root / lib / server / masterd.py @ 87ed6b79
History | View | Annotate | Download (23.9 kB)
1 |
#
|
---|---|
2 |
#
|
3 |
|
4 |
# Copyright (C) 2006, 2007, 2010, 2011, 2012 Google Inc.
|
5 |
#
|
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.
|
10 |
#
|
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.
|
15 |
#
|
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
|
19 |
# 02110-1301, USA.
|
20 |
|
21 |
|
22 |
"""Master daemon program.
|
23 |
|
24 |
Some classes deviates from the standard style guide since the
|
25 |
inheritance from parent classes requires it.
|
26 |
|
27 |
"""
|
28 |
|
29 |
# pylint: disable=C0103
|
30 |
# C0103: Invalid name ganeti-masterd
|
31 |
|
32 |
import grp |
33 |
import os |
34 |
import pwd |
35 |
import sys |
36 |
import socket |
37 |
import time |
38 |
import tempfile |
39 |
import logging |
40 |
|
41 |
from optparse import OptionParser |
42 |
|
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 |
import ganeti.rpc.errors as rpcerr |
52 |
from ganeti import utils |
53 |
from ganeti import errors |
54 |
from ganeti import ssconf |
55 |
from ganeti import workerpool |
56 |
import ganeti.rpc.node as rpc |
57 |
import ganeti.rpc.client as rpccl |
58 |
from ganeti import bootstrap |
59 |
from ganeti import netutils |
60 |
from ganeti import objects |
61 |
from ganeti import query |
62 |
from ganeti import runtime |
63 |
from ganeti import pathutils |
64 |
from ganeti import ht |
65 |
|
66 |
from ganeti.utils import version |
67 |
|
68 |
|
69 |
CLIENT_REQUEST_WORKERS = 16
|
70 |
|
71 |
EXIT_NOTMASTER = constants.EXIT_NOTMASTER |
72 |
EXIT_NODESETUP_ERROR = constants.EXIT_NODESETUP_ERROR |
73 |
|
74 |
|
75 |
def _LogNewJob(status, info, ops): |
76 |
"""Log information about a recently submitted job.
|
77 |
|
78 |
"""
|
79 |
op_summary = utils.CommaJoin(op.Summary() for op in ops) |
80 |
|
81 |
if status:
|
82 |
logging.info("New job with id %s, summary: %s", info, op_summary)
|
83 |
else:
|
84 |
logging.info("Failed to submit job, reason: '%s', summary: %s",
|
85 |
info, op_summary) |
86 |
|
87 |
|
88 |
class ClientRequestWorker(workerpool.BaseWorker): |
89 |
# pylint: disable=W0221
|
90 |
def RunTask(self, server, message, client): |
91 |
"""Process the request.
|
92 |
|
93 |
"""
|
94 |
client_ops = ClientOps(server) |
95 |
|
96 |
try:
|
97 |
(method, args, ver) = rpccl.ParseRequest(message) |
98 |
except rpcerr.ProtocolError, err:
|
99 |
logging.error("Protocol Error: %s", err)
|
100 |
client.close_log() |
101 |
return
|
102 |
|
103 |
success = False
|
104 |
try:
|
105 |
# Verify client's version if there was one in the request
|
106 |
if ver is not None and ver != constants.LUXI_VERSION: |
107 |
raise errors.LuxiError("LUXI version mismatch, server %s, request %s" % |
108 |
(constants.LUXI_VERSION, ver)) |
109 |
|
110 |
result = client_ops.handle_request(method, args) |
111 |
success = True
|
112 |
except errors.GenericError, err:
|
113 |
logging.exception("Unexpected exception")
|
114 |
success = False
|
115 |
result = errors.EncodeException(err) |
116 |
except:
|
117 |
logging.exception("Unexpected exception")
|
118 |
err = sys.exc_info() |
119 |
result = "Caught exception: %s" % str(err[1]) |
120 |
|
121 |
try:
|
122 |
reply = rpccl.FormatResponse(success, result) |
123 |
client.send_message(reply) |
124 |
# awake the main thread so that it can write out the data.
|
125 |
server.awaker.signal() |
126 |
except: # pylint: disable=W0702 |
127 |
logging.exception("Send error")
|
128 |
client.close_log() |
129 |
|
130 |
|
131 |
class MasterClientHandler(daemon.AsyncTerminatedMessageStream): |
132 |
"""Handler for master peers.
|
133 |
|
134 |
"""
|
135 |
_MAX_UNHANDLED = 1
|
136 |
|
137 |
def __init__(self, server, connected_socket, client_address, family): |
138 |
daemon.AsyncTerminatedMessageStream.__init__(self, connected_socket,
|
139 |
client_address, |
140 |
constants.LUXI_EOM, |
141 |
family, self._MAX_UNHANDLED)
|
142 |
self.server = server
|
143 |
|
144 |
def handle_message(self, message, _): |
145 |
self.server.request_workers.AddTask((self.server, message, self)) |
146 |
|
147 |
|
148 |
class _MasterShutdownCheck: |
149 |
"""Logic for master daemon shutdown.
|
150 |
|
151 |
"""
|
152 |
#: How long to wait between checks
|
153 |
_CHECK_INTERVAL = 5.0
|
154 |
|
155 |
#: How long to wait after all jobs are done (e.g. to give clients time to
|
156 |
#: retrieve the job status)
|
157 |
_SHUTDOWN_LINGER = 5.0
|
158 |
|
159 |
def __init__(self): |
160 |
"""Initializes this class.
|
161 |
|
162 |
"""
|
163 |
self._had_active_jobs = None |
164 |
self._linger_timeout = None |
165 |
|
166 |
def __call__(self, jq_prepare_result): |
167 |
"""Determines if master daemon is ready for shutdown.
|
168 |
|
169 |
@param jq_prepare_result: Result of L{jqueue.JobQueue.PrepareShutdown}
|
170 |
@rtype: None or number
|
171 |
@return: None if master daemon is ready, timeout if the check must be
|
172 |
repeated
|
173 |
|
174 |
"""
|
175 |
if jq_prepare_result:
|
176 |
# Check again shortly
|
177 |
logging.info("Job queue has been notified for shutdown but is still"
|
178 |
" busy; next check in %s seconds", self._CHECK_INTERVAL) |
179 |
self._had_active_jobs = True |
180 |
return self._CHECK_INTERVAL |
181 |
|
182 |
if not self._had_active_jobs: |
183 |
# Can shut down as there were no active jobs on the first check
|
184 |
return None |
185 |
|
186 |
# No jobs are running anymore, but maybe some clients want to collect some
|
187 |
# information. Give them a short amount of time.
|
188 |
if self._linger_timeout is None: |
189 |
self._linger_timeout = utils.RunningTimeout(self._SHUTDOWN_LINGER, True) |
190 |
|
191 |
remaining = self._linger_timeout.Remaining()
|
192 |
|
193 |
logging.info("Job queue no longer busy; shutting down master daemon"
|
194 |
" in %s seconds", remaining)
|
195 |
|
196 |
# TODO: Should the master daemon socket be closed at this point? Doing so
|
197 |
# wouldn't affect existing connections.
|
198 |
|
199 |
if remaining < 0: |
200 |
return None |
201 |
else:
|
202 |
return remaining
|
203 |
|
204 |
|
205 |
class MasterServer(daemon.AsyncStreamServer): |
206 |
"""Master Server.
|
207 |
|
208 |
This is the main asynchronous master server. It handles connections to the
|
209 |
master socket.
