diff --git a/cts/CM_ais.py b/cts/CM_ais.py index 3ed6f88980..6ec5a48165 100644 --- a/cts/CM_ais.py +++ b/cts/CM_ais.py @@ -1,441 +1,441 @@ '''CTS: Cluster Testing System: AIS dependent modules... ''' __copyright__=''' Copyright (C) 2007 Andrew Beekhof ''' # # This program is free software; you can redistribute it and/or # modify it under the terms of the GNU General Public License # as published by the Free Software Foundation; either version 2 # of the License, or (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program; if not, write to the Free Software # Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. import os, sys, warnings from cts.CTSvars import * from cts.CM_lha import crm_lha from cts.CTS import Process ####################################################################### # # LinuxHA v2 dependent modules # ####################################################################### class crm_ais(crm_lha): ''' The crm version 3 cluster manager class. It implements the things we need to talk to and manipulate crm clusters running on top of openais ''' def __init__(self, Environment, randseed=None): crm_lha.__init__(self, Environment, randseed=randseed) self.update({ "Name" : "crm-ais", "EpocheCmd" : "crm_node -e --openais", "QuorumCmd" : "crm_node -q --openais", "ParitionCmd" : "crm_node -p --openais", "Pat:They_stopped" : "%s crmd.*Node %s\[.*state is now lost", "Pat:ChildExit" : "Child process .* exited", # Bad news Regexes. Should never occur. "BadRegexes" : ( r" trace:", 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", exiting\.", r"(WARN|warn).*Ignoring HA message.*vote.*not in our membership list", r"pengine.*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"Child process .* terminated with signal", - r"Executing .* fencing operation", r"LogActions: 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"crm_write_blackbox", r"pacemakerd.*Could not connect to Cluster Configuration Database API", #r"crm_ipc_send:.*Request .* failed", #r"crm_ipc_send:.*Sending to .* is disabled until pending reply is recieved", # Not inherently bad, but worth tracking #r"No need to invoke the TE", #r"ping.*: DEBUG: Updated connected = 0", #r"Digest mis-match:", r"te_graph_trigger: Transition failed: terminated", - r"fence_pcmk.* Call to fence", - r"fence_pcmk", + #r"Executing .* fencing operation", + #r"fence_pcmk.* Call to fence", + #r"fence_pcmk", r"cman killed by node", ), }) def errorstoignore(self): # At some point implement a more elegant solution that # also produces a report at the end '''Return list of errors which are known and very noisey should be ignored''' if 1: return [ 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: log_operation: Operation 'reboot' .* with device 'FencingFail' returned:", r"Child process .* terminated with signal 9", ] return [] def NodeUUID(self, node): return node def ais_components(self): fullcomplist = {} self.complist = [] self.common_ignore = [ "Pending action:", "error: crm_log_message_adv:", "error: MSG: No message to dump", "pending LRM operations at shutdown", "Lost connection to the CIB service", "Connection to the CIB terminated...", "Sending message to CIB service FAILED", "apply_xml_diff: Diff application failed!", "crmd.*Action A_RECOVER .* not supported", "unconfirmed_actions: Waiting on .* unconfirmed actions", "cib_native_msgready: Message pending on command channel", "crmd.*do_exit: Performing A_EXIT_1 - forcefully exiting the CRMd", "verify_stopped: Resource .* was active at shutdown. You may ignore this error if it is unmanaged.", "error: attrd_connection_destroy: Lost connection to attrd", "info: te_fence_node: Executing .* fencing operation", # "error: native_create_actions: Resource .*stonith::.* is active on 2 nodes attempting recovery", # "error: process_pe_message: Transition .* ERRORs found during PE processing", ] cib_ignore = [ "lrmd.*error: crm_ipc_read: Connection to stonith-ng failed", "lrmd.*error: mainloop_gio_callback: Connection to stonith-ng.* closed", "lrmd.*error: stonith_connection_destroy_cb: LRMD lost STONITH connection", "lrmd.*error: stonith_connection_failed: STONITH connection failed, finalizing .* pending operations", ] fullcomplist["cib"] = Process(self, "cib", pats = [ "State transition .* S_RECOVERY", "Respawning .* crmd", "Respawning .* attrd", "error: crm_ipc_read: Connection to cib_.* failed", "error: mainloop_gio_callback: Connection to cib_.* closed", "Connection to the CIB terminated...", "Child process crmd exited .* rc=2", "Child process attrd exited .* rc=1", "crmd.*Input I_TERMINATE from do_recover", "crmd.*I_ERROR.*crmd_cib_connection_destroy", "crmd.*do_exit: Could not recover from internal error", ], badnews_ignore = cib_ignore, common_ignore = self.common_ignore) fullcomplist["lrmd"] = Process(self, "lrmd", pats = [ "State transition .* S_RECOVERY", "LRM Connection failed", "Respawning .* crmd", "error: crm_ipc_read: Connection to lrmd failed", "error: mainloop_gio_callback: Connection to lrmd.* closed", "crmd.*I_ERROR.*lrm_connection_destroy", "Child process crmd exited .* rc=2", "crmd.*Input I_TERMINATE from do_recover", "crmd.*do_exit: Could not recover from internal error", ], badnews_ignore = self.common_ignore) fullcomplist["crmd"] = Process(self, "crmd", pats = [ # "WARN: determine_online_status: Node .* is unclean", # "Scheduling Node .* for STONITH", # "Executing .* fencing operation", # Only if the node wasn't the DC: "State transition S_IDLE", "State transition .* -> S_IDLE", ], badnews_ignore = self.common_ignore) fullcomplist["attrd"] = Process(self, "attrd", pats = [ ], badnews_ignore = self.common_ignore) fullcomplist["pengine"] = Process(self, "pengine", dc_pats = [ "State transition .* S_RECOVERY", "Respawning .* crmd", "Child process crmd exited .* rc=2", "crm_ipc_read: Connection to pengine failed", "error: mainloop_gio_callback: Connection to pengine.* closed", "crit: pe_ipc_destroy: Connection to the Policy Engine failed", "crmd.*I_ERROR.*save_cib_contents", "crmd.*Input I_TERMINATE from do_recover", "crmd.*do_exit: Could not recover from internal error", ], badnews_ignore = self.common_ignore) stonith_ignore = [ "LogActions: Recover Fencing", "update_failcount: Updating failcount for Fencing", "error: te_connect_stonith: Sign-in failed: triggered a retry", "stonith_connection_failed: STONITH connection failed, finalizing .* pending operations.", "process_lrm_event: LRM operation Fencing.* Error" ] stonith_ignore.extend(self.common_ignore) fullcomplist["stonith-ng"] = Process(self, "stonith-ng", process="stonithd", pats = [ "crm_ipc_read: Connection to stonith-ng failed", "stonith_connection_destroy_cb: LRMD lost STONITH connection", "mainloop_gio_callback: Connection to stonith-ng.* closed", "tengine_stonith_connection_destroy: Fencing daemon connection failed", "crmd.*stonith_api_add_notification: Callback already present", ], badnews_ignore = stonith_ignore) vgrind = self.Env["valgrind-procs"].split() for key in fullcomplist.keys(): if self.Env["valgrind-tests"]: if key in vgrind: # Processes running under valgrind can't be shot with "killall -9 processname" self.log("Filtering %s from the component list as it is being profiled by valgrind" % key) continue if key == "stonith-ng" and not self.Env["DoFencing"]: continue self.complist.append(fullcomplist[key]) #self.complist = [ fullcomplist["pengine"] ] return self.complist class crm_whitetank(crm_ais): ''' The crm version 3 cluster manager class. It implements the things we need to talk to and manipulate crm clusters running on top of openais ''' def __init__(self, Environment, randseed=None): crm_ais.