = NULL; sorted_op_list = NULL; for (rsc_op = pcmk__xe_first_child(rsc_entry); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) { if (pcmk__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, pcmk__str_none)) { crm_xml_add(rsc_op, "resource", rsc); crm_xml_add(rsc_op, XML_ATTR_UNAME, node); op_list = g_list_prepend(op_list, rsc_op); } } if (op_list == NULL) { /* if there are no operations, there is nothing to do */ return NULL; } sorted_op_list = g_list_sort(op_list, sort_op_by_callid); /* create active recurring operations as optional */ if (active_filter == FALSE) { return sorted_op_list; } op_list = NULL; calculate_active_ops(sorted_op_list, &start_index, &stop_index); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; if (start_index < stop_index) { crm_trace("Skipping %s: not active", ID(rsc_entry)); break; } else if (counter < start_index) { crm_trace("Skipping %s: old", ID(rsc_op)); continue; } op_list = g_list_append(op_list, rsc_op); } g_list_free(sorted_op_list); return op_list; } GList * find_operations(const char *rsc, const char *node, gboolean active_filter, pe_working_set_t * data_set) { GList *output = NULL; GList *intermediate = NULL; xmlNode *tmp = NULL; xmlNode *status = find_xml_node(data_set->input, XML_CIB_TAG_STATUS, TRUE); pe_node_t *this_node = NULL; xmlNode *node_state = NULL; for (node_state = pcmk__xe_first_child(status); node_state != NULL; node_state = pcmk__xe_next(node_state)) { if (pcmk__str_eq((const char *)node_state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { const char *uname = crm_element_value(node_state, XML_ATTR_UNAME); if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) { continue; } this_node = pe_find_node(data_set->nodes, uname); if(this_node == NULL) { CRM_LOG_ASSERT(this_node != NULL); continue; } else if (pe__is_guest_or_remote_node(this_node)) { determine_remote_online_status(data_set, this_node); } else { determine_online_status(node_state, this_node, data_set); } if (this_node->details->online || pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { /* offline nodes run no resources... * unless stonith is enabled in which case we need to * make sure rsc start events happen after the stonith */ xmlNode *lrm_rsc = NULL; tmp = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE); tmp = find_xml_node(tmp, XML_LRM_TAG_RESOURCES, FALSE); for (lrm_rsc = pcmk__xe_first_child(tmp); lrm_rsc != NULL; lrm_rsc = pcmk__xe_next(lrm_rsc)) { if (pcmk__str_eq((const char *)lrm_rsc->name, XML_LRM_TAG_RESOURCE, pcmk__str_none)) { const char *rsc_id = crm_element_value(lrm_rsc, XML_ATTR_ID); if (rsc != NULL && !pcmk__str_eq(rsc_id, rsc, pcmk__str_casei)) { continue; } intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter); output = g_list_concat(output, intermediate); } } } } } return output; } diff --git a/tools/crm_rule.c b/tools/crm_rule.c index 0880415f5e..6b2694a6c5 100644 --- a/tools/crm_rule.c +++ b/tools/crm_rule.c @@ -1,371 +1,372 @@ /* * Copyright 2019-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #define SUMMARY "evaluate rules from the Pacemaker configuration" enum crm_rule_mode { crm_rule_mode_none, crm_rule_mode_check }; struct { char *date; char *input_xml; enum crm_rule_mode mode; char *rule; } options = { .mode = crm_rule_mode_none }; static int crm_rule_check(pe_working_set_t *data_set, const char *rule_id, crm_time_t *effective_date); static gboolean mode_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); static GOptionEntry mode_entries[] = { { "check", 'c', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, mode_cb, "Check whether a rule is in effect", NULL }, { NULL } }; static GOptionEntry data_entries[] = { { "xml-text", 'X', 0, G_OPTION_ARG_STRING, &options.input_xml, "Use argument for XML (or stdin if '-')", NULL }, { NULL } }; static GOptionEntry addl_entries[] = { { "date", 'd', 0, G_OPTION_ARG_STRING, &options.date, "Whether the rule is in effect on a given date", NULL }, { "rule", 'r', 0, G_OPTION_ARG_STRING, &options.rule, "The ID of the rule to check", NULL }, { NULL } }; static gboolean mode_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (strcmp(option_name, "c")) { options.mode = crm_rule_mode_check; } return TRUE; } /*! * \internal * \brief Evaluate a date expression for a specific time * * \param[in] time_expr date_expression XML * \param[in] now Time for which to evaluate expression * \param[out] next_change If not NULL, set to when evaluation will change * * \return Standard Pacemaker return code */ static int eval_date_expression(xmlNode *expr, crm_time_t *now, crm_time_t *next_change) { pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; return pe__eval_date_expr(expr, &rule_data, next_change); } static int crm_rule_check(pe_working_set_t *data_set, const char *rule_id, crm_time_t *effective_date) { xmlNode *cib_constraints = NULL; xmlNode *match = NULL; xmlXPathObjectPtr xpathObj = NULL; char *xpath = NULL; int rc = pcmk_rc_ok; int max = 0; /* Rules are under the constraints node in the XML, so first find that. */ cib_constraints = get_object_root(XML_CIB_TAG_CONSTRAINTS, data_set->input); /* Get all rules matching the given ID which are also simple enough for us to check. * For the moment, these rules must only have a single date_expression child and: * - Do not have a date_spec operation, or * - Have a date_spec operation that contains years= but does not contain moon=. * * We do this in steps to provide better error messages. First, check that there's * any rule with the given ID. */ xpath = crm_strdup_printf("//rule[@id='%s']", rule_id); xpathObj = xpath_search(cib_constraints, xpath); max = numXpathResults(xpathObj); if (max == 0) { CMD_ERR("No rule found with ID=%s", rule_id); rc = ENXIO; goto done; } else if (max > 1) { CMD_ERR("More than one rule with ID=%s found", rule_id); rc = ENXIO; goto done; } free(xpath); freeXpathObject(xpathObj); /* Next, make sure it has exactly one date_expression. */ xpath = crm_strdup_printf("//rule[@id='%s']//date_expression", rule_id); xpathObj = xpath_search(cib_constraints, xpath); max = numXpathResults(xpathObj); if (max != 1) { CMD_ERR("Can't check rule %s because it does not have exactly one date_expression", rule_id); rc = EOPNOTSUPP; goto done; } free(xpath); freeXpathObject(xpathObj); /* Then, check that it's something we actually support. */ xpath = crm_strdup_printf("//rule[@id='%s']//date_expression[@operation!='date_spec']", rule_id); xpathObj = xpath_search(cib_constraints, xpath); max = numXpathResults(xpathObj); if (max == 0) { free(xpath); freeXpathObject(xpathObj); xpath = crm_strdup_printf("//rule[@id='%s']//date_expression[@operation='date_spec' and date_spec/@years and not(date_spec/@moon)]", rule_id); xpathObj = xpath_search(cib_constraints, xpath); max = numXpathResults(xpathObj); if (max == 0) { CMD_ERR("Rule either must not use date_spec, or use date_spec with years= but not moon="); rc = ENXIO; goto done; } } match = getXpathResult(xpathObj, 0); /* We should have ensured both of these pass with the xpath query above, but * double checking can't hurt. */ CRM_ASSERT(match != NULL); CRM_ASSERT(find_expression_type(match) == time_expr); rc = eval_date_expression(match, effective_date, NULL); if (rc == pcmk_rc_within_range) { printf("Rule %s is still in effect\n", rule_id); rc = pcmk_rc_ok; } else if (rc == pcmk_rc_ok) { printf("Rule %s satisfies conditions\n", rule_id); } else if (rc == pcmk_rc_after_range) { printf("Rule %s is expired\n", rule_id); } else if (rc == pcmk_rc_before_range) { printf("Rule %s has not yet taken effect\n", rule_id); } else if (rc == pcmk_rc_op_unsatisfied) { printf("Rule %s does not satisfy conditions\n", rule_id); } else { printf("Could not determine whether rule %s is expired\n", rule_id); } done: free(xpath); freeXpathObject(xpathObj); return rc; } static GOptionContext * build_arg_context(pcmk__common_args_t *args) { GOptionContext *context = NULL; const char *description = "This tool is currently experimental.