diff --git a/cts/CTStests.py b/cts/CTStests.py index 3cb5fae310..3f6c7b123b 100644 --- a/cts/CTStests.py +++ b/cts/CTStests.py @@ -1,3124 +1,3124 @@ """ Test-specific classes for Pacemaker's Cluster Test Suite (CTS) """ -__copyright__ = "Copyright 2000-2020 the Pacemaker project contributors" +__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)) + 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: " + 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) devnull = open(os.devnull, 'wb') subprocess.check_call(["dd", "if=/dev/urandom", "of=%s" % keyfile, "bs=4096", "count=1"], stdout=devnull, stderr=devnull) devnull.close() # 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() 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 AllTestClasses.append(RemoteRscFailure) # vim:ts=4:sw=4:et: diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c index 0122e2baa6..5adb38af45 100644 --- a/daemons/controld/controld_execd.c +++ b/daemons/controld/controld_execd.c @@ -1,2854 +1,2855 @@ /* - * 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 // 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, "reload", CRMD_ACTION_STATUS, NULL)) { if (entry->stop_params) { g_hash_table_destroy(entry->stop_params); } entry->stop_params = crm_str_table_new(); 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; } } /* 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"); } 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; } static char * build_parameter_list(const lrmd_event_data_t *op, const struct ra_metadata_s *metadata, xmlNode *result, enum ra_param_flags_e param_type, bool invert_for_xml) { char *list = NULL; size_t len = 0; for (GList *iter = metadata->ra_params; iter != NULL; iter = iter->next) { struct ra_param_s *param = (struct ra_param_s *) iter->data; bool accept = pcmk_is_set(param->rap_flags, param_type); if (accept) { 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 (result && (invert_for_xml? !accept : accept)) { 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)) { restart = create_xml_node(NULL, XML_TAG_PARAMS); /* Add any parameters with unique="1" to the "op-force-restart" list. * * (Currently, we abuse "unique=0" to indicate reloadability. This is * nonstandard and should eventually be replaced once the OCF standard * is updated with something better.) */ list = build_parameter_list(op, metadata, restart, ra_param_unique, FALSE); } else { /* Resource does not support 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 */ secure = create_xml_node(NULL, XML_TAG_PARAMS); list = build_parameter_list(op, metadata, secure, ra_param_private, TRUE); 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; 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; } metadata = metadata_cache_get(lrm_state->metadata_cache, rsc); if (metadata == NULL) { /* For now, we always collect resource agent meta-data via a local, * synchronous, direct execution of the agent. This has multiple issues: * the executor should execute agents, not the controller; meta-data for * Pacemaker Remote nodes should be collected on those nodes, not * locally; and the meta-data call shouldn't eat into the timeout of the * real action being performed. * * These issues are planned to be addressed by having the scheduler * schedule a meta-data cache check at the beginning of each transition. * Once that is working, this block will only be a fallback in case the * initial collection fails. */ char *metadata_str = NULL; int rc = lrm_state_get_metadata(lrm_state, rsc->standard, rsc->provider, rsc->type, &metadata_str, 0); if (rc != pcmk_ok) { crm_warn("Failed to get metadata for %s (%s:%s:%s)", rsc->id, rsc->standard, rsc->provider, rsc->type); return TRUE; } metadata = metadata_cache_update(lrm_state->metadata_cache, rsc, metadata_str); free(metadata_str); if (metadata == NULL) { crm_warn("Failed to update metadata for %s (%s:%s:%s)", rsc->id, rsc->standard, rsc->provider, rsc->type); 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); + 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 ENABLE_ACL 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; } #endif 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")); call = crm_parse_int(call_id, "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; #if ENABLE_ACL 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; } #endif 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); #if ENABLE_ACL user_name = pcmk__update_acl_user(input->msg, F_CRM_USER, NULL); #endif 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); } #if ENABLE_ACL crm_trace("Executor %s command from %s as user %s", crm_op, from_sys, user_name); #else crm_trace("Executor %s command from %s", crm_op, from_sys); #endif 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 { 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 = metadata_cache_get(lrm_state->metadata_cache, rsc); 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)) { op->timeout = crm_parse_int(value, "0"); } else if (pcmk__str_eq(key, XML_OP_ATTR_START_DELAY, pcmk__str_casei)) { op->start_delay = crm_parse_int(value, "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 = crm_str_table_new(); 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); op->start_delay = crm_parse_int(op_delay, "0"); op_timeout = crm_meta_value(params, XML_ATTR_TIMEOUT); op->timeout = crm_parse_int(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 = crm_str_table_new(); 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_MASTER: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_MASTER: // 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/lib/pengine/native.c b/lib/pengine/native.c index 95c3da905b..f8b5c6be88 100644 --- a/lib/pengine/native.c +++ b/lib/pengine/native.c @@ -1,1356 +1,1354 @@ /* - * 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 #define VARIANT_NATIVE 1 #include "./variant.h" /*! * \internal * \brief Check whether a resource is active on multiple nodes */ static bool is_multiply_active(pe_resource_t *rsc) { unsigned int count = 0; if (rsc->variant == pe_native) { pe__find_active_requires(rsc, &count); } return count > 1; } static void native_priority_to_node(pe_resource_t * rsc, pe_node_t * node) { int priority = 0; if (rsc->priority == 0) { return; } if (rsc->role == RSC_ROLE_MASTER) { // Promoted instance takes base priority + 1 priority = rsc->priority + 1; } else { priority = rsc->priority; } node->details->priority += priority; pe_rsc_trace(rsc, "Node '%s' now has priority %d with %s'%s' (priority: %d%s)", node->details->uname, node->details->priority, rsc->role == RSC_ROLE_MASTER ? "promoted " : "", rsc->id, rsc->priority, rsc->role == RSC_ROLE_MASTER ? " + 1" : ""); /* Priority of a resource running on a guest node is added to the cluster * node as well. */ if (node->details->remote_rsc && node->details->remote_rsc->container) { GListPtr gIter = node->details->remote_rsc->container->running_on; for (; gIter != NULL; gIter = gIter->next) { pe_node_t *a_node = gIter->data; a_node->details->priority += priority; pe_rsc_trace(rsc, "Node '%s' now has priority %d with %s'%s' (priority: %d%s) " "from guest node '%s'", a_node->details->uname, a_node->details->priority, rsc->role == RSC_ROLE_MASTER ? "promoted " : "", rsc->id, rsc->priority, rsc->role == RSC_ROLE_MASTER ? " + 1" : "", node->details->uname); } } } void native_add_running(pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set) { GListPtr gIter = rsc->running_on; CRM_CHECK(node != NULL, return); for (; gIter != NULL; gIter = gIter->next) { pe_node_t *a_node = (pe_node_t *) gIter->data; CRM_CHECK(a_node != NULL, return); if (pcmk__str_eq(a_node->details->id, node->details->id, pcmk__str_casei)) { return; } } pe_rsc_trace(rsc, "Adding %s to %s %s", rsc->id, node->details->uname, pcmk_is_set(rsc->flags, pe_rsc_managed)? "" : "(unmanaged)"); rsc->running_on = g_list_append(rsc->running_on, node); if (rsc->variant == pe_native) { node->details->running_rsc = g_list_append(node->details->running_rsc, rsc); native_priority_to_node(rsc, node); } if (rsc->variant == pe_native && node->details->maintenance) { pe__clear_resource_flags(rsc, pe_rsc_managed); } if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe_resource_t *p = rsc->parent; pe_rsc_info(rsc, "resource %s isn't managed", rsc->id); resource_location(rsc, node, INFINITY, "not_managed_default", data_set); while(p && node->details->online) { /* add without the additional location constraint */ p->running_on = g_list_append(p->running_on, node); p = p->parent; } return; } if (is_multiply_active(rsc)) { switch (rsc->recovery_type) { case recovery_stop_only: { GHashTableIter gIter; pe_node_t *local_node = NULL; /* make sure it doesn't come up again */ if (rsc->allowed_nodes != NULL) { g_hash_table_destroy(rsc->allowed_nodes); } rsc->allowed_nodes = pe__node_list2table(data_set->nodes); g_hash_table_iter_init(&gIter, rsc->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&local_node)) { local_node->weight = -INFINITY; } } break; case recovery_stop_start: break; case recovery_block: pe__clear_resource_flags(rsc, pe_rsc_managed); pe__set_resource_flags(rsc, pe_rsc_block); /* If the resource belongs to a group or bundle configured with * multiple-active=block, block the entire entity. */ if (rsc->parent && (rsc->parent->variant == pe_group || rsc->parent->variant == pe_container) && rsc->parent->recovery_type == recovery_block) { GListPtr gIter = rsc->parent->children; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child = (pe_resource_t *) gIter->data; pe__clear_resource_flags(child, pe_rsc_managed); pe__set_resource_flags(child, pe_rsc_block); } } break; } crm_debug("%s is active on multiple nodes including %s: %s", rsc->id, node->details->uname, recovery2text(rsc->recovery_type)); } else { pe_rsc_trace(rsc, "Resource %s is active on: %s", rsc->id, node->details->uname); } if (rsc->parent != NULL) { native_add_running(rsc->parent, node, data_set); } } static