__init__(self, Environment, randseed=randseed) self.update({ "Name" : "crm-whitetank", "StartCmd" : "service openais start", "StopCmd" : "service openais stop", "Pat:We_stopped" : "%s.*openais.*pcmk_shutdown: Shutdown complete", "Pat:They_stopped" : "%s crmd.*Node %s\[.*state is now lost", "Pat:They_dead" : "openais:.*Node %s is now: lost", "Pat:ChildKilled" : "%s openais.*Child process %s terminated with signal 9", "Pat:ChildRespawn" : "%s openais.*Respawning failed child process: %s", "Pat:ChildExit" : "Child process .* exited", }) def Components(self): self.ais_components() aisexec_ignore = [ "error: ais_dispatch: Receiving message .* failed", "crmd.*I_ERROR.*crmd_cib_connection_destroy", "cib.*error: cib_ais_destroy: AIS connection terminated", #"crmd.*error: crm_ais_destroy: AIS connection terminated", "crmd.*do_exit: Could not recover from internal error", "crmd.*I_TERMINATE.*do_recover", "attrd.*attrd_ais_destroy: Lost connection to OpenAIS service!", "stonithd.*error: AIS connection terminated", ] aisexec_ignore.extend(self.common_ignore) self.complist.append(Process(self, "aisexec", pats = [ "error: ais_dispatch: AIS connection failed", "crmd.*error: do_exit: Could not recover from internal error", "pengine.*Scheduling Node .* for STONITH", "stonithd.*requests a STONITH operation RESET on node", "stonithd.*Succeeded to STONITH the node", ], badnews_ignore = aisexec_ignore)) class crm_cs_v0(crm_ais): ''' The crm version 3 cluster manager class. It implements the things we need to talk to and manipulate crm clusters running against version 0 of our plugin ''' def __init__(self, Environment, randseed=None): crm_ais.__init__(self, Environment, randseed=randseed) self.update({ "Name" : "crm-plugin-v0", "StartCmd" : "service corosync start", "StopCmd" : "service corosync stop", # The next pattern is too early # "Pat:We_stopped" : "%s.*Service engine unloaded: Pacemaker Cluster Manager", # The next pattern would be preferred, but it doesn't always come out # "Pat:We_stopped" : "%s.*Corosync Cluster Engine exiting with status", "Pat:We_stopped" : "%s.*Service engine unloaded: corosync cluster quorum service", "Pat:They_stopped" : "%s crmd.*Node %s\[.*state is now lost", "Pat:They_dead" : "corosync:.*Node %s is now: lost", "Pat:ChildKilled" : "%s corosync.*Child process %s terminated with signal 9", "Pat:ChildRespawn" : "%s corosync.*Respawning failed child process: %s", }) def Components(self): self.ais_components() corosync_ignore = [ r"pacemakerd.*error: pcmk_child_exit: Child process .* exited", r"error: send_cpg_message: Sending message via cpg FAILED", r"crmd.*error: verify_stopped: Resource .* was active at shutdown. You may ignore this error if it is unmanaged.", r"error: pcmk_cpg_dispatch: Connection to the CPG API failed:", r"error: crm_ipc_read: Connection to cib_shm failed", r"error: mainloop_gio_callback: Connection to .* closed", r"crmd_cib_connection_destroys:", r"crmd.*error: do_recover: Action A_RECOVER .* not supported", r"error: do_log: FSA: Input I_TERMINATE.*do_recover", r"error: cib_ais_destroy: Corosync connection lost! Exiting.", r"attrd.*error: attrd_cib_connection_destroy: Connection to the CIB terminated...", r"error: send_ais_text: Sending message .* via cpg: FAILED", r"error: crm_ipc_read: Connection to stonith-ng failed", r"crit: tengine_stonith_connection_destroy: Fencing daemon connection failed", r"error: stonith_connection_destroy_cb: LRMD lost STONITH connection", r"error: stonith_connection_failed: STONITH connection failed", r"error: te_connect_stonith: Sign-in failed: triggered a retry", r"error: process_lrm_event: LRM operation Fencing.*", r"error: do_log: FSA: Input I_ERROR from crmd_cib_connection_destroy.* received in state", r"error: do_log: FSA: Input I_ERROR from do_shutdown_req.* received in state", r"warning: do_state_transition: State transition .* S_RECOVERY .*origin=crmd_cib_connection_destroy", r"warning: do_state_transition: State transition .* S_RECOVERY .*origin=do_shutdown_req", r"crmd.*error: cib_native_perform_op_delegate: Couldn't perform cib_slave operation.*: Transport endpoint is not connected", r"crmd.*error: cib_native_perform_op_delegate: CIB disconnected", ] self.complist.append(Process(self, "corosync", pats = [ r"pacemakerd.*error: cfg_connection_destroy: Connection destroyed", r"pacemakerd.*error: cpg_connection_destroy: Connection destroyed", r"attrd_ais_destroy: Lost connection to Corosync service!", r"stonith_peer_ais_destroy: AIS connection terminated", r"cib_ais_destroy: Corosync connection lost! Exiting.", r"crmd.*do_exit: Could not recover from internal error", r"pengine.*Scheduling Node .* for STONITH", r"log_operation: Operation .* for host .* with device .* returned: 0", r"tengine_stonith_notify: Peer .* was terminated .*: OK", ], badnews_ignore = corosync_ignore, common_ignore = self.common_ignore)) return self.complist class crm_cs_v1(crm_cs_v0): ''' The crm version 3 cluster manager class. It implements the things we need to talk to and manipulate crm clusters running on top of version 1 of our plugin ''' def __init__(self, Environment, randseed=None): crm_cs_v0.__init__(self, Environment, randseed=randseed) self.update({ "Name" : "crm-plugin-v1", "StartCmd" : "service corosync start && service pacemaker start", "StopCmd" : "service pacemaker stop; service corosync stop", "EpocheCmd" : "crm_node -e", "QuorumCmd" : "crm_node -q", "ParitionCmd" : "crm_node -p", "Pat:We_stopped" : "%s.*Service engine unloaded: corosync cluster quorum service", "Pat:They_stopped" : "%s crmd.*Node %s\[.*state is now lost", "Pat:They_dead" : "crmd.*Node %s\[.*state is now lost", "Pat:ChildKilled" : "%s pacemakerd.*Child process %s terminated with signal 9", "Pat:ChildRespawn" : "%s pacemakerd.*Respawning failed child process: %s", }) class crm_mcp(crm_cs_v0): ''' The crm version 4 cluster manager class. It implements the things we need to talk to and manipulate crm clusters running on top of native corosync (no plugins) ''' def __init__(self, Environment, randseed=None): crm_cs_v0.__init__(self, Environment, randseed=randseed) self.update({ "Name" : "crm-mcp", "StartCmd" : "service corosync start && service pacemaker start", "StopCmd" : "service pacemaker stop; service corosync stop", "EpocheCmd" : "crm_node -e", "QuorumCmd" : "crm_node -q", "ParitionCmd" : "crm_node -p", # Close enough... "Corosync Cluster Engine exiting normally" isn't printed # reliably and there's little interest in doing anything it "Pat:We_stopped" : "%s.*Unloading all Corosync service engines", "Pat:They_stopped" : "%s crmd.*Node %s\[.*state is now lost", "Pat:They_dead" : "crmd.*Node %s\[.*state is now lost", "Pat:ChildKilled" : "%s pacemakerd.*Child process %s terminated with signal 9", "Pat:ChildRespawn" : "%s pacemakerd.*Respawning failed child process: %s", "Pat:InfraUp" : "%s corosync.*Initializing transport", "Pat:PacemakerUp" : "%s pacemakerd.*Starting Pacemaker", }) class crm_cman(crm_cs_v0): ''' The crm version 3 cluster manager class. It implements the things we need to talk to and manipulate crm clusters running on top of openais ''' def __init__(self, Environment, randseed=None): crm_cs_v0.__init__(self, Environment, randseed=randseed) self.update({ "Name" : "crm-cman", "StartCmd" : "service cman start && service pacemaker start", "StopCmd" : "service pacemaker stop; service cman stop;", "EpocheCmd" : "crm_node -e --cman", "QuorumCmd" : "crm_node -q --cman", "ParitionCmd" : "crm_node -p --cman", "Pat:We_stopped" : "%s.*Unloading all Corosync service engines", "Pat:They_stopped" : "%s crmd.*Node %s\[.*state is now lost", "Pat:They_dead" : "crmd.*Node %s\[.