diff --git a/python/pacemaker/_cts/clustermanager.py b/python/pacemaker/_cts/clustermanager.py index 99f6d25fd3..96fbb63b28 100644 --- a/python/pacemaker/_cts/clustermanager.py +++ b/python/pacemaker/_cts/clustermanager.py @@ -1,883 +1,888 @@ """ ClusterManager class for Pacemaker's Cluster Test Suite (CTS) """ __all__ = ["ClusterManager"] __copyright__ = """Copyright 2000-2023 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 pacemaker.buildoptions import BuildOptions from pacemaker._cts.CTS import NodeStatus, Process from pacemaker._cts.audits import AuditResource from pacemaker._cts.cib import ConfigFactory from pacemaker._cts.environment import EnvFactory from pacemaker._cts.logging import LogFactory from pacemaker._cts.patterns import PatternSelector from pacemaker._cts.remote import RemoteFactory from pacemaker._cts.watcher import LogWatcher 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 _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.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.__instance_errors_to_ignore = [] self.cib_installed = 0 self.config = None self.use_short_names = 1 self._finalConditions() self.check_transitions = 0 self.check_elections = 0 self.CIBsync = {} self.CibFactory = ConfigFactory(self) self.cib = self.CibFactory.create_config(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(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 clear_instance_errors_to_ignore(self): + """ Reset instance-specific errors to ignore on each iteration """ - def instance_errorstoignore(self): - '''Return list of errors which are 'normal' for a specific test instance''' - return self.__instance_errorstoignore + self.__instance_errors_to_ignore = [] + + @property + def instance_errors_to_ignore(self): + """ Return a list of known errors that should be ignored for a specific + test instance + """ + + return self.__instance_errors_to_ignore + + @property + def errors_to_ignore(self): + """ Return a list of known error messages that should be ignored """ + + return self.templates.get_patterns("BadNewsIgnore") 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, BuildOptions.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, self.Env["nodes"], self.Env["LogWatcher"], "StartupFencing", 0) stonith.set_watch() 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) + self.__instance_errors_to_ignore.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) + self.__instance_errors_to_ignore.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.wait_for_node(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, self.Env["nodes"], self.Env["LogWatcher"], "StartaCM", self.Env["StartTime"]+10) 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.set_watch() (rc, _) = self.rsh(node, self.templates["StartCmd"]) if rc != 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.look_for_all() 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.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=False) 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 (rc, _) = self.rsh(node, self.templates["StopCmd"]) if rc == 0: # Make sure we can continue even if corosync leaks 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=False) self.ShouldBeStatus[node] = "down" 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.wait_for_all_nodes(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, self.Env["nodes"], self.Env["LogWatcher"], "fast-start", self.Env["DeadTime"]+10) watch.set_watch() if not self.StartaCM(nodelist[0], verbose=verbose): return 0 for node in nodelist: self.StartaCMnoBlock(node, verbose=verbose) watch.look_for_all() 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 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=False) self.rsh(node, self.templates["FixCommCmd"] % self.key_for_node(target), synchronous=False) self.debug("Communication restored between %s and %s" % (target, node)) 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* 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 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 self.templates.get_patterns("BadNewsIgnore") - def install_config(self, node): if not self.ns.wait_for_node(node): self.log("Node %s is not up." % node) return None if not node in self.CIBsync and self.Env["ClobberCIB"]: self.CIBsync[node] = 1 self.rsh(node, "rm -f " + BuildOptions.CIB_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 self.rsh.copy(self.Env["CIBfilename"], "root@" + (self.templates["CIBfile"] % node)) != 0: raise ValueError("Can not scp file to %s %d"%(node)) self.rsh(node, "chown " + BuildOptions.DAEMON_USER + " " + BuildOptions.CIB_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, [node], self.Env["LogWatcher"], "ClusterIdle") idle_watch.set_watch() (_, out) = self.rsh(node, self.templates["StatusCmd"]%node, verbose=1) if not out: out = "" else: out = out[0].strip() self.debug("Node %s status: '%s'" %(node, out)) if 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, nodes.split(), self.Env["LogWatcher"], "ClusterStable", timeout) idle_watch.set_watch() for node in nodes.split(): # have each node dump its current state self.rsh(node, self.templates["StatusCmd"] % node, verbose=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: (_, out) = self.rsh(node, self.templates["StatusCmd"]%node, verbose=1) if out: status_line = out[0].strip() 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): (_, output) = self.rsh(node, "crm_resource -c", verbose=1) 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) 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": (_, out) = self.rsh(node, self.templates["PartitionCmd"], verbose=1) if not out: self.log("no partition details for %s" % node) continue partition = out[0].strip() if 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"], verbose=1) quorum = quorum[0].strip() 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", 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", "State transition S_STARTING -> S_PENDING", ], badnews_ignore = common_ignore) based = Process(self, "pacemaker-based", 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", 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", pats = [ "State transition .