diff --git a/cts/CTS.py b/cts/CTS.py index ed42634451..2418991837 100644 --- a/cts/CTS.py +++ b/cts/CTS.py @@ -1,1791 +1,1795 @@ '''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 * from threading import Thread trace_rsh=None trace_lw=None has_log_stats = {} log_stats_bin = CTSvars.CRM_DAEMON_DIR + "/cts_log_stats.sh" log_stats = """ #!/bin/bash # Tool for generating system load reports while CTS runs trap "" 1 f=$1; shift action=$1; shift base=`basename $0` if [ ! -e $f ]; then echo "Time, Load 1, Load 5, Load 15, Test Marker" > $f fi function killpid() { if [ -e $f.pid ]; then kill -9 `cat $f.pid` rm -f $f.pid fi } function status() { if [ -e $f.pid ]; then kill -0 `cat $f.pid` return $? else return 1 fi } function start() { # Is it already running? if status then return fi echo Active as $$ echo $$ > $f.pid while [ 1 = 1 ]; do uptime | sed s/up.*:/,/ | tr '\\n' ',' >> $f #top -b -c -n1 | grep -e usr/libexec/pacemaker | grep -v -e grep -e python | head -n 1 | sed s@/usr/libexec/pacemaker/@@ | awk '{print " 0, "$9", "$10", "$12}' | tr '\\n' ',' >> $f echo 0 >> $f sleep 5 done } case $action in start) start ;; start-bg|bg) # Use c --ssh -- ./stats.sh file start-bg nohup $0 $f start >/dev/null 2>&1 > $f echo " $*" >> $f start ;; *) echo "Unknown action: $action." ;; esac """ 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)) self.StatsMark(0) 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() self.StatsExtract() return 1 #ClusterManager.oprofileSave(Iterations) Scenario.TearDown() self.StatsExtract() 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"]) def StatsExtract(self): if not self["stats"]: return for host in self["nodes"]: log_stats_file = "%s/cts-stats.csv" % CTSvars.CRM_DAEMON_DIR if has_log_stats.has_key(host): self.rsh(host, '''bash %s %s stop''' % (log_stats_bin, log_stats_file)) (rc, lines) = self.rsh(host, '''cat %s''' % log_stats_file, stdout=2) self.rsh(host, '''bash %s %s delete''' % (log_stats_bin, log_stats_file)) fname = "cts-stats-%d-nodes-%s.csv" % (len(self["nodes"]), host) print "Extracted stats: %s" % fname fd = open(fname, "a") fd.writelines(lines) fd.close() def StatsMark(self, testnum): '''Mark the test number in the stats log''' global has_log_stats if not self["stats"]: return for host in self["nodes"]: log_stats_file = "%s/cts-stats.csv" % CTSvars.CRM_DAEMON_DIR if not has_log_stats.has_key(host): global log_stats global log_stats_bin script=log_stats #script = re.sub("\\\\", "\\\\", script) script = re.sub('\"', '\\\"', script) script = re.sub("'", "\'", script) script = re.sub("`", "\`", script) script = re.sub("\$", "\\\$", script) self.debug("Installing %s on %s" % (log_stats_bin, host)) self.rsh(host, '''echo "%s" > %s''' % (script, log_stats_bin), silent=True) self.rsh(host, '''bash %s %s delete''' % (log_stats_bin, log_stats_file)) has_log_stats[host] = 1 # Now mark it self.rsh(host, '''bash %s %s mark %s''' % (log_stats_bin, log_stats_file, testnum), synchronous=0) 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 AsyncWaitProc(Thread): def __init__(self, proc, node, command, Env): self.Env = Env self.proc = proc self.node = node self.command = command Thread.__init__(self) def log(self, args): if not self.Env: print (args) else: self.Env.log(args) def debug(self, args): if not self.Env: print (args) else: self.Env.debug(args) def run(self): self.debug("cmd: async: target=%s, pid=%d: %s" % (self.node, self.proc.pid, self.command)) self.proc.wait() self.debug("cmd: pid %d returned %d" % (self.proc.pid, self.proc.returncode)) if self.proc.stderr: lines = self.proc.stderr.readlines() self.proc.stderr.close() for line in lines: self.debug("cmd: stderr[%d]: %s" % (self.proc.pid, line)) if self.proc.stdout: lines = self.proc.stdout.readlines() self.proc.stdout.close() for line in lines: self.debug("cmd: stdout[%d]: %s" % (self.proc.pid, line)) 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.async = [] self.silent = silent if trace_rsh: self.silent = False # -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. ''' if trace_rsh: silent = False rc = 0 result = None proc = Popen(self._cmd([node, command]), stdout = PIPE, stderr = PIPE, close_fds = True, shell = True) if not synchronous and proc.pid > 0 and not self.silent: aproc = AsyncWaitProc(proc, node, command, self.Env) aproc.start() return 0 #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: if stdout == 3: result.append("error: "+err) else: 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 trace_rsh: silent = False 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.offset = "EOF" if host == None: host = "localhost" 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): self.log("Not implemented") class FileObj(SearchObj): def __init__(self, Env, filename, host=None, name=None): global has_log_watcher SearchObj.__init__(self, Env, filename, host, name) 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 next(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 JournalObj(SearchObj): def __init__(self, Env, host=None, name=None): SearchObj.__init__(self, Env, name, host, name) self.next() def next(self): cache = [] command = "journalctl -q --after-cursor='%s' --show-cursor" % (self.offset) if self.offset == "EOF": command = "journalctl -q -n 0 --show-cursor" (rc, lines) = self.Env.rsh(self.host, command, stdout=None, silent=True, blocking=False) for line in lines: match = re.search("^-- cursor: ([^.]+)", line) if match: last_offset = self.offset self.offset = match.group(1) - if last_offset == "EOF": self.debug("Got %d lines, new cursor: %s" % (len(lines), self.offset)) + self.debug("Got %d lines, new cursor: %s" % (len(lines), self.offset)) 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, hosts=None, kind=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) if kind: self.kind = kind else: self.kind = self.Env["LogWatcher"] + if log: + self.filename = log + else: + self.filename = self.Env["LogFileName"] + 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 hosts: self.hosts = hosts else: self.hosts = self.Env["nodes"] if trace_lw: self.debug_level = 3 silent = False 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.kind == "remote": for node in self.hosts: self.file_list.append(FileObj(self.Env, self.filename, node, self.name)) elif self.kind == "journal": for node in self.hosts: self.file_list.append(JournalObj(self.Env, node, self.name)) else: self.file_list.append(FileObj(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 if trace_lw: silent = False lines=0 needlines=True 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 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 > 3: 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(): if self.debug_level > 1: self.debug("lines during timeout") time.sleep(1) self.__get_lines() elif needlines: # Grab any relevant messages that might have arrived since # the last time the buffer was populated if self.debug_level > 1: self.debug("lines without timeout") self.__get_lines() # Don't come back here again needlines = False 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 trace_lw: silent = False 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 upnode=None 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 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) elif self.Env["Stack"] == "corosync (cman)": # There is a delay between gaining quorum and CMAN starting fencing # This can mean that even nodes that are fully up get fenced # There is no use fighting it, just look for everyone so that CTS doesn't get confused stonithPats.append(self["Pat:Fencing_ok"] % peer) stonithPats.append(self["Pat:Fencing_start"] % peer) if peer != node and not upnode and self.ShouldBeStatus[peer] == "up": upnode = peer # Look for STONITH ops, depending on Env["at-boot"] we might need to change the nodes status if not upnode: return None stonith = LogWatcher(self.Env, self["LogFileName"], stonithPats, "StartupFencing", 0, hosts=[upnode]) stonith.setwatch() return stonith def fencing_cleanup(self, node, stonith): peer_list = [] peer_state = {} self.debug("Looking for nodes that were fenced as a result of %s starting" % node) # If we just started a node, we may now have quorum (and permission to fence) if not stonith: self.debug("Nothing to do") return peer_list q = self.HasQuorum(None) if not q and len(self.Env["nodes"]) > 2: # We didn't gain quorum - we shouldn't have shot anyone self.debug("Quorum: %d Len: %d" % (q, len(self.Env["nodes"]))) return peer_list # Now see if any states need to be updated self.debug("looking for: " + repr(stonith.regexes)) shot = stonith.look(0) while shot: line = repr(shot) self.debug("Found: "+ line) del stonith.regexes[stonith.whichmatch] # Extract node name for n in self.Env["nodes"]: if re.search(self["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"] % n, shot): peer = n peer_state[peer] = "in-progress" self.__instance_errorstoignore.append(self["Pat:Fencing_start"] % peer) if not peer: self.log("ERROR: Unknown stonith match: %s" % line) elif not peer in peer_list: self.debug("Found peer: "+ peer) peer_list.append(peer) # Get the next one shot = stonith.look(60) for peer in peer_list: self.debug(" Peer %s was fenced as a result of %s starting: %s" % (peer, node, peer_state[peer])) if self.Env["at-boot"]: self.ShouldBeStatus[peer] = "up" else: self.ShouldBeStatus[peer] = "down" if peer_state[peer] == "in-progress": # Wait for any in-progress operations to complete shot = stonith.look(60) while len(stonith.regexes) and shot: line = repr(shot) self.debug("Found: "+ line) del stonith.regexes[stonith.whichmatch] shot = stonith.look(60) # Now make sure the node is alive too self.ns.WaitForNodeToComeUp(peer, self["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 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)) self.fencing_cleanup(node, stonith) 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") self.install_config(node) 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, force=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" and force == False: 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 ("ERROR: Could not stop %s on node %s" %(self["Name"], node)) return None def StopaCMnoBlock(self, node): '''Stop the cluster manager on a given node with none-block mode''' self.debug("Stopping %s on node %s" %(self["Name"], node)) self.rsh(node, self["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"] % node, 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, quick=False): '''Start the cluster manager on every node in the cluster. We can do it on a subset of the cluster if nodelist is not None. ''' map = {} if not nodelist: nodelist=self.Env["nodes"] for node in nodelist: if self.ShouldBeStatus[node] == "down": self.ns.WaitForAllNodesToComeUp(nodelist, 300) if not quick: if not self.StartaCM(node, verbose=verbose): return 0 return 1 # Approximation of SimulStartList for --boot watchpats = [ ] watchpats.append(self["Pat:DC_IDLE"]) for node in nodelist: watchpats.append(self["Pat:Local_started"] % node) watchpats.append(self["Pat:InfraUp"] % node) watchpats.append(self["Pat:PacemakerUp"] % node) # Start all the nodes - at about the same time... watch = LogWatcher(self.Env, self["LogFileName"], watchpats, "fast-start", self["DeadTime"]+10) watch.setwatch() if not self.StartaCM(nodelist[0], verbose=verbose): return 0 for node in nodelist: self.StartaCMnoBlock(node, verbose=verbose) watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.log ("Warn: Startup pattern not found: %s" %(regex)) if not self.cluster_stable(): self.log("Cluster did not stabilize") return 0 return 1 def stopall(self, nodelist=None, verbose=False, force=False): '''Stop the cluster managers on every node in the cluster. We can do it on a subset of the cluster if nodelist is not None. ''' ret = 1 map = {} if not nodelist: nodelist=self.Env["nodes"] for node in self.Env["nodes"]: if self.ShouldBeStatus[node] == "up" or force == True: if not self.StopaCM(node, verbose=verbose, force=force): ret = 0 return ret def rereadall(self, nodelist=None): '''Force the cluster managers on every node in the cluster to reread their config files. We can do it on a subset of the cluster if nodelist is not None. ''' map = {} if not nodelist: nodelist=self.Env["nodes"] for node in self.Env["nodes"]: if self.ShouldBeStatus[node] == "up": self.RereadCM(node) def statall(self, nodelist=None): '''Return the status of the cluster managers in the cluster. We can do it on a subset of the cluster if nodelist is not None. ''' result={} if not nodelist: nodelist=self.Env["nodes"] for node in nodelist: if self.StataCM(node): result[node] = "up" else: result[node] = "down" return result def isolate_node(self, target, nodes=None): '''isolate the communication between the nodes''' if not nodes: nodes = self.Env["nodes"] for node in nodes: if node != target: rc = self.rsh(target, self["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 diff --git a/cts/CTStests.py b/cts/CTStests.py index df060d7066..35cc69626d 100644 --- a/cts/CTStests.py +++ b/cts/CTStests.py @@ -1,2864 +1,2869 @@ '''CTS: Cluster Testing System: Tests module There are a few things we want to do here: ''' __copyright__=''' Copyright (C) 2000, 2001 Alan Robertson Licensed under the GNU GPL. Add RecourceRecover testcase Zhao Kai ''' # # 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. # # SPECIAL NOTE: # # Tests may NOT implement any cluster-manager-specific code in them. # EXTEND the ClusterManager object to provide the base capabilities # the test needs if you need to do something that the current CM classes # do not. Otherwise you screw up the whole point of the object structure # in CTS. # # Thank you. # import time, os, re, types, string, tempfile, sys from stat import * from cts import CTS from cts.CTSaudits import * from cts.CTSvars import * +from cts.patterns import PatternSelector AllTestClasses = [ ] class CTSTest: ''' A Cluster test. We implement the basic set of properties and behaviors for a generic cluster test. Cluster tests track their own statistics. We keep each of the kinds of counts we track as separate {name,value} pairs. ''' def __init__(self, cm): #self.name="the unnamed test" self.Stats = {"calls":0 , "success":0 , "failure":0 , "skipped":0 , "auditfail":0} # if not issubclass(cm.