diff --git a/cts/lab/CTStests.py b/cts/lab/CTStests.py index f4be998cfb..5535177b95 100644 --- a/cts/lab/CTStests.py +++ b/cts/lab/CTStests.py @@ -1,3142 +1,3142 @@ """ Test-specific classes for Pacemaker's Cluster Test Suite (CTS) """ __copyright__ = "Copyright 2000-2021 the Pacemaker project contributors" __license__ = "GNU General Public License version 2 or later (GPLv2+) WITHOUT ANY WARRANTY" # # SPECIAL NOTE: # # Tests may NOT implement any cluster-manager-specific code in them. # EXTEND the ClusterManager object to provide the base capabilities # the test needs if you need to do something that the current CM classes # do not. Otherwise you screw up the whole point of the object structure # in CTS. # # Thank you. # import os import re import time import subprocess import tempfile from stat import * from cts import CTS from cts.CTSaudits import * from cts.CTSvars import * from cts.patterns import PatternSelector from cts.logging import LogFactory from cts.remote import RemoteFactory 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('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 = self.create_watch(["No fencing will be done"], 60) managed.setwatch() self._set_unmanaged(node) if not managed.lookforall(): self.logger.log("Patterns not found: " + repr(managed.unmatched)) return self.failure("Resource management not disabled") pats = [] pats.append(self.templates["Pat:RscOpOK"] % ("start", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("stop", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("promote", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("demote", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("migrate", ".*")) watch = self.create_watch(pats, 60, "ShutdownActivity") watch.setwatch() self.debug("Shutting down the cluster") ret = self.stopall(None) if not ret: self._set_managed(node) return self.failure("Couldn't shut down the cluster") self.debug("Bringing the cluster back up") ret = self.startall(None) time.sleep(5) # allow ping to update the CIB if not ret: self._set_managed(node) return self.failure("Couldn't restart the cluster") if self.local_badnews("ResourceActivity:", watch): self._set_managed(node) return self.failure("Resources stopped or started during cluster restart") watch = self.create_watch(pats, 60, "StartupActivity") watch.setwatch() # Re-enable resource management (and verify it happened). self._set_managed(node) self.CM.cluster_stable() if not self._is_managed(node): return self.failure("Could not re-enable resource management") # Ignore actions for STONITH resources ignore = [] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rclass == "stonith": self.debug("Ignoring start actions for %s" % r.id) ignore.append(self.templates["Pat:RscOpOK"] % ("start", r.id)) if self.local_badnews("ResourceActivity:", watch, ignore): return self.failure("Resources stopped or started after resource management was re-enabled") return ret def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"resource( was|s were) active at shutdown", ] def is_applicable(self): return 1 AllTestClasses.append(Reattach) class SpecialTest1(CTSTest): '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SpecialTest1" self.startall = SimulStartLite(cm) self.restart1 = RestartTest(cm) self.stopall = SimulStopLite(cm) def __call__(self, node): '''Perform the 'SpecialTest1' test for Andrew. ''' self.incr("calls") # Shut down all the nodes... ret = self.stopall(None) if not ret: return self.failure("Could not stop all nodes") # Test config recovery when the other nodes come up self.rsh(node, "rm -f "+CTSvars.CRM_CONFIG_DIR+"/cib*") # Start the selected node ret = self.restart1(node) if not ret: return self.failure("Could not start "+node) # Start all remaining nodes ret = self.startall(None) if not ret: return self.failure("Could not start the remaining nodes") return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' # Errors that occur as a result of the CIB being wiped return [ r"error.*: v1 patchset error, patch failed to apply: Application of an update diff failed", r"error.*: Resource start-up disabled since no STONITH resources have been defined", r"error.*: Either configure some or disable STONITH with the stonith-enabled option", r"error.*: NOTE: Clusters with shared data need STONITH to ensure data integrity", ] AllTestClasses.append(SpecialTest1) class HAETest(CTSTest): '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "HAETest" self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) self.is_loop = 1 def setup(self, node): # Start all remaining nodes ret = self.startall(None) if not ret: return self.failure("Couldn't start all nodes") return self.success() def teardown(self, node): # Stop everything ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") return self.success() def wait_on_state(self, node, resource, expected_clones, attempts=240): while attempts > 0: active = 0 (rc, lines) = self.rsh(node, "crm_resource -r %s -W -Q" % resource, stdout=None) # Hack until crm_resource does the right thing if rc == 0 and lines: active = len(lines) if len(lines) == expected_clones: return 1 elif rc == 1: self.debug("Resource %s is still inactive" % resource) elif rc == 234: self.logger.log("Unknown resource %s" % resource) return 0 elif rc == 246: self.logger.log("Cluster is inactive") return 0 elif rc != 0: self.logger.log("Call to crm_resource failed, rc=%d" % rc) return 0 else: self.debug("Resource %s is active on %d times instead of %d" % (resource, active, expected_clones)) attempts -= 1 time.sleep(1) return 0 def find_dlm(self, node): self.r_dlm = None (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rtype == "controld" and r.parent != "NA": self.debug("Found dlm: %s" % self.r_dlm) self.r_dlm = r.parent return 1 return 0 def find_hae_resources(self, node): self.r_dlm = None self.r_o2cb = None self.r_ocfs2 = [] if self.find_dlm(node): self.find_ocfs2_resources(node) def is_applicable(self): if not self.is_applicable_common(): return 0 if self.Env["Schema"] == "hae": return 1 return None class HAERoleTest(HAETest): def __init__(self, cm): '''Lars' mount/unmount test for the HA extension. ''' HAETest.__init__(self,cm) self.name = "HAERoleTest" def change_state(self, node, resource, target): rc = self.rsh(node, "crm_resource -V -r %s -p target-role -v %s --meta" % (resource, target)) return rc def __call__(self, node): self.incr("calls") lpc = 0 failed = 0 delay = 2 done = time.time() + self.Env["loop-minutes"]*60 self.find_hae_resources(node) clone_max = len(self.Env["nodes"]) while time.time() <= done and not failed: lpc = lpc + 1 self.change_state(node, self.r_dlm, "Stopped") if not self.wait_on_state(node, self.r_dlm, 0): self.failure("%s did not go down correctly" % self.r_dlm) failed = lpc self.change_state(node, self.r_dlm, "Started") if not self.wait_on_state(node, self.r_dlm, clone_max): self.failure("%s did not come up correctly" % self.r_dlm) failed = lpc if not self.wait_on_state(node, self.r_o2cb, clone_max): self.failure("%s did not come up correctly" % self.r_o2cb) failed = lpc for fs in self.r_ocfs2: if not self.wait_on_state(node, fs, clone_max): self.failure("%s did not come up correctly" % fs) failed = lpc if failed: return self.failure("iteration %d failed" % failed) return self.success() AllTestClasses.append(HAERoleTest) class HAEStandbyTest(HAETest): '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): HAETest.__init__(self,cm) self.name = "HAEStandbyTest" def change_state(self, node, resource, target): rc = self.rsh(node, "crm_standby -V -l reboot -v %s" % (target)) return rc def __call__(self, node): self.incr("calls") lpc = 0 failed = 0 done = time.time() + self.Env["loop-minutes"]*60 self.find_hae_resources(node) clone_max = len(self.Env["nodes"]) while time.time() <= done and not failed: lpc = lpc + 1 self.change_state(node, self.r_dlm, "true") if not self.wait_on_state(node, self.r_dlm, clone_max-1): self.failure("%s did not go down correctly" % self.r_dlm) failed = lpc self.change_state(node, self.r_dlm, "false") if not self.wait_on_state(node, self.r_dlm, clone_max): self.failure("%s did not come up correctly" % self.r_dlm) failed = lpc if not self.wait_on_state(node, self.r_o2cb, clone_max): self.failure("%s did not come up correctly" % self.r_o2cb) failed = lpc for fs in self.r_ocfs2: if not self.wait_on_state(node, fs, clone_max): self.failure("%s did not come up correctly" % fs) failed = lpc if failed: return self.failure("iteration %d failed" % failed) return self.success() AllTestClasses.append(HAEStandbyTest) class NearQuorumPointTest(CTSTest): ''' This test brings larger clusters near the quorum point (50%). In addition, it will test doing starts and stops at the same time. Here is how I think it should work: - loop over the nodes and decide randomly which will be up and which will be down Use a 50% probability for each of up/down. - figure out what to do to get into that state from the current state - in parallel, bring up those going up and bring those going down. ''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "NearQuorumPoint" def __call__(self, dummy): '''Perform the 'NearQuorumPoint' test. ''' self.incr("calls") startset = [] stopset = [] stonith = self.CM.prepare_fencing_watcher("NearQuorumPoint") #decide what to do with each node for node in self.Env["nodes"]: action = self.Env.RandomGen.choice(["start","stop"]) #action = self.Env.RandomGen.choice(["start","stop","no change"]) if action == "start" : startset.append(node) elif action == "stop" : stopset.append(node) self.debug("start nodes:" + repr(startset)) self.debug("stop nodes:" + repr(stopset)) #add search patterns watchpats = [ ] for node in stopset: if self.CM.ShouldBeStatus[node] == "up": watchpats.append(self.templates["Pat:We_stopped"] % node) for node in startset: if self.CM.ShouldBeStatus[node] == "down": #watchpats.append(self.templates["Pat:NonDC_started"] % node) watchpats.append(self.templates["Pat:Local_started"] % node) else: for stopping in stopset: if self.CM.ShouldBeStatus[stopping] == "up": watchpats.append(self.templates["Pat:They_stopped"] % (node, self.CM.key_for_node(stopping))) if len(watchpats) == 0: return self.skipped() if len(startset) != 0: watchpats.append(self.templates["Pat:DC_IDLE"]) watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() #begin actions for node in stopset: if