*state is now lost", "Pat:ChildKilled" : "%s pacemakerd.*Child process %s terminated with signal 9", "Pat:ChildRespawn" : "%s pacemakerd.*Respawning failed child process: %s", }) diff --git a/cts/CM_lha.py b/cts/CM_lha.py index 410c8c7147..9e62ea8328 100755 --- a/cts/CM_lha.py +++ b/cts/CM_lha.py @@ -1,631 +1,627 @@ '''CTS: Cluster Testing System: LinuxHA v2 dependent modules... ''' __copyright__=''' Author: Huang Zhen Copyright (C) 2004 International Business Machines Additional Audits, Revised Start action, Default Configuration: Copyright (C) 2004 Andrew Beekhof ''' # # This program is free software; you can redistribute it and/or # modify it under the terms of the GNU General Public License # as published by the Free Software Foundation; either version 2 # of the License, or (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program; if not, write to the Free Software # Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. import os, sys, warnings from cts import CTS from cts.CTSvars import * from cts.CTS import * from cts.CIB import * from cts.CTStests import AuditResource try: from xml.dom.minidom import * except ImportError: sys.__stdout__.write("Python module xml.dom.minidom not found\n") sys.__stdout__.write("Please install python-xml or similar before continuing\n") sys.__stdout__.flush() sys.exit(1) ####################################################################### # # LinuxHA v2 dependent modules # ####################################################################### class crm_lha(ClusterManager): ''' The linux-ha version 2 cluster manager class. It implements the things we need to talk to and manipulate linux-ha version 2 clusters ''' def __init__(self, Environment, randseed=None): ClusterManager.__init__(self, Environment, randseed=randseed) #HeartbeatCM.__init__(self, Environment, randseed=randseed) self.fastfail = 0 self.clear_cache = 0 self.cib_installed = 0 self.config = None self.cluster_monitor = 0 self.use_short_names = 1 self.update({ "Name" : "crm-lha", "DeadTime" : 300, "StartTime" : 300, # Max time to start up "StableTime" : 30, "StartCmd" : "service heartbeat start > /dev/null 2>&1", "StopCmd" : "service heartbeat stop > /dev/null 2>&1", "StatusCmd" : "crmadmin -t 60000 -S %s 2>/dev/null", "EpocheCmd" : "crm_node -H -e", "QuorumCmd" : "crm_node -H -q", "ParitionCmd" : "crm_node -H -p", "CibQuery" : "cibadmin -Ql", "CibAddXml" : "cibadmin --modify -c --xml-text %s", "CibDelXpath" : "cibadmin --delete --xpath %s", # 300,000 == 5 minutes "RscRunning" : CTSvars.CRM_DAEMON_DIR + "/lrmd_test -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", "LogFileName" : Environment["LogFileName"], "UUIDQueryCmd" : "crmadmin -N", "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", # Patterns to look for in the log files for various occasions... "Pat:DC_IDLE" : "crmd.*State transition.*-> S_IDLE", # This wont work if we have multiple partitions "Pat:Local_started" : "%s .*The local CRM is operational", "Pat:Slave_started" : "%s .*State transition.*-> S_NOT_DC", "Pat:Master_started" : "%s .* State transition.*-> S_IDLE", "Pat:We_stopped" : "heartbeat.*%s.*Heartbeat shutdown complete", "Pat:Logd_stopped" : "%s logd:.*Exiting write process", "Pat:They_stopped" : "%s .*LOST:.* %s ", "Pat:They_dead" : "node %s.*: is dead", "Pat:TransitionComplete" : "Transition status: Complete: complete", "Pat:ChildKilled" : "%s heartbeat.*%s.*killed by signal 9", "Pat:ChildRespawn" : "%s heartbeat.*Respawning client.*%s", "Pat:ChildExit" : "(ERROR|error): Client .* exited with return code", "Pat:Fencing_start" : "Initiating remote operation .* for %s", - "Pat:Fencing_start_offset" : " for ", - "Pat:Fencing_ok" : "stonith.* log_operation: Operation .* for host '%s' with device .* returned: 0", - "Pat:Fencing_ok_offset" : "for host '", - # Bad news Regexes. Should never occur. "BadRegexes" : ( r" trace:", 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", exiting\.", r"WARN.*Ignoring HA message.*vote.*not in our membership list", r"pengine.*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"No need to invoke the TE", r"global_timer_callback:", r"Faking parameter digest creation", r"Parameters to .* action changed:", r"Parameters to .* changed", ), }) if self.Env["DoBSC"]: del self["Pat:They_stopped"] del self["Pat:Logd_stopped"] self.Env["use_logd"] = 0 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 errorstoignore(self): # At some point implement a more elegant solution that # also produces a report at the end '''Return list of errors which are known and very noisey should be ignored''' if 1: return [ "(ERROR|error): crm_abort: crm_glib_handler: ", "(ERROR|error): Message hist queue is filling up", "stonithd.*CRIT: external_hostlist: 'vmware gethosts' returned an empty hostlist", "stonithd.*(ERROR|error): Could not list nodes for stonith RA external/vmware.", "pengine.*Preventing .* from re-starting", ] return [] def install_config(self, node): if not self.ns.WaitForNodeToComeUp(node): self.log("Node %s is not up." % node) return None if not self.CIBsync.has_key(node) 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["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] = "" 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["Pat:Slave_started"]%node) watchpats.append(self["Pat:Master_started"]%node) idle_watch = CTS.LogWatcher(self.Env, self["LogFileName"], watchpats, "ClusterIdle") idle_watch.setwatch() out = self.rsh(node, self["StatusCmd"]%node, 1) self.debug("Node %s status: '%s'" %(node, out)) if not out or string.find(out, '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 looses messages if string.find(out, 'S_NOT_DC') != -1: # Up and stable return 2 if string.find(out, '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["Pat:DC_IDLE"]) self.debug("Waiting for cluster stability...") if timeout == None: timeout = self["DeadTime"] idle_watch = CTS.LogWatcher(self.Env, self["LogFileName"], watchpats, "ClusterStable", timeout) idle_watch.setwatch() any_up = 0 for node in self.Env["nodes"]: # have each node dump its current state if self.ShouldBeStatus[node] == "up": self.rsh(node, self["StatusCmd"] %node, 1) any_up = 1 if any_up == 0: self.debug("Cluster is inactive") return 1 ret = idle_watch.look() while ret: self.debug(ret) for node in nodes: 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["StatusCmd"]%node, 1) if not status_line: rc = 0 elif string.find(status_line, 'S_IDLE') != -1: rc = 1 elif string.find(status_line, 'S_INTEGRATION') != -1: rc = 1 elif string.find(status_line, 'S_FINALIZE_JOIN') != -1: rc = 1 elif string.find(status_line, 'S_POLICY_ENGINE') != -1: rc = 1 elif string.find(status_line, 'S_TRANSITION_ENGINE') != -1: rc = 1 return rc def active_resources(self, node): # [SM].* {node} matches Started, Slave, Master # Stopped wont be matched as it wont include {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["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["ParitionCmd"], 1) if not partition: self.log("no partition details for %s" %node) elif len(partition) > 2: partition = partition[:-1] 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) 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["QuorumCmd"], 1) if string.find(quorum, "1") != -1: return 1 elif string.find(quorum, "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 service", "Connection to the CIB terminated...", "Sending message to CIB service 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", "do_exit: Performing A_EXIT_1 - forcefully exiting the CRMd", "verify_stopped: Resource .