diff --git a/cts/CIB.py b/cts/CIB.py index 08ad210df2..8fbae393eb 100644 --- a/cts/CIB.py +++ b/cts/CIB.py @@ -1,565 +1,565 @@ '''CTS: Cluster Testing System: CIB generator ''' __copyright__=''' Author: Andrew Beekhof Copyright (C) 2008 Andrew Beekhof ''' from UserDict import UserDict import sys, time, types, syslog, os, struct, string, signal, traceback, warnings from cts.CTSvars import * from cts.CTS import ClusterManager class CibBase: cts_cib = None cib_tmpfile = None version = "unknown" feature_set = "unknown" target = None def __init__(self, CM, tmpfile=None): self.CM = CM #self.target = self.CM.Env["nodes"][0] if not tmpfile: warnings.filterwarnings("ignore") self.cib_tmpfile=os.tmpnam() warnings.resetwarnings() else: self.cib_tmpfile = tmpfile def version(self): return self.version def NextIP(self): fields = string.split(self.CM.Env["IPBase"], '.') fields[3] = str(int(fields[3])+1) ip = string.join(fields, '.') self.CM.Env["IPBase"] = ip return ip class CIB06(CibBase): version = "transitional-0.6" coloc_template = """""" cib_template =''' %s %s %s ''' cib_option_template = ''' ''' lsb_resource = ''' ''' clustermon_location_constraint = ''' ''' resource_group_template = '''%s %s %s''' per_node_constraint_template = ''' ''' pingd_constraint_template = ''' ''' dummy_resource_template = ''' ''' clustermon_resource_template = ''' ''' master_slave_resource = ''' ''' pingd_resource_template = """ """ stonith_resource_template = """ """ bsc_template = ''' ''' def NewIP(self, name=None): template = ''' ''' ip = self.NextIP() if not name: name = "r"+ip return template % (name, name, name, name, name, ip) def NewHBIP(self, name=None): template = ''' ''' ip = self.NextIP() if not name: name = "r"+ip return template % (name, name, name, name, ip) def NewDummy(self, name): return self.dummy_resource_template % (name, name, name, name) def install(self, target): self.CM.rsh("localhost", "echo \'" + self.contents(target) + "\' > " + self.cib_tmpfile) rc = self.CM.rsh.cp(cib_file, "root@%s:%s/cib.xml" + (target, CTSvars.CRM_CONFIG_DIR)) if rc != 0: raise ValueError("Can not copy %s to %s (%d)"%(self.cib_tmpfile, target, rc)) self.CM.rsh(target, "chown "+CTSvars.CRM_DAEMON_USER+" "+CTSvars.CRM_CONFIG_DIR+"/cib.xml") self.CM.rsh("localhost", "rm -f "+self.cib_tmpfile) def contents(self, target=None): # fencing resource if self.cts_cib: return self.cts_cib nodelist = "" num_nodes = 0 for node in self.CM.Env["nodes"]: nodelist += node + " " num_nodes = num_nodes + 1 no_quorum = "stop" if num_nodes < 3: no_quorum = "ignore" self.CM.debug("Cluster only has %d nodes, ignoring quorum" % num_nodes) #make up crm config cib_options = self.cib_option_template % (self.CM.Env["DoFencing"], no_quorum) #create resources and their constraints resources = "" constraints = "" if self.CM.Env["DoBSC"] == 1: cib_options = cib_options + self.bsc_template if self.CM.Env["CIBResource"] != 1: # generate cib self.cts_cib = self.cib_template % (cib_options, resources, constraints) return self.cts_cib if self.CM.cluster_monitor == 1: resources += self.clustermon_resource_template constraints += self.clustermon_location_constraint ip1_rsc = self.NewIP() ip2_rsc = self.NewHBIP() ip3_rsc = self.NewIP() resources += self.resource_group_template % (ip1_rsc, ip2_rsc, ip3_rsc) # lsb resource resources += self.lsb_resource # Migrator resources += self.NewDummy("migrator") constraints += self.coloc_template % ("group-with-master", "group-1", "master-1", "Master", "INFINITY") constraints += self.coloc_template % ("lsb-with-group", "lsb_dummy", "group-1", "Started", "INFINITY") # per node resource for node in self.CM.Env["nodes"]: per_node_resources = self.NewIP("rsc_"+node) per_node_constraint = self.per_node_constraint_template % (node, "rsc_"+node, node) resources += per_node_resources constraints += per_node_constraint # Ping the test master resources += self.pingd_resource_template % os.uname()[1] # Require conectivity to run constraints += self.pingd_constraint_template % ("master-1", "master-1", "m", "Started", "m", 1) if self.CM.Env["DoFencing"]: p_name = None p_value = None entries = string.split(self.CM.Env["stonith-params"], ',') for entry in entries: (p_name, p_value) = string.split(entry, '=') if p_name == "hostlist" and p_value == "all": p_value = string.join(self.CM.Env["nodes"], " ") stonith_resource = self.stonith_resource_template % (self.CM.Env["stonith-type"], p_name, p_value) resources += stonith_resource #master slave resource