self.CM.ShouldBeStatus[node] == "up": self.CM.StopaCMnoBlock(node) for node in startset: if self.CM.ShouldBeStatus[node] == "down": self.CM.StartaCMnoBlock(node) #get the result if watch.lookforall(): self.CM.cluster_stable() self.CM.fencing_cleanup("NearQuorumPoint", stonith) return self.success() self.logger.log("Warn: Patterns not found: " + repr(watch.unmatched)) #get the "bad" nodes upnodes = [] for node in stopset: if self.CM.StataCM(node) == 1: upnodes.append(node) downnodes = [] for node in startset: if self.CM.StataCM(node) == 0: downnodes.append(node) self.CM.fencing_cleanup("NearQuorumPoint", stonith) if upnodes == [] and downnodes == []: self.CM.cluster_stable() # Make sure they're completely down with no residule for node in stopset: self.rsh(node, self.templates["StopCmd"]) return self.success() if len(upnodes) > 0: self.logger.log("Warn: Unstoppable nodes: " + repr(upnodes)) if len(downnodes) > 0: self.logger.log("Warn: Unstartable nodes: " + repr(downnodes)) return self.failure() def is_applicable(self): return 1 AllTestClasses.append(NearQuorumPointTest) class RollingUpgradeTest(CTSTest): '''Perform a rolling upgrade of the cluster''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "RollingUpgrade" self.start = StartTest(cm) self.stop = StopTest(cm) self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) def setup(self, node): # Start all remaining nodes ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") for node in self.Env["nodes"]: if not self.downgrade(node, None): return self.failure("Couldn't downgrade %s" % node) ret = self.startall(None) if not ret: return self.failure("Couldn't start all nodes") return self.success() def teardown(self, node): # Stop everything ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") for node in self.Env["nodes"]: if not self.upgrade(node, None): return self.failure("Couldn't upgrade %s" % node) return self.success() def install(self, node, version, start=1, flags="--force"): target_dir = "/tmp/rpm-%s" % version src_dir = "%s/%s" % (self.Env["rpm-dir"], version) self.logger.log("Installing %s on %s with %s" % (version, node, flags)) if not self.stop(node): return self.failure("stop failure: "+node) rc = self.rsh(node, "mkdir -p %s" % target_dir) rc = self.rsh(node, "rm -f %s/*.rpm" % target_dir) (rc, lines) = self.rsh(node, "ls -1 %s/*.rpm" % src_dir, None) for line in lines: line = line[:-1] rc = self.rsh.cp("%s" % (line), "%s:%s/" % (node, target_dir)) rc = self.rsh(node, "rpm -Uvh %s %s/*.rpm" % (flags, target_dir)) if start and not self.start(node): return self.failure("start failure: "+node) return self.success() def upgrade(self, node, start=1): return self.install(node, self.Env["current-version"], start) def downgrade(self, node, start=1): return self.install(node, self.Env["previous-version"], start, "--force --nodeps") def __call__(self, node): '''Perform the 'Rolling Upgrade' test. ''' self.incr("calls") for node in self.Env["nodes"]: if self.upgrade(node): return self.failure("Couldn't upgrade %s" % node) self.CM.cluster_stable() return self.success() def is_applicable(self): if not self.is_applicable_common(): return None if not "rpm-dir" in list(self.Env.keys()): return None if not "current-version" in list(self.Env.keys()): return None if not "previous-version" in list(self.Env.keys()): return None return 1 # Register RestartTest as a good test to run AllTestClasses.append(RollingUpgradeTest) class BSC_AddResource(CTSTest): '''Add a resource to the cluster''' def __init__(self, cm): CTSTest.__init__(self, cm) self.name = "AddResource" self.resource_offset = 0 self.cib_cmd = """cibadmin -C -o %s -X '%s' """ def __call__(self, node): self.incr("calls") self.resource_offset = self.resource_offset + 1 r_id = "bsc-rsc-%s-%d" % (node, self.resource_offset) start_pat = "pacemaker-controld.*%s_start_0.*confirmed.*ok" patterns = [] patterns.append(start_pat % r_id) watch = self.create_watch(patterns, self.Env["DeadTime"]) watch.setwatch() ip = self.NextIP() if not self.make_ip_resource(node, r_id, "ocf", "IPaddr", ip): return self.failure("Make resource %s failed" % r_id) failed = 0 watch_result = watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.logger.log ("Warn: Pattern not found: %s" % (regex)) failed = 1 if failed: return self.failure("Resource pattern(s) not found") if not self.CM.cluster_stable(self.Env["DeadTime"]): return self.failure("Unstable cluster") return self.success() def NextIP(self): ip = self.Env["IPBase"] if ":" in ip: fields = ip.rpartition(":") fields[2] = str(hex(int(fields[2], 16)+1)) print(str(hex(int(f[2], 16)+1))) else: fields = ip.rpartition('.') fields[2] = str(int(fields[2])+1) ip = fields[0] + fields[1] + fields[3]; self.Env["IPBase"] = ip return ip.strip() def make_ip_resource(self, node, id, rclass, type, ip): self.logger.log("Creating %s:%s:%s (%s) on %s" % (rclass,type,id,ip,node)) rsc_xml=""" """ % (id, rclass, type, id, id, ip) node_constraint = """ """ % (id, id, id, id, node) rc = 0 (rc, lines) = self.rsh(node, self.cib_cmd % ("constraints", node_constraint), None) if rc != 0: self.logger.log("Constraint creation failed: %d" % rc) return None (rc, lines) = self.rsh(node, self.cib_cmd % ("resources", rsc_xml), None) if rc != 0: self.logger.log("Resource creation failed: %d" % rc) return None return 1 def is_applicable(self): if self.Env["DoBSC"]: return 1 return None AllTestClasses.append(BSC_AddResource) class SimulStopLite(CTSTest): '''Stop any active nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SimulStopLite" def __call__(self, dummy): '''Perform the 'SimulStopLite' setup work. ''' self.incr("calls") self.debug("Setup: " + self.name) # We ignore the "node" parameter... watchpats = [ ] for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.incr("WasStarted") watchpats.append(self.templates["Pat:We_stopped"] % node) if len(watchpats) == 0: return self.success() # Stop all the nodes - at about the same time... watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() self.set_timer() for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.CM.StopaCMnoBlock(node) if watch.lookforall(): # Make sure they're completely down with no residule for node in self.Env["nodes"]: self.rsh(node, self.templates["StopCmd"]) return self.success() did_fail = 0 up_nodes = [] for node in self.Env["nodes"]: if self.CM.StataCM(node) == 1: did_fail = 1 up_nodes.append(node) if did_fail: return self.failure("Active nodes exist: " + repr(up_nodes)) self.logger.log("Warn: All nodes stopped but CTS didn't detect: " + repr(watch.unmatched)) return self.failure("Missing log message: "+repr(watch.unmatched)) def is_applicable(self): '''SimulStopLite is a setup test and never applicable''' return 0 class SimulStartLite(CTSTest): '''Start any stopped nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SimulStartLite" def __call__(self, dummy): '''Perform the 'SimulStartList' setup work. ''' self.incr("calls") self.debug("Setup: " + self.name) # We ignore the "node" parameter... node_list = [] for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "down": self.incr("WasStopped") node_list.append(node) self.set_timer() while len(node_list) > 0: # Repeat until all nodes come up watchpats = [ ] uppat = self.templates["Pat:NonDC_started"] if self.CM.upcount() == 0: uppat = self.templates["Pat:Local_started"] watchpats.append(self.templates["Pat:DC_IDLE"]) for node in node_list: watchpats.append(uppat % node) watchpats.append(self.templates["Pat:InfraUp"] % node) watchpats.append(self.templates["Pat:PacemakerUp"] % node) # Start all the nodes - at about the same time... watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() stonith = self.CM.prepare_fencing_watcher(self.name) for node in node_list: self.CM.StartaCMnoBlock(node) watch.lookforall() node_list = self.CM.fencing_cleanup(self.name, stonith) if node_list == None: return self.failure("Cluster did not stabilize") # Remove node_list messages from watch.unmatched for node in node_list: self.logger.debug("Dealing with stonith operations for %s" % repr(node_list)) if watch.unmatched: try: watch.unmatched.remove(uppat % node) except: self.debug("Already matched: %s" % (uppat % node)) try: watch.unmatched.remove(self.templates["Pat:InfraUp"] % node) except: self.debug("Already matched: %s" % (self.templates["Pat:InfraUp"] % node)) try: watch.unmatched.remove(self.templates["Pat:PacemakerUp"] % node) except: self.debug("Already matched: %s" % (self.templates["Pat:PacemakerUp"] % node)) if watch.unmatched: for regex in watch.unmatched: self.logger.log ("Warn: Startup pattern not found: %s" %(regex)) if not self.CM.cluster_stable(): return self.failure("Cluster did not stabilize") did_fail = 0 unstable = [] for node in self.Env["nodes"]: if self.CM.StataCM(node) == 0: did_fail = 1 unstable.append(node) if did_fail: return self.failure("Unstarted nodes exist: " + repr(unstable)) unstable = [] for node in self.Env["nodes"]: if not self.CM.node_stable(node): did_fail = 1 unstable.append(node) if did_fail: return self.failure("Unstable cluster nodes exist: " + repr(unstable)) return self.success() def is_applicable(self): '''SimulStartLite is a setup test and never applicable''' return 0 def TestList(cm, audits): result = [] for testclass in AllTestClasses: bound_test = testclass(cm) if bound_test.is_applicable(): bound_test.Audits = audits result.append(bound_test) return result class RemoteLXC(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "RemoteLXC" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.num_containers = 2 self.is_container = 1 self.is_docker_unsafe = 1 self.failed = 0 self.fail_string = "" def start_lxc_simple(self, node): # restore any artifacts laying around from a previous test. self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -s -R &>/dev/null") # generate the containers, put them in the config, add some resources to them pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("start", "lxc1")) pats.append(self.templates["Pat:RscOpOK"] % ("start", "lxc2")) pats.append(self.templates["Pat:RscOpOK"] % ("start", "lxc-ms")) pats.append(self.templates["Pat:RscOpOK"] % ("promote", "lxc-ms")) self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -g -a -m -s -c %d &>/dev/null" % self.num_containers) self.set_timer("remoteSimpleInit") watch.lookforall() self.log_timer("remoteSimpleInit") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 def cleanup_lxc_simple(self, node): pats = [ ] # if the test failed, attempt to clean up the cib and libvirt environment # as best as possible if self.failed == 1: # restore libvirt and cib self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -s -R &>/dev/null") return watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("stop", "container1")) pats.append(self.templates["Pat:RscOpOK"] % ("stop", "container2")) self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -p &>/dev/null") self.set_timer("remoteSimpleCleanup") watch.lookforall() self.log_timer("remoteSimpleCleanup") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 # cleanup libvirt self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -s -R &>/dev/null") def __call__(self, node): '''Perform the 'RemoteLXC' test. ''' self.incr("calls") ret = self.startall(None) if not ret: return self.failure("Setup failed, start all nodes failed.") rc = self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -v &>/dev/null") if rc == 1: self.log("Environment test for lxc support failed.") return self.skipped() self.start_lxc_simple(node) self.cleanup_lxc_simple(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed == 1: return self.failure(self.fail_string) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"Updating failcount for ping", r"schedulerd.