root / daemons / ganeti-watcher @ 78f3bd30
History | View | Annotate | Download (12.5 kB)
1 |
#!/usr/bin/python |
---|---|
2 |
# |
3 |
|
4 |
# Copyright (C) 2006, 2007 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 |
"""Tool to restart erronously downed virtual machines. |
23 |
|
24 |
This program and set of classes implement a watchdog to restart |
25 |
virtual machines in a Ganeti cluster that have crashed or been killed |
26 |
by a node reboot. Run from cron or similar. |
27 |
|
28 |
""" |
29 |
|
30 |
import os |
31 |
import sys |
32 |
import re |
33 |
import time |
34 |
import fcntl |
35 |
import errno |
36 |
import logging |
37 |
from optparse import OptionParser |
38 |
|
39 |
from ganeti import utils |
40 |
from ganeti import constants |
41 |
from ganeti import serializer |
42 |
from ganeti import ssconf |
43 |
from ganeti import errors |
44 |
|
45 |
|
46 |
MAXTRIES = 5 |
47 |
BAD_STATES = ['stopped'] |
48 |
HELPLESS_STATES = ['(node down)'] |
49 |
NOTICE = 'NOTICE' |
50 |
ERROR = 'ERROR' |
51 |
KEY_RESTART_COUNT = "restart_count" |
52 |
KEY_RESTART_WHEN = "restart_when" |
53 |
KEY_BOOT_ID = "bootid" |
54 |
|
55 |
|
56 |
class Error(Exception): |
57 |
"""Generic custom error class.""" |
58 |
|
59 |
|
60 |
class NotMasterError(Error): |
61 |
"""Exception raised when this host is not the master.""" |
62 |
|
63 |
|
64 |
def Indent(s, prefix='| '): |
65 |
"""Indent a piece of text with a given prefix before each line. |
66 |
|
67 |
Args: |
68 |
s: The string to indent |
69 |
prefix: The string to prepend each line. |
70 |
|
71 |
""" |
72 |
return "%s%s\n" % (prefix, ('\n' + prefix).join(s.splitlines())) |
73 |
|
74 |
|
75 |
def DoCmd(cmd): |
76 |
"""Run a shell command. |
77 |
|
78 |
Args: |
79 |
cmd: the command to run. |
80 |
|
81 |
Raises CommandError with verbose commentary on error. |
82 |
|
83 |
""" |
84 |
res = utils.RunCmd(cmd) |
85 |
|
86 |
if res.failed: |
87 |
raise Error("Command %s failed:\n%s\nstdout:\n%sstderr:\n%s" % |
88 |
(repr(cmd), |
89 |
Indent(res.fail_reason), |
90 |
Indent(res.stdout), |
91 |
Indent(res.stderr))) |
92 |
|
93 |
return res |
94 |
|
95 |
|
96 |
class WatcherState(object): |
97 |
"""Interface to a state file recording restart attempts. |
98 |
|
99 |
""" |
100 |
def __init__(self): |
101 |
"""Open, lock, read and parse the file. |
102 |
|
103 |
Raises StandardError on lock contention. |
104 |
|
105 |
""" |
106 |
# The two-step dance below is necessary to allow both opening existing |
107 |
# file read/write and creating if not existing. Vanilla open will truncate |
108 |
# an existing file -or- allow creating if not existing. |
109 |
f = os.open(constants.WATCHER_STATEFILE, os.O_RDWR | os.O_CREAT) |
110 |
f = os.fdopen(f, 'w+') |
111 |
|
112 |
try: |
113 |
fcntl.flock(f.fileno(), fcntl.LOCK_EX|fcntl.LOCK_NB) |
114 |
except IOError, x: |
115 |
if x.errno == errno.EAGAIN: |
116 |
raise StandardError("State file already locked") |
117 |
raise |
118 |
|
119 |
self.statefile = f |
120 |
|
121 |
try: |
122 |
self.data = serializer.Load(self.statefile.read()) |
123 |
except Exception, msg: |
124 |
# Ignore errors while loading the file and treat it as empty |
125 |
self.data = {} |
126 |
logging.warning(("Empty or invalid state file. Using defaults." |
127 |
" Error message: %s"), msg) |
128 |
|
129 |
if "instance" not in self.data: |
130 |
self.data["instance"] = {} |
131 |
if "node" not in self.data: |
132 |
self.data["node"] = {} |
133 |
|
134 |
def __del__(self): |
135 |
"""Called on destruction. |
136 |
|
137 |
""" |
138 |
if self.statefile: |
139 |
self._Close() |
140 |
|
141 |
def _Close(self): |
142 |
"""Unlock configuration file and close it. |
143 |
|
144 |
""" |
145 |
assert self.statefile |
146 |
|
147 |
fcntl.flock(self.statefile.fileno(), fcntl.LOCK_UN) |
148 |
|
149 |
self.statefile.close() |
150 |
self.statefile = None |
151 |
|
152 |
def GetNodeBootID(self, name): |
153 |
"""Returns the last boot ID of a node or None. |
154 |
|
155 |
""" |
156 |
ndata = self.data["node"] |
157 |
|
158 |
if name in ndata and KEY_BOOT_ID in ndata[name]: |
159 |
return ndata[name][KEY_BOOT_ID] |
160 |
return None |
161 |
|
162 |
def SetNodeBootID(self, name, bootid): |
163 |
"""Sets the boot ID of a node. |
164 |
|
165 |
""" |
166 |
assert bootid |
167 |
|
168 |
ndata = self.data["node"] |
169 |
|
170 |
if name not in ndata: |
171 |
ndata[name] = {} |
172 |
|
173 |
ndata[name][KEY_BOOT_ID] = bootid |
174 |
|
175 |
def NumberOfRestartAttempts(self, instance): |
176 |
"""Returns number of previous restart attempts. |
177 |
|
178 |
Args: |
179 |
instance - the instance to look up. |
180 |
|
181 |
""" |
182 |
idata = self.data["instance"] |
183 |
|
184 |
if instance.name in idata: |
185 |
return idata[instance.name][KEY_RESTART_COUNT] |
186 |
|
187 |
return 0 |
188 |
|
189 |
def RecordRestartAttempt(self, instance): |
190 |
"""Record a restart attempt. |
191 |
|
192 |
Args: |
193 |
instance - the instance being restarted |
194 |
|
195 |
""" |
196 |
idata = self.data["instance"] |
197 |
|
198 |
if instance.name not in idata: |
199 |
inst = idata[instance.name] = {} |
200 |
else: |
201 |
inst = idata[instance.name] |
202 |
|
203 |
inst[KEY_RESTART_WHEN] = time.time() |
204 |
inst[KEY_RESTART_COUNT] = inst.get(KEY_RESTART_COUNT, 0) + 1 |
205 |
|
206 |
def RemoveInstance(self, instance): |
207 |
"""Update state to reflect that a machine is running, i.e. remove record. |
208 |
|
209 |
Args: |
210 |
instance - the instance to remove from books |
211 |
|
212 |
This method removes the record for a named instance. |
213 |
|
214 |
""" |
215 |
idata = self.data["instance"] |
216 |
|
217 |
if instance.name in idata: |
218 |
del idata[instance.name] |
219 |
|
220 |
def Save(self): |
221 |
"""Save state to file, then unlock and close it. |
222 |
|
223 |
""" |
224 |
assert self.statefile |
225 |
|
226 |
self.statefile.seek(0) |
227 |
self.statefile.truncate() |
228 |
|
229 |
self.statefile.write(serializer.Dump(self.data)) |
230 |
|
231 |
self._Close() |
232 |
|
233 |
|
234 |
class Instance(object): |
235 |
"""Abstraction for a Virtual Machine instance. |
236 |
|
237 |
Methods: |
238 |
Restart(): issue a command to restart the represented machine. |
239 |
|
240 |
""" |
241 |
def __init__(self, name, state, autostart): |
242 |
self.name = name |
243 |
self.state = state |
244 |
self.autostart = autostart |
245 |
|
246 |
def Restart(self): |
247 |
"""Encapsulates the start of an instance. |
248 |
|
249 |
""" |
250 |
DoCmd(['gnt-instance', 'startup', '--lock-retries=15', self.name]) |
251 |
|
252 |
def ActivateDisks(self): |
253 |
"""Encapsulates the activation of all disks of an instance. |
254 |
|
255 |
""" |
256 |
DoCmd(['gnt-instance', 'activate-disks', '--lock-retries=15', self.name]) |
257 |
|
258 |
|
259 |
def _RunListCmd(cmd): |
260 |
"""Runs a command and parses its output into lists. |
261 |
|
262 |
""" |
263 |
for line in DoCmd(cmd).stdout.splitlines(): |
264 |
yield line.split(':') |
265 |
|
266 |
|
267 |
def GetInstanceList(with_secondaries=None): |
268 |
"""Get a list of instances on this cluster. |
269 |
|
270 |
""" |
271 |
cmd = ['gnt-instance', 'list', '--lock-retries=15', '--no-headers', |
272 |
'--separator=:'] |
273 |
|
274 |
fields = 'name,oper_state,admin_state' |
275 |
|
276 |
if with_secondaries is not None: |
277 |
fields += ',snodes' |
278 |
|
279 |
cmd.append('-o') |
280 |
cmd.append(fields) |
281 |
|
282 |
instances = [] |
283 |
for fields in _RunListCmd(cmd): |
284 |
if with_secondaries is not None: |
285 |
(name, status, autostart, snodes) = fields |
286 |
|
287 |
if snodes == "-": |
288 |
continue |
289 |
|
290 |
for node in with_secondaries: |
291 |
if node in snodes.split(','): |
292 |
break |
293 |
else: |
294 |
continue |
295 |
|
296 |
else: |
297 |
(name, status, autostart) = fields |
298 |
|
299 |
instances.append(Instance(name, status, autostart != "no")) |
300 |
|
301 |
return instances |
302 |
|
303 |
|
304 |
def GetNodeBootIDs(): |
305 |
"""Get a dict mapping nodes to boot IDs. |
306 |
|
307 |
""" |
308 |
cmd = ['gnt-node', 'list', '--lock-retries=15', '--no-headers', |
309 |
'--separator=:', '-o', 'name,bootid'] |
310 |
|
311 |
ids = {} |
312 |
for fields in _RunListCmd(cmd): |
313 |
(name, bootid) = fields |
314 |
ids[name] = bootid |
315 |
|
316 |
return ids |
317 |
|
318 |
|
319 |
class Watcher(object): |
320 |
"""Encapsulate the logic for restarting erronously halted virtual machines. |
321 |
|
322 |
The calling program should periodically instantiate me and call Run(). |
323 |
This will traverse the list of instances, and make up to MAXTRIES attempts |
324 |
to restart machines that are down. |
325 |
|
326 |
""" |
327 |
def __init__(self): |
328 |
sstore = ssconf.SimpleStore() |
329 |
master = sstore.GetMasterNode() |
330 |
if master != utils.HostInfo().name: |
331 |
raise NotMasterError("This is not the master node") |
332 |
self.instances = GetInstanceList() |
333 |
self.bootids = GetNodeBootIDs() |
334 |
self.started_instances = set() |
335 |
|
336 |
def Run(self): |
337 |
notepad = WatcherState() |
338 |
try: |
339 |
self.CheckInstances(notepad) |
340 |
self.CheckDisks(notepad) |
341 |
self.VerifyDisks() |
342 |
finally: |
343 |
notepad.Save() |
344 |
|
345 |
def CheckDisks(self, notepad): |
346 |
"""Check all nodes for restarted ones. |
347 |
|
348 |
""" |
349 |
check_nodes = [] |
350 |
for name, id in self.bootids.iteritems(): |
351 |
old = notepad.GetNodeBootID(name) |
352 |
if old != id: |
353 |
# Node's boot ID has changed, proably through a reboot. |
354 |
check_nodes.append(name) |
355 |
|
356 |
if check_nodes: |
357 |
# Activate disks for all instances with any of the checked nodes as a |
358 |
# secondary node. |
359 |
for instance in GetInstanceList(with_secondaries=check_nodes): |
360 |
if not instance.autostart: |
361 |
logging.info(("Skipping disk activation for non-autostart" |
362 |
" instance %s"), instance.name) |
363 |
continue |
364 |
if instance.name in self.started_instances: |
365 |
# we already tried to start the instance, which should have |
366 |
# activated its drives (if they can be at all) |
367 |
continue |
368 |
try: |
369 |
logging.info("Activating disks for instance %s", instance.name) |
370 |
instance.ActivateDisks() |
371 |
except Error, err: |
372 |
logging.error(str(err), exc_info=True) |
373 |
|
374 |
# Keep changed boot IDs |
375 |
for name in check_nodes: |
376 |
notepad.SetNodeBootID(name, self.bootids[name]) |
377 |
|
378 |
def CheckInstances(self, notepad): |
379 |
"""Make a pass over the list of instances, restarting downed ones. |
380 |
|
381 |
""" |
382 |
for instance in self.instances: |
383 |
# Don't care about manually stopped instances |
384 |
