4 # Copyright (C) 2006, 2007, 2010, 2011, 2012 Google Inc.
6 # This program is free software; you can redistribute it and/or modify
7 # it under the terms of the GNU General Public License as published by
8 # the Free Software Foundation; either version 2 of the License, or
9 # (at your option) any later version.
11 # This program is distributed in the hope that it will be useful, but
12 # WITHOUT ANY WARRANTY; without even the implied warranty of
13 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 # General Public License for more details.
16 # You should have received a copy of the GNU General Public License
17 # along with this program; if not, write to the Free Software
18 # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
22 """Master daemon program.
24 Some classes deviates from the standard style guide since the
25 inheritance from parent classes requires it.
29 # pylint: disable=C0103
30 # C0103: Invalid name ganeti-masterd
41 from optparse import OptionParser
43 from ganeti import config
44 from ganeti import constants
45 from ganeti import daemon
46 from ganeti import mcpu
47 from ganeti import opcodes
48 from ganeti import jqueue
49 from ganeti import locking
50 from ganeti import luxi
51 from ganeti import utils
52 from ganeti import errors
53 from ganeti import ssconf
54 from ganeti import workerpool
55 from ganeti import rpc
56 from ganeti import bootstrap
57 from ganeti import netutils
58 from ganeti import objects
59 from ganeti import query
60 from ganeti import runtime
61 from ganeti import pathutils
65 CLIENT_REQUEST_WORKERS = 16
67 EXIT_NOTMASTER = constants.EXIT_NOTMASTER
68 EXIT_NODESETUP_ERROR = constants.EXIT_NODESETUP_ERROR
71 def _LogNewJob(status, info, ops):
72 """Log information about a recently submitted job.
75 op_summary = utils.CommaJoin(op.Summary() for op in ops)
78 logging.info("New job with id %s, summary: %s", info, op_summary)
80 logging.info("Failed to submit job, reason: '%s', summary: %s",
84 class ClientRequestWorker(workerpool.BaseWorker):
85 # pylint: disable=W0221
86 def RunTask(self, server, message, client):
87 """Process the request.
90 client_ops = ClientOps(server)
93 (method, args, version) = luxi.ParseRequest(message)
94 except luxi.ProtocolError, err:
95 logging.error("Protocol Error: %s", err)
101 # Verify client's version if there was one in the request
102 if version is not None and version != constants.LUXI_VERSION:
103 raise errors.LuxiError("LUXI version mismatch, server %s, request %s" %
104 (constants.LUXI_VERSION, version))
106 result = client_ops.handle_request(method, args)
108 except errors.GenericError, err:
109 logging.exception("Unexpected exception")
111 result = errors.EncodeException(err)
113 logging.exception("Unexpected exception")
115 result = "Caught exception: %s" % str(err[1])
118 reply = luxi.FormatResponse(success, result)
119 client.send_message(reply)
120 # awake the main thread so that it can write out the data.
121 server.awaker.signal()
122 except: # pylint: disable=W0702
123 logging.exception("Send error")
127 class MasterClientHandler(daemon.AsyncTerminatedMessageStream):
128 """Handler for master peers.
133 def __init__(self, server, connected_socket, client_address, family):
134 daemon.AsyncTerminatedMessageStream.__init__(self, connected_socket,
137 family, self._MAX_UNHANDLED)
140 def handle_message(self, message, _):
141 self.server.request_workers.AddTask((self.server, message, self))
144 class _MasterShutdownCheck:
145 """Logic for master daemon shutdown.
148 #: How long to wait between checks
149 _CHECK_INTERVAL = 5.0
151 #: How long to wait after all jobs are done (e.g. to give clients time to
152 #: retrieve the job status)
153 _SHUTDOWN_LINGER = 5.0
156 """Initializes this class.
159 self._had_active_jobs = None
160 self._linger_timeout = None
162 def __call__(self, jq_prepare_result):
163 """Determines if master daemon is ready for shutdown.