__class__, ClusterManager): # raise ValueError("Must be a ClusterManager object") self.CM = cm - self.rsh = cm.rsh self.Env = cm.Env - self.log = cm.log - self.debug = cm.debug - self.log_patterns = cm + self.rsh = cm.Env.rsh + self.templates = PatternSelector(cm["Name"]) self.Audits = [] self.timeout=120 self.passed = 1 self.is_loop = 0 self.is_unsafe = 0 self.is_experimental = 0 self.is_container = 0 self.is_valgrind = 0 self.benchmark = 0 # which tests to benchmark self.timer = {} # timers + def log(self, args): + self.Env.log(args) + + def debug(self, args): + self.Env.debug(args + 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 log_mark(self, msg): self.debug("MARK: test %s %s %d" % (self.name,msg,time.time())) return def get_timer(self,key = "test"): try: return self.timer[key] except: return 0 def set_timer(self,key = "test"): self.timer[key] = time.time() return self.timer[key] def log_timer(self,key = "test"): elapsed = 0 if key in self.timer: elapsed = time.time() - self.timer[key] s = key == "test" and self.name or "%s:%s" %(self.name,key) self.debug("%s runtime: %.2f" % (s, elapsed)) del self.timer[key] return elapsed def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not self.Stats.has_key(name): self.Stats[name]=0 self.Stats[name] = self.Stats[name]+1 # Reset the test passed boolean if name == "calls": self.passed = 1 def failure(self, reason="none"): '''Increment the failure count''' self.passed = 0 self.incr("failure") self.log(("Test %s" % self.name).ljust(35) +" FAILED: %s" % reason) return None def success(self): '''Increment the success count''' self.incr("success") return 1 def skipped(self): '''Increment the skipped count''' self.incr("skipped") return 1 def __call__(self, node): '''Perform the given test''' raise ValueError("Abstract Class member (__call__)") self.incr("calls") return self.failure() def audit(self): passed = 1 if len(self.Audits) > 0: for audit in self.Audits: if not audit(): self.log("Internal %s Audit %s FAILED." % (self.name, audit.name())) self.incr("auditfail") passed = 0 return passed def setup(self, node): '''Setup the given test''' return self.success() def teardown(self, node): '''Tear down the given test''' return self.success() def create_watch(self, patterns, timeout, name=None): if not name: name = self.name - return CTS.LogWatcher(self.Env, self.log_patterns["LogFileName"], patterns, name, timeout) + return CTS.LogWatcher(self.Env, None, patterns, name, timeout) def local_badnews(self, prefix, watch, local_ignore=[]): errcount = 0 if not prefix: prefix = "LocalBadNews:" ignorelist = [] ignorelist.append(" CTS: ") ignorelist.append(prefix) ignorelist.extend(local_ignore) while errcount < 100: match=watch.look(0) if match: add_err = 1 for ignore in ignorelist: if add_err == 1 and re.search(ignore, match): add_err = 0 if add_err == 1: self.log(prefix + " " + match) errcount=errcount+1 else: break else: self.log("Too many errors!") return errcount def is_applicable(self): return self.is_applicable_common() def is_applicable_common(self): '''Return TRUE if we are applicable in the current test configuration''' #raise ValueError("Abstract Class member (is_applicable)") if self.is_loop and not self.Env["loop-tests"]: return 0 elif self.is_unsafe and not self.Env["unsafe-tests"]: return 0 elif self.is_valgrind and not self.Env["valgrind-tests"]: return 0 elif self.is_experimental and not self.Env["experimental-tests"]: return 0 elif self.is_container and not self.Env["container-tests"]: return 0 elif self.Env["benchmark"] and self.benchmark == 0: return 0 return 1 def find_ocfs2_resources(self, node): self.r_o2cb = None self.r_ocfs2 = [] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rtype == "o2cb" and r.parent != "NA": self.debug("Found o2cb: %s" % self.r_o2cb) self.r_o2cb = r.parent if re.search("^Constraint", line): c = AuditConstraint(self.CM, line) if c.type == "rsc_colocation" and c.target == self.r_o2cb: self.r_ocfs2.append(c.rsc) self.debug("Found ocfs2 filesystems: %s" % repr(self.r_ocfs2)) return len(self.r_ocfs2) def canrunnow(self, node): '''Return TRUE if we can meaningfully run right now''' return 1 def errorstoignore(self): '''Return list of errors which are 'normal' and should be ignored''' return [] ################################################################### class StopTest(CTSTest): ################################################################### '''Stop (deactivate) the cluster manager on a node''' def __init__(self, cm): CTSTest.__init__(self, cm) self.name="Stop" def __call__(self, node): '''Perform the 'stop' test. ''' self.incr("calls") if self.CM.ShouldBeStatus[node] != "up": return self.skipped() patterns = [] # Technically we should always be able to notice ourselves stopping - patterns.append(self.log_patterns["Pat:We_stopped"] % node) + patterns.append(self.templates["Pat:We_stopped"] % node) #if self.Env["use_logd"]: - # patterns.append(self.log_patterns["Pat:Logd_stopped"] % node) + # patterns.append(self.templates["Pat:Logd_stopped"] % node) # Any active node needs to notice this one left # NOTE: This wont work if we have multiple partitions for other in self.Env["nodes"]: if self.CM.ShouldBeStatus[other] == "up" and other != node: - patterns.append(self.log_patterns["Pat:They_stopped"] %(other, self.CM.key_for_node(node))) + patterns.append(self.templates["Pat:They_stopped"] %(other, self.CM.key_for_node(node))) #self.debug("Checking %s will notice %s left"%(other, node)) - watch = self.create_watch(patterns, self.log_patterns["DeadTime"]) + watch = self.create_watch(patterns, self.templates["DeadTime"]) watch.setwatch() if node == self.CM.OurNode: self.incr("us") else: if self.CM.upcount() <= 1: self.incr("all") else: self.incr("them") self.CM.StopaCM(node) watch_result = watch.lookforall() failreason=None UnmatchedList = "||" if watch.unmatched: (rc, output) = self.rsh(node, "/bin/ps axf", None) for line in output: self.debug(line) (rc, output) = self.rsh(node, "/usr/sbin/dlm_tool dump", None) for line in output: self.debug(line) for regex in watch.unmatched: self.log ("ERROR: Shutdown pattern not found: %s" % (regex)) UnmatchedList += regex + "||"; failreason="Missing shutdown pattern" - self.CM.cluster_stable(self.log_patterns["DeadTime"]) + self.CM.cluster_stable(self.templates["DeadTime"]) if not watch.unmatched or self.CM.upcount() == 0: return self.success() if len(watch.unmatched) >= self.CM.upcount(): return self.failure("no match against (%s)" % UnmatchedList) if failreason == None: return self.success() else: return self.failure(failreason) # # We don't register StopTest because it's better when called by # another test... # ################################################################### class StartTest(CTSTest): ################################################################### '''Start (activate) the cluster manager on a node''' def __init__(self, cm, debug=None): CTSTest.__init__(self,cm) self.name="start" self.debug = debug def __call__(self, node): '''Perform the 'start' test. ''' self.incr("calls") if self.CM.upcount() == 0: self.incr("us") else: self.incr("them") if self.CM.ShouldBeStatus[node] != "down": return self.skipped() elif self.CM.StartaCM(node): return self.success() else: return self.failure("Startup %s on node %s failed" - %(self.log_patterns["Name"], node)) + %(self.templates["Name"], node)) # # We don't register StartTest because it's better when called by # another test... # ################################################################### class FlipTest(CTSTest): ################################################################### '''If it's running, stop it. If it's stopped start it. Overthrow the status quo... ''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="Flip" self.start = StartTest(cm) self.stop = StopTest(cm) def __call__(self, node): '''Perform the 'Flip' test. ''' self.incr("calls") if self.CM.ShouldBeStatus[node] == "up": self.incr("stopped") ret = self.stop(node) type="up->down" # Give the cluster time to recognize it's gone... - time.sleep(self.log_patterns["StableTime"]) + time.sleep(self.templates["StableTime"]) elif self.CM.ShouldBeStatus[node] == "down": self.incr("started") ret = self.start(node) type="down->up" else: return self.skipped() self.incr(type) if ret: return self.success() else: return self.failure("%s failure" % type) # Register FlipTest as a good test to run AllTestClasses.append(FlipTest) ################################################################### class RestartTest(CTSTest): ################################################################### '''Stop and restart a node''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="Restart" self.start = StartTest(cm) self.stop = StopTest(cm) self.benchmark = 1 def __call__(self, node): '''Perform the 'restart' test. ''' self.incr("calls") self.incr("node:" + node) ret1 = 1 if self.CM.StataCM(node): self.incr("WasStopped") if not self.start(node): return self.failure("start (setup) failure: "+node) self.set_timer() if not self.stop(node): return self.failure("stop failure: "+node) if not self.start(node): return self.failure("start failure: "+node) return self.success() # Register RestartTest as a good test to run AllTestClasses.append(RestartTest) ################################################################### class StonithdTest(CTSTest): ################################################################### def __init__(self, cm): CTSTest.__init__(self, cm) self.name="Stonithd" self.startall = SimulStartLite(cm) self.benchmark = 1 def __call__(self, node): self.incr("calls") if len(self.Env["nodes"]) < 2: return self.skipped() ret = self.startall(None) if not ret: return self.failure("Setup failed") is_dc = self.CM.is_node_dc(node) watchpats = [] watchpats.append("log_operation: Operation .* for host '%s' with device .* returned: 0" % node) watchpats.append("tengine_stonith_notify: Peer %s was terminated .*: OK" % node) if self.Env["at-boot"] == 0: self.debug("Expecting %s to stay down" % node) self.CM.ShouldBeStatus[node]="down" else: self.debug("Expecting %s to come up again %d" % (node, self.Env["at-boot"])) watchpats.append("%s .*do_state_transition: .* S_STARTING -> S_PENDING" % node) watchpats.append("%s .*do_state_transition: .