X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/5188ab3773f1b4d11a82bafda3a0a91bf54f3c7e..c8fcde472922e4ee664d904e0bf1a583f1d5040d:/daemons/ganeti-watcher?ds=sidebyside diff --git a/daemons/ganeti-watcher b/daemons/ganeti-watcher index 94c6b48..42a2eaf 100755 --- a/daemons/ganeti-watcher +++ b/daemons/ganeti-watcher @@ -30,24 +30,21 @@ by a node reboot. Run from cron or similar. import os import sys import time -import fcntl -import errno import logging from optparse import OptionParser from ganeti import utils from ganeti import constants from ganeti import serializer -from ganeti import ssconf from ganeti import errors from ganeti import opcodes -from ganeti import logger from ganeti import cli +from ganeti import luxi MAXTRIES = 5 -BAD_STATES = ['stopped'] -HELPLESS_STATES = ['(node down)'] +BAD_STATES = ['ERROR_down'] +HELPLESS_STATES = ['ERROR_nodedown', 'ERROR_nodeoffline'] NOTICE = 'NOTICE' ERROR = 'ERROR' KEY_RESTART_COUNT = "restart_count" @@ -66,14 +63,23 @@ class NotMasterError(errors.GenericError): def Indent(s, prefix='| '): """Indent a piece of text with a given prefix before each line. - Args: - s: The string to indent - prefix: The string to prepend each line. + @param s: the string to indent + @param prefix: the string to prepend each line """ return "%s%s\n" % (prefix, ('\n' + prefix).join(s.splitlines())) +def StartMaster(): + """Try to start the master daemon. + + """ + result = utils.RunCmd(['ganeti-masterd']) + if result.failed: + logging.error("Can't start the master daemon: output '%s'", result.output) + return not result.failed + + class WatcherState(object): """Interface to a state file recording restart attempts. @@ -93,11 +99,15 @@ class WatcherState(object): utils.LockFile(self.statefile.fileno()) try: - self._data = serializer.Load(self.statefile.read()) + state_data = self.statefile.read() + if not state_data: + self._data = {} + else: + self._data = serializer.Load(state_data) except Exception, msg: # Ignore errors while loading the file and treat it as empty self._data = {} - logging.warning(("Empty or invalid state file. Using defaults." + logging.warning(("Invalid state file. Using defaults." " Error message: %s"), msg) if "instance" not in self._data: @@ -162,8 +172,8 @@ class WatcherState(object): def NumberOfRestartAttempts(self, instance): """Returns number of previous restart attempts. - Args: - instance - the instance to look up. + @type instance: L{Instance} + @param instance: the instance to look up """ idata = self._data["instance"] @@ -176,8 +186,8 @@ class WatcherState(object): def RecordRestartAttempt(self, instance): """Record a restart attempt. - Args: - instance - the instance being restarted + @type instance: L{Instance} + @param instance: the instance being restarted """ idata = self._data["instance"] @@ -191,12 +201,13 @@ class WatcherState(object): inst[KEY_RESTART_COUNT] = inst.get(KEY_RESTART_COUNT, 0) + 1 def RemoveInstance(self, instance): - """Update state to reflect that a machine is running, i.e. remove record. + """Update state to reflect that a machine is running. - Args: - instance - the instance to remove from books + This method removes the record for a named instance (as we only + track down instances). - This method removes the record for a named instance. + @type instance: L{Instance} + @param instance: the instance to remove from books """ idata = self._data["instance"] @@ -208,9 +219,6 @@ class WatcherState(object): class Instance(object): """Abstraction for a Virtual Machine instance. - Methods: - Restart(): issue a command to restart the represented machine. - """ def __init__(self, name, state, autostart): self.name = name @@ -221,9 +229,7 @@ class Instance(object): """Encapsulates the start of an instance. """ - op = opcodes.OpStartupInstance(instance_name=self.name, - force=False, - extra_args=None) + op = opcodes.OpStartupInstance(instance_name=self.name, force=False) cli.SubmitOpCode(op, cl=client) def ActivateDisks(self): @@ -234,45 +240,42 @@ class Instance(object): cli.SubmitOpCode(op, cl=client) -def GetInstanceList(with_secondaries=None): +def GetClusterData(): """Get a list of instances on this cluster. """ - fields = ["name", "oper_state", "admin_state"] - - if with_secondaries is not None: - fields.append("snodes") + op1_fields = ["name", "status", "admin_state", "snodes"] + op1 = opcodes.OpQueryInstances(output_fields=op1_fields, names=[], + use_locking=True) + op2_fields = ["name", "bootid", "offline"] + op2 = opcodes.OpQueryNodes(output_fields=op2_fields, names=[], + use_locking=True) - result = client.QueryInstances([], fields) + job_id = client.SubmitJob([op1, op2]) - instances = [] - for fields in result: - if with_secondaries is not None: - (name, status, autostart, snodes) = fields + all_results = cli.PollJob(job_id, cl=client, feedback_fn=logging.debug) - if not snodes: - continue + result = all_results[0] + smap = {} - for node in with_secondaries: - if node in snodes: - break - else: - continue - - else: - (name, status, autostart) = fields + instances = {} + for fields in result: + (name, status, autostart, snodes) = fields - instances.append(Instance(name, status, autostart)) + # update the secondary node map + for node in snodes: + if node not in smap: + smap[node] = [] + smap[node].append(name) - return instances + instances[name] = Instance(name, status, autostart) + nodes = dict([(name, (bootid, offline)) + for name, bootid, offline in all_results[1]]) -def GetNodeBootIDs(): - """Get a dict mapping nodes to boot IDs. + client.ArchiveJob(job_id) - """ - result = client.QueryNodes([], ["name", "bootid"]) - return dict([(name, bootid) for name, bootid in result]) + return instances, nodes, smap class Watcher(object): @@ -283,31 +286,45 @@ class Watcher(object): to restart machines that are down. """ - def __init__(self): - sstore = ssconf.SimpleStore() - master = sstore.GetMasterNode() + def __init__(self, opts, notepad): + self.notepad = notepad + master = client.QueryConfigValues(["master_node"])[0] if master != utils.HostInfo().name: raise NotMasterError("This is not the master node") - self.instances = GetInstanceList() - self.bootids = GetNodeBootIDs() + self.instances, self.bootids, self.smap = GetClusterData() self.started_instances = set() + self.opts = opts def Run(self): - notepad = WatcherState() - try: - self.CheckInstances(notepad) - self.CheckDisks(notepad) - self.VerifyDisks() - finally: - notepad.Save() + """Watcher run sequence. + + """ + notepad = self.notepad + self.ArchiveJobs(self.opts.job_age) + self.CheckInstances(notepad) + self.CheckDisks(notepad) + self.VerifyDisks() + + def ArchiveJobs(self, age): + """Archive old jobs. + + """ + arch_count, left_count = client.AutoArchiveJobs(age) + logging.debug("Archived %s jobs, left %s" % (arch_count, left_count)) def CheckDisks(self, notepad): """Check all nodes for restarted ones. """ check_nodes = [] - for name, new_id in self.bootids.iteritems(): + for name, (new_id, offline) in self.bootids.iteritems(): old = notepad.GetNodeBootID(name) + if new_id is None: + # Bad node, not returning a boot id + if not offline: + logging.debug("Node %s missing boot id, skipping secondary checks", + name) + continue if old != new_id: # Node's boot ID has changed, proably through a reboot. check_nodes.append(name) @@ -315,34 +332,35 @@ class Watcher(object): if check_nodes: # Activate disks for all instances with any of the checked nodes as a # secondary node. - for instance in GetInstanceList(with_secondaries=check_nodes): - if not instance.autostart: - logging.info(("Skipping disk activation for non-autostart" - " instance %s"), instance.name) - continue - if instance.name in self.started_instances: - # we already tried to start the instance, which should have - # activated its drives (if they can be at all) + for node in check_nodes: + if node not in self.smap: continue - try: - logging.info("Activating disks for instance %s", instance.name) - instance.ActivateDisks() - except Exception, err: - logging.error(str(err), exc_info=True) + for instance_name in self.smap[node]: + instance = self.instances[instance_name] + if not instance.autostart: + logging.info(("Skipping disk activation for non-autostart" + " instance %s"), instance.name) + continue + if instance.name in self.started_instances: + # we already tried to start the instance, which should have + # activated its drives (if they can be at all) + continue + try: + logging.info("Activating disks for instance %s", instance.name) + instance.ActivateDisks() + except Exception: + logging.exception("Error while activating disks for instance %s", + instance.name) # Keep changed boot IDs for name in check_nodes: - notepad.SetNodeBootID(name, self.bootids[name]) + notepad.SetNodeBootID(name, self.bootids[name][0]) def CheckInstances(self, notepad): """Make a pass over the list of instances, restarting downed ones. """ - for instance in self.instances: - # Don't care about manually stopped instances - if not instance.autostart: - continue - + for instance in self.instances.values(): if instance.state in BAD_STATES: n = notepad.NumberOfRestartAttempts(instance) @@ -361,8 +379,9 @@ class Watcher(object): instance.name, last) instance.Restart() self.started_instances.add(instance.name) - except Exception, err: - logging.error(str(err), exc_info=True) + except Exception: + logging.exception("Error while restarting instance %s", + instance.name) notepad.RecordRestartAttempt(instance) elif instance.state in HELPLESS_STATES: @@ -373,12 +392,15 @@ class Watcher(object): notepad.RemoveInstance(instance) logging.info("Restart of %s succeeded", instance.name) - def VerifyDisks(self): + @staticmethod + def VerifyDisks(): """Run gnt-cluster verify-disks. """ op = opcodes.OpVerifyDisks() - result = cli.SubmitOpCode(op, cl=client) + job_id = client.SubmitJob([op]) + result = cli.PollJob(job_id, cl=client, feedback_fn=logging.debug)[0] + client.ArchiveJob(job_id) if not isinstance(result, (tuple, list)): logging.error("Can't get a valid result from verify-disks") return @@ -400,8 +422,7 @@ class Watcher(object): def ParseOptions(): """Parse the command line options. - Returns: - (options, args) as from OptionParser.parse_args() + @return: (options, args) as from OptionParser.parse_args() """ parser = OptionParser(description="Ganeti cluster watcher", @@ -412,7 +433,11 @@ def ParseOptions(): parser.add_option("-d", "--debug", dest="debug", help="Write all messages to stderr", default=False, action="store_true") + parser.add_option("-A", "--job-age", dest="job_age", + help="Autoarchive jobs older than this age (default" + " 6 hours)", default=6*3600) options, args = parser.parse_args() + options.job_age = cli.ParseTimespec(options.job_age) return options, args @@ -424,19 +449,41 @@ def main(): options, args = ParseOptions() - logger.SetupLogging(constants.LOG_WATCHER, debug=options.debug, - stderr_logging=options.debug) + utils.SetupLogging(constants.LOG_WATCHER, debug=options.debug, + stderr_logging=options.debug) + update_file = True try: - client = cli.GetClient() - + notepad = WatcherState() try: - watcher = Watcher() - except errors.ConfigurationError: - # Just exit if there's no configuration - sys.exit(constants.EXIT_SUCCESS) - - watcher.Run() + try: + client = cli.GetClient() + except errors.OpPrereqError: + # this is, from cli.GetClient, a not-master case + logging.debug("Not on master, exiting") + sys.exit(constants.EXIT_SUCCESS) + except luxi.NoMasterError, err: + logging.warning("Master seems to be down (%s), trying to restart", + str(err)) + if not StartMaster(): + logging.critical("Can't start the master, exiting") + update_file = False + sys.exit(constants.EXIT_FAILURE) + # else retry the connection + client = cli.GetClient() + + try: + watcher = Watcher(options, notepad) + except errors.ConfigurationError: + # Just exit if there's no configuration + sys.exit(constants.EXIT_SUCCESS) + + watcher.Run() + finally: + if update_file: + notepad.Save() + else: + logging.debug("Not updating status file due to failure") except SystemExit: raise except NotMasterError: