diff --git a/cts/CTS.py b/cts/CTS.py index 6ff7ff4507..8fccbdde8d 100755 --- a/cts/CTS.py +++ b/cts/CTS.py @@ -1,1362 +1,1363 @@ '''CTS: Cluster Testing System: Main module Classes related to testing high-availability clusters... ''' __copyright__=''' Copyright (C) 2000, 2001 Alan Robertson Licensed under the GNU GPL. ''' # # This program is free software; you can redistribute it and/or # modify it under the terms of the GNU General Public License # as published by the Free Software Foundation; either version 2 # of the License, or (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program; if not, write to the Free Software # Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. import types, string, select, sys, time, re, os, struct, signal import base64, pickle, binascii, fcntl from UserDict import UserDict from syslog import * from subprocess import Popen,PIPE from CTSvars import * 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): + def __init__(self, Env=None, silent=False): self.Env = Env + self.silent = silent # -n: no stdin, -x: no X11 self.Command = "ssh -l root -n -x" # -B: batch mode, -q: no stats (quiet) self.CpCommand = "scp -B -q" self.OurNode=string.lower(os.uname()[1]) def enable_qarsh(self): # http://nstraz.wordpress.com/2008/12/03/introducing-qarsh/ self.log("Using QARSH for connections to cluster nodes") self.Command = "qarsh -l root HOME=/root" self.CpCommand = "qacp" def _fixcmd(self, cmd): return re.sub("\'", "'\\''", cmd) def _cmd(self, *args): '''Compute the string that will run the given command on the given remote system''' args= args[0] sysname = args[0] command = args[1] #print "sysname: %s, us: %s" % (sysname, self.OurNode) if sysname == None or string.lower(sysname) == self.OurNode or sysname == "localhost": - ret = self._fixcmd(command) + 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.Env: - print (args) - else: - self.Env.log(args) + if not self.silent: + 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) + if not self.silent: + if not self.Env: + print (args) + else: + self.Env.debug(args) def __call__(self, node, command, stdout=0, blocking=1): '''Run the given command on the given remote system If you call this class like a function, this is the function that gets called. It just runs it roughly as though it were a system() call on the remote machine. The first argument is name of the machine to run it on. ''' rc = 0 result = None if not blocking: proc = Popen(self._cmd([node, command]), stdout = PIPE, stderr = PIPE, close_fds = True, shell = True) self.debug("cmd: async: target=%s, rc=%d: %s" % (node, proc.pid, command)) if proc.pid > 0: return 0 return -1 proc = Popen(self._cmd([node, command]), stdout = PIPE, stderr = PIPE, close_fds = True, shell = True) if stdout == 1: result = proc.stdout.readline() else: result = proc.stdout.readlines() proc.stdout.close() rc = proc.wait() 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() for err in errors: self.debug("cmd: stderr: %s" % err) if stdout == 0: if result: for line in result: self.debug("cmd: stdout: %s" % line) return rc return (rc, result) def cp(self, *args): '''Perform a remote copy''' cpstring = self.CpCommand for arg in args: cpstring = cpstring + " \'" + arg + "\'" rc = os.system(cpstring) self.debug("cmd: rc=%d: %s" % (rc, cpstring)) return rc has_log_watcher = {} log_watcher_bin = "/tmp/cts_log_watcher.py" log_watcher = """ import sys, os, fcntl ''' Remote logfile reader for CTS Reads a specified number of lines from the supplied offset Returns the current offset Contains logic for handling truncation ''' -count = 5 +limit = 5 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 - count = int(args[i+1]) + limit = int(args[i+1]) elif args[i] == '-f' or args[i] == '--filename': skipthis=1 filename = args[i+1] elif args[i] == '-o' or args[i] == '--offset': skipthis=1 offset = args[i+1] elif args[i] == '-p' or args[i] == '--prefix': skipthis=1 prefix = args[i+1] logfile=open(filename, 'r') logfile.seek(0, os.SEEK_END) newsize=logfile.tell() if offset != 'EOF': offset = int(offset) if newsize >= offset: logfile.seek(offset) else: print prefix + 'File truncated' logfile.seek(0) # Don't block when we reach EOF fcntl.fcntl(logfile.fileno(), fcntl.F_SETFL, os.O_NONBLOCK) +count = limit while count > 0: count -= 1 line = logfile.readline() if line: print line.strip() else: break -print prefix + 'Last read: %d' % logfile.tell() +print prefix + 'Last read: %d %d' % (logfile.tell(), limit - count) logfile.close() """ class SearchObj: def __init__(self, Env, filename, host=None): self.Env = Env self.host = host self.filename = filename self.cache = [] self.offset = "EOF" self.rsh = RemoteExec(Env) if host == None: host = "localhost" global has_log_watcher if not has_log_watcher.has_key(host): global log_watcher global log_watcher_bin - self.log("Installing %s on %s" % (log_watcher_bin, host)) + self.debug("Installing %s on %s" % (log_watcher_bin, host)) self.rsh(host, '''echo "%s" > %s''' % (log_watcher, log_watcher_bin)) has_log_watcher[host] = 1 self.next() def __str__(self): if self.host: return "%s:%s" % (self.host, self.filename) return self.filename def log(self, args): message = "lw: %s: %s" % (self, args) if not self.Env: print (message) else: self.Env.log(message) def debug(self, args): message = "lw: %s: %s" % (self, args) if not self.Env: print (message) else: self.Env.debug(message) def next(self): + cache = [] if not len(self.cache): global log_watcher_bin (rc, lines) = self.rsh( self.host, "python %s -p CTSwatcher: -f %s -o %s -l 20" % (log_watcher_bin, self.filename, self.offset), stdout=None) for line in lines: if self.host: self.debug("Got: %s" % line) match = re.search("^CTSwatcher:Last read: (\d+)", line) if match: self.offset = match.group(1) #self.debug("New offset: %s" % self.offset) elif re.search("^CTSwatcher:", line): self.debug("Got control line: "+ line) else: - self.cache.append(line) - - if self.cache: - line = self.cache[0] - self.cache.remove(line) - return line + cache.append(line) - return None + return cache class LogWatcher(RemoteExec): '''This class watches logs for messages that fit certain regular expressions. Watching logs for events isn't the ideal way to do business, but it's better than nothing :-) On the other hand, this class is really pretty cool ;-) The way you use this class is as follows: Construct a LogWatcher object Call setwatch() when you want to start watching the log Call look() to scan the log looking for the patterns ''' def __init__(self, Env, log, regexes, name="Anon", timeout=10, debug_level=None): '''This is the constructor for the LogWatcher class. It takes a log name to watch, and a list of regular expressions to watch for." ''' RemoteExec.__init__(self, Env) # Validate our arguments. Better sooner than later ;-) for regex in regexes: assert re.compile(regex) self.name = name self.regexes = regexes self.filename = log self.debug_level = debug_level self.whichmatch = -1 self.unmatched = None self.file_list = [] self.line_cache = [] for regex in self.regexes: self.debug("Looking for regex: "+regex) self.Timeout = int(timeout) self.returnonlymatch = None if not os.access(log, os.R_OK): raise ValueError("File [" + log + "] not accessible (r)") def debug(self, args): message = "lw: %s: %s" % (self.name, args) if not self.Env: print (message) else: self.Env.debug(message) def setwatch(self): '''Mark the place to start watching the log from. ''' if self.Env["LogWatcher"] == "remote": for node in self.Env["nodes"]: self.file_list.append(SearchObj(self.Env, self.filename, node)) else: self.file_list.append(SearchObj(self.Env, self.filename)) def __del__(self): if self.debug_level > 1: self.debug("Destroy") def ReturnOnlyMatch(self, onlymatch=1): '''Specify one or more subgroups of the match to return rather than the whole string http://www.python.org/doc/2.5.2/lib/match-objects.html ''' self.returnonlymatch = onlymatch def __get_line(self): if not len(self.line_cache): if not len(self.file_list): raise ValueError("No sources to read from") for f in self.file_list: - line = f.next() - if line: - self.line_cache.append(line) + lines = f.next() + self.line_cache.extend(lines) if self.line_cache: line = self.line_cache[0] self.line_cache.remove(line) return line return None def look(self, timeout=None): '''Examine the log looking for the given patterns. It starts looking from the place marked by setwatch(). This function looks in the file in the fashion of tail -f. It properly recovers from log file truncation, but not from removing and recreating the log. It would be nice if it recovered from this as well :-) We return the first line which matches any of our patterns. ''' if timeout == None: timeout = self.Timeout done=time.time()+timeout+1 if self.debug_level > 2: self.debug("starting single search: timeout=%d" % timeout) while (timeout <= 0 or time.time() <= done): line = self.