resources += self.master_slave_resource % (num_nodes, 1, 1, 1) # generate cib self.cts_cib = self.cib_template % (cib_options, resources, constraints) return self.cts_cib class CIB10(CibBase): feature_set = "3.0" version = "pacemaker-1.0" cib_template = ''' ''' def _create(self, command): fixed = "HOME=/root CIB_file="+self.cib_tmpfile+" crm --force configure " + command rc = self.CM.rsh(self.target, fixed) if rc != 0: self.CM.log("Configure call failed: "+fixed) sys.exit(1) def _show(self, command=""): output = "" (rc, result) = self.CM.rsh(self.target, "HOME=/root CIB_file="+self.cib_tmpfile+" crm configure show "+command, None, ) for line in result: output += line self.CM.debug("Generated Config: "+line) return output def NewIP(self, name=None, standard="ocf:heartbeat"): ip = self.NextIP() if not name: name = "r"+ip if not standard: standard = "" else: standard += ":" self._create('''primitive %s %sIPaddr params ip=%s cidr_netmask=32 op monitor interval=5s''' % (name, standard, ip)) return name def install(self, target): old = self.cib_tmpfile # Force a rebuild self.cts_cib = None self.cib_tmpfile = CTSvars.CRM_CONFIG_DIR+"/cib.xml" self.contents(target) self.CM.rsh(self.target, "chown "+CTSvars.CRM_DAEMON_USER+" "+self.cib_tmpfile) self.cib_tmpfile = old def contents(self, target=None): # fencing resource if self.cts_cib: return self.cts_cib if not target: self.target = self.CM.Env["nodes"][0] else: self.target = target cib_base = self.cib_template % (self.feature_set, self.version, ''' remote-tls-port='9898' remote-clear-port='9999' ''') self.CM.rsh(self.target, '''echo "%s" > %s''' % (cib_base, self.cib_tmpfile)) #self.CM.rsh.cp(self.cib_tmpfile, "root@%s:%s" % (self.target, self.cib_tmpfile)) nodelist = "" self.num_nodes = 0 for node in self.CM.Env["nodes"]: nodelist += node + " " self.num_nodes = self.num_nodes + 1 no_quorum = "stop" if self.num_nodes < 3: no_quorum = "ignore" self.CM.debug("Cluster only has %d nodes, ignoring quorum" % self.num_nodes) # The shell no longer functions when the lrmd isn't running, how wonderful # Start one here and let the cluster clean it up when the full stack starts # Just hope target has the same location for lrmd - self.CM.rsh(self.target, CTSvars.CRM_DAEMON_DIR+"/lrmd", blocking=0) + self.CM.rsh(self.target, CTSvars.CRM_DAEMON_DIR+"/lrmd", synchronous=0) # Tell the shell to mind its own business, we know what we're doing self.CM.rsh(self.target, "crm options check-mode relaxed") # Now stop the shell from rejecting every update because we've not defined stonith resources yet self._create('''property stonith-enabled=false''') self._create('''property start-failure-is-fatal=false pe-input-series-max=5000''') self._create('''property shutdown-escalation=5min startup-fencing=false batch-limit=10 dc-deadtime=5s''') self._create('''property no-quorum-policy=%s expected-quorum-votes=%d''' % (no_quorum, self.num_nodes)) if self.CM.Env["DoBSC"] == 1: self._create('''property ident-string="Linux-HA TEST configuration file - REMOVEME!!"''') # Add resources? if self.CM.Env["CIBResource"] == 1: self.add_resources() # Fencing resource if self.CM.Env["DoFencing"]: params = None entries = string.split(self.CM.Env["stonith-params"], ',') for entry in entries: (name, value) = string.split(entry, '=') if name == "hostlist" and value == "all": value = string.join(self.CM.Env["nodes"], " ") if params: params = ("""%s '%s="%s"' """ % (params, name, value)) else: params = ("""'%s="%s"' """ % (name, value)) if params: params = "params %s" % params else: params = "" self._create('''primitive FencingChild stonith::%s %s op monitor interval=120s timeout=300 op start interval=0 timeout=180s op stop interval=0 timeout=180s''' % (self.CM.Env["stonith-type"], params)) # Set a threshold for unreliable stonith devices such as the vmware one self._create('''clone Fencing FencingChild meta globally-unique=false migration-threshold=5''') if self.CM.cluster_monitor == 1: self._create('''primitive cluster_mon ocf:pacemaker:ClusterMon params update=10 extra_options="-r -n" user=abeekhof htmlfile=/suse/abeekhof/Export/cluster.html op start interval=0 requires=nothing op monitor interval=5s requires=nothing''') self._create('''location prefer-dc cluster_mon rule -INFINITY: \#is_dc eq false''') self._create('''property stonith-enabled=%s''' % (self.CM.Env["DoFencing"])) # generate cib self.cts_cib = self._show("xml") if self.cib_tmpfile != CTSvars.CRM_CONFIG_DIR+"/cib.xml": self.CM.rsh(self.target, "rm -f "+self.cib_tmpfile) return self.cts_cib def add_resources(self): # Group Resource r1 = self.NewIP() ip = self.NextIP() r2 = "r"+ip self._create('''primitive %s heartbeat::IPaddr params 1=%s/32 op monitor interval=5s''' % (r2, ip)) r3 = self.NewIP() self._create('''group group-1 %s %s %s''' % (r1, r2, r3)) # Per-node resources for node in self.CM.Env["nodes"]: r = self.NewIP("rsc_"+node) self._create('''location prefer-%s %s rule 100: \#uname eq %s''' % (node, r, node)) # LSB resource lsb_agent=CTSvars.CTS_home+'''/LSBDummy''' self.CM.log("Installing LSB agent %s on %s" % (lsb_agent, repr(self.CM.Env["nodes"]))) for node in self.CM.Env["nodes"]: self.CM.rsh(node, "mkdir -p %s" % CTSvars.CTS_home) self.CM.rsh.cp(lsb_agent, "root@%s:%s" % (node, lsb_agent)) self._create('''primitive lsb-dummy lsb::''' +lsb_agent+ ''' op monitor interval=5s''') self._create('''colocation lsb-with-group INFINITY: lsb-dummy group-1''') self._create('''order lsb-after-group mandatory: group-1 lsb-dummy symmetrical=true''') # Migrator self._create('''primitive migrator ocf:pacemaker:Dummy meta allow-migrate=1 op monitor interval=P10S''') # Ping the test master self._create('''primitive ping-1 ocf:pacemaker:ping params host_list=%s name=connected debug=true op monitor interval=120s''' % os.uname()[1]) self._create('''clone Connectivity ping-1 meta globally-unique=false''') #master slave resource self._create('''primitive stateful-1 ocf:pacemaker:Stateful op monitor interval=15s op monitor interval=16s role=Master''') self._create('''ms master-1 stateful-1 meta clone-max=%d clone-node-max=%d master-max=%d master-node-max=%d''' % (self.num_nodes, 1, 1, 1)) # Require conectivity to run the master self._create('''location %s-is-connected %s rule -INFINITY: connected lt %d or not_defined connected''' % ("m1", "master-1", 1)) # Group with the master self._create('''colocation group-with-master INFINITY: group-1 master-1:Master''') self._create('''order group-after-master mandatory: master-1:promote group-1:start symmetrical=true''') class HASI(CIB10): def add_resources(self): # DLM resource self._create('''primitive dlm ocf:pacemaker:controld op monitor interval=120s''') self._create('''clone dlm-clone dlm meta globally-unique=false interleave=true''') # O2CB resource self._create('''primitive o2cb ocf:ocfs2:o2cb op monitor interval=120s''') self._create('''clone o2cb-clone o2cb meta globally-unique=false interleave=true''') self._create('''colocation o2cb-with-dlm INFINITY: o2cb-clone dlm-clone''') self._create('''order start-o2cb-after-dlm mandatory: dlm-clone o2cb-clone''') class ConfigFactory: def __init__(self, CM): self.CM = CM self.register("pacemaker06", CIB06, CM) self.register("pacemaker10", CIB10, CM) self.register("hae", HASI, CM) def register(self, methodName, constructor, *args, **kargs): """register a constructor""" _args = [constructor] _args.extend(args) setattr(self, methodName, apply(ConfigFactoryItem,_args, kargs)) def unregister(self, methodName): """unregister a constructor""" delattr(self, methodName) def createConfig(self, name="pacemaker-1.0"): if name == "pacemaker-0.6": name = "pacemaker06"; elif name == "pacemaker-1.0": name = "pacemaker10"; elif name == "hasi": name = "hae"; if hasattr(self, name): return getattr(self, name)() else: self.CM.log("Configuration variant '%s' is unknown. Defaulting to latest config" % name) return self.pacemaker10() class ConfigFactoryItem: def __init__(self, function, *args, **kargs): assert callable(function), "function should be a callable obj" self._function = function self._args = args self._kargs = kargs def __call__(self, *args, **kargs): """call function""" _args = list(self._args) _args.extend(args) _kargs = self._kargs.copy() _kargs.update(kargs) return apply(self._function,_args,_kargs) #CibFactory = ConfigFactory() diff --git a/cts/CTS.py b/cts/CTS.py index 0ca78b94ad..de8a0320f1 100644 --- a/cts/CTS.py +++ b/cts/CTS.py @@ -1,1340 +1,1352 @@ '''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., 59 Temple Place - Suite 330, Boston, MA 02111-1307, 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"] = None self["CMclass"] = None self["logrestartcmd"] = "/etc/init.d/syslog-ng restart 2>&1 > /dev/null" 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" + 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 -l root" self.CpCommand = "qacp" 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, blocking=1, silent=False): + 