165 @param jq_prepare_result: Result of L{jqueue.JobQueue.PrepareShutdown}
166 @rtype: None or number
167 @return: None if master daemon is ready, timeout if the check must be
171 if jq_prepare_result:
172 # Check again shortly
173 logging.info("Job queue has been notified for shutdown but is still"
174 " busy; next check in %s seconds", self._CHECK_INTERVAL)
175 self._had_active_jobs = True
176 return self._CHECK_INTERVAL
178 if not self._had_active_jobs:
179 # Can shut down as there were no active jobs on the first check
182 # No jobs are running anymore, but maybe some clients want to collect some
183 # information. Give them a short amount of time.
184 if self._linger_timeout is None:
185 self._linger_timeout = utils.RunningTimeout(self._SHUTDOWN_LINGER, True)
187 remaining = self._linger_timeout.Remaining()
189 logging.info("Job queue no longer busy; shutting down master daemon"
190 " in %s seconds", remaining)
192 # TODO: Should the master daemon socket be closed at this point? Doing so
193 # wouldn't affect existing connections.
201 class MasterServer(daemon.AsyncStreamServer):
204 This is the main asynchronous master server. It handles connections to the
208 family = socket.AF_UNIX
210 def __init__(self, address, uid, gid):
211 """MasterServer constructor
213 @param address: the unix socket address to bind the MasterServer to
214 @param uid: The uid of the owner of the socket
215 @param gid: The gid of the owner of the socket
218 temp_name = tempfile.mktemp(dir=os.path.dirname(address))
219 daemon.AsyncStreamServer.__init__(self, self.family, temp_name)
220 os.chmod(temp_name, 0770)
221 os.chown(temp_name, uid, gid)
222 os.rename(temp_name, address)
224 self.awaker = daemon.AsyncAwaker()
226 # We'll only start threads once we've forked.
228 self.request_workers = None
230 self._shutdown_check = None
232 def handle_connection(self, connected_socket, client_address):
233 # TODO: add connection count and limit the number of open connections to a
234 # maximum number to avoid breaking for lack of file descriptors or memory.
235 MasterClientHandler(self, connected_socket, client_address, self.family)
237 def setup_queue(self):
238 self.context = GanetiContext()
239 self.request_workers = workerpool.WorkerPool("ClientReq",
240 CLIENT_REQUEST_WORKERS,
243 def WaitForShutdown(self):
244 """Prepares server for shutdown.
247 if self._shutdown_check is None:
248 self._shutdown_check = _MasterShutdownCheck()
250 return self._shutdown_check(self.context.jobqueue.PrepareShutdown())
252 def server_cleanup(self):
253 """Cleanup the server.
255 This involves shutting down the processor threads and the master
262 if self.request_workers:
263 self.request_workers.TerminateWorkers()
265 self.context.jobqueue.Shutdown()
269 """Class holding high-level client operations."""