*: Recover (ping|lxc-ms|container)\s*\(.*\)", # The orphaned lxc-ms resource causes an expected transition error # that is a result of the scheduler not having knowledge that the # promotable resource used to be a clone. As a result, it looks like that # resource is running in multiple locations when it shouldn't... But in # this instance we know why this error is occurring and that it is expected. r"Calculated [Tt]ransition .*pe-error", r"Resource lxc-ms .* is active on 2 nodes attempting recovery", r"Unknown operation: fail", r"VirtualDomain.*ERROR: Unable to determine emulator", ] AllTestClasses.append(RemoteLXC) class RemoteDriver(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = self.__class__.__name__ self.is_docker_unsafe = 1 self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.stop = StopTest(cm) self.remote_rsc = "remote-rsc" self.cib_cmd = """cibadmin -C -o %s -X '%s' """ self.reset() def reset(self): self.pcmk_started = 0 self.failed = False self.fail_string = "" self.remote_node_added = 0 self.remote_rsc_added = 0 self.remote_use_reconnect_interval = self.Env.RandomGen.choice([True,False]) def fail(self, msg): """ Mark test as failed. """ self.failed = True # Always log the failure. self.logger.log(msg) # Use first failure as test status, as it's likely to be most useful. if not self.fail_string: self.fail_string = msg def get_othernode(self, node): for othernode in self.Env["nodes"]: if othernode == node: # we don't want to try and use the cib that we just shutdown. # find a cluster node that is not our soon to be remote-node. continue else: return othernode def del_rsc(self, node, rsc): othernode = self.get_othernode(node) rc = self.rsh(othernode, "crm_resource -D -r %s -t primitive" % (rsc)) if rc != 0: self.fail("Removal of resource '%s' failed" % rsc) def add_rsc(self, node, rsc_xml): othernode = self.get_othernode(node) rc = self.rsh(othernode, self.cib_cmd % ("resources", rsc_xml)) if rc != 0: self.fail("resource creation failed") def add_primitive_rsc(self, node): rsc_xml = """ """ % { "node": self.remote_rsc } self.add_rsc(node, rsc_xml) if not self.failed: self.remote_rsc_added = 1 def add_connection_rsc(self, node): rsc_xml = """ """ % { "node": self.remote_node, "server": node } if self.remote_use_reconnect_interval: # Set reconnect interval on resource rsc_xml = rsc_xml + """ """ % (self.remote_node) rsc_xml = rsc_xml + """ """ % { "node": self.remote_node } self.add_rsc(node, rsc_xml) if not self.failed: self.remote_node_added = 1 def disable_services(self, node): self.corosync_enabled = self.Env.service_is_enabled(node, "corosync") if self.corosync_enabled: self.Env.disable_service(node, "corosync") self.pacemaker_enabled = self.Env.service_is_enabled(node, "pacemaker") if self.pacemaker_enabled: self.Env.disable_service(node, "pacemaker") def restore_services(self, node): if self.corosync_enabled: self.Env.enable_service(node, "corosync") if self.pacemaker_enabled: self.Env.enable_service(node, "pacemaker") def stop_pcmk_remote(self, node): # disable pcmk remote for i in range(10): rc = self.rsh(node, "service pacemaker_remote stop") if rc != 0: time.sleep(6) else: break def start_pcmk_remote(self, node): for i in range(10): rc = self.rsh(node, "service pacemaker_remote start") if rc != 0: time.sleep(6) else: self.pcmk_started = 1 break def freeze_pcmk_remote(self, node): """ Simulate a Pacemaker Remote daemon failure. """ # We freeze the process. self.rsh(node, "killall -STOP pacemaker-remoted") def resume_pcmk_remote(self, node): # We resume the process. self.rsh(node, "killall -CONT pacemaker-remoted") def start_metal(self, node): # Cluster nodes are reused as remote nodes in remote tests. If cluster # services were enabled at boot, in case the remote node got fenced, the # cluster node would join instead of the expected remote one. Meanwhile # pacemaker_remote would not be able to start. Depending on the chances, # the situations might not be able to be orchestrated gracefully any more. # # Temporarily disable any enabled cluster serivces. self.disable_services(node) pcmk_started = 0 # make sure the resource doesn't already exist for some reason self.rsh(node, "crm_resource -D -r %s -t primitive" % (self.remote_rsc)) self.rsh(node, "crm_resource -D -r %s -t primitive" % (self.remote_node)) if not self.stop(node): self.fail("Failed to shutdown cluster node %s" % node) return self.start_pcmk_remote(node) if self.pcmk_started == 0: self.fail("Failed to start pacemaker_remote on node %s" % node) return # Convert node to baremetal now that it has shutdown the cluster stack pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("start", self.remote_node)) pats.append(self.templates["Pat:DC_IDLE"]) self.add_connection_rsc(node) self.set_timer("remoteMetalInit") watch.lookforall() self.log_timer("remoteMetalInit") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) def migrate_connection(self, node): if self.failed: return pats = [ ] pats.append(self.templates["Pat:RscOpOK"] % ("migrate_to", self.remote_node)) pats.append(self.templates["Pat:RscOpOK"] % ("migrate_from", self.remote_node)) pats.append(self.templates["Pat:DC_IDLE"]) watch = self.create_watch(pats, 120) watch.setwatch() (rc, lines) = self.rsh(node, "crm_resource -M -r %s" % (self.remote_node), None) if rc != 0: self.fail("failed to move remote node connection resource") return self.set_timer("remoteMetalMigrate") watch.lookforall() self.log_timer("remoteMetalMigrate") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) return def fail_rsc(self, node): if self.failed: return watchpats = [ ] watchpats.append(self.templates["Pat:RscRemoteOpOK"] % ("stop", self.remote_rsc, self.remote_node)) watchpats.append(self.templates["Pat:RscRemoteOpOK"] % ("start", self.remote_rsc, self.remote_node)) watchpats.append(self.templates["Pat:DC_IDLE"]) watch = self.create_watch(watchpats, 120) watch.setwatch() self.debug("causing dummy rsc to fail.") rc = self.rsh(node, "rm -f /var/run/resource-agents/Dummy*") self.set_timer("remoteRscFail") watch.lookforall() self.log_timer("remoteRscFail") if watch.unmatched: self.fail("Unmatched patterns during rsc fail: %s" % watch.unmatched) def fail_connection(self, node): if self.failed: return watchpats = [ ] watchpats.append(self.templates["Pat:FenceOpOK"] % self.remote_node) watchpats.append(self.templates["Pat:NodeFenced"] % self.remote_node) watch = self.create_watch(watchpats, 120) watch.setwatch() # freeze the pcmk remote daemon. this will result in fencing self.debug("Force stopped active remote node") self.freeze_pcmk_remote(node) self.debug("Waiting for remote node to be fenced.") self.set_timer("remoteMetalFence") watch.lookforall() self.log_timer("remoteMetalFence") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) return self.debug("Waiting for the remote node to come back up") self.CM.ns.WaitForNodeToComeUp(node, 120); pats = [ ] watch = self.create_watch(pats, 240) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("start", self.remote_node)) if self.remote_rsc_added == 1: pats.append(self.templates["Pat:RscRemoteOpOK"] % ("start", self.remote_rsc, self.remote_node)) # start the remote node again watch it integrate back into cluster. self.start_pcmk_remote(node) if self.pcmk_started == 0: self.fail("Failed to start pacemaker_remote on node %s" % node) return self.debug("Waiting for remote node to rejoin cluster after being fenced.") self.set_timer("remoteMetalRestart") watch.lookforall() self.log_timer("remoteMetalRestart") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) return def add_dummy_rsc(self, node): if self.failed: return # verify we can put a resource on the remote node pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscRemoteOpOK"] % ("start", self.remote_rsc, self.remote_node)) pats.append(self.templates["Pat:DC_IDLE"]) # Add a resource that must live on remote-node self.add_primitive_rsc(node) # force that rsc to prefer the remote node. (rc, line) = self.CM.rsh(node, "crm_resource -M -r %s -N %s -f" % (self.remote_rsc, self.remote_node), None) if rc != 0: self.fail("Failed to place remote resource on remote node.") return self.set_timer("remoteMetalRsc") watch.lookforall() self.log_timer("remoteMetalRsc") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) def test_attributes(self, node): if self.failed: return # This verifies permanent attributes can be set on a remote-node. It also # verifies the remote-node can edit its own cib node section remotely. (rc, line) = self.CM.rsh(node, "crm_attribute -l forever -n testattr -v testval -N %s" % (self.remote_node), None) if rc != 0: self.fail("Failed to set remote-node attribute. rc:%s output:%s" % (rc, line)) return (rc, line) = self.CM.rsh(node, "crm_attribute -l forever -n testattr -q -N %s" % (self.remote_node), None) if rc != 0: self.fail("Failed to get remote-node attribute") return (rc, line) = self.CM.rsh(node, "crm_attribute -l forever -n testattr -D -N %s" % (self.remote_node), None) if rc != 0: self.fail("Failed to delete