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 blocking: + if not synchronous: proc = Popen(self._cmd([node, command]), stdout = PIPE, stderr = PIPE, close_fds = True, shell = True) 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 stdout == 1: - result = proc.stdout.readline() + #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: - result = proc.stdout.readlines() + self.log("No stdout stream") - proc.stdout.close() 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 = "/tmp/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 = 100 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] 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' - logfile.seek(0) + 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 = limit while count > 0: count -= 1 line = logfile.readline() if line: print line.strip() else: break print prefix + 'Last read: %d %d' % (logfile.tell(), limit - count) logfile.close() """ class SearchObj: def __init__(self, Env, filename, host=None): self.Env = Env self.host = host 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 -p CTSwatcher: -f %s -o %s" % (log_watcher_bin, self.filename, self.offset), - stdout=None, silent=True) + 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): '''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 = [] 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)) 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_line(self): if not len(self.line_cache): if not len(self.file_list): raise ValueError("No sources to read from") for f in self.file_list: lines = f.next() self.line_cache.extend(lines) if self.line_cache: line = self.line_cache[0] self.line_cache.remove(line) return line return None def look(self, timeout=None): '''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 now=time.time() done=now+timeout+1 if self.debug_level > 2: self.debug("starting single search: timeout=%d, now=%d, limit=%d" % (timeout, now, done)) while (timeout <= 0 or time.time() <= done): line = self.__get_line() if 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: time.sleep(1) #time.sleep(0.025) else: self.debug("End of file") return None - self.debug("Single search timed out: timeout=%d, now=%d, limit=%d" % (timeout, now, done)) + self.debug("Single search timed out: timeout=%d, start=%d, limit=%d, now=%d" % (timeout, now, done, time.time())) 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, "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]) def key_for_node(self, node): return node 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_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 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) watch.setwatch() 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"]) 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: #self.debug("Found match: "+ repr(watch_result)) self.cluster_stable(self["DeadTime"]) return 1 if self.StataCM(node) and self.cluster_stable(self["DeadTime"]): 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, blocking=0) + 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: 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"], blocking=0) + 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": 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), blocking=0) - self.rsh(node, self["FixCommCmd"] % self.key_for_node(target), blocking=0) + 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=[], 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.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 diff --git a/cts/CTSaudits.py b/cts/CTSaudits.py index 344ae6c893..754a031b27 100755 --- a/cts/CTSaudits.py +++ b/cts/CTSaudits.py @@ -1,782 +1,782 @@ '''CTS: Cluster Testing System: Audit module ''' __copyright__=''' Copyright (C) 2000, 2001,2005 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., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. import time, os, string, re import CTS class ClusterAudit: def __init__(self, cm): self.CM = cm def __call__(self): raise ValueError("Abstract Class member (__call__)") def is_applicable(self): '''Return TRUE if we are applicable in the current test configuration''' raise ValueError("Abstract Class member (is_applicable)") return 1 def log(self, args): self.CM.log("audit: %s" % args) def debug(self, args): self.CM.debug("audit: %s" % args) def name(self): raise ValueError("Abstract Class member (name)") AllAuditClasses = [ ] class LogAudit(ClusterAudit): def name(self): return "LogAudit" def __init__(self, cm): self.CM = cm def RestartClusterLogging(self, nodes=None): if not nodes: nodes = self.CM.Env["nodes"] self.CM.log("Restarting logging on: %s" % repr(nodes)) for node in nodes: cmd=self.CM.Env["logrestartcmd"] - if self.CM.rsh(node, cmd, blocking=0) != 0: + if self.CM.rsh(node, cmd, synchronous=0) != 0: self.CM.log ("ERROR: Cannot restart logging on %s [%s failed]" % (node, cmd)) def TestLogging(self): patterns = [] prefix = "Test message from" watch_syslog = None watch_remote = None for node in self.CM.Env["nodes"]: # Look for the node name in two places to make sure # that syslog is logging with the correct hostname patterns.append("%s.