* was active at shutdown. You may ignore this error if it is unmanaged.", ] stonith_ignore = [ "(ERROR|error): stonithd_signon: ", "update_failcount: Updating failcount for child_DoFencing", "(ERROR|error): te_connect_stonith: Sign-in failed: triggered a retry", "lrmd.*(ERROR|error): cl_get_value: wrong argument (reply)", "lrmd.*(ERROR|error): is_expected_msg:.* null message", "lrmd.*(ERROR|error): stonithd_receive_ops_result failed.", ] stonith_ignore.extend(common_ignore) ccm_ignore = [ "(ERROR|error): get_channel_token: No reply message - disconnected" ] ccm_ignore.extend(common_ignore) ccm = Process(self, "ccm", triggersreboot=self.fastfail, pats = [ "State transition .* S_RECOVERY", "CCM connection appears to have failed", "crmd.*Action A_RECOVER .* not supported", "crmd.*Input I_TERMINATE from do_recover", "Exiting to recover from CCM connection failure", "crmd.*do_exit: Could not recover from internal error", "crmd.*I_ERROR.*(ccm_dispatch|crmd_cib_connection_destroy)", "crmd.*exited with return code 2.", "attrd.*exited with return code 1.", "cib.*exited with return code 2.", # Not if it was fenced # "A new node joined the cluster", # "WARN: determine_online_status: Node .* is unclean", # "Scheduling Node .* for STONITH", # "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 = ccm_ignore) cib = Process(self, "cib", triggersreboot=self.fastfail, pats = [ "State transition .* S_RECOVERY", "Lost connection to the CIB service", "Connection to the CIB terminated...", "crmd.*Input I_TERMINATE from do_recover", "crmd.*I_ERROR.*crmd_cib_connection_destroy", "crmd.*do_exit: Could not recover from internal error", "crmd.*exited with return code 2.", "attrd.*exited with return code 1.", ], badnews_ignore = common_ignore) lrmd = Process(self, "lrmd", triggersreboot=self.fastfail, pats = [ "State transition .* S_RECOVERY", "LRM Connection failed", "crmd.*I_ERROR.*lrm_connection_destroy", "State transition S_STARTING -> S_PENDING", "crmd.*Input I_TERMINATE from do_recover", "crmd.*do_exit: Could not recover from internal error", "crmd.*exited with return code 2.", ], badnews_ignore = common_ignore) crmd = Process(self, "crmd", triggersreboot=self.fastfail, pats = [ # "WARN: determine_online_status: Node .* is unclean", # "Scheduling Node .* for STONITH", # "Executing .* fencing operation", # "tengine_stonith_callback: .*result=0", "State transition .* S_IDLE", "State transition S_STARTING -> S_PENDING", ], badnews_ignore = common_ignore) pengine = Process(self, "pengine", triggersreboot=self.fastfail, pats = [ "State transition .* S_RECOVERY", "crmd.*exited with return code 2.", "crmd.*Input I_TERMINATE from do_recover", "crmd.*do_exit: Could not recover from internal error", "crmd.*CRIT: pe_connection_destroy: Connection to the Policy Engine failed", "crmd.*I_ERROR.*save_cib_contents", "crmd.*exited with return code 2.", ], badnews_ignore = common_ignore, dc_only=1) if self.Env["DoFencing"] == 1 : complist.append(Process(self, "stoniths", triggersreboot=self.fastfail, dc_pats = [ "crmd.*CRIT: tengine_stonith_connection_destroy: Fencing daemon connection failed", "Attempting connection to fencing daemon", "te_connect_stonith: Connected", ], badnews_ignore = stonith_ignore)) if self.fastfail == 0: ccm.pats.extend([ "attrd .* exited with return code 1", "(ERROR|error): Respawning client .*attrd", "cib.* exited with return code 2", "(ERROR|error): Respawning client .*cib", "crmd.* exited with return code 2", "(ERROR|error): Respawning client .*crmd" ]) cib.pats.extend([ "attrd.* exited with return code 1", "(ERROR|error): Respawning client .*attrd", "crmd.* exited with return code 2", "(ERROR|error): Respawning client .*crmd" ]) lrmd.pats.extend([ "crmd.* exited with return code 2", "(ERROR|error): Respawning client .*crmd" ]) pengine.pats.extend([ "(ERROR|error): Respawning client .*crmd" ]) complist.append(ccm) complist.append(cib) complist.append(lrmd) complist.append(crmd) complist.append(pengine) return complist def NodeUUID(self, node): lines = self.rsh(node, self["UUIDQueryCmd"], 1) for line in lines: self.debug("UUIDLine:"+ line) m = re.search(r'%s.+\((.+)\)' % node, line) if m: return m.group(1) return "" def StandbyStatus(self, node): out=self.rsh(node, self["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["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['CibAddXml'] % (rsc_xml)) self.rsh(node, self['CibAddXml'] % (constraint_xml)) def RemoveDummyRsc(self, node, rid): constraint = "\"//rsc_location[@rsc='%s']\"" % (rid) rsc = "\"//primitive[@id='%s']\"" % (rid) self.rsh(node, self['CibDelXpath'] % constraint) self.rsh(node, self['CibDelXpath'] % rsc) ####################################################################### # # A little test code... # # Which you are advised to completely ignore... # ####################################################################### if __name__ == '__main__': pass diff --git a/cts/CTS.py b/cts/CTS.py index 6a546e8d10..68fb1ab9a0 100644 --- a/cts/CTS.py +++ b/cts/CTS.py @@ -1,1509 +1,1507 @@ '''CTS: Cluster Testing System: Main module Classes related to testing high-availability clusters... ''' __copyright__=''' Copyright (C) 2000, 2001 Alan Robertson Licensed under the GNU GPL. ''' # # This program is free software; you can redistribute it and/or # modify it under the terms of the GNU General Public License # as published by the Free Software Foundation; either version 2 # of the License, or (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program; if not, write to the Free Software # Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. import types, string, select, sys, time, re, os, struct, signal import time, syslog, random, traceback, base64, pickle, binascii, fcntl from socket import gethostbyname_ex from UserDict import UserDict from subprocess import Popen,PIPE from cts.CTSvars import * class CtsLab(UserDict): '''This class defines the Lab Environment for the Cluster Test System. It defines those things which are expected to change from test environment to test environment for the same cluster manager. It is where you define the set of nodes that are in your test lab what kind of reset mechanism you use, etc. This class is derived from a UserDict because we hold many different parameters of different kinds, and this provides provide a uniform and extensible interface useful for any kind of communication between the user/administrator/tester and CTS. At this point in time, it is the intent of this class to model static configuration and/or environmental data about the environment which doesn't change as the tests proceed. Well-known names (keys) are an important concept in this class. The HasMinimalKeys member function knows the minimal set of well-known names for the class. The following names are standard (well-known) at this time: nodes An array of the nodes in the cluster reset A ResetMechanism object logger An array of objects that log strings... CMclass The type of ClusterManager we are running (This is a class object, not a class instance) RandSeed Random seed. It is a triple of bytes. (optional) The CTS code ignores names it doesn't know about/need. The individual tests have access to this information, and it is perfectly acceptable to provide hints, tweaks, fine-tuning directions or other information to the tests through this mechanism. ''' def __init__(self): self.data = {} self.rsh = RemoteExec(self) self.RandomGen = random.Random() self.Scenario = None # Get a random seed for the random number generator. self["LogWatcher"] = "any" self["LogFileName"] = "/var/log/messages" self["OutputFile"] = None self["SyslogFacility"] = "daemon" self["CMclass"] = None self["logger"] = ([StdErrLog(self)]) self.SeedRandom() def