__get_line() if line: which=-1 + if re.search("CTS:", line): + continue if self.debug_level > 2: self.debug("Processing: "+ line) for regex in self.regexes: which=which+1 if self.debug_level > 2: self.debug("Comparing line to: "+ regex) #matchobj = re.search(string.lower(regex), string.lower(line)) matchobj = re.search(regex, line) if matchobj: self.whichmatch=which if self.returnonlymatch: return matchobj.group(self.returnonlymatch) else: self.debug("Matched: "+line) if self.debug_level > 1: self.debug("With: "+ regex) return line elif timeout > 0: - time.sleep(2) + time.sleep(5) #time.sleep(0.025) else: self.debug("End of file") return None self.debug("Timeout") return None def lookforall(self, timeout=None, allow_multiple_matches=None): '''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 = [] 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) == 0 def IsSshdUp(self, node): #return self.rsh(node, "true") == 0; rc = self.Env.rsh(node, "true") return rc == 0 def WaitForNodeToComeUp(self, node, Timeout=300): '''Return TRUE when given node comes up, or None/FALSE if timeout''' timeout=Timeout anytimeouts=0 while timeout > 0: if self.IsNodeBooted(node) and self.IsSshdUp(node): if anytimeouts: # Fudge to wait for the system to finish coming up time.sleep(30) self.Env.debug("Node %s now up" % node) return 1 time.sleep(30) if (not anytimeouts): self.Env.debug("Waiting for node %s to come up" % node) anytimeouts=1 timeout = timeout - 1 self.Env.log("%s did not come up within %d tries" % (node, Timeout)) answer = raw_input('Continue? [nY]') if answer and answer == "n": raise ValueError("%s did not come up within %d tries" % (node, Timeout)) def WaitForAllNodesToComeUp(self, nodes, timeout=300): '''Return TRUE when all nodes come up, or FALSE if timeout''' for node in nodes: if not self.WaitForNodeToComeUp(node, timeout): return None return 1 class ClusterManager(UserDict): '''The Cluster Manager class. This is an subclass of the Python dictionary class. (this is because it contains lots of {name,value} pairs, not because it's behavior is that terribly similar to a dictionary in other ways.) This is an abstract class which class implements high-level operations on the cluster and/or its cluster managers. Actual cluster managers classes are subclassed from this type. One of the things we do is track the state we think every node should be in. ''' def __InitialConditions(self): #if os.geteuid() != 0: # raise ValueError("Must Be Root!") None def _finalConditions(self): for key in self.keys(): if self[key] == None: raise ValueError("Improper derivation: self[" + key + "] must be overridden by subclass.") def __init__(self, Environment, randseed=None): self.Env = Environment self.__InitialConditions() self.clear_cache = 0 self.TestLoggingLevel=0 self.data = { "up" : "up", # Status meaning up "down" : "down", # Status meaning down "StonithCmd" : "stonith -t baytech -p '10.10.10.100 admin admin' %s", "DeadTime" : 30, # Max time to detect dead node... "StartTime" : 90, # Max time to start up # # These next values need to be overridden in the derived class. # "Name" : None, "StartCmd" : None, "StopCmd" : None, "StatusCmd" : None, #"RereadCmd" : None, "BreakCommCmd" : None, "FixCommCmd" : None, #"TestConfigDir" : None, "LogFileName" : None, #"Pat:Master_started" : None, #"Pat:Slave_started" : None, "Pat:We_stopped" : None, "Pat:They_stopped" : None, "BadRegexes" : None, # A set of "bad news" regexes # to apply to the log } self.rsh = self.Env.rsh self.ShouldBeStatus={} self.OurNode=string.lower(os.uname()[1]) self.ShouldBeStatus={} self.ns = NodeStatus(self.Env) def errorstoignore(self): '''Return list of errors which are 'normal' and should be ignored''' return [] def log(self, args): self.Env.log(args) def debug(self, args): self.Env.debug(args) def prepare(self): '''Finish the Initialization process. Prepare to test...''' for node in self.Env["nodes"]: if self.StataCM(node): self.ShouldBeStatus[node]="up" else: self.ShouldBeStatus[node]="down" self.unisolate_node(node) def upcount(self): '''How many nodes are up?''' count=0 for node in self.Env["nodes"]: if self.ShouldBeStatus[node]=="up": count=count+1 return count def install_config(self, node): return None def clear_all_caches(self): if self.clear_cache: for node in self.Env["nodes"]: if self.ShouldBeStatus[node] == "down": self.debug("Removing cache file on: "+node) self.rsh(node, "rm -f "+CTSvars.HA_VARLIBHBDIR+"/hostcache") else: self.debug("NOT Removing cache file on: "+node) def StartaCM(self, node): '''Start up the cluster manager on a given node''' self.debug("Starting %s on node %s" %(self["Name"], node)) ret = 1 if not self.ShouldBeStatus.has_key(node): self.ShouldBeStatus[node] = "down" if self.ShouldBeStatus[node] != "down": return 1 patterns = [] # Technically we should always be able to notice ourselves starting patterns.append(self["Pat:Local_started"] % node) if self.upcount() == 0: patterns.append(self["Pat:Master_started"] % node) else: patterns.append(self["Pat:Slave_started"] % node) watch = LogWatcher( self.Env, self["LogFileName"], patterns, "StartaCM", self["StartTime"]+10) watch.setwatch() self.install_config(node) self.ShouldBeStatus[node] = "any" if self.StataCM(node) and self.cluster_stable(self["DeadTime"]): self.log ("%s was already started" %(node)) return 1 # Clear out the host cache so autojoin can be exercised if self.clear_cache: self.debug("Removing cache file on: "+node) self.rsh(node, "rm -f "+CTSvars.HA_VARLIBHBDIR+"/hostcache") if not(self.Env["valgrind-tests"]): startCmd = self["StartCmd"] else: if self.Env["valgrind-prefix"]: prefix = self.Env["valgrind-prefix"] else: prefix = "cts" startCmd = """G_SLICE=always-malloc HA_VALGRIND_ENABLED='%s' VALGRIND_OPTS='%s --log-file=/tmp/%s-%s.valgrind' %s""" % ( self.Env["valgrind-procs"], self.Env["valgrind-opts"], prefix, """%p""", self["StartCmd"]) if self.rsh(node, startCmd) != 0: self.log ("Warn: Start command failed on node %s" %(node)) return None self.ShouldBeStatus[node]="up" watch_result = watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.log ("Warn: Startup pattern not found: %s" %(regex)) if watch_result: #self.debug("Found match: "+ repr(watch_result)) self.cluster_stable(self["DeadTime"]) return 1 if self.StataCM(node) and self.cluster_stable(self["DeadTime"]): return 1 self.log ("Warn: Start failed for node %s" %(node)) return None def StartaCMnoBlock(self, node): '''Start up the cluster manager on a given node with none-block mode''' self.debug("Starting %s on node %s" %(self["Name"], node)) # Clear out the host cache so autojoin can be exercised if self.clear_cache: self.debug("Removing cache file on: "+node) self.rsh(node, "rm -f "+CTSvars.HA_VARLIBHBDIR+"/hostcache") if not(self.Env["valgrind-tests"]): startCmd = self["StartCmd"] else: if self.Env["valgrind-prefix"]: prefix = self.Env["valgrind-prefix"] else: prefix = "cts" startCmd = """G_SLICE=always-malloc HA_VALGRIND_ENABLED='%s' VALGRIND_OPTS='%s --log-file=/tmp/%s-%s.valgrind' %s""" % ( self.Env["valgrind-procs"], self.Env["valgrind-opts"], prefix, """%p""", self["StartCmd"]) self.rsh(node, startCmd, blocking=0) self.ShouldBeStatus[node]="up" return 1 def StopaCM(self, node): '''Stop the cluster manager on a given node''' self.debug("Stopping %s on node %s" %(self["Name"], node)) if self.ShouldBeStatus[node] != "up": return 1 if self.rsh(node, self["StopCmd"]) == 0: self.ShouldBeStatus[node]="down" self.cluster_stable(self["DeadTime"]) return 1 else: self.log ("Could not stop %s on node %s" %(self["Name"], node)) return None def StopaCMnoBlock(self, node): '''Stop the cluster manager on a given node with none-block mode''' self.debug("Stopping %s on node %s" %(self["Name"], node)) self.rsh(node, self["StopCmd"], blocking=0) self.ShouldBeStatus[node]="down" return 1 def cluster_stable(self, timeout = None): time.sleep(self["StableTime"]) return 1 def node_stable(self, node): return 1 def RereadCM(self, node): '''Force the cluster manager on a given node to reread its config This may be a no-op on certain cluster managers. ''' rc=self.rsh(node, self["RereadCmd"]) if rc == 0: return 1 else: self.log ("Could not force %s on node %s to reread its config" % (self["Name"], node)) return None def StataCM(self, node): '''Report the status of the cluster manager on a given node''' out=self.rsh(node, self["StatusCmd"], 1) ret= (string.find(out, 'stopped') == -1) try: if ret: if self.ShouldBeStatus[node] == "down": self.log( "Node status for %s is %s but we think it should be %s" % (node, "up", self.ShouldBeStatus[node])) else: if self.ShouldBeStatus[node] == "up": self.log( "Node status for %s is %s but we think it should be %s" % (node, "down", self.ShouldBeStatus[node])) except KeyError: pass if ret: self.ShouldBeStatus[node]="up" else: self.ShouldBeStatus[node]="down" return ret def startall(self, nodelist=None): '''Start the cluster manager on every node in the cluster. We can do it on a subset of the cluster if nodelist is not None. ''' ret = 1 map = {} if not nodelist: nodelist=self.Env["nodes"] for node in nodelist: if self.ShouldBeStatus[node] == "down": if not self.StartaCM(node): ret = 0 return ret def stopall(self, nodelist=None): '''Stop the cluster managers on every node in the cluster. We can do it on a subset of the cluster if nodelist is not None. ''' ret = 1 map = {} if not nodelist: nodelist=self.Env["nodes"] for node in self.Env["nodes"]: if self.ShouldBeStatus[node] == "up": if