*%s %s" % (node, prefix, node)) watch_pref = self.CM.Env["LogWatcher"] if watch_pref == "any" or watch_pref == "syslog": self.CM.Env["LogWatcher"] = "syslog" if watch_pref == "any": self.CM.log("Testing for %s logs" % self.CM.Env["LogWatcher"]) watch_syslog = CTS.LogWatcher(self.CM.Env, self.CM.Env["LogFileName"], patterns, "LogAudit", 30) watch_syslog.setwatch() if watch_pref == "any" or watch_pref == "remote": self.CM.Env["LogWatcher"] = "remote" if watch_pref == "any": self.CM.log("Testing for %s logs" % self.CM.Env["LogWatcher"]) watch_remote = CTS.LogWatcher(self.CM.Env, self.CM.Env["LogFileName"], patterns, "LogAudit", 30) watch_remote.setwatch() for node in self.CM.Env["nodes"]: cmd="logger -p %s.info %s %s" % (self.CM.Env["SyslogFacility"], prefix, node) - if self.CM.rsh(node, cmd, blocking=0, silent=True) != 0: + if self.CM.rsh(node, cmd, synchronous=0, silent=True) != 0: self.CM.log ("ERROR: Cannot execute remote command [%s] on %s" % (cmd, node)) if watch_syslog: watch = watch_syslog self.CM.Env["LogWatcher"] = "syslog" watch_result = watch.lookforall(silent=True) if not watch.unmatched: if watch_pref == "any": self.CM.log ("Continuing with %s-based log reader" % (self.CM.Env["LogWatcher"])) return 1 if watch_remote: watch = watch_remote self.CM.Env["LogWatcher"] = "remote" watch_result = watch.lookforall(silent=True) if not watch.unmatched: if watch_pref == "any": self.CM.log ("Continuing with %s-based log reader" % (self.CM.Env["LogWatcher"])) return 1 if watch_syslog and not watch_syslog.unmatched: for regex in watch_syslog.unmatched: self.CM.log ("Test message [%s] not found in syslog logs." % regex) if watch_remote and not watch_remote.unmatched: for regex in watch_remote.unmatched: self.CM.log ("Test message [%s] not found in remote logs." % regex) return 0 def __call__(self): max=3 attempt=0 self.CM.ns.WaitForAllNodesToComeUp(self.CM.Env["nodes"]) while attempt <= max and self.TestLogging() == 0: attempt = attempt + 1 self.RestartClusterLogging() time.sleep(60*attempt) if attempt > max: self.CM.log("ERROR: Cluster logging unrecoverable.") return 0 return 1 def is_applicable(self): if self.CM.Env["DoBSC"]: return 0 return 1 class DiskAudit(ClusterAudit): def name(self): return "DiskspaceAudit" def __init__(self, cm): self.CM = cm def __call__(self): result=1 dfcmd="df -k /var/log | tail -1 | tr -s ' ' | cut -d' ' -f2" self.CM.ns.WaitForAllNodesToComeUp(self.CM.Env["nodes"]) for node in self.CM.Env["nodes"]: dfout=self.CM.rsh(node, dfcmd, 1) if not dfout: self.CM.log ("ERROR: Cannot execute remote df command [%s] on %s" % (dfcmd, node)) else: try: idfout = int(dfout) except (ValueError, TypeError): self.CM.log("Warning: df output from %s was invalid [%s]" % (node, dfout)) else: if idfout == 0: self.CM.log("CRIT: Completely out of log disk space on %s" % node) result=None elif idfout <= 1000: self.CM.log("WARN: Low on log disk space (%d Mbytes) on %s" % (idfout, node)) return result def is_applicable(self): if self.CM.Env["DoBSC"]: return 0 return 1 class AuditResource: def __init__(self, cm, line): fields = line.split() self.CM = cm self.line = line self.type = fields[1] self.id = fields[2] self.clone_id = fields[3] self.parent = fields[4] self.rprovider = fields[5] self.rclass = fields[6] self.rtype = fields[7] self.host = fields[8] self.needs_quorum = fields[9] self.flags = int(fields[10]) self.flags_s = fields[11] if self.parent == "NA": self.parent = None def unique(self): if self.flags & int("0x00000020", 16): return 1 return 0 def orphan(self): if self.flags & int("0x00000001", 16): return 1 return 0 def managed(self): if self.flags & int("0x00000002", 16): return 1 return 0 class AuditConstraint: def __init__(self, cm, line): fields = line.split() self.CM = cm self.line = line self.type = fields[1] self.id = fields[2] self.rsc = fields[3] self.target = fields[4] self.score = fields[5] self.rsc_role = fields[6] self.target_role = fields[7] if self.rsc_role == "NA": self.rsc_role = None if self.target_role == "NA": self.target_role = None class PrimitiveAudit(ClusterAudit): def name(self): return "PrimitiveAudit" def __init__(self, cm): self.CM = cm def doResourceAudit(self, resource, quorum): rc=1 active = self.CM.ResourceLocation(resource.id) if len(active) == 1: if quorum: self.debug("Resource %s active on %s" % (resource.id, repr(active))) elif resource.needs_quorum == 1: self.CM.log("Resource %s active without quorum: %s" % (resource.id, repr(active))) rc=0 elif not resource.managed(): self.CM.log("Resource %s not managed. Active on %s" % (resource.id, repr(active))) elif not resource.unique(): # TODO: Figure out a clever way to actually audit these resource types if len(active) > 1: self.debug("Non-unique resource %s is active on: %s" % (resource.id, repr(active))) else: self.debug("Non-unique resource %s is not active" % resource.id) elif len(active) > 1: self.CM.log("Resource %s is active multiple times: %s" % (resource.id, repr(active))) rc=0 elif resource.orphan(): self.debug("Resource %s is an inactive orphan" % resource.id) elif len(self.inactive_nodes) == 0: self.CM.log("WARN: Resource %s not served anywhere" % resource.id) rc=0 elif self.CM.Env["warn-inactive"] == 1: if quorum or not resource.needs_quorum: self.CM.log("WARN: Resource %s not served anywhere (Inactive nodes: %s)" % (resource.id, repr(self.inactive_nodes))) else: self.debug("Resource %s not served anywhere (Inactive nodes: %s)" % (resource.id, repr(self.inactive_nodes))) elif quorum or not resource.needs_quorum: self.debug("Resource %s not served anywhere (Inactive nodes: %s)" % (resource.id, repr(self.inactive_nodes))) return rc def setup(self): self.target = None self.resources = [] self.constraints = [] self.active_nodes = [] self.inactive_nodes = [] for node in self.CM.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.active_nodes.append(node) else: self.inactive_nodes.append(node) for node in self.CM.Env["nodes"]: if self.target == None and self.CM.ShouldBeStatus[node] == "up": self.target = node if not self.target: # TODO: In Pacemaker 1.0 clusters we'll be able to run crm_resource # with CIB_file=/path/to/cib.xml even when the cluster isn't running self.debug("No nodes active - skipping %s" % self.name()) return 0 (rc, lines) = self.CM.rsh(self.target, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): self.resources.append(AuditResource(self.CM, line)) elif re.search("^Constraint", line): self.constraints.append(AuditConstraint(self.CM, line)) else: self.CM.log("Unknown entry: %s" % line); return 1 def __call__(self): rc = 1 if not self.setup(): return 1 quorum = self.CM.HasQuorum(None) for resource in self.resources: if resource.type == "primitive": if self.doResourceAudit(resource, quorum) == 0: rc = 0 return rc def is_applicable(self): if self.CM["Name"] == "crm-lha": return 1 if self.CM["Name"] == "crm-ais": return 1 return 0 class GroupAudit(PrimitiveAudit): def name(self): return "GroupAudit" def __call__(self): rc = 1 if not self.setup(): return 1 for group in self.resources: if group.type == "group": first_match = 1 group_location = None for child in self.resources: if child.parent == group.id: nodes = self.CM.ResourceLocation(child.id) if first_match and len(nodes) > 0: group_location = nodes[0] first_match = 0 if len(nodes) > 1: rc = 0 self.CM.log("Child %s of %s is active more than once: %s" % (child.id, group.id, repr(nodes))) elif len(nodes) == 0: # Groups are allowed to be partially active # However we do need to make sure later children aren't running group_location = None self.debug("Child %s of %s is stopped" % (child.id, group.id)) elif nodes[0] != group_location: rc = 0 self.CM.log("Child %s of %s is active on the wrong node (%s) expected %s" % (child.id, group.id, nodes[0], group_location)) else: self.debug("Child %s of %s is active on %s" % (child.id, group.id, nodes[0])) return rc class CloneAudit(PrimitiveAudit): def name(self): return "CloneAudit" def __call__(self): rc = 1 if not self.setup(): return 1 for clone in self.resources: if clone.type == "clone": for child in self.resources: if child.parent == clone.id and child.type == "primitive": self.debug("Checking child %s of %s..." % (child.id, clone.id)) # Check max and node_max # Obtain with: # crm_resource -g clone_max --meta -r child.id # crm_resource -g clone_node_max --meta -r child.id return rc class ColocationAudit(PrimitiveAudit): def name(self): return "ColocationAudit" def crm_location(self, resource): (rc, lines) = self.CM.rsh(self.target, "crm_resource -W -r %s -Q"%resource, None) hosts = [] if rc == 0: for line in lines: fields = line.split() hosts.append(fields[0]) return hosts def __call__(self): rc = 1 if not self.setup(): return 1 for coloc in self.constraints: if coloc.type == "rsc_colocation": source = self.crm_location(coloc.rsc) target = self.crm_location(coloc.target) if len(source) == 0: self.debug("Colocation audit (%s): %s not running" % (coloc.id, coloc.rsc)) else: for node in source: if not node in target: rc = 0 self.CM.log("Colocation audit (%s): %s running on %s (not in %s)" % (coloc.id, coloc.rsc, node, repr(target))) else: self.debug("Colocation audit (%s): %s running on %s (in %s)" % (coloc.id, coloc.rsc, node, repr(target))) return rc class CrmdStateAudit(ClusterAudit): def __init__(self, cm): self.CM = cm self.Stats = {"calls":0 , "success":0 , "failure":0 , "skipped":0 , "auditfail":0} def has_key(self, key): return self.Stats.has_key(key) def __setitem__(self, key, value): self.Stats[key] = value def __getitem__(self, key): return self.Stats[key] def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not self.Stats.has_key(name): self.Stats[name]=0 self.Stats[name] = self.Stats[name]+1 def __call__(self): passed = 1 up_are_down = 0 down_are_up = 0 unstable_list = [] for node in self.CM.Env["nodes"]: should_be = self.CM.ShouldBeStatus[node] rc = self.CM.test_node_CM(node) if rc > 0: if should_be == "down": down_are_up = down_are_up + 1 if rc == 1: unstable_list.append(node) elif should_be == "up": up_are_down = up_are_down + 1 if len(unstable_list) > 0: passed = 0 self.CM.log("Cluster is not stable: %d (of %d): %s" %(len(unstable_list), self.CM.upcount(), repr(unstable_list))) if up_are_down > 0: passed = 0 self.CM.log("%d (of %d) nodes expected to be up were down." %(up_are_down, len(self.CM.Env["nodes"]))) if down_are_up > 0: passed = 0 self.CM.log("%d (of %d) nodes expected to be down were up." %(down_are_up, len(self.CM.Env["nodes"]))) return passed def name(self): return "CrmdStateAudit" def is_applicable(self): if self.CM["Name"] == "crm-lha": return 1 if self.CM["Name"] == "crm-ais": return 1 return 0 class CIBAudit(ClusterAudit): def __init__(self, cm): self.CM = cm self.Stats = {"calls":0 , "success":0 , "failure":0 , "skipped":0 , "auditfail":0} def has_key(self, key): return self.Stats.has_key(key) def __setitem__(self, key, value): self.Stats[key] = value def __getitem__(self, key): return self.Stats[key] def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not self.Stats.has_key(name): self.Stats[name]=0 self.Stats[name] = self.Stats[name]+1 def __call__(self): passed = 1 ccm_partitions = self.CM.find_partitions() if len(ccm_partitions) == 0: self.debug("\tNo partitions to audit") return 1 for partition in ccm_partitions: self.debug("\tAuditing CIB consistency for: %s" %partition) partition_passed = 0 if self.audit_cib_contents(partition) == 0: passed = 0 return passed def audit_cib_contents(self, hostlist): passed = 1 node0 = None node0_xml = None partition_hosts = hostlist.split() for node in partition_hosts: node_xml = self.store_remote_cib(node, node0) if node_xml == None: self.CM.log("Could not perform audit: No configuration from %s" % node) passed = 0 elif node0 == None: node0 = node node0_xml = node_xml elif node0_xml == None: self.CM.log("Could not perform audit: No configuration from %s" % node0) passed = 0 else: (rc, result) = self.CM.rsh( node0, "crm_diff -VV -cf --new %s --original %s" % (node_xml, node0_xml), None) if rc != 0: self.CM.log("Diff between %s and %s failed: %d" % (node0_xml, node_xml, rc)) passed = 0 for line in result: if not re.search("", line): passed = 0 self.debug("CibDiff[%s-%s]: %s" % (node0, node, line)) else: self.debug("CibDiff[%s-%s] Ignoring: %s" % (node0, node, line)) # self.CM.rsh(node0, "rm -f %s" % node_xml) # self.CM.rsh(node0, "rm -f %s" % node0_xml) return passed def store_remote_cib(self, node, target): combined = "" filename="/tmp/ctsaudit.