* S_PENDING -> S_NOT_DC" % node) - watch = self.create_watch(watchpats, 30 + self.log_patterns["DeadTime"] + self.log_patterns["StableTime"] + self.log_patterns["StartTime"]) + watch = self.create_watch(watchpats, 30 + self.templates["DeadTime"] + self.templates["StableTime"] + self.templates["StartTime"]) watch.setwatch() origin = self.Env.RandomGen.choice(self.Env["nodes"]) rc = self.rsh(origin, "stonith_admin --reboot %s -VVVVVV" % node) if rc == 194: # 194 - 256 = -62 = Timer expired # # Look for the patterns, usually this means the required # device was running on the node to be fenced - or that # the required devices were in the process of being loaded # and/or moved # # Effectively the node committed suicide so there will be # no confirmation, but pacemaker should be watching and # fence the node again self.log("Fencing command on %s to fence %s timed out" % (origin, node)) elif origin != node and rc != 0: self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.debug("Waiting STONITHd node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.Env["nodes"], 600) self.log("Fencing command on %s failed to fence %s (rc=%d)" % (origin, node, rc)) elif origin == node and rc != 255: # 255 == broken pipe, ie. the node was fenced as epxected self.log("Logcally originated fencing returned %d" % rc) self.set_timer("fence") matched = watch.lookforall() self.log_timer("fence") self.set_timer("reform") if watch.unmatched: self.log("Patterns not found: " + repr(watch.unmatched)) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.debug("Waiting STONITHd node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.Env["nodes"], 600) self.debug("Waiting for the cluster to re-stabilize with all nodes") - is_stable = self.CM.cluster_stable(self.log_patterns["StartTime"]) + is_stable = self.CM.cluster_stable(self.templates["StartTime"]) if not matched: return self.failure("Didn't find all expected patterns") elif not is_stable: return self.failure("Cluster did not become stable") self.log_timer("reform") return self.success() def errorstoignore(self): return [ - self.log_patterns["Pat:Fencing_start"] % ".*", - self.log_patterns["Pat:Fencing_ok"] % ".*", + self.templates["Pat:Fencing_start"] % ".*", + self.templates["Pat:Fencing_ok"] % ".*", "error: native_create_actions: Resource .*stonith::.* is active on 2 nodes attempting recovery", "error: remote_op_done: Operation reboot of .*by .* for stonith_admin.*: Timer expired", ] def is_applicable(self): if not self.is_applicable_common(): return 0 if self.Env.has_key("DoFencing"): return self.Env["DoFencing"] return 1 AllTestClasses.append(StonithdTest) ################################################################### class StartOnebyOne(CTSTest): ################################################################### '''Start all the nodes ~ one by one''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="StartOnebyOne" self.stopall = SimulStopLite(cm) self.start = StartTest(cm) self.ns=CTS.NodeStatus(cm.Env) def __call__(self, dummy): '''Perform the 'StartOnebyOne' test. ''' self.incr("calls") # We ignore the "node" parameter... # Shut down all the nodes... ret = self.stopall(None) if not ret: return self.failure("Test setup failed") failed=[] self.set_timer() for node in self.Env["nodes"]: if not self.start(node): failed.append(node) if len(failed) > 0: return self.failure("Some node failed to start: " + repr(failed)) return self.success() # Register StartOnebyOne as a good test to run AllTestClasses.append(StartOnebyOne) ################################################################### class SimulStart(CTSTest): ################################################################### '''Start all the nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="SimulStart" self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) def __call__(self, dummy): '''Perform the 'SimulStart' test. ''' self.incr("calls") # We ignore the "node" parameter... # Shut down all the nodes... ret = self.stopall(None) if not ret: return self.failure("Setup failed") self.CM.clear_all_caches() if not self.startall(None): return self.failure("Startall failed") return self.success() # Register SimulStart as a good test to run AllTestClasses.append(SimulStart) ################################################################### class SimulStop(CTSTest): ################################################################### '''Stop all the nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="SimulStop" self.startall = SimulStartLite(cm) self.stopall = SimulStopLite(cm) def __call__(self, dummy): '''Perform the 'SimulStop' test. ''' self.incr("calls") # We ignore the "node" parameter... # Start up all the nodes... ret = self.startall(None) if not ret: return self.failure("Setup failed") if not self.stopall(None): return self.failure("Stopall failed") return self.success() # Register SimulStop as a good test to run AllTestClasses.append(SimulStop) ################################################################### class StopOnebyOne(CTSTest): ################################################################### '''Stop all the nodes in order''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="StopOnebyOne" self.startall = SimulStartLite(cm) self.stop = StopTest(cm) def __call__(self, dummy): '''Perform the 'StopOnebyOne' test. ''' self.incr("calls") # We ignore the "node" parameter... # Start up all the nodes... ret = self.startall(None) if not ret: return self.failure("Setup failed") failed=[] self.set_timer() for node in self.Env["nodes"]: if not self.stop(node): failed.append(node) if len(failed) > 0: return self.failure("Some node failed to stop: " + repr(failed)) self.CM.clear_all_caches() return self.success() # Register StopOnebyOne as a good test to run AllTestClasses.append(StopOnebyOne) ################################################################### class RestartOnebyOne(CTSTest): ################################################################### '''Restart all the nodes in order''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="RestartOnebyOne" self.startall = SimulStartLite(cm) def __call__(self, dummy): '''Perform the 'RestartOnebyOne' test. ''' self.incr("calls") # We ignore the "node" parameter... # Start up all the nodes... ret = self.startall(None) if not ret: return self.failure("Setup failed") did_fail=[] self.set_timer() self.restart = RestartTest(self.CM) for node in self.Env["nodes"]: if not self.restart(node): did_fail.append(node) if did_fail: return self.failure("Could not restart %d nodes: %s" %(len(did_fail), repr(did_fail))) return self.success() # Register StopOnebyOne as a good test to run AllTestClasses.append(RestartOnebyOne) ################################################################### class PartialStart(CTSTest): ################################################################### '''Start a node - but tell it to stop before it finishes starting up''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="PartialStart" self.startall = SimulStartLite(cm) self.stopall = SimulStopLite(cm) self.stop = StopTest(cm) #self.is_unsafe = 1 def __call__(self, node): '''Perform the 'PartialStart' test. ''' self.incr("calls") ret = self.stopall(None) if not ret: return self.failure("Setup failed") # FIXME! This should use the CM class to get the pattern # then it would be applicable in general watchpats = [] watchpats.append("crmd.*Connecting to cluster infrastructure") - watch = self.create_watch(watchpats, self.log_patterns["DeadTime"]+10) + watch = self.create_watch(watchpats, self.templates["DeadTime"]+10) watch.setwatch() self.CM.StartaCMnoBlock(node) ret = watch.lookforall() if not ret: self.log("Patterns not found: " + repr(watch.unmatched)) return self.failure("Setup of %s failed" % node) ret = self.stop(node) if not ret: return self.failure("%s did not stop in time" % node) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' # We might do some fencing in the 2-node case if we make it up far enough return [ """Executing reboot fencing operation""" ] # Register StopOnebyOne as a good test to run AllTestClasses.append(PartialStart) ####################################################################### class StandbyTest(CTSTest): ####################################################################### def __init__(self, cm): CTSTest.__init__(self,cm) self.name="Standby" self.benchmark = 1 self.start = StartTest(cm) self.startall = SimulStartLite(cm) # make sure the node is active # set the node to standby mode # check resources, none resource should be running on the node # set the node to active mode # check resouces, resources should have been migrated back (SHOULD THEY?) def __call__(self, node): self.incr("calls") ret=self.startall(None) if not ret: return self.failure("Start all nodes failed") self.debug("Make sure node %s is active" % node) if self.CM.StandbyStatus(node) != "off": if not self.CM.SetStandbyMode(node, "off"): return self.failure("can't set node %s to active mode" % node) self.CM.cluster_stable() status = self.CM.StandbyStatus(node) if status != "off": return self.failure("standby status of %s is [%s] but we expect [off]" % (node, status)) self.debug("Getting resources running on node %s" % node) rsc_on_node = self.CM.active_resources(node) watchpats = [] watchpats.append("do_state_transition:.*-> S_POLICY_ENGINE") - watch = self.create_watch(watchpats, self.log_patterns["DeadTime"]+10) + watch = self.create_watch(watchpats, self.templates["DeadTime"]+10) watch.setwatch() self.debug("Setting node %s to standby mode" % node) if not self.CM.SetStandbyMode(node, "on"): return self.failure("can't set node %s to standby mode" % node) self.set_timer("on") ret = watch.lookforall() if not ret: self.log("Patterns not found: " + repr(watch.unmatched)) self.CM.SetStandbyMode(node, "off") return self.failure("cluster didn't react to standby change on %s" % node) self.CM.cluster_stable() status = self.CM.StandbyStatus(node) if status != "on": return self.failure("standby status of %s is [%s] but we expect [on]" % (node, status)) self.log_timer("on") self.debug("Checking resources") bad_run = self.CM.active_resources(node) if len(bad_run) > 0: rc = self.failure("%s set to standby, %s is still running on it" % (node, repr(bad_run))) self.debug("Setting node %s to active mode" % node) self.CM.SetStandbyMode(node, "off") return rc self.debug("Setting node %s to active mode" % node) if not self.CM.SetStandbyMode(node, "off"): return self.failure("can't set node %s to active mode" % node) self.set_timer("off") self.CM.cluster_stable() status = self.CM.StandbyStatus(node) if status != "off": return self.failure("standby status of %s is [%s] but we expect [off]" % (node, status)) self.log_timer("off") return self.success() AllTestClasses.append(StandbyTest) ####################################################################### class ValgrindTest(CTSTest): ####################################################################### '''Check for memory leaks''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="Valgrind" self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) self.is_valgrind = 1 self.is_loop = 1 def setup(self, node): self.incr("calls") ret=self.stopall(None) if not ret: return self.failure("Stop all nodes failed") # Enable valgrind self.logPat = "/tmp/%s-*.valgrind" % self.name self.Env["valgrind-prefix"] = self.name self.rsh(node, "rm -f %s" % self.logPat, None) ret=self.startall(None) if not ret: return self.failure("Start all nodes failed") for node in self.Env["nodes"]: (rc, output) = self.rsh(node, "ps u --ppid `pidofproc aisexec`", None) for line in output: self.debug(line) return self.success() def teardown(self, node): # Disable valgrind self.Env["valgrind-prefix"] = None # Return all nodes to normal ret=self.stopall(None) if not ret: return self.failure("Stop all nodes failed") return self.success() def find_leaks(self): # Check for leaks leaked = [] self.stop = StopTest(self.CM) for node in self.Env["nodes"]: (rc, ps_out) = self.rsh(node, "ps u --ppid `pidofproc aisexec`", None) rc = self.stop(node) if not rc: self.failure("Couldn't shut down %s" % node) rc = self.rsh(node, "grep -e indirectly.*lost:.*[1-9] -e definitely.*lost:.*[1-9] -e (ERROR|error).*SUMMARY:.*[1-9].*errors %s" % self.logPat, 0) if rc != 1: leaked.append(node) self.failure("Valgrind errors detected on %s" % node) for line in ps_out: self.log(line) (rc, output) = self.rsh(node, "grep -e lost: -e SUMMARY: %s" % self.logPat, None) for line in output: self.log(line) (rc, output) = self.rsh(node, "cat %s" % self.logPat, None) for line in output: self.debug(line) self.rsh(node, "rm -f %s" % self.logPat, None) return leaked def __call__(self, node): leaked = self.find_leaks() if len(leaked) > 0: return self.failure("Nodes %s leaked" % repr(leaked)) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ """cib:.*readCibXmlFile:""", """HA_VALGRIND_ENABLED""" ] ####################################################################### class StandbyLoopTest(ValgrindTest): ####################################################################### '''Check for memory leaks by putting a node in and out of standby for an hour''' def __init__(self, cm): ValgrindTest.