root / daemons / ganeti-masterd @ 26d3fd2f
History | View | Annotate | Download (18 kB)
1 |
#!/usr/bin/python |
---|---|
2 |
# |
3 |
|
4 |
# Copyright (C) 2006, 2007, 2010 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-msg=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 |
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 |
|
59 |
|
60 |
CLIENT_REQUEST_WORKERS = 16 |
61 |
|
62 |
EXIT_NOTMASTER = constants.EXIT_NOTMASTER |
63 |
EXIT_NODESETUP_ERROR = constants.EXIT_NODESETUP_ERROR |
64 |
|
65 |
|
66 |
class ClientRequestWorker(workerpool.BaseWorker): |
67 |
# pylint: disable-msg=W0221 |
68 |
def RunTask(self, server, message, client): |
69 |
"""Process the request. |
70 |
|
71 |
""" |
72 |
client_ops = ClientOps(server) |
73 |
|
74 |
try: |
75 |
(method, args) = luxi.ParseRequest(message) |
76 |
except luxi.ProtocolError, err: |
77 |
logging.error("Protocol Error: %s", err) |
78 |
client.close_log() |
79 |
return |
80 |
|
81 |
success = False |
82 |
try: |
83 |
result = client_ops.handle_request(method, args) |
84 |
success = True |
85 |
except errors.GenericError, err: |
86 |
logging.exception("Unexpected exception") |
87 |
success = False |
88 |
result = errors.EncodeException(err) |
89 |
except: |
90 |
logging.exception("Unexpected exception") |
91 |
err = sys.exc_info() |
92 |
result = "Caught exception: %s" % str(err[1]) |
93 |
|
94 |
try: |
95 |
reply = luxi.FormatResponse(success, result) |
96 |
client.send_message(reply) |
97 |
# awake the main thread so that it can write out the data. |
98 |
server.awaker.signal() |
99 |
except: # pylint: disable-msg=W0702 |
100 |
logging.exception("Send error") |
101 |
client.close_log() |
102 |
|
103 |
|
104 |
class MasterClientHandler(daemon.AsyncTerminatedMessageStream): |
105 |
"""Handler for master peers. |
106 |
|
107 |
""" |
108 |
_MAX_UNHANDLED = 1 |
109 |
def __init__(self, server, connected_socket, client_address, family): |
110 |
daemon.AsyncTerminatedMessageStream.__init__(self, connected_socket, |
111 |
client_address, |
112 |
constants.LUXI_EOM, |
113 |
family, self._MAX_UNHANDLED) |
114 |
self.server = server |
115 |
|
116 |
def handle_message(self, message, _): |
117 |
self.server.request_workers.AddTask((self.server, message, self)) |
118 |
|
119 |
|
120 |
class MasterServer(daemon.AsyncStreamServer): |
121 |
"""Master Server. |
122 |
|
123 |
This is the main asynchronous master server. It handles connections to the |
124 |
master socket. |
125 |
|
126 |
""" |
127 |
family = socket.AF_UNIX |
128 |
|
129 |
def __init__(self, mainloop, address, uid, gid): |
130 |
"""MasterServer constructor |
131 |
|
132 |
@type mainloop: ganeti.daemon.Mainloop |
133 |
@param mainloop: Mainloop used to poll for I/O events |
134 |
@param address: the unix socket address to bind the MasterServer to |
135 |
@param uid: The uid of the owner of the socket |
136 |
@param gid: The gid of the owner of the socket |
137 |
|
138 |
""" |
139 |
temp_name = tempfile.mktemp(dir=os.path.dirname(address)) |
140 |
daemon.AsyncStreamServer.__init__(self, self.family, temp_name) |
141 |
os.chmod(temp_name, 0770) |
142 |
os.chown(temp_name, uid, gid) |
143 |
os.rename(temp_name, address) |
144 |
|
145 |
self.mainloop = mainloop |
146 |
self.awaker = daemon.AsyncAwaker() |
147 |
|
148 |
# We'll only start threads once we've forked. |
149 |
self.context = None |
150 |
self.request_workers = None |
151 |
|
152 |
def handle_connection(self, connected_socket, client_address): |
153 |
# TODO: add connection count and limit the number of open connections to a |
154 |
# maximum number to avoid breaking for lack of file descriptors or memory. |
155 |
MasterClientHandler(self, connected_socket, client_address, self.family) |
156 |
|
157 |
def setup_queue(self): |
158 |
self.context = GanetiContext() |
159 |
self.request_workers = workerpool.WorkerPool("ClientReq", |
160 |
CLIENT_REQUEST_WORKERS, |
161 |
ClientRequestWorker) |
162 |
|
163 |
def server_cleanup(self): |
164 |
"""Cleanup the server. |
165 |
|
166 |
This involves shutting down the processor threads and the master |
167 |
socket. |
168 |
|
169 |
""" |
170 |
try: |
171 |
self.close() |
172 |
finally: |
173 |
if self.request_workers: |
174 |
self.request_workers.TerminateWorkers() |
175 |
if self.context: |
176 |
self.context.jobqueue.Shutdown() |
177 |
|
178 |
|
179 |
class ClientOps: |
180 |
"""Class holding high-level client operations.""" |
181 |
def __init__(self, server): |
182 |
self.server = server |
183 |
|
184 |
def handle_request(self, method, args): # pylint: disable-msg=R0911 |
185 |
queue = self.server.context.jobqueue |
186 |
|
187 |
# TODO: Parameter validation |
188 |
|
189 |
# TODO: Rewrite to not exit in each 'if/elif' branch |
190 |
|
191 |
if method == luxi.REQ_SUBMIT_JOB: |
192 |
logging.info("Received new job") |
193 |
ops = [opcodes.OpCode.LoadOpCode(state) for state in args] |
194 |
return queue.SubmitJob(ops) |
195 |
|
196 |
if method == luxi.REQ_SUBMIT_MANY_JOBS: |
197 |
logging.info("Received multiple jobs") |
198 |
jobs = [] |
199 |
for ops in args: |
200 |
jobs.append([opcodes.OpCode.LoadOpCode(state) for state in ops]) |
201 |
return queue.SubmitManyJobs(jobs) |
202 |
|
203 |
elif method == luxi.REQ_CANCEL_JOB: |
204 |
job_id = args |
205 |
logging.info("Received job cancel request for %s", job_id) |
206 |
return queue.CancelJob(job_id) |
207 |
|
208 |
elif method == luxi.REQ_ARCHIVE_JOB: |
209 |
job_id = args |
210 |
logging.info("Received job archive request for %s", job_id) |
211 |
return queue.ArchiveJob(job_id) |
212 |
|
213 |
elif method == luxi.REQ_AUTOARCHIVE_JOBS: |
214 |
(age, timeout) = args |
215 |
logging.info("Received job autoarchive request for age %s, timeout %s", |
216 |
age, timeout) |
217 |
return queue.AutoArchiveJobs(age, timeout) |
218 |
|
219 |
elif method == luxi.REQ_WAIT_FOR_JOB_CHANGE: |
220 |
(job_id, fields, prev_job_info, prev_log_serial, timeout) = args |
221 |
logging.info("Received job poll request for %s", job_id) |
222 |
return queue.WaitForJobChanges(job_id, fields, prev_job_info, |
223 |
prev_log_serial, timeout) |
224 |
|
225 |
elif method == luxi.REQ_QUERY_JOBS: |
226 |
(job_ids, fields) = args |
227 |
if isinstance(job_ids, (tuple, list)) and job_ids: |
228 |
msg = utils.CommaJoin(job_ids) |
229 |
else: |
230 |
msg = str(job_ids) |
231 |
logging.info("Received job query request for %s", msg) |
232 |
return queue.QueryJobs(job_ids, fields) |
233 |
|
234 |
elif method == luxi.REQ_QUERY_INSTANCES: |
235 |
(names, fields, use_locking) = args |
236 |
logging.info("Received instance query request for %s", names) |
237 |
if use_locking: |
238 |
raise errors.OpPrereqError("Sync queries are not allowed", |
239 |
errors.ECODE_INVAL) |
240 |
op = opcodes.OpQueryInstances(names=names, output_fields=fields, |
241 |
use_locking=use_locking) |
242 |
return self._Query(op) |
243 |
|
244 |
elif method == luxi.REQ_QUERY_NODES: |
245 |
(names, fields, use_locking) = args |
246 |
logging.info("Received node query request for %s", names) |
247 |
if use_locking: |
248 |
raise errors.OpPrereqError("Sync queries are not allowed", |
249 |
errors.ECODE_INVAL) |
250 |
op = opcodes.OpQueryNodes(names=names, output_fields=fields, |
251 |
use_locking=use_locking) |
252 |
return self._Query(op) |
253 |
|
254 |
elif method == luxi.REQ_QUERY_EXPORTS: |
255 |
nodes, use_locking = args |
256 |
if use_locking: |
257 |
raise errors.OpPrereqError("Sync queries are not allowed", |
258 |
errors.ECODE_INVAL) |
259 |
logging.info("Received exports query request") |
260 |
op = opcodes.OpQueryExports(nodes=nodes, use_locking=use_locking) |
261 |
return self._Query(op) |
262 |
|
263 |
elif method == luxi.REQ_QUERY_CONFIG_VALUES: |
264 |
fields = args |
265 |
logging.info("Received config values query request for %s", fields) |
266 |
op = opcodes.OpQueryConfigValues(output_fields=fields) |
267 |
return self._Query(op) |
268 |
|
269 |
elif method == luxi.REQ_QUERY_CLUSTER_INFO: |
270 |
logging.info("Received cluster info query request") |
271 |
op = opcodes.OpQueryClusterInfo() |
272 |
return self._Query(op) |
273 |
|
274 |
elif method == luxi.REQ_QUERY_TAGS: |
275 |
kind, name = args |
276 |
logging.info("Received tags query request") |
277 |
op = opcodes.OpGetTags(kind=kind, name=name) |
278 |
return self._Query(op) |
279 |
|
280 |
elif method == luxi.REQ_QUERY_LOCKS: |
281 |
(fields, sync) = args |
282 |
logging.info("Received locks query request") |
283 |
return self.server.context.glm.QueryLocks(fields, sync) |
284 |
|
285 |
elif method == luxi.REQ_QUEUE_SET_DRAIN_FLAG: |
286 |
drain_flag = args |
287 |
logging.info("Received queue drain flag change request to %s", |
288 |
drain_flag) |
289 |
return queue.SetDrainFlag(drain_flag) |
290 |
|
291 |
elif method == luxi.REQ_SET_WATCHER_PAUSE: |
292 |
(until, ) = args |
293 |
|
294 |
if until is None: |
295 |
logging.info("Received request to no longer pause the watcher") |
296 |
else: |
297 |
if not isinstance(until, (int, float)): |
298 |
raise TypeError("Duration must be an integer or float") |
299 |
|
300 |
if until < time.time(): |
301 |
raise errors.GenericError("Unable to set pause end time in the past") |
302 |
|
303 |
logging.info("Received request to pause the watcher until %s", until) |
304 |
|
305 |
return _SetWatcherPause(until) |
306 |
|
307 |
else: |
308 |
logging.info("Received invalid request '%s'", method) |
309 |
raise ValueError("Invalid operation '%s'" % method) |
310 |
|
311 |
def _Query(self, op): |
312 |
"""Runs the specified opcode and returns the result. |
313 |
|
314 |
""" |
315 |
# Queries don't have a job id |
316 |
proc = mcpu.Processor(self.server.context, None) |
317 |
|
318 |
# TODO: Executing an opcode using locks will acquire them in blocking mode. |
319 |
# Consider using a timeout for retries. |
320 |
return proc.ExecOpCode(op, None) |
321 |
|
322 |
|
323 |
class GanetiContext(object): |
324 |
"""Context common to all ganeti threads. |
325 |
|
326 |
This class creates and holds common objects shared by all threads. |
327 |
|
328 |
""" |
329 |
# pylint: disable-msg=W0212 |
330 |
# we do want to ensure a singleton here |
331 |
_instance = None |
332 |
|
333 |
def __init__(self): |
334 |
"""Constructs a new GanetiContext object. |
335 |
|
336 |
There should be only a GanetiContext object at any time, so this |
337 |
function raises an error if this is not the case. |
338 |
|
339 |
""" |
340 |
assert self.__class__._instance is None, "double GanetiContext instance" |
341 |
|
342 |
# Create global configuration object |
343 |
self.cfg = config.ConfigWriter() |
344 |
|
345 |
# Locking manager |
346 |
self.glm = locking.GanetiLockManager( |
347 |
self.cfg.GetNodeList(), |
348 |
self.cfg.GetInstanceList()) |
349 |
|
350 |
# Job queue |
351 |
self.jobqueue = jqueue.JobQueue(self) |
352 |
|
353 |
# setting this also locks the class against attribute modifications |
354 |
self.__class__._instance = self |
355 |
|
356 |
def __setattr__(self, name, value): |
357 |
"""Setting GanetiContext attributes is forbidden after initialization. |
358 |
|
359 |
""" |
360 |
assert self.__class__._instance is None, "Attempt to modify Ganeti Context" |
361 |
object.__setattr__(self, name, value) |
362 |
|
363 |
def AddNode(self, node, ec_id): |
364 |
"""Adds a node to the configuration and lock manager. |
365 |
|
366 |
""" |
367 |
# Add it to the configuration |
368 |
self.cfg.AddNode(node, ec_id) |
369 |
|
370 |
# If preseeding fails it'll not be added |
371 |
self.jobqueue.AddNode(node) |
372 |
|
373 |
# Add the new node to the Ganeti Lock Manager |
374 |
self.glm.add(locking.LEVEL_NODE, node.name) |
375 |
|
376 |
def ReaddNode(self, node): |
377 |
"""Updates a node that's already in the configuration |
378 |
|
379 |
""" |
380 |
# Synchronize the queue again |
381 |
self.jobqueue.AddNode(node) |
382 |
|
383 |
def RemoveNode(self, name): |
384 |
"""Removes a node from the configuration and lock manager. |
385 |
|
386 |
""" |
387 |
# Remove node from configuration |
388 |
self.cfg.RemoveNode(name) |
389 |
|
390 |
# Notify job queue |
391 |
self.jobqueue.RemoveNode(name) |
392 |
|
393 |
# Remove the node from the Ganeti Lock Manager |
394 |
self.glm.remove(locking.LEVEL_NODE, name) |
395 |
|
396 |
|
397 |
def _SetWatcherPause(until): |
398 |
"""Creates or removes the watcher pause file. |
399 |
|
400 |
@type until: None or int |
401 |
@param until: Unix timestamp saying until when the watcher shouldn't run |
402 |
|
403 |
""" |
404 |
if until is None: |
405 |
utils.RemoveFile(constants.WATCHER_PAUSEFILE) |
406 |
else: |
407 |
utils.WriteFile(constants.WATCHER_PAUSEFILE, |
408 |
data="%d\n" % (until, )) |
409 |
|
410 |
return until |
411 |
|
412 |
|
413 |
@rpc.RunWithRPC |
414 |
def CheckAgreement(): |
415 |
"""Check the agreement on who is the master. |
416 |
|
417 |
The function uses a very simple algorithm: we must get more positive |
418 |
than negative answers. Since in most of the cases we are the master, |
419 |
we'll use our own config file for getting the node list. In the |
420 |
future we could collect the current node list from our (possibly |
421 |
obsolete) known nodes. |
422 |
|
423 |
In order to account for cold-start of all nodes, we retry for up to |
424 |
a minute until we get a real answer as the top-voted one. If the |
425 |
nodes are more out-of-sync, for now manual startup of the master |
426 |
should be attempted. |
427 |
|
428 |
Note that for a even number of nodes cluster, we need at least half |
429 |
of the nodes (beside ourselves) to vote for us. This creates a |
430 |
problem on two-node clusters, since in this case we require the |
431 |
other node to be up too to confirm our status. |
432 |
|
433 |
""" |
434 |
myself = netutils.Hostname.GetSysName() |
435 |
#temp instantiation of a config writer, used only to get the node list |
436 |
cfg = config.ConfigWriter() |
437 |
node_list = cfg.GetNodeList() |
438 |
del cfg |
439 |
retries = 6 |
440 |
while retries > 0: |
441 |
votes = bootstrap.GatherMasterVotes(node_list) |
442 |
if not votes: |
443 |
# empty node list, this is a one node cluster |
444 |
return True |
445 |
if votes[0][0] is None: |
446 |
retries -= 1 |
447 |
time.sleep(10) |
448 |
continue |
449 |
break |
450 |
if retries == 0: |
451 |
logging.critical("Cluster inconsistent, most of the nodes didn't answer" |
452 |
" after multiple retries. Aborting startup") |
453 |
logging.critical("Use the --no-voting option if you understand what" |
454 |
" effects it has on the cluster state") |
455 |
return False |
456 |
# here a real node is at the top of the list |
457 |
all_votes = sum(item[1] for item in votes) |
458 |
top_node, top_votes = votes[0] |
459 |
|
460 |
result = False |
461 |
if top_node != myself: |
462 |
logging.critical("It seems we are not the master (top-voted node" |
463 |
" is %s with %d out of %d votes)", top_node, top_votes, |
464 |
all_votes) |
465 |
elif top_votes < all_votes - top_votes: |
466 |
logging.critical("It seems we are not the master (%d votes for," |
467 |
" %d votes against)", top_votes, all_votes - top_votes) |
468 |
else: |
469 |
result = True |
470 |
|
471 |
return result |
472 |
|
473 |
|
474 |
@rpc.RunWithRPC |
475 |
def ActivateMasterIP(): |
476 |
# activate ip |
477 |
master_node = ssconf.SimpleStore().GetMasterNode() |
478 |
result = rpc.RpcRunner.call_node_start_master(master_node, False, False) |
479 |
msg = result.fail_msg |
480 |
if msg: |
481 |
logging.error("Can't activate master IP address: %s", msg) |
482 |
|
483 |
|
484 |
def CheckMasterd(options, args): |
485 |
"""Initial checks whether to run or exit with a failure. |
486 |
|
487 |
""" |
488 |
if args: # masterd doesn't take any arguments |
489 |
print >> sys.stderr, ("Usage: %s [-f] [-d]" % sys.argv[0]) |
490 |
sys.exit(constants.EXIT_FAILURE) |
491 |
|
492 |
ssconf.CheckMaster(options.debug) |
493 |
|
494 |
try: |
495 |
options.uid = pwd.getpwnam(constants.MASTERD_USER).pw_uid |
496 |
options.gid = grp.getgrnam(constants.DAEMONS_GROUP).gr_gid |
497 |
except KeyError: |
498 |
print >> sys.stderr, ("User or group not existing on system: %s:%s" % |
499 |
(constants.MASTERD_USER, constants.DAEMONS_GROUP)) |
500 |
sys.exit(constants.EXIT_FAILURE) |
501 |
|
502 |
|
503 |
# If CheckMaster didn't fail we believe we are the master, but we have to |
504 |
# confirm with the other nodes. |
505 |
if options.no_voting: |
506 |
if options.yes_do_it: |
507 |
return |
508 |
|
509 |
sys.stdout.write("The 'no voting' option has been selected.\n") |
510 |
sys.stdout.write("This is dangerous, please confirm by" |
511 |
" typing uppercase 'yes': ") |
512 |
sys.stdout.flush() |
513 |
|
514 |
confirmation = sys.stdin.readline().strip() |
515 |
if confirmation != "YES": |
516 |
print >> sys.stderr, "Aborting." |
517 |
sys.exit(constants.EXIT_FAILURE) |
518 |
|
519 |
return |
520 |
|
521 |
# CheckAgreement uses RPC and threads, hence it needs to be run in a separate |
522 |
# process before we call utils.Daemonize in the current process. |
523 |
if not utils.RunInSeparateProcess(CheckAgreement): |
524 |
sys.exit(constants.EXIT_FAILURE) |
525 |
|
526 |
# ActivateMasterIP also uses RPC/threads, so we run it again via a |
527 |
# separate process. |
528 |
|
529 |
# TODO: decide whether failure to activate the master IP is a fatal error |
530 |
utils.RunInSeparateProcess(ActivateMasterIP) |
531 |
|
532 |
|
533 |
def ExecMasterd(options, args): # pylint: disable-msg=W0613 |
534 |
"""Main master daemon function, executed with the PID file held. |
535 |
|
536 |
""" |
537 |
# This is safe to do as the pid file guarantees against |
538 |
# concurrent execution. |
539 |
utils.RemoveFile(constants.MASTER_SOCKET) |
540 |
|
541 |
mainloop = daemon.Mainloop() |
542 |
master = MasterServer(mainloop, constants.MASTER_SOCKET, |
543 |
options.uid, options.gid) |
544 |
try: |
545 |
rpc.Init() |
546 |
try: |
547 |
master.setup_queue() |
548 |
try: |
549 |
mainloop.Run() |
550 |
finally: |
551 |
master.server_cleanup() |
552 |
finally: |
553 |
rpc.Shutdown() |
554 |
finally: |
555 |
utils.RemoveFile(constants.MASTER_SOCKET) |
556 |
|
557 |
|
558 |
def main(): |
559 |
"""Main function""" |
560 |
parser = OptionParser(description="Ganeti master daemon", |
561 |
usage="%prog [-f] [-d]", |
562 |
version="%%prog (ganeti) %s" % |
563 |
constants.RELEASE_VERSION) |
564 |
parser.add_option("--no-voting", dest="no_voting", |
565 |
help="Do not check that the nodes agree on this node" |
566 |
" being the master and start the daemon unconditionally", |
567 |
default=False, action="store_true") |
568 |
parser.add_option("--yes-do-it", dest="yes_do_it", |
569 |
help="Override interactive check for --no-voting", |
570 |
default=False, action="store_true") |
571 |
daemon.GenericMain(constants.MASTERD, parser, CheckMasterd, ExecMasterd, |
572 |
multithreaded=True) |
573 |
|
574 |
|
575 |
if __name__ == "__main__": |
576 |
main() |