|
210 |
|
211 |
"""
|
212 |
family = socket.AF_UNIX |
213 |
|
214 |
def __init__(self, address, uid, gid): |
215 |
"""MasterServer constructor
|
216 |
|
217 |
@param address: the unix socket address to bind the MasterServer to
|
218 |
@param uid: The uid of the owner of the socket
|
219 |
@param gid: The gid of the owner of the socket
|
220 |
|
221 |
"""
|
222 |
temp_name = tempfile.mktemp(dir=os.path.dirname(address)) |
223 |
daemon.AsyncStreamServer.__init__(self, self.family, temp_name) |
224 |
os.chmod(temp_name, 0770)
|
225 |
os.chown(temp_name, uid, gid) |
226 |
os.rename(temp_name, address) |
227 |
|
228 |
self.awaker = daemon.AsyncAwaker()
|
229 |
|
230 |
# We'll only start threads once we've forked.
|
231 |
self.context = None |
232 |
self.request_workers = None |
233 |
|
234 |
self._shutdown_check = None |
235 |
|
236 |
def handle_connection(self, connected_socket, client_address): |
237 |
# TODO: add connection count and limit the number of open connections to a
|
238 |
# maximum number to avoid breaking for lack of file descriptors or memory.
|
239 |
MasterClientHandler(self, connected_socket, client_address, self.family) |
240 |
|
241 |
def setup_context(self): |
242 |
self.context = GanetiContext()
|
243 |
self.request_workers = workerpool.WorkerPool("ClientReq", |
244 |
CLIENT_REQUEST_WORKERS, |
245 |
ClientRequestWorker) |
246 |
|
247 |
def WaitForShutdown(self): |
248 |
"""Prepares server for shutdown.
|
249 |
|
250 |
"""
|
251 |
if self._shutdown_check is None: |
252 |
self._shutdown_check = _MasterShutdownCheck()
|
253 |
|
254 |
return self._shutdown_check(self.context.jobqueue.PrepareShutdown()) |
255 |
|
256 |
def server_cleanup(self): |
257 |
"""Cleanup the server.
|
258 |
|
259 |
This involves shutting down the processor threads and the master
|
260 |
socket.
|
261 |
|
262 |
"""
|
263 |
try:
|
264 |
self.close()
|
265 |
finally:
|
266 |
if self.request_workers: |
267 |
self.request_workers.TerminateWorkers()
|
268 |
if self.context: |
269 |
self.context.jobqueue.Shutdown()
|
270 |
self.context.livelock.close()
|
271 |
|
272 |
|
273 |
class ClientOps: |
274 |
"""Class holding high-level client operations."""
|
275 |
def __init__(self, server): |
276 |
self.server = server
|
277 |
|
278 |
def handle_request(self, method, args): # pylint: disable=R0911 |
279 |
context = self.server.context
|
280 |
queue = context.jobqueue |
281 |
|
282 |
# TODO: Parameter validation
|
283 |
if not isinstance(args, (tuple, list)): |
284 |
logging.info("Received invalid arguments of type '%s'", type(args)) |
285 |
raise ValueError("Invalid arguments type '%s'" % type(args)) |
286 |
|
287 |
if method not in luxi.REQ_ALL: |
288 |
logging.info("Received invalid request '%s'", method)
|
289 |
raise ValueError("Invalid operation '%s'" % method) |
290 |
|
291 |
# TODO: Rewrite to not exit in each 'if/elif' branch
|
292 |
|
293 |
if method == luxi.REQ_SUBMIT_JOB:
|
294 |
logging.info("Receiving new job")
|
295 |
(job_def, ) = args |
296 |
ops = [opcodes.OpCode.LoadOpCode(state) for state in job_def] |
297 |
job_id = queue.SubmitJob(ops) |
298 |
_LogNewJob(True, job_id, ops)
|
299 |
return job_id
|
300 |
|
301 |
elif method == luxi.REQ_PICKUP_JOB:
|
302 |
logging.info("Picking up new job from queue")
|
303 |
(job_id, ) = args |
304 |
queue.PickupJob(job_id) |
305 |
|
306 |
