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)
LogFactory().debug("Node %s now up" % node)
return 1
time.sleep(30)
if (not anytimeouts):
LogFactory().debug("Waiting for node %s to come up" % node)
anytimeouts = 1
timeout = timeout - 1
LogFactory().log("%s did not come up within %d tries" % (node, Timeout))
if self.Env["continue"] == 1:
answer = "Y"
else:
try:
answer = raw_input('Continue? [nY]')
except EOFError as e:
answer = "n"
if answer and answer == "n":
raise ValueError("%s did not come up within %d tries" % (node, Timeout))
def __init__(self, node, command, completionDelegate=None):
self.proc = None
self.node = node
self.command = command
self.logger = LogFactory()
self.delegate = completionDelegate
Thread.__init__(self)
def __init__(self, Environment, randseed=None):
self.Env = EnvFactory().getInstance()
self.templates = PatternSelector(self.Env["Name"])
self.__InitialConditions()
self.logger = LogFactory()
self.TestLoggingLevel = 0
self.data = {}
self.name = self.Env["Name"]
self.rsh = RemoteFactory().getInstance()
self.ShouldBeStatus = {}
self.ns = NodeStatus(self.Env)
self.OurNode = os.uname()[1].lower()
self.__instance_errorstoignore = []
self.fastfail = 0
self.cib_installed = 0
self.config = None
self.cluster_monitor = 0
self.use_short_names = 1
if self.Env["DoBSC"]:
del self.templates["Pat:They_stopped"]
self._finalConditions()
self.check_transitions = 0
self.check_elections = 0
self.CIBsync = {}
self.CibFactory = ConfigFactory(self)
self.cib = self.CibFactory.createConfig(self.Env["Schema"])
def __init__(self, rsh, silent=False):
print repr(self)
self. async = []
self.rsh = rsh
self.silent = silent
self.logger = LogFactory()
if trace_rsh:
self.silent = False
self.OurNode = string.lower(os.uname()[1])
def __init__(self, Environment, randseed=None):
self.Env = EnvFactory().getInstance()
self.templates = PatternSelector(self.Env["Name"])
self.__InitialConditions()
self.logger = LogFactory()
self.TestLoggingLevel=0
self.data = {}
self.name = self.Env["Name"]
self.rsh = RemoteFactory().getInstance()
self.ShouldBeStatus={}
self.ns = NodeStatus(self.Env)
self.OurNode = os.uname()[1].lower()
self.__instance_errorstoignore = []
def __init__(self, node, command, completionDelegate=None):
self.proc = None
self.node = node
self.command = command
self.logger = LogFactory()
self.delegate = completionDelegate;
Thread.__init__(self)
def __init__(self, rsh, silent=False):
print repr(self)
self.async = []
self.rsh = rsh
self.silent = silent
self.logger = LogFactory()
if trace_rsh:
self.silent = False
self.OurNode=string.lower(os.uname()[1])
class AsyncRemoteCmd(Thread):
def __init__(self, node, command, completionDelegate=None):
self.proc = None
self.node = node
self.command = command
self.logger = LogFactory()
self.delegate = completionDelegate;
Thread.__init__(self)
def run(self):
outLines = None
errLines = None
self.proc = Popen(self.command, stdout = PIPE, stderr = PIPE, close_fds = True, shell = True)
self.logger.debug("cmd: async: target=%s, pid=%d: %s" % (self.node, self.proc.pid, self.command))
self.proc.wait()
self.logger.debug("cmd: pid %d returned %d to %s" % (self.proc.pid, self.proc.returncode, repr(self.delegate)))
if self.proc.stderr:
errLines = self.proc.stderr.readlines()
self.proc.stderr.close()
for line in errLines:
self.logger.debug("cmd: stderr[%d]: %s" % (self.proc.pid, line))
if self.proc.stdout:
outLines = self.proc.stdout.readlines()
self.proc.stdout.close()
# for line in outLines:
# self.logger.log("cmd: stdout[%d]: %s" % (self.proc.pid, line))
if self.delegate:
self.delegate.async_complete(self.proc.pid, self.proc.returncode, outLines, errLines)
class AsyncWaitProc(Thread):
def __init__(self, proc, node, command, completionDelegate=None):
self.proc = proc
self.node = node
self.command = command
self.logger = LogFactory()
self.delegate = completionDelegate;
Thread.__init__(self)
def run(self):
outLines = None
errLines = None
self.logger.debug("cmd: async: target=%s, pid=%d: %s" % (self.node, self.proc.pid, self.command))
self.proc.wait()
self.logger.debug("cmd: pid %d returned %d" % (self.proc.pid, self.proc.returncode))
if self.proc.stderr:
errLines = self.proc.stderr.readlines()
self.proc.stderr.close()
for line in errLines:
self.logger.debug("cmd: stderr[%d]: %s" % (self.proc.pid, line))
if self.proc.stdout:
outLines = self.proc.stdout.readlines()
self.proc.stdout.close()
# for line in outLines:
# self.logger.debug("cmd: stdout[%d]: %s" % (self.proc.pid, line))
if self.delegate:
self.delegate.async_complete(self.proc.pid, self.proc.returncode, outLines, errLines)
class AsyncRemoteCmd(Thread):
def __init__(self, node, command, completionDelegate=None):
self.proc = None
self.node = node
self.command = command
self.logger = LogFactory()
self.delegate = completionDelegate
Thread.__init__(self)
def run(self):
outLines = None
errLines = None
self.proc = Popen(self.command,
stdout=PIPE,
stderr=PIPE,
close_fds=True,
shell=True)
self.logger.debug("cmd: async: target=%s, pid=%d: %s" %
(self.node, self.proc.pid, self.command))
self.proc.wait()
self.logger.debug(
"cmd: pid %d returned %d to %s" %
(self.proc.pid, self.proc.returncode, repr(self.delegate)))
if self.proc.stderr:
errLines = self.proc.stderr.readlines()
self.proc.stderr.close()
for line in errLines:
self.logger.debug("cmd: stderr[%d]: %s" %
(self.proc.pid, line))
if self.proc.stdout:
outLines = self.proc.stdout.readlines()
self.proc.stdout.close()
# for line in outLines:
# self.logger.log("cmd: stdout[%d]: %s" % (self.proc.pid, line))
if self.delegate:
self.delegate.async_complete(self.proc.pid, self.proc.returncode,
outLines, errLines)
class AsyncWaitProc(Thread):
def __init__(self, proc, node, command, completionDelegate=None):
self.proc = proc
self.node = node
self.command = command
self.logger = LogFactory()
self.delegate = completionDelegate
Thread.__init__(self)
def run(self):
outLines = None
errLines = None
self.logger.debug("cmd: async: target=%s, pid=%d: %s" %
(self.node, self.proc.pid, self.command))
self.proc.wait()
self.logger.debug("cmd: pid %d returned %d" %
(self.proc.pid, self.proc.returncode))
if self.proc.stderr:
errLines = self.proc.stderr.readlines()
self.proc.stderr.close()
for line in errLines:
self.logger.debug("cmd: stderr[%d]: %s" %
(self.proc.pid, line))
if self.proc.stdout:
outLines = self.proc.stdout.readlines()
self.proc.stdout.close()
# for line in outLines:
# self.logger.debug("cmd: stdout[%d]: %s" % (self.proc.pid, line))
if self.delegate:
self.delegate.async_complete(self.proc.pid, self.proc.returncode,
outLines, errLines)
def sig_handler(signum, frame):
LogFactory().log("Interrupted by signal %d" % signum)
if scenario: scenario.summarize()
if signum == 15:
if scenario: scenario.TearDown()
sys.exit(1)
sys.stderr.write("(check your install and PYTHONPATH)\n")
# Now do it again to get more details
from cts.CTSvars import *
from cts.CM_ais import *
from cts.CM_lha import crm_lha
from cts.CTSaudits import AuditList
from cts.CTStests import TestList
from cts.CTSscenarios import *
from cts.logging import LogFactory
sys.exit(-1)
cm = None
scenario = None
LogFactory().add_stderr()
def sig_handler(signum, frame):
LogFactory().log("Interrupted by signal %d" % signum)
if scenario: scenario.summarize()
if signum == 15:
if scenario: scenario.TearDown()
sys.exit(1)
if __name__ == '__main__':
Environment = CtsLab(sys.argv[1:])
NumIter = Environment["iterations"]
Tests = []
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 list(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 = EnvFactory().getInstance()
self.templates = PatternSelector(self.Env["Name"])
self.__InitialConditions()
self.logger = LogFactory()
self.clear_cache = 0
self.TestLoggingLevel=0
self.data = {}
self.name = self.Env["Name"]
self.rsh = RemoteFactory().getInstance()
self.ShouldBeStatus={}
self.ns = NodeStatus(self.Env)
self.OurNode = string.lower(os.uname()[1])
self.__instance_errorstoignore = []
def __getitem__(self, key):
if key == "Name":
return self.name
print("FIXME: Getting %s from %s" % (key, repr(self)))
if key in self.data:
return self.data[key]
return self.templates.get_patterns(self.Env["Name"], key)
def __setitem__(self, key, value):
print("FIXME: Setting %s=%s on %s" % (key, value, repr(self)))
self.data[key] = value
def key_for_node(self, node):
return node
def instance_errorstoignore_clear(self):
'''Allows the test scenario to reset instance errors to ignore on each iteration.'''
self.__instance_errorstoignore = []
def instance_errorstoignore(self):
'''Return list of errors which are 'normal' for a specific test instance'''
return self.__instance_errorstoignore
def errorstoignore(self):
'''Return list of errors which are 'normal' and should be ignored'''
return []
def log(self, args):
self.logger.log(args)
def debug(self, args):
self.logger.debug(args)
def prepare(self):
'''Finish the Initialization process. Prepare to test...'''
print(repr(self)+"prepare")
for node in self.Env["nodes"]:
if self.StataCM(node):
self.ShouldBeStatus[node] = "up"
else:
self.ShouldBeStatus[node] = "down"
self.unisolate_node(node)
def upcount(self):
'''How many nodes are up?'''
count = 0
for node in self.Env["nodes"]:
if self.ShouldBeStatus[node] == "up":
count = count + 1
return count
def install_helper(self, filename, destdir=None, nodes=None, sourcedir=None):
if sourcedir == None:
sourcedir = CTSvars.CTS_home
file_with_path = "%s/%s" % (sourcedir, filename)
if not nodes:
nodes = self.Env["nodes"]
if not destdir:
destdir = CTSvars.CTS_home
self.debug("Installing %s to %s on %s" % (filename, destdir, repr(self.Env["nodes"])))
for node in nodes:
self.rsh(node, "mkdir -p %s" % destdir)
self.rsh.cp(file_with_path, "root@%s:%s/%s" % (node, destdir, filename))
return file_with_path
def install_config(self, node):
return None
def clear_all_caches(self):
if self.clear_cache:
for node in self.Env["nodes"]:
if self.ShouldBeStatus[node] == "down":
self.debug("Removing cache file on: "+node)
self.rsh(node, "rm -f "+CTSvars.HA_VARLIBHBDIR+"/hostcache")
else:
self.debug("NOT Removing cache file on: "+node)
def prepare_fencing_watcher(self, name):
# If we don't have quorum now but get it as a result of starting this node,
# then a bunch of nodes might get fenced
upnode = None
if self.HasQuorum(None):
self.debug("Have quorum")
return None
if not self.templates["Pat:Fencing_start"]:
print("No start pattern")
return None
if not self.templates["Pat:Fencing_ok"]:
print("No ok pattern")
return None
stonith = None
stonithPats = []
for peer in self.Env["nodes"]:
if self.ShouldBeStatus[peer] != "up":
stonithPats.append(self.templates["Pat:Fencing_ok"] % peer)
stonithPats.append(self.templates["Pat:Fencing_start"] % peer)
elif self.Env["Stack"] == "corosync (cman)":
# There is a delay between gaining quorum and CMAN starting fencing
# This can mean that even nodes that are fully up get fenced
# There is no use fighting it, just look for everyone so that CTS doesn't get confused
stonithPats.append(self.templates["Pat:Fencing_ok"] % peer)
stonithPats.append(self.templates["Pat:Fencing_start"] % peer)
stonith = LogWatcher(self.Env["LogFileName"], stonithPats, "StartupFencing", 0, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"])
stonith.setwatch()
return stonith
def fencing_cleanup(self, node, stonith):
peer_list = []
peer_state = {}
self.debug("Looking for nodes that were fenced as a result of %s starting" % node)
# If we just started a node, we may now have quorum (and permission to fence)
if not stonith:
self.debug("Nothing to do")
return peer_list
q = self.HasQuorum(None)
if not q and len(self.Env["nodes"]) > 2:
# We didn't gain quorum - we shouldn't have shot anyone
self.debug("Quorum: %d Len: %d" % (q, len(self.Env["nodes"])))
return peer_list
for n in self.Env["nodes"]:
peer_state[n] = "unknown"
# Now see if any states need to be updated
self.debug("looking for: " + repr(stonith.regexes))
shot = stonith.look(0)
while shot:
line = repr(shot)
self.debug("Found: " + line)
del stonith.regexes[stonith.whichmatch]
# Extract node name
for n in self.Env["nodes"]:
if re.search(self.templates["Pat:Fencing_ok"] % n, shot):
peer = n
peer_state[peer] = "complete"
self.__instance_errorstoignore.append(self.templates["Pat:Fencing_ok"] % peer)
elif peer_state[n] != "complete" and re.search(self.templates["Pat:Fencing_start"] % n, shot):
# TODO: Correctly detect multiple fencing operations for the same host
peer = n
peer_state[peer] = "in-progress"
self.__instance_errorstoignore.append(self.templates["Pat:Fencing_start"] % peer)
if not peer:
self.logger.log("ERROR: Unknown stonith match: %s" % line)
elif not peer in peer_list:
self.debug("Found peer: " + peer)
peer_list.append(peer)
# Get the next one
shot = stonith.look(60)
for peer in peer_list:
self.debug(" Peer %s was fenced as a result of %s starting: %s" % (peer, node, peer_state[peer]))
if self.Env["at-boot"]:
self.ShouldBeStatus[peer] = "up"
else:
self.ShouldBeStatus[peer] = "down"
if peer_state[peer] == "in-progress":
# Wait for any in-progress operations to complete
shot = stonith.look(60)
while len(stonith.regexes) and shot:
line = repr(shot)
self.debug("Found: " + line)
del stonith.regexes[stonith.whichmatch]
shot = stonith.look(60)
# Now make sure the node is alive too
self.ns.WaitForNodeToComeUp(peer, self.Env["DeadTime"])
# Poll until it comes up
if self.Env["at-boot"]:
if not self.StataCM(peer):
time.sleep(self.Env["StartTime"])
if not self.StataCM(peer):
self.logger.log("ERROR: Peer %s failed to restart after being fenced" % peer)
return None
return peer_list
def StartaCM(self, node, verbose=False):
'''Start up the cluster manager on a given node'''
if verbose: self.logger.log("Starting %s on node %s" % (self.templates["Name"], node))
else: self.debug("Starting %s on node %s" % (self.templates["Name"], node))
ret = 1
if not node in self.ShouldBeStatus:
self.ShouldBeStatus[node] = "down"
if self.ShouldBeStatus[node] != "down":
return 1
patterns = []
# Technically we should always be able to notice ourselves starting
patterns.append(self.templates["Pat:Local_started"] % node)
if self.upcount() == 0:
patterns.append(self.templates["Pat:Master_started"] % node)
else:
patterns.append(self.templates["Pat:Slave_started"] % node)
watch = LogWatcher(
self.Env["LogFileName"], patterns, "StartaCM", self.Env["StartTime"]+10, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"])
self.install_config(node)
self.ShouldBeStatus[node] = "any"
if self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]):
self.logger.log ("%s was already started" % (node))
return 1
# 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.templates["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.templates["StartCmd"])
stonith = self.prepare_fencing_watcher(node)
watch.setwatch()
if self.rsh(node, startCmd) != 0:
self.logger.log ("Warn: Start command failed on node %s" % (node))
self.fencing_cleanup(node, stonith)
return None
self.ShouldBeStatus[node] = "up"
watch_result = watch.lookforall()
if watch.unmatched:
for regex in watch.unmatched:
self.logger.log ("Warn: Startup pattern not found: %s" % (regex))
if watch_result and self.cluster_stable(self.Env["DeadTime"]):
#self.debug("Found match: "+ repr(watch_result))
self.fencing_cleanup(node, stonith)
return 1
elif self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]):
self.fencing_cleanup(node, stonith)
return 1
self.logger.log ("Warn: Start failed for node %s" % (node))
return None
def StartaCMnoBlock(self, node, verbose=False):
'''Start up the cluster manager on a given node with none-block mode'''
if verbose: self.logger.log("Starting %s on node %s" % (self["Name"], node))
else: self.debug("Starting %s on node %s" % (self["Name"], node))
# Clear out the host cache so autojoin can be exercised
if self.clear_cache:
self.debug("Removing cache file on: "+node)
self.rsh(node, "rm -f "+CTSvars.HA_VARLIBHBDIR+"/hostcache")
self.install_config(node)
if not(self.Env["valgrind-tests"]):
startCmd = self.templates["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.templates["StartCmd"])
self.rsh(node, startCmd, synchronous=0)
self.ShouldBeStatus[node] = "up"
return 1
def StopaCM(self, node, verbose=False, force=False):
'''Stop the cluster manager on a given node'''
if verbose: self.logger.log("Stopping %s on node %s" % (self["Name"], node))
else: self.debug("Stopping %s on node %s" % (self["Name"], node))
if self.ShouldBeStatus[node] != "up" and force == False:
return 1
if self.rsh(node, self.templates["StopCmd"]) == 0:
# Make sure we can continue even if corosync leaks
# fdata-* is the old name
#self.rsh(node, "rm -f /dev/shm/qb-* /dev/shm/fdata-*")
self.ShouldBeStatus[node] = "down"
self.cluster_stable(self.Env["DeadTime"])
return 1
else:
self.logger.log ("ERROR: Could not stop %s on node %s" % (self["Name"], node))
return None
def StopaCMnoBlock(self, node):
'''Stop the cluster manager on a given node with none-block mode'''
self.debug("Stopping %s on node %s" % (self["Name"], node))
self.rsh(node, self.templates["StopCmd"], synchronous=0)
self.ShouldBeStatus[node] = "down"
return 1
def cluster_stable(self, timeout = None):
time.sleep(self.Env["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.templates["RereadCmd"])
if rc == 0:
return 1
else:
self.logger.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.templates["StatusCmd"] % node, 1)
ret= (string.find(out, 'stopped') == -1)
try:
if ret:
if self.ShouldBeStatus[node] == "down":
self.logger.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.logger.log(
"Node status for %s is %s but we think it should be %s"
% (node, "down", self.ShouldBeStatus[node]))
except KeyError: pass
if ret:
self.ShouldBeStatus[node] = "up"
else:
self.ShouldBeStatus[node] = "down"
return ret
def startall(self, nodelist=None, verbose=False, quick=False):
'''Start the cluster manager on every node in the cluster.
We can do it on a subset of the cluster if nodelist is not None.
'''
map = {}
if not nodelist:
nodelist = self.Env["nodes"]
for node in nodelist:
if self.ShouldBeStatus[node] == "down":
self.ns.WaitForAllNodesToComeUp(nodelist, 300)
if not quick:
# This is used for "basic sanity checks", so only start one node ...
if not self.StartaCM(node, verbose=verbose):
return 0
return 1
# Approximation of SimulStartList for --boot
watchpats = [ ]
watchpats.append(self.templates["Pat:DC_IDLE"])
for node in nodelist:
watchpats.append(self.templates["Pat:Local_started"] % node)
watchpats.append(self.templates["Pat:InfraUp"] % node)
watchpats.append(self.templates["Pat:PacemakerUp"] % node)
# Start all the nodes - at about the same time...
watch = LogWatcher(self.Env["LogFileName"], watchpats, "fast-start", self.Env["DeadTime"]+10, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"])
watch.setwatch()
if not self.StartaCM(nodelist[0], verbose=verbose):
return 0
for node in nodelist:
self.StartaCMnoBlock(node, verbose=verbose)
watch.lookforall()
if watch.unmatched:
for regex in watch.unmatched:
self.logger.log ("Warn: Startup pattern not found: %s" % (regex))
if not self.cluster_stable():
self.logger.log("Cluster did not stabilize")
return 0
return 1
def stopall(self, nodelist=None, verbose=False, force=False):
'''Stop the cluster managers on every node in the cluster.
We can do it on a subset of the cluster if nodelist is not None.
'''
ret = 1
map = {}
if not nodelist:
nodelist = self.Env["nodes"]
for node in self.Env["nodes"]:
if self.ShouldBeStatus[node] == "up" or force == True:
if not self.StopaCM(node, verbose=verbose, force=force):
ret = 0
return ret
def 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.templates["BreakCommCmd"] % self.key_for_node(node))
if rc != 0:
self.logger.log("Could not break the communication between %s and %s: %d" % (target, node, rc))
return None
else:
self.debug("Communication cut between %s and %s" % (target, node))
return 1
def unisolate_node(self, target, nodes=None):
'''fix the communication between the nodes'''
if not nodes:
nodes = self.Env["nodes"]
for node in nodes:
if node != target:
restored = 0
# Limit the amount of time we have asynchronous connectivity for
# Restore both sides as simultaneously as possible
self.rsh(target, self.templates["FixCommCmd"] % self.key_for_node(node), synchronous=0)
self.rsh(node, self.templates["FixCommCmd"] % self.key_for_node(target), synchronous=0)
self.debug("Communication restored between %s and %s" % (target, node))
def reducecomm_node(self,node):
'''reduce the communication between the nodes'''
rc = self.rsh(node, self.templates["ReduceCommCmd"]%(self.Env["XmitLoss"],self.Env["RecvLoss"]))
if rc == 0:
return 1
else:
self.logger.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.templates["RestoreCommCmd"]);
if rc == 0:
return 1
else:
self.logger.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")
def StatsExtract(self):
if not self.Env["stats"]:
return
for host in self.Env["nodes"]:
log_stats_file = "%s/cts-stats.csv" % CTSvars.CRM_DAEMON_DIR
if host in has_log_stats:
self.rsh(host, '''bash %s %s stop''' % (log_stats_bin, log_stats_file))
(rc, lines) = self.rsh(host, '''cat %s''' % log_stats_file, stdout=2)
self.rsh(host, '''bash %s %s delete''' % (log_stats_bin, log_stats_file))
fname = "cts-stats-%d-nodes-%s.csv" % (len(self.Env["nodes"]), host)
print("Extracted stats: %s" % fname)
fd = open(fname, "a")
fd.writelines(lines)
fd.close()
def StatsMark(self, testnum):
'''Mark the test number in the stats log'''
global has_log_stats
if not self.Env["stats"]:
return
for host in self.Env["nodes"]:
log_stats_file = "%s/cts-stats.csv" % CTSvars.CRM_DAEMON_DIR
if not host in has_log_stats:
global log_stats
global log_stats_bin
script=log_stats
#script = re.sub("\\\\", "\\\\", script)
script = re.sub('\"', '\\\"', script)
script = re.sub("'", "\'", script)
script = re.sub("`", "\`", script)
script = re.sub("\$", "\\\$", script)
self.debug("Installing %s on %s" % (log_stats_bin, host))
self.rsh(host, '''echo "%s" > %s''' % (script, log_stats_bin), silent=True)
self.rsh(host, '''bash %s %s delete''' % (log_stats_bin, log_stats_file))
has_log_stats[host] = 1
# Now mark it
self.rsh(host, '''bash %s %s mark %s''' % (log_stats_bin, log_stats_file, testnum), synchronous=0)
class 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, rsh, silent=False):
print repr(self)
self. async = []
self.rsh = rsh
self.silent = silent
self.logger = LogFactory()
if trace_rsh:
self.silent = False
self.OurNode = string.lower(os.uname()[1])
def _fixcmd(self, cmd):
return re.sub("\'", "'\\''", cmd)
def _cmd(self, *args):
'''Compute the string that will run the given command on the
given remote system'''
args = args[0]
sysname = args[0]
command = args[1]
#print "sysname: %s, us: %s" % (sysname, self.OurNode)
if sysname == None or string.lower(
sysname) == self.OurNode or sysname == "localhost":
ret = command
else:
ret = self.rsh.Command + " " + sysname + " '" + self._fixcmd(
command) + "'"
#print ("About to run %s\n" % ret)
return ret
def log(self, args):
if not self.silent:
self.logger.log(args)
def debug(self, args):
if not self.silent:
self.logger.debug(args)
def call_async(self, node, command, completionDelegate=None):
#if completionDelegate: print "Waiting for %d on %s: %s" % (proc.pid, node, command)
aproc = AsyncRemoteCmd(node,
self._cmd([node, command]),
completionDelegate=completionDelegate)
aproc.start()
return aproc
def __call__(self,
node,
command,
stdout=0,
synchronous=1,
silent=False,
blocking=True,
completionDelegate=None):
'''Run the given command on the given remote system
If you call this class like a function, this is the function that gets
called. It just runs it roughly as though it were a system() call
on the remote machine. The first argument is name of the machine to
run it on.
'''
if trace_rsh:
silent = False
rc = 0
result = None
proc = Popen(self._cmd([node, command]),
stdout=PIPE,
stderr=PIPE,
close_fds=True,
shell=True)
#if completionDelegate: print "Waiting for %d on %s: %s" % (proc.pid, node, command)
if not synchronous and proc.pid > 0 and not self.silent:
aproc = AsyncWaitProc(proc,
node,
command,
completionDelegate=completionDelegate)
aproc.start()
return 0
#if not blocking:
# fcntl.fcntl(proc.stdout.fileno(), fcntl.F_SETFL, os.O_NONBLOCK)
if proc.stdout:
if stdout == 1:
result = proc.stdout.readline()
else:
result = proc.stdout.readlines()
proc.stdout.close()
else:
self.log("No stdout stream")
rc = proc.wait()
if not silent:
self.debug("cmd: target=%s, rc=%d: %s" % (node, rc, command))
if stdout == 1:
return result
if proc.stderr:
errors = proc.stderr.readlines()
proc.stderr.close()
if completionDelegate:
completionDelegate.async_complete(proc.pid, proc.returncode,
result, errors)
if not silent:
for err in errors:
if stdout == 3:
result.append("error: " + err)
else:
self.debug("cmd: stderr: %s" % err)
if stdout == 0:
if not silent and result:
for line in result:
self.debug("cmd: stdout: %s" % line)
return rc
return (rc, result)
def cp(self, source, target, silent=False):
'''Perform a remote copy'''
cpstring = self.rsh.CpCommand + " \'" + source + "\'" + " \'" + target + "\'"
rc = os.system(cpstring)
if trace_rsh:
silent = False
if not silent: self.debug("cmd: rc=%d: %s" % (rc, cpstring))
return rc
def __init__(self, args=None):
self.Env = EnvFactory().getInstance(args)
self.Scenario = None
self.logger = LogFactory()
self.rsh = RemoteFactory().getInstance()
class CtsLab:
'''This class defines the Lab Environment for the Cluster Test System.
It defines those things which are expected to change from test
environment to test environment for the same cluster manager.
It is where you define the set of nodes that are in your test lab
what kind of reset mechanism you use, etc.
This class is derived from a UserDict because we hold many
different parameters of different kinds, and this provides
provide a uniform and extensible interface useful for any kind of
communication between the user/administrator/tester and CTS.
At this point in time, it is the intent of this class to model static
configuration and/or environmental data about the environment which
doesn't change as the tests proceed.
Well-known names (keys) are an important concept in this class.
The HasMinimalKeys member function knows the minimal set of
well-known names for the class.
The following names are standard (well-known) at this time:
nodes An array of the nodes in the cluster
reset A ResetMechanism object
logger An array of objects that log strings...
CMclass The type of ClusterManager we are running
(This is a class object, not a class instance)
RandSeed Random seed. It is a triple of bytes. (optional)
The CTS code ignores names it doesn't know about/need.
The individual tests have access to this information, and it is
perfectly acceptable to provide hints, tweaks, fine-tuning
directions or other information to the tests through this mechanism.
'''
def __init__(self, args=None):
self.Env = EnvFactory().getInstance(args)
self.Scenario = None
self.logger = LogFactory()
self.rsh = RemoteFactory().getInstance()
def dump(self):
self.Env.dump()
def has_key(self, key):
return key in self.Env.keys()
def __getitem__(self, key):
return self.Env[key]
def __setitem__(self, key, value):
self.Env[key] = value
def run(self, Scenario, Iterations):
if not Scenario:
self.logger.log("No scenario was defined")
return 1
self.logger.log("Cluster nodes: ")
for node in self.Env["nodes"]:
self.logger.log(" * %s" % (node))
if not Scenario.SetUp():
return 1
try :
Scenario.run(Iterations)
except :
self.logger.log("Exception by %s" % sys.exc_info()[0])
self.logger.traceback(traceback)
Scenario.summarize()
Scenario.TearDown()
return 1
#ClusterManager.oprofileSave(Iterations)
Scenario.TearDown()
Scenario.summarize()
if Scenario.Stats["failure"] > 0:
return Scenario.Stats["failure"]
elif Scenario.Stats["success"] != Iterations:
self.logger.log("No failure count but success != requested iterations")
return 1
return 0
def __CheckNode(self, node):
"Raise a ValueError if the given node isn't valid"
if not self.IsValidNode(node):
raise ValueError("Invalid node [%s] in CheckNode" % node)
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, rsh, silent=False):
print repr(self)
self.async = []
self.rsh = rsh
self.silent = silent
self.logger = LogFactory()
if trace_rsh:
self.silent = False
self.OurNode=string.lower(os.uname()[1])
def _fixcmd(self, cmd):
return re.sub("\'", "'\\''", cmd)
def _cmd(self, *args):
'''Compute the string that will run the given command on the
given remote system'''
args= args[0]
sysname = args[0]
command = args[1]
#print "sysname: %s, us: %s" % (sysname, self.OurNode)
if sysname == None or string.lower(sysname) == self.OurNode or sysname == "localhost":
ret = command
else:
ret = self.rsh.Command + " " + sysname + " '" + self._fixcmd(command) + "'"
#print ("About to run %s\n" % ret)
return ret
def log(self, args):
if not self.silent:
self.logger.log(args)
def debug(self, args):
if not self.silent:
self.logger.debug(args)
def call_async(self, node, command, completionDelegate=None):
#if completionDelegate: print "Waiting for %d on %s: %s" % (proc.pid, node, command)
aproc = AsyncRemoteCmd(node, self._cmd([node, command]), completionDelegate=completionDelegate)
aproc.start()
return aproc
def __call__(self, node, command, stdout=0, synchronous=1, silent=False, blocking=True, completionDelegate=None):
'''Run the given command on the given remote system
If you call this class like a function, this is the function that gets
called. It just runs it roughly as though it were a system() call
on the remote machine. The first argument is name of the machine to
run it on.
'''
if trace_rsh:
silent = False
rc = 0
result = None
proc = Popen(self._cmd([node, command]),
stdout = PIPE, stderr = PIPE, close_fds = True, shell = True)
#if completionDelegate: print "Waiting for %d on %s: %s" % (proc.pid, node, command)
if not synchronous and proc.pid > 0 and not self.silent:
aproc = AsyncWaitProc(proc, node, command, completionDelegate=completionDelegate)
aproc.start()
return 0
#if not blocking:
# fcntl.fcntl(proc.stdout.fileno(), fcntl.F_SETFL, os.O_NONBLOCK)
if proc.stdout:
if stdout == 1:
result = proc.stdout.readline()
else:
result = proc.stdout.readlines()
proc.stdout.close()
else:
self.log("No stdout stream")
rc = proc.wait()
if not silent: self.debug("cmd: target=%s, rc=%d: %s" % (node, rc, command))
if stdout == 1:
return result
if proc.stderr:
errors = proc.stderr.readlines()
proc.stderr.close()
if completionDelegate:
completionDelegate.async_complete(proc.pid, proc.returncode, result, errors)
if not silent:
for err in errors:
if stdout == 3:
result.append("error: "+err)
else:
self.debug("cmd: stderr: %s" % err)
if stdout == 0:
if not silent and result:
for line in result:
self.debug("cmd: stdout: %s" % line)
return rc
return (rc, result)
def cp(self, source, target, silent=False):
'''Perform a remote copy'''
cpstring = self.rsh.CpCommand + " \'" + source + "\'" + " \'" + target + "\'"
rc = os.system(cpstring)
if trace_rsh:
silent = False
if not silent: self.debug("cmd: rc=%d: %s" % (rc, cpstring))
return rc
def __init__(self, args=None):
self.Env = EnvFactory().getInstance(args)
self.Scenario = None
self.logger = LogFactory()
self.rsh = RemoteFactory().getInstance()
class CtsLab:
'''This class defines the Lab Environment for the Cluster Test System.
It defines those things which are expected to change from test
environment to test environment for the same cluster manager.
It is where you define the set of nodes that are in your test lab
what kind of reset mechanism you use, etc.
This class is derived from a UserDict because we hold many
different parameters of different kinds, and this provides
provide a uniform and extensible interface useful for any kind of
communication between the user/administrator/tester and CTS.
At this point in time, it is the intent of this class to model static
configuration and/or environmental data about the environment which
doesn't change as the tests proceed.
Well-known names (keys) are an important concept in this class.
The HasMinimalKeys member function knows the minimal set of
well-known names for the class.
The following names are standard (well-known) at this time:
nodes An array of the nodes in the cluster
reset A ResetMechanism object
logger An array of objects that log strings...
CMclass The type of ClusterManager we are running
(This is a class object, not a class instance)
RandSeed Random seed. It is a triple of bytes. (optional)
The CTS code ignores names it doesn't know about/need.
The individual tests have access to this information, and it is
perfectly acceptable to provide hints, tweaks, fine-tuning
directions or other information to the tests through this mechanism.
'''
def __init__(self, args=None):
self.Env = EnvFactory().getInstance(args)
self.Scenario = None
self.logger = LogFactory()
self.rsh = RemoteFactory().getInstance()
def dump(self):
self.Env.dump()
def has_key(self, key):
return key in self.Env.keys()
def __getitem__(self, key):
return self.Env[key]
def __setitem__(self, key, value):
self.Env[key] = value
def run(self, Scenario, Iterations):
if not Scenario:
self.logger.log("No scenario was defined")
return 1
self.logger.log("Cluster nodes: ")
for node in self.Env["nodes"]:
self.logger.log(" * %s" % (node))
if not Scenario.SetUp():
return 1
try :
Scenario.run(Iterations)
except :
self.logger.log("Exception by %s" % sys.exc_info()[0])
self.logger.traceback(traceback)
Scenario.summarize()
Scenario.TearDown()
return 1
#ClusterManager.oprofileSave(Iterations)
Scenario.TearDown()
Scenario.summarize()
if Scenario.Stats["failure"] > 0:
return Scenario.Stats["failure"]
elif Scenario.Stats["success"] != Iterations:
self.logger.log("No failure count but success != requested iterations")
return 1
return 0
def __CheckNode(self, node):
"Raise a ValueError if the given node isn't valid"
if not self.IsValidNode(node):
raise ValueError("Invalid node [%s] in CheckNode" % node)
class RATesterScenarioComponent(ScenarioComponent):
"""Assertion-friendly base class for scenario setup/teardown.
"""
def __init__(self, environment, verbose=True):
self.rsh = RemoteFactory().getInstance()
self.logger = LogFactory()
self.Env = environment
self.verbose = verbose
self.ratemplates = RATemplates()
def copy_to_nodes(self, files, create_dir=False, owner=False, perm=False):
for node in self.Env["nodes"]:
for localfile, remotefile in files:
if create_dir:
remotedir = os.path.dirname(remotefile)
rc = self.rsh(node, "mkdir -p %s" % remotedir)
assert rc == 0, 'create dir "%s" on remote node "%s"' % (remotedir, node)
src = os.path.join(os.path.dirname(os.path.abspath(__file__)), localfile)
rc = self.rsh.cp(src, "root@%s:%s" % (node, remotefile))
assert rc == 0, 'copy test data "%s" on remote node "%s"' % (src, node)
if owner:
rc = self.rsh(node, "chown %s %s" % (owner, remotefile))
assert rc == 0, 'change ownership of "%s" on remote node "%s"' % (src, node)
if perm:
rc = self.rsh(node, "chmod %s %s" % (perm, remotefile))
assert rc == 0, 'change permission of "%s" on remote node "%s"' % (src, node)
def get_candidate_path(self, candidates, is_dir=False):
testopt = "-f" if is_dir is False else "-d"
target = False
for candidate in candidates:
if self.rsh(self.Env["nodes"][0], "test %s %s" % (testopt, candidate)) == 0:
return candidate
assert target
def IsApplicable(self):
return 1
def SetUp(self, cluster_manager):
try:
self.setup_scenario(cluster_manager)
return 1
except AssertionError as e:
print("Setup of scenario %s failed: %s" % (self.__class__.__name__, str(e)))
return 0
def TearDown(self, cluster_manager):
try:
self.teardown_scenario(cluster_manager)
return 1
except AssertionError as e:
print("Teardown of scenario %s failed: %s" % (self.__class__.__name__, str(e)))
return 0
def log(self, args):
self.logger.log(args)
def debug(self, args):
self.logger.debug(args)
def rsh_check(self, target, command, expected=0):
if self.verbose:
self.log("> [%s] %s" % (target, command))
res = self.rsh(target, command + " &>/dev/null")
assert res == expected, '"%s" returned %d' % (command, res)
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 list(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 = EnvFactory().getInstance()
self.templates = PatternSelector(self.Env["Name"])
self.__InitialConditions()
self.logger = LogFactory()
self.TestLoggingLevel = 0
self.data = {}
self.name = self.Env["Name"]
self.rsh = RemoteFactory().getInstance()
self.ShouldBeStatus = {}
self.ns = NodeStatus(self.Env)
self.OurNode = os.uname()[1].lower()
self.__instance_errorstoignore = []
self.fastfail = 0
self.cib_installed = 0
self.config = None
self.cluster_monitor = 0
self.use_short_names = 1
if self.Env["DoBSC"]:
del self.templates["Pat:They_stopped"]
self._finalConditions()
self.check_transitions = 0
self.check_elections = 0
self.CIBsync = {}
self.CibFactory = ConfigFactory(self)
self.cib = self.CibFactory.createConfig(self.Env["Schema"])
def __getitem__(self, key):
if key == "Name":
return self.name
print("FIXME: Getting %s from %s" % (key, repr(self)))
if key in self.data:
return self.data[key]
return self.templates.get_patterns(self.Env["Name"], key)
def __setitem__(self, key, value):
print("FIXME: Setting %s=%s on %s" % (key, value, repr(self)))
self.data[key] = value
def key_for_node(self, node):
return node
def instance_errorstoignore_clear(self):
'''Allows the test scenario to reset instance errors to ignore on each iteration.'''
self.__instance_errorstoignore = []
def instance_errorstoignore(self):
'''Return list of errors which are 'normal' for a specific test instance'''
return self.__instance_errorstoignore
def log(self, args):
self.logger.log(args)
def debug(self, args):
self.logger.debug(args)
def upcount(self):
'''How many nodes are up?'''
count = 0
for node in self.Env["nodes"]:
if self.ShouldBeStatus[node] == "up":
count = count + 1
return count
def install_support(self, command="install"):
for node in self.Env["nodes"]:
self.rsh(node, CTSvars.CRM_DAEMON_DIR + "/cts-support " + command)
def prepare_fencing_watcher(self, name):
# If we don't have quorum now but get it as a result of starting this node,
# then a bunch of nodes might get fenced
upnode = None
if self.HasQuorum(None):
self.debug("Have quorum")
return None
if not self.templates["Pat:Fencing_start"]:
print("No start pattern")
return None
if not self.templates["Pat:Fencing_ok"]:
print("No ok pattern")
return None
stonith = None
stonithPats = []
for peer in self.Env["nodes"]:
if self.ShouldBeStatus[peer] != "up":
stonithPats.append(self.templates["Pat:Fencing_ok"] % peer)
stonithPats.append(self.templates["Pat:Fencing_start"] % peer)
stonith = LogWatcher(self.Env["LogFileName"],
stonithPats,
"StartupFencing",
0,
hosts=self.Env["nodes"],
kind=self.Env["LogWatcher"])
stonith.setwatch()
return stonith
def fencing_cleanup(self, node, stonith):
peer_list = []
peer_state = {}
self.debug(
"Looking for nodes that were fenced as a result of %s starting" %
node)
# If we just started a node, we may now have quorum (and permission to fence)
if not stonith:
self.debug("Nothing to do")
return peer_list
q = self.HasQuorum(None)
if not q and len(self.Env["nodes"]) > 2:
# We didn't gain quorum - we shouldn't have shot anyone
self.debug("Quorum: %d Len: %d" % (q, len(self.Env["nodes"])))
return peer_list
for n in self.Env["nodes"]:
peer_state[n] = "unknown"
# Now see if any states need to be updated
self.debug("looking for: " + repr(stonith.regexes))
shot = stonith.look(0)
while shot:
line = repr(shot)
self.debug("Found: " + line)
del stonith.regexes[stonith.whichmatch]
# Extract node name
for n in self.Env["nodes"]:
if re.search(self.templates["Pat:Fencing_ok"] % n, shot):
peer = n
peer_state[peer] = "complete"
self.__instance_errorstoignore.append(
self.templates["Pat:Fencing_ok"] % peer)
elif peer_state[n] != "complete" and re.search(
self.templates["Pat:Fencing_start"] % n, shot):
# TODO: Correctly detect multiple fencing operations for the same host
peer = n
peer_state[peer] = "in-progress"
self.__instance_errorstoignore.append(
self.templates["Pat:Fencing_start"] % peer)
if not peer:
self.logger.log("ERROR: Unknown stonith match: %s" % line)
elif not peer in peer_list:
self.debug("Found peer: " + peer)
peer_list.append(peer)
# Get the next one
shot = stonith.look(60)
for peer in peer_list:
self.debug(" Peer %s was fenced as a result of %s starting: %s" %
(peer, node, peer_state[peer]))
if self.Env["at-boot"]:
self.ShouldBeStatus[peer] = "up"
else:
self.ShouldBeStatus[peer] = "down"
if peer_state[peer] == "in-progress":
# Wait for any in-progress operations to complete
shot = stonith.look(60)
while len(stonith.regexes) and shot:
line = repr(shot)
self.debug("Found: " + line)
del stonith.regexes[stonith.whichmatch]
shot = stonith.look(60)
# Now make sure the node is alive too
self.ns.WaitForNodeToComeUp(peer, self.Env["DeadTime"])
# Poll until it comes up
if self.Env["at-boot"]:
if not self.StataCM(peer):
time.sleep(self.Env["StartTime"])
if not self.StataCM(peer):
self.logger.log(
"ERROR: Peer %s failed to restart after being fenced" %
peer)
return None
return peer_list
def StartaCM(self, node, verbose=False):
'''Start up the cluster manager on a given node'''
if verbose:
self.logger.log("Starting %s on node %s" %
(self.templates["Name"], node))
else:
self.debug("Starting %s on node %s" %
(self.templates["Name"], node))
ret = 1
if not node in self.ShouldBeStatus:
self.ShouldBeStatus[node] = "down"
if self.ShouldBeStatus[node] != "down":
return 1
patterns = []
# Technically we should always be able to notice ourselves starting
patterns.append(self.templates["Pat:Local_started"] % node)
if self.upcount() == 0:
patterns.append(self.templates["Pat:DC_started"] % node)
else:
patterns.append(self.templates["Pat:NonDC_started"] % node)
watch = LogWatcher(self.Env["LogFileName"],
patterns,
"StartaCM",
self.Env["StartTime"] + 10,
hosts=self.Env["nodes"],
kind=self.Env["LogWatcher"])
self.install_config(node)
self.ShouldBeStatus[node] = "any"
if self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]):
self.logger.log("%s was already started" % (node))
return 1
stonith = self.prepare_fencing_watcher(node)
watch.setwatch()
if self.rsh(node, self.templates["StartCmd"]) != 0:
self.logger.log("Warn: Start command failed on node %s" % (node))
self.fencing_cleanup(node, stonith)
return None
self.ShouldBeStatus[node] = "up"
watch_result = watch.lookforall()
if watch.unmatched:
for regex in watch.unmatched:
self.logger.log("Warn: Startup pattern not found: %s" %
(regex))
if watch_result and self.cluster_stable(self.Env["DeadTime"]):
#self.debug("Found match: "+ repr(watch_result))
self.fencing_cleanup(node, stonith)
return 1
elif self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]):
self.fencing_cleanup(node, stonith)
return 1
self.logger.log("Warn: Start failed for node %s" % (node))
return None
def StartaCMnoBlock(self, node, verbose=False):
'''Start up the cluster manager on a given node with none-block mode'''
if verbose:
self.logger.log("Starting %s on node %s" % (self["Name"], node))
else:
self.debug("Starting %s on node %s" % (self["Name"], node))
self.install_config(node)
self.rsh(node, self.templates["StartCmd"], synchronous=0)
self.ShouldBeStatus[node] = "up"
return 1
def StopaCM(self, node, verbose=False, force=False):
'''Stop the cluster manager on a given node'''
if verbose:
self.logger.log("Stopping %s on node %s" % (self["Name"], node))
else:
self.debug("Stopping %s on node %s" % (self["Name"], node))
if self.ShouldBeStatus[node] != "up" and force == False:
return 1
if self.rsh(node, self.templates["StopCmd"]) == 0:
# Make sure we can continue even if corosync leaks
# fdata-* is the old name
#self.rsh(node, "rm -rf /dev/shm/qb-* /dev/shm/fdata-*")
self.ShouldBeStatus[node] = "down"
self.cluster_stable(self.Env["DeadTime"])
return 1
else:
self.logger.log("ERROR: Could not stop %s on node %s" %
(self["Name"], node))
return None
def StopaCMnoBlock(self, node):
'''Stop the cluster manager on a given node with none-block mode'''
self.debug("Stopping %s on node %s" % (self["Name"], node))
self.rsh(node, self.templates["StopCmd"], synchronous=0)
self.ShouldBeStatus[node] = "down"
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.templates["RereadCmd"])
if rc == 0:
return 1
else:
self.logger.log(
"Could not force %s on node %s to reread its config" %
(self["Name"], node))
return None
def startall(self, nodelist=None, verbose=False, quick=False):
'''Start the cluster manager on every node in the cluster.
We can do it on a subset of the cluster if nodelist is not None.
'''
map = {}
if not nodelist:
nodelist = self.Env["nodes"]
for node in nodelist:
if self.ShouldBeStatus[node] == "down":
self.ns.WaitForAllNodesToComeUp(nodelist, 300)
if not quick:
# This is used for "basic sanity checks", so only start one node ...
if not self.StartaCM(node, verbose=verbose):
return 0
return 1
# Approximation of SimulStartList for --boot
watchpats = []
watchpats.append(self.templates["Pat:DC_IDLE"])
for node in nodelist:
watchpats.append(self.templates["Pat:InfraUp"] % node)
watchpats.append(self.templates["Pat:PacemakerUp"] % node)
watchpats.append(self.templates["Pat:Local_started"] % node)
watchpats.append(self.templates["Pat:They_up"] %
(nodelist[0], node))
# Start all the nodes - at about the same time...
watch = LogWatcher(self.Env["LogFileName"],
watchpats,
"fast-start",
self.Env["DeadTime"] + 10,
hosts=self.Env["nodes"],
kind=self.Env["LogWatcher"])
watch.setwatch()
if not self.StartaCM(nodelist[0], verbose=verbose):
return 0
for node in nodelist:
self.StartaCMnoBlock(node, verbose=verbose)
watch.lookforall()
if watch.unmatched:
for regex in watch.unmatched:
self.logger.log("Warn: Startup pattern not found: %s" %
(regex))
if not self.cluster_stable():
self.logger.log("Cluster did not stabilize")
return 0
return 1
def stopall(self, nodelist=None, verbose=False, force=False):
'''Stop the cluster managers on every node in the cluster.
We can do it on a subset of the cluster if nodelist is not None.
'''
ret = 1
map = {}
if not nodelist:
nodelist = self.Env["nodes"]
for node in self.Env["nodes"]:
if self.ShouldBeStatus[node] == "up" or force == True:
if not self.StopaCM(node, verbose=verbose, force=force):
ret = 0
return ret
def 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.templates["BreakCommCmd"] % self.key_for_node(node))
if rc != 0:
self.logger.log(
"Could not break the communication between %s and %s: %d"
% (target, node, rc))
return None
else:
self.debug("Communication cut between %s and %s" %
(target, node))
return 1
def unisolate_node(self, target, nodes=None):
'''fix the communication between the nodes'''
if not nodes:
nodes = self.Env["nodes"]
for node in nodes:
if node != target:
restored = 0
# Limit the amount of time we have asynchronous connectivity for
# Restore both sides as simultaneously as possible
self.rsh(target,
self.templates["FixCommCmd"] %
self.key_for_node(node),
synchronous=0)
self.rsh(node,
self.templates["FixCommCmd"] %
self.key_for_node(target),
synchronous=0)
self.debug("Communication restored between %s and %s" %
(target, node))
def reducecomm_node(self, node):
'''reduce the communication between the nodes'''
rc = self.rsh(
node, self.templates["ReduceCommCmd"] %
(self.Env["XmitLoss"], self.Env["RecvLoss"]))
if rc == 0:
return 1
else:
self.logger.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.templates["RestoreCommCmd"])
if rc == 0:
return 1
else:
self.logger.log(
"Could not restore the communication between the nodes from node: %s"
% node)
return None
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:<directory>/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")
def StatsExtract(self):
if not self.Env["stats"]:
return
for host in self.Env["nodes"]:
log_stats_file = "%s/cts-stats.csv" % CTSvars.CRM_DAEMON_DIR
if host in has_log_stats:
self.rsh(
host,
'''bash %s %s stop''' % (log_stats_bin, log_stats_file))
(rc, lines) = self.rsh(host,
'''cat %s''' % log_stats_file,
stdout=2)
self.rsh(
host,
'''bash %s %s delete''' % (log_stats_bin, log_stats_file))
fname = "cts-stats-%d-nodes-%s.csv" % (len(
self.Env["nodes"]), host)
print("Extracted stats: %s" % fname)
fd = open(fname, "a")
fd.writelines(lines)
fd.close()
def StatsMark(self, testnum):
'''Mark the test number in the stats log'''
global has_log_stats
if not self.Env["stats"]:
return
for host in self.Env["nodes"]:
log_stats_file = "%s/cts-stats.csv" % CTSvars.CRM_DAEMON_DIR
if not host in has_log_stats:
global log_stats
global log_stats_bin
script = log_stats
#script = re.sub("\\\\", "\\\\", script)
script = re.sub('\"', '\\\"', script)
script = re.sub("'", "\'", script)
script = re.sub("`", "\`", script)
script = re.sub("\$", "\\\$", script)
self.debug("Installing %s on %s" % (log_stats_bin, host))
self.rsh(host,
'''echo "%s" > %s''' % (script, log_stats_bin),
silent=True)
self.rsh(
host,
'''bash %s %s delete''' % (log_stats_bin, log_stats_file))
has_log_stats[host] = 1
# Now mark it
self.rsh(host,
'''bash %s %s mark %s''' %
(log_stats_bin, log_stats_file, testnum),
synchronous=0)
def errorstoignore(self):
# At some point implement a more elegant solution that
# also produces a report at the end
""" Return a list of known error messages that should be ignored """
return PatternSelector().get_patterns(self.name, "BadNewsIgnore")
def install_config(self, node):
if not self.ns.WaitForNodeToComeUp(node):
self.log("Node %s is not up." % node)
return None
if not node in self.CIBsync and self.Env["ClobberCIB"] == 1:
self.CIBsync[node] = 1
self.rsh(node, "rm -f " + CTSvars.CRM_CONFIG_DIR + "/cib*")
# Only install the CIB on the first node, all the other ones will pick it up from there
if self.cib_installed == 1:
return None
self.cib_installed = 1
if self.Env["CIBfilename"] == None:
self.log("Installing Generated CIB on node %s" % (node))
self.cib.install(node)
else:
self.log("Installing CIB (%s) on node %s" %
(self.Env["CIBfilename"], node))
if 0 != self.rsh.cp(
self.Env["CIBfilename"], "root@" +
(self.templates["CIBfile"] % node)):
raise ValueError("Can not scp file to %s %d" % (node))
self.rsh(
node, "chown " + CTSvars.CRM_DAEMON_USER + " " +
CTSvars.CRM_CONFIG_DIR + "/cib.xml")
def prepare(self):
'''Finish the Initialization process. Prepare to test...'''
self.partitions_expected = 1
for node in self.Env["nodes"]:
self.ShouldBeStatus[node] = ""
if self.Env["experimental-tests"]:
self.unisolate_node(node)
self.StataCM(node)
def test_node_CM(self, node):
'''Report the status of the cluster manager on a given node'''
watchpats = []
watchpats.append("Current ping state: (S_IDLE|S_NOT_DC)")
watchpats.append(self.templates["Pat:NonDC_started"] % node)
watchpats.append(self.templates["Pat:DC_started"] % node)
idle_watch = LogWatcher(self.Env["LogFileName"],
watchpats,
"ClusterIdle",
hosts=[node],
kind=self.Env["LogWatcher"])
idle_watch.setwatch()
out = self.rsh(node, self.templates["StatusCmd"] % node, 1)
self.debug("Node %s status: '%s'" % (node, out))
if not out or (out.find('ok') < 0):
if self.ShouldBeStatus[node] == "up":
self.log(
"Node status for %s is %s but we think it should be %s" %
(node, "down", self.ShouldBeStatus[node]))
self.ShouldBeStatus[node] = "down"
return 0
if self.ShouldBeStatus[node] == "down":
self.log(
"Node status for %s is %s but we think it should be %s: %s" %
(node, "up", self.ShouldBeStatus[node], out))
self.ShouldBeStatus[node] = "up"
# check the output first - because syslog-ng loses messages
if out.find('S_NOT_DC') != -1:
# Up and stable
return 2
if out.find('S_IDLE') != -1:
# Up and stable
return 2
# fall back to syslog-ng and wait
if not idle_watch.look():
# just up
self.debug("Warn: Node %s is unstable: %s" % (node, out))
return 1
# Up and stable
return 2
# Is the node up or is the node down
def StataCM(self, node):
'''Report the status of the cluster manager on a given node'''
if self.test_node_CM(node) > 0:
return 1
return None
# Being up and being stable is not the same question...
def node_stable(self, node):
'''Report the status of the cluster manager on a given node'''
if self.test_node_CM(node) == 2:
return 1
self.log("Warn: Node %s not stable" % (node))
return None
def partition_stable(self, nodes, timeout=None):
watchpats = []
watchpats.append("Current ping state: S_IDLE")
watchpats.append(self.templates["Pat:DC_IDLE"])
self.debug("Waiting for cluster stability...")
if timeout == None:
timeout = self.Env["DeadTime"]
if len(nodes) < 3:
self.debug("Cluster is inactive")
return 1
idle_watch = LogWatcher(self.Env["LogFileName"],
watchpats,
"ClusterStable",
timeout,
hosts=nodes.split(),
kind=self.Env["LogWatcher"])
idle_watch.setwatch()
for node in nodes.split():
# have each node dump its current state
self.rsh(node, self.templates["StatusCmd"] % node, 1)
ret = idle_watch.look()
while ret:
self.debug(ret)
for node in nodes.split():
if re.search(node, ret):
return 1
ret = idle_watch.look()
self.debug("Warn: Partition %s not IDLE after %ds" %
(repr(nodes), timeout))
return None
def cluster_stable(self, timeout=None, double_check=False):
partitions = self.find_partitions()
for partition in partitions:
if not self.partition_stable(partition, timeout):
return None
if double_check:
# Make sure we are really stable and that all resources,
# including those that depend on transient node attributes,
# are started if they were going to be
time.sleep(5)
for partition in partitions:
if not self.partition_stable(partition, timeout):
return None
return 1
def is_node_dc(self, node, status_line=None):
rc = 0
if not status_line:
status_line = self.rsh(node, self.templates["StatusCmd"] % node, 1)
if not status_line:
rc = 0
elif status_line.find('S_IDLE') != -1:
rc = 1
elif status_line.find('S_INTEGRATION') != -1:
rc = 1
elif status_line.find('S_FINALIZE_JOIN') != -1:
rc = 1
elif status_line.find('S_POLICY_ENGINE') != -1:
rc = 1
elif status_line.find('S_TRANSITION_ENGINE') != -1:
rc = 1
return rc
def active_resources(self, node):
(rc, output) = self.rsh(node, """crm_resource -c""", None)
resources = []
for line in output:
if re.search("^Resource", line):
tmp = AuditResource(self, line)
if tmp.type == "primitive" and tmp.host == node:
resources.append(tmp.id)
return resources
def ResourceLocation(self, rid):
ResourceNodes = []
for node in self.Env["nodes"]:
if self.ShouldBeStatus[node] == "up":
cmd = self.templates["RscRunning"] % (rid)
(rc, lines) = self.rsh(node, cmd, None)
if rc == 127:
self.log(
"Command '%s' failed. Binary or pacemaker-cts package not installed?"
% cmd)
for line in lines:
self.log("Output: " + line)
elif rc == 0:
ResourceNodes.append(node)
return ResourceNodes
def find_partitions(self):
ccm_partitions = []
for node in self.Env["nodes"]:
if self.ShouldBeStatus[node] == "up":
partition = self.rsh(node, self.templates["PartitionCmd"], 1)
if not partition:
self.log("no partition details for %s" % node)
elif len(partition) > 2:
nodes = partition.split()
nodes.sort()
partition = ' '.join(nodes)
found = 0
for a_partition in ccm_partitions:
if partition == a_partition:
found = 1
if found == 0:
self.debug("Adding partition from %s: %s" %
(node, partition))
ccm_partitions.append(partition)
else:
self.debug(
"Partition '%s' from %s is consistent with existing entries"
% (partition, node))
else:
self.log("bad partition details for %s" % node)
else:
self.debug("Node %s is down... skipping" % node)
self.debug("Found partitions: %s" % repr(ccm_partitions))
return ccm_partitions
def HasQuorum(self, node_list):
# If we are auditing a partition, then one side will
# have quorum and the other not.
# So the caller needs to tell us which we are checking
# If no value for node_list is specified... assume all nodes
if not node_list:
node_list = self.Env["nodes"]
for node in node_list:
if self.ShouldBeStatus[node] == "up":
quorum = self.rsh(node, self.templates["QuorumCmd"], 1)
if quorum.find("1") != -1:
return 1
elif quorum.find("0") != -1:
return 0
else:
self.debug("WARN: Unexpected quorum test result from " +
node + ":" + quorum)
return 0
def Components(self):
complist = []
common_ignore = [
"Pending action:",
"(ERROR|error): crm_log_message_adv:",
"(ERROR|error): MSG: No message to dump",
"pending LRM operations at shutdown",
"Lost connection to the CIB manager",
"Connection to the CIB terminated...",
"Sending message to the CIB manager FAILED",
"Action A_RECOVER .* not supported",
"(ERROR|error): stonithd_op_result_ready: not signed on",
"pingd.*(ERROR|error): send_update: Could not send update",
"send_ipc_message: IPC Channel to .* is not connected",
"unconfirmed_actions: Waiting on .* unconfirmed actions",
"cib_native_msgready: Message pending on command channel",
r": Performing A_EXIT_1 - forcefully exiting ",
r"Resource .* was active at shutdown. You may ignore this error if it is unmanaged.",
]
stonith_ignore = [
r"Updating failcount for child_DoFencing",
r"error.*: Fencer connection failed \(will retry\)",
"pacemaker-execd.*(ERROR|error): stonithd_receive_ops_result failed.",
]
stonith_ignore.extend(common_ignore)
ccm = Process(
self,
"ccm",
triggersreboot=self.fastfail,
pats=[
"State transition .* S_RECOVERY",
"pacemaker-controld.*Action A_RECOVER .* not supported",
r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover",
r"pacemaker-controld.*: Could not recover from internal error",
"pacemaker-controld.*I_ERROR.*crmd_cib_connection_destroy",
# these status numbers are likely wrong now
r"pacemaker-controld.*exited with status 2",
r"attrd.*exited with status 1",
r"cib.*exited with status 2",
# Not if it was fenced
# "A new node joined the cluster",
# "WARN: determine_online_status: Node .* is unclean",
# "Scheduling Node .* for STONITH",
# "Executing .* fencing operation",
# "tengine_stonith_callback: .*result=0",
# "Processing I_NODE_JOIN:.* cause=C_HA_MESSAGE",
# "State transition S_.* -> S_INTEGRATION.*input=I_NODE_JOIN",
"State transition S_STARTING -> S_PENDING",
],
badnews_ignore=common_ignore)
based = Process(
self,
"pacemaker-based",
triggersreboot=self.fastfail,
pats=[
"State transition .* S_RECOVERY",
"Lost connection to the CIB manager",
"Connection to the CIB manager terminated",
r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover",
"pacemaker-controld.*I_ERROR.*crmd_cib_connection_destroy",
r"pacemaker-controld.*: Could not recover from internal error",
# these status numbers are likely wrong now
r"pacemaker-controld.*exited with status 2",
r"attrd.*exited with status 1",
],
badnews_ignore=common_ignore)
execd = Process(
self,
"pacemaker-execd",
triggersreboot=self.fastfail,
pats=[
"State transition .* S_RECOVERY",
"LRM Connection failed",
"pacemaker-controld.*I_ERROR.*lrm_connection_destroy",
"State transition S_STARTING -> S_PENDING",
r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover",
r"pacemaker-controld.*: Could not recover from internal error",
# this status number is likely wrong now
r"pacemaker-controld.*exited with status 2",
],
badnews_ignore=common_ignore)
controld = Process(
self,
"pacemaker-controld",
triggersreboot=self.fastfail,
pats=[
# "WARN: determine_online_status: Node .* is unclean",
# "Scheduling Node .* for STONITH",
# "Executing .* fencing operation",
# "tengine_stonith_callback: .*result=0",
"State transition .* S_IDLE",
"State transition S_STARTING -> S_PENDING",
],
badnews_ignore=common_ignore)
schedulerd = Process(
self,
"pacemaker-schedulerd",
triggersreboot=self.fastfail,
pats=[
"State transition .* S_RECOVERY",
r"pacemaker-controld.*: Input I_TERMINATE .*from do_recover",
r"pacemaker-controld.*: Could not recover from internal error",
r"pacemaker-controld.*CRIT.*: Connection to the scheduler failed",
"pacemaker-controld.*I_ERROR.*save_cib_contents",
# this status number is likely wrong now
r"pacemaker-controld.*exited with status 2",
],
badnews_ignore=common_ignore,
dc_only=1)
if self.Env["DoFencing"] == 1:
complist.append(
Process(
self,
"stoniths",
triggersreboot=self.fastfail,
dc_pats=[
r"pacemaker-controld.*CRIT.*: Fencing daemon connection failed",
"Attempting connection to fencing daemon",
],
badnews_ignore=stonith_ignore))
if self.fastfail == 0:
ccm.pats.extend([
# these status numbers are likely wrong now
r"attrd.*exited with status 1",
r"pacemaker-(based|controld).*exited with status 2",
])
based.pats.extend([
# these status numbers are likely wrong now
r"attrd.*exited with status 1",
r"pacemaker-controld.*exited with status 2",
])
execd.pats.extend([
# these status numbers are likely wrong now
r"pacemaker-controld.*exited with status 2",
])
complist.append(ccm)
complist.append(based)
complist.append(execd)
complist.append(controld)
complist.append(schedulerd)
return complist
def StandbyStatus(self, node):
out = self.rsh(node, self.templates["StandbyQueryCmd"] % node, 1)
if not out:
return "off"
out = out[:-1]
self.debug("Standby result: " + out)
return out
# status == "on" : Enter Standby mode
# status == "off": Enter Active mode
def SetStandbyMode(self, node, status):
current_status = self.StandbyStatus(node)
cmd = self.templates["StandbyCmd"] % (node, status)
ret = self.rsh(node, cmd)
return True
def AddDummyRsc(self, node, rid):
rsc_xml = """ '<resources>
<primitive class=\"ocf\" id=\"%s\" provider=\"pacemaker\" type=\"Dummy\">
<operations>
<op id=\"%s-interval-10s\" interval=\"10s\" name=\"monitor\"/
</operations>
</primitive>
</resources>'""" % (rid, rid)
constraint_xml = """ '<constraints>
<rsc_location id=\"location-%s-%s\" node=\"%s\" rsc=\"%s\" score=\"INFINITY\"/>
</constraints>'
""" % (rid, node, node, rid)
self.rsh(node, self.templates['CibAddXml'] % (rsc_xml))
self.rsh(node, self.templates['CibAddXml'] % (constraint_xml))
def RemoveDummyRsc(self, node, rid):
constraint = "\"//rsc_location[@rsc='%s']\"" % (rid)
rsc = "\"//primitive[@id='%s']\"" % (rid)
self.rsh(node, self.templates['CibDelXpath'] % constraint)
self.rsh(node, self.templates['CibDelXpath'] % rsc)
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 list(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 = EnvFactory().getInstance()
self.templates = PatternSelector(self.Env["Name"])
self.__InitialConditions()
self.logger = LogFactory()
self.clear_cache = 0
self.TestLoggingLevel=0
self.data = {}
self.name = self.Env["Name"]
self.rsh = RemoteFactory().getInstance()
self.ShouldBeStatus={}
self.ns = NodeStatus(self.Env)
self.OurNode = string.lower(os.uname()[1])
self.__instance_errorstoignore = []
def __getitem__(self, key):
if key == "Name":
return self.name
print("FIXME: Getting %s from %s" % (key, repr(self)))
if key in self.data:
return self.data[key]
return self.templates.get_patterns(self.Env["Name"], key)
def __setitem__(self, key, value):
print("FIXME: Setting %s=%s on %s" % (key, value, repr(self)))
self.data[key] = value
def key_for_node(self, node):
return node
def instance_errorstoignore_clear(self):
'''Allows the test scenario to reset instance errors to ignore on each iteration.'''
self.__instance_errorstoignore = []
def instance_errorstoignore(self):
'''Return list of errors which are 'normal' for a specific test instance'''
return self.__instance_errorstoignore
def errorstoignore(self):
'''Return list of errors which are 'normal' and should be ignored'''
return []
def log(self, args):
self.logger.log(args)
def debug(self, args):
self.logger.debug(args)
def prepare(self):
'''Finish the Initialization process. Prepare to test...'''
print(repr(self)+"prepare")
for node in self.Env["nodes"]:
if self.StataCM(node):
self.ShouldBeStatus[node] = "up"
else:
self.ShouldBeStatus[node] = "down"
self.unisolate_node(node)
def upcount(self):
'''How many nodes are up?'''
count = 0
for node in self.Env["nodes"]:
if self.ShouldBeStatus[node] == "up":
count = count + 1
return count
def install_helper(self, filename, destdir=None, nodes=None, sourcedir=None):
if sourcedir == None:
sourcedir = CTSvars.CTS_home
file_with_path = "%s/%s" % (sourcedir, filename)
if not nodes:
nodes = self.Env["nodes"]
if not destdir:
destdir = CTSvars.CTS_home
self.debug("Installing %s to %s on %s" % (filename, destdir, repr(self.Env["nodes"])))
for node in nodes:
self.rsh(node, "mkdir -p %s" % destdir)
self.rsh.cp(file_with_path, "root@%s:%s/%s" % (node, destdir, filename))
return file_with_path
def install_config(self, node):
return None
def clear_all_caches(self):
if self.clear_cache:
for node in self.Env["nodes"]:
if self.ShouldBeStatus[node] == "down":
self.debug("Removing cache file on: "+node)
self.rsh(node, "rm -f "+CTSvars.HA_VARLIBHBDIR+"/hostcache")
else:
self.debug("NOT Removing cache file on: "+node)
def prepare_fencing_watcher(self, name):
# If we don't have quorum now but get it as a result of starting this node,
# then a bunch of nodes might get fenced
upnode = None
if self.HasQuorum(None):
self.debug("Have quorum")
return None
if not self.templates["Pat:Fencing_start"]:
print("No start pattern")
return None
if not self.templates["Pat:Fencing_ok"]:
print("No ok pattern")
return None
stonith = None
stonithPats = []
for peer in self.Env["nodes"]:
if self.ShouldBeStatus[peer] != "up":
stonithPats.append(self.templates["Pat:Fencing_ok"] % peer)
stonithPats.append(self.templates["Pat:Fencing_start"] % peer)
elif self.Env["Stack"] == "corosync (cman)":
# There is a delay between gaining quorum and CMAN starting fencing
# This can mean that even nodes that are fully up get fenced
# There is no use fighting it, just look for everyone so that CTS doesn't get confused
stonithPats.append(self.templates["Pat:Fencing_ok"] % peer)
stonithPats.append(self.templates["Pat:Fencing_start"] % peer)
stonith = LogWatcher(self.Env["LogFileName"], stonithPats, "StartupFencing", 0, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"])
stonith.setwatch()
return stonith
def fencing_cleanup(self, node, stonith):
peer_list = []
peer_state = {}
self.debug("Looking for nodes that were fenced as a result of %s starting" % node)
# If we just started a node, we may now have quorum (and permission to fence)
if not stonith:
self.debug("Nothing to do")
return peer_list
q = self.HasQuorum(None)
if not q and len(self.Env["nodes"]) > 2:
# We didn't gain quorum - we shouldn't have shot anyone
self.debug("Quorum: %d Len: %d" % (q, len(self.Env["nodes"])))
return peer_list
for n in self.Env["nodes"]:
peer_state[n] = "unknown"
# Now see if any states need to be updated
self.debug("looking for: " + repr(stonith.regexes))
shot = stonith.look(0)
while shot:
line = repr(shot)
self.debug("Found: " + line)
del stonith.regexes[stonith.whichmatch]
# Extract node name
for n in self.Env["nodes"]:
if re.search(self.templates["Pat:Fencing_ok"] % n, shot):
peer = n
peer_state[peer] = "complete"
self.__instance_errorstoignore.append(self.templates["Pat:Fencing_ok"] % peer)
elif peer_state[n] != "complete" and re.search(self.templates["Pat:Fencing_start"] % n, shot):
# TODO: Correctly detect multiple fencing operations for the same host
peer = n
peer_state[peer] = "in-progress"
self.__instance_errorstoignore.append(self.templates["Pat:Fencing_start"] % peer)
if not peer:
self.logger.log("ERROR: Unknown stonith match: %s" % line)
elif not peer in peer_list:
self.debug("Found peer: " + peer)
peer_list.append(peer)
# Get the next one
shot = stonith.look(60)
for peer in peer_list:
self.debug(" Peer %s was fenced as a result of %s starting: %s" % (peer, node, peer_state[peer]))
if self.Env["at-boot"]:
self.ShouldBeStatus[peer] = "up"
else:
self.ShouldBeStatus[peer] = "down"
if peer_state[peer] == "in-progress":
# Wait for any in-progress operations to complete
shot = stonith.look(60)
while len(stonith.regexes) and shot:
line = repr(shot)
self.debug("Found: " + line)
del stonith.regexes[stonith.whichmatch]
shot = stonith.look(60)
# Now make sure the node is alive too
self.ns.WaitForNodeToComeUp(peer, self.Env["DeadTime"])
# Poll until it comes up
if self.Env["at-boot"]:
if not self.StataCM(peer):
time.sleep(self.Env["StartTime"])
if not self.StataCM(peer):
self.logger.log("ERROR: Peer %s failed to restart after being fenced" % peer)
return None
return peer_list
def StartaCM(self, node, verbose=False):
'''Start up the cluster manager on a given node'''
if verbose: self.logger.log("Starting %s on node %s" % (self.templates["Name"], node))
else: self.debug("Starting %s on node %s" % (self.templates["Name"], node))
ret = 1
if not node in self.ShouldBeStatus:
self.ShouldBeStatus[node] = "down"
if self.ShouldBeStatus[node] != "down":
return 1
patterns = []
# Technically we should always be able to notice ourselves starting
patterns.append(self.templates["Pat:Local_started"] % node)
if self.upcount() == 0:
patterns.append(self.templates["Pat:Master_started"] % node)
else:
patterns.append(self.templates["Pat:Slave_started"] % node)
watch = LogWatcher(
self.Env["LogFileName"], patterns, "StartaCM", self.Env["StartTime"]+10, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"])
self.install_config(node)
self.ShouldBeStatus[node] = "any"
if self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]):
self.logger.log ("%s was already started" % (node))
return 1
# 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.templates["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.templates["StartCmd"])
stonith = self.prepare_fencing_watcher(node)
watch.setwatch()
if self.rsh(node, startCmd) != 0:
self.logger.log ("Warn: Start command failed on node %s" % (node))
self.fencing_cleanup(node, stonith)
return None
self.ShouldBeStatus[node] = "up"
watch_result = watch.lookforall()
if watch.unmatched:
for regex in watch.unmatched:
self.logger.log ("Warn: Startup pattern not found: %s" % (regex))
if watch_result and self.cluster_stable(self.Env["DeadTime"]):
#self.debug("Found match: "+ repr(watch_result))
self.fencing_cleanup(node, stonith)
return 1
elif self.StataCM(node) and self.cluster_stable(self.Env["DeadTime"]):
self.fencing_cleanup(node, stonith)
return 1
self.logger.log ("Warn: Start failed for node %s" % (node))
return None
def StartaCMnoBlock(self, node, verbose=False):
'''Start up the cluster manager on a given node with none-block mode'''
if verbose: self.logger.log("Starting %s on node %s" % (self["Name"], node))
else: self.debug("Starting %s on node %s" % (self["Name"], node))
# Clear out the host cache so autojoin can be exercised
if self.clear_cache:
self.debug("Removing cache file on: "+node)
self.rsh(node, "rm -f "+CTSvars.HA_VARLIBHBDIR+"/hostcache")
self.install_config(node)
if not(self.Env["valgrind-tests"]):
startCmd = self.templates["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.templates["StartCmd"])
self.rsh(node, startCmd, synchronous=0)
self.ShouldBeStatus[node] = "up"
return 1
def StopaCM(self, node, verbose=False, force=False):
'''Stop the cluster manager on a given node'''
if verbose: self.logger.log("Stopping %s on node %s" % (self["Name"], node))
else: self.debug("Stopping %s on node %s" % (self["Name"], node))
if self.ShouldBeStatus[node] != "up" and force == False:
return 1
if self.rsh(node, self.templates["StopCmd"]) == 0:
# Make sure we can continue even if corosync leaks
# fdata-* is the old name
#self.rsh(node, "rm -f /dev/shm/qb-* /dev/shm/fdata-*")
self.ShouldBeStatus[node] = "down"
self.cluster_stable(self.Env["DeadTime"])
return 1
else:
self.logger.log ("ERROR: Could not stop %s on node %s" % (self["Name"], node))
return None
def StopaCMnoBlock(self, node):
'''Stop the cluster manager on a given node with none-block mode'''
self.debug("Stopping %s on node %s" % (self["Name"], node))
self.rsh(node, self.templates["StopCmd"], synchronous=0)
self.ShouldBeStatus[node] = "down"
return 1
def cluster_stable(self, timeout = None):
time.sleep(self.Env["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.templates["RereadCmd"])
if rc == 0:
return 1
else:
self.logger.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.templates["StatusCmd"] % node, 1)
ret= (string.find(out, 'stopped') == -1)
try:
if ret:
if self.ShouldBeStatus[node] == "down":
self.logger.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.logger.log(
"Node status for %s is %s but we think it should be %s"
% (node, "down", self.ShouldBeStatus[node]))
except KeyError: pass
if ret:
self.ShouldBeStatus[node] = "up"
else:
self.ShouldBeStatus[node] = "down"
return ret
def startall(self, nodelist=None, verbose=False, quick=False):
'''Start the cluster manager on every node in the cluster.
We can do it on a subset of the cluster if nodelist is not None.
'''
map = {}
if not nodelist:
nodelist = self.Env["nodes"]
for node in nodelist:
if self.ShouldBeStatus[node] == "down":
self.ns.WaitForAllNodesToComeUp(nodelist, 300)
if not quick:
if not self.StartaCM(node, verbose=verbose):
return 0
return 1
# Approximation of SimulStartList for --boot
watchpats = [ ]
watchpats.append(self.templates["Pat:DC_IDLE"])
for node in nodelist:
watchpats.append(self.templates["Pat:Local_started"] % node)
watchpats.append(self.templates["Pat:InfraUp"] % node)
watchpats.append(self.templates["Pat:PacemakerUp"] % node)
# Start all the nodes - at about the same time...
watch = LogWatcher(self.Env["LogFileName"], watchpats, "fast-start", self.Env["DeadTime"]+10, hosts=self.Env["nodes"], kind=self.Env["LogWatcher"])
watch.setwatch()
if not self.StartaCM(nodelist[0], verbose=verbose):
return 0
for node in nodelist:
self.StartaCMnoBlock(node, verbose=verbose)
watch.lookforall()
if watch.unmatched:
for regex in watch.unmatched:
self.logger.log ("Warn: Startup pattern not found: %s" % (regex))
if not self.cluster_stable():
self.logger.log("Cluster did not stabilize")
return 0
return 1
def stopall(self, nodelist=None, verbose=False, force=False):
'''Stop the cluster managers on every node in the cluster.
We can do it on a subset of the cluster if nodelist is not None.
'''
ret = 1
map = {}
if not nodelist:
nodelist = self.Env["nodes"]
for node in self.Env["nodes"]:
if self.ShouldBeStatus[node] == "up" or force == True:
if not self.StopaCM(node, verbose=verbose, force=force):
ret = 0
return ret
def 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.templates["BreakCommCmd"] % self.key_for_node(node))
if rc != 0:
self.logger.log("Could not break the communication between %s and %s: %d" % (target, node, rc))
return None
else:
self.debug("Communication cut between %s and %s" % (target, node))
return 1
def unisolate_node(self, target, nodes=None):
'''fix the communication between the nodes'''
if not nodes:
nodes = self.Env["nodes"]
for node in nodes:
if node != target:
restored = 0
# Limit the amount of time we have asynchronous connectivity for
# Restore both sides as simultaneously as possible
self.rsh(target, self.templates["FixCommCmd"] % self.key_for_node(node), synchronous=0)
self.rsh(node, self.templates["FixCommCmd"] % self.key_for_node(target), synchronous=0)
self.debug("Communication restored between %s and %s" % (target, node))
def reducecomm_node(self,node):
'''reduce the communication between the nodes'''
rc = self.rsh(node, self.templates["ReduceCommCmd"]%(self.Env["XmitLoss"],self.Env["RecvLoss"]))
if rc == 0:
return 1
else:
self.logger.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.templates["RestoreCommCmd"]);
if rc == 0:
return 1
else:
self.logger.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")
def StatsExtract(self):
if not self.Env["stats"]:
return
for host in self.Env["nodes"]:
log_stats_file = "%s/cts-stats.csv" % CTSvars.CRM_DAEMON_DIR
if host in has_log_stats:
self.rsh(host, '''bash %s %s stop''' % (log_stats_bin, log_stats_file))
(rc, lines) = self.rsh(host, '''cat %s''' % log_stats_file, stdout=2)
self.rsh(host, '''bash %s %s delete''' % (log_stats_bin, log_stats_file))
fname = "cts-stats-%d-nodes-%s.csv" % (len(self.Env["nodes"]), host)
print("Extracted stats: %s" % fname)
fd = open(fname, "a")
fd.writelines(lines)
fd.close()
def StatsMark(self, testnum):
'''Mark the test number in the stats log'''
global has_log_stats
if not self.Env["stats"]:
return
for host in self.Env["nodes"]:
log_stats_file = "%s/cts-stats.csv" % CTSvars.CRM_DAEMON_DIR
if not host in has_log_stats:
global log_stats
global log_stats_bin
script=log_stats
#script = re.sub("\\\\", "\\\\", script)
script = re.sub('\"', '\\\"', script)
script = re.sub("'", "\'", script)
script = re.sub("`", "\`", script)
script = re.sub("\$", "\\\$", script)
self.debug("Installing %s on %s" % (log_stats_bin, host))
self.rsh(host, '''echo "%s" > %s''' % (script, log_stats_bin), silent=True)
self.rsh(host, '''bash %s %s delete''' % (log_stats_bin, log_stats_file))
has_log_stats[host] = 1
# Now mark it
self.rsh(host, '''bash %s %s mark %s''' % (log_stats_bin, log_stats_file, testnum), synchronous=0)
from cts.CM_ais import *
from cts.CM_lha import crm_lha
from cts.CTSaudits import AuditList
from cts.CTStests import TestList
from cts.CTSscenarios import *
from cts.logging import LogFactory
except ImportError as e:
sys.stderr.write("abort: %s\n" % e)
sys.stderr.write("check your install and PYTHONPATH; couldn't find cts libraries in:\n%s\n" %
' '.join(sys.path))
sys.exit(1)
# These are globals so they can be used by the signal handler.
cm = None
scenario = None
LogFactory().add_stderr()
def sig_handler(signum, frame) :
LogFactory().log("Interrupted by signal %d"%signum)
if scenario: scenario.summarize()
if signum == 15 :
if scenario: scenario.TearDown()
sys.exit(1)
if __name__ == '__main__':
Environment = CtsLab(sys.argv[1:])
NumIter = Environment["iterations"]
Tests = []
def __init__(self, environment, verbose=True):
self.rsh = RemoteFactory().getInstance()
self.logger = LogFactory()
self.Env = environment
self.verbose = verbose
self.ratemplates = RATemplates()