\n" - "The interface, behavior, and output may change with any version of pacemaker."; + "The interface, behavior, and output may change " + "with any version of Pacemaker."; context = pcmk__build_arg_context(args, NULL, NULL, NULL); g_option_context_set_description(context, description); pcmk__add_arg_group(context, "modes", "Modes (mutually exclusive):", "Show modes of operation", mode_entries); pcmk__add_arg_group(context, "data", "Data:", "Show data options", data_entries); pcmk__add_arg_group(context, "additional", "Additional Options:", "Show additional options", addl_entries); return context; } int main(int argc, char **argv) { cib_t *cib_conn = NULL; pe_working_set_t *data_set = NULL; crm_time_t *rule_date = NULL; xmlNode *input = NULL; int rc = pcmk_ok; crm_exit_t exit_code = CRM_EX_OK; GError *error = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); GOptionContext *context = build_arg_context(args); gchar **processed_args = pcmk__cmdline_preproc(argv, "drX"); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } pcmk__cli_init_logging("crm_rule", args->verbosity); if (args->version) { g_strfreev(processed_args); pcmk__free_arg_context(context); /* FIXME: When crm_rule is converted to use formatted output, this can go. */ pcmk__cli_help('v', CRM_EX_OK); } /* Check command line arguments before opening a connection to * the CIB manager or doing anything else important. */ switch(options.mode) { case crm_rule_mode_check: if (options.rule == NULL) { CMD_ERR("--check requires use of --rule="); exit_code = CRM_EX_USAGE; goto done; } break; default: CMD_ERR("No mode operation given"); exit_code = CRM_EX_USAGE; goto done; break; } /* Set up some defaults. */ rule_date = crm_time_new(options.date); if (rule_date == NULL) { CMD_ERR("No --date given and can't determine current date"); exit_code = CRM_EX_DATAERR; goto done; } /* Where does the XML come from? If one of various command line options were * given, use those. Otherwise, connect to the CIB and use that. */ if (pcmk__str_eq(options.input_xml, "-", pcmk__str_casei)) { input = stdin2xml(); if (input == NULL) { CMD_ERR("Couldn't parse input from STDIN\n"); exit_code = CRM_EX_DATAERR; goto done; } } else if (options.input_xml != NULL) { input = string2xml(options.input_xml); if (input == NULL) { CMD_ERR("Couldn't parse input string: %s\n", options.input_xml); exit_code = CRM_EX_DATAERR; goto done; } } else { // Establish a connection to the CIB cib_conn = cib_new(); rc = cib_conn->cmds->signon(cib_conn, crm_system_name, cib_command); if (rc != pcmk_ok) { CMD_ERR("Could not connect to CIB: %s", pcmk_strerror(rc)); exit_code = crm_errno2exit(rc); goto done; } } /* Populate working set from CIB query */ if (input == NULL) { rc = cib_conn->cmds->query(cib_conn, NULL, &input, cib_scope_local | cib_sync_call); if (rc != pcmk_ok) { exit_code = crm_errno2exit(rc); goto done; } } /* Populate the working set instance */ data_set = pe_new_working_set(); if (data_set == NULL) { exit_code = crm_errno2exit(ENOMEM); goto done; } pe__set_working_set_flags(data_set, pe_flag_no_counts|pe_flag_no_compat); data_set->input = input; data_set->now = rule_date; /* Unpack everything. */ cluster_status(data_set); /* Now do whichever operation mode was asked for. There's only one at the * moment so this looks a little silly, but I expect there will be more * modes in the future. */ switch(options.mode) { case crm_rule_mode_check: rc = crm_rule_check(data_set, options.rule, rule_date); if (rc > 0) { CMD_ERR("Error checking rule: %s", pcmk_rc_str(rc)); } exit_code = pcmk_rc2exitc(rc); break; default: break; } done: if (cib_conn != NULL) { cib_conn->cmds->signoff(cib_conn); cib_delete(cib_conn); } g_strfreev(processed_args); pcmk__free_arg_context(context); pe_free_working_set(data_set); pcmk__output_and_clear_error(error, NULL); crm_exit(exit_code); }