elif method == luxi.REQ_SUBMIT_JOB_TO_DRAINED_QUEUE:
|
307 |
logging.info("Forcefully receiving new job")
|
308 |
(job_def, ) = args |
309 |
ops = [opcodes.OpCode.LoadOpCode(state) for state in job_def] |
310 |
job_id = queue.SubmitJobToDrainedQueue(ops) |
311 |
_LogNewJob(True, job_id, ops)
|
312 |
return job_id
|
313 |
|
314 |
elif method == luxi.REQ_SUBMIT_MANY_JOBS:
|
315 |
logging.info("Receiving multiple jobs")
|
316 |
(job_defs, ) = args |
317 |
jobs = [] |
318 |
for ops in job_defs: |
319 |
jobs.append([opcodes.OpCode.LoadOpCode(state) for state in ops]) |
320 |
job_ids = queue.SubmitManyJobs(jobs) |
321 |
for ((status, job_id), ops) in zip(job_ids, jobs): |
322 |
_LogNewJob(status, job_id, ops) |
323 |
return job_ids
|
324 |
|
325 |
elif method == luxi.REQ_CANCEL_JOB:
|
326 |
(job_id, ) = args |
327 |
logging.info("Received job cancel request for %s", job_id)
|
328 |
return queue.CancelJob(job_id)
|
329 |
|
330 |
elif method == luxi.REQ_CHANGE_JOB_PRIORITY:
|
331 |
(job_id, priority) = args |
332 |
logging.info("Received request to change priority for job %s to %s",
|
333 |
job_id, priority) |
334 |
return queue.ChangeJobPriority(job_id, priority)
|
335 |
|
336 |
elif method == luxi.REQ_ARCHIVE_JOB:
|
337 |
(job_id, ) = args |
338 |
logging.info("Received job archive request for %s", job_id)
|
339 |
return queue.ArchiveJob(job_id)
|
340 |
|
341 |
elif method == luxi.REQ_AUTO_ARCHIVE_JOBS:
|
342 |
(age, timeout) = args |
343 |
logging.info("Received job autoarchive request for age %s, timeout %s",
|
344 |
age, timeout) |
345 |
return queue.AutoArchiveJobs(age, timeout)
|
346 |
|
347 |
elif method == luxi.REQ_WAIT_FOR_JOB_CHANGE:
|
348 |
(job_id, fields, prev_job_info, prev_log_serial, timeout) = args |
349 |
logging.info("Received job poll request for %s", job_id)
|
350 |
return queue.WaitForJobChanges(job_id, fields, prev_job_info,
|
351 |
prev_log_serial, timeout) |
352 |
|
353 |
elif method == luxi.REQ_QUERY:
|
354 |
(what, fields, qfilter) = args |
355 |
|
356 |
if what in constants.QR_VIA_OP: |
357 |
result = self._Query(opcodes.OpQuery(what=what, fields=fields,
|
358 |
qfilter=qfilter)) |
359 |
elif what == constants.QR_LOCK:
|
360 |
if qfilter is not None: |
361 |
raise errors.OpPrereqError("Lock queries can't be filtered", |
362 |
errors.ECODE_INVAL) |
363 |
return context.glm.QueryLocks(fields)
|
364 |
elif what == constants.QR_JOB:
|
365 |
return queue.QueryJobs(fields, qfilter)
|
366 |
elif what in constants.QR_VIA_LUXI: |
367 |
luxi_client = runtime.GetClient() |
368 |
result = luxi_client.Query(what, fields, qfilter).ToDict() |
369 |
else:
|
370 |
raise errors.OpPrereqError("Resource type '%s' unknown" % what, |
371 |
errors.ECODE_INVAL) |
372 |
|
373 |
return result
|
374 |
|
375 |
elif method == luxi.REQ_QUERY_FIELDS:
|
376 |
(what, fields) = args |
377 |
req = objects.QueryFieldsRequest(what=what, fields=fields) |
378 |
|
379 |
try:
|
380 |
fielddefs = query.ALL_FIELDS[req.what] |
381 |
except KeyError: |
382 |
raise errors.OpPrereqError("Resource type '%s' unknown" % req.what, |
383 |
errors.ECODE_INVAL) |
384 |
|
385 |
return query.QueryFields(fielddefs, req.fields)
|
386 |
|
387 |
elif method == luxi.REQ_QUERY_JOBS:
|
388 |
(job_ids, fields) = args |
389 |
if isinstance(job_ids, (tuple, list)) and job_ids: |
390 |
msg = utils.CommaJoin(job_ids) |
391 |
else:
|
392 |
msg = str(job_ids)