270 def __init__(self, server):
273 def handle_request(self, method, args): # pylint: disable=R0911
274 context = self.server.context
275 queue = context.jobqueue
277 # TODO: Parameter validation
278 if not isinstance(args, (tuple, list)):
279 logging.info("Received invalid arguments of type '%s'", type(args))
280 raise ValueError("Invalid arguments type '%s'" % type(args))
282 # TODO: Rewrite to not exit in each 'if/elif' branch
284 if method == luxi.REQ_SUBMIT_JOB:
285 logging.info("Receiving new job")
287 ops = [opcodes.OpCode.LoadOpCode(state) for state in job_def]
288 job_id = queue.SubmitJob(ops)
289 _LogNewJob(True, job_id, ops)
292 elif method == luxi.REQ_SUBMIT_MANY_JOBS:
293 logging.info("Receiving multiple jobs")
297 jobs.append([opcodes.OpCode.LoadOpCode(state) for state in ops])
298 job_ids = queue.SubmitManyJobs(jobs)
299 for ((status, job_id), ops) in zip(job_ids, jobs):
300 _LogNewJob(status, job_id, ops)
303 elif method == luxi.REQ_CANCEL_JOB:
305 logging.info("Received job cancel request for %s", job_id)
306 return queue.CancelJob(job_id)
308 elif method == luxi.REQ_CHANGE_JOB_PRIORITY:
309 (job_id, priority) = args
310 logging.info("Received request to change priority for job %s to %s",
312 return queue.ChangeJobPriority(job_id, priority)
314 elif method == luxi.REQ_ARCHIVE_JOB:
316 logging.info("Received job archive request for %s", job_id)
317 return queue.ArchiveJob(job_id)
319 elif method == luxi.REQ_AUTO_ARCHIVE_JOBS:
320 (age, timeout) = args
321 logging.info("Received job autoarchive request for age %s, timeout %s",
323 return queue.AutoArchiveJobs(age, timeout)
325 elif method == luxi.REQ_WAIT_FOR_JOB_CHANGE:
326 (job_id, fields, prev_job_info, prev_log_serial, timeout) = args
327 logging.info("Received job poll request for %s", job_id)
328 return queue.WaitForJobChanges(job_id, fields, prev_job_info,
329 prev_log_serial, timeout)
331 elif method == luxi.REQ_QUERY:
332 (what, fields, qfilter) = args
334 if what in constants.QR_VIA_OP:
335 result = self._Query(opcodes.OpQuery(what=what, fields=fields,
337 elif what == constants.QR_LOCK:
338 if qfilter is not None:
339 raise errors.OpPrereqError("Lock queries can't be filtered",
341 return context.glm.QueryLocks(fields)
342 elif what == constants.QR_JOB:
343 return queue.QueryJobs(fields, qfilter)
344 elif what in constants.QR_VIA_LUXI:
345 raise NotImplementedError
347 raise errors.OpPrereqError("Resource type '%s' unknown" % what,
352 elif method == luxi.REQ_QUERY_FIELDS:
353 (what, fields) = args
354 req = objects.QueryFieldsRequest(what=what, fields=fields)
357 fielddefs = query.ALL_FIELDS[req.what]
359 raise errors.OpPrereqError("Resource type '%s' unknown" % req.what,
362 return query.QueryFields(fielddefs, req.fields)
364 elif method == luxi.REQ_QUERY_JOBS:
365 (job_ids, fields) = args
366 if isinstance(job_ids, (tuple, list)) and job_ids:
367 msg = utils.CommaJoin(job_ids)
370 logging.info("Received job query request for %s", msg)
371 return queue.OldStyleQueryJobs(job_ids, fields)
373 elif method == luxi.REQ_QUERY_INSTANCES:
374 (names, fields, use_locking) = args
375 logging.info("Received instance query request for %s", names)
377 raise errors.OpPrereqError("Sync queries are not allowed",
379 op = opcodes.OpInstanceQuery(names=names, output_fields=fields,
380 use_locking=use_locking)
381 return self._Query(op)
383 elif method == luxi.REQ_QUERY_NODES:
384 (names, fields, use_locking) = args
385 logging.info("Received node query request for %s", names)
387 raise errors.OpPrereqError("Sync queries are not allowed",
389 op = opcodes.OpNodeQuery(names=names, output_fields=fields,
390 use_locking=use_locking)
391 return self._Query(op)
393 elif method == luxi.REQ_QUERY_GROUPS:
394 (names, fields, use_locking) = args
395 logging.info("Received group query request for %s", names)