__init__(self,cm) self.name="StandbyLoop" def __call__(self, node): lpc = 0 delay = 2 failed = 0 done=time.time() + self.Env["loop-minutes"]*60 while time.time() <= done and not failed: lpc = lpc + 1 time.sleep(delay) if not self.CM.SetStandbyMode(node, "on"): self.failure("can't set node %s to standby mode" % node) failed = lpc time.sleep(delay) if not self.CM.SetStandbyMode(node, "off"): self.failure("can't set node %s to active mode" % node) failed = lpc leaked = self.find_leaks() if failed: return self.failure("Iteration %d failed" % failed) elif len(leaked) > 0: return self.failure("Nodes %s leaked" % repr(leaked)) return self.success() AllTestClasses.append(StandbyLoopTest) ############################################################################## class BandwidthTest(CTSTest): ############################################################################## # Tests should not be cluster-manager-specific # If you need to find out cluster manager configuration to do this, then # it should be added to the generic cluster manager API. '''Test the bandwidth which heartbeat uses''' def __init__(self, cm): CTSTest.__init__(self, cm) self.name = "Bandwidth" self.start = StartTest(cm) self.__setitem__("min",0) self.__setitem__("max",0) self.__setitem__("totalbandwidth",0) self.tempfile = tempfile.mktemp(".cts") self.startall = SimulStartLite(cm) def __call__(self, node): '''Perform the Bandwidth test''' self.incr("calls") if self.CM.upcount()<1: return self.skipped() Path = self.CM.InternalCommConfig() if "ip" not in Path["mediatype"]: return self.skipped() port = Path["port"][0] port = int(port) ret = self.startall(None) if not ret: return self.failure("Test setup failed") time.sleep(5) # We get extra messages right after startup. fstmpfile = "/var/run/band_estimate" dumpcmd = "tcpdump -p -n -c 102 -i any udp port %d > %s 2>&1" \ % (port, fstmpfile) rc = self.rsh(node, dumpcmd) if rc == 0: farfile = "root@%s:%s" % (node, fstmpfile) self.rsh.cp(farfile, self.tempfile) Bandwidth = self.countbandwidth(self.tempfile) if not Bandwidth: self.log("Could not compute bandwidth.") return self.success() intband = int(Bandwidth + 0.5) self.log("...bandwidth: %d bits/sec" % intband) self.Stats["totalbandwidth"] = self.Stats["totalbandwidth"] + Bandwidth if self.Stats["min"] == 0: self.Stats["min"] = Bandwidth if Bandwidth > self.Stats["max"]: self.Stats["max"] = Bandwidth if Bandwidth < self.Stats["min"]: self.Stats["min"] = Bandwidth self.rsh(node, "rm -f %s" % fstmpfile) os.unlink(self.tempfile) return self.success() else: return self.failure("no response from tcpdump command [%d]!" % rc) def countbandwidth(self, file): fp = open(file, "r") fp.seek(0) count = 0 sum = 0 while 1: line = fp.readline() if not line: return None if re.search("udp",line) or re.search("UDP,", line): count=count+1 linesplit = string.split(line," ") for j in range(len(linesplit)-1): if linesplit[j]=="udp": break if linesplit[j]=="length:": break try: sum = sum + int(linesplit[j+1]) except ValueError: self.log("Invalid tcpdump line: %s" % line) return None T1 = linesplit[0] timesplit = string.split(T1,":") time2split = string.split(timesplit[2],".") time1 = (long(timesplit[0])*60+long(timesplit[1]))*60+long(time2split[0])+long(time2split[1])*0.000001 break while count < 100: line = fp.readline() if not line: return None if re.search("udp",line) or re.search("UDP,", line): count = count+1 linessplit = string.split(line," ") for j in range(len(linessplit)-1): if linessplit[j] =="udp": break if linesplit[j]=="length:": break try: sum=int(linessplit[j+1])+sum except ValueError: self.log("Invalid tcpdump line: %s" % line) return None T2 = linessplit[0] timesplit = string.split(T2,":") time2split = string.split(timesplit[2],".") time2 = (long(timesplit[0])*60+long(timesplit[1]))*60+long(time2split[0])+long(time2split[1])*0.000001 time = time2-time1 if (time <= 0): return 0 return (sum*8)/time def is_applicable(self): '''BandwidthTest never applicable''' return 0 AllTestClasses.append(BandwidthTest) ################################################################### class MaintenanceMode(CTSTest): ################################################################### def __init__(self, cm): CTSTest.__init__(self,cm) self.name="MaintenanceMode" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.max=30 #self.is_unsafe = 1 self.benchmark = 1 self.action = "asyncmon" self.interval = 0 self.rid="maintenanceDummy" def toggleMaintenanceMode(self, node, action): pats = [] - pats.append(self.log_patterns["Pat:DC_IDLE"]) + pats.append(self.templates["Pat:DC_IDLE"]) # fail the resource right after turning Maintenance mode on # verify it is not recovered until maintenance mode is turned off if action == "On": pats.append("Updating failcount for %s on .* after .* %s" % (self.rid, self.action)) else: - pats.append(self.log_patterns["Pat:RscOpOK"] % (self.rid, "stop_0")) - pats.append(self.log_patterns["Pat:RscOpOK"] % (self.rid, "start_0")) + pats.append(self.templates["Pat:RscOpOK"] % (self.rid, "stop_0")) + pats.append(self.templates["Pat:RscOpOK"] % (self.rid, "start_0")) watch = self.create_watch(pats, 60) watch.setwatch() self.debug("Turning maintenance mode %s" % action) - self.rsh(node, self.log_patterns["MaintenanceMode%s" % (action)]) + self.rsh(node, self.templates["MaintenanceMode%s" % (action)]) if (action == "On"): self.rsh(node, "crm_resource -V -F -r %s -H %s &>/dev/null" % (self.rid, node)) self.set_timer("recover%s" % (action)) watch.lookforall() self.log_timer("recover%s" % (action)) if watch.unmatched: self.debug("Failed to find patterns when turning maintenance mode %s" % action) return repr(watch.unmatched) return "" def insertMaintenanceDummy(self, node): pats = [] - pats.append(("%s.*" % node) + (self.log_patterns["Pat:RscOpOK"] % (self.rid, "start_0"))) + pats.append(("%s.*" % node) + (self.templates["Pat:RscOpOK"] % (self.rid, "start_0"))) watch = self.create_watch(pats, 60) watch.setwatch() self.CM.AddDummyRsc(node, self.rid) self.set_timer("addDummy") watch.lookforall() self.log_timer("addDummy") if watch.unmatched: self.debug("Failed to find patterns when adding maintenance dummy resource") return repr(watch.unmatched) return "" def removeMaintenanceDummy(self, node): pats = [] - pats.append(self.log_patterns["Pat:RscOpOK"] % (self.rid, "stop_0")) + pats.append(self.templates["Pat:RscOpOK"] % (self.rid, "stop_0")) watch = self.create_watch(pats, 60) watch.setwatch() self.CM.RemoveDummyRsc(node, self.rid) self.set_timer("removeDummy") watch.lookforall() self.log_timer("removeDummy") if watch.unmatched: self.debug("Failed to find patterns when removing maintenance dummy resource") return repr(watch.unmatched) return "" def managedRscList(self, node): rscList = [] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): tmp = AuditResource(self.CM, line) if tmp.managed(): rscList.append(tmp.id) return rscList def verifyResources(self, node, rscList, managed): managedList = list(rscList) managed_str = "managed" if not managed: managed_str = "unmanaged" (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): tmp = AuditResource(self.CM, line) if managed and not tmp.managed(): continue elif not managed and tmp.managed(): continue elif managedList.count(tmp.id): managedList.remove(tmp.id) if len(managedList) == 0: self.debug("Found all %s resources on %s" % (managed_str, node)) return True self.log("Could not find all %s resources on %s. %s" % (managed_str, node, managedList)) return False def __call__(self, node): '''Perform the 'MaintenanceMode' test. ''' self.incr("calls") verify_managed = False verify_unmanaged = False failPat = "" ret = self.startall(None) if not ret: return self.failure("Setup failed") # get a list of all the managed resources. We use this list # after enabling maintenance mode to verify all managed resources # become un-managed. After maintenance mode is turned off, we use # this list to verify all the resources become managed again. managedResources = self.managedRscList(node) if len(managedResources) == 0: self.log("No managed resources on %s" % node) return self.skipped() # insert a fake resource we can fail during maintenance mode # so we can verify recovery does not take place until after maintenance # mode is disabled. failPat = failPat + self.insertMaintenanceDummy(node) # toggle maintenance mode ON, then fail dummy resource. failPat = failPat + self.toggleMaintenanceMode(node, "On") # verify all the resources are now unmanaged if self.verifyResources(node, managedResources, False): verify_unmanaged = True # Toggle maintenance mode OFF, verify dummy is recovered. failPat = failPat + self.toggleMaintenanceMode(node, "Off") # verify all the resources are now managed again if self.verifyResources(node, managedResources, True): verify_managed = True # Remove our maintenance dummy resource. failPat = failPat + self.removeMaintenanceDummy(node) self.CM.cluster_stable() if failPat != "": return self.failure("Unmatched patterns: %s" % (failPat)) elif verify_unmanaged is False: return self.failure("Failed to verify resources became unmanaged during maintenance mode") elif verify_managed is False: return self.failure("Failed to verify resources switched back to managed after disabling maintenance mode") return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ """Updating failcount for %s""" % self.rid, """LogActions: Recover %s""" % self.rid, """Unknown operation: fail""", """(ERROR|error): sending stonithRA op to stonithd failed.""", - self.log_patterns["Pat:RscOpOK"] % (self.rid, ("%s_%d" % (self.action, self.interval))), + self.templates["Pat:RscOpOK"] % (self.rid, ("%s_%d" % (self.action, self.interval))), """(ERROR|error): process_graph_event: Action %s_%s_%d .* initiated outside of a transition""" % (self.rid, self.action, self.interval), ] AllTestClasses.append(MaintenanceMode) ################################################################### class ResourceRecover(CTSTest): ################################################################### def __init__(self, cm): CTSTest.__init__(self,cm) self.name="ResourceRecover" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.max=30 self.rid=None self.rid_alt=None #self.is_unsafe = 1 self.benchmark = 1 # these are the values used for the new LRM API call self.action = "asyncmon" self.interval = 0 def __call__(self, node): '''Perform the 'ResourceRecover' test. ''' self.incr("calls") ret = self.startall(None) if not ret: return self.failure("Setup failed") resourcelist = self.CM.active_resources(node) # if there are no resourcelist, return directly if len(resourcelist)==0: self.log("No active resources on %s" % node) return self.skipped() self.rid = self.Env.RandomGen.choice(resourcelist) self.rid_alt = self.rid rsc = None (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): tmp = AuditResource(self.CM, line) if tmp.id == self.rid: rsc = tmp # Handle anonymous clones that get renamed self.rid = rsc.clone_id break if not rsc: return self.failure("Could not find %s in the resource list" % self.rid) self.debug("Shooting %s aka. %s" % (rsc.clone_id, rsc.id)) pats = [] pats.append("Updating failcount for %s on .* after .* %s" % (self.rid, self.action)) if rsc.managed(): - pats.append(self.log_patterns["Pat:RscOpOK"] % (self.rid, "stop_0")) + pats.append(self.templates["Pat:RscOpOK"] % (self.rid, "stop_0")) if rsc.unique(): - pats.append(self.log_patterns["Pat:RscOpOK"] % (self.rid, "start_0")) + pats.append(self.templates["Pat:RscOpOK"] % (self.rid, "start_0")) else: # Anonymous clones may get restarted with a different clone number - pats.append(self.log_patterns["Pat:RscOpOK"] % (".*", "start_0")) + pats.append(self.templates["Pat:RscOpOK"] % (".*", "start_0")) watch = self.create_watch(pats, 60) watch.setwatch() self.rsh(node, "crm_resource -V -F -r %s -H %s &>/dev/null" % (self.rid, node)) self.set_timer("recover") watch.lookforall() self.log_timer("recover") self.CM.cluster_stable() recovered=self.CM.ResourceLocation(self.rid) if watch.unmatched: return self.failure("Patterns not found: %s" % repr(watch.unmatched)) elif rsc.unique() and len(recovered) > 1: return self.failure("%s is now active on more than one node: %s"%(self.rid, repr(recovered))) elif len(recovered) > 0: self.debug("%s is running on: %s" %(self.rid, repr(recovered))) elif rsc.managed(): return self.failure("%s was not recovered and is inactive" % self.rid) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ """Updating failcount for %s""" % self.rid, """LogActions: Recover %s""" % self.rid, """LogActions: Recover %s""" % self.rid_alt, """Unknown operation: fail""", """(ERROR|error): sending stonithRA op to stonithd failed.""", - self.log_patterns["Pat:RscOpOK"] % (self.rid, ("%s_%d" % (self.action, self.interval))), + self.templates["Pat:RscOpOK"] % (self.rid, ("%s_%d" % (self.action, self.interval))), """(ERROR|error): process_graph_event: Action %s_%s_%d .* initiated outside of a transition""" % (self.rid, self.action, self.interval), ] AllTestClasses.append(ResourceRecover) ################################################################### class ComponentFail(CTSTest): ################################################################### def __init__(self, cm): CTSTest.