void recursive_clear_unique(pe_resource_t *rsc) { pe__clear_resource_flags(rsc, pe_rsc_unique); add_hash_param(rsc->meta, XML_RSC_ATTR_UNIQUE, XML_BOOLEAN_FALSE); for (GList *child = rsc->children; child != NULL; child = child->next) { recursive_clear_unique((pe_resource_t *) child->data); } } gboolean native_unpack(pe_resource_t * rsc, pe_working_set_t * data_set) { pe_resource_t *parent = uber_parent(rsc); native_variant_data_t *native_data = NULL; const char *standard = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); uint32_t ra_caps = pcmk_get_ra_caps(standard); pe_rsc_trace(rsc, "Processing resource %s...", rsc->id); native_data = calloc(1, sizeof(native_variant_data_t)); rsc->variant_opaque = native_data; // Only some agent standards support unique and promotable clones if (!pcmk_is_set(ra_caps, pcmk_ra_cap_unique) && pcmk_is_set(rsc->flags, pe_rsc_unique) && pe_rsc_is_clone(parent)) { /* @COMPAT We should probably reject this situation as an error (as we * do for promotable below) rather than warn and convert, but that would * be a backward-incompatible change that we should probably do with a * transform at a schema major version bump. */ pe__force_anon(standard, parent, rsc->id, data_set); /* Clear globally-unique on the parent and all its descendents unpacked * so far (clearing the parent should make any future children unpacking * correct). We have to clear this resource explicitly because it isn't * hooked into the parent's children yet. */ recursive_clear_unique(parent); recursive_clear_unique(rsc); } if (!pcmk_is_set(ra_caps, pcmk_ra_cap_promotable) && pcmk_is_set(parent->flags, pe_rsc_promotable)) { pe_err("Resource %s is of type %s and therefore " "cannot be used as a promotable clone resource", rsc->id, standard); return FALSE; } return TRUE; } static bool rsc_is_on_node(pe_resource_t *rsc, const pe_node_t *node, int flags) { pe_rsc_trace(rsc, "Checking whether %s is on %s", rsc->id, node->details->uname); if (pcmk_is_set(flags, pe_find_current) && rsc->running_on) { for (GListPtr iter = rsc->running_on; iter; iter = iter->next) { pe_node_t *loc = (pe_node_t *) iter->data; if (loc->details == node->details) { return TRUE; } } } else if (pcmk_is_set(flags, pe_find_inactive) && (rsc->running_on == NULL)) { return TRUE; } else if (!pcmk_is_set(flags, pe_find_current) && rsc->allocated_to && (rsc->allocated_to->details == node->details)) { return TRUE; } return FALSE; } pe_resource_t * native_find_rsc(pe_resource_t * rsc, const char *id, const pe_node_t *on_node, int flags) { bool match = FALSE; pe_resource_t *result = NULL; CRM_CHECK(id && rsc && rsc->id, return NULL); if (flags & pe_find_clone) { const char *rid = ID(rsc->xml); if (!pe_rsc_is_clone(uber_parent(rsc))) { match = FALSE; } else if (!strcmp(id, rsc->id) || pcmk__str_eq(id, rid, pcmk__str_casei)) { match = TRUE; } } else if (!strcmp(id, rsc->id)) { match = TRUE; } else if (pcmk_is_set(flags, pe_find_renamed) && rsc->clone_name && strcmp(rsc->clone_name, id) == 0) { match = TRUE; } else if (pcmk_is_set(flags, pe_find_any) || (pcmk_is_set(flags, pe_find_anon) && !pcmk_is_set(rsc->flags, pe_rsc_unique))) { match = pe_base_name_eq(rsc, id); } if (match && on_node) { bool match_node = rsc_is_on_node(rsc, on_node, flags); if (match_node == FALSE) { match = FALSE; } } if (match) { return rsc; } for (GListPtr gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child = (pe_resource_t *) gIter->data; result = rsc->fns->find_rsc(child, id, on_node, flags); if (result) { return result; } } return NULL; } // create is ignored char * native_parameter(pe_resource_t * rsc, pe_node_t * node, gboolean create, const char *name, pe_working_set_t * data_set) { char *value_copy = NULL; const char *value = NULL; GHashTable *params = NULL; CRM_CHECK(rsc != NULL, return NULL); CRM_CHECK(name != NULL && strlen(name) != 0, return NULL); pe_rsc_trace(rsc, "Looking up %s in %s", name, rsc->id); params = pe_rsc_params(rsc, node, data_set); value = g_hash_table_lookup(params, name); if (value == NULL) { /* try meta attributes instead */ value = g_hash_table_lookup(rsc->meta, name); } if (value != NULL) { value_copy = strdup(value); } return value_copy; } gboolean native_active(pe_resource_t * rsc, gboolean all) { for (GList *gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) { pe_node_t *a_node = (pe_node_t *) gIter->data; if (a_node->details->unclean) { pe_rsc_trace(rsc, "Resource %s: node %s is unclean", rsc->id, a_node->details->uname); return TRUE; } else if (a_node->details->online == FALSE) { pe_rsc_trace(rsc, "Resource %s: node %s is offline", rsc->id, a_node->details->uname); } else { pe_rsc_trace(rsc, "Resource %s active on %s", rsc->id, a_node->details->uname); return TRUE; } } return FALSE; } struct print_data_s { long options; void *print_data; }; static const char * native_pending_state(pe_resource_t * rsc) { const char *pending_state = NULL; if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_START, pcmk__str_casei)) { pending_state = "Starting"; } else if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_STOP, pcmk__str_casei)) { pending_state = "Stopping"; } else if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_MIGRATE, pcmk__str_casei)) { pending_state = "Migrating"; } else if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_MIGRATED, pcmk__str_casei)) { /* Work might be done in here. */ pending_state = "Migrating"; } else if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_PROMOTE, pcmk__str_casei)) { pending_state = "Promoting"; } else if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_DEMOTE, pcmk__str_casei)) { pending_state = "Demoting"; } return pending_state; } static const char * native_pending_task(pe_resource_t * rsc) { const char *pending_task = NULL; if (pcmk__str_eq(rsc->pending_task, CRMD_ACTION_STATUS, pcmk__str_casei)) { pending_task = "Monitoring"; /* Pending probes are not printed, even if pending * operations are requested. If someone ever requests that * behavior, uncomment this and the corresponding part of * unpack.c:unpack_rsc_op(). */ /* } else if (pcmk__str_eq(rsc->pending_task, "probe", pcmk__str_casei)) { pending_task = "Checking"; */ } return pending_task; } static enum rsc_role_e native_displayable_role(pe_resource_t *rsc) { enum rsc_role_e role = rsc->role; if ((role == RSC_ROLE_STARTED) && pcmk_is_set(uber_parent(rsc)->flags, pe_rsc_promotable)) { role = RSC_ROLE_SLAVE; } return role; } static const char * native_displayable_state(pe_resource_t *rsc, long options) { const char *rsc_state = NULL; if (options & pe_print_pending) { rsc_state = native_pending_state(rsc); } if (rsc_state == NULL) { rsc_state = role2text(native_displayable_role(rsc)); } return rsc_state; } static void native_print_xml(pe_resource_t * rsc, const char *pre_text, long options, void *print_data) { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER); const char *rsc_state = native_displayable_state(rsc, options); const char *target_role = NULL; /* resource information. */ status_print("%sxml, XML_ATTR_TYPE)); + class, ((prov == NULL)? "" : ":"), ((prov == NULL)? "" : prov), + crm_element_value(rsc->xml, XML_ATTR_TYPE)); status_print("role=\"%s\" ", rsc_state); if (rsc->meta) { target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); } if (target_role) { status_print("target_role=\"%s\" ", target_role); } status_print("active=\"%s\" ", pcmk__btoa(rsc->fns->active(rsc, TRUE))); status_print("orphaned=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_orphan)); status_print("blocked=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_block)); status_print("managed=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_managed)); status_print("failed=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_failed)); status_print("failure_ignored=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_failure_ignored)); status_print("nodes_running_on=\"%d\" ", g_list_length(rsc->running_on)); if (options & pe_print_pending) { const char *pending_task = native_pending_task(rsc); if (pending_task) { status_print("pending=\"%s\" ", pending_task); } } /* print out the nodes this resource is running on */ if (options & pe_print_rsconly) { status_print("/>\n"); /* do nothing */ } else if (rsc->running_on != NULL) { GListPtr gIter = rsc->running_on; status_print(">\n"); for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; status_print("%s \n", pre_text, node->details->uname, node->details->id, pcmk__btoa(node->details->online == FALSE)); } status_print("%s\n", pre_text); } else { status_print("/>\n"); } } // Append a flag to resource description string's flags list static bool add_output_flag(GString *s, const char *flag_desc, bool have_flags) { g_string_append(s, (have_flags? ", " : " (")); g_string_append(s, flag_desc); return true; } // Append a node name to resource description string's node list static bool add_output_node(GString *s, const char *node, bool have_nodes) { g_string_append(s, (have_nodes? " " : " [ ")); g_string_append(s, node); return true; } /*! * \internal * \brief Create a string description of a resource * * \param[in] rsc Resource to describe * \param[in] name Desired identifier for the resource * \param[in] node If not NULL, node that resource is "on" * \param[in] options Bitmask of pe_print_* * \param[in] target_role Resource's target role * \param[in] show_nodes Whether to display nodes when multiply active * * \return Newly allocated string description of resource * \note Caller must free the result with g_free(). */ gchar * pcmk__native_output_string(pe_resource_t *rsc, const char *name, pe_node_t *node, long options, const char *target_role, bool show_nodes) { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); const char *provider = NULL; const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE); gchar *retval = NULL; GString *outstr = NULL; bool have_flags = false; if (rsc->variant != pe_native) { return NULL; } CRM_CHECK(name != NULL, name = "unknown"); CRM_CHECK(kind != NULL, kind = "unknown"); CRM_CHECK(class != NULL, class = "unknown"); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) { provider = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER); } if ((node == NULL) && (rsc->lock_node != NULL)) { node = rsc->lock_node; } if (pcmk_is_set(options, pe_print_rsconly) || pcmk__list_of_multiple(rsc->running_on)) { node = NULL; } // We need a string of at least this size outstr = g_string_sized_new(strlen(name) + strlen(class) + strlen(kind) + (provider? (strlen(provider) + 2) : 0) + (node? strlen(node->details->uname) + 1 : 0) + 11); // Resource name and agent g_string_printf(outstr, "%s\t(%s%s%s:%s):\t", name, class, - /* @COMPAT This should be a single ':' (see CLBZ#5395) but - * to avoid breaking anything relying on it, we're keeping - * it like this until the next minor version bump. - */ - (provider? "::" : ""), (provider? provider : ""), kind); + ((provider == NULL)? "" : ":"), + ((provider == NULL)? "" : provider), kind); // State on node if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) { g_string_append(outstr, " ORPHANED"); } if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { enum rsc_role_e role = native_displayable_role(rsc); if (role > RSC_ROLE_SLAVE) { g_string_append_printf(outstr, " FAILED %s", role2text(role)); } else { g_string_append(outstr, " FAILED"); } } else { g_string_append_printf(outstr, " %s", native_displayable_state(rsc, options)); } if (node) { g_string_append_printf(outstr, " %s", node->details->uname); } // Flags, as: ( [...]) if (node && !(node->details->online) && node->details->unclean) { have_flags = add_output_flag(outstr, "UNCLEAN", have_flags); } if (node && (node == rsc->lock_node)) { have_flags = add_output_flag(outstr, "LOCKED", have_flags); } if (pcmk_is_set(options, pe_print_pending)) { const char *pending_task = native_pending_task(rsc); if (pending_task) { have_flags = add_output_flag(outstr, pending_task, have_flags); } } if (target_role) { enum rsc_role_e target_role_e = text2role(target_role); /* Only show target role if it limits our abilities (i.e. ignore * Started, as it is the default anyways, and doesn't prevent the * resource from becoming Master). */ if (target_role_e == RSC_ROLE_STOPPED) { have_flags = add_output_flag(outstr, "disabled", have_flags); } else if (pcmk_is_set(uber_parent(rsc)->flags, pe_rsc_promotable) && target_role_e == RSC_ROLE_SLAVE) { have_flags = add_output_flag(outstr, "target-role:", have_flags); g_string_append(outstr, target_role); } } if (pcmk_is_set(rsc->flags, pe_rsc_block)) { have_flags = add_output_flag(outstr, "blocked", have_flags); } else if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { have_flags = add_output_flag(outstr, "unmanaged", have_flags); } if (pcmk_is_set(rsc->flags, pe_rsc_failure_ignored)) { have_flags = add_output_flag(outstr, "failure ignored", have_flags); } if (have_flags) { g_string_append(outstr, ")"); } // User-supplied description if (pcmk_is_set(options, pe_print_rsconly) || pcmk__list_of_multiple(rsc->running_on)) { const char *desc = crm_element_value(rsc->xml, XML_ATTR_DESC); if (desc) { g_string_append_printf(outstr, " %s", desc); } } if (show_nodes && !pcmk_is_set(options, pe_print_rsconly) && pcmk__list_of_multiple(rsc->running_on)) { bool have_nodes = false; for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pe_node_t *n = (pe_node_t *) iter->data; have_nodes = add_output_node(outstr, n->details->uname, have_nodes); } if (have_nodes) { g_string_append(outstr, " ]"); } } retval = outstr->str; g_string_free(outstr, FALSE); return retval; } int pe__common_output_html(pcmk__output_t *out, pe_resource_t * rsc, const char *name, pe_node_t *node, long options) { const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE); const char *target_role = NULL; xmlNodePtr list_node = NULL; const char *cl = NULL; CRM_ASSERT(rsc->variant == pe_native); CRM_ASSERT(kind != NULL); if (rsc->meta) { const char *is_internal = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERNAL_RSC); if (crm_is_true(is_internal) && !pcmk_is_set(options, pe_print_implicit)) { crm_trace("skipping print of internal resource %s", rsc->id); return pcmk_rc_no_output; } target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); } if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { cl = "rsc-managed"; } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { cl = "rsc-failed"; } else if (rsc->variant == pe_native && (rsc->running_on == NULL)) { cl = "rsc-failed"; } else if (pcmk__list_of_multiple(rsc->running_on)) { cl = "rsc-multiple"; } else if (pcmk_is_set(rsc->flags, pe_rsc_failure_ignored)) { cl = "rsc-failure-ignored"; } else { cl = "rsc-ok"; } { gchar *s = pcmk__native_output_string(rsc, name, node, options, target_role, true); list_node = pcmk__output_create_html_node(out, "li", NULL, NULL, NULL); pcmk_create_html_node(list_node, "span", NULL, cl, s); g_free(s); } return pcmk_rc_ok; } int