* S_IDLE", "State transition S_STARTING -> S_PENDING", ], badnews_ignore = common_ignore) schedulerd = Process(self, "pacemaker-schedulerd", 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=True) if self.Env["DoFencing"]: complist.append(Process(self, "stoniths", dc_pats = [ r"pacemaker-controld.*CRIT.*: Fencing daemon connection failed", "Attempting connection to fencing daemon", ], badnews_ignore = stonith_ignore)) 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, verbose=1) if not out: return "off" out = out[0].strip() 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) 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/python/pacemaker/_cts/scenarios.py b/python/pacemaker/_cts/scenarios.py index b3eef1fcf0..f5d70a2595 100644 --- a/python/pacemaker/_cts/scenarios.py +++ b/python/pacemaker/_cts/scenarios.py @@ -1,408 +1,408 @@ """ Test scenario classes for Pacemaker's Cluster Test Suite (CTS) """ __all__ = [ "AllOnce", "Boot", "BootCluster", "LeaveBooted", "RandomTests", "Sequence" ] __copyright__ = "Copyright 2000-2023 the Pacemaker project contributors" __license__ = "GNU General Public License version 2 or later (GPLv2+) WITHOUT ANY WARRANTY" import re import time from pacemaker._cts.audits import ClusterAudit from pacemaker._cts.input import should_continue from pacemaker._cts.tests.ctstest import CTSTest from pacemaker._cts.watcher import LogWatcher class ScenarioComponent: """ The base class for all scenario components. A scenario component is one single step in a scenario. Each component is basically just a setup and teardown method. """ def __init__(self, cm, env): """ Create a new ScenarioComponent instance Arguments: cm -- A ClusterManager instance env -- An Environment instance """ # pylint: disable=invalid-name self._cm = cm self._env = env def is_applicable(self): """ Return True if this component is applicable in the given Environment. This method must be provided by all subclasses. """ raise NotImplementedError def setup(self): """ Set up the component, returning True on success. This method must be provided by all subclasses. """ raise NotImplementedError def teardown(self): """ Tear down the given component. This method must be provided by all subclasses. """ raise NotImplementedError class Scenario: """ The base class for scenario. A scenario is an ordered list of ScenarioComponent objects. A scenario proceeds by setting up all its components in sequence, running a list of tests and audits, and then tearing down its components in reverse. """ def __init__(self, cm, components, audits, tests): """ Create a new Scenario instance Arguments: cm -- A ClusterManager instance components -- A list of ScenarioComponents comprising this Scenario audits -- A list of ClusterAudits that will be performed as part of this Scenario tests -- A list of CTSTests that will be run """ # pylint: disable=invalid-name self.stats = { "success": 0, "failure": 0, "BadNews": 0, "skipped": 0 } self.tests = tests self._audits = audits self._bad_news = None self._cm = cm self._components = components for comp in components: if not issubclass(comp.__class__, ScenarioComponent): raise ValueError("Init value must be subclass of ScenarioComponent") for audit in audits: if not issubclass(audit.__class__, ClusterAudit): raise ValueError("Init value must be subclass of ClusterAudit") for test in tests: if not issubclass(test.__class__, CTSTest): raise ValueError("Init value must be a subclass of CTSTest") def is_applicable(self): """ Return True if all ScenarioComponents are applicable """ for comp in self._components: if not comp.is_applicable(): return False return True def setup(self): """ Set up the scenario, returning True on success. If setup fails at some point, tear down those components that did successfully set up. """ self._cm.prepare() self.audit() # Also detects remote/local log config self._cm.ns.wait_for_all_nodes(self._cm.Env["nodes"]) self.audit() self._cm.install_support() self._bad_news = LogWatcher(self._cm.Env["LogFileName"], self._cm.templates.get_patterns("BadNews"), self._cm.Env["nodes"], self._cm.Env["LogWatcher"], "BadNews", 0) self._bad_news.set_watch() # 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(): # OOPS! We failed. Tear partial setups down. self.audit() self._cm.log("Tearing down partial setup") self.teardown(j) return False j += 1 self.audit() return True def teardown(self, n_components=None): """ Tear down the scenario in the reverse order it was set up. If n_components is not None, only tear down that many components. """ if not n_components: n_components = len(self._components)-1 j = n_components while j >= 0: self._components[j].teardown() j -= 1 self.audit() self._cm.install_support("uninstall") def incr(self, name): """ Increment the given stats key """ if not name in self.stats: self.stats[name] = 0 self.stats[name] += 1 def run(self, iterations): """ Run all tests in the scenario the given number of times """ self._cm.oprofileStart() try: self._run_loop(iterations) self._cm.oprofileStop() except: self._cm.oprofileStop() raise def _run_loop(self, iterations): """ Do the hard part of the run method - actually run all the tests the given number of times. """ raise NotImplementedError def run_test(self, test, testcount): """ Run the given test. testcount is the number of tests (including this one) that have been run across all iterations. """ nodechoice = self._cm.Env.random_node() ret = True did_run = False - self._cm.instance_errorstoignore_clear() + self._cm.clear_instance_errors_to_ignore() choice = "(%s)" % nodechoice self._cm.log("Running test {:<22} {:<15} [{:>3}]".format(test.name, choice, testcount)) starttime = test.set_timer() if not test.setup(nodechoice): self._cm.log("Setup failed") ret = False elif not test.can_run_now(nodechoice): self._cm.log("Skipped") test.skipped() else: did_run = True ret = test(nodechoice) if not test.teardown(nodechoice): self._cm.log("Teardown failed") if not should_continue(self._cm.Env): raise ValueError("Teardown of %s on %s failed" % (test.name, nodechoice)) ret = False stoptime = time.time() self._cm.oprofileSave(testcount) elapsed_time = stoptime - starttime test_time = stoptime - test.get_timer() if "min_time" not in test.stats: test.stats["elapsed_time"] = elapsed_time test.stats["min_time"] = test_time test.stats["max_time"] = test_time else: test.stats["elapsed_time"] += elapsed_time if test_time < test.stats["min_time"]: test.stats["min_time"] = test_time if test_time > test.stats["max_time"]: test.stats["max_time"] = test_time if ret: self.incr("success") test.log_timer() else: self.incr("failure") self._cm.statall() did_run = True # Force the test count to be incremented anyway so test extraction works self.audit(test.errors_to_ignore) return did_run def summarize(self): """ Output scenario results """ self._cm.log("****************") self._cm.log("Overall Results:%r" % self.stats) self._cm.log("****************") stat_filter = { "calls": 0, "failure": 0, "skipped": 0, "auditfail": 0 } self._cm.log("Test Summary") for test in self.tests: for key in stat_filter: stat_filter[key] = test.stats[key] name = "Test %s:" % test.name self._cm.log("{:<25} {!r}".format(name, stat_filter)) self._cm.debug("Detailed Results") for test in self.tests: name = "Test %s:" % test.name self._cm.debug("{:<25} {!r}".format(name, stat_filter)) self._cm.log("<<<<<<<<<<<<<<<< TESTS COMPLETED") def audit(self, local_ignore=None): """ Perform all scenario audits and log results. If there are too many failures, prompt the user to confirm that the scenario should continue running. """ errcount = 0 ignorelist = ["CTS:"] if local_ignore: ignorelist.extend(local_ignore) - ignorelist.extend(self._cm.errorstoignore()) - ignorelist.extend(self._cm.instance_errorstoignore()) + ignorelist.extend(self._cm.errors_to_ignore) + ignorelist.extend(self._cm.instance_errors_to_ignore) # This makes sure everything is stabilized before starting... failed = 0 for audit in self._audits: if not audit(): self._cm.log("Audit %s FAILED." % audit.name) failed += 1 else: self._cm.debug("Audit %s passed." % audit.name) while errcount < 1000: match = None if self._bad_news: match = self._bad_news.look(0) if match: add_err = True for ignore in ignorelist: if add_err and re.search(ignore, match): add_err = False if add_err: self._cm.log("BadNews: %s" % match) self.incr("BadNews") errcount += 1 else: break else: print("Big problems") if not should_continue(self._cm.Env): self._cm.log("Shutting down.") self.summarize() self.teardown() raise ValueError("Looks like we hit a BadNews jackpot!") if self._bad_news: self._bad_news.end() return failed class AllOnce(Scenario): """ Every Test Once """ def _run_loop(self, iterations): testcount = 1 for test in self.tests: self.run_test(test, testcount) testcount += 1 class RandomTests(Scenario): """ Random Test Execution """ def _run_loop(self, iterations): testcount = 1 while testcount <= iterations: test = self._cm.Env.random_gen.choice(self.tests) self.run_test(test, testcount) testcount += 1 class Sequence(Scenario): """ Named Tests in Sequence """ def _run_loop(self, iterations): testcount = 1 while testcount <= iterations: for test in self.tests: self.run_test(test, testcount) testcount += 1 class Boot(Scenario): """ Start the Cluster """ def _run_loop(self, iterations): return class BootCluster(ScenarioComponent): """ The BootCluster component simply starts the cluster manager on all nodes, waiting for each to come up before starting given that a node might have been rebooted or crashed beforehand. """ def is_applicable(self): """ BootCluster is always applicable """ return True def setup(self): """ Set up the component, returning True on success """ self._cm.prepare() # Clear out the cobwebs ;-) self._cm.stopall(verbose=True, force=True) # Now start the Cluster Manager on all the nodes. self._cm.log("Starting Cluster Manager on all nodes.") return self._cm.startall(verbose=True, quick=True) def teardown(self): """ Tear down the component """ self._cm.log("Stopping Cluster Manager on all nodes") self._cm.stopall(verbose=True, force=False) class LeaveBooted(BootCluster): """ The LeaveBooted component leaves all nodes up when the scenario is complete. """ def teardown(self): """ Tear down the component """ self._cm.log("Leaving Cluster running on all nodes")