#!/usr/bin/python
#
-# Copyright (C) 2006, 2007 Google Inc.
+# Copyright (C) 2006, 2007, 2008 Google Inc.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
import os
import sys
-import re
import time
import fcntl
import errno
-import simplejson
+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
MAXTRIES = 5
KEY_BOOT_ID = "bootid"
-class Error(Exception):
- """Generic custom error class."""
+# Global client object
+client = None
-class NotMasterError(Error):
+class NotMasterError(errors.GenericError):
"""Exception raised when this host is not the master."""
res = utils.RunCmd(cmd)
if res.failed:
- raise Error("Command %s failed:\n%s\nstdout:\n%sstderr:\n%s" %
- (repr(cmd),
- Indent(res.fail_reason),
- Indent(res.stdout),
- Indent(res.stderr)))
+ msg = ("Command %s failed:\n%s\nstdout:\n%sstderr:\n%s" %
+ (repr(cmd),
+ Indent(res.fail_reason),
+ Indent(res.stdout),
+ Indent(res.stderr)))
+ raise errors.CommandError(msg)
return res
def __init__(self):
"""Open, lock, read and parse the file.
- Raises StandardError on lock contention.
+ Raises exception on lock contention.
"""
# 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.
- f = os.open(constants.WATCHER_STATEFILE, os.O_RDWR | os.O_CREAT)
- f = os.fdopen(f, 'w+')
+ fd = os.open(constants.WATCHER_STATEFILE, os.O_RDWR | os.O_CREAT)
+ self.statefile = os.fdopen(fd, 'w+')
- try:
- fcntl.flock(f.fileno(), fcntl.LOCK_EX|fcntl.LOCK_NB)
- except IOError, x:
- if x.errno == errno.EAGAIN:
- raise StandardError("State file already locked")
- raise
-
- self.statefile = f
+ utils.LockFile(self.statefile.fileno())
try:
- self.data = simplejson.load(self.statefile)
+ self._data = serializer.Load(self.statefile.read())
except Exception, msg:
# Ignore errors while loading the file and treat it as empty
- self.data = {}
- sys.stderr.write("Empty or invalid state file."
- " Using defaults. Error message: %s\n" % msg)
+ self._data = {}
+ logging.warning(("Empty or invalid state file. Using defaults."
+ " Error message: %s"), msg)
+
+ if "instance" not in self._data:
+ self._data["instance"] = {}
+ if "node" not in self._data:
+ self._data["node"] = {}
- if "instance" not in self.data:
- self.data["instance"] = {}
- if "node" not in self.data:
- self.data["node"] = {}
+ self._orig_data = serializer.Dump(self._data)
- def __del__(self):
- """Called on destruction.
+ def Save(self):
+ """Save state to file, then unlock and close it.
"""
- if self.statefile:
- self._Close()
+ assert self.statefile
+
+ serialized_form = serializer.Dump(self._data)
+ if self._orig_data == serialized_form:
+ logging.debug("Data didn't change, just touching status file")
+ os.utime(constants.WATCHER_STATEFILE, None)
+ return
+
+ # We need to make sure the file is locked before renaming it, otherwise
+ # starting ganeti-watcher again at the same time will create a conflict.
+ fd = utils.WriteFile(constants.WATCHER_STATEFILE,
+ data=serialized_form,
+ prewrite=utils.LockFile, close=False)
+ self.statefile = os.fdopen(fd, 'w+')
- def _Close(self):
+ def Close(self):
"""Unlock configuration file and close it.
"""
assert self.statefile
- fcntl.flock(self.statefile.fileno(), fcntl.LOCK_UN)
-
+ # Files are automatically unlocked when closing them
self.statefile.close()
self.statefile = None
"""Returns the last boot ID of a node or None.
"""
- ndata = self.data["node"]
+ ndata = self._data["node"]
if name in ndata and KEY_BOOT_ID in ndata[name]:
return ndata[name][KEY_BOOT_ID]
"""
assert bootid
- ndata = self.data["node"]
+ ndata = self._data["node"]
if name not in ndata:
ndata[name] = {}
instance - the instance to look up.