not self.StopaCM(node): 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"] % 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"] % node, blocking=0) self.rsh(node, self["FixCommCmd"] % target, blocking=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, name, dc_only, pats, dc_pats, badnews_ignore, triggersreboot, cm): self.name = str(name) self.dc_only = dc_only self.pats = pats self.dc_pats = dc_pats self.CM = cm self.badnews_ignore = badnews_ignore self.triggersreboot = triggersreboot self.KillCmd = "killall -9 " + self.name 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 class ScenarioComponent: def __init__(self, Env): self.Env = Env def IsApplicable(self): '''Return TRUE if the current ScenarioComponent is applicable in the given LabEnvironment given to the constructor. ''' raise ValueError("Abstract Class member (IsApplicable)") def SetUp(self, CM): '''Set up the given ScenarioComponent''' raise ValueError("Abstract Class member (Setup)") def TearDown(self, CM): '''Tear down (undo) the given ScenarioComponent''' raise ValueError("Abstract Class member (Setup)") class Scenario: ( '''The basic idea of a scenario is that of an ordered list of ScenarioComponent objects. Each ScenarioComponent is SetUp() in turn, and then after the tests have been run, they are torn down using TearDown() (in reverse order). A Scenario is applicable to a particular cluster manager iff each ScenarioComponent is applicable. A partially set up scenario is torn down if it fails during setup. ''') def __init__(self, Components): "Initialize the Scenario from the list of ScenarioComponents" for comp in Components: if not issubclass(comp.__class__, ScenarioComponent): raise ValueError("Init value must be subclass of" " ScenarioComponent") self.Components = Components def IsApplicable(self): ( '''A Scenario IsApplicable() iff each of its ScenarioComponents IsApplicable() ''' ) for comp in self.Components: if not comp.IsApplicable(): return None return 1 def SetUp(self, CM): '''Set up the Scenario. Return TRUE on success.''' j=0 while j < len(self.Components): if not self.Components[j].SetUp(CM): # OOPS! We failed. Tear partial setups down. CM.log("Tearing down partial setup") self.TearDown(CM, j) return None j=j+1 return 1 def TearDown(self, CM, max=None): '''Tear Down the Scenario - in reverse order.''' if max == None: max = len(self.Components)-1 j=max while j >= 0: self.Components[j].TearDown(CM) j=j-1 class InitClusterManager(ScenarioComponent): ( '''InitClusterManager is the most basic of ScenarioComponents. This ScenarioComponent simply starts the cluster manager on all the nodes. It is fairly robust as it waits for all nodes to come up before starting as they might have been rebooted or crashed for some reason beforehand. ''') def __init__(self, Env): pass def IsApplicable(self): '''InitClusterManager is so generic it is always Applicable''' return 1 def SetUp(self, CM): '''Basic Cluster Manager startup. Start everything''' CM.prepare() # Clear out the cobwebs ;-) self.TearDown(CM) # Now start the Cluster Manager on all the nodes. CM.log("Starting Cluster Manager on all nodes.") return CM.startall() def TearDown(self, CM): '''Set up the given ScenarioComponent''' # Stop the cluster manager everywhere CM.log("Stopping Cluster Manager on all nodes") return CM.stopall() class PingFest(ScenarioComponent): ( '''PingFest does a flood ping to each node in the cluster from the test machine. If the LabEnvironment Parameter PingSize is set, it will be used as the size of ping packet requested (via the -s option). If it is not set, it defaults to 1024 bytes. According to the manual page for ping: Outputs packets as fast as they come back or one hundred times per second, whichever is more. For every ECHO_REQUEST sent a period ``.'' is printed, while for every ECHO_REPLY received a backspace is printed. This provides a rapid display of how many packets are being dropped. Only the super-user may use this option. This can be very hard on a net- work and should be used with caution. ''' ) def __init__(self, Env): self.Env = Env def IsApplicable(self): '''PingFests are always applicable ;-) ''' return 1 def SetUp(self, CM): '''Start the PingFest!''' self.PingSize=1024 if CM.Env.has_key("PingSize"): self.PingSize=CM.Env["PingSize"] CM.log("Starting %d byte flood pings" % self.PingSize) self.PingPids=[] for node in CM.Env["nodes"]: self.PingPids.append(self._pingchild(node)) CM.log("Ping PIDs: " + repr(self.PingPids)) return 1 def TearDown(self, CM): '''Stop it right now! My ears are pinging!!''' for pid in self.PingPids: if pid != None: CM.log("Stopping ping process %d" % pid) os.kill(pid, signal.SIGKILL) def _pingchild(self, node): Args = ["ping", "-qfn", "-s", str(self.PingSize), node] sys.stdin.flush() sys.stdout.flush() sys.stderr.flush() pid = os.fork() if pid < 0: self.Env.log("Cannot fork ping child") return None if pid > 0: return pid # Otherwise, we're the child process. os.execvp("ping", Args) self.Env.log("Cannot execvp ping: " + repr(Args)) sys.exit(1) class PacketLoss(ScenarioComponent): ( ''' It would be useful to do some testing of CTS with a modest amount of packet loss enabled - so we could see that everything runs like it should with a certain amount of packet loss present. ''') def IsApplicable(self): '''always Applicable''' return 1 def SetUp(self, CM): '''Reduce the reliability of communications''' if float(CM.Env["XmitLoss"]) == 0 and float(CM.Env["RecvLoss"]) == 0 : return 1 for node in CM.Env["nodes"]: CM.reducecomm_node(node) CM.log("Reduce the reliability of communications") return 1 def TearDown(self, CM): '''Fix the reliability of communications''' if float(CM.Env["XmitLoss"]) == 0 and float(CM.Env["RecvLoss"]) == 0 : return 1 for node in CM.Env["nodes"]: CM.unisolate_node(node) CM.log("Fix the reliability of communications") class BasicSanityCheck(ScenarioComponent): ( ''' ''') def IsApplicable(self): return self.Env["DoBSC"] def SetUp(self, CM): CM.prepare() # Clear out the cobwebs self.TearDown(CM) # Now start the Cluster Manager on all the nodes. CM.log("Starting Cluster Manager on BSC node(s).") return CM.startall() def TearDown(self, CM): CM.log("Stopping Cluster Manager on BSC node(s).") return CM.stopall() class Benchmark(ScenarioComponent): ( ''' ''') def IsApplicable(self): return self.Env["benchmark"] def SetUp(self, CM): CM.prepare() # Clear out the cobwebs self.TearDown(CM) # Now start the Cluster Manager on all the nodes. CM.log("Starting Cluster Manager on all node(s).") return CM.startall() def TearDown(self, CM): CM.log("Stopping Cluster Manager on all node(s).") return CM.stopall() class RollingUpgrade(ScenarioComponent): ( ''' Test a rolling upgrade between two versions of the stack ''') def __init__(self, Env): self.Env = Env def IsApplicable(self): if not self.Env["rpm-dir"]: return None if not self.Env["current-version"]: return None if not self.Env["previous-version"]: return None return 1 def install(self, node, version): target_dir = "/tmp/rpm-%s" % version src_dir = "%s/%s" % (self.CM.Env["rpm-dir"], version) rc = self.CM.rsh(node, "mkdir -p %s" % target_dir) rc = self.CM.cp("%s/*.rpm %s:%s" % (src_dir, node, target_dir)) rc = self.CM.rsh(node, "rpm -Uvh --force %s/*.rpm" % (target_dir)) return self.success() def upgrade(self, node): return self.install(node, self.CM.Env["current-version"]) def downgrade(self, node): return self.install(node, self.CM.Env["previous-version"]) def SetUp(self, CM): CM.prepare() # Clear out the cobwebs CM.stopall() CM.log("Downgrading all nodes to %s." % self.Env["previous-version"]) for node in self.Env["nodes"]: if not self.downgrade(node): CM.log("Couldn't downgrade %s" % node) return None return 1 def TearDown(self, CM): # Stop everything CM.log("Stopping Cluster Manager on Upgrade nodes.") CM.stopall() CM.log("Upgrading all nodes to %s." % self.Env["current-version"]) for node in self.Env["nodes"]: if not self.upgrade(node): CM.log("Couldn't upgrade %s" % node) return None return 1 diff --git a/cts/CTSaudits.py b/cts/CTSaudits.py index cb9f7cd7fe..18dd6d2d10 100755 --- a/cts/CTSaudits.py +++ b/cts/CTSaudits.py @@ -1,777 +1,782 @@ '''CTS: Cluster Testing System: Audit module ''' __copyright__=''' Copyright (C) 2000, 2001,2005 Alan Robertson Licensed under the GNU GPL. ''' # # This program is free software; you can redistribute it and/or # modify it under the terms of the GNU General Public License # as published by the Free Software Foundation; either version 2 # of the License, or (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program; if not, write to the Free Software # Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. import time, os, string, re import CTS class ClusterAudit: def __init__(self, cm): self.CM = cm def __call__(self): raise ValueError("Abstract Class member (__call__)") def is_applicable(self): '''Return TRUE if we are applicable in the current test configuration''' raise ValueError("Abstract Class member (is_applicable)") return 1 def log(self, args): self.CM.log("audit: %s" % args) def debug(self, args): self.CM.debug("audit: %s" % args) def name(self): raise ValueError("Abstract Class member (name)") AllAuditClasses = [ ] class LogAudit(ClusterAudit): def name(self): return "LogAudit" def __init__(self, cm): self.CM = cm def RestartClusterLogging(self, nodes=None): if not nodes: nodes = self.CM.Env["nodes"] self.CM.log("Restarting logging on: %s" % repr(nodes)) for node in nodes: cmd=self.CM.Env["logrestartcmd"] if self.CM.rsh(node, cmd, blocking=0) != 0: self.CM.log ("ERROR: Cannot restart logging on %s [%s failed]" % (node, cmd)) def TestLogging(self): patterns = [] prefix = "Test message from" watch_syslog = None watch_remote = None for node in self.CM.Env["nodes"]: # Look for the node name in two places to make sure # that syslog is logging with the correct hostname patterns.append("%s.