SeedRandom(self, seed=None): if not seed: seed = int(time.time()) if self.has_key("RandSeed"): self.log("New random seed is: " + str(seed)) else: self.log("Random seed is: " + str(seed)) self["RandSeed"] = seed self.RandomGen.seed(str(seed)) def HasMinimalKeys(self): 'Return TRUE if our object has the minimal set of keys/values in it' result = 1 for key in self.MinimalKeys: if not self.has_key(key): result = None return result def log(self, args): "Log using each of the supplied logging methods" for logfcn in self._logfunctions: logfcn(string.strip(args)) def debug(self, args): "Log using each of the supplied logging methods" for logfcn in self._logfunctions: if logfcn.name() != "StdErrLog": logfcn("debug: %s" % string.strip(args)) def dump(self): keys = [] for key in self.keys(): keys.append(key) keys.sort() for key in keys: self.debug("Environment["+key+"]:\t"+str(self[key])) def run(self, Scenario, Iterations): if not Scenario: self.log("No scenario was defined") return 1 self.log("Cluster nodes: ") for node in self["nodes"]: self.log(" * %s" % (node)) if not Scenario.SetUp(): return 1 try : Scenario.run(Iterations) except : self.log("Exception by %s" % sys.exc_info()[0]) for logmethod in self["logger"]: traceback.print_exc(50, logmethod) Scenario.summarize() Scenario.TearDown() return 1 #ClusterManager.oprofileSave(Iterations) Scenario.TearDown() Scenario.summarize() if Scenario.Stats["failure"] > 0: return Scenario.Stats["failure"] elif Scenario.Stats["success"] != Iterations: self.log("No failure count but success != requested iterations") return 1 return 0 def __setitem__(self, key, value): '''Since this function gets called whenever we modify the dictionary (object), we can (and do) validate those keys that we know how to validate. For the most part, we know how to validate the "MinimalKeys" elements. ''' # # List of nodes in the system # if key == "nodes": self.Nodes = {} for node in value: # I don't think I need the IP address, etc. but this validates # the node name against /etc/hosts and/or DNS, so it's a # GoodThing(tm). try: self.Nodes[node] = gethostbyname_ex(node) except: print node+" not found in DNS... aborting" raise # # List of Logging Mechanism(s) # elif key == "logger": if len(value) < 1: raise ValueError("Must have at least one logging mechanism") for logger in value: if not callable(logger): raise ValueError("'logger' elements must be callable") self._logfunctions = value # # Cluster Manager Class # elif key == "CMclass": if value and not issubclass(value, ClusterManager): raise ValueError("'CMclass' must be a subclass of" " ClusterManager") # # Initial Random seed... # #elif key == "RandSeed": # if len(value) != 3: # raise ValueError("'Randseed' must be a 3-element list/tuple") # for elem in value: # if not isinstance(elem, types.IntType): # raise ValueError("'Randseed' list must all be ints") self.data[key] = value def IsValidNode(self, node): 'Return TRUE if the given node is valid' return self.Nodes.has_key(node) def __CheckNode(self, node): "Raise a ValueError if the given node isn't valid" if not self.IsValidNode(node): raise ValueError("Invalid node [%s] in CheckNode" % node) def RandomNode(self): '''Choose a random node from the cluster''' return self.RandomGen.choice(self["nodes"]) class Logger: TimeFormat = "%b %d %H:%M:%S\t" def __call__(self, lines): raise ValueError("Abstract class member (__call__)") def write(self, line): return self(line.rstrip()) def writelines(self, lines): for s in lines: self.write(s) return 1 def flush(self): return 1 def isatty(self): return None class SysLog(Logger): # http://docs.python.org/lib/module-syslog.html defaultsource="CTS" map = { "kernel": syslog.LOG_KERN, "user": syslog.LOG_USER, "mail": syslog.LOG_MAIL, "daemon": syslog.LOG_DAEMON, "auth": syslog.LOG_AUTH, "lpr": syslog.LOG_LPR, "news": syslog.LOG_NEWS, "uucp": syslog.LOG_UUCP, "cron": syslog.LOG_CRON, "local0": syslog.LOG_LOCAL0, "local1": syslog.LOG_LOCAL1, "local2": syslog.LOG_LOCAL2, "local3": syslog.LOG_LOCAL3, "local4": syslog.LOG_LOCAL4, "local5": syslog.LOG_LOCAL5, "local6": syslog.LOG_LOCAL6, "local7": syslog.LOG_LOCAL7, } def __init__(self, labinfo): if labinfo.has_key("syslogsource"): self.source=labinfo["syslogsource"] else: self.source=SysLog.defaultsource self.facility="daemon" if labinfo.has_key("SyslogFacility") and labinfo["SyslogFacility"]: if SysLog.map.has_key(labinfo["SyslogFacility"]): self.facility=labinfo["SyslogFacility"] else: raise ValueError("%s: bad syslog facility"%labinfo["SyslogFacility"]) self.facility=SysLog.map[self.facility] syslog.openlog(self.source, 0, self.facility) def setfacility(self, facility): self.facility = facility if SysLog.map.has_key(self.facility): self.facility=SysLog.map[self.facility] syslog.closelog() syslog.openlog(self.source, 0, self.facility) def __call__(self, lines): if isinstance(lines, types.StringType): syslog.syslog(lines) else: for line in lines: syslog.syslog(line) def name(self): return "Syslog" class StdErrLog(Logger): def __init__(self, labinfo): pass def __call__(self, lines): t = time.strftime(Logger.TimeFormat, time.localtime(time.time())) if isinstance(lines, types.StringType): sys.__stderr__.writelines([t, lines, "\n"]) else: for line in lines: sys.__stderr__.writelines([t, line, "\n"]) sys.__stderr__.flush() def name(self): return "StdErrLog" class FileLog(Logger): def __init__(self, labinfo, filename=None): if filename == None: filename=labinfo["LogFileName"] self.logfile=filename import os self.hostname = os.uname()[1]+" " self.source = "CTS: " def __call__(self, lines): fd = open(self.logfile, "a") t = time.strftime(Logger.TimeFormat, time.localtime(time.time())) if isinstance(lines, types.StringType): fd.writelines([t, self.hostname, self.source, lines, "\n"]) else: for line in lines: fd.writelines([t, self.hostname, self.source, line, "\n"]) fd.close() def name(self): return "FileLog" class RemoteExec: '''This is an abstract remote execution class. It runs a command on another machine - somehow. The somehow is up to us. This particular class uses ssh. Most of the work is done by fork/exec of ssh or scp. ''' def __init__(self, Env=None, silent=False): self.Env = Env self.silent = silent # -n: no stdin, -x: no X11, # -o ServerAliveInterval=5 disconnect after 3*5s if the server stops responding self.Command = "ssh -l root -n -x -o ServerAliveInterval=5 -o ConnectTimeout=10 -o TCPKeepAlive=yes -o ServerAliveCountMax=3 " # -B: batch mode, -q: no stats (quiet) self.CpCommand = "scp -B -q" self.OurNode=string.lower(os.uname()[1]) def enable_qarsh(self): # http://nstraz.wordpress.com/2008/12/03/introducing-qarsh/ self.log("Using QARSH for connections to cluster nodes") self.Command = "qarsh -t 300 -l root" self.CpCommand = "qacp -q" def _fixcmd(self, cmd): return re.sub("\'", "'\\''", cmd) def _cmd(self, *args): '''Compute the string that will run the given command on the given remote system''' args= args[0] sysname = args[0] command = args[1] #print "sysname: %s, us: %s" % (sysname, self.OurNode) if sysname == None or string.lower(sysname) == self.OurNode or sysname == "localhost": ret = command else: ret = self.Command + " " + sysname + " '" + self._fixcmd(command) + "'" #print ("About to run %s\n" % ret) return ret def log(self, args): if not self.silent: if not self.Env: print (args) else: self.Env.log(args) def debug(self, args): if not self.silent: if not self.Env: print (args) else: self.Env.debug(args) def __call__(self, node, command, stdout=0, synchronous=1, silent=False, blocking=True): '''Run the given command on the given remote system If you call this class like a function, this is the function that gets called. It just runs it roughly as though it were a system() call on the remote machine. The first argument is name of the machine to run it on. ''' rc = 0 result = None if not synchronous: proc = Popen(self._cmd([node, command]), stdout = PIPE, stderr = PIPE, close_fds = True, shell = True) if not silent: self.debug("cmd: async: target=%s, rc=%d: %s" % (node, proc.pid, command)) if proc.pid > 0: return 0 return -1 proc = Popen(self._cmd([node, command]), stdout = PIPE, stderr = PIPE, close_fds = True, shell = True) #if not blocking: # fcntl.fcntl(proc.stdout.fileno(), fcntl.F_SETFL, os.O_NONBLOCK) if proc.stdout: if stdout == 1: result = proc.stdout.readline() else: result = proc.stdout.readlines() proc.stdout.close() else: self.log("No stdout stream") rc = proc.wait() if not silent: self.debug("cmd: target=%s, rc=%d: %s" % (node, rc, command)) if stdout == 1: return result if proc.stderr: errors = proc.stderr.readlines() proc.stderr.close() if not silent: for err in errors: self.debug("cmd: stderr: %s" % err) if stdout == 0: if not silent and result: for line in result: self.debug("cmd: stdout: %s" % line) return rc return (rc, result) def cp(self, source, target, silent=False): '''Perform a remote copy''' cpstring = self.CpCommand + " \'" + source + "\'" + " \'" + target + "\'" rc = os.system(cpstring) if not silent: self.debug("cmd: rc=%d: %s" % (rc, cpstring)) return rc has_log_watcher = {} log_watcher_bin = CTSvars.CRM_DAEMON_DIR + "/cts_log_watcher.py" log_watcher = """ import sys, os, fcntl ''' Remote logfile reader for CTS Reads a specified number of lines from the supplied offset Returns the current offset Contains logic for handling truncation ''' limit = 0 offset = 0 prefix = '' filename = '/var/log/messages' skipthis=None args=sys.argv[1:] for i in range(0, len(args)): if skipthis: skipthis=None continue elif args[i] == '-l' or args[i] == '--limit': skipthis=1 limit = int(args[i+1]) elif args[i] == '-f' or args[i] == '--filename': skipthis=1 filename = args[i+1] elif args[i] == '-o' or args[i] == '--offset': skipthis=1 offset = args[i+1] elif args[i] == '-p' or args[i] == '--prefix': skipthis=1 prefix = args[i+1] elif args[i] == '-t' or args[i] == '--tag': skipthis=1 if not os.access(filename, os.R_OK): print prefix + 'Last read: %d, limit=%d, count=%d - unreadable' % (0, limit, 0) sys.exit(1) logfile=open(filename, 'r') logfile.seek(0, os.SEEK_END) newsize=logfile.tell() if offset != 'EOF': offset = int(offset) if newsize >= offset: logfile.seek(offset) else: print prefix + ('File truncated from %d to %d' % (offset, newsize)) if (newsize*1.05) < offset: logfile.seek(0) # else: we probably just lost a few logs after a fencing op # continue from the new end # TODO: accept a timestamp and discard all messages older than it # Don't block when we reach EOF fcntl.fcntl(logfile.fileno(), fcntl.F_SETFL, os.O_NONBLOCK) count = 0 while True: if logfile.tell() >= newsize: break elif limit and count >= limit: break line = logfile.readline() if not line: break print line.strip() count += 1 print prefix + 'Last read: %d, limit=%d, count=%d' % (logfile.tell(), limit, count) logfile.close() """ class SearchObj: def __init__(self, Env, filename, host=None, name=None): self.Env = Env self.host = host self.name = name self.filename = filename self.cache = [] self.offset = "EOF" if host == None: host = "localhost" global has_log_watcher if not has_log_watcher.has_key(host): global log_watcher global log_watcher_bin self.debug("Installing %s on %s" % (log_watcher_bin, host)) self.Env.rsh(host, '''echo "%s" > %s''' % (log_watcher, log_watcher_bin), silent=True) has_log_watcher[host] = 1 self.next() def __str__(self): if self.host: return "%s:%s" % (self.host, self.filename) return self.filename def log(self, args): message = "lw: %s: %s" % (self, args) if not self.Env: print (message) else: self.Env.log(message) def debug(self, args): message = "lw: %s: %s" % (self, args) if not self.Env: print (message) else: self.Env.debug(message) def next(self): cache = [] if not len(self.cache): global log_watcher_bin (rc, lines) = self.Env.rsh( self.host, "python %s -t %s -p CTSwatcher: -f %s -o %s" % (log_watcher_bin, self.name, self.filename, self.offset), stdout=None, silent=True, blocking=False) for line in lines: match = re.search("^CTSwatcher:Last read: (\d+)", line) if match: last_offset = self.offset self.offset = match.group(1) #if last_offset == "EOF": self.debug("Got %d lines, new offset: %s" % (len(lines), self.offset)) elif re.search("^CTSwatcher:.*truncated", line): self.log(line) elif re.search("^CTSwatcher:", line): self.debug("Got control line: "+ line) else: cache.append(line) return cache class LogWatcher(RemoteExec): '''This class watches logs for messages that fit certain regular expressions. Watching logs for events isn't the ideal way to do business, but it's better than nothing :-) On the other hand, this class is really pretty cool ;-) The way you use this class is as follows: Construct a LogWatcher object Call setwatch() when you want to start watching the log Call look() to scan the log looking for the patterns ''' def __init__(self, Env, log, regexes, name="Anon", timeout=10, debug_level=None, silent=False): '''This is the constructor for the LogWatcher class. It takes a log name to watch, and a list of regular expressions to watch for." ''' RemoteExec.__init__(self, Env) # Validate our arguments. Better sooner than later ;-) for regex in regexes: assert re.compile(regex) self.name = name self.regexes = regexes self.filename = log self.debug_level = debug_level self.whichmatch = -1 self.unmatched = None self.file_list = [] self.line_cache = [] if not silent: for regex in self.regexes: self.debug("Looking for regex: "+regex) self.Timeout = int(timeout) self.returnonlymatch = None def debug(self, args): message = "lw: %s: %s" % (self.name, args) if not self.Env: print (message) else: self.Env.debug(message) def setwatch(self): '''Mark the place to start watching the log from. ''' if self.Env["LogWatcher"] == "remote": for node in self.Env["nodes"]: self.file_list.append(SearchObj(self.Env, self.filename, node, self.name)) else: self.file_list.append(SearchObj(self.Env, self.filename)) def __del__(self): if self.debug_level > 1: self.debug("Destroy") def ReturnOnlyMatch(self, onlymatch=1): '''Specify one or more subgroups of the match to return rather than the whole string http://www.python.org/doc/2.5.2/lib/match-objects.html ''' self.returnonlymatch = onlymatch def __get_lines(self): if not len(self.file_list): raise ValueError("No sources to read from") for f in self.file_list: lines = f.next() if len(lines): self.line_cache.extend(lines) def look(self, timeout=None, silent=False): '''Examine the log looking for the given patterns. It starts looking from the place marked by setwatch(). This function looks in the file in the fashion of tail -f. It properly recovers from log file truncation, but not from removing and recreating the log. It would be nice if it recovered from this as well :-) We return the first line which matches any of our patterns. ''' if timeout == None: timeout = self.Timeout lines=0 begin=time.time() end=begin+timeout+1 if self.debug_level > 2: self.debug("starting single search: timeout=%d, begin=%d, end=%d" % (timeout, begin, end)) if not self.regexes: self.debug("Nothing to look for") return None self.