"""
- idata = self.data["instance"]
+ idata = self._data["instance"]
if instance.name in idata:
return idata[instance.name][KEY_RESTART_COUNT]
instance - the instance being restarted
"""
- idata = self.data["instance"]
+ idata = self._data["instance"]
if instance.name not in idata:
inst = idata[instance.name] = {}
This method removes the record for a named instance.
"""
- idata = self.data["instance"]
+ idata = self._data["instance"]
if instance.name in idata:
del idata[instance.name]
- def Save(self):
- """Save state to file, then unlock and close it.
-
- """
- assert self.statefile
-
- self.statefile.seek(0)
- self.statefile.truncate()
-
- simplejson.dump(self.data, self.statefile)
-
- self._Close()
-
class Instance(object):
"""Abstraction for a Virtual Machine instance.
"""Encapsulates the start of an instance.
"""
- DoCmd(['gnt-instance', 'startup', '--lock-retries=15', self.name])
+ op = opcodes.OpStartupInstance(instance_name=self.name,
+ force=False,
+ extra_args=None)
+ cli.SubmitOpCode(op, cl=client)
def ActivateDisks(self):
"""Encapsulates the activation of all disks of an instance.
"""
- DoCmd(['gnt-instance', 'activate-disks', '--lock-retries=15', self.name])
-
-
-def _RunListCmd(cmd):
- """Runs a command and parses its output into lists.
-
- """
- for line in DoCmd(cmd).stdout.splitlines():
- yield line.split(':')
+ op = opcodes.OpActivateInstanceDisks(instance_name=self.name)
+ cli.SubmitOpCode(op, cl=client)
def GetInstanceList(with_secondaries=None):
"""Get a list of instances on this cluster.
"""
- cmd = ['gnt-instance', 'list', '--lock-retries=15', '--no-headers',
- '--separator=:']
-
- fields = 'name,oper_state,admin_state'
+ fields = ["name", "oper_state", "admin_state"]
if with_secondaries is not None:
- fields += ',snodes'
+ fields.append("snodes")
- cmd.append('-o')
- cmd.append(fields)
+ result = client.QueryInstances([], fields)
instances = []
- for fields in _RunListCmd(cmd):
+ for fields in result:
if with_secondaries is not None:
(name, status, autostart, snodes) = fields
- if snodes == "-":
+ if not snodes:
continue
for node in with_secondaries:
- if node in snodes.split(','):
+ if node in snodes:
break
else:
continue
else:
(name, status, autostart) = fields
- instances.append(Instance(name, status, autostart != "no"))
+ instances.append(Instance(name, status, autostart))
return instances
"""Get a dict mapping nodes to boot IDs.
"""
- cmd = ['gnt-node', 'list', '--lock-retries=15', '--no-headers',
- '--separator=:', '-o', 'name,bootid']
-
- ids = {}
- for fields in _RunListCmd(cmd):
- (name, bootid) = fields
- ids[name] = bootid
-
- return ids
-
-
-class Message(object):
- """Encapsulation of a notice or error message.
-
- """
- def __init__(self, level, msg):
- self.level = level
- self.msg = msg
- self.when = time.time()
-
- def __str__(self):
- return self.level + ' ' + time.ctime(self.when) + '\n' + Indent(self.msg)
+ result = client.QueryNodes([], ["name", "bootid"])
+ return dict([(name, bootid) for name, bootid in result])
class Watcher(object):
raise NotMasterError("This is not the master node")
self.instances = GetInstanceList()
self.bootids = GetNodeBootIDs()
- self.messages = []
self.started_instances = set()
def Run(self):
notepad = WatcherState()
- self.CheckInstances(notepad)
- self.CheckDisks(notepad)
- self.VerifyDisks()
- notepad.Save()
+ try:
+ self.CheckInstances(notepad)
+ self.CheckDisks(notepad)
+ self.VerifyDisks()
+ finally:
+ notepad.Save()
def CheckDisks(self, notepad):
"""Check all nodes for restarted ones.
"""
check_nodes = []
- for name, id in self.bootids.iteritems():
+ for name, new_id in self.bootids.iteritems():
old = notepad.GetNodeBootID(name)
- if old != id:
+ if old != new_id:
# Node's boot ID has changed, proably through a reboot.
check_nodes.append(name)
# secondary node.
for instance in GetInstanceList(with_secondaries=check_nodes):
if not instance.autostart:
- self.messages.append(Message(NOTICE,
- ("Skipping disk activation for"
- " non-autostart instance '%s'." %
- instance.name)))
+ 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:
- self.messages.append(Message(NOTICE, ("Activating disks for %s." %
- instance.name)))
+ logging.info("Activating disks for instance %s", instance.name)
instance.ActivateDisks()
- except Error, x:
- self.messages.append(Message(ERROR, str(x)))
+ except Exception, err:
+ logging.error(str(err), exc_info=True)
# Keep changed boot IDs
for name in check_nodes:
last = " (Attempt #%d)" % (n + 1)
else:
notepad.RecordRestartAttempt(instance)
- self.messages.append(Message(ERROR, "Could not restart %s for %d"
- " times, giving up..." %
- (instance.name, MAXTRIES)))
+ logging.error("Could not restart %s after %d attempts, giving up",
+ instance.name, MAXTRIES)
continue
try:
- self.messages.append(Message(NOTICE, ("Restarting %s%s." %
- (instance.name, last))))
+ logging.info("Restarting %s%s",
+ instance.name, last)
instance.Restart()
self.started_instances.add(instance.name)
- except Error, x:
- self.messages.append(Message(ERROR, str(x)))
+ except Exception, err:
+ logging.error(str(err), exc_info=True)
notepad.RecordRestartAttempt(instance)
elif instance.state in HELPLESS_STATES:
else:
if notepad.NumberOfRestartAttempts(instance):
notepad.RemoveInstance(instance)
- msg = Message(NOTICE, "Restart of %s succeeded." % instance.name)
- self.messages.append(msg)
+ logging.info("Restart of %s succeeded", instance.name)
def VerifyDisks(self):
"""Run gnt-cluster verify-disks.
"""
- result = DoCmd(['gnt-cluster', 'verify-disks', '--lock-retries=15'])
+ # TODO: What should we do here?
+ result = DoCmd(['gnt-cluster', 'verify-disks'])
if result.output:
- self.messages.append(Message(NOTICE, result.output))
-
- def WriteReport(self, logfile):
- """Log all messages to file.
-
- Args:
- logfile: file object open for writing (the log file)
-
- """
- for msg in self.messages:
- print >> logfile, str(msg)
+ logging.info(result.output)
def ParseOptions():
constants.RELEASE_VERSION)
parser.add_option("-d", "--debug", dest="debug",
- help="Don't redirect messages to the log file",
+ help="Write all messages to stderr",
default=False, action="store_true")
options, args = parser.parse_args()
return options, args
"""Main function.
"""
+ global client
+
options, args = ParseOptions()
- if not options.debug:
- sys.stderr = sys.stdout = open(constants.LOG_WATCHER, 'a')
+ logger.SetupLogging(constants.LOG_WATCHER, debug=options.debug)
try:
+ client = cli.GetClient()
+
try:
watcher = Watcher()
except errors.ConfigurationError:
# Just exit if there's no configuration
sys.exit(constants.EXIT_SUCCESS)
+
watcher.Run()
- watcher.WriteReport(sys.stdout)
+ except SystemExit:
+ raise
except NotMasterError:
- if options.debug:
- sys.stderr.write("Not master, exiting.\n")
+ logging.debug("Not master, exiting")
sys.exit(constants.EXIT_NOTMASTER)
except errors.ResolverError, err:
- sys.stderr.write("Cannot resolve hostname '%s', exiting.\n" % err.args[0])
+ logging.error("Cannot resolve hostname '%s', exiting.", err.args[0])
sys.exit(constants.EXIT_NODESETUP_ERROR)
- except Error, err:
- print err
+ except Exception, err:
+ logging.error(str(err), exc_info=True)
+ sys.exit(constants.EXIT_FAILURE)
if __name__ == '__main__':