pe__common_output_text(pcmk__output_t *out, pe_resource_t * rsc, const char *name, pe_node_t *node, long options) { const char *target_role = NULL; CRM_ASSERT(rsc->variant == pe_native); if (rsc->meta) { const char *is_internal = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERNAL_RSC); if (crm_is_true(is_internal) && !pcmk_is_set(options, pe_print_implicit)) { crm_trace("skipping print of internal resource %s", rsc->id); return pcmk_rc_no_output; } target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); } { gchar *s = pcmk__native_output_string(rsc, name, node, options, target_role, true); out->list_item(out, NULL, "%s", s); g_free(s); } return pcmk_rc_ok; } void common_print(pe_resource_t * rsc, const char *pre_text, const char *name, pe_node_t *node, long options, void *print_data) { const char *target_role = NULL; CRM_ASSERT(rsc->variant == pe_native); if (rsc->meta) { const char *is_internal = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERNAL_RSC); if (crm_is_true(is_internal) && !pcmk_is_set(options, pe_print_implicit)) { crm_trace("skipping print of internal resource %s", rsc->id); return; } target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); } if (options & pe_print_xml) { native_print_xml(rsc, pre_text, options, print_data); return; } if ((pre_text == NULL) && (options & pe_print_printf)) { pre_text = " "; } if (options & pe_print_html) { if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { status_print(""); } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { status_print(""); } else if (rsc->running_on == NULL) { status_print(""); } else if (pcmk__list_of_multiple(rsc->running_on)) { status_print(""); } else if (pcmk_is_set(rsc->flags, pe_rsc_failure_ignored)) { status_print(""); } else { status_print(""); } } { gchar *resource_s = pcmk__native_output_string(rsc, name, node, options, target_role, false); status_print("%s%s", (pre_text? pre_text : ""), resource_s); g_free(resource_s); } #if CURSES_ENABLED if (pcmk_is_set(options, pe_print_ncurses) && !pcmk_is_set(options, pe_print_rsconly) && !pcmk__list_of_multiple(rsc->running_on)) { /* coverity[negative_returns] False positive */ move(-1, 0); } #endif if (pcmk_is_set(options, pe_print_html)) { status_print(" "); } if (!pcmk_is_set(options, pe_print_rsconly) && pcmk__list_of_multiple(rsc->running_on)) { GListPtr gIter = rsc->running_on; int counter = 0; if (options & pe_print_html) { status_print("
    \n"); } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print("["); } for (; gIter != NULL; gIter = gIter->next) { pe_node_t *n = (pe_node_t *) gIter->data; counter++; if (options & pe_print_html) { status_print("
  • \n%s", n->details->uname); } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print(" %s", n->details->uname); } else if ((options & pe_print_log)) { status_print("\t%d : %s", counter, n->details->uname); } else { status_print("%s", n->details->uname); } if (options & pe_print_html) { status_print("
    • \n"); } } if (options & pe_print_html) { status_print("
\n"); } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print(" ]"); } } if (options & pe_print_html) { status_print("
\n"); } else if (options & pe_print_suppres_nl) { /* nothing */ } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print("\n"); } } void native_print(pe_resource_t * rsc, const char *pre_text, long options, void *print_data) { pe_node_t *node = NULL; CRM_ASSERT(rsc->variant == pe_native); if (options & pe_print_xml) { native_print_xml(rsc, pre_text, options, print_data); return; } node = pe__current_node(rsc); if (node == NULL) { // This is set only if a non-probe action is pending on this node node = rsc->pending_node; } common_print(rsc, pre_text, rsc_printable_id(rsc), node, options, print_data); } PCMK__OUTPUT_ARGS("primitive", "unsigned int", "pe_resource_t *", "GList *", "GList *") int pe__resource_xml(pcmk__output_t *out, va_list args) { unsigned int options = va_arg(args, unsigned int); pe_resource_t *rsc = va_arg(args, pe_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER); const char *rsc_state = native_displayable_state(rsc, options); long is_print_pending = options & pe_print_pending; char ra_name[LINE_MAX]; char *nodes_running_on = NULL; char *priority = NULL; int rc = pcmk_rc_no_output; const char *target_role = NULL; if (rsc->meta != NULL) { target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); } CRM_ASSERT(rsc->variant == pe_native); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } /* resource information. */ - sprintf(ra_name, "%s%s%s:%s", class, prov ? "::" : "", prov ? prov : "" - , crm_element_value(rsc->xml, XML_ATTR_TYPE)); + sprintf(ra_name, "%s%s%s:%s", + class, ((prov == NULL)? "" : ":"), ((prov == NULL)? "" : prov), + crm_element_value(rsc->xml, XML_ATTR_TYPE)); nodes_running_on = crm_itoa(g_list_length(rsc->running_on)); priority = crm_ftoa(rsc->priority); rc = pe__name_and_nvpairs_xml(out, true, "resource", 12, "id", rsc_printable_id(rsc), "resource_agent", ra_name, "role", rsc_state, "target_role", target_role, "active", pcmk__btoa(rsc->fns->active(rsc, TRUE)), "orphaned", pe__rsc_bool_str(rsc, pe_rsc_orphan), "blocked", pe__rsc_bool_str(rsc, pe_rsc_block), "managed", pe__rsc_bool_str(rsc, pe_rsc_managed), "failed", pe__rsc_bool_str(rsc, pe_rsc_failed), "failure_ignored", pe__rsc_bool_str(rsc, pe_rsc_failure_ignored), "nodes_running_on", nodes_running_on, "pending", (is_print_pending? native_pending_task(rsc) : NULL)); free(priority); free(nodes_running_on); CRM_ASSERT(rc == pcmk_rc_ok); if (rsc->running_on != NULL) { GList *gIter = rsc->running_on; for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; rc = pe__name_and_nvpairs_xml(out, false, "node", 3, "name", node->details->uname, "id", node->details->id, "cached", pcmk__btoa(node->details->online)); CRM_ASSERT(rc == pcmk_rc_ok); } } pcmk__output_xml_pop_parent(out); return rc; } PCMK__OUTPUT_ARGS("primitive", "unsigned int", "pe_resource_t *", "GList *", "GList *") int pe__resource_html(pcmk__output_t *out, va_list args) { unsigned int options = va_arg(args, unsigned int); pe_resource_t *rsc = va_arg(args, pe_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); pe_node_t *node = pe__current_node(rsc); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } CRM_ASSERT(rsc->variant == pe_native); if (node == NULL) { // This is set only if a non-probe action is pending on this node node = rsc->pending_node; } return pe__common_output_html(out, rsc, rsc_printable_id(rsc), node, options); } PCMK__OUTPUT_ARGS("primitive", "unsigned int", "pe_resource_t *", "GList *", "GList *") int pe__resource_text(pcmk__output_t *out, va_list args) { unsigned int options = va_arg(args, unsigned int); pe_resource_t *rsc = va_arg(args, pe_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); pe_node_t *node = pe__current_node(rsc); CRM_ASSERT(rsc->variant == pe_native); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } if (node == NULL) { // This is set only if a non-probe action is pending on this node node = rsc->pending_node; } return pe__common_output_text(out, rsc, rsc_printable_id(rsc), node, options); } void native_free(pe_resource_t * rsc) { pe_rsc_trace(rsc, "Freeing resource action list (not the data)"); common_free(rsc); } enum rsc_role_e native_resource_state(const pe_resource_t * rsc, gboolean current) { enum rsc_role_e role = rsc->next_role; if (current) { role = rsc->role; } pe_rsc_trace(rsc, "%s state: %s", rsc->id, role2text(role)); return role; } /*! * \internal * \brief List nodes where a resource (or any of its children) is * * \param[in] rsc Resource to check * \param[out] list List to add result to * \param[in] current 0 = where known, 1 = running, 2 = running or pending * * \return If list contains only one node, that node */ pe_node_t * native_location(const pe_resource_t *rsc, GList **list, int current) { pe_node_t *one = NULL; GListPtr result = NULL; if (rsc->children) { GListPtr gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child = (pe_resource_t *) gIter->data; child->fns->location(child, &result, current); } } else if (current) { if (rsc->running_on) { result = g_list_copy(rsc->running_on); } if ((current == 2) && rsc->pending_node && !pe_find_node_id(result, rsc->pending_node->details->id)) { result = g_list_append(result, rsc->pending_node); } } else if (current == FALSE && rsc->allocated_to) { result = g_list_append(NULL, rsc->allocated_to); } if (result && (result->next == NULL)) { one = result->data; } if (list) { GListPtr gIter = result; for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; if (*list == NULL || pe_find_node_id(*list, node->details->id) == NULL) { *list = g_list_append(*list, node); } } } g_list_free(result); return one; } static void get_rscs_brief(GListPtr rsc_list, GHashTable * rsc_table, GHashTable * active_table) { GListPtr gIter = rsc_list; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE); int offset = 0; char buffer[LINE_MAX]; int *rsc_counter = NULL; int *active_counter = NULL; if (rsc->variant != pe_native) { continue; } offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", class); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) { const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER); - offset += snprintf(buffer + offset, LINE_MAX - offset, "::%s", prov); + offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s", prov); } offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s", kind); CRM_LOG_ASSERT(offset > 0); if (rsc_table) { rsc_counter = g_hash_table_lookup(rsc_table, buffer); if (rsc_counter == NULL) { rsc_counter = calloc(1, sizeof(int)); *rsc_counter = 0; g_hash_table_insert(rsc_table, strdup(buffer), rsc_counter); } (*rsc_counter)++; } if (active_table) { GListPtr gIter2 = rsc->running_on; for (; gIter2 != NULL; gIter2 = gIter2->next) { pe_node_t *node = (pe_node_t *) gIter2->data; GHashTable *node_table = NULL; if (node->details->unclean == FALSE && node->details->online == FALSE) { continue; } node_table = g_hash_table_lookup(active_table, node->details->uname); if (node_table == NULL) { node_table = crm_str_table_new(); g_hash_table_insert(active_table, strdup(node->details->uname), node_table); } active_counter = g_hash_table_lookup(node_table, buffer); if (active_counter == NULL) { active_counter = calloc(1, sizeof(int)); *active_counter = 0; g_hash_table_insert(node_table, strdup(buffer), active_counter); } (*active_counter)++; } } } } static void destroy_node_table(gpointer data) { GHashTable *node_table = data; if (node_table) { g_hash_table_destroy(node_table); } } void print_rscs_brief(GListPtr rsc_list, const char *pre_text, long options, void *print_data, gboolean print_all) { GHashTable *rsc_table = crm_str_table_new(); GHashTable *active_table = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_node_table); GHashTableIter hash_iter; char *type = NULL; int *rsc_counter = NULL; get_rscs_brief(rsc_list, rsc_table, active_table); g_hash_table_iter_init(&hash_iter, rsc_table); while (g_hash_table_iter_next(&hash_iter, (gpointer *)&type, (gpointer *)&rsc_counter)) { GHashTableIter hash_iter2; char *node_name = NULL; GHashTable *node_table = NULL; int active_counter_all = 0; g_hash_table_iter_init(&hash_iter2, active_table); while (g_hash_table_iter_next(&hash_iter2, (gpointer *)&node_name, (gpointer *)&node_table)) { int *active_counter = g_hash_table_lookup(node_table, type); if (active_counter == NULL || *active_counter == 0) { continue; } else { active_counter_all += *active_counter; } if (options & pe_print_rsconly) { node_name = NULL; } if (options & pe_print_html) { status_print("
  • \n"); } if (print_all) { status_print("%s%d/%d\t(%s):\tActive %s\n", pre_text ? pre_text : "", active_counter ? *active_counter : 0, rsc_counter ? *rsc_counter : 0, type, active_counter && (*active_counter > 0) && node_name ? node_name : ""); } else { status_print("%s%d\t(%s):\tActive %s\n", pre_text ? pre_text : "", active_counter ? *active_counter : 0, type, active_counter && (*active_counter > 0) && node_name ? node_name : ""); } if (options & pe_print_html) { status_print("
    • \n"); } } if (print_all && active_counter_all == 0) { if (options & pe_print_html) { status_print("
  • \n"); } status_print("%s%d/%d\t(%s):\tActive\n", pre_text ? pre_text : "", active_counter_all, rsc_counter ? *rsc_counter : 0, type); if (options & pe_print_html) { status_print("
    • \n"); } } } if (rsc_table) { g_hash_table_destroy(rsc_table); rsc_table = NULL; } if (active_table) { g_hash_table_destroy(active_table); active_table = NULL; } } int pe__rscs_brief_output(pcmk__output_t *out, GListPtr rsc_list, long options, gboolean print_all) { GHashTable *rsc_table = crm_str_table_new(); GHashTable *active_table = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_node_table); GListPtr sorted_rscs; int rc = pcmk_rc_no_output; get_rscs_brief(rsc_list, rsc_table, active_table); /* Make a list of the rsc_table keys so that it can be sorted. This is to make sure * output order stays consistent between systems. */ sorted_rscs = g_hash_table_get_keys(rsc_table); sorted_rscs = g_list_sort(sorted_rscs, (GCompareFunc) strcmp); for (GListPtr gIter = sorted_rscs; gIter; gIter = gIter->next) { char *type = (char *) gIter->data; int *rsc_counter = g_hash_table_lookup(rsc_table, type); GHashTableIter hash_iter2; char *node_name = NULL; GHashTable *node_table = NULL; int active_counter_all = 0; g_hash_table_iter_init(&hash_iter2, active_table); while (g_hash_table_iter_next(&hash_iter2, (gpointer *)&node_name, (gpointer *)&node_table)) { int *active_counter = g_hash_table_lookup(node_table, type); if (active_counter == NULL || *active_counter == 0) { continue; } else { active_counter_all += *active_counter; } if (options & pe_print_rsconly) { node_name = NULL; } if (print_all) { out->list_item(out, NULL, "%d/%d\t(%s):\tActive %s", *active_counter, rsc_counter ? *rsc_counter : 0, type, (*active_counter > 0) && node_name ? node_name : ""); } else { out->list_item(out, NULL, "%d\t(%s):\tActive %s", *active_counter, type, (*active_counter > 0) && node_name ? node_name : ""); } rc = pcmk_rc_ok; } if (print_all && active_counter_all == 0) { out->list_item(out, NULL, "%d/%d\t(%s):\tActive", active_counter_all, rsc_counter ? *rsc_counter : 0, type); rc = pcmk_rc_ok; } } if (rsc_table) { g_hash_table_destroy(rsc_table); rsc_table = NULL; } if (active_table) { g_hash_table_destroy(active_table); active_table = NULL; } if (sorted_rscs) { g_list_free(sorted_rscs); } return rc; } gboolean pe__native_is_filtered(pe_resource_t *rsc, GListPtr only_rsc, gboolean check_parent) { if (pcmk__str_in_list(only_rsc, rsc_printable_id(rsc)) || pcmk__str_in_list(only_rsc, rsc->id)) { return FALSE; } else if (check_parent) { pe_resource_t *up = uber_parent(rsc); if (pe_rsc_is_bundled(rsc)) { return up->parent->fns->is_filtered(up->parent, only_rsc, FALSE); } else { return up->fns->is_filtered(up, only_rsc, FALSE); } } return TRUE; }