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 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) elif self.Env["Stack"] == "corosync (cman)": # There is a delay between gaining quorum and CMAN starting fencing # This can mean that even nodes that are fully up get fenced # There is no use fighting it, just look for everyone so that CTS doesn't get confused 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 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 SetUp(self): '''Set up the Scenario. Return TRUE on success.''' self.ClusterManager.prepare() self.audit() # Also detects remote/local log config self.ClusterManager.StatsMark(0) self.ClusterManager.ns.WaitForAllNodesToComeUp( self.ClusterManager.Env["nodes"]) self.audit() self.ClusterManager.install_support() self.BadNews = LogWatcher(self.ClusterManager.Env["LogFileName"], self.ClusterManager.templates.get_patterns( self.ClusterManager.Env["Name"], "BadNews"), "BadNews", 0, kind=self.ClusterManager.Env["LogWatcher"], hosts=self.ClusterManager.Env["nodes"]) self.BadNews.setwatch( ) # Call after we've figured out what type of log watching to do in LogAudit j = 0 while j < len(self.Components): if not self.Components[j].SetUp(self.ClusterManager): # OOPS! We failed. Tear partial setups down. self.audit() self.ClusterManager.log("Tearing down partial setup") self.TearDown(j) return None j = j + 1 self.audit() return 1
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 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
def TestLogging(self): patterns = [] prefix = "Test message from" suffix = str(uuid.uuid4()) watch = {} for node in self.CM.Env["nodes"]: # Look for the node name in two places to make sure # that syslog is logging with the correct hostname m = re.search("^([^.]+).*", node) if m: simple = m.group(1) else: simple = node patterns.append("%s.*%s %s %s" % (simple, prefix, node, suffix)) watch_pref = self.CM.Env["LogWatcher"] if watch_pref == "any": for k in self.kinds: watch[k] = LogWatcher(self.CM.Env["LogFileName"], patterns, "LogAudit", 5, silent=True, hosts=self.CM.Env["nodes"], kind=k) watch[k].setwatch() else: k = watch_pref watch[k] = LogWatcher(self.CM.Env["LogFileName"], patterns, "LogAudit", 5, silent=True, hosts=self.CM.Env["nodes"], kind=k) watch[k].setwatch() if watch_pref == "any": self.CM.log("Writing log with key: %s" % (suffix)) for node in self.CM.Env["nodes"]: cmd = "logger -p %s.info %s %s %s" % ( self.CM.Env["SyslogFacility"], prefix, node, suffix) if self.CM.rsh(node, cmd, synchronous=0, silent=True) != 0: self.CM.log("ERROR: Cannot execute remote command [%s] on %s" % (cmd, node)) for k in self.kinds: if k in watch: w = watch[k] if watch_pref == "any": self.CM.log("Testing for %s logs" % (k)) w.lookforall(silent=True) if not w.unmatched: if watch_pref == "any": self.CM.log("Continuing with %s-based log reader" % (w.kind)) self.CM.Env["LogWatcher"] = w.kind return 1 for k in list(watch.keys()): w = watch[k] if w.unmatched: for regex in w.unmatched: self.CM.log("Test message [%s] not found in %s logs." % (regex, w.kind)) return 0
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 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:Master_started"] % node) else: patterns.append(self.templates["Pat:Slave_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 # Clear out the host cache so autojoin can be exercised if self.clear_cache: self.debug("Removing cache file on: "+node) self.rsh(node, "rm -f "+CTSvars.HA_VARLIBHBDIR+"/hostcache") if not(self.Env["valgrind-tests"]): startCmd = self.templates["StartCmd"] else: if self.Env["valgrind-prefix"]: prefix = self.Env["valgrind-prefix"] else: prefix = "cts" startCmd = """G_SLICE=always-malloc HA_VALGRIND_ENABLED='%s' VALGRIND_OPTS='%s --log-file=/tmp/%s-%s.valgrind' %s""" % ( self.Env["valgrind-procs"], self.Env["valgrind-opts"], prefix, """%p""", self.templates["StartCmd"]) stonith = self.prepare_fencing_watcher(node) watch.setwatch() if self.rsh(node, 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