%s.xml" % node if not target: target = node (rc, lines) = self.CM.rsh(node, self.CM["CibQuery"], None) if rc != 0: self.CM.log("Could not retrieve configuration") return None self.CM.rsh("localhost", "rm -f %s" % filename) for line in lines: self.CM.rsh("localhost", "echo \'%s\' >> %s" % (line[:-1], filename), silent=True) if self.CM.rsh.cp(filename, "root@%s:%s" % (target, filename), silent=True) != 0: self.CM.log("Could not store configuration") return None return filename def name(self): return "CibAudit" def is_applicable(self): if self.CM["Name"] == "crm-lha": return 1 if self.CM["Name"] == "crm-ais": return 1 return 0 class PartitionAudit(ClusterAudit): def __init__(self, cm): self.CM = cm self.Stats = {"calls":0 , "success":0 , "failure":0 , "skipped":0 , "auditfail":0} self.NodeEpoche={} self.NodeState={} self.NodeQuorum={} def has_key(self, key): return self.Stats.has_key(key) def __setitem__(self, key, value): self.Stats[key] = value def __getitem__(self, key): return self.Stats[key] def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not self.Stats.has_key(name): self.Stats[name]=0 self.Stats[name] = self.Stats[name]+1 def __call__(self): passed = 1 ccm_partitions = self.CM.find_partitions() if ccm_partitions == None or len(ccm_partitions) == 0: return 1 self.CM.cluster_stable(double_check=True) if len(ccm_partitions) != self.CM.partitions_expected: self.CM.log("ERROR: %d cluster partitions detected:" %len(ccm_partitions)) passed = 0 for partition in ccm_partitions: self.CM.log("\t %s" %partition) for partition in ccm_partitions: partition_passed = 0 if self.audit_partition(partition) == 0: passed = 0 return passed def trim_string(self, avalue): if not avalue: return None if len(avalue) > 1: return avalue[:-1] def trim2int(self, avalue): if not avalue: return None if len(avalue) > 1: return int(avalue[:-1]) def audit_partition(self, partition): passed = 1 dc_found = [] dc_allowed_list = [] lowest_epoche = None node_list = partition.split() self.debug("Auditing partition: %s" %(partition)) for node in node_list: if self.CM.ShouldBeStatus[node] != "up": self.CM.log("Warn: Node %s appeared out of nowhere" %(node)) self.CM.ShouldBeStatus[node] = "up" # not in itself a reason to fail the audit (not what we're # checking for in this audit) self.NodeState[node] = self.CM.rsh(node, self.CM["StatusCmd"]%node, 1) self.NodeEpoche[node] = self.CM.rsh(node, self.CM["EpocheCmd"], 1) self.NodeQuorum[node] = self.CM.rsh(node, self.CM["QuorumCmd"], 1) self.debug("Node %s: %s - %s - %s." %(node, self.NodeState[node], self.NodeEpoche[node], self.NodeQuorum[node])) self.NodeState[node] = self.trim_string(self.NodeState[node]) self.NodeEpoche[node] = self.trim2int(self.NodeEpoche[node]) self.NodeQuorum[node] = self.trim_string(self.NodeQuorum[node]) if not self.NodeEpoche[node]: self.CM.log("Warn: Node %s dissappeared: cant determin epoche" %(node)) self.CM.ShouldBeStatus[node] = "down" # not in itself a reason to fail the audit (not what we're # checking for in this audit) elif lowest_epoche == None or self.NodeEpoche[node] < lowest_epoche: lowest_epoche = self.NodeEpoche[node] if not lowest_epoche: self.CM.log("Lowest epoche not determined in %s" % (partition)) passed = 0 for node in node_list: if self.CM.ShouldBeStatus[node] == "up": if self.CM.is_node_dc(node, self.NodeState[node]): dc_found.append(node) if self.NodeEpoche[node] == lowest_epoche: self.debug("%s: OK" % node) elif not self.NodeEpoche[node]: self.debug("Check on %s ignored: no node epoche" % node) elif not lowest_epoche: self.debug("Check on %s ignored: no lowest epoche" % node) else: self.CM.log("DC %s is not the oldest node (%d vs. %d)" %(node, self.NodeEpoche[node], lowest_epoche)) passed = 0 if len(dc_found) == 0: self.CM.log("DC not found on any of the %d allowed nodes: %s (of %s)" %(len(dc_allowed_list), str(dc_allowed_list), str(node_list))) elif len(dc_found) > 1: self.CM.log("%d DCs (%s) found in cluster partition: %s" %(len(dc_found), str(dc_found), str(node_list))) passed = 0 if passed == 0: for node in node_list: if self.CM.ShouldBeStatus[node] == "up": self.CM.log("epoche %s : %s" %(self.NodeEpoche[node], self.NodeState[node])) return passed def name(self): return "PartitionAudit" def is_applicable(self): if self.CM["Name"] == "crm-lha": return 1 if self.CM["Name"] == "crm-ais": return 1 return 0 AllAuditClasses.append(DiskAudit) AllAuditClasses.append(LogAudit) AllAuditClasses.append(CrmdStateAudit) AllAuditClasses.append(PartitionAudit) AllAuditClasses.append(PrimitiveAudit) AllAuditClasses.append(GroupAudit) AllAuditClasses.append(CloneAudit) AllAuditClasses.append(ColocationAudit) AllAuditClasses.append(CIBAudit) def AuditList(cm): result = [] for auditclass in AllAuditClasses: a = auditclass(cm) if a.is_applicable(): result.append(a) return result