__get_lines() while True: if len(self.line_cache): lines += 1 line = self.line_cache[0] self.line_cache.remove(line) which=-1 if re.search("CTS:", line): continue if self.debug_level > 2: self.debug("Processing: "+ line) for regex in self.regexes: which=which+1 if self.debug_level > 2: self.debug("Comparing line to: "+ regex) #matchobj = re.search(string.lower(regex), string.lower(line)) matchobj = re.search(regex, line) if matchobj: self.whichmatch=which if self.returnonlymatch: return matchobj.group(self.returnonlymatch) else: self.debug("Matched: "+line) if self.debug_level > 1: self.debug("With: "+ regex) return line elif timeout > 0 and end > time.time(): time.sleep(1) self.__get_lines() elif timeout > 0: # Grab any relevant messages that might have arrived since # the last time the buffer was populated self.__get_lines() # Don't come back here again timeout = 0 else: self.debug("Single search terminated: start=%d, end=%d, now=%d, lines=%d" % (begin, end, time.time(), lines)) return None self.debug("How did we get here") return None def lookforall(self, timeout=None, allow_multiple_matches=None, silent=False): '''Examine the log looking for ALL of the given patterns. It starts looking from the place marked by setwatch(). We return when the timeout is reached, or when we have found ALL of the regexes that were part of the watch ''' if timeout == None: timeout = self.Timeout save_regexes = self.regexes returnresult = [] if not silent: self.debug("starting search: timeout=%d" % timeout) for regex in self.regexes: if self.debug_level > 2: self.debug("Looking for regex: "+regex) while (len(self.regexes) > 0): oneresult = self.look(timeout) if not oneresult: self.unmatched = self.regexes self.matched = returnresult self.regexes = save_regexes return None returnresult.append(oneresult) if not allow_multiple_matches: del self.regexes[self.whichmatch] else: # Allow multiple regexes to match a single line tmp_regexes = self.regexes self.regexes = [] which = 0 for regex in tmp_regexes: matchobj = re.search(regex, oneresult) if not matchobj: self.regexes.append(regex) self.unmatched = None self.matched = returnresult self.regexes = save_regexes return returnresult class NodeStatus: def __init__(self, Env): self.Env = Env def IsNodeBooted(self, node): '''Return TRUE if the given node is booted (responds to pings)''' return self.Env.rsh("localhost", "ping -nq -c1 -w1 %s" % node, silent=True) == 0 def IsSshdUp(self, node): rc = self.Env.rsh(node, "true", silent=True) return rc == 0 def WaitForNodeToComeUp(self, node, Timeout=300): '''Return TRUE when given node comes up, or None/FALSE if timeout''' timeout=Timeout anytimeouts=0 while timeout > 0: if self.IsNodeBooted(node) and self.IsSshdUp(node): if anytimeouts: # Fudge to wait for the system to finish coming up time.sleep(30) self.Env.debug("Node %s now up" % node) return 1 time.sleep(30) if (not anytimeouts): self.Env.debug("Waiting for node %s to come up" % node) anytimeouts=1 timeout = timeout - 1 self.Env.log("%s did not come up within %d tries" % (node, Timeout)) answer = raw_input('Continue? [nY]') if answer and answer == "n": raise ValueError("%s did not come up within %d tries" % (node, Timeout)) def WaitForAllNodesToComeUp(self, nodes, timeout=300): '''Return TRUE when all nodes come up, or FALSE if timeout''' for node in nodes: if not self.WaitForNodeToComeUp(node, timeout): return None return 1 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 self.keys(): if self[key] == None: raise ValueError("Improper derivation: self[" + key + "] must be overridden by subclass.") def __init__(self, Environment, randseed=None): self.Env = Environment self.__InitialConditions() self.clear_cache = 0 self.TestLoggingLevel=0 self.data = { "up" : "up", # Status meaning up "down" : "down", # Status meaning down "StonithCmd" : "stonith -t baytech -p '10.10.10.100 admin admin' %s", "DeadTime" : 30, # Max time to detect dead node... "StartTime" : 90, # Max time to start up # # These next values need to be overridden in the derived class. # "Name" : None, "StartCmd" : None, "StopCmd" : None, "StatusCmd" : None, #"RereadCmd" : None, "BreakCommCmd" : None, "FixCommCmd" : None, #"TestConfigDir" : None, "LogFileName" : None, #"Pat:Master_started" : None, #"Pat:Slave_started" : None, "Pat:We_stopped" : None, "Pat:They_stopped" : None, "Pat:InfraUp" : "%s", "Pat:PacemakerUp" : "%s", "BadRegexes" : None, # A set of "bad news" regexes # to apply to the log } self.rsh = self.Env.rsh self.ShouldBeStatus={} self.ns = NodeStatus(self.Env) self.OurNode=string.lower(os.uname()[1]) self.__instance_errorstoignore = [] 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 errorstoignore(self): '''Return list of errors which are 'normal' and should be ignored''' return [] def log(self, args): self.Env.log(args) def debug(self, args): self.Env.debug(args) def prepare(self): '''Finish the Initialization process. Prepare to test...''' for node in self.Env["nodes"]: if self.StataCM(node): self.ShouldBeStatus[node]="up" else: self.ShouldBeStatus[node]="down" self.unisolate_node(node) 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_helper(self, filename, destdir=None, nodes=None, sourcedir=None): if sourcedir == None: sourcedir = CTSvars.CTS_home file_with_path="%s/%s" % (sourcedir, filename) if not nodes: nodes = self.Env["nodes"] if not destdir: destdir=CTSvars.CTS_home self.debug("Installing %s to %s on %s" % (filename, destdir, repr(self.Env["nodes"]))) for node in nodes: self.rsh(node, "mkdir -p %s" % destdir) self.rsh.cp(file_with_path, "root@%s:%s/%s" % (node, destdir, filename)) return file_with_path def install_config(self, node): return None def clear_all_caches(self): if self.clear_cache: for node in self.Env["nodes"]: if self.ShouldBeStatus[node] == "down": self.debug("Removing cache file on: "+node) self.rsh(node, "rm -f "+CTSvars.HA_VARLIBHBDIR+"/hostcache") else: self.debug("NOT Removing cache file on: "+node) def prepare_fencing_watcher(self, node): # 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 if self.HasQuorum(None): return None if not self.has_key("Pat:Fencing_start"): return None if not self.has_key("Pat:Fencing_ok"): return None if not self.has_key("Pat:Fencing_ok_offset"): return None stonith = None stonithPats = [] for peer in self.Env["nodes"]: if peer != node and self.ShouldBeStatus[peer] != "up": stonithPats.append(self["Pat:Fencing_ok"] % peer) stonithPats.append(self["Pat:Fencing_start"] % peer) # Look for STONITH ops, depending on Env["at-boot"] we might need to change the nodes status stonith = LogWatcher(self.Env, self["LogFileName"], stonithPats, "StartaCM", 0) 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) # Make sure everyone is online before continuing self.ns.WaitForAllNodesToComeUp(self.Env["nodes"]) if not stonith: return peer_list if not self.HasQuorum(None) and len(self.Env["nodes"]) > 2: # We didn't gain quorum - we shouldn't have shot anyone return peer_list # 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 - # Order is important here - if re.search(self["Pat:Fencing_ok_offset"], shot): - start = line.find(self["Pat:Fencing_ok_offset"]) + len(self["Pat:Fencing_ok_offset"]) - peer = line[start:].split("' ")[0] - peer_state[peer] = "complete" + for n in self.Env["nodes"]: + if re.search(self["Pat:Fencing_ok"] % n, shot): + peer = n + peer_state[peer] = "complete" + self.__instance_errorstoignore.append(self["Pat:Fencing_ok"] % peer) - elif re.search(self["Pat:Fencing_start_offset"], shot): - start = line.find(self["Pat:Fencing_start_offset"]) + len(self["Pat:Fencing_start_offset"]) - peer = line[start:].split(": ")[0] - peer_state[peer] = "in-progress" + elif re.search(self["Pat:Fencing_start"] % n, shot): + peer = n + peer_state[peer] = "in-progress" + self.__instance_errorstoignore.append(self["Pat:Fencing_start"] % peer) - else: - self.log("ERROR: Unknown stonith match: %s" % line) + else: + self.log("ERROR: Unknown stonith match: %s" % line) if 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.