397 raise errors.OpPrereqError("Sync queries are not allowed",
399 op = opcodes.OpGroupQuery(names=names, output_fields=fields)
400 return self._Query(op)
402 elif method == luxi.REQ_QUERY_NETWORKS:
403 (names, fields, use_locking) = args
404 logging.info("Received network query request for %s", names)
406 raise errors.OpPrereqError("Sync queries are not allowed",
408 op = opcodes.OpNetworkQuery(names=names, output_fields=fields)
409 return self._Query(op)
411 elif method == luxi.REQ_QUERY_EXPORTS:
412 (nodes, use_locking) = args
414 raise errors.OpPrereqError("Sync queries are not allowed",
416 logging.info("Received exports query request")
417 op = opcodes.OpBackupQuery(nodes=nodes, use_locking=use_locking)
418 return self._Query(op)
420 elif method == luxi.REQ_QUERY_CONFIG_VALUES:
422 logging.info("Received config values query request for %s", fields)
423 op = opcodes.OpClusterConfigQuery(output_fields=fields)
424 return self._Query(op)
426 elif method == luxi.REQ_QUERY_CLUSTER_INFO:
427 logging.info("Received cluster info query request")
428 op = opcodes.OpClusterQuery()
429 return self._Query(op)
431 elif method == luxi.REQ_QUERY_TAGS:
433 logging.info("Received tags query request")
434 op = opcodes.OpTagsGet(kind=kind, name=name, use_locking=False)
435 return self._Query(op)
437 elif method == luxi.REQ_SET_DRAIN_FLAG:
438 (drain_flag, ) = args
439 logging.info("Received queue drain flag change request to %s",
441 return queue.SetDrainFlag(drain_flag)
443 elif method == luxi.REQ_SET_WATCHER_PAUSE:
446 return _SetWatcherPause(context, until)
449 logging.info("Received invalid request '%s'", method)
450 raise ValueError("Invalid operation '%s'" % method)
452 def _Query(self, op):
453 """Runs the specified opcode and returns the result.
456 # Queries don't have a job id
457 proc = mcpu.Processor(self.server.context, None, enable_locks=False)
459 # TODO: Executing an opcode using locks will acquire them in blocking mode.
460 # Consider using a timeout for retries.
461 return proc.ExecOpCode(op, None)
464 class GanetiContext(object):
465 """Context common to all ganeti threads.
467 This class creates and holds common objects shared by all threads.
470 # pylint: disable=W0212
471 # we do want to ensure a singleton here
475 """Constructs a new GanetiContext object.
477 There should be only a GanetiContext object at any time, so this
478 function raises an error if this is not the case.
481 assert self.__class__._instance is None, "double GanetiContext instance"
483 # Create global configuration object
484 self.cfg = config.ConfigWriter()
487 self.glm = locking.GanetiLockManager(
488 self.cfg.GetNodeList(),
489 self.cfg.GetNodeGroupList(),
490 self.cfg.GetInstanceList(),
491 self.cfg.GetNetworkList())
493 self.cfg.SetContext(self)
496 self.rpc = rpc.RpcRunner(self.cfg, self.glm.AddToLockMonitor)
499 self.jobqueue = jqueue.JobQueue(self)
501 # setting this also locks the class against attribute modifications
502 self.__class__._instance = self
504 def __setattr__(self, name, value):
505 """Setting GanetiContext attributes is forbidden after initialization.
508 assert self.__class__._instance is None, "Attempt to modify Ganeti Context"
509 object.__setattr__(self, name, value)
511 def AddNode(self, node, ec_id):
512 """Adds a node to the configuration and lock manager.
515 # Add it to the configuration
516 self.cfg.AddNode(node, ec_id)
518 # If preseeding fails it'll not be added
519 self.jobqueue.AddNode(node)
521 # Add the new node to the Ganeti Lock Manager
522 self.glm.add(locking.LEVEL_NODE, node.name)
523 self.glm.add(locking.LEVEL_NODE_RES, node.name)
525 def ReaddNode(self, node):
526 """Updates a node that's already in the configuration
529 # Synchronize the queue again
530 self.jobqueue.AddNode(node)
532 def RemoveNode(self, name):
533 """Removes a node from the configuration and lock manager.
536 # Remove node from configuration
537 self.cfg.RemoveNode(name)
540 self.jobqueue.RemoveNode(name)
542 # Remove the node from the Ganeti Lock Manager
543 self.glm.remove(locking.LEVEL_NODE, name)
544 self.glm.remove(locking.LEVEL_NODE_RES, name)
547 def _SetWatcherPause(context, until):
548 """Creates or removes the watcher pause file.
550 @type context: L{GanetiContext}
551 @param context: Global Ganeti context
552 @type until: None or int
553 @param until: Unix timestamp saying until when the watcher shouldn't run
556 node_names = context.cfg.GetNodeList()
559 logging.info("Received request to no longer pause watcher")
561 if not ht.TNumber(until):
562 raise TypeError("Duration must be numeric")
564 if until < time.time():
565 raise errors.GenericError("Unable to set pause end time in the past")
567 logging.info("Received request to pause watcher until %s", until)
569 result = context.rpc.call_set_watcher_pause(node_names, until)
571 errmsg = utils.CommaJoin("%s (%s)" % (node_name, nres.fail_msg)
572 for (node_name, nres) in result.items()
573 if nres.fail_msg and not nres.offline)
575 raise errors.OpExecError("Watcher pause was set where possible, but failed"
576 " on the following node(s): %s" % errmsg)
582 def CheckAgreement():
583 """Check the agreement on who is the master.
585 The function uses a very simple algorithm: we must get more positive
586 than negative answers. Since in most of the cases we are the master,
587 we'll use our own config file for getting the node list. In the
588 future we could collect the current node list from our (possibly
589 obsolete) known nodes.
591 In order to account for cold-start of all nodes, we retry for up to
592 a minute until we get a real answer as the top-voted one. If the
593 nodes are more out-of-sync, for now manual startup of the master
596 Note that for a even number of nodes cluster, we need at least half
597 of the nodes (beside ourselves) to vote for us. This creates a
598 problem on two-node clusters, since in this case we require the
599 other node to be up too to confirm our status.
602 myself = netutils.Hostname.GetSysName()
603 #temp instantiation of a config writer, used only to get the node list
604 cfg = config.ConfigWriter()
605 node_list = cfg.GetNodeList()
609 votes = bootstrap.GatherMasterVotes(node_list)
611 # empty node list, this is a one node cluster
613 if votes[0][0] is None:
619 logging.critical("Cluster inconsistent, most of the nodes didn't answer"
620 " after multiple retries. Aborting startup")
621 logging.critical("Use the --no-voting option if you understand what"
622 " effects it has on the cluster state")
624 # here a real node is at the top of the list
625 all_votes = sum(item[1] for item in votes)
626 top_node, top_votes = votes[0]
629 if top_node != myself:
630 logging.critical("It seems we are not the master (top-voted node"
631 " is %s with %d out of %d votes)", top_node, top_votes,
633 elif top_votes < all_votes - top_votes:
634 logging.critical("It seems we are not the master (%d votes for,"
635 " %d votes against)", top_votes, all_votes - top_votes)
643 def ActivateMasterIP():
645 cfg = config.ConfigWriter()
646 master_params = cfg.GetMasterNetworkParameters()
647 ems = cfg.GetUseExternalMipScript()
648 runner = rpc.BootstrapRunner()
649 result = runner.call_node_activate_master_ip(master_params.name,
652 msg = result.fail_msg
654 logging.error("Can't activate master IP address: %s", msg)
657 def CheckMasterd(options, args):
658 """Initial checks whether to run or exit with a failure.