*%s %s" % (node, prefix, node)) - if self.Env["LogWatcher"] == "any" or self.Env["LogWatcher"] == "syslog": - self.Env["LogWatcher"] = "syslog" - watch_syslog = CTS.LogWatcher(self.CM.Env, self.CM.Env["LogFileName"], patterns, "LogAudit", 60) + watch_pref = self.CM.Env["LogWatcher"] + if watch_pref == "any" or watch_pref == "syslog": + self.CM.Env["LogWatcher"] = "syslog" + self.CM.log("Testing for %s logs" % self.CM.Env["LogWatcher"]) + watch_syslog = CTS.LogWatcher(self.CM.Env, self.CM.Env["LogFileName"], patterns, "LogAudit", 30) watch_syslog.setwatch() - if self.Env["LogWatcher"] == "any" or self.Env["LogWatcher"] == "remote": - self.Env["LogWatcher"] = "remote" - watch_remote = CTS.LogWatcher(self.CM.Env, self.CM.Env["LogFileName"], patterns, "LogAudit", 60) + if watch_pref == "any" or watch_pref == "remote": + self.CM.Env["LogWatcher"] = "remote" + self.CM.log("Testing for %s logs" % self.CM.Env["LogWatcher"]) + watch_remote = CTS.LogWatcher(self.CM.Env, self.CM.Env["LogFileName"], patterns, "LogAudit", 30) watch_remote.setwatch() for node in self.CM.Env["nodes"]: cmd="logger -p %s.info %s %s" % (self.CM.Env["SyslogFacility"], prefix, node) if self.CM.rsh(node, cmd, blocking=0) != 0: self.CM.log ("ERROR: Cannot execute remote command [%s] on %s" % (cmd, node)) if watch_syslog: watch = watch_syslog - self.Env["LogWatcher"] = "syslog" + self.CM.Env["LogWatcher"] = "syslog" watch_result = watch.lookforall() - if watch.unmatched: - for regex in watch.unmatched: - self.CM.log ("Test message [%s] not found in %s logs." % (regex, self.Env["LogWatcher"])) - else: - self.CM.log ("Continuing with %s-based log reader" % (self.Env["LogWatcher"])) + if not watch.unmatched: + self.CM.log ("Continuing with %s-based log reader" % (self.CM.Env["LogWatcher"])) return 1 if watch_remote: watch = watch_remote - self.Env["LogWatcher"] = "remote" + self.CM.Env["LogWatcher"] = "remote" watch_result = watch.lookforall() - if watch.unmatched: - for regex in watch.unmatched: - self.CM.log ("Test message [%s] not found in %s logs." % (regex, self.Env["LogWatcher"])) - else: - self.CM.log ("Continuing with %s-based log reader" % (self.Env["LogWatcher"])) + if not watch.unmatched: + self.CM.log ("Continuing with %s-based log reader" % (self.CM.Env["LogWatcher"])) return 1 + if watch_syslog and not watch_syslog.unmatched: + for regex in watch_syslog.unmatched: + self.CM.log ("Test message [%s] not found in syslog logs." % regex) + + if watch_remote and not watch_remote.unmatched: + for regex in watch_remote.unmatched: + self.CM.log ("Test message [%s] not found in remote logs." % regex) + return 0 def __call__(self): max=3 attempt=0 self.CM.ns.WaitForAllNodesToComeUp(self.CM.Env["nodes"]) while attempt <= max and self.TestLogging() == 0: attempt = attempt + 1 self.RestartClusterLogging() time.sleep(60*attempt) if attempt > max: self.CM.log("ERROR: Cluster logging unrecoverable.") return 0 return 1 def is_applicable(self): if self.CM.Env["DoBSC"]: return 0 return 1 class DiskAudit(ClusterAudit): def name(self): return "DiskspaceAudit" def __init__(self, cm): self.CM = cm def __call__(self): result=1 dfcmd="df -k /var/log | tail -1 | tr -s ' ' | cut -d' ' -f2" self.CM.ns.WaitForAllNodesToComeUp(self.CM.Env["nodes"]) for node in self.CM.Env["nodes"]: dfout=self.CM.rsh(node, dfcmd, 1) if not dfout: self.CM.log ("ERROR: Cannot execute remote df command [%s] on %s" % (dfcmd, node)) else: try: idfout = int(dfout) except (ValueError, TypeError): self.CM.log("Warning: df output from %s was invalid [%s]" % (node, dfout)) else: if idfout == 0: self.CM.log("CRIT: Completely out of log disk space on %s" % node) result=None elif idfout <= 1000: self.CM.log("WARN: Low on log disk space (%d Mbytes) on %s" % (idfout, node)) return result def is_applicable(self): if self.CM.Env["DoBSC"]: return 0 return 1 class AuditResource: def __init__(self, cm, line): fields = line.split() self.CM = cm self.line = line self.type = fields[1] self.id = fields[2] self.clone_id = fields[3] self.parent = fields[4] self.rprovider = fields[5] self.rclass = fields[6] self.rtype = fields[7] self.host = fields[8] self.needs_quorum = fields[9] self.flags = int(fields[10]) self.flags_s = fields[11] if self.parent == "NA": self.parent = None def unique(self): if self.flags & int("0x00000020", 16): return 1 return 0 def orphan(self): if self.flags & int("0x00000001", 16): return 1 return 0 def managed(self): if self.flags & int("0x00000002", 16): return 1 return 0 class AuditConstraint: def __init__(self, cm, line): fields = line.split() self.CM = cm self.line = line self.type = fields[1] self.id = fields[2] self.rsc = fields[3] self.target = fields[4] self.score = fields[5] self.rsc_role = fields[6] self.target_role = fields[7] if self.rsc_role == "NA": self.rsc_role = None if self.target_role == "NA": self.target_role = None class PrimitiveAudit(ClusterAudit): def name(self): return "PrimitiveAudit" def __init__(self, cm): self.CM = cm def doResourceAudit(self, resource): rc=1 active = self.CM.ResourceLocation(resource.id) if len(active) == 1: if self.CM.HasQuorum(None): self.debug("Resource %s active on %s" % (resource.id, repr(active))) elif resource.needs_quorum == 1: self.CM.log("Resource %s active without quorum: %s" % (resource.id, repr(active))) rc=0 elif not resource.managed(): self.CM.log("Resource %s not managed. Active on %s" % (resource.id, repr(active))) elif not resource.unique(): # TODO: Figure out a clever way to actually audit these resource types if len(active) > 1: self.debug("Non-unique resource %s is active on: %s" % (resource.id, repr(active))) else: self.debug("Non-unique resource %s is not active" % resource.id) elif len(active) > 1: self.CM.log("Resource %s is active multiple times: %s" % (resource.id, repr(active))) rc=0 elif resource.orphan(): self.debug("Resource %s is an inactive orphan" % resource.id) elif len(self.inactive_nodes) == 0: self.CM.log("WARN: Resource %s not served anywhere" % resource.id) rc=0 elif self.CM.Env["warn-inactive"] == 1: if self.CM.HasQuorum(None) or not resource.needs_quorum: self.CM.log("WARN: Resource %s not served anywhere (Inactive nodes: %s)" % (resource.id, repr(self.inactive_nodes))) else: self.debug("Resource %s not served anywhere (Inactive nodes: %s)" % (resource.id, repr(self.inactive_nodes))) elif self.CM.HasQuorum(None) or not resource.needs_quorum: self.debug("Resource %s not served anywhere (Inactive nodes: %s)" % (resource.id, repr(self.inactive_nodes))) return rc def setup(self): self.target = None self.resources = [] self.constraints = [] self.active_nodes = [] self.inactive_nodes = [] self.debug("Do Audit %s"%self.name()) for node in self.CM.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.active_nodes.append(node) else: self.inactive_nodes.append(node) for node in self.CM.Env["nodes"]: if self.target == None and self.CM.ShouldBeStatus[node] == "up": self.target = node if not self.target: # TODO: In Pacemaker 1.0 clusters we'll be able to run crm_resource # with CIB_file=/path/to/cib.xml even when the cluster isn't running self.debug("No nodes active - skipping %s" % self.name()) return 0 (rc, lines) = self.CM.rsh(self.target, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): self.resources.append(AuditResource(self.CM, line)) elif re.search("^Constraint", line): self.constraints.append(AuditConstraint(self.CM, line)) else: self.CM.log("Unknown entry: %s" % line); return 1 def __call__(self): rc = 1 if not self.setup(): return 1 for resource in self.resources: if resource.type == "primitive": if self.doResourceAudit(resource) == 0: rc = 0 return rc def is_applicable(self): if self.CM["Name"] == "crm-lha": return 1 if self.CM["Name"] == "crm-ais": return 1 return 0 class GroupAudit(PrimitiveAudit): def name(self): return "GroupAudit" def __call__(self): rc = 1 if not self.setup(): return 1 for group in self.resources: if group.type == "group": first_match = 1 group_location = None for child in self.resources: if child.parent == group.id: nodes = self.CM.ResourceLocation(child.id) if first_match and len(nodes) > 0: group_location = nodes[0] first_match = 0 if len(nodes) > 1: rc = 0 self.CM.log("Child %s of %s is active more than once: %s" % (child.id, group.id, repr(nodes))) elif len(nodes) == 0: # Groups are allowed to be partially active # However we do need to make sure later children aren't running group_location = None self.debug("Child %s of %s is stopped" % (child.id, group.id)) elif nodes[0] != group_location: rc = 0 self.CM.log("Child %s of %s is active on the wrong node (%s) expected %s" % (child.id, group.id, nodes[0], group_location)) else: self.debug("Child %s of %s is active on %s" % (child.id, group.id, nodes[0])) return rc class CloneAudit(PrimitiveAudit): def name(self): return "CloneAudit" def __call__(self): rc = 1 if not self.setup(): return 1 for clone in self.resources: if clone.type == "clone": for child in self.resources: if child.parent == clone.id and child.type == "primitive": self.debug("Checking child %s of %s..." % (child.id, clone.id)) # Check max and node_max # Obtain with: # crm_resource -g clone_max --meta -r child.id # crm_resource -g clone_node_max --meta -r child.id return rc class ColocationAudit(PrimitiveAudit): def name(self): return "ColocationAudit" def crm_location(self, resource): (rc, lines) = self.CM.rsh(self.target, "crm_resource -W -r %s -Q"%resource, None) hosts = [] if rc == 0: for line in lines: fields = line.split() hosts.append(fields[0]) return hosts def __call__(self): rc = 1 if not self.setup(): return 1 for coloc in self.constraints: if coloc.type == "rsc_colocation": source = self.crm_location(coloc.rsc) target = self.crm_location(coloc.target) if len(source) == 0: self.debug("Colocation audit (%s): %s not running" % (coloc.id, coloc.rsc)) else: for node in source: if not node in target: rc = 0 self.CM.log("Colocation audit (%s): %s running on %s (not in %s)" % (coloc.id, coloc.rsc, node, repr(target))) else: self.debug("Colocation audit (%s): %s running on %s (in %s)" % (coloc.id, coloc.rsc, node, repr(target))) return rc class CrmdStateAudit(ClusterAudit): def __init__(self, cm): self.CM = cm self.Stats = {"calls":0 , "success":0 , "failure":0 , "skipped":0 , "auditfail":0} def has_key(self, key): return self.Stats.has_key(key) def __setitem__(self, key, value): self.Stats[key] = value def __getitem__(self, key): return self.Stats[key] def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not self.Stats.has_key(name): self.Stats[name]=0 self.Stats[name] = self.Stats[name]+1 def __call__(self): passed = 1 up_are_down = 0 down_are_up = 0 unstable_list = [] self.debug("Do Audit %s"%self.name()) for node in self.CM.Env["nodes"]: should_be = self.CM.ShouldBeStatus[node] rc = self.CM.test_node_CM(node) if rc > 0: if should_be == "down": down_are_up = down_are_up + 1 if rc == 1: unstable_list.append(node) elif should_be == "up": up_are_down = up_are_down + 1 if len(unstable_list) > 0: passed = 0 self.CM.log("Cluster is not stable: %d (of %d): %s" %(len(unstable_list), self.CM.upcount(), repr(unstable_list))) if up_are_down > 0: passed = 0 self.CM.log("%d (of %d) nodes expected to be up were down." %(up_are_down, len(self.CM.Env["nodes"]))) if down_are_up > 0: passed = 0 self.CM.log("%d (of %d) nodes expected to be down were up." %(down_are_up, len(self.CM.Env["nodes"]))) return passed def name(self): return "CrmdStateAudit" def is_applicable(self): if self.CM["Name"] == "crm-lha": return 1 if self.CM["Name"] == "crm-ais": return 1 return 0 class CIBAudit(ClusterAudit): def __init__(self, cm): self.CM = cm self.Stats = {"calls":0 , "success":0 , "failure":0 , "skipped":0 , "auditfail":0} def has_key(self, key): return self.Stats.has_key(key) def __setitem__(self, key, value): self.Stats[key] = value def __getitem__(self, key): return self.Stats[key] def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not self.Stats.has_key(name): self.Stats[name]=0 self.Stats[name] = self.Stats[name]+1 def __call__(self): self.debug("Do Audit %s"%self.name()) passed = 1 ccm_partitions = self.CM.find_partitions() if len(ccm_partitions) == 0: self.debug("\tNo partitions to audit") return 1 for partition in ccm_partitions: self.debug("\tAuditing CIB consistency for: %s" %partition) partition_passed = 0 if self.audit_cib_contents(partition) == 0: passed = 0 return passed def audit_cib_contents(self, hostlist): passed = 1 node0 = None node0_xml = None partition_hosts = hostlist.split() for node in partition_hosts: node_xml = self.store_remote_cib(node, node0) if node_xml == None: self.CM.log("Could not perform audit: No configuration from %s" % node) passed = 0 elif node0 == None: node0 = node node0_xml = node_xml elif node0_xml == None: self.CM.log("Could not perform audit: No configuration from %s" % node0) passed = 0 else: (rc, result) = self.CM.rsh( node0, "crm_diff -VV -cf --new %s --original %s" % (node_xml, node0_xml), None) if rc != 0: self.CM.log("Diff between %s and %s failed: %d" % (node0_xml, node_xml, rc)) passed = 0 for line in result: if not re.search("", line): passed = 0 self.debug("CibDiff[%s-%s]: %s" % (node0, node, line)) else: self.debug("CibDiff[%s-%s] Ignoring: %s" % (node0, node, line)) # self.CM.rsh(node0, "rm -f %s" % node_xml) # self.CM.rsh(node0, "rm -f %s" % node0_xml) return passed def store_remote_cib(self, node, target): combined = "" filename="/tmp/ctsaudit.%s.xml" % node if not target: target = node (rc, lines) = self.CM.rsh(node, self.CM["CibQuery"], None) if rc != 0: self.CM.log("Could not retrieve configuration") return None self.CM.rsh("localhost", "rm -f %s" % filename) for line in lines: self.CM.rsh("localhost", "echo \'%s\' >> %s" % (line[:-1], filename)) if self.CM.rsh.cp(filename, "root@%s:%s" % (target, filename)) != 0: self.CM.log("Could not store configuration") return None return filename def name(self): return "CibAudit" def is_applicable(self): if self.CM["Name"] == "crm-lha": return 1 if self.CM["Name"] == "crm-ais": return 1 return 0 class PartitionAudit(ClusterAudit): def __init__(self, cm): self.CM = cm self.Stats = {"calls":0 , "success":0 , "failure":0 , "skipped":0 , "auditfail":0} self.NodeEpoche={} self.NodeState={} self.NodeQuorum={} def has_key(self, key): return self.Stats.has_key(key) def __setitem__(self, key, value): self.Stats[key] = value def __getitem__(self, key): return self.Stats[key] def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not self.Stats.has_key(name): self.Stats[name]=0 self.Stats[name] = self.Stats[name]+1 def __call__(self): self.debug("Do Audit %s"%self.name()) passed = 1 ccm_partitions = self.CM.find_partitions() if ccm_partitions == None or len(ccm_partitions) == 0: return 1 if len(ccm_partitions) != self.CM.partitions_expected: self.CM.log("ERROR: %d cluster partitions detected:" %len(ccm_partitions)) passed = 0 for partition in ccm_partitions: self.CM.log("\t %s" %partition) for partition in ccm_partitions: partition_passed = 0 if self.audit_partition(partition) == 0: passed = 0 return passed def trim_string(self, avalue): if not avalue: return None if len(avalue) > 1: return avalue[:-1] def trim2int(self, avalue): if not avalue: return None if len(avalue) > 1: return int(avalue[:-1]) def audit_partition(self, partition): passed = 1 dc_found = [] dc_allowed_list = [] lowest_epoche = None node_list = partition.split() self.debug("Auditing partition: %s" %(partition)) for node in node_list: if self.CM.ShouldBeStatus[node] != "up": self.CM.log("Warn: Node %s appeared out of nowhere" %(node)) self.CM.ShouldBeStatus[node] = "up" # not in itself a reason to fail the audit (not what we're # checking for in this audit) self.NodeState[node] = self.CM.rsh(node, self.CM["StatusCmd"]%node, 1) self.NodeEpoche[node] = self.CM.rsh(node, self.CM["EpocheCmd"], 1) self.NodeQuorum[node] = self.CM.rsh(node, self.CM["QuorumCmd"], 1) self.debug("Node %s: %s - %s - %s." %(node, self.NodeState[node], self.NodeEpoche[node], self.NodeQuorum[node])) self.NodeState[node] = self.trim_string(self.NodeState[node]) self.NodeEpoche[node] = self.trim2int(self.NodeEpoche[node]) self.NodeQuorum[node] = self.trim_string(self.NodeQuorum[node]) if not self.NodeEpoche[node]: self.CM.log("Warn: Node %s dissappeared: cant determin epoche" %(node)) self.CM.ShouldBeStatus[node] = "down" # not in itself a reason to fail the audit (not what we're # checking for in this audit) elif lowest_epoche == None or self.NodeEpoche[node] < lowest_epoche: lowest_epoche = self.NodeEpoche[node] if not lowest_epoche: self.CM.log("Lowest epoche not determined in %s" % (partition)) passed = 0 for node in node_list: if self.CM.ShouldBeStatus[node] == "up": if self.CM.is_node_dc(node, self.NodeState[node]): dc_found.append(node) if self.NodeEpoche[node] == lowest_epoche: self.debug("%s: OK" % node) elif not self.NodeEpoche[node]: self.debug("Check on %s ignored: no node epoche" % node) elif not lowest_epoche: self.debug("Check on %s ignored: no lowest epoche" % node) else: self.CM.log("DC %s is not the oldest node (%d vs. %d)" %(node, self.NodeEpoche[node], lowest_epoche)) passed = 0 if len(dc_found) == 0: self.CM.log("DC not found on any of the %d allowed nodes: %s (of %s)" %(len(dc_allowed_list), str(dc_allowed_list), str(node_list))) elif len(dc_found) > 1: self.CM.log("%d DCs (%s) found in cluster partition: %s" %(len(dc_found), str(dc_found), str(node_list))) passed = 0 if passed == 0: for node in node_list: if self.CM.ShouldBeStatus[node] == "up": self.CM.log("epoche %s : %s" %(self.NodeEpoche[node], self.NodeState[node])) return passed def name(self): return "PartitionAudit" def is_applicable(self): if self.CM["Name"] == "crm-lha": return 1 if self.CM["Name"] == "crm-ais": return 1 return 0 AllAuditClasses.append(DiskAudit) AllAuditClasses.append(LogAudit) AllAuditClasses.append(CrmdStateAudit) AllAuditClasses.append(PartitionAudit) AllAuditClasses.append(PrimitiveAudit) AllAuditClasses.append(GroupAudit) AllAuditClasses.append(CloneAudit) AllAuditClasses.append(ColocationAudit) AllAuditClasses.append(CIBAudit) def AuditList(cm): result = [] for auditclass in AllAuditClasses: result.append(auditclass(cm)) return result diff --git a/cts/CTStests.py b/cts/CTStests.py index 0551784eeb..9339186d87 100644 --- a/cts/CTStests.py +++ b/cts/CTStests.py @@ -1,2411 +1,2411 @@ '''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., 59 Temple Place - Suite 330, Boston, MA 02111-1307, 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 CTS import CTSaudits import time, os, re, types, string, tempfile, sys from CTSaudits import * from stat import * # List of all class objects for tests which we ought to # consider running. class AllTests: ''' A collection of tests which are run at random. ''' def __init__(self, scenario, cm, tests, Audits): self.CM = cm self.Env = cm.Env self.Scenario = scenario self.Tests = [] self.Audits = [] self.ns=CTS.NodeStatus(self.Env) self.Stats = {"success":0, "failure":0, "BadNews":0, "skipped":0} self.IndividualStats= {} for audit in Audits: if not issubclass(audit.