if not instance.autostart: |
385 |
continue |
386 |
|
387 |
if instance.state in BAD_STATES: |
388 |
n = notepad.NumberOfRestartAttempts(instance) |
389 |
|
390 |
if n > MAXTRIES: |
391 |
# stay quiet. |
392 |
continue |
393 |
elif n < MAXTRIES: |
394 |
last = " (Attempt #%d)" % (n + 1) |
395 |
else: |
396 |
notepad.RecordRestartAttempt(instance) |
397 |
logging.error("Could not restart %s after %d attempts, giving up", |
398 |
instance.name, MAXTRIES) |
399 |
continue |
400 |
try: |
401 |
logging.info("Restarting %s%s", |
402 |
instance.name, last) |
403 |
instance.Restart() |
404 |
self.started_instances.add(instance.name) |
405 |
except Error, err: |
406 |
logging.error(str(err), exc_info=True) |
407 |
|
408 |
notepad.RecordRestartAttempt(instance) |
409 |
elif instance.state in HELPLESS_STATES: |
410 |
if notepad.NumberOfRestartAttempts(instance): |
411 |
notepad.RemoveInstance(instance) |
412 |
else: |
413 |
if notepad.NumberOfRestartAttempts(instance): |
414 |
notepad.RemoveInstance(instance) |
415 |
logging.info("Restart of %s succeeded", instance.name) |
416 |
|
417 |
def VerifyDisks(self): |
418 |
"""Run gnt-cluster verify-disks. |
419 |
|
420 |
""" |
421 |
result = DoCmd(['gnt-cluster', 'verify-disks', '--lock-retries=15']) |
422 |
if result.output: |
423 |
logging.info(result.output) |
424 |
|
425 |
|
426 |
def ParseOptions(): |
427 |
"""Parse the command line options. |
428 |
|
429 |
Returns: |
430 |
(options, args) as from OptionParser.parse_args() |
431 |
|
432 |
""" |
433 |
parser = OptionParser(description="Ganeti cluster watcher", |
434 |
usage="%prog [-d]", |
435 |
version="%%prog (ganeti) %s" % |
436 |
constants.RELEASE_VERSION) |
437 |
|
438 |
parser.add_option("-d", "--debug", dest="debug", |
439 |
help="Write all messages to stderr", |
440 |
default=False, action="store_true") |
441 |
options, args = parser.parse_args() |
442 |
return options, args |
443 |
|
444 |
|
445 |
def SetupLogging(debug): |
446 |
"""Configures the logging module. |
447 |
|
448 |
""" |
449 |
formatter = logging.Formatter("%(asctime)s: %(message)s") |
450 |
|
451 |
logfile_handler = logging.FileHandler(constants.LOG_WATCHER) |
452 |
logfile_handler.setFormatter(formatter) |
453 |
logfile_handler.setLevel(logging.INFO) |
454 |
|
455 |
stderr_handler = logging.StreamHandler() |
456 |
stderr_handler.setFormatter(formatter) |
457 |
if debug: |
458 |
stderr_handler.setLevel(logging.NOTSET) |
459 |
else: |
460 |
stderr_handler.setLevel(logging.CRITICAL) |
461 |
|
462 |
root_logger = logging.getLogger("") |
463 |
root_logger.setLevel(logging.NOTSET) |
464 |
root_logger.addHandler(logfile_handler) |
465 |
root_logger.addHandler(stderr_handler) |
466 |
|
467 |
|
468 |
def main(): |
469 |
"""Main function. |
470 |
|
471 |
""" |
472 |
options, args = ParseOptions() |
473 |
|
474 |
SetupLogging(options.debug) |
475 |
|
476 |
try: |
477 |
try: |
478 |
watcher = Watcher() |
479 |
except errors.ConfigurationError: |
480 |
# Just exit if there's no configuration |
481 |
sys.exit(constants.EXIT_SUCCESS) |
482 |
watcher.Run() |
483 |
except NotMasterError: |
484 |
logging.debug("Not master, exiting") |
485 |
sys.exit(constants.EXIT_NOTMASTER) |
486 |
except errors.ResolverError, err: |
487 |
logging.error("Cannot resolve hostname '%s', exiting.", err.args[0]) |
488 |
sys.exit(constants.EXIT_NODESETUP_ERROR) |
489 |
except Exception, err: |
490 |
logging.error(str(err), exc_info=True) |
491 |
sys.exit(constants.EXIT_FAILURE) |
492 |
|
493 |
|
494 |
if __name__ == '__main__': |
495 |
main() |