661 if args: # masterd doesn't take any arguments
662 print >> sys.stderr, ("Usage: %s [-f] [-d]" % sys.argv[0])
663 sys.exit(constants.EXIT_FAILURE)
665 ssconf.CheckMaster(options.debug)
668 options.uid = pwd.getpwnam(constants.MASTERD_USER).pw_uid
669 options.gid = grp.getgrnam(constants.DAEMONS_GROUP).gr_gid
671 print >> sys.stderr, ("User or group not existing on system: %s:%s" %
672 (constants.MASTERD_USER, constants.DAEMONS_GROUP))
673 sys.exit(constants.EXIT_FAILURE)
675 # Determine static runtime architecture information
676 runtime.InitArchInfo()
678 # Check the configuration is sane before anything else
680 config.ConfigWriter()
681 except errors.ConfigVersionMismatch, err:
682 v1 = "%s.%s.%s" % constants.SplitVersion(err.args[0])
683 v2 = "%s.%s.%s" % constants.SplitVersion(err.args[1])
684 print >> sys.stderr, \
685 ("Configuration version mismatch. The current Ganeti software"
686 " expects version %s, but the on-disk configuration file has"
687 " version %s. This is likely the result of upgrading the"
688 " software without running the upgrade procedure. Please contact"
689 " your cluster administrator or complete the upgrade using the"
690 " cfgupgrade utility, after reading the upgrade notes." %
692 sys.exit(constants.EXIT_FAILURE)
693 except errors.ConfigurationError, err:
694 print >> sys.stderr, \
695 ("Configuration error while opening the configuration file: %s\n"
696 "This might be caused by an incomplete software upgrade or"
697 " by a corrupted configuration file. Until the problem is fixed"
698 " the master daemon cannot start." % str(err))
699 sys.exit(constants.EXIT_FAILURE)
701 # If CheckMaster didn't fail we believe we are the master, but we have to
702 # confirm with the other nodes.
703 if options.no_voting:
704 if not options.yes_do_it:
705 sys.stdout.write("The 'no voting' option has been selected.\n")
706 sys.stdout.write("This is dangerous, please confirm by"
707 " typing uppercase 'yes': ")
710 confirmation = sys.stdin.readline().strip()
711 if confirmation != "YES":
712 print >> sys.stderr, "Aborting."
713 sys.exit(constants.EXIT_FAILURE)
716 # CheckAgreement uses RPC and threads, hence it needs to be run in
717 # a separate process before we call utils.Daemonize in the current
719 if not utils.RunInSeparateProcess(CheckAgreement):
720 sys.exit(constants.EXIT_FAILURE)
722 # ActivateMasterIP also uses RPC/threads, so we run it again via a
725 # TODO: decide whether failure to activate the master IP is a fatal error
726 utils.RunInSeparateProcess(ActivateMasterIP)
729 def PrepMasterd(options, _):
730 """Prep master daemon function, executed with the PID file held.
733 # This is safe to do as the pid file guarantees against
734 # concurrent execution.
735 utils.RemoveFile(pathutils.MASTER_SOCKET)
737 mainloop = daemon.Mainloop()
738 master = MasterServer(pathutils.MASTER_SOCKET, options.uid, options.gid)
739 return (mainloop, master)
742 def ExecMasterd(options, args, prep_data): # pylint: disable=W0613
743 """Main master daemon function, executed with the PID file held.
746 (mainloop, master) = prep_data
752 mainloop.Run(shutdown_wait_fn=master.WaitForShutdown)
754 master.server_cleanup()
758 utils.RemoveFile(pathutils.MASTER_SOCKET)
760 logging.info("Clean master daemon shutdown")
765 parser = OptionParser(description="Ganeti master daemon",
766 usage="%prog [-f] [-d]",
767 version="%%prog (ganeti) %s" %
768 constants.RELEASE_VERSION)
769 parser.add_option("--no-voting", dest="no_voting",
770 help="Do not check that the nodes agree on this node"
771 " being the master and start the daemon unconditionally",
772 default=False, action="store_true")
773 parser.add_option("--yes-do-it", dest="yes_do_it",
774 help="Override interactive check for --no-voting",
775 default=False, action="store_true")
776 daemon.GenericMain(constants.MASTERD, parser, CheckMasterd, PrepMasterd,
777 ExecMasterd, multithreaded=True)