__init__(self,cm) self.name="ComponentFail" self.startall = SimulStartLite(cm) self.complist = cm.Components() self.patterns = [] self.okerrpatterns = [] self.is_unsafe = 1 def __call__(self, node): '''Perform the 'ComponentFail' test. ''' self.incr("calls") self.patterns = [] self.okerrpatterns = [] # start all nodes ret = self.startall(None) if not ret: return self.failure("Setup failed") - if not self.CM.cluster_stable(self.log_patterns["StableTime"]): + if not self.CM.cluster_stable(self.templates["StableTime"]): return self.failure("Setup failed - unstable") node_is_dc = self.CM.is_node_dc(node, None) # select a component to kill chosen = self.Env.RandomGen.choice(self.complist) while chosen.dc_only == 1 and node_is_dc == 0: chosen = self.Env.RandomGen.choice(self.complist) self.debug("...component %s (dc=%d,boot=%d)" % (chosen.name, node_is_dc,chosen.triggersreboot)) self.incr(chosen.name) if chosen.name != "aisexec" and chosen.name != "corosync": - if self.log_patterns["Name"] != "crm-lha" or chosen.name != "pengine": - self.patterns.append(self.log_patterns["Pat:ChildKilled"] %(node, chosen.name)) - self.patterns.append(self.log_patterns["Pat:ChildRespawn"] %(node, chosen.name)) + if self.templates["Name"] != "crm-lha" or chosen.name != "pengine": + self.patterns.append(self.templates["Pat:ChildKilled"] %(node, chosen.name)) + self.patterns.append(self.templates["Pat:ChildRespawn"] %(node, chosen.name)) self.patterns.extend(chosen.pats) if node_is_dc: self.patterns.extend(chosen.dc_pats) # In an ideal world, this next stuff should be in the "chosen" object as a member function - if self.log_patterns["Name"] == "crm-lha" and chosen.triggersreboot: + if self.templates["Name"] == "crm-lha" and chosen.triggersreboot: # Make sure the node goes down and then comes back up if it should reboot... for other in self.Env["nodes"]: if other != node: - self.patterns.append(self.log_patterns["Pat:They_stopped"] %(other, self.CM.key_for_node(node))) - self.patterns.append(self.log_patterns["Pat:Slave_started"] % node) - self.patterns.append(self.log_patterns["Pat:Local_started"] % node) + self.patterns.append(self.templates["Pat:They_stopped"] %(other, self.CM.key_for_node(node))) + self.patterns.append(self.templates["Pat:Slave_started"] % node) + self.patterns.append(self.templates["Pat:Local_started"] % node) if chosen.dc_only: # Sometimes these will be in the log, and sometimes they won't... self.okerrpatterns.append("%s .*Process %s:.* exited" %(node, chosen.name)) self.okerrpatterns.append("%s .*I_ERROR.*crmdManagedChildDied" %node) self.okerrpatterns.append("%s .*The %s subsystem terminated unexpectedly" %(node, chosen.name)) self.okerrpatterns.append("(ERROR|error): Client .* exited with return code") else: # Sometimes this won't be in the log... - self.okerrpatterns.append(self.log_patterns["Pat:ChildKilled"] %(node, chosen.name)) - self.okerrpatterns.append(self.log_patterns["Pat:ChildRespawn"] %(node, chosen.name)) - self.okerrpatterns.append(self.log_patterns["Pat:ChildExit"]) + self.okerrpatterns.append(self.templates["Pat:ChildKilled"] %(node, chosen.name)) + self.okerrpatterns.append(self.templates["Pat:ChildRespawn"] %(node, chosen.name)) + self.okerrpatterns.append(self.templates["Pat:ChildExit"]) # supply a copy so self.patterns doesnt end up empty tmpPats = [] tmpPats.extend(self.patterns) self.patterns.extend(chosen.badnews_ignore) # Look for STONITH ops, depending on Env["at-boot"] we might need to change the nodes status stonithPats = [] - stonithPats.append(self.log_patterns["Pat:Fencing_ok"] % node) + stonithPats.append(self.templates["Pat:Fencing_ok"] % node) stonith = self.create_watch(stonithPats, 0) stonith.setwatch() # set the watch for stable watch = self.create_watch( - tmpPats, self.log_patterns["DeadTime"] + self.log_patterns["StableTime"] + self.log_patterns["StartTime"]) + tmpPats, self.templates["DeadTime"] + self.templates["StableTime"] + self.templates["StartTime"]) watch.setwatch() # kill the component chosen.kill(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.debug("Waiting for any STONITHd node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.Env["nodes"], 600) self.debug("Waiting for the cluster to re-stabilize with all nodes") - self.CM.cluster_stable(self.log_patterns["StartTime"]) + self.CM.cluster_stable(self.templates["StartTime"]) self.debug("Checking if %s was shot" % node) shot = stonith.look(60) if shot: self.debug("Found: "+ repr(shot)) - self.okerrpatterns.append(self.log_patterns["Pat:Fencing_start"] % node) + self.okerrpatterns.append(self.templates["Pat:Fencing_start"] % node) if self.Env["at-boot"] == 0: self.CM.ShouldBeStatus[node]="down" # If fencing occurred, chances are many (if not all) the expected logs # will not be sent - or will be lost when the node reboots return self.success() # check for logs indicating a graceful recovery matched = watch.lookforall(allow_multiple_matches=1) if watch.unmatched: self.log("Patterns not found: " + repr(watch.unmatched)) self.debug("Waiting for the cluster to re-stabilize with all nodes") - is_stable = self.CM.cluster_stable(self.log_patterns["StartTime"]) + is_stable = self.CM.cluster_stable(self.templates["StartTime"]) if not matched: return self.failure("Didn't find all expected %s patterns" % chosen.name) elif not is_stable: return self.failure("Cluster did not become stable after killing %s" % chosen.name) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' # Note that okerrpatterns refers to the last time we ran this test # The good news is that this works fine for us... self.okerrpatterns.extend(self.patterns) return self.okerrpatterns AllTestClasses.append(ComponentFail) #################################################################### class SplitBrainTest(CTSTest): #################################################################### '''It is used to test split-brain. when the path between the two nodes break check the two nodes both take over the resource''' def __init__(self,cm): CTSTest.__init__(self,cm) self.name = "SplitBrain" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.is_experimental = 1 def isolate_partition(self, partition): other_nodes = [] other_nodes.extend(self.Env["nodes"]) for node in partition: try: other_nodes.remove(node) except ValueError: self.log("Node "+node+" not in " + repr(self.Env["nodes"]) + " from " +repr(partition)) if len(other_nodes) == 0: return 1 self.debug("Creating partition: " + repr(partition)) self.debug("Everyone else: " + repr(other_nodes)) for node in partition: if not self.CM.isolate_node(node, other_nodes): self.log("Could not isolate %s" % node) return 0 return 1 def heal_partition(self, partition): other_nodes = [] other_nodes.extend(self.Env["nodes"]) for node in partition: try: other_nodes.remove(node) except ValueError: self.log("Node "+node+" not in " + repr(self.Env["nodes"])) if len(other_nodes) == 0: return 1 self.debug("Healing partition: " + repr(partition)) self.debug("Everyone else: " + repr(other_nodes)) for node in partition: self.CM.unisolate_node(node, other_nodes) def __call__(self, node): '''Perform split-brain test''' self.incr("calls") self.passed = 1 partitions = {} ret = self.startall(None) if not ret: return self.failure("Setup failed") while 1: # Retry until we get multiple partitions partitions = {} p_max = len(self.Env["nodes"]) for node in self.Env["nodes"]: p = self.Env.RandomGen.randint(1, p_max) if not partitions.has_key(p): partitions[p]= [] partitions[p].append(node) p_max = len(partitions.keys()) if p_max > 1: break # else, try again self.debug("Created %d partitions" % p_max) for key in partitions.keys(): self.debug("Partition["+str(key)+"]:\t"+repr(partitions[key])) # Disabling STONITH to reduce test complexity for now self.rsh(node, "crm_attribute -V -n stonith-enabled -v false") for key in partitions.keys(): self.isolate_partition(partitions[key]) count = 30 while count > 0: if len(self.CM.find_partitions()) != p_max: time.sleep(10) else: break else: self.failure("Expected partitions were not created") # Target number of partitions formed - wait for stability if not self.CM.cluster_stable(): self.failure("Partitioned cluster not stable") # Now audit the cluster state self.CM.partitions_expected = p_max if not self.audit(): self.failure("Audits failed") self.CM.partitions_expected = 1 # And heal them again for key in partitions.keys(): self.heal_partition(partitions[key]) # Wait for a single partition to form count = 30 while count > 0: if len(self.CM.find_partitions()) != 1: time.sleep(10) count -= 1 else: break else: self.failure("Cluster did not reform") # Wait for it to have the right number of members count = 30 while count > 0: members = [] partitions = self.CM.find_partitions() if len(partitions) > 0: members = partitions[0].split() if len(members) != len(self.Env["nodes"]): time.sleep(10) count -= 1 else: break else: self.failure("Cluster did not completely reform") # Wait up to 20 minutes - the delay is more preferable than # trying to continue with in a messed up state if not self.CM.cluster_stable(1200): self.failure("Reformed cluster not stable") answer = raw_input('Continue? [nY]') if answer and answer == "n": raise ValueError("Reformed cluster not stable") # Turn fencing back on if self.Env["DoFencing"]: self.rsh(node, "crm_attribute -V -D -n stonith-enabled") self.CM.cluster_stable() if self.passed: return self.success() return self.failure("See previous errors") def errorstoignore(self): '''Return list of errors which are 'normal' and should be ignored''' return [ "Another DC detected:", "(ERROR|error): attrd_cib_callback: .*Application of an update diff failed", "crmd_ha_msg_callback:.*not in our membership list", "CRIT:.*node.*returning after partition", ] def is_applicable(self): if not self.is_applicable_common(): return 0 return len(self.Env["nodes"]) > 2 AllTestClasses.append(SplitBrainTest) #################################################################### class Reattach(CTSTest): #################################################################### def __init__(self, cm): CTSTest.__init__(self,cm) self.name="Reattach" self.startall = SimulStartLite(cm) self.restart1 = RestartTest(cm) self.stopall = SimulStopLite(cm) self.is_unsafe = 0 # Handled by canrunnow() def setup(self, node): attempt=0 if not self.startall(None): return None # Make sure we are really _really_ stable and that all # resources, including those that depend on transient node # attributes, are started while not self.CM.cluster_stable(double_check=True): if attempt < 5: attempt += 1 self.debug("Not stable yet, re-testing") else: self.log("Cluster is not stable") return None return 1 def teardown(self, node): # Make sure 'node' is up start = StartTest(self.CM) start(node) is_managed = self.rsh(node, "crm_attribute -Q -G -t crm_config -n is-managed-default -d true", 1) is_managed = is_managed[:-1] # Strip off the newline if is_managed != "true": self.log("Attempting to re-enable resource management on %s (%s)" % (node, is_managed)) managed = self.create_watch(["is-managed-default"], 60) managed.setwatch() self.rsh(node, "crm_attribute -V -D -n is-managed-default") if not managed.lookforall(): self.log("Patterns not found: " + repr(managed.unmatched)) self.log("Could not re-enable resource management") return 0 return 1 def canrunnow(self, node): '''Return TRUE if we can meaningfully run right now''' if self.find_ocfs2_resources(node): self.log("Detach/Reattach scenarios are not possible with OCFS2 services present") return 0 return 1 def __call__(self, node): self.incr("calls") pats = [] managed = self.create_watch(["is-managed-default"], 60) managed.setwatch() self.debug("Disable resource management") self.rsh(node, "crm_attribute -V -n is-managed-default -v false") if not managed.lookforall(): self.log("Patterns not found: " + repr(managed.unmatched)) return self.failure("Resource management not disabled") pats = [] - pats.append(self.log_patterns["Pat:RscOpOK"] % (".*", "start")) - pats.append(self.log_patterns["Pat:RscOpOK"] % (".*", "stop")) - pats.append(self.log_patterns["Pat:RscOpOK"] % (".*", "promote")) - pats.append(self.log_patterns["Pat:RscOpOK"] % (".*", "demote")) - pats.append(self.log_patterns["Pat:RscOpOK"] % (".*", "migrate")) + pats.append(self.templates["Pat:RscOpOK"] % (".*", "start")) + pats.append(self.templates["Pat:RscOpOK"] % (".*", "stop")) + pats.append(self.templates["Pat:RscOpOK"] % (".*", "promote")) + pats.append(self.templates["Pat:RscOpOK"] % (".*", "demote")) + pats.append(self.templates["Pat:RscOpOK"] % (".