__instance_errorstoignore.append(self["Pat:Fencing_ok"] % peer) - self.__instance_errorstoignore.append(self["Pat:Fencing_start"] % peer) self.ShouldBeStatus[peer]="down" self.debug(" Peer %s was fenced as a result of %s starting: %s" % (peer, node, peer_state[peer])) self.ns.WaitForNodeToComeUp(peer, self["DeadTime"]) # Poll until it comes up if self.Env["at-boot"]: if not self.StataCM(peer): time.sleep(self["StartTime"]) if not self.StataCM(peer): self.log("ERROR: Peer %s failed to restart after being fenced" % peer) return None self.ShouldBeStatus[peer]="up" return peer_list def StartaCM(self, node, verbose=False): '''Start up the cluster manager on a given node''' if verbose: self.log("Starting %s on node %s" %(self["Name"], node)) else: self.debug("Starting %s on node %s" %(self["Name"], node)) ret = 1 if not self.ShouldBeStatus.has_key(node): 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["Pat:Local_started"] % node) if self.upcount() == 0: patterns.append(self["Pat:Master_started"] % node) else: patterns.append(self["Pat:Slave_started"] % node) watch = LogWatcher( self.Env, self["LogFileName"], patterns, "StartaCM", self["StartTime"]+10) self.install_config(node) self.ShouldBeStatus[node] = "any" if self.StataCM(node) and self.cluster_stable(self["DeadTime"]): self.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["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["StartCmd"]) stonith = self.prepare_fencing_watcher(node) watch.setwatch() if self.rsh(node, startCmd) != 0: self.log ("Warn: Start command failed on node %s" %(node)) return None self.ShouldBeStatus[node]="up" watch_result = watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.log ("Warn: Startup pattern not found: %s" %(regex)) if watch_result and self.cluster_stable(self["DeadTime"]): #self.debug("Found match: "+ repr(watch_result)) self.fencing_cleanup(node, stonith) return 1 elif self.StataCM(node) and self.cluster_stable(self["DeadTime"]): self.fencing_cleanup(node, stonith) return 1 self.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.log("Starting %s on node %s" %(self["Name"], node)) else: self.debug("Starting %s on node %s" %(self["Name"], node)) # 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["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["StartCmd"]) self.rsh(node, startCmd, synchronous=0) self.ShouldBeStatus[node]="up" return 1 def StopaCM(self, node, verbose=False): '''Stop the cluster manager on a given node''' if verbose: self.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": return 1 if self.rsh(node, self["StopCmd"]) == 0: # Make sure we can continue even if corosync leaks # fdata-* is the old name #self.rsh(node, "rm -f /dev/shm/qb-* /dev/shm/fdata-*") self.ShouldBeStatus[node]="down" self.cluster_stable(self["DeadTime"]) return 1 else: self.log ("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["StopCmd"], synchronous=0) self.ShouldBeStatus[node]="down" return 1 def cluster_stable(self, timeout = None): time.sleep(self["StableTime"]) return 1 def node_stable(self, node): 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["RereadCmd"]) if rc == 0: return 1 else: self.log ("Could not force %s on node %s to reread its config" % (self["Name"], node)) return None def StataCM(self, node): '''Report the status of the cluster manager on a given node''' out=self.rsh(node, self["StatusCmd"], 1) ret= (string.find(out, 'stopped') == -1) try: if ret: if self.ShouldBeStatus[node] == "down": self.log( "Node status for %s is %s but we think it should be %s" % (node, "up", self.ShouldBeStatus[node])) else: 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])) except KeyError: pass if ret: self.ShouldBeStatus[node]="up" else: self.ShouldBeStatus[node]="down" return ret def startall(self, nodelist=None, verbose=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. ''' ret = 1 map = {} if not nodelist: nodelist=self.Env["nodes"] for node in nodelist: if self.ShouldBeStatus[node] == "down": self.ns.WaitForAllNodesToComeUp(nodelist, 300) if not self.StartaCM(node, verbose=verbose): ret = 0 return ret def stopall(self, nodelist=None, verbose=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": if not self.StopaCM(node, verbose=verbose): 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["BreakCommCmd"] % self.key_for_node(node)) if rc != 0: self.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["FixCommCmd"] % self.key_for_node(node), synchronous=0) self.rsh(node, self["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["ReduceCommCmd"]%(self.Env["XmitLoss"],self.Env["RecvLoss"])) if rc == 0: return 1 else: self.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["RestoreCommCmd"]); if rc == 0: return 1 else: self.log("Could not restore the communication between the nodes from node: %s" % node) return None def HasQuorum(self, node_list): "Return TRUE if the cluster currently has quorum" # 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 raise ValueError("Abstract Class member (HasQuorum)") def Components(self): raise ValueError("Abstract Class member (Components)") 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:/usr/lib/heartbeat/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") class Resource: ''' This is an HA resource (not a resource group). A resource group is just an ordered list of Resource objects. ''' def __init__(self, cm, rsctype=None, instance=None): self.CM = cm self.ResourceType = rsctype self.Instance = instance self.needs_quorum = 1 def Type(self): return self.ResourceType def Instance(self, nodename): return self.Instance def IsRunningOn(self, nodename): ''' This member function returns true if our resource is running on the given node in the cluster. It is analagous to the "status" operation on SystemV init scripts and heartbeat scripts. FailSafe calls it the "exclusive" operation. ''' raise ValueError("Abstract Class member (IsRunningOn)") return None def IsWorkingCorrectly(self, nodename): ''' This member function returns true if our resource is operating correctly on the given node in the cluster. Heartbeat does not require this operation, but it might be called the Monitor operation, which is what FailSafe calls it. For remotely monitorable resources (like IP addresses), they *should* be monitored remotely for testing. ''' raise ValueError("Abstract Class member (IsWorkingCorrectly)") return None def Start(self, nodename): ''' This member function starts or activates the resource. ''' raise ValueError("Abstract Class member (Start)") return None def Stop(self, nodename): ''' This member function stops or deactivates the resource. ''' raise ValueError("Abstract Class member (Stop)") return None def __repr__(self): if (self.Instance and len(self.Instance) > 1): return "{" + self.ResourceType + "::" + self.Instance + "}" else: return "{" + self.ResourceType + "}" class Component: def kill(self, node): None class Process(Component): def __init__(self, cm, name, process=None, dc_only=0, pats=[], dc_pats=[], badnews_ignore=[], common_ignore=[], triggersreboot=0): self.name = str(name) self.dc_only = dc_only self.pats = pats self.dc_pats = dc_pats self.CM = cm self.badnews_ignore = badnews_ignore self.badnews_ignore.extend(common_ignore) self.triggersreboot = triggersreboot if process: self.proc = str(process) else: self.proc = str(name) self.KillCmd = "killall -9 " + self.proc def kill(self, node): if self.CM.rsh(node, self.KillCmd) != 0: self.CM.log ("ERROR: Kill %s failed on node %s" %(self.name,node)) return None return 1