__class__, ClusterAudit): raise ValueError("Init value must be a subclass of ClusterAudit") if audit.is_applicable(): self.Audits.append(audit) for test in tests: if not issubclass(test.__class__, CTSTest): raise ValueError("Init value must be a subclass of CTSTest") if test.is_applicable(): self.Tests.append(test) if not scenario.IsApplicable(): raise ValueError("Scenario not applicable in" " given Environment") def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not self.Stats.has_key(name): self.Stats[name]=0 self.Stats[name] = self.Stats[name]+1 def audit(self, BadNews, test): errcount=0 BadNewsDebug=0 #BadNews.debug=1 ignorelist = [] ignorelist.append(" CTS: ") ignorelist.append("BadNews:") ignorelist.extend(self.CM.errorstoignore()) if test: ignorelist.extend(test.errorstoignore()) while errcount < 1000: if BadNewsDebug: print "Looking for BadNews" match=BadNews.look(0) if match: if BadNewsDebug: print "BadNews found: "+match add_err = 1 for ignore in ignorelist: if add_err == 1 and re.search(ignore, match): if BadNewsDebug: print "Ignoring based on pattern: ("+ignore+")" add_err = 0 if add_err == 1: self.CM.log("BadNews: " + match) self.incr("BadNews") errcount=errcount+1 else: break else: answer = raw_input('Big problems. Continue? [nY]') if answer and answer == "n": self.CM.log("Shutting down.") self.CM.stopall() self.summarize() raise ValueError("Looks like we hit a BadNews jackpot!") for audit in self.Audits: if not audit(): self.CM.log("Audit " + audit.name() + " FAILED.") self.incr("auditfail") if test: test.incr("auditfail") def summarize(self): self.CM.log("****************") self.CM.log("Overall Results:" + repr(self.Stats)) self.CM.log("****************") stat_filter = { "calls":0, "failure":0, "skipped":0, "auditfail":0, } self.CM.log("Test Summary") for test in self.Tests: for key in stat_filter.keys(): stat_filter[key] = test.Stats[key] self.CM.log(("Test %s: "%test.name).ljust(25) + " %s"%repr(stat_filter)) self.CM.debug("Detailed Results") for test in self.Tests: self.CM.debug(("Test %s: "%test.name).ljust(25) + " %s"%repr(test.Stats)) self.CM.log("<<<<<<<<<<<<<<<< TESTS COMPLETED") def test_loop(self, BadNews, max): testcount=1 self.CM.log("Executing all tests once") for test in self.Tests: if self.run_test(BadNews, test, testcount): testcount += 1 return testcount def run_test(self, BadNews, test, testcount): nodechoice = self.Env.RandomNode() ret = 1 where = "" did_run = 0 self.CM.log(("Running test %s" % test.name).ljust(35) + (" (%s) " % nodechoice).ljust(15) +"["+ ("%d" % testcount).rjust(3) +"]") starttime = test.set_timer() if not test.setup(nodechoice): self.CM.log("Setup failed") ret = 0 elif not test.canrunnow(nodechoice): self.CM.log("Skipped") test.skipped() else: did_run = 1 ret = test(nodechoice) if not test.teardown(nodechoice): self.CM.log("Teardown failed") ret = 0 stoptime=time.time() self.CM.oprofileSave(testcount) elapsed_time = stoptime - starttime test_time = stoptime - test.get_timer() if not test.has_key("min_time"): test["elapsed_time"] = elapsed_time test["min_time"] = test_time test["max_time"] = test_time else: test["elapsed_time"] = test["elapsed_time"] + elapsed_time if test_time < test["min_time"]: test["min_time"] = test_time if test_time > test["max_time"]: test["max_time"] = test_time if ret: self.incr("success") test.log_timer() else: self.incr("failure") self.CM.statall() did_run = 1 # Force the test count to be incrimented anyway so test extraction works self.audit(BadNews, test) return did_run def run(self, max=1): ( ''' Set up the given scenario, then run the selected tests at random for the selected number of iterations. ''') BadNews=CTS.LogWatcher(self.CM.Env, self.CM["LogFileName"], self.CM["BadRegexes"], "BadNews", 0) - BadNews.setwatch() self.CM.ns.WaitForAllNodesToComeUp(self.CM.Env["nodes"]) self.CM.oprofileStop() self.CM.oprofileStart() if not self.CM.Env["DoBSC"]: audit = LogAudit(self.CM) if not audit(): self.CM.log("Audit " + audit.name() + " FAILED.") return (None, None) else: self.CM.log("Audit " + audit.name() + " passed.") audit = DiskAudit(self.CM) if not audit(): self.CM.log("Audit " + audit.name() + " FAILED.") return (None, None) else: self.CM.log("Audit " + audit.name() + " passed.") + BadNews.setwatch() # Call after we've figured out what type of log watching to do in LogAudit if not self.Scenario.SetUp(self.CM): return (None, None) self.CM.oprofileSave(0) time.sleep(30) # This makes sure everything is stabilized before starting... self.audit(BadNews, None) testcount = self.test_loop(BadNews, max) self.Scenario.TearDown(self.CM) self.CM.oprofileSave(testcount) self.CM.oprofileStop() self.audit(BadNews, None) for test in self.Tests: self.IndividualStats[test.name] = test.Stats return self.Stats, self.IndividualStats class RandomTests(AllTests): def test_loop(self, BadNews, max): testcount=1 self.CM.log("Executing tests at random") while testcount <= max: test = self.Env.RandomGen.choice(self.Tests) if self.run_test(BadNews, test, testcount): testcount += 1 return testcount class BenchTests(AllTests): ''' Nothing (yet) here. ''' 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.Audits = [] self.timeout=120 self.passed = 1 self.is_loop = 0 self.is_unsafe = 0 self.is_experimental = 0 self.is_valgrind = 0 self.benchmark = 0 # which tests to benchmark self.timer = {} # timers 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.CM.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.CM.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.CM.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.CM.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.CM.Env, self.CM["LogFileName"], 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.CM.log(prefix + " " + match) errcount=errcount+1 else: break else: self.CM.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.CM.Env["loop-tests"]: return 0 elif self.is_unsafe and not self.CM.Env["unsafe-tests"]: return 0 elif self.is_valgrind and not self.CM.Env["valgrind-tests"]: return 0 elif self.is_experimental and not self.CM.Env["experimental-tests"]: return 0 return 1 def find_ocfs2_resources(self, node): self.r_o2cb = None self.r_ocfs2 = [] (rc, lines) = self.CM.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.CM.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.CM.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.CM["Pat:We_stopped"] % node) #if self.CM.Env["use_logd"]: # patterns.append(self.CM["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.CM.Env["nodes"]: if self.CM.ShouldBeStatus[other] == "up" and other != node: patterns.append(self.CM["Pat:They_stopped"] %(other, node)) #self.debug("Checking %s will notice %s left"%(other, node)) watch = self.create_watch(patterns, self.CM["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.CM.rsh(node, "/bin/ps axf", None) for line in output: self.CM.debug(line) for regex in watch.unmatched: self.CM.log ("ERROR: Shutdown pattern not found: %s" % (regex)) UnmatchedList += regex + "||"; failreason="Missing shutdown pattern" self.CM.cluster_stable(self.CM["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.CM["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.CM["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.CM.Env["nodes"]) < 2: return self.skipped() ret = self.startall(None) if not ret: return self.failure("Setup failed") watchpats = [] watchpats.append("Forcing node %s to be terminated" % node) watchpats.append("Scheduling Node %s for STONITH" % node) watchpats.append("Executing .* fencing operation") watchpats.append("sending fencing op RESET for %s" % node) if not self.CM.is_node_dc(node): # Won't be found if the DC is shot (and there's no equivalent message from stonithd) watchpats.append("tengine_stonith_callback: .*result=0") if self.CM.Env["at-boot"] == 0: self.CM.debug("Expecting %s to stay down" % node) self.CM.ShouldBeStatus[node]="down" else: self.CM.debug("Expecting %s to come up again %d" % (node, self.CM.Env["at-boot"])) watchpats.append("%s crmd: .* S_STARTING -> S_PENDING" % node) watchpats.append("%s crmd: .* S_PENDING -> S_NOT_DC" % node) watch = self.create_watch(watchpats, self.CM["DeadTime"] + self.CM["StableTime"] + self.CM["StartTime"]) watch.setwatch() self.CM.rsh(node, "crm_attribute --node %s --type status --attr-name terminate --attr-value true" % node) self.set_timer("fence") matched = watch.lookforall() self.log_timer("fence") self.set_timer("reform") if matched: self.CM.debug("Found: "+ repr(matched)) else: self.CM.log("Patterns not found: " + repr(watch.unmatched)) self.CM.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.CM.debug("Waiting STONITHd node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.CM.Env["nodes"], 600) self.CM.debug("Waiting for the cluster to re-stabilize with all nodes") is_stable = self.CM.cluster_stable(self.CM["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 [ "Executing .* fencing operation" ] def is_applicable(self): if not self.is_applicable_common(): return 0 if self.CM.Env.has_key("DoStonith"): return self.CM.Env["DoStonith"] 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.CM.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.CM.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.CM.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.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("Starting crmd") watch = self.create_watch(watchpats, self.CM["DeadTime"]+10) watch.setwatch() self.CM.StartaCMnoBlock(node) ret = watch.lookforall() if not ret: self.CM.log("Patterns not found: " + repr(watch.unmatched)) return self.failure("Setup of %s failed" % node) ret = self.stopall(None) if not ret: return self.failure("%s did not stop in time" % node) return self.success() # 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.CM.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.CM.debug("Getting resources running on node %s" % node) rsc_on_node = self.CM.active_resources(node) self.CM.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") time.sleep(1) # Allow time for the update to be applied and cause something 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.CM.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.CM.debug("Setting node %s to active mode" % node) self.CM.SetStandbyMode(node, "off") return rc self.CM.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.CM.Env["valgrind-prefix"] = self.name self.CM.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.CM.Env["nodes"]: (rc, output) = self.CM.rsh(node, "ps u --ppid `pidofproc aisexec`", None) for line in output: self.CM.debug(line) return self.success() def teardown(self, node): # Disable valgrind self.CM.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.CM.Env["nodes"]: (rc, ps_out) = self.CM.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.CM.rsh(node, "grep -e indirectly.*lost:.*[1-9] -e definitely.*lost:.*[1-9] -e 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.CM.log(line) (rc, output) = self.CM.rsh(node, "grep -e lost: -e SUMMARY: %s" % self.logPat, None) for line in output: self.CM.log(line) (rc, output) = self.CM.rsh(node, "cat %s" % self.logPat, None) for line in output: self.CM.debug(line) self.CM.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.CM.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.CM.rsh(node, dumpcmd) if rc == 0: farfile = "root@%s:%s" % (node, fstmpfile) self.CM.rsh.cp(farfile, self.tempfile) Bandwidth = self.countbandwidth(self.tempfile) if not Bandwidth: self.CM.log("Could not compute bandwidth.") return self.success() intband = int(Bandwidth + 0.5) self.CM.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.CM.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.CM.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.CM.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 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.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.CM.log("No active resources on %s" % node) return self.skipped() self.rid = self.CM.Env.RandomGen.choice(resourcelist) rsc = None (rc, lines) = self.CM.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.CM.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("crmd:.* Performing .* op=%s_stop_0" % self.rid) if rsc.unique(): pats.append("crmd:.* Performing .* op=%s_start_0" % self.rid) pats.append("crmd:.* LRM operation %s_start_0.*confirmed.*ok" % self.rid) else: # Anonymous clones may get restarted with a different clone number pats.append("crmd:.* Performing .* op=.*_start_0") pats.append("crmd:.* LRM operation .*_start_0.*confirmed.*ok") watch = self.create_watch(pats, 60) watch.setwatch() self.CM.rsh(node, "crm_resource -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.CM.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, """Unknown operation: fail""", """ERROR: sending stonithRA op to stonithd failed.""", """ERROR: process_lrm_event: LRM operation %s_%s_%d""" % (self.rid, self.action, self.interval), """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.CM["StableTime"]): return self.failure("Setup failed - unstable") node_is_dc = self.CM.is_node_dc(node, None) # select a component to kill chosen = self.CM.Env.RandomGen.choice(self.complist) while chosen.dc_only == 1 and node_is_dc == 0: chosen = self.CM.Env.RandomGen.choice(self.complist) self.CM.debug("...component %s (dc=%d,boot=%d)" % (chosen.name, node_is_dc,chosen.triggersreboot)) self.incr(chosen.name) if chosen.name != "aisexec": if self.CM["Name"] != "crm-lha" or chosen.name != "pengine": self.patterns.append(self.CM["Pat:ChildKilled"] %(node, chosen.name)) self.patterns.append(self.CM["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.CM["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.CM.Env["nodes"]: if other != node: self.patterns.append(self.CM["Pat:They_stopped"] %(other, node)) self.patterns.append(self.CM["Pat:Slave_started"] % node) self.patterns.append(self.CM["Pat:Local_started"] % node) if chosen.dc_only: # Sometimes these will be in the log, and sometimes they won't... self.okerrpatterns.append("%s crmd:.*Process %s:.* exited" %(node, chosen.name)) self.okerrpatterns.append("%s crmd:.*I_ERROR.*crmdManagedChildDied" %node) self.okerrpatterns.append("%s crmd:.*The %s subsystem terminated unexpectedly" %(node, chosen.name)) self.okerrpatterns.append("ERROR: Client .* exited with return code") else: # Sometimes this won't be in the log... self.okerrpatterns.append(self.CM["Pat:ChildKilled"] %(node, chosen.name)) self.okerrpatterns.append(self.CM["Pat:ChildRespawn"] %(node, chosen.name)) self.okerrpatterns.append(self.CM["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("sending fencing op RESET for %s" % node) stonith = self.create_watch(stonithPats, 0) stonith.setwatch() # set the watch for stable watch = self.create_watch( tmpPats, self.CM["DeadTime"] + self.CM["StableTime"] + self.CM["StartTime"]) watch.setwatch() # kill the component chosen.kill(node) # check to see Heartbeat noticed matched = watch.lookforall(allow_multiple_matches=1) if matched: self.CM.debug("Found: "+ repr(matched)) else: self.CM.log("Patterns not found: " + repr(watch.unmatched)) if self.CM.Env["at-boot"] == 0: self.CM.debug("Checking if %s was shot" % node) shot = stonith.look(60) if shot: self.CM.debug("Found: "+ repr(shot)) self.CM.ShouldBeStatus[node]="down" self.CM.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.CM.debug("Waiting for any STONITHd node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.CM.Env["nodes"], 600) self.CM.debug("Waiting for the cluster to re-stabilize with all nodes") is_stable = self.CM.cluster_stable(self.CM["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") 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.CM.Env["nodes"]) for node in partition: try: other_nodes.remove(node) except ValueError: self.CM.log("Node "+node+" not in " + repr(self.CM.Env["nodes"]) + " from " +repr(partition)) if len(other_nodes) == 0: return 1 self.CM.debug("Creating partition: " + repr(partition)) self.CM.debug("Everyone else: " + repr(other_nodes)) for node in partition: if not self.CM.isolate_node(node, other_nodes): self.CM.log("Could not isolate %s" % node) return 0 return 1 def heal_partition(self, partition): other_nodes = [] other_nodes.extend(self.CM.Env["nodes"]) for node in partition: try: other_nodes.remove(node) except ValueError: self.CM.log("Node "+node+" not in " + repr(self.CM.Env["nodes"])) if len(other_nodes) == 0: return 1 self.CM.debug("Healing partition: " + repr(partition)) self.CM.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.CM.Env["nodes"]) for node in self.CM.Env["nodes"]: p = self.CM.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.CM.debug("Created %d partitions" % p_max) for key in partitions.keys(): self.CM.debug("Partition["+str(key)+"]:\t"+repr(partitions[key])) # Disabling STONITH to reduce test complexity for now self.CM.rsh(node, "crm_attribute -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.CM.