*", "migrate")) watch = self.create_watch(pats, 60, "ShutdownActivity") watch.setwatch() self.debug("Shutting down the cluster") ret = self.stopall(None) if not ret: self.debug("Re-enable resource management") self.rsh(node, "crm_attribute -V -D -n is-managed-default") return self.failure("Couldn't shut down the cluster") self.debug("Bringing the cluster back up") ret = self.startall(None) time.sleep(5) # allow ping to update the CIB if not ret: self.debug("Re-enable resource management") self.rsh(node, "crm_attribute -V -D -n is-managed-default") return self.failure("Couldn't restart the cluster") if self.local_badnews("ResourceActivity:", watch): self.debug("Re-enable resource management") self.rsh(node, "crm_attribute -V -D -n is-managed-default") return self.failure("Resources stopped or started during cluster restart") watch = self.create_watch(pats, 60, "StartupActivity") watch.setwatch() managed = self.create_watch(["is-managed-default"], 60) managed.setwatch() self.debug("Re-enable resource management") self.rsh(node, "crm_attribute -V -D -n is-managed-default") if not managed.lookforall(): self.log("Patterns not found: " + repr(managed.unmatched)) return self.failure("Resource management not enabled") self.CM.cluster_stable() # Ignore actions for STONITH resources ignore = [] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rclass == "stonith": self.debug("Ignoring start actions for %s" % r.id) - ignore.append(self.log_patterns["Pat:RscOpOK"] % (r.id, "start_0")) + ignore.append(self.templates["Pat:RscOpOK"] % (r.id, "start_0")) if self.local_badnews("ResourceActivity:", watch, ignore): return self.failure("Resources stopped or started after resource management was re-enabled") return ret def errorstoignore(self): '''Return list of errors which should be ignored''' return [ "resources were active at shutdown", "pingd: .*(ERROR|error): send_ipc_message:", "pingd: .*(ERROR|error): send_update:", "lrmd: .*(ERROR|error): notify_client:", ] def is_applicable(self): - if self.log_patterns["Name"] == "crm-lha": + if self.templates["Name"] == "crm-lha": return None return 1 AllTestClasses.append(Reattach) #################################################################### class SpecialTest1(CTSTest): #################################################################### '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="SpecialTest1" self.startall = SimulStartLite(cm) self.restart1 = RestartTest(cm) self.stopall = SimulStopLite(cm) def __call__(self, node): '''Perform the 'SpecialTest1' test for Andrew. ''' self.incr("calls") # Shut down all the nodes... ret = self.stopall(None) if not ret: return self.failure("Could not stop all nodes") # Test config recovery when the other nodes come up self.rsh(node, "rm -f "+CTSvars.CRM_CONFIG_DIR+"/cib*") # Start the selected node ret = self.restart1(node) if not ret: return self.failure("Could not start "+node) # Start all remaining nodes ret = self.startall(None) if not ret: return self.failure("Could not start the remaining nodes") return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' # Errors that occur as a result of the CIB being wiped return [ """warning: retrieveCib: Cluster configuration not found:""", """error: cib_perform_op: v1 patchset error, patch failed to apply: Application of an update diff failed""", """error: unpack_resources: Resource start-up disabled since no STONITH resources have been defined""", """error: unpack_resources: Either configure some or disable STONITH with the stonith-enabled option""", """error: unpack_resources: NOTE: Clusters with shared data need STONITH to ensure data integrity""", ] AllTestClasses.append(SpecialTest1) #################################################################### class HAETest(CTSTest): #################################################################### '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="HAETest" self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) self.is_loop = 1 def setup(self, node): # Start all remaining nodes ret = self.startall(None) if not ret: return self.failure("Couldn't start all nodes") return self.success() def teardown(self, node): # Stop everything ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") return self.success() def wait_on_state(self, node, resource, expected_clones, attempts=240): while attempts > 0: active=0 (rc, lines) = self.rsh(node, "crm_resource -r %s -W -Q" % resource, stdout=None) # Hack until crm_resource does the right thing if rc == 0 and lines: active = len(lines) if len(lines) == expected_clones: return 1 elif rc == 1: self.debug("Resource %s is still inactive" % resource) elif rc == 234: self.log("Unknown resource %s" % resource) return 0 elif rc == 246: self.log("Cluster is inactive") return 0 elif rc != 0: self.log("Call to crm_resource failed, rc=%d" % rc) return 0 else: self.debug("Resource %s is active on %d times instead of %d" % (resource, active, expected_clones)) attempts -= 1 time.sleep(1) return 0 def find_dlm(self, node): self.r_dlm = None (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rtype == "controld" and r.parent != "NA": self.debug("Found dlm: %s" % self.r_dlm) self.r_dlm = r.parent return 1 return 0 def find_hae_resources(self, node): self.r_dlm = None self.r_o2cb = None self.r_ocfs2 = [] if self.find_dlm(node): self.find_ocfs2_resources(node) def is_applicable(self): if not self.is_applicable_common(): return 0 if self.Env["Schema"] == "hae": return 1 return None #################################################################### class HAERoleTest(HAETest): #################################################################### def __init__(self, cm): '''Lars' mount/unmount test for the HA extension. ''' HAETest.__init__(self,cm) self.name="HAERoleTest" def change_state(self, node, resource, target): rc = self.rsh(node, "crm_resource -V -r %s -p target-role -v %s --meta" % (resource, target)) return rc def __call__(self, node): self.incr("calls") lpc = 0 failed = 0 delay = 2 done=time.time() + self.Env["loop-minutes"]*60 self.find_hae_resources(node) clone_max = len(self.Env["nodes"]) while time.time() <= done and not failed: lpc = lpc + 1 self.change_state(node, self.r_dlm, "Stopped") if not self.wait_on_state(node, self.r_dlm, 0): self.failure("%s did not go down correctly" % self.r_dlm) failed = lpc self.change_state(node, self.r_dlm, "Started") if not self.wait_on_state(node, self.r_dlm, clone_max): self.failure("%s did not come up correctly" % self.r_dlm) failed = lpc if not self.wait_on_state(node, self.r_o2cb, clone_max): self.failure("%s did not come up correctly" % self.r_o2cb) failed = lpc for fs in self.r_ocfs2: if not self.wait_on_state(node, fs, clone_max): self.failure("%s did not come up correctly" % fs) failed = lpc if failed: return self.failure("iteration %d failed" % failed) return self.success() AllTestClasses.append(HAERoleTest) #################################################################### class HAEStandbyTest(HAETest): #################################################################### '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): HAETest.__init__(self,cm) self.name="HAEStandbyTest" def change_state(self, node, resource, target): rc = self.rsh(node, "crm_standby -V -l reboot -v %s" % (target)) return rc def __call__(self, node): self.incr("calls") lpc = 0 failed = 0 done=time.time() + self.Env["loop-minutes"]*60 self.find_hae_resources(node) clone_max = len(self.Env["nodes"]) while time.time() <= done and not failed: lpc = lpc + 1 self.change_state(node, self.r_dlm, "true") if not self.wait_on_state(node, self.r_dlm, clone_max-1): self.failure("%s did not go down correctly" % self.r_dlm) failed = lpc self.change_state(node, self.r_dlm, "false") if not self.wait_on_state(node, self.r_dlm, clone_max): self.failure("%s did not come up correctly" % self.r_dlm) failed = lpc if not self.wait_on_state(node, self.r_o2cb, clone_max): self.failure("%s did not come up correctly" % self.r_o2cb) failed = lpc for fs in self.r_ocfs2: if not self.wait_on_state(node, fs, clone_max): self.failure("%s did not come up correctly" % fs) failed = lpc if failed: return self.failure("iteration %d failed" % failed) return self.success() AllTestClasses.append(HAEStandbyTest) ################################################################### class NearQuorumPointTest(CTSTest): ################################################################### ''' This test brings larger clusters near the quorum point (50%). In addition, it will test doing starts and stops at the same time. Here is how I think it should work: - loop over the nodes and decide randomly which will be up and which will be down Use a 50% probability for each of up/down. - figure out what to do to get into that state from the current state - in parallel, bring up those going up and bring those going down. ''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="NearQuorumPoint" def __call__(self, dummy): '''Perform the 'NearQuorumPoint' test. ''' self.incr("calls") startset = [] stopset = [] stonith = self.CM.prepare_fencing_watcher("NearQuorumPoint") #decide what to do with each node for node in self.Env["nodes"]: action = self.Env.RandomGen.choice(["start","stop"]) #action = self.Env.RandomGen.choice(["start","stop","no change"]) if action == "start" : startset.append(node) elif action == "stop" : stopset.append(node) self.debug("start nodes:" + repr(startset)) self.debug("stop nodes:" + repr(stopset)) #add search patterns watchpats = [ ] for node in stopset: if self.CM.ShouldBeStatus[node] == "up": - watchpats.append(self.log_patterns["Pat:We_stopped"] % node) + watchpats.append(self.templates["Pat:We_stopped"] % node) for node in startset: if self.CM.ShouldBeStatus[node] == "down": - #watchpats.append(self.log_patterns["Pat:Slave_started"] % node) - watchpats.append(self.log_patterns["Pat:Local_started"] % node) + #watchpats.append(self.templates["Pat:Slave_started"] % node) + watchpats.append(self.templates["Pat:Local_started"] % node) else: for stopping in stopset: if self.CM.ShouldBeStatus[stopping] == "up": - watchpats.append(self.log_patterns["Pat:They_stopped"] % (node, self.CM.key_for_node(stopping))) + watchpats.append(self.templates["Pat:They_stopped"] % (node, self.CM.key_for_node(stopping))) if len(watchpats) == 0: return self.skipped() if len(startset) != 0: - watchpats.append(self.log_patterns["Pat:DC_IDLE"]) + watchpats.append(self.templates["Pat:DC_IDLE"]) - watch = self.create_watch(watchpats, self.log_patterns["DeadTime"]+10) + watch = self.create_watch(watchpats, self.templates["DeadTime"]+10) watch.setwatch() #begin actions for node in stopset: if self.CM.ShouldBeStatus[node] == "up": self.CM.StopaCMnoBlock(node) for node in startset: if self.CM.ShouldBeStatus[node] == "down": self.CM.StartaCMnoBlock(node) #get the result if watch.lookforall(): self.CM.cluster_stable() self.CM.fencing_cleanup("NearQuorumPoint", stonith) return self.success() self.log("Warn: Patterns not found: " + repr(watch.unmatched)) #get the "bad" nodes upnodes = [] for node in stopset: if self.CM.StataCM(node) == 1: upnodes.append(node) downnodes = [] for node in startset: if self.CM.StataCM(node) == 0: downnodes.append(node) self.CM.fencing_cleanup,("NearQuorumPoint", stonith) if upnodes == [] and downnodes == []: self.CM.cluster_stable() # Make sure they're completely down with no residule for node in stopset: - self.rsh(node, self.log_patterns["StopCmd"]) + self.rsh(node, self.templates["StopCmd"]) return self.success() if len(upnodes) > 0: self.log("Warn: Unstoppable nodes: " + repr(upnodes)) if len(downnodes) > 0: self.log("Warn: Unstartable nodes: " + repr(downnodes)) return self.failure() def is_applicable(self): - if self.log_patterns["Name"] == "crm-cman": + if self.templates["Name"] == "crm-cman": return None return 1 AllTestClasses.append(NearQuorumPointTest) ################################################################### class RollingUpgradeTest(CTSTest): ################################################################### '''Perform a rolling upgrade of the cluster''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="RollingUpgrade" self.start = StartTest(cm) self.stop = StopTest(cm) self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) def setup(self, node): # Start all remaining nodes ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") for node in self.Env["nodes"]: if not self.downgrade(node, None): return self.failure("Couldn't downgrade %s" % node) ret = self.startall(None) if not ret: return self.failure("Couldn't start all nodes") return self.success() def teardown(self, node): # Stop everything ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") for node in self.Env["nodes"]: if not self.upgrade(node, None): return self.failure("Couldn't upgrade %s" % node) return self.success() def install(self, node, version, start=1, flags="--force"): target_dir = "/tmp/rpm-%s" % version src_dir = "%s/%s" % (self.Env["rpm-dir"], version) self.log("Installing %s on %s with %s" % (version, node, flags)) if not self.stop(node): return self.failure("stop failure: "+node) rc = self.rsh(node, "mkdir -p %s" % target_dir) rc = self.rsh(node, "rm -f %s/*.rpm" % target_dir) (rc, lines) = self.rsh(node, "ls -1 %s/*.rpm" % src_dir, None) for line in lines: line = line[:-1] rc = self.rsh.cp("%s" % (line), "%s:%s/" % (node, target_dir)) rc = self.rsh(node, "rpm -Uvh %s %s/*.rpm" % (flags, target_dir)) if start and not self.start(node): return self.failure("start failure: "+node) return self.success() def upgrade(self, node, start=1): return self.install(node, self.Env["current-version"], start) def downgrade(self, node, start=1): return self.install(node, self.Env["previous-version"], start, "--force --nodeps") def __call__(self, node): '''Perform the 'Rolling Upgrade' test. ''' self.incr("calls") for node in self.Env["nodes"]: if self.upgrade(node): return self.failure("Couldn't upgrade %s" % node) self.CM.cluster_stable() return self.success() def is_applicable(self): if not self.is_applicable_common(): return None if not self.Env.has_key("rpm-dir"): return None if not self.Env.has_key("current-version"): return None if not self.Env.has_key("previous-version"): return None return 1 # Register RestartTest as a good test to run AllTestClasses.append(RollingUpgradeTest) ################################################################### class BSC_AddResource(CTSTest): ################################################################### '''Add a resource to the cluster''' def __init__(self, cm): CTSTest.__init__(self, cm) self.name="AddResource" self.resource_offset = 0 self.cib_cmd="""cibadmin -C -o %s -X '%s' """ def __call__(self, node): self.incr("calls") self.resource_offset = self.resource_offset + 1 r_id = "bsc-rsc-%s-%d" % (node, self.resource_offset) start_pat = "crmd.*%s_start_0.*confirmed.*ok" patterns = [] patterns.append(start_pat % r_id) - watch = self.create_watch(patterns, self.log_patterns["DeadTime"]) + watch = self.create_watch(patterns, self.templates["DeadTime"]) watch.setwatch() ip = self.NextIP() if not self.make_ip_resource(node, r_id, "ocf", "IPaddr", ip): return self.failure("Make resource %s failed" % r_id) failed = 0 watch_result = watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.log ("Warn: Pattern not found: %s" % (regex)) failed = 1 if failed: return self.failure("Resource pattern(s) not found") - if not self.CM.cluster_stable(self.log_patterns["DeadTime"]): + if not self.CM.cluster_stable(self.templates["DeadTime"]): return self.failure("Unstable cluster") return self.success() def NextIP(self): ip = self.Env["IPBase"] if ":" in ip: fields = ip.rpartition(":") fields[2] = str(hex(int(fields[2], 16)+1)) print str(hex(int(f[2], 16)+1)) else: fields = ip.rpartition('.') fields[2] = str(int(fields[2])+1) ip = fields[0] + fields[1] + fields[3]; self.Env["IPBase"] = ip return ip.strip() def make_ip_resource(self, node, id, rclass, type, ip): self.log("Creating %s::%s:%s (%s) on %s" % (rclass,type,id,ip,node)) rsc_xml=""" """ % (id, rclass, type, id, id, ip) node_constraint=""" """ % (id, id, id, id, node) rc = 0 (rc, lines) = self.rsh(node, self.cib_cmd % ("constraints", node_constraint), None) if rc != 0: self.log("Constraint creation failed: %d" % rc) return None (rc, lines) = self.rsh(node, self.cib_cmd % ("resources", rsc_xml), None) if rc != 0: self.log("Resource creation failed: %d" % rc) return None return 1 def is_applicable(self): if self.Env["DoBSC"]: return 1 return None AllTestClasses.append(BSC_AddResource) class SimulStopLite(CTSTest): ################################################################### '''Stop any active nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="SimulStopLite" def __call__(self, dummy): '''Perform the 'SimulStopLite' setup work. ''' self.incr("calls") self.debug("Setup: " + self.name) # We ignore the "node" parameter... watchpats = [ ] for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.incr("WasStarted") - watchpats.append(self.log_patterns["Pat:We_stopped"] % node) + watchpats.append(self.templates["Pat:We_stopped"] % node) #if self.Env["use_logd"]: - # watchpats.append(self.log_patterns["Pat:Logd_stopped"] % node) + # watchpats.append(self.templates["Pat:Logd_stopped"] % node) if len(watchpats) == 0: self.CM.clear_all_caches() return self.success() # Stop all the nodes - at about the same time... - watch = self.create_watch(watchpats, self.log_patterns["DeadTime"]+10) + watch = self.create_watch(watchpats, self.templates["DeadTime"]+10) watch.setwatch() self.set_timer() for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.CM.StopaCMnoBlock(node) if watch.lookforall(): self.CM.clear_all_caches() # Make sure they're completely down with no residule for node in self.Env["nodes"]: - self.rsh(node, self.log_patterns["StopCmd"]) + self.rsh(node, self.templates["StopCmd"]) return self.success() did_fail=0 up_nodes = [] for node in self.Env["nodes"]: if self.CM.StataCM(node) == 1: did_fail=1 up_nodes.append(node) if did_fail: return self.failure("Active nodes exist: " + repr(up_nodes)) self.log("Warn: All nodes stopped but CTS didnt detect: " + repr(watch.unmatched)) self.CM.clear_all_caches() return self.failure("Missing log message: "+repr(watch.unmatched)) def is_applicable(self): '''SimulStopLite is a setup test and never applicable''' return 0 ################################################################### class SimulStartLite(CTSTest): ################################################################### '''Start any stopped nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name="SimulStartLite" def __call__(self, dummy): '''Perform the 'SimulStartList' setup work. ''' self.incr("calls") self.debug("Setup: " + self.name) # We ignore the "node" parameter... node_list = [] for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "down": self.incr("WasStopped") node_list.append(node) self.set_timer() while len(node_list) > 0: watchpats = [ ] - uppat = self.log_patterns["Pat:Slave_started"] + uppat = self.templates["Pat:Slave_started"] if self.CM.upcount() == 0: - uppat = self.log_patterns["Pat:Local_started"] + uppat = self.templates["Pat:Local_started"] - watchpats.append(self.log_patterns["Pat:DC_IDLE"]) + watchpats.append(self.templates["Pat:DC_IDLE"]) for node in node_list: watchpats.append(uppat % node) - watchpats.append(self.log_patterns["Pat:InfraUp"] % node) - watchpats.append(self.log_patterns["Pat:PacemakerUp"] % node) + watchpats.append(self.templates["Pat:InfraUp"] % node) + watchpats.append(self.templates["Pat:PacemakerUp"] % node) # Start all the nodes - at about the same time... - watch = self.create_watch(watchpats, self.log_patterns["DeadTime"]+10) + watch = self.create_watch(watchpats, self.templates["DeadTime"]+10) watch.setwatch() stonith = self.CM.prepare_fencing_watcher(self.name) for node in node_list: self.CM.StartaCMnoBlock(node) watch.lookforall() node_list = self.CM.fencing_cleanup(self.name, stonith) # Remove node_list messages from watch.unmatched for node in node_list: if watch.unmatched: watch.unmatched.remove(uppat % node) if watch.unmatched: for regex in watch.unmatched: self.log ("Warn: Startup pattern not found: %s" %(regex)) if not self.CM.cluster_stable(): return self.failure("Cluster did not stabilize") did_fail=0 unstable = [] for node in self.Env["nodes"]: if self.CM.StataCM(node) == 0: did_fail=1 unstable.append(node) if did_fail: return self.failure("Unstarted nodes exist: " + repr(unstable)) unstable = [] for node in self.Env["nodes"]: if not self.CM.node_stable(node): did_fail=1 unstable.append(node) if did_fail: return self.failure("Unstable cluster nodes exist: " + repr(unstable)) return self.success() def is_applicable(self): '''SimulStartLite is a setup test and never applicable''' return 0 def TestList(cm, audits): result = [] for testclass in AllTestClasses: bound_test = testclass(cm) if bound_test.is_applicable(): bound_test.Audits = audits result.append(bound_test) return result ################################################################### class RemoteLXC(CTSTest): ################################################################### def __init__(self, cm): CTSTest.__init__(self,cm) self.name="RemoteLXC" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.num_containers = 2 self.is_container = 1 self.failed = 0 self.fail_string = "" def start_lxc_simple(self, node): rc = self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -v &>/dev/null") if rc == 1: return self.skipped() # restore any artifacts laying around from a previous test. self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -R &>/dev/null") # generate the containers, put them in the config, add some resources to them pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() - pats.append(self.log_patterns["Pat:RscOpOK"] % ("lxc1", "start_0")) - pats.append(self.log_patterns["Pat:RscOpOK"] % ("lxc2", "start_0")) - pats.append(self.log_patterns["Pat:RscOpOK"] % ("lxc-ms", "start_0")) - pats.append(self.log_patterns["Pat:RscOpOK"] % ("lxc-ms", "promote_0")) + pats.append(self.templates["Pat:RscOpOK"] % ("lxc1", "start_0")) + pats.append(self.templates["Pat:RscOpOK"] % ("lxc2", "start_0")) + pats.append(self.templates["Pat:RscOpOK"] % ("lxc-ms", "start_0")) + pats.append(self.templates["Pat:RscOpOK"] % ("lxc-ms", "promote_0")) self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -g -a -m -s -c %d &>/dev/null" % self.num_containers) self.set_timer("remoteSimpleInit") watch.lookforall() self.log_timer("remoteSimpleInit") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 def cleanup_lxc_simple(self, node): pats = [ ] # if the test failed, attempt to clean up the cib and libvirt environment # as best as possible if self.failed == 1: # restore libvirt and cib self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -R &>/dev/null") self.rsh(node, "crm_resource -C -r container1 &>/dev/null") self.rsh(node, "crm_resource -C -r container2 &>/dev/null") self.rsh(node, "crm_resource -C -r lxc1 &>/dev/null") self.rsh(node, "crm_resource -C -r lxc2 &>/dev/null") self.rsh(node, "crm_resource -C -r lxc-ms &>/dev/null") time.sleep(20) return watch = self.create_watch(pats, 120) watch.setwatch() - pats.append(self.log_patterns["Pat:RscOpOK"] % ("container1", "stop_0")) - pats.append(self.log_patterns["Pat:RscOpOK"] % ("container2", "stop_0")) + pats.append(self.templates["Pat:RscOpOK"] % ("container1", "stop_0")) + pats.append(self.templates["Pat:RscOpOK"] % ("container2", "stop_0")) self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -p &>/dev/null") self.set_timer("remoteSimpleCleanup") watch.lookforall() self.log_timer("remoteSimpleCleanup") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 # cleanup libvirt self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -R &>/dev/null") def __call__(self, node): '''Perform the 'RemoteLXC' test. ''' self.incr("calls") ret = self.startall(None) if not ret: return self.failure("Setup failed, start all nodes failed.") self.start_lxc_simple(node) self.cleanup_lxc_simple(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed == 1: return self.failure(self.fail_string) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ """Updating failcount for ping""", """LogActions: Recover ping""", """LogActions: Recover lxc-ms""", """LogActions: Recover container""", # The orphaned lxc-ms resource causes an expected transition error # that is a result of the pengine not having knowledge that the # ms resource used to be a clone. As a result it looks like that # resource is running in multiple locations when it shouldn't... But in # this instance we know why this error is occurring and that it is expected. """Calculated Transition .* /var/lib/pacemaker/pengine/pe-error""", """Resource lxc-ms .* is active on 2 nodes attempting recovery""", """Unknown operation: fail""", """notice: operation_finished: ping-""", """notice: operation_finished: container""", """notice: operation_finished: .*_monitor_0:.*:stderr""", """(ERROR|error): sending stonithRA op to stonithd failed.""", ] AllTestClasses.append(RemoteLXC) ################################################################### class RemoteBaremetal(CTSTest): ################################################################### def __init__(self, cm): CTSTest.__init__(self,cm) self.name="RemoteBaremetal" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.stop = StopTest(cm) self.pcmk_started=0 self.failed = 0 self.fail_string = "" self.remote_node_added = 0 self.remote_node="remote1" self.remote_rsc_added = 0 self.remote_rsc="remote1-rsc" self.cib_cmd="""cibadmin -C -o %s -X '%s' """ def del_rsc(self, node, rsc): for othernode in self.Env["nodes"]: if othernode == node: # we don't want to try and use the cib that we just shutdown. # find a cluster node that is not our soon to be remote-node. continue rc = self.rsh(othernode, "crm_resource -D -r %s -t primitive" % (rsc)) if rc != 0: self.fail_string = ("Removal of resource '%s' failed" % (rsc)) self.failed = 1 else: self.fail_string = "" self.failed = 0 break def add_rsc(self, node, rsc_xml): failed=0 fail_string="" for othernode in self.Env["nodes"]: if othernode == node: # we don't want to try and use the cib that we just shutdown. # find a cluster node that is not our soon to be remote-node. continue rc = self.rsh(othernode, self.cib_cmd % ("resources", rsc_xml)) if rc != 0: fail_string = "resource creation failed" failed = 1 else: fail_string = "" failed = 0 break if failed == 1: self.failed=failed self.fail_string=fail_string def add_primitive_rsc(self, node): rsc_xml=""" """ % (self.remote_rsc) self.add_rsc(node, rsc_xml) if self.failed == 0: self.remote_rsc_added=1 def add_connection_rsc(self, node): rsc_xml=""" """ % (self.remote_node, node) self.add_rsc(node, rsc_xml) if self.failed == 0: self.remote_node_added=1 def step1_start_metal(self, node): pcmk_started=0 # make sure the resource doesn't already exist for some reason self.rsh(node, "crm_resource -D -r %s -t primitive" % (self.remote_rsc)) self.rsh(node, "crm_resource -D -r %s -t primitive" % (self.remote_node)) if not self.stop(node): self.failed = 1 self.fail_string = "Failed to shutdown cluster node %s" % (node) return for i in range(10): rc = self.rsh(node, "service pacemaker_remote start") if rc != 0: time.sleep(6) else: self.pcmk_started = 1 break if self.pcmk_started == 0: self.failed = 1 self.fail_string = "Failed to start pacemaker_remote on node %s" % (node) return # convert node to baremetal node now that it has shutdow the cluster stack pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() - pats.append(self.log_patterns["Pat:RscOpOK"] % (self.remote_node, "start")) + pats.append(self.templates["Pat:RscOpOK"] % (self.remote_node, "start")) self.add_connection_rsc(node) self.set_timer("remoteMetalInit") watch.lookforall() self.log_timer("remoteMetalInit") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 def step2_add_rsc(self, node): if self.failed == 1: return # verify we can put a resource on the remote node pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() pats.append("process_lrm_event: Operation %s_start_0.*node=%s, .*confirmed.*ok" % (self.remote_rsc, self.remote_node)) # Add a resource that must live on remote-node self.add_primitive_rsc(node) # this crm_resource command actually occurs on the remote node # which verifies that the ipc proxy works rc = self.rsh(node, "crm_resource -M -r remote1-rsc -N %s" % (self.remote_node)) if rc != 0: self.fail_string = "Failed to place primitive on remote-node" self.failed = 1 return self.set_timer("remoteMetalRsc") watch.lookforall() self.log_timer("remoteMetalRsc") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 def cleanup_metal(self, node): if self.pcmk_started == 0: return pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() if self.remote_rsc_added == 1: - pats.append(self.log_patterns["Pat:RscOpOK"] % (self.remote_rsc, "stop")) + pats.append(self.templates["Pat:RscOpOK"] % (self.remote_rsc, "stop")) if self.remote_node_added == 1: - pats.append(self.log_patterns["Pat:RscOpOK"] % (self.remote_node, "stop")) + pats.append(self.templates["Pat:RscOpOK"] % (self.remote_node, "stop")) self.set_timer("remoteMetalCleanup") if self.remote_rsc_added == 1: self.rsh(node, "crm_resource -U -r remote1-rsc -N %s" % (self.remote_node)) self.del_rsc(node, self.remote_rsc) if self.remote_node_added == 1: self.del_rsc(node, self.remote_node) watch.lookforall() self.log_timer("remoteMetalCleanup") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 # disable pcmk remote for i in range(10): rc = self.rsh(node, "service pacemaker_remote stop") if rc != 0: time.sleep(6) else: break def setup_env(self): sync_key = 0 # we are assuming if all nodes have a key, that it is # the right key... If any node doesn't have a remote # key, we regenerate it everywhere. for node in self.Env["nodes"]: rc = self.rsh(node, "ls /etc/pacemaker/authkey") if rc != 0: sync_key = 1 break if sync_key == 0: return # create key locally os.system("/usr/share/pacemaker/tests/cts/lxc_autogen.sh -k &> /dev/null") # sync key throughout the cluster for node in self.Env["nodes"]: rc = self.rsh(node, "mkdir /etc/pacemaker") self.rsh.cp("/etc/pacemaker/authkey", "%s:/etc/pacemaker/authkey" % (node)) def __call__(self, node): '''Perform the 'RemoteBaremetal' test. ''' self.incr("calls") ret = self.startall(None) if not ret: return self.failure("Setup failed, start all nodes failed.") self.setup_env() self.step1_start_metal(node) self.step2_add_rsc(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed == 1: return self.failure(self.fail_string) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ """is running on remote1 which isn't allowed""", """Connection terminated""", """Failed to send remote""", ] AllTestClasses.append(RemoteBaremetal) # vim:ts=4:sw=4:et: diff --git a/cts/Makefile.am b/cts/Makefile.am index 399f18338d..d9af1a4b4e 100644 --- a/cts/Makefile.am +++ b/cts/Makefile.am @@ -1,48 +1,49 @@ # # heartbeat: Linux-HA heartbeat code # # Copyright (C) 2001 Michael Moerz # # 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. # MAINTAINERCLEANFILES = Makefile.in CLEANFILES = LSBDummy EXTRA_DIST = $(cts_SCRIPTS) $(cts_DATA) ctsdir = $(datadir)/$(PACKAGE)/tests/cts ctslibdir = $(pyexecdir)/cts ctslib_PYTHON = __init__.py \ CTSvars.py \ CM_lha.py \ CM_ais.py \ CTS.py \ CTSaudits.py \ CTStests.py \ + CTSpatt.py \ CTSscenarios.py \ CIB.py \ cib_xml.py cts_DATA = README cts.supp cts_SCRIPTS = cts \ CTSlab.py \ lxc_autogen.sh \ LSBDummy \ $(top_srcdir)/fencing/fence_dummy SUBDIRS = benchmark diff --git a/cts/patterns.py b/cts/patterns.py new file mode 100644 index 0000000000..f0a52e088d --- /dev/null +++ b/cts/patterns.py @@ -0,0 +1,360 @@ +from UserDict import UserDict +import sys, time, types, syslog, os, struct, string, signal, traceback, warnings, socket + +patternvariants = {} +class BasePatterns: + def __init__(self, name): + self.name = name + patternvariants[name] = self + self.ignore = [] + self.BadNews = [] + self.commands = { + "DeadTime" : 300, + "StartTime" : 300, # Max time to start up + "StableTime" : 30, + "StatusCmd" : "crmadmin -t 60000 -S %s 2>/dev/null", + "CibQuery" : "cibadmin -Ql", + "CibAddXml" : "cibadmin --modify -c --xml-text %s", + "CibDelXpath" : "cibadmin --delete --xpath %s", + # 300,000 == 5 minutes + "RscRunning" : CTSvars.CRM_DAEMON_DIR + "/lrmd_test -R -r %s", + "CIBfile" : "%s:"+CTSvars.CRM_CONFIG_DIR+"/cib.xml", + "TmpDir" : "/tmp", + + "BreakCommCmd" : "iptables -A INPUT -s %s -j DROP >/dev/null 2>&1", + "FixCommCmd" : "iptables -D INPUT -s %s -j DROP >/dev/null 2>&1", + +# tc qdisc add dev lo root handle 1: cbq avpkt 1000 bandwidth 1000mbit +# tc class add dev lo parent 1: classid 1:1 cbq rate "$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", + + "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", + } + self.search = { + "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:Fencing_start" : "Initiating remote operation .* for %s", + "Pat:Fencing_ok" : "stonith.* remote_op_done: Operation .* of %s by .*: OK", + + "Pat:RscOpOK" : "process_lrm_event: Operation %s_%s.*ok.*confirmed", + } + + def get_patterns(self, key): + if key == "BadNews": + return self.BadNews + elif key == "BadNewsIgnore": + return self.ignore + elif key == "Commands": + return self.commands + elif key == "Search": + return self.search + + def __getitem__(self, key): + if self.commands.has_key(key): + return self.commands[key] + elif self.search.has_key(key): + return self.search[key] + else: + print "Unknown template '%s' for %s" % (key, self.name) + return None + +class crm_lha(BasePatterns): + def __init__(self, name): + BasePatterns.__init__(self, name) + + self.commands.update({ + "StartCmd" : "service heartbeat start > /dev/null 2>&1", + "StopCmd" : "service heartbeat stop > /dev/null 2>&1", + "EpocheCmd" : "crm_node -H -e", + "QuorumCmd" : "crm_node -H -q", + "ParitionCmd" : "crm_node -H -p", + }) + + self.search.update({ + # Patterns to look for in the log files for various occasions... + "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", + }) + self.BadNews = [ + 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", + ] + + self.ignore = [ + "(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", + ] + +class crm_cs_v0(BasePatterns): + def __init__(self, name): + BasePatterns.__init__(self, name) + + self.commands.update({ + "EpocheCmd" : "crm_node -e --openais", + "QuorumCmd" : "crm_node -q --openais", + "ParitionCmd" : "crm_node -p --openais", + "StartCmd" : "service corosync start", + "StopCmd" : "service corosync stop", + }) + + self.search.update({ +# 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:ChildExit" : "Child process .* exited", + "Pat:ChildKilled" : "%s corosync.*Child process %s terminated with signal 9", + "Pat:ChildRespawn" : "%s corosync.*Respawning failed child process: %s", + }) + + self.ignore = [ + r"crm_mon:", + r"crmadmin:", + r"update_trace_data", + r"async_notify: strange, client not found", + r"Parse error: Ignoring unknown option .*nodename", + r"error: log_operation: Operation 'reboot' .* with device 'FencingFail' returned:", + r"Child process .* terminated with signal 9", + r"getinfo response error: 1$", + ] + + self.BadNews = [ + 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"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"Receiving messages from a node we think is dead", + r"share the same cluster nodeid", + r"share the same name", + + #r"crm_ipc_send:.*Request .* failed", + #r"crm_ipc_send:.*Sending to .* is disabled until pending reply is received", + + # Not inherently bad, but worth tracking + #r"No need to invoke the TE", + #r"ping.*: DEBUG: Updated connected = 0", + #r"Digest mis-match:", + r"te_graph_trigger: Transition failed: terminated", + r"process_ping_reply", + r"retrieveCib", + r"cib_process_replace", + #r"Executing .* fencing operation", + #r"fence_pcmk.* Call to fence", + #r"fence_pcmk", + r"cman killed by node", + r"Election storm", + r"stalled the FSA with pending inputs", + ] + +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, name): + crm_cs_v0.__init__(self, name) + + self.name = "crm-plugin-v0" + self.commands.update({ + "StartCmd" : "service corosync start && service pacemaker start", + "StopCmd" : "service pacemaker stop; service pacemaker_remote stop; service corosync stop", + + "EpocheCmd" : "crm_node -e", + "QuorumCmd" : "crm_node -q", + "ParitionCmd" : "crm_node -p", + }) + + self.search.update({ + # 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", + }) + +# if self.Env["have_systemd"]: +# self.update({ +# # When systemd is in use, we can look for this instead +# "Pat:We_stopped" : "%s.*Stopped Corosync Cluster Engine", +# }) + +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, name): + crm_cs_v0.__init__(self, name) + + self.commands.update({ + "StartCmd" : "service pacemaker start", + "StopCmd" : "service pacemaker stop; service pacemaker_remote 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", + }) + + +class PatternSelector: + + def __init__(self, name): + self.name = name + self.base = BasePatterns("crm-base") + + if not name: + crm_cs_v0("crm-plugin-v0") + crm_cman("crm-cman") + crm_mcp("crm-mcp") + elif name == "crm-lha": + crm_lha(name) + elif name == "crm-plugin-v0": + crm_cs_v0(name) + elif name == "crm-cman": + crm_cman(name) + elif name == "crm-mcp": + crm_mcp(name) + + def get_variant(self, variant): + if patternvariants.has_key(variant): + return patternvariants[variant] + return self.base + + def get_patterns(self, variant, kind): + return self.get_variant(variant).get_patterns(kind) + + def get_template(self, variant, key): + v = self.get_variant(variant) + return v[key] + + def __getitem__(self, key): + return self.get_template(self.name, key) + +# python cts/CTSpatt.py -k crm-mcp -t StartCmd +if __name__ == '__main__': + + pdir=os.path.dirname(sys.path[0]) + sys.path.insert(0, pdir) # So that things work from the source directory + + from cts.CTSvars import * + + kind=None + template=None + + skipthis=None + args=sys.argv[1:] + for i in range(0, len(args)): + if skipthis: + skipthis=None + continue + + elif args[i] == "-k" or args[i] == "--kind": + skipthis=1 + kind = args[i+1] + + elif args[i] == "-t" or args[i] == "--template": + skipthis=1 + template = args[i+1] + + else: + print "Illegal argument " + args[i] + + + print PatternSelector(kind)[template]