|
393 |
logging.info("Received job query request for %s", msg)
|
394 |
return queue.OldStyleQueryJobs(job_ids, fields)
|
395 |
|
396 |
elif method == luxi.REQ_QUERY_CONFIG_VALUES:
|
397 |
(fields, ) = args |
398 |
logging.info("Received config values query request for %s", fields)
|
399 |
op = opcodes.OpClusterConfigQuery(output_fields=fields) |
400 |
return self._Query(op) |
401 |
|
402 |
elif method == luxi.REQ_QUERY_CLUSTER_INFO:
|
403 |
logging.info("Received cluster info query request")
|
404 |
op = opcodes.OpClusterQuery() |
405 |
return self._Query(op) |
406 |
|
407 |
elif method == luxi.REQ_QUERY_TAGS:
|
408 |
(kind, name) = args |
409 |
logging.info("Received tags query request")
|
410 |
op = opcodes.OpTagsGet(kind=kind, name=name, use_locking=False)
|
411 |
return self._Query(op) |
412 |
|
413 |
elif method == luxi.REQ_SET_DRAIN_FLAG:
|
414 |
(drain_flag, ) = args |
415 |
logging.info("Received queue drain flag change request to %s",
|
416 |
drain_flag) |
417 |
return queue.SetDrainFlag(drain_flag)
|
418 |
|
419 |
elif method == luxi.REQ_SET_WATCHER_PAUSE:
|
420 |
(until, ) = args |
421 |
|
422 |
return _SetWatcherPause(context, until)
|
423 |
|
424 |
else:
|
425 |
logging.critical("Request '%s' in luxi.REQ_ALL, but not known", method)
|
426 |
raise errors.ProgrammerError("Operation '%s' in luxi.REQ_ALL," |
427 |
" but not implemented" % method)
|
428 |
|
429 |
def _Query(self, op): |
430 |
"""Runs the specified opcode and returns the result.
|
431 |
|
432 |
"""
|
433 |
# Queries don't have a job id
|
434 |
proc = mcpu.Processor(self.server.context, None, enable_locks=False) |
435 |
|
436 |
# TODO: Executing an opcode using locks will acquire them in blocking mode.
|
437 |
# Consider using a timeout for retries.
|
438 |
return proc.ExecOpCode(op, None) |
439 |
|
440 |
|
441 |
class GanetiContext(object): |
442 |
"""Context common to all ganeti threads.
|
443 |
|
444 |
This class creates and holds common objects shared by all threads.
|
445 |
|
446 |
"""
|
447 |
# pylint: disable=W0212
|
448 |
# we do want to ensure a singleton here
|
449 |
_instance = None
|
450 |
|
451 |
def __init__(self): |
452 |
"""Constructs a new GanetiContext object.
|
453 |
|
454 |
There should be only a GanetiContext object at any time, so this
|
455 |
function raises an error if this is not the case.
|
456 |
|
457 |
"""
|
458 |
assert self.__class__._instance is None, "double GanetiContext instance" |
459 |
|
460 |
# Create a livelock file
|
461 |
self.livelock = utils.livelock.LiveLock("masterd") |
462 |
|
463 |
# Create global configuration object
|
464 |
self.cfg = config.ConfigWriter()
|
465 |
|
466 |
# Locking manager
|
467 |
self.glm = locking.GanetiLockManager(
|
468 |
self.cfg.GetNodeList(),
|
469 |
self.cfg.GetNodeGroupList(),
|
470 |
[inst.name for inst in self.cfg.GetAllInstancesInfo().values()], |
471 |
self.cfg.GetNetworkList())
|
472 |
|
473 |
self.cfg.SetContext(self) |
474 |
|
475 |
# RPC runner
|
476 |
self.rpc = rpc.RpcRunner(self.cfg, self.glm.AddToLockMonitor) |
477 |
|
478 |
# Job queue
|
479 |
self.jobqueue = jqueue.JobQueue(self) |
480 |
|
481 |
# setting this also locks the class against attribute modifications
|
482 |
self.__class__._instance = self |
483 |
|
484 |
def __setattr__(self, name, value): |
485 |
"""Setting GanetiContext attributes is forbidden after initialization.
|
486 |
|
487 |
"""
|
488 |
assert self.__class__._instance is None, "Attempt to modify Ganeti Context" |
489 |
object.__setattr__(self, name, value) |
490 |
|
491 |
def AddNode(self, node, ec_id): |
492 |
"""Adds a node to the configuration.
|
493 |
|
494 |
"""
|
495 |
# Add it to the configuration
|
496 |
self.cfg.AddNode(node, ec_id)
|
497 |
|
498 |
# If preseeding fails it'll not be added
|
499 |
self.jobqueue.AddNode(node)
|
500 |
|
501 |
def ReaddNode(self, node): |
502 |
"""Updates a node that's already in the configuration
|
503 |
|
504 |
"""
|
505 |
# Synchronize the queue again
|
506 |
self.jobqueue.AddNode(node)
|
507 |
|
508 |
def RemoveNode(self, node): |
509 |
"""Removes a node from the configuration and lock manager.
|
510 |
|
511 |
"""
|
512 |
# Remove node from configuration
|
513 |
self.cfg.RemoveNode(node.uuid)
|
514 |
|
515 |
# Notify job queue
|
516 |
self.jobqueue.RemoveNode(node.name)
|
517 |
|
518 |
|
519 |
def _SetWatcherPause(context, until): |
520 |
"""Creates or removes the watcher pause file.
|
521 |
|
522 |
@type context: L{GanetiContext}
|
523 |
@param context: Global Ganeti context
|
524 |
@type until: None or int
|
525 |
@param until: Unix timestamp saying until when the watcher shouldn't run
|
526 |
|
527 |
"""
|
528 |
node_names = context.cfg.GetNodeList() |
529 |
|
530 |
if until is None: |
531 |
logging.info("Received request to no longer pause watcher")
|
532 |
else:
|
533 |
if not ht.TNumber(until): |
534 |
raise TypeError("Duration must be numeric") |
535 |
|
536 |
if until < time.time():
|
537 |
raise errors.GenericError("Unable to set pause end time in the past") |
538 |
|
539 |
logging.info("Received request to pause watcher until %s", until)
|
540 |
|
541 |
result = context.rpc.call_set_watcher_pause(node_names, until) |
542 |
|
543 |
errmsg = utils.CommaJoin("%s (%s)" % (node_name, nres.fail_msg)
|
544 |
for (node_name, nres) in result.items() |
545 |
if nres.fail_msg and not nres.offline) |
546 |
if errmsg:
|
547 |
raise errors.OpExecError("Watcher pause was set where possible, but failed" |
548 |
" on the following node(s): %s" % errmsg)
|
549 |
|
550 |
return until
|
551 |
|
552 |
|
553 |
@rpc.RunWithRPC
|
554 |
def CheckAgreement(): |
555 |
"""Check the agreement on who is the master.
|
556 |
|
557 |
The function uses a very simple algorithm: we must get more positive
|
558 |
than negative answers. Since in most of the cases we are the master,
|
559 |
we'll use our own config file for getting the node list. In the
|
560 |
future we could collect the current node list from our (possibly
|
561 |
obsolete) known nodes.
|
562 |
|
563 |
In order to account for cold-start of all nodes, we retry for up to
|
564 |
a minute until we get a real answer as the top-voted one. If the
|
565 |
nodes are more out-of-sync, for now manual startup of the master
|
566 |
should be attempted.
|
567 |
|
568 |
Note that for a even number of nodes cluster, we need at least half
|
569 |
of the nodes (beside ourselves) to vote for us. This creates a
|
570 |
problem on two-node clusters, since in this case we require the
|
571 |
other node to be up too to confirm our status.
|
572 |
|
573 |
"""
|
574 |
myself = netutils.Hostname.GetSysName() |
575 |
#temp instantiation of a config writer, used only to get the node list
|
576 |
cfg = config.ConfigWriter() |
577 |
node_names = cfg.GetNodeNames(cfg.GetNodeList()) |
578 |
del cfg
|
579 |
retries = 6
|
580 |
while retries > 0: |
581 |
votes = bootstrap.GatherMasterVotes(node_names) |
582 |
if not votes: |
583 |
# empty node list, this is a one node cluster
|
584 |
return True |
585 |
if votes[0][0] is None: |
586 |
retries -= 1
|
587 |
time.sleep(10)
|
588 |
continue
|
589 |
break
|
590 |
if retries == 0: |
591 |
logging.critical("Cluster inconsistent, most of the nodes didn't answer"
|
592 |
" after multiple retries. Aborting startup")
|
593 |
logging.critical("Use the --no-voting option if you understand what"
|
594 |
" effects it has on the cluster state")
|
595 |
return False |
596 |
# here a real node is at the top of the list
|
597 |
all_votes = sum(item[1] for item in votes) |
598 |
top_node, top_votes = votes[0]
|
599 |
|
600 |
result = False
|
601 |
if top_node != myself:
|
602 |
logging.critical("It seems we are not the master (top-voted node"
|
603 |
" is %s with %d out of %d votes)", top_node, top_votes,
|
604 |
all_votes) |
605 |
elif top_votes < all_votes - top_votes:
|
606 |
logging.critical("It seems we are not the master (%d votes for,"
|
607 |
" %d votes against)", top_votes, all_votes - top_votes)
|
608 |
else:
|
609 |
result = True
|
610 |
|
611 |
return result
|
612 |
|
613 |
|
614 |
@rpc.RunWithRPC
|
615 |
def ActivateMasterIP(): |
616 |
# activate ip
|
617 |
cfg = config.ConfigWriter() |
618 |
master_params = cfg.GetMasterNetworkParameters() |
619 |
ems = cfg.GetUseExternalMipScript() |
620 |
runner = rpc.BootstrapRunner() |
621 |
# we use the node name, as the configuration is only available here yet
|
622 |
result = runner.call_node_activate_master_ip( |
623 |
cfg.GetNodeName(master_params.uuid), master_params, ems) |
624 |
|
625 |
msg = result.fail_msg |
626 |
if msg:
|
627 |
logging.error("Can't activate master IP address: %s", msg)
|
628 |
|
629 |
|
630 |
def CheckMasterd(options, args): |
631 |
"""Initial checks whether to run or exit with a failure.
|
632 |
|
633 |
"""
|
634 |
if args: # masterd doesn't take any arguments |
635 |
print >> sys.stderr, ("Usage: %s [-f] [-d]" % sys.argv[0]) |
636 |
sys.exit(constants.EXIT_FAILURE) |
637 |
|
638 |
ssconf.CheckMaster(options.debug) |
639 |
|
640 |
try:
|
641 |
options.uid = pwd.getpwnam(constants.MASTERD_USER).pw_uid |
642 |
options.gid = grp.getgrnam(constants.DAEMONS_GROUP).gr_gid |
643 |
except KeyError: |
644 |
print >> sys.stderr, ("User or group not existing on system: %s:%s" % |
645 |
(constants.MASTERD_USER, constants.DAEMONS_GROUP)) |
646 |
sys.exit(constants.EXIT_FAILURE) |
647 |
|
648 |
# Determine static runtime architecture information
|
649 |
runtime.InitArchInfo() |
650 |
|
651 |
# Check the configuration is sane before anything else
|
652 |
try:
|
653 |
config.ConfigWriter() |
654 |
except errors.ConfigVersionMismatch, err:
|
655 |
v1 = "%s.%s.%s" % version.SplitVersion(err.args[0]) |
656 |
v2 = "%s.%s.%s" % version.SplitVersion(err.args[1]) |
657 |
print >> sys.stderr, \
|
658 |
("Configuration version mismatch. The current Ganeti software"
|
659 |
" expects version %s, but the on-disk configuration file has"
|
660 |
" version %s. This is likely the result of upgrading the"
|
661 |
" software without running the upgrade procedure. Please contact"
|
662 |
" your cluster administrator or complete the upgrade using the"
|
663 |
" cfgupgrade utility, after reading the upgrade notes." %
|
664 |
(v1, v2)) |
665 |
sys.exit(constants.EXIT_FAILURE) |
666 |
except errors.ConfigurationError, err:
|
667 |
print >> sys.stderr, \
|
668 |
("Configuration error while opening the configuration file: %s\n"
|
669 |
"This might be caused by an incomplete software upgrade or"
|
670 |
" by a corrupted configuration file. Until the problem is fixed"