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.CM.Env["DoStonith"]: self.CM.rsh(node, "crm_attribute -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: 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.CM.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): return self.startall(None) def canrunnow(self, node): '''Return TRUE if we can meaningfully run right now''' if self.find_ocfs2_resources(node): self.CM.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.CM.debug("Disable resource management") self.CM.rsh(node, "crm_attribute -n is-managed-default -v false") if not managed.lookforall(): self.CM.log("Patterns not found: " + repr(managed.unmatched)) return self.failure("Resource management not disabled") pats = [] pats.append("crmd:.*Performing.*_stop_0") pats.append("crmd:.*Performing.*_start_0") pats.append("crmd:.*Performing.*_promote_0") pats.append("crmd:.*Performing.*_demote_0") pats.append("crmd:.*Performing.*_migrate_.*_0") watch = self.create_watch(pats, 60, "ShutdownActivity") watch.setwatch() self.CM.debug("Shutting down the cluster") ret = self.stopall(None) if not ret: self.CM.debug("Re-enable resource management") self.CM.rsh(node, "crm_attribute -D -n is-managed-default") return self.failure("Couldn't shut down the cluster") self.CM.debug("Bringing the cluster back up") ret = self.startall(None) if not ret: self.CM.debug("Re-enable resource management") self.CM.rsh(node, "crm_attribute -D -n is-managed-default") return self.failure("Couldn't restart the cluster") if self.local_badnews("ResourceActivity:", watch): self.CM.debug("Re-enable resource management") self.CM.rsh(node, "crm_attribute -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.CM.debug("Re-enable resource management") self.CM.rsh(node, "crm_attribute -D -n is-managed-default") if not managed.lookforall(): self.CM.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.CM.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.CM.debug("Ignoring: crmd:.*Performing.*op=%s_.*_0" % r.id) ignore.append("crmd:.*Performing.*op=%s_.*_0" % r.id) 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 [ "You may ignore this error if it is unmanaged.", "pingd: .*ERROR: send_ipc_message:", "pingd: .*ERROR: send_update:", "lrmd: .*ERROR: notify_client:", ] def is_applicable(self): if self.CM["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 ret # Start the selected node ret = self.restart1(node) if not ret: return ret # Start all remaining nodes ret = self.startall(None) return ret 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.CM.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.CM.debug("Resource %s is still inactive" % resource) elif rc == 234: self.CM.log("Unknown resource %s" % resource) return 0 elif rc == 246: self.CM.log("Cluster is inactive") return 0 elif rc != 0: self.CM.log("Call to crm_resource failed, rc=%d" % rc) return 0 else: self.CM.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.CM.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.CM.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.CM.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.CM.rsh(node, "crm_resource -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.CM.Env["loop-minutes"]*60 self.find_hae_resources(node) clone_max = len(self.CM.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.CM.rsh(node, "crm_standby -l reboot -v %s" % (target)) return rc def __call__(self, node): self.incr("calls") lpc = 0 failed = 0 done=time.time() + self.CM.Env["loop-minutes"]*60 self.find_hae_resources(node) clone_max = len(self.CM.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 = [] #decide what to do with each node for node in self.CM.Env["nodes"]: action = self.CM.Env.RandomGen.choice(["start","stop"]) #action = self.CM.Env.RandomGen.choice(["start","stop","no change"]) if action == "start" : startset.append(node) elif action == "stop" : stopset.append(node) self.CM.debug("start nodes:" + repr(startset)) self.CM.debug("stop nodes:" + repr(stopset)) #add search patterns watchpats = [ ] for node in stopset: if self.CM.ShouldBeStatus[node] == "up": watchpats.append(self.CM["Pat:We_stopped"] % node) for node in startset: if self.CM.ShouldBeStatus[node] == "down": #watchpats.append(self.CM["Pat:Slave_started"] % node) watchpats.append(self.CM["Pat:Local_started"] % node) else: for stopping in stopset: if self.CM.ShouldBeStatus[stopping] == "up": watchpats.append(self.CM["Pat:They_stopped"] % (node, stopping)) if len(watchpats) == 0: return self.skipped() if len(startset) != 0: watchpats.append(self.CM["Pat:DC_IDLE"]) watch = self.create_watch(watchpats, self.CM["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() return self.success() self.CM.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) if upnodes == [] and downnodes == []: self.CM.cluster_stable() # Make sure they're completely down with no residule for node in stopset: self.CM.rsh(node, self.CM["StopCmd"]) return self.success() if len(upnodes) > 0: self.CM.log("Warn: Unstoppable nodes: " + repr(upnodes)) if len(downnodes) > 0: self.CM.log("Warn: Unstartable nodes: " + repr(downnodes)) return self.failure() 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.CM.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.CM.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.CM.Env["rpm-dir"], version) self.CM.log("Installing %s on %s with %s" % (version, node, flags)) if not self.stop(node): return self.failure("stop failure: "+node) rc = self.CM.rsh(node, "mkdir -p %s" % target_dir) rc = self.CM.rsh(node, "rm -f %s/*.rpm" % target_dir) (rc, lines) = self.CM.rsh(node, "ls -1 %s/*.rpm" % src_dir, None) for line in lines: line = line[:-1] rc = self.CM.rsh.cp("%s" % (line), "%s:%s/" % (node, target_dir)) rc = self.CM.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.CM.Env["current-version"], start) def downgrade(self, node, start=1): return self.install(node, self.CM.Env["previous-version"], start, "--force --nodeps") def __call__(self, node): '''Perform the 'Rolling Upgrade' test. ''' self.incr("calls") for node in self.CM.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.CM.Env.has_key("rpm-dir"): return None if not self.CM.Env.has_key("current-version"): return None if not self.CM.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.CM["DeadTime"]) watch.setwatch() fields = string.split(self.CM.Env["IPBase"], '.') fields[3] = str(int(fields[3])+1) ip = string.join(fields, '.') self.CM.Env["IPBase"] = ip 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.CM.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.CM["DeadTime"]): return self.failure("Unstable cluster") return self.success() def make_ip_resource(self, node, id, rclass, type, ip): self.CM.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.CM.rsh(node, self.cib_cmd % ("constraints", node_constraint), None) if rc != 0: self.CM.log("Constraint creation failed: %d" % rc) return None (rc, lines) = self.CM.rsh(node, self.cib_cmd % ("resources", rsc_xml), None) if rc != 0: self.CM.log("Resource creation failed: %d" % rc) return None return 1 def is_applicable(self): if self.CM.Env["DoBSC"]: return 1 return None 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.CM.debug("Setup: " + self.name) # We ignore the "node" parameter... watchpats = [ ] for node in self.CM.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.incr("WasStarted") watchpats.append(self.CM["Pat:We_stopped"] % node) #if self.CM.Env["use_logd"]: # watchpats.append(self.CM["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.CM["DeadTime"]+10) watch.setwatch() self.set_timer() for node in self.CM.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.CM.Env["nodes"]: self.CM.rsh(node, self.CM["StopCmd"]) return self.success() did_fail=0 up_nodes = [] for node in self.CM.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.CM.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.CM.debug("Setup: " + self.name) # We ignore the "node" parameter... watchpats = [ ] uppat = self.CM["Pat:Slave_started"] if self.CM.upcount() == 0: uppat = self.CM["Pat:Local_started"] for node in self.CM.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "down": self.incr("WasStopped") watchpats.append(uppat % node) if len(watchpats) == 0: return self.success() watchpats.append(self.CM["Pat:DC_IDLE"]) # Start all the nodes - at about the same time... watch = self.create_watch(watchpats, self.CM["DeadTime"]+10) watch.setwatch() self.set_timer() for node in self.CM.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "down": self.CM.StartaCMnoBlock(node) if watch.lookforall(): for attempt in (1, 2, 3, 4, 5): if self.CM.cluster_stable(): return self.success() return self.failure("Cluster did not stabilize") did_fail=0 unstable = [] for node in self.CM.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.CM.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)) self.CM.log("ERROR: All nodes started but CTS didnt detect: " + repr(watch.unmatched)) return self.failure() 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 def BenchTestList(cm, audits): all = TestList(cm, audits) result = [] for test in all: if test.benchmark: result.append(test) return result # vim:ts=4:sw=4:et: