# 02110-1301, USA.
-"""Tool to restart erronously downed virtual machines.
+"""Tool to restart erroneously downed virtual machines.
This program and set of classes implement a watchdog to restart
virtual machines in a Ganeti cluster that have crashed or been killed
"""
+# pylint: disable-msg=C0103,W0142
+
+# C0103: Invalid name ganeti-watcher
+
import os
import sys
import time
import logging
-import errno
from optparse import OptionParser
from ganeti import utils
from ganeti import opcodes
from ganeti import cli
from ganeti import luxi
+from ganeti import ssconf
+from ganeti import bdev
+from ganeti import hypervisor
+from ganeti.confd import client as confd_client
MAXTRIES = 5
"""Exception raised when this host is not the master."""
-def Indent(s, prefix='| '):
- """Indent a piece of text with a given prefix before each line.
+def ShouldPause():
+ """Check whether we should pause.
+
+ """
+ return bool(utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE))
+
- @param s: the string to indent
- @param prefix: the string to prepend each line
+def StartNodeDaemons():
+ """Start all the daemons that should be running on all nodes.
"""
- return "%s%s\n" % (prefix, ('\n' + prefix).join(s.splitlines()))
+ # on master or not, try to start the node daemon
+ utils.EnsureDaemon(constants.NODED)
+ # start confd as well. On non candidates it will be in disabled mode.
+ utils.EnsureDaemon(constants.CONFD)
-def ShouldPause():
- """Check whether we should pause.
+def RunWatcherHooks():
+ """Run the watcher hooks.
"""
- return bool(utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE))
+ hooks_dir = utils.PathJoin(constants.HOOKS_BASE_DIR,
+ constants.HOOKS_NAME_WATCHER)
+ if not os.path.isdir(hooks_dir):
+ return
+
+ try:
+ results = utils.RunParts(hooks_dir)
+ except Exception, msg: # pylint: disable-msg=W0703
+ logging.critical("RunParts %s failed: %s", hooks_dir, msg)
+
+ for (relname, status, runresult) in results:
+ if status == constants.RUNPARTS_SKIP:
+ logging.debug("Watcher hook %s: skipped", relname)
+ elif status == constants.RUNPARTS_ERR:
+ logging.warning("Watcher hook %s: error (%s)", relname, runresult)
+ elif status == constants.RUNPARTS_RUN:
+ if runresult.failed:
+ logging.warning("Watcher hook %s: failed (exit: %d) (output: %s)",
+ relname, runresult.exit_code, runresult.output)
+ else:
+ logging.debug("Watcher hook %s: success (output: %s)", relname,
+ runresult.output)
-def EnsureDaemon(name):
- """Check for and start daemon if not alive.
+class NodeMaintenance(object):
+ """Talks to confd daemons and possible shutdown instances/drbd devices.
"""
- result = utils.RunCmd([constants.DAEMON_UTIL, "check-and-start", name])
- if result.failed:
- logging.error("Can't start daemon '%s', failure %s, output: %s",
- name, result.fail_reason, result.output)
- return False
+ def __init__(self):
+ self.store_cb = confd_client.StoreResultCallback()
+ self.filter_cb = confd_client.ConfdFilterCallback(self.store_cb)
+ self.confd_client = confd_client.GetConfdClient(self.filter_cb)
+
+ @staticmethod
+ def ShouldRun():
+ """Checks whether node maintenance should run.
+
+ """
+ try:
+ return ssconf.SimpleStore().GetMaintainNodeHealth()
+ except errors.ConfigurationError, err:
+ logging.error("Configuration error, not activating node maintenance: %s",
+ err)
+ return False
+
+ @staticmethod
+ def GetRunningInstances():
+ """Compute list of hypervisor/running instances.
+
+ """
+ hyp_list = ssconf.SimpleStore().GetHypervisorList()
+ results = []
+ for hv_name in hyp_list:
+ try:
+ hv = hypervisor.GetHypervisor(hv_name)
+ ilist = hv.ListInstances()
+ results.extend([(iname, hv_name) for iname in ilist])
+ except: # pylint: disable-msg=W0702
+ logging.error("Error while listing instances for hypervisor %s",
+ hv_name, exc_info=True)
+ return results
+
+ @staticmethod
+ def GetUsedDRBDs():
+ """Get list of used DRBD minors.
+
+ """
+ return bdev.DRBD8.GetUsedDevs().keys()
+
+ @classmethod
+ def DoMaintenance(cls, role):
+ """Maintain the instance list.
- return True
+ """
+ if role == constants.CONFD_NODE_ROLE_OFFLINE:
+ inst_running = cls.GetRunningInstances()
+ cls.ShutdownInstances(inst_running)
+ drbd_running = cls.GetUsedDRBDs()
+ cls.ShutdownDRBD(drbd_running)
+ else:
+ logging.debug("Not doing anything for role %s", role)
+
+ @staticmethod
+ def ShutdownInstances(inst_running):
+ """Shutdown running instances.
+
+ """
+ names_running = set([i[0] for i in inst_running])
+ if names_running:
+ logging.info("Following instances should not be running,"
+ " shutting them down: %s", utils.CommaJoin(names_running))
+ # this dictionary will collapse duplicate instance names (only
+ # xen pvm/vhm) into a single key, which is fine
+ i2h = dict(inst_running)
+ for name in names_running:
+ hv_name = i2h[name]
+ hv = hypervisor.GetHypervisor(hv_name)
+ hv.StopInstance(None, force=True, name=name)
+
+ @staticmethod
+ def ShutdownDRBD(drbd_running):
+ """Shutdown active DRBD devices.
+
+ """
+ if drbd_running:
+ logging.info("Following DRBD minors should not be active,"
+ " shutting them down: %s", utils.CommaJoin(drbd_running))
+ for minor in drbd_running:
+ # pylint: disable-msg=W0212
+ # using the private method as is, pending enhancements to the DRBD
+ # interface
+ bdev.DRBD8._ShutdownAll(minor)
+
+ def Exec(self):
+ """Check node status versus cluster desired state.
+
+ """
+ my_name = utils.HostInfo().name
+ req = confd_client.ConfdClientRequest(type=
+ constants.CONFD_REQ_NODE_ROLE_BYNAME,
+ query=my_name)
+ self.confd_client.SendRequest(req, async=False, coverage=-1)
+ timed_out, _, _ = self.confd_client.WaitForReply(req.rsalt)
+ if not timed_out:
+ # should have a valid response
+ status, result = self.store_cb.GetResponse(req.rsalt)
+ assert status, "Missing result but received replies"
+ if not self.filter_cb.consistent[req.rsalt]:
+ logging.warning("Inconsistent replies, not doing anything")
+ return
+ self.DoMaintenance(result.server_reply.answer)
+ else:
+ logging.warning("Confd query timed out, cannot do maintenance actions")
class WatcherState(object):
"""Interface to a state file recording restart attempts.
"""
- def __init__(self):
+ def __init__(self, statefile):
"""Open, lock, read and parse the file.
- Raises exception on lock contention.
+ @type statefile: file
+ @param statefile: State file object
"""
- # The two-step dance below is necessary to allow both opening existing
- # file read/write and creating if not existing. Vanilla open will truncate
- # an existing file -or- allow creating if not existing.
- fd = os.open(constants.WATCHER_STATEFILE, os.O_RDWR | os.O_CREAT)
- self.statefile = os.fdopen(fd, 'w+')
-
- utils.LockFile(self.statefile.fileno())
+ self.statefile = statefile
try:
state_data = self.statefile.read()
self._data = {}
else:
self._data = serializer.Load(state_data)
- except Exception, msg:
+ except Exception, msg: # pylint: disable-msg=W0703
# Ignore errors while loading the file and treat it as empty
self._data = {}
logging.warning(("Invalid state file. Using defaults."
class Watcher(object):
- """Encapsulate the logic for restarting erronously halted virtual machines.
+ """Encapsulate the logic for restarting erroneously halted virtual machines.
The calling program should periodically instantiate me and call Run().
This will traverse the list of instances, and make up to MAXTRIES attempts
"""
arch_count, left_count = client.AutoArchiveJobs(age)
- logging.debug("Archived %s jobs, left %s" % (arch_count, left_count))
+ logging.debug("Archived %s jobs, left %s", arch_count, left_count)
def CheckDisks(self, notepad):
"""Check all nodes for restarted ones.
try:
logging.info("Activating disks for instance %s", instance.name)
instance.ActivateDisks()
- except Exception:
+ except Exception: # pylint: disable-msg=W0703
logging.exception("Error while activating disks for instance %s",
instance.name)
instance.name, last)
instance.Restart()
self.started_instances.add(instance.name)
- except Exception:
+ except Exception: # pylint: disable-msg=W0703
logging.exception("Error while restarting instance %s",
instance.name)
# nothing to do
return
logging.debug("Will activate disks for instances %s",
- ", ".join(offline_disk_instances))
+ utils.CommaJoin(offline_disk_instances))
# we submit only one job, and wait for it. not optimal, but spams
# less the job queue
job = [opcodes.OpActivateInstanceDisks(instance_name=name)
for name in offline_disk_instances]
job_id = cli.SendJob(job, cl=client)
- cli.PollJob(job_id, cl=client, feedback_fn=logging.debug)
+ try:
+ cli.PollJob(job_id, cl=client, feedback_fn=logging.debug)
+ except Exception: # pylint: disable-msg=W0703
+ logging.exception("Error while activating disks")
+
+
+def OpenStateFile(path):
+ """Opens the state file and acquires a lock on it.
+
+ @type path: string
+ @param path: Path to state file
+
+ """
+ # The two-step dance below is necessary to allow both opening existing
+ # file read/write and creating if not existing. Vanilla open will truncate
+ # an existing file -or- allow creating if not existing.
+ statefile_fd = os.open(path, os.O_RDWR | os.O_CREAT)
+
+ # Try to acquire lock on state file. If this fails, another watcher instance
+ # might already be running or another program is temporarily blocking the
+ # watcher from running.
+ try:
+ utils.LockFile(statefile_fd)
+ except errors.LockError, err:
+ logging.error("Can't acquire lock on state file %s: %s", path, err)
+ return None
+
+ return os.fdopen(statefile_fd, "w+")
def ParseOptions():
"""Main function.
"""
- global client
+ global client # pylint: disable-msg=W0603
options, args = ParseOptions()
+ if args: # watcher doesn't take any arguments
+ print >> sys.stderr, ("Usage: %s [-f] " % sys.argv[0])
+ sys.exit(constants.EXIT_FAILURE)
+
utils.SetupLogging(constants.LOG_WATCHER, debug=options.debug,
stderr_logging=options.debug)
logging.debug("Pause has been set, exiting")
sys.exit(constants.EXIT_SUCCESS)
+ statefile = OpenStateFile(constants.WATCHER_STATEFILE)
+ if not statefile:
+ sys.exit(constants.EXIT_FAILURE)
+
update_file = False
try:
- # on master or not, try to start the node dameon
- EnsureDaemon(constants.NODED)
-
- notepad = WatcherState()
+ StartNodeDaemons()
+ RunWatcherHooks()
+ # run node maintenance in all cases, even if master, so that old
+ # masters can be properly cleaned up too
+ if NodeMaintenance.ShouldRun():
+ NodeMaintenance().Exec()
+
+ notepad = WatcherState(statefile)
try:
try:
client = cli.GetClient()
except luxi.NoMasterError, err:
logging.warning("Master seems to be down (%s), trying to restart",
str(err))
- if not EnsureDaemon(constants.MASTERD):
+ if not utils.EnsureDaemon(constants.MASTERD):
logging.critical("Can't start the master, exiting")
sys.exit(constants.EXIT_FAILURE)
# else retry the connection
client = cli.GetClient()
# we are on master now
- EnsureDaemon(constants.RAPI)
+ utils.EnsureDaemon(constants.RAPI)
try:
watcher = Watcher(options, notepad)
except errors.JobQueueDrainError:
logging.error("Job queue is drained, can't maintain cluster state")
except Exception, err:
- logging.error(str(err), exc_info=True)
+ logging.exception(str(err))
sys.exit(constants.EXIT_FAILURE)
projects / ganeti-local / blobdiff