remote-node attribute") return def cleanup_metal(self, node): self.restore_services(node) if self.pcmk_started == 0: return pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() if self.remote_rsc_added == 1: pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.remote_rsc)) if self.remote_node_added == 1: pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.remote_node)) self.set_timer("remoteMetalCleanup") self.resume_pcmk_remote(node) if self.remote_rsc_added == 1: # Remove dummy resource added for remote node tests self.debug("Cleaning up dummy rsc put on remote node") self.rsh(self.get_othernode(node), "crm_resource -U -r %s" % self.remote_rsc) self.del_rsc(node, self.remote_rsc) if self.remote_node_added == 1: # Remove remote node's connection resource self.debug("Cleaning up remote node connection resource") self.rsh(self.get_othernode(node), "crm_resource -U -r %s" % (self.remote_node)) self.del_rsc(node, self.remote_node) watch.lookforall() self.log_timer("remoteMetalCleanup") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) self.stop_pcmk_remote(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.remote_node_added == 1: # Remove remote node itself self.debug("Cleaning up node entry for remote node") self.rsh(self.get_othernode(node), "crm_node --force --remove %s" % self.remote_node) def setup_env(self, node): self.remote_node = "remote-%s" % (node) # we are assuming if all nodes have a key, that it is # the right key... If any node doesn't have a remote # key, we regenerate it everywhere. if self.rsh.exists_on_all("/etc/pacemaker/authkey", self.Env["nodes"]): return # create key locally (handle, keyfile) = tempfile.mkstemp(".cts") os.close(handle) subprocess.check_call(["dd", "if=/dev/urandom", "of=%s" % keyfile, "bs=4096", "count=1"], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL) # sync key throughout the cluster for node in self.Env["nodes"]: self.rsh(node, "mkdir -p --mode=0750 /etc/pacemaker") self.rsh.cp(keyfile, "root@%s:/etc/pacemaker/authkey" % node) self.rsh(node, "chgrp haclient /etc/pacemaker /etc/pacemaker/authkey") self.rsh(node, "chmod 0640 /etc/pacemaker/authkey") os.unlink(keyfile) def is_applicable(self): if not self.is_applicable_common(): return False for node in self.Env["nodes"]: rc = self.rsh(node, "which pacemaker-remoted >/dev/null 2>&1") if rc != 0: return False return True def start_new_test(self, node): self.incr("calls") self.reset() ret = self.startall(None) if not ret: return self.failure("setup failed: could not start all nodes") self.setup_env(node) self.start_metal(node) self.add_dummy_rsc(node) return True def __call__(self, node): return self.failure("This base class is not meant to be called directly.") def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"""is running on remote.*which isn't allowed""", r"""Connection terminated""", r"""Could not send remote""", ] # RemoteDriver is just a base class for other tests, so it is not added to AllTestClasses class RemoteBasic(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteBaremetal' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) self.test_attributes(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() AllTestClasses.append(RemoteBasic) class RemoteStonithd(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteStonithd' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) self.fail_connection(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() def is_applicable(self): if not RemoteDriver.is_applicable(self): return False if "DoFencing" in list(self.Env.keys()): return self.Env["DoFencing"] return True def errorstoignore(self): ignore_pats = [ r"Lost connection to Pacemaker Remote node", r"Software caused connection abort", r"pacemaker-controld.*:\s+error.*: Operation remote-.*_monitor", r"pacemaker-controld.*:\s+error.*: Result of monitor operation for remote-.*", r"schedulerd.*:\s+Recover remote-.*\s*\(.*\)", r"error: Result of monitor operation for .* on remote-.*: Internal communication failure", ] ignore_pats.extend(RemoteDriver.errorstoignore(self)) return ignore_pats AllTestClasses.append(RemoteStonithd) class RemoteMigrate(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteMigrate' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) self.migrate_connection(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() def is_applicable(self): if not RemoteDriver.is_applicable(self): return 0 # This test requires at least three nodes: one to convert to a # remote node, one to host the connection originally, and one # to migrate the connection to. if len(self.Env["nodes"]) < 3: return 0 return 1 AllTestClasses.append(RemoteMigrate) class RemoteRscFailure(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteRscFailure' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) # This is an important step. We are migrating the connection # before failing the resource. This verifies that the migration # has properly maintained control over the remote-node. self.migrate_connection(node) self.fail_rsc(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() def errorstoignore(self): ignore_pats = [ r"schedulerd.*: Recover remote-rsc\s*\(.*\)", r"Dummy.*: No process state file found", ] ignore_pats.extend(RemoteDriver.errorstoignore(self)) return ignore_pats def is_applicable(self): if not RemoteDriver.is_applicable(self): return 0 # This test requires at least three nodes: one to convert to a # remote node, one to host the connection originally, and one # to migrate the connection to. if len(self.Env["nodes"]) < 3: return 0 return 1 AllTestClasses.append(RemoteRscFailure) # vim:ts=4:sw=4:et: diff --git a/cts/lab/ClusterManager.py b/cts/lab/ClusterManager.py index 8daffaebb1..002d45b5ec 100644 --- a/cts/lab/ClusterManager.py +++ b/cts/lab/ClusterManager.py @@ -1,1067 +1,1067 @@ """ ClusterManager class for Pacemaker's Cluster Test Suite (CTS) """ -__copyright__ = """Copyright 2000-2021 the Pacemaker project contributors. +__copyright__ = """Copyright 2000-2022 the Pacemaker project contributors. Certain portions by Huang Zhen are copyright 2004 International Business Machines. The version control history for this file may have further details.""" __license__ = "GNU General Public License version 2 or later (GPLv2+) WITHOUT ANY WARRANTY" import os import re import time from collections import UserDict from cts.CTSvars import * from cts.CIB import ConfigFactory from cts.CTS import NodeStatus, Process from cts.CTStests import AuditResource from cts.logging import LogFactory from cts.watcher import LogWatcher from cts.remote import RemoteFactory from cts.environment import EnvFactory from cts.patterns import PatternSelector has_log_stats = {} log_stats_bin = CTSvars.CRM_DAEMON_DIR + "/cts_log_stats.sh" log_stats = """ #!%s # Tool for generating system load reports while CTS runs trap "" 1 f=$1; shift action=$1; shift base=`basename $0` if [ ! -e $f ]; then echo "Time, Load 1, Load 5, Load 15, Test Marker" > $f fi function killpid() { if [ -e $f.pid ]; then kill -9 `cat $f.pid` rm -f $f.pid fi } function status() { if [ -e $f.pid ]; then kill -0 `cat $f.pid` return $? else return 1 fi } function start() { # Is it already running? if status then return fi echo Active as $$ echo $$ > $f.pid while [ 1 = 1 ]; do uptime | sed s/up.*:/,/ | tr '\\n' ',' >> $f #top -b -c -n1 | grep -e usr/libexec/pacemaker | grep -v -e grep -e python | head -n 1 | sed s@/usr/libexec/pacemaker/@@ | awk '{print " 0, "$9", "$10", "$12}' | tr '\\n' ',' >> $f echo 0 >> $f sleep 5 done } case $action in start) start ;; start-bg|bg) # Use c --ssh -- ./stats.sh file start-bg nohup $0 $f start >/dev/null 2>&1 > $f echo " $*" >> $f start ;; *) echo "Unknown action: $action." ;; esac """ % (CTSvars.BASH_PATH) class ClusterManager(UserDict): '''The Cluster Manager class. This is an subclass of the Python dictionary class. (this is because it contains lots of {name,value} pairs, not because it's behavior is that terribly similar to a dictionary in other ways.) This is an abstract class which class implements high-level operations on the cluster and/or its cluster managers. Actual cluster managers classes are subclassed from this type. One of the things we do is track the state we think every node should be in. ''' def __InitialConditions(self): #if os.geteuid() != 0: # raise ValueError("Must Be Root!") None def _finalConditions(self): for key in list(self.keys()): if self[key] == None: raise ValueError("Improper derivation: self[" + key + "] must be overridden by subclass.") def __init__(self): self.Env = EnvFactory().getInstance() self.templates = PatternSelector(self.Env["Name"]) self.__InitialConditions() self.logger = LogFactory() self.TestLoggingLevel=0 self.data = {} self.name = self.Env["Name"] self.rsh = RemoteFactory().getInstance() self.ShouldBeStatus={} self.ns = NodeStatus(self.Env) self.OurNode = os.uname()[1].lower() self.__instance_errorstoignore = [] self.fastfail = 0 self.cib_installed = 0 self.config = None self.cluster_monitor = 0 self.use_short_names = 1 if self.Env["DoBSC"]: del self.templates["Pat:They_stopped"] self._finalConditions() self.check_transitions = 0 self.check_elections = 0 self.CIBsync = {} self.CibFactory = ConfigFactory(self) self.cib = self.CibFactory.createConfig(self.Env["Schema"]) def __getitem__(self, key): if key == "Name": return self.name print("FIXME: Getting %s from %s" % (key, repr(self))) if key in self.data: return self.data[key] return self.templates.get_patterns(self.Env["Name"], key) def __setitem__(self, key, value): print("FIXME: Setting %s=%s on %s" % (key, value, repr(self))) self.data[key] = value def key_for_node(self, node): return node def instance_errorstoignore_clear(self): '''Allows the test scenario to reset instance errors to ignore on each iteration.''' self.__instance_errorstoignore = [] def instance_errorstoignore(self): '''Return list of errors which are 'normal' for a specific test instance''' return self.__instance_errorstoignore def log(self, args): self.logger.log(args) def debug(self, args): self.logger.debug(args) def upcount(self): '''How many nodes are up?''' count = 0 for node in self.Env["nodes"]: if self.ShouldBeStatus[node] == "up": count = count + 1 return count def install_support(self, command="install"): for node in self.Env["nodes"]: self.rsh(node, CTSvars.CRM_DAEMON_DIR + "/cts-support " + command) def prepare_fencing_watcher(self, name): # If we don't have quorum now but get it as a result of starting this node, # then a bunch of nodes might get fenced upnode = None if self.HasQuorum(None): self.debug("Have quorum") return None if not self.templates["Pat:Fencing_start"]: print("No start pattern") return None if not self.templates["Pat:Fencing_ok"]: print("No ok pattern") return None stonith = None stonithPats = [] for peer in self.Env["nodes"]: if self.ShouldBeStatus[peer] != "up": stonithPats.append(self.templates["Pat:Fencing_ok"] % peer) stonithPats.append(self.templates["Pat:Fencing_start"] % peer) stonith = LogWatcher(self.Env["LogFileName"], stonithPats, "StartupFencing", 0, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"]) stonith.setwatch() return stonith def fencing_cleanup(self, node, stonith): peer_list = [] peer_state = {} self.debug("Looking for nodes that were fenced as a result of %s starting" % node) # If we just started a node, we may now have quorum (and permission to fence) if not stonith: self.debug("Nothing to do") return peer_list q = self.HasQuorum(None) if not q and len(self.Env["nodes"]) > 2: # We didn't gain quorum - we shouldn't have shot anyone self.debug("Quorum: %d Len: %d" % (q, len(self.Env["nodes"]))) return peer_list for n in self.Env["nodes"]: peer_state[n] = "unknown" # Now see if any states need to be updated self.debug("looking for: " + repr(stonith.regexes)) shot = stonith.look(0) while shot: line = repr(shot) self.debug("Found: " + line) del stonith.regexes[stonith.whichmatch] # Extract node name for n in self.Env["nodes"]: if re.search(self.templates["Pat:Fencing_ok"] % n, shot): peer = n peer_state[peer] = "complete" self.__instance_errorstoignore.append(self.templates["Pat:Fencing_ok"] % peer) elif peer_state[n] != "complete" and re.search(self.templates["Pat:Fencing_start"] % n, shot): # TODO: Correctly detect multiple fencing operations for the same host peer = n peer_state[peer] = "in-progress" self.__instance_errorstoignore.append(self.templates["Pat:Fencing_start"] % peer) if not peer: self.logger.log("ERROR: Unknown stonith match: %s" % line) elif not peer in peer_list: self.debug("Found peer: " + peer) peer_list.append(peer) # Get the next one shot = stonith.look(60) for peer in peer_list: self.debug(" Peer %s was fenced as a result of %s starting: %s" % (peer, node, peer_state[peer])) if self.Env["at-boot"]: self.ShouldBeStatus[peer] = "up" else: self.ShouldBeStatus[peer] = "down" if peer_state[peer] == "in-progress": # Wait for any in-progress operations to complete shot = stonith.look(60) while len(stonith.regexes) and shot: line = repr(shot) self.debug("Found: " + line) del stonith.regexes[stonith.whichmatch] shot = stonith.look(60) # Now make sure the node is alive too self.ns.WaitForNodeToComeUp(peer, self.Env["DeadTime"]) # Poll until it comes up if self.Env["at-boot"]: if not self.StataCM(peer): time.sleep(self.Env["StartTime"]) if not self.StataCM(peer): self.logger.log("ERROR: Peer %s failed to restart after being fenced" % peer) return None return peer_list def StartaCM(self, node, verbose=False): '''Start up the cluster manager on a given node''' if verbose: self.logger.log("Starting %s on node %s" % (self.templates["Name"], node)) else: self.debug("Starting %s on node %s" % (self.templates["Name"], node)) ret = 1 if not node in self.ShouldBeStatus: self.ShouldBeStatus[node] = "down" if self.ShouldBeStatus[node] != "down": return 1 patterns = [] # Technically we should always be able to notice ourselves starting patterns.append(self.templates["Pat:Local_started"] % node) if self.upcount() == 0: patterns.append(self.templates["Pat:DC_started"] % node) else: patterns.append(self.templates["Pat:NonDC_started"] % node) watch = LogWatcher( self.Env["LogFileName"], patterns, "StartaCM", self.Env["StartTime"]+10, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"]) self.install_config(node) self.ShouldBeStatus[node] = "any" if self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]): self.logger.log ("%s was already started" % (node)) return 1 stonith = self.prepare_fencing_watcher(node) watch.setwatch() if self.rsh(node, self.templates["StartCmd"]) != 0: self.logger.log ("Warn: Start command failed on node %s" % (node)) self.fencing_cleanup(node, stonith) return None self.ShouldBeStatus[node] = "up" watch_result = watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.logger.log ("Warn: Startup pattern not found: %s" % (regex)) if watch_result and self.cluster_stable(self.Env["DeadTime"]): #self.debug("Found match: "+ repr(watch_result)) self.fencing_cleanup(node, stonith) return 1 elif self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]): self.fencing_cleanup(node, stonith) return 1 self.logger.log ("Warn: Start failed for node %s" % (node)) return None def StartaCMnoBlock(self, node, verbose=False): '''Start up the cluster manager on a given node with none-block mode''' if verbose: self.logger.log("Starting %s on node %s" % (self["Name"], node)) else: self.debug("Starting %s on node %s" % (self["Name"], node)) self.install_config(node) self.rsh(node, self.templates["StartCmd"], synchronous=0) self.ShouldBeStatus[node] = "up" return 1 def StopaCM(self, node, verbose=False, force=False): '''Stop the cluster manager on a given node''' if verbose: self.logger.log("Stopping %s on node %s" % (self["Name"], node)) else: self.debug("Stopping %s on node %s" % (self["Name"], node)) if self.ShouldBeStatus[node] != "up" and force == False: return 1 if self.rsh(node, self.templates["StopCmd"]) == 0: # Make sure we can continue even if corosync leaks # fdata-* is the old name #self.rsh(node, "rm -rf /dev/shm/qb-* /dev/shm/fdata-*") self.ShouldBeStatus[node] = "down" self.cluster_stable(self.Env["DeadTime"]) return 1 else: self.logger.log ("ERROR: Could not stop %s on node %s" % (self["Name"], node)) return None def StopaCMnoBlock(self, node): '''Stop the cluster manager on a given node with none-block mode''' self.debug("Stopping %s on node %s" % (self["Name"], node)) self.rsh(node, self.templates["StopCmd"], synchronous=0) self.ShouldBeStatus[node] = "down" return 1 def RereadCM(self, node): '''Force the cluster manager on a given node to reread its config This may be a no-op on certain cluster managers. ''' rc=self.rsh(node, self.templates["RereadCmd"]) if rc == 0: return 1 else: self.logger.log ("Could not force %s on node %s to reread its config" % (self["Name"], node)) return None def startall(self, nodelist=None, verbose=False, quick=False): '''Start the cluster manager on every node in the cluster. We can do it on a subset of the cluster if nodelist is not None. ''' map = {} if not nodelist: nodelist = self.Env["nodes"] for node in nodelist: if self.ShouldBeStatus[node] == "down": self.ns.WaitForAllNodesToComeUp(nodelist, 300) if not quick: # This is used for "basic sanity checks", so only start one node ... if not self.StartaCM(node, verbose=verbose): return 0 return 1 # Approximation of SimulStartList for --boot watchpats = [ ] watchpats.append(self.templates["Pat:DC_IDLE"]) for node in nodelist: watchpats.append(self.templates["Pat:InfraUp"] % node) watchpats.append(self.templates["Pat:PacemakerUp"] % node) watchpats.append(self.templates["Pat:Local_started"] % node) watchpats.append(self.templates["Pat:They_up"] % (nodelist[0], node)) # Start all the nodes - at about the same time... watch = LogWatcher(self.Env["LogFileName"], watchpats, "fast-start", self.Env["DeadTime"]+10, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"]) watch.setwatch() if not self.StartaCM(nodelist[0], verbose=verbose): return 0 for node in nodelist: self.StartaCMnoBlock(node, verbose=verbose) watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.logger.log ("Warn: Startup pattern not found: %s" % (regex)) if not self.cluster_stable(): self.logger.log("Cluster did not stabilize") return 0 return 1 def stopall(self, nodelist=None, verbose=False, force=False): '''Stop the cluster managers on every node in the cluster. We can do it on a subset of the cluster if nodelist is not None. ''' ret = 1 map = {} if not nodelist: nodelist = self.Env["nodes"] for node in self.Env["nodes"]: if self.ShouldBeStatus[node] == "up" or force == True: if not self.StopaCM(node, verbose=verbose, force=force): ret = 0 return ret def rereadall(self, nodelist=None): '''Force the cluster managers on every node in the cluster to reread their config files. We can do it on a subset of the cluster if nodelist is not None. ''' map = {} if not nodelist: nodelist = self.Env["nodes"] for node in self.Env["nodes"]: if self.ShouldBeStatus[node] == "up": self.RereadCM(node) def statall(self, nodelist=None): '''Return the status of the cluster managers in the cluster. We can do it on a subset of the cluster if nodelist is not None. ''' result = {} if not nodelist: nodelist = self.Env["nodes"] for node in nodelist: if self.StataCM(node): result[node] = "up" else: result[node] = "down" return result def isolate_node(self, target, nodes=None): '''isolate the communication between the nodes''' if not nodes: nodes = self.Env["nodes"] for node in nodes: if node != target: rc = self.rsh(target, self.templates["BreakCommCmd"] % self.key_for_node(node)) if rc != 0: self.logger.log("Could not break the communication between %s and %s: %d" % (target, node, rc)) return None else: self.debug("Communication cut between %s and %s" % (target, node)) return 1 def unisolate_node(self, target, nodes=None): '''fix the communication between the nodes''' if not nodes: nodes = self.Env["nodes"] for node in nodes: if node != target: restored = 0 # Limit the amount of time we have asynchronous connectivity for # Restore both sides as simultaneously as possible self.rsh(target, self.templates["FixCommCmd"] % self.key_for_node(node), synchronous=0) self.rsh(node, self.templates["FixCommCmd"] % self.key_for_node(target), synchronous=0) self.debug("Communication restored between %s and %s" % (target, node)) def reducecomm_node(self,node): '''reduce the communication between the nodes''' rc = self.rsh(node, self.templates["ReduceCommCmd"]%(self.Env["XmitLoss"],self.Env["RecvLoss"])) if rc == 0: return 1 else: self.logger.log("Could not reduce the communication between the nodes from node: %s" % node) return None def restorecomm_node(self,node): '''restore the saved communication between the nodes''' rc = 0 if float(self.Env["XmitLoss"]) != 0 or float(self.Env["RecvLoss"]) != 0 : rc = self.rsh(node, self.templates["RestoreCommCmd"]); if rc == 0: return 1 else: self.logger.log("Could not restore the communication between the nodes from node: %s" % node) return None def oprofileStart(self, node=None): if not node: for n in self.Env["oprofile"]: self.oprofileStart(n) elif node in self.Env["oprofile"]: self.debug("Enabling oprofile on %s" % node) self.rsh(node, "opcontrol --init") self.rsh(node, "opcontrol --setup --no-vmlinux --separate=lib --callgraph=20 --image=all") self.rsh(node, "opcontrol --start") self.rsh(node, "opcontrol --reset") def oprofileSave(self, test, node=None): if not node: for n in self.Env["oprofile"]: self.oprofileSave(test, n) elif node in self.Env["oprofile"]: self.rsh(node, "opcontrol --dump") self.rsh(node, "opcontrol --save=cts.%d" % test) # Read back with: opreport -l session:cts.0 image:/c* if None: self.rsh(node, "opcontrol --reset") else: self.oprofileStop(node) self.oprofileStart(node) def oprofileStop(self, node=None): if not node: for n in self.Env["oprofile"]: self.oprofileStop(n) elif node in self.Env["oprofile"]: self.debug("Stopping oprofile on %s" % node) self.rsh(node, "opcontrol --reset") self.rsh(node, "opcontrol --shutdown 2>&1 > /dev/null") def StatsExtract(self): if not self.Env["stats"]: return for host in self.Env["nodes"]: log_stats_file = "%s/cts-stats.csv" % CTSvars.CRM_DAEMON_DIR if host in has_log_stats: self.rsh(host, '''bash %s %s stop''' % (log_stats_bin, log_stats_file)) (rc, lines) = self.rsh(host, '''cat %s''' % log_stats_file, stdout=2) self.rsh(host, '''bash %s %s delete''' % (log_stats_bin, log_stats_file)) fname = "cts-stats-%d-nodes-%s.csv" % (len(self.Env["nodes"]), host) print("Extracted stats: %s" % fname) fd = open(fname, "a") fd.writelines(lines) fd.close() def StatsMark(self, testnum): '''Mark the test number in the stats log''' global has_log_stats if not self.Env["stats"]: return for host in self.Env["nodes"]: log_stats_file = "%s/cts-stats.csv" % CTSvars.CRM_DAEMON_DIR if not host in has_log_stats: global log_stats global log_stats_bin script=log_stats #script = re.sub("\\\\", "\\\\", script) script = re.sub('\"', '\\\"', script) script = re.sub("'", "\'", script) script = re.sub("`", "\`", script) script = re.sub("\$", "\\\$", script) self.debug("Installing %s on %s" % (log_stats_bin, host)) self.rsh(host, '''echo "%s" > %s''' % (script, log_stats_bin), silent=True) self.rsh(host, '''bash %s %s delete''' % (log_stats_bin, log_stats_file)) has_log_stats[host] = 1 # Now mark it self.rsh(host, '''bash %s %s mark %s''' % (log_stats_bin, log_stats_file, testnum), synchronous=0) def errorstoignore(self): # At some point implement a more elegant solution that # also produces a report at the end """ Return a list of known error messages that should be ignored """ return PatternSelector().get_patterns(self.name, "BadNewsIgnore") def install_config(self, node): if not self.ns.WaitForNodeToComeUp(node): self.log("Node %s is not up." % node) return None if not node in self.CIBsync and self.Env["ClobberCIB"] == 1: self.CIBsync[node] = 1 self.rsh(node, "rm -f "+CTSvars.CRM_CONFIG_DIR+"/cib*") # Only install the CIB on the first node, all the other ones will pick it up from there if self.cib_installed == 1: return None self.cib_installed = 1 if self.Env["CIBfilename"] == None: self.log("Installing Generated CIB on node %s" % (node)) self.cib.install(node) else: self.log("Installing CIB (%s) on node %s" % (self.Env["CIBfilename"], node)) if 0 != self.rsh.cp(self.Env["CIBfilename"], "root@" + (self.templates["CIBfile"] % node)): raise ValueError("Can not scp file to %s %d"%(node)) self.rsh(node, "chown "+CTSvars.CRM_DAEMON_USER+" "+CTSvars.CRM_CONFIG_DIR+"/cib.xml") def prepare(self): '''Finish the Initialization process. Prepare to test...''' self.partitions_expected = 1 for node in self.Env["nodes"]: self.ShouldBeStatus[node] = "" if self.Env["experimental-tests"]: self.unisolate_node(node) self.StataCM(node) def test_node_CM(self, node): '''Report the status of the cluster manager on a given node''' watchpats = [ ] watchpats.append("Current ping state: (S_IDLE|S_NOT_DC)") watchpats.append(self.templates["Pat:NonDC_started"] % node) watchpats.append(self.templates["Pat:DC_started"] % node) idle_watch = LogWatcher(self.Env["LogFileName"], watchpats, "ClusterIdle", hosts=[node], kind=self.Env["LogWatcher"]) idle_watch.setwatch() out = self.rsh(node, self.templates["StatusCmd"]%node, 1) self.debug("Node %s status: '%s'" %(node, out)) if not out or (out.find('ok') < 0): if self.ShouldBeStatus[node] == "up": self.log( "Node status for %s is %s but we think it should be %s" % (node, "down", self.ShouldBeStatus[node])) self.ShouldBeStatus[node] = "down" return 0 if self.ShouldBeStatus[node] == "down": self.log( "Node status for %s is %s but we think it should be %s: %s" % (node, "up", self.ShouldBeStatus[node], out)) self.ShouldBeStatus[node] = "up" # check the output first - because syslog-ng loses messages if out.find('S_NOT_DC') != -1: # Up and stable return 2 if out.find('S_IDLE') != -1: # Up and stable return 2 # fall back to syslog-ng and wait if not idle_watch.look(): # just up self.debug("Warn: Node %s is unstable: %s" % (node, out)) return 1 # Up and stable return 2 # Is the node up or is the node down def StataCM(self, node): '''Report the status of the cluster manager on a given node''' if self.test_node_CM(node) > 0: return 1 return None # Being up and being stable is not the same question... def node_stable(self, node): '''Report the status of the cluster manager on a given node''' if self.test_node_CM(node) == 2: return 1 self.log("Warn: Node %s not stable" % (node)) return None def partition_stable(self, nodes, timeout=None): watchpats = [ ] watchpats.append("Current ping state: S_IDLE") watchpats.append(self.templates["Pat:DC_IDLE"]) self.debug("Waiting for cluster stability...") if timeout == None: timeout = self.Env["DeadTime"] if len(nodes) < 3: self.debug("Cluster is inactive") return 1 idle_watch = LogWatcher(self.Env["LogFileName"], watchpats, "ClusterStable", timeout, hosts=nodes.split(), kind=self.Env["LogWatcher"]) idle_watch.setwatch() for node in nodes.split(): # have each node dump its current state self.rsh(node, self.templates["StatusCmd"] % node, 1) ret = idle_watch.look() while ret: self.debug(ret) for node in nodes.split(): if re.search(node, ret): return 1 ret = idle_watch.look() self.debug("Warn: Partition %s not IDLE after %ds" % (repr(nodes), timeout)) return None def cluster_stable(self, timeout=None, double_check=False): partitions = self.find_partitions() for partition in partitions: if not self.partition_stable(partition, timeout): return None if double_check: # Make sure we are really stable and that all resources, # including those that depend on transient node attributes, # are started if they were going to be time.sleep(5) for partition in partitions: if not self.partition_stable(partition, timeout): return None return 1 def is_node_dc(self, node, status_line=None): rc = 0 if not status_line: status_line = self.rsh(node, self.templates["StatusCmd"]%node, 1) if not status_line: rc = 0 elif status_line.find('S_IDLE') != -1: rc = 1 elif status_line.find('S_INTEGRATION') != -1: rc = 1 elif status_line.find('S_FINALIZE_JOIN') != -1: rc = 1 elif status_line.find('S_POLICY_ENGINE') != -1: rc = 1 elif status_line.find('S_TRANSITION_ENGINE') != -1: rc = 1 return rc def active_resources(self, node): (rc, output) = self.rsh(node, """crm_resource -c""", None) resources = [] for line in output: if re.search("^Resource", line): tmp = AuditResource(self, line) if tmp.type == "primitive" and tmp.host == node: resources.append(tmp.id) return resources def ResourceLocation(self, rid): ResourceNodes = [] for node in self.Env["nodes"]: if self.ShouldBeStatus[node] == "up": cmd = self.templates["RscRunning"] % (rid) (rc, lines) = self.rsh(node, cmd, None) if rc == 127: self.log("Command '%s' failed. Binary or pacemaker-cts package not installed?" % cmd) for line in lines: self.log("Output: "+line) elif rc == 0: ResourceNodes.append(node) return ResourceNodes def find_partitions(self): ccm_partitions = [] for node in self.Env["nodes"]: if self.ShouldBeStatus[node] == "up": partition = self.rsh(node, self.templates["PartitionCmd"], 1) if not partition: self.log("no partition details for %s" % node) elif len(partition) > 2: nodes = partition.split() nodes.sort() partition = ' '.join(nodes) found = 0 for a_partition in ccm_partitions: if partition == a_partition: found = 1 if found == 0: self.debug("Adding partition from %s: %s" % (node, partition)) ccm_partitions.append(partition) else: self.debug("Partition '%s' from %s is consistent with existing entries" % (partition, node)) else: self.log("bad partition details for %s" % node) else: self.debug("Node %s is down... skipping" % node) self.debug("Found partitions: %s" % repr(ccm_partitions) ) return ccm_partitions def HasQuorum(self, node_list): # If we are auditing a partition, then one side will # have quorum and the other not. # So the caller needs to tell us which we are checking # If no value for node_list is specified... assume all nodes if not node_list: node_list = self.Env["nodes"] for node in node_list: if self.ShouldBeStatus[node] == "up": quorum = self.rsh(node, self.templates["QuorumCmd"], 1) if quorum.find("1") != -1: return 1 elif quorum.find("0") != -1: return 0 else: self.debug("WARN: Unexpected quorum test result from " + node + ":" + quorum) return 0 def Components(self): complist = [] common_ignore = [ "Pending action:", "(ERROR|error): crm_log_message_adv:", "(ERROR|error): MSG: No message to dump", "pending LRM operations at shutdown", "Lost connection to the CIB manager", "Connection to the CIB terminated...", "Sending message to the CIB manager FAILED", "Action A_RECOVER .* not supported", "(ERROR|error): stonithd_op_result_ready: not signed on", "pingd.*(ERROR|error): send_update: Could not send update", "send_ipc_message: IPC Channel to .* is not connected", "unconfirmed_actions: Waiting on .* unconfirmed actions", "cib_native_msgready: Message pending on command channel", r": Performing A_EXIT_1 - forcefully exiting ", r"Resource .* was active at shutdown. You may ignore this error if it is unmanaged.", ] stonith_ignore = [ r"Updating failcount for child_DoFencing", r"error.*: Fencer connection failed \(will retry\)", "pacemaker-execd.*(ERROR|error): stonithd_receive_ops_result failed.", ] stonith_ignore.extend(common_ignore) ccm = Process(self, "ccm", triggersreboot=self.fastfail, pats = [ "State transition .* S_RECOVERY", "pacemaker-controld.*Action A_RECOVER .* not supported", r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover", r"pacemaker-controld.*: Could not recover from internal error", "pacemaker-controld.*I_ERROR.*crmd_cib_connection_destroy", # these status numbers are likely wrong now r"pacemaker-controld.*exited with status 2", r"attrd.*exited with status 1", r"cib.*exited with status 2", # Not if it was fenced # "A new node joined the cluster", # "WARN: determine_online_status: Node .* is unclean", -# "Scheduling Node .* for STONITH", +# "Scheduling node .* for fencing", # "Executing .* fencing operation", # "tengine_stonith_callback: .*result=0", # "Processing I_NODE_JOIN:.* cause=C_HA_MESSAGE", # "State transition S_.* -> S_INTEGRATION.*input=I_NODE_JOIN", "State transition S_STARTING -> S_PENDING", ], badnews_ignore = common_ignore) based = Process(self, "pacemaker-based", triggersreboot=self.fastfail, pats = [ "State transition .* S_RECOVERY", "Lost connection to the CIB manager", "Connection to the CIB manager terminated", r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover", "pacemaker-controld.*I_ERROR.*crmd_cib_connection_destroy", r"pacemaker-controld.*: Could not recover from internal error", # these status numbers are likely wrong now r"pacemaker-controld.*exited with status 2", r"attrd.*exited with status 1", ], badnews_ignore = common_ignore) execd = Process(self, "pacemaker-execd", triggersreboot=self.fastfail, pats = [ "State transition .* S_RECOVERY", "LRM Connection failed", "pacemaker-controld.*I_ERROR.*lrm_connection_destroy", "State transition S_STARTING -> S_PENDING", r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover", r"pacemaker-controld.*: Could not recover from internal error", # this status number is likely wrong now r"pacemaker-controld.*exited with status 2", ], badnews_ignore = common_ignore) controld = Process(self, "pacemaker-controld", triggersreboot=self.fastfail, pats = [ # "WARN: determine_online_status: Node .* is unclean", -# "Scheduling Node .* for STONITH", +# "Scheduling node .* for fencing", # "Executing .* fencing operation", # "tengine_stonith_callback: .*result=0", "State transition .* S_IDLE", "State transition S_STARTING -> S_PENDING", ], badnews_ignore = common_ignore) schedulerd = Process(self, "pacemaker-schedulerd", triggersreboot=self.fastfail, pats = [ "State transition .* S_RECOVERY", r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover", r"pacemaker-controld.*: Could not recover from internal error", r"pacemaker-controld.*CRIT.*: Connection to the scheduler failed", "pacemaker-controld.*I_ERROR.*save_cib_contents", # this status number is likely wrong now r"pacemaker-controld.*exited with status 2", ], badnews_ignore = common_ignore, dc_only=1) if self.Env["DoFencing"] == 1 : complist.append(Process(self, "stoniths", triggersreboot=self.fastfail, dc_pats = [ r"pacemaker-controld.*CRIT.*: Fencing daemon connection failed", "Attempting connection to fencing daemon", ], badnews_ignore = stonith_ignore)) if self.fastfail == 0: ccm.pats.extend([ # these status numbers are likely wrong now r"attrd.*exited with status 1", r"pacemaker-(based|controld).*exited with status 2", ]) based.pats.extend([ # these status numbers are likely wrong now r"attrd.*exited with status 1", r"pacemaker-controld.*exited with status 2", ]) execd.pats.extend([ # these status numbers are likely wrong now r"pacemaker-controld.*exited with status 2", ]) complist.append(ccm) complist.append(based) complist.append(execd) complist.append(controld) complist.append(schedulerd) return complist def StandbyStatus(self, node): out=self.rsh(node, self.templates["StandbyQueryCmd"] % node, 1) if not out: return "off" out = out[:-1] self.debug("Standby result: "+out) return out # status == "on" : Enter Standby mode # status == "off": Enter Active mode def SetStandbyMode(self, node, status): current_status = self.StandbyStatus(node) cmd = self.templates["StandbyCmd"] % (node, status) ret = self.rsh(node, cmd) return True def AddDummyRsc(self, node, rid): rsc_xml = """ ' '""" % (rid, rid) constraint_xml = """ ' ' """ % (rid, node, node, rid) self.rsh(node, self.templates['CibAddXml'] % (rsc_xml)) self.rsh(node, self.templates['CibAddXml'] % (constraint_xml)) def RemoveDummyRsc(self, node, rid): constraint = "\"//rsc_location[@rsc='%s']\"" % (rid) rsc = "\"//primitive[@id='%s']\"" % (rid) self.rsh(node, self.templates['CibDelXpath'] % constraint) self.rsh(node, self.templates['CibDelXpath'] % rsc) diff --git a/cts/lab/patterns.py b/cts/lab/patterns.py index 38d6e1065a..90cac735d8 100644 --- a/cts/lab/patterns.py +++ b/cts/lab/patterns.py @@ -1,415 +1,415 @@ """ Pattern-holding classes for Pacemaker's Cluster Test Suite (CTS) """ __copyright__ = "Copyright 2008-2022 the Pacemaker project contributors" __license__ = "GNU General Public License version 2 or later (GPLv2+) WITHOUT ANY WARRANTY" import sys, os from cts.CTSvars import * patternvariants = {} class BasePatterns(object): def __init__(self, name): self.name = name patternvariants[name] = self self.ignore = [ "avoid confusing Valgrind", # Logging bug in some versions of libvirtd r"libvirtd.*: internal error: Failed to parse PCI config address", # pcs can log this when node is fenced, but fencing is OK in some # tests (and we will catch it in pacemaker logs when not OK) r"pcs.daemon:No response from: .* request: get_configs, error:", ] self.BadNews = [] self.components = {} self.commands = { "StatusCmd" : "crmadmin -t 60 -S %s 2>/dev/null", "CibQuery" : "cibadmin -Ql", "CibAddXml" : "cibadmin --modify -c --xml-text %s", "CibDelXpath" : "cibadmin --delete --xpath %s", # 300,000 == 5 minutes "RscRunning" : CTSvars.CRM_DAEMON_DIR + "/cts-exec-helper -R -r %s", "CIBfile" : "%s:"+CTSvars.CRM_CONFIG_DIR+"/cib.xml", "TmpDir" : "/tmp", "BreakCommCmd" : "iptables -A INPUT -s %s -j DROP >/dev/null 2>&1", "FixCommCmd" : "iptables -D INPUT -s %s -j DROP >/dev/null 2>&1", # tc qdisc add dev lo root handle 1: cbq avpkt 1000 bandwidth 1000mbit # tc class add dev lo parent 1: classid 1:1 cbq rate "$RATE"kbps allot 17000 prio 5 bounded isolated # tc filter add dev lo parent 1: protocol ip prio 16 u32 match ip dst 127.0.0.1 match ip sport $PORT 0xFFFF flowid 1:1 # tc qdisc add dev lo parent 1: netem delay "$LATENCY"msec "$(($LATENCY/4))"msec 10% 2> /dev/null > /dev/null "ReduceCommCmd" : "", "RestoreCommCmd" : "tc qdisc del dev lo root", "MaintenanceModeOn" : "cibadmin --modify -c --xml-text ''", "MaintenanceModeOff" : "cibadmin --delete --xpath \"//nvpair[@name='maintenance-mode']\"", "StandbyCmd" : "crm_attribute -Vq -U %s -n standby -l forever -v %s 2>/dev/null", "StandbyQueryCmd" : "crm_attribute -qG -U %s -n standby -l forever -d off 2>/dev/null", } self.search = { "Pat:DC_IDLE" : "pacemaker-controld.*State transition.*-> S_IDLE", # This won't work if we have multiple partitions "Pat:Local_started" : "%s\W.*controller successfully started", "Pat:NonDC_started" : r"%s\W.*State transition.*-> S_NOT_DC", "Pat:DC_started" : r"%s\W.*State transition.*-> S_IDLE", "Pat:We_stopped" : "%s\W.*OVERRIDE THIS PATTERN", "Pat:They_stopped" : "%s\W.*LOST:.* %s ", "Pat:They_dead" : "node %s.*: is dead", "Pat:They_up" : "%s %s\W.*OVERRIDE THIS PATTERN", "Pat:TransitionComplete" : "Transition status: Complete: complete", "Pat:Fencing_start" : r"Requesting peer fencing .* targeting %s", "Pat:Fencing_ok" : r"pacemaker-fenced.*:\s*Operation .* targeting %s by .* for .*@.*: OK", "Pat:Fencing_recover" : r"pacemaker-schedulerd.*: Recover %s", "Pat:Fencing_active" : r"stonith resource .* is active on 2 nodes (attempting recovery)", "Pat:Fencing_probe" : r"pacemaker-controld.* Result of probe operation for %s on .*: Error", "Pat:RscOpOK" : r"pacemaker-controld.*:\s+Result of %s operation for %s.*: (0 \()?ok", "Pat:RscOpFail" : r"pacemaker-schedulerd.*:.*Unexpected result .* recorded for %s of %s ", "Pat:CloneOpFail" : r"pacemaker-schedulerd.*:.*Unexpected result .* recorded for %s of (%s|%s) ", "Pat:RscRemoteOpOK" : r"pacemaker-controld.*:\s+Result of %s operation for %s on %s: (0 \()?ok", "Pat:NodeFenced" : r"pacemaker-controld.*:\s* Peer %s was terminated \(.*\) by .* on behalf of .*: OK", "Pat:FenceOpOK" : "Operation .* targeting %s using .* returned 0", } def get_component(self, key): if key in self.components: return self.components[key] print("Unknown component '%s' for %s" % (key, self.name)) return [] def get_patterns(self, key): if key == "BadNews": return self.BadNews elif key == "BadNewsIgnore": return self.ignore elif key == "Commands": return self.commands elif key == "Search": return self.search elif key == "Components": return self.components def __getitem__(self, key): if key == "Name": return self.name elif key in self.commands: return self.commands[key] elif key in self.search: return self.search[key] else: print("Unknown template '%s' for %s" % (key, self.name)) return None class crm_corosync(BasePatterns): ''' Patterns for Corosync version 2 cluster manager class ''' def __init__(self, name): BasePatterns.__init__(self, name) self.commands.update({ "StartCmd" : "service corosync start && service pacemaker start", "StopCmd" : "service pacemaker stop; [ ! -e /usr/sbin/pacemaker-remoted ] || service pacemaker_remote stop; service corosync stop", "EpochCmd" : "crm_node -e", "QuorumCmd" : "crm_node -q", "PartitionCmd" : "crm_node -p", }) self.search.update({ # Close enough ... "Corosync Cluster Engine exiting normally" isn't # printed reliably. "Pat:We_stopped" : "%s\W.*Unloading all Corosync service engines", "Pat:They_stopped" : "%s\W.*pacemaker-controld.*Node %s(\[|\s).*state is now lost", "Pat:They_dead" : "pacemaker-controld.*Node %s(\[|\s).*state is now lost", "Pat:They_up" : "\W%s\W.*pacemaker-controld.*Node %s state is now member", "Pat:ChildExit" : r"\[[0-9]+\] exited with status [0-9]+ \(", # "with signal 9" == pcmk_child_exit(), "$" == check_active_before_startup_processes() "Pat:ChildKilled" : r"%s\W.*pacemakerd.*%s\[[0-9]+\] terminated( with signal 9|$)", "Pat:ChildRespawn" : "%s\W.*pacemakerd.*Respawning %s subdaemon after unexpected exit", "Pat:InfraUp" : "%s\W.*corosync.*Initializing transport", "Pat:PacemakerUp" : "%s\W.*pacemakerd.*Starting Pacemaker", }) self.ignore = self.ignore + [ r"crm_mon:", r"crmadmin:", r"update_trace_data", r"async_notify:.*strange, client not found", r"Parse error: Ignoring unknown option .*nodename", r"error.*: Operation 'reboot' .* using FencingFail returned ", r"getinfo response error: 1$", r"sbd.* error: inquisitor_child: DEBUG MODE IS ACTIVE", r"sbd.* pcmk:\s*error:.*Connection to cib_ro.* (failed|closed)", ] self.BadNews = [ r"[^(]error:", r"crit:", r"ERROR:", r"CRIT:", r"Shutting down...NOW", r"Timer I_TERMINATE just popped", r"input=I_ERROR", r"input=I_FAIL", r"input=I_INTEGRATED cause=C_TIMER_POPPED", r"input=I_FINALIZED cause=C_TIMER_POPPED", r"input=I_ERROR", r"(pacemakerd|pacemaker-execd|pacemaker-controld):.*, exiting", r"schedulerd.*Attempting recovery of resource", r"is taking more than 2x its timeout", r"Confirm not received from", r"Welcome reply not received from", r"Attempting to schedule .* after a stop", r"Resource .* was active at shutdown", r"duplicate entries for call_id", r"Search terminated:", r":global_timer_callback", r"Faking parameter digest creation", r"Parameters to .* action changed:", r"Parameters to .* changed", r"pacemakerd.*\[[0-9]+\] terminated( with signal| as IPC server|$)", r"pacemaker-schedulerd.*Recover .*\(.* -\> .*\)", r"rsyslogd.* imuxsock lost .* messages from pid .* due to rate-limiting", r"Peer is not part of our cluster", r"We appear to be in an election loop", r"Unknown node -> we will not deliver message", r"(Blackbox dump requested|Problem detected)", r"pacemakerd.*Could not connect to Cluster Configuration Database API", r"Receiving messages from a node we think is dead", r"share the same cluster nodeid", r"share the same name", #r"crm_ipc_send:.*Request .* failed", #r"crm_ipc_send:.*Sending to .* is disabled until pending reply is received", # Not inherently bad, but worth tracking #r"No need to invoke the TE", #r"ping.*: DEBUG: Updated connected = 0", #r"Digest mis-match:", r"pacemaker-controld:.*Transition failed: terminated", r"Local CIB .* differs from .*:", r"warn.*:\s*Continuing but .* will NOT be used", r"warn.*:\s*Cluster configuration file .* is corrupt", #r"Executing .* fencing operation", r"Election storm", r"stalled the FSA with pending inputs", ] self.components["common-ignore"] = [ r"Pending action:", r"resource( was|s were) active at shutdown", r"pending LRM operations at shutdown", r"Lost connection to the CIB manager", r"pacemaker-controld.*:\s*Action A_RECOVER .* not supported", r"pacemaker-controld.*:\s*Performing A_EXIT_1 - forcefully exiting ", r".*:\s*Requesting fencing \([^)]+\) of node ", r"(Blackbox dump requested|Problem detected)", ] self.components["corosync-ignore"] = [ r"Could not connect to Corosync CFG: CS_ERR_LIBRARY", r"error:.*Connection to the CPG API failed: Library error", r"\[[0-9]+\] exited with status [0-9]+ \(", r"\[[0-9]+\] terminated with signal 15", r"pacemaker-based.*error:.*Corosync connection lost", r"pacemaker-fenced.*error:.*Corosync connection terminated", r"pacemaker-controld.*State transition .* S_RECOVERY", r"pacemaker-controld.*error:.*Input (I_ERROR|I_TERMINATE ) .*received in state", r"pacemaker-controld.*error:.*Could not recover from internal error", r"error:.*Connection to cib_(shm|rw).* (failed|closed)", r"error:.*cib_(shm|rw) IPC provider disconnected while waiting", r"error:.*Connection to (fencer|stonith-ng).* (closed|failed|lost)", r"crit: Fencing daemon connection failed", # This is overbroad, but we don't have a way to say that only # certain transition errors are acceptable (if the fencer respawns, # fence devices may appear multiply active). We have to rely on # other causes of a transition error logging their own error # message, which is the usual practice. r"pacemaker-schedulerd.* Calculated transition .*/pe-error", ] self.components["corosync"] = [ # We expect each daemon to lose its cluster connection. # However, if the CIB manager loses its connection first, # it's possible for another daemon to lose that connection and # exit before losing the cluster connection. r"pacemakerd.*:\s*warning:.*Lost connection to cluster layer", r"pacemaker-attrd.*:\s*(crit|error):.*Lost connection to (cluster layer|the CIB manager)", r"pacemaker-based.*:\s*(crit|error):.*Lost connection to cluster layer", r"pacemaker-controld.*:\s*(crit|error):.*Lost connection to (cluster layer|the CIB manager)", r"pacemaker-fenced.*:\s*(crit|error):.*Lost connection to (cluster layer|the CIB manager)", - r"schedulerd.*Scheduling Node .* for STONITH", + r"schedulerd.*Scheduling node .* for fencing", r"pacemaker-controld.*:\s*Peer .* was terminated \(.*\) by .* on behalf of .*:\s*OK", ] self.components["pacemaker-based"] = [ r"pacemakerd.* pacemaker-attrd\[[0-9]+\] exited with status 102", r"pacemakerd.* pacemaker-controld\[[0-9]+\] exited with status 1", r"pacemakerd.* Respawning pacemaker-attrd subdaemon after unexpected exit", r"pacemakerd.* Respawning pacemaker-based subdaemon after unexpected exit", r"pacemakerd.* Respawning pacemaker-controld subdaemon after unexpected exit", r"pacemakerd.* Respawning pacemaker-fenced subdaemon after unexpected exit", r"pacemaker-.* Connection to cib_.* (failed|closed)", r"pacemaker-attrd.*:.*Lost connection to the CIB manager", r"pacemaker-controld.*:.*Lost connection to the CIB manager", r"pacemaker-controld.*I_ERROR.*crmd_cib_connection_destroy", r"pacemaker-controld.* State transition .* S_RECOVERY", r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover", r"pacemaker-controld.*Could not recover from internal error", ] self.components["pacemaker-based-ignore"] = [ r"pacemaker-execd.*Connection to (fencer|stonith-ng).* (closed|failed|lost)", # This is overbroad, but we don't have a way to say that only # certain transition errors are acceptable (if the fencer respawns, # fence devices may appear multiply active). We have to rely on # other causes of a transition error logging their own error # message, which is the usual practice. r"pacemaker-schedulerd.* Calculated transition .*/pe-error", ] self.components["pacemaker-execd"] = [ r"pacemaker-controld.*Connection to executor failed", r"pacemaker-controld.*I_ERROR.*lrm_connection_destroy", r"pacemaker-controld.*State transition .* S_RECOVERY", r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover", r"pacemaker-controld.*Could not recover from internal error", r"pacemakerd.*pacemaker-controld\[[0-9]+\] exited with status 1", r"pacemakerd.* Respawning pacemaker-execd subdaemon after unexpected exit", r"pacemakerd.* Respawning pacemaker-controld subdaemon after unexpected exit", ] self.components["pacemaker-execd-ignore"] = [ r"pacemaker-(attrd|controld).*Connection to lrmd.* (failed|closed)", r"pacemaker-(attrd|controld).*Could not execute alert", ] self.components["pacemaker-controld"] = [ # "WARN: determine_online_status: Node .* is unclean", -# "Scheduling Node .* for STONITH", +# "Scheduling node .* for fencing", # Only if the node wasn't the DC: "State transition S_IDLE", "State transition .* -> S_IDLE", ] self.components["pacemaker-controld-ignore"] = [] self.components["pacemaker-attrd"] = [] self.components["pacemaker-attrd-ignore"] = [] self.components["pacemaker-schedulerd"] = [ "State transition .* S_RECOVERY", r"pacemakerd.* Respawning pacemaker-controld subdaemon after unexpected exit", r"pacemaker-controld\[[0-9]+\] exited with status 1 \(", r"Connection to the scheduler failed", "pacemaker-controld.*I_ERROR.*save_cib_contents", r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover", "pacemaker-controld.*Could not recover from internal error", ] self.components["pacemaker-schedulerd-ignore"] = [ r"Connection to pengine.* (failed|closed)", ] self.components["pacemaker-fenced"] = [ r"error:.*Connection to (fencer|stonith-ng).* (closed|failed|lost)", r"Fencing daemon connection failed", r"pacemaker-controld.*Fencer successfully connected", ] self.components["pacemaker-fenced-ignore"] = [ r"error:.*Connection to (fencer|stonith-ng).* (closed|failed|lost)", r"crit:.*Fencing daemon connection failed", r"error:.*Fencer connection failed \(will retry\)", r"Connection to (fencer|stonith-ng) failed, finalizing .* pending operations", r"pacemaker-controld.*:\s+Result of .* operation for Fencing.*Error", # This is overbroad, but we don't have a way to say that only # certain transition errors are acceptable (if the fencer respawns, # fence devices may appear multiply active). We have to rely on # other causes of a transition error logging their own error # message, which is the usual practice. r"pacemaker-schedulerd.* Calculated transition .*/pe-error", ] self.components["pacemaker-fenced-ignore"].extend(self.components["common-ignore"]) class crm_corosync_docker(crm_corosync): ''' Patterns for Corosync version 2 cluster manager class ''' def __init__(self, name): crm_corosync.__init__(self, name) self.commands.update({ "StartCmd" : "pcmk_start", "StopCmd" : "pcmk_stop", }) class PatternSelector(object): def __init__(self, name=None): self.name = name self.base = BasePatterns("crm-base") if not name: crm_corosync("crm-corosync") elif name == "crm-corosync": crm_corosync(name) elif name == "crm-corosync-docker": crm_corosync_docker(name) def get_variant(self, variant): if variant in patternvariants: return patternvariants[variant] print("defaulting to crm-base for %s" % variant) return self.base def get_patterns(self, variant, kind): return self.get_variant(variant).get_patterns(kind) def get_template(self, variant, key): v = self.get_variant(variant) return v[key] def get_component(self, variant, kind): return self.get_variant(variant).get_component(kind) def __getitem__(self, key): return self.get_template(self.name, key) # python cts/CTSpatt.py -k crm-corosync -t StartCmd if __name__ == '__main__': pdir=os.path.dirname(sys.path[0]) sys.path.insert(0, pdir) # So that things work from the source directory kind=None template=None skipthis=None args=sys.argv[1:] for i in range(0, len(args)): if skipthis: skipthis=None continue elif args[i] == "-k" or args[i] == "--kind": skipthis=1 kind = args[i+1] elif args[i] == "-t" or args[i] == "--template": skipthis=1 template = args[i+1] else: print("Illegal argument " + args[i]) print(PatternSelector(kind)[template])