|
671 |
" the master daemon cannot start." % str(err)) |
672 |
sys.exit(constants.EXIT_FAILURE) |
673 |
|
674 |
# If CheckMaster didn't fail we believe we are the master, but we have to
|
675 |
# confirm with the other nodes.
|
676 |
if options.no_voting:
|
677 |
if not options.yes_do_it: |
678 |
sys.stdout.write("The 'no voting' option has been selected.\n")
|
679 |
sys.stdout.write("This is dangerous, please confirm by"
|
680 |
" typing uppercase 'yes': ")
|
681 |
sys.stdout.flush() |
682 |
|
683 |
confirmation = sys.stdin.readline().strip() |
684 |
if confirmation != "YES": |
685 |
print >> sys.stderr, "Aborting." |
686 |
sys.exit(constants.EXIT_FAILURE) |
687 |
|
688 |
else:
|
689 |
# CheckAgreement uses RPC and threads, hence it needs to be run in
|
690 |
# a separate process before we call utils.Daemonize in the current
|
691 |
# process.
|
692 |
if not utils.RunInSeparateProcess(CheckAgreement): |
693 |
sys.exit(constants.EXIT_FAILURE) |
694 |
|
695 |
# ActivateMasterIP also uses RPC/threads, so we run it again via a
|
696 |
# separate process.
|
697 |
|
698 |
# TODO: decide whether failure to activate the master IP is a fatal error
|
699 |
utils.RunInSeparateProcess(ActivateMasterIP) |
700 |
|
701 |
|
702 |
def PrepMasterd(options, _): |
703 |
"""Prep master daemon function, executed with the PID file held.
|
704 |
|
705 |
"""
|
706 |
# This is safe to do as the pid file guarantees against
|
707 |
# concurrent execution.
|
708 |
utils.RemoveFile(pathutils.MASTER_SOCKET) |
709 |
|
710 |
mainloop = daemon.Mainloop() |
711 |
master = MasterServer(pathutils.MASTER_SOCKET, options.uid, options.gid) |
712 |
return (mainloop, master)
|
713 |
|
714 |
|
715 |
def ExecMasterd(options, args, prep_data): # pylint: disable=W0613 |
716 |
"""Main master daemon function, executed with the PID file held.
|
717 |
|
718 |
"""
|
719 |
(mainloop, master) = prep_data |
720 |
try:
|
721 |
rpc.Init() |
722 |
try:
|
723 |
master.setup_context() |
724 |
try:
|
725 |
mainloop.Run(shutdown_wait_fn=master.WaitForShutdown) |
726 |
finally:
|
727 |
master.server_cleanup() |
728 |
finally:
|
729 |
rpc.Shutdown() |
730 |
finally:
|
731 |
utils.RemoveFile(pathutils.MASTER_SOCKET) |
732 |
|
733 |
logging.info("Clean master daemon shutdown")
|
734 |
|
735 |
|
736 |
def Main(): |
737 |
"""Main function"""
|
738 |
parser = OptionParser(description="Ganeti master daemon",
|
739 |
usage="%prog [-f] [-d]",
|
740 |
version="%%prog (ganeti) %s" %
|
741 |
constants.RELEASE_VERSION) |
742 |
parser.add_option("--no-voting", dest="no_voting", |
743 |
help="Do not check that the nodes agree on this node"
|
744 |
" being the master and start the daemon unconditionally",
|
745 |
default=False, action="store_true") |
746 |
parser.add_option("--yes-do-it", dest="yes_do_it", |
747 |
help="Override interactive check for --no-voting",
|
748 |
default=False, action="store_true") |
749 |
daemon.GenericMain(constants.MASTERD, parser, CheckMasterd, PrepMasterd, |
750 |
ExecMasterd, multithreaded=True)
|