root / daemons / ganeti-watcher @ 2ec08468
History | View | Annotate | Download (9.2 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 |
LOGFILE = '/var/log/ganeti/watcher.log' |
31 |
MAXTRIES = 5 |
32 |
BAD_STATES = ['stopped'] |
33 |
HELPLESS_STATES = ['(node down)'] |
34 |
NOTICE = 'NOTICE' |
35 |
ERROR = 'ERROR' |
36 |
|
37 |
import os |
38 |
import sys |
39 |
import time |
40 |
import fcntl |
41 |
import errno |
42 |
import socket |
43 |
from optparse import OptionParser |
44 |
|
45 |
|
46 |
from ganeti import utils |
47 |
from ganeti import constants |
48 |
from ganeti import ssconf |
49 |
|
50 |
|
51 |
class Error(Exception): |
52 |
"""Generic custom error class.""" |
53 |
|
54 |
|
55 |
class NotMasterError(Error): |
56 |
"""Exception raised when this host is not the master.""" |
57 |
|
58 |
|
59 |
def Indent(s, prefix='| '): |
60 |
"""Indent a piece of text with a given prefix before each line. |
61 |
|
62 |
Args: |
63 |
s: The string to indent |
64 |
prefix: The string to prepend each line. |
65 |
|
66 |
""" |
67 |
return "%s%s\n" % (prefix, ('\n' + prefix).join(s.splitlines())) |
68 |
|
69 |
|
70 |
def DoCmd(cmd): |
71 |
"""Run a shell command. |
72 |
|
73 |
Args: |
74 |
cmd: the command to run. |
75 |
|
76 |
Raises CommandError with verbose commentary on error. |
77 |
|
78 |
""" |
79 |
res = utils.RunCmd(cmd) |
80 |
|
81 |
if res.failed: |
82 |
raise Error("Command %s failed:\n%s\nstdout:\n%sstderr:\n%s" % |
83 |
(repr(cmd), |
84 |
Indent(res.fail_reason), |
85 |
Indent(res.stdout), |
86 |
Indent(res.stderr))) |
87 |
|
88 |
return res |
89 |
|
90 |
|
91 |
class RestarterState(object): |
92 |
"""Interface to a state file recording restart attempts. |
93 |
|
94 |
Methods: |
95 |
Open(): open, lock, read and parse the file. |
96 |
Raises StandardError on lock contention. |
97 |
|
98 |
NumberOfAttempts(name): returns the number of times in succession |
99 |
a restart has been attempted of the named instance. |
100 |
|
101 |
RecordAttempt(name, when): records one restart attempt of name at |
102 |
time in when. |
103 |
|
104 |
Remove(name): remove record given by name, if exists. |
105 |
|
106 |
Save(name): saves all records to file, releases lock and closes file. |
107 |
|
108 |
""" |
109 |
def __init__(self): |
110 |
# The two-step dance below is necessary to allow both opening existing |
111 |
# file read/write and creating if not existing. Vanilla open will truncate |
112 |
# an existing file -or- allow creating if not existing. |
113 |
f = os.open(constants.WATCHER_STATEFILE, os.O_RDWR | os.O_CREAT) |
114 |
f = os.fdopen(f, 'w+') |
115 |
|
116 |
try: |
117 |
fcntl.flock(f.fileno(), fcntl.LOCK_EX|fcntl.LOCK_NB) |
118 |
except IOError, x: |
119 |
if x.errno == errno.EAGAIN: |
120 |
raise StandardError("State file already locked") |
121 |
raise |
122 |
|
123 |
self.statefile = f |
124 |
self.inst_map = {} |
125 |
|
126 |
for line in f: |
127 |
name, when, count = line.rstrip().split(':') |
128 |
|
129 |
when = int(when) |
130 |
count = int(count) |
131 |
|
132 |
self.inst_map[name] = (when, count) |
133 |
|
134 |
def NumberOfAttempts(self, instance): |
135 |
"""Returns number of previous restart attempts. |
136 |
|
137 |
Args: |
138 |
instance - the instance to look up. |
139 |
|
140 |
""" |
141 |
assert self.statefile |
142 |
|
143 |
if instance.name in self.inst_map: |
144 |
return self.inst_map[instance.name][1] |
145 |
|
146 |
return 0 |
147 |
|
148 |
def RecordAttempt(self, instance): |
149 |
"""Record a restart attempt. |
150 |
|
151 |
Args: |
152 |
instance - the instance being restarted |
153 |
|
154 |
""" |
155 |
assert self.statefile |
156 |
|
157 |
when = time.time() |
158 |
|
159 |
self.inst_map[instance.name] = (when, 1 + self.NumberOfAttempts(instance)) |
160 |
|
161 |
def Remove(self, instance): |
162 |
"""Update state to reflect that a machine is running, i.e. remove record. |
163 |
|
164 |
Args: |
165 |
instance - the instance to remove from books |
166 |
|
167 |
This method removes the record for a named instance. |
168 |
|
169 |
""" |
170 |
assert self.statefile |
171 |
|
172 |
if instance.name in self.inst_map: |
173 |
del self.inst_map[instance.name] |
174 |
|
175 |
def Save(self): |
176 |
"""Save records to file, then unlock and close file. |
177 |
|
178 |
""" |
179 |
assert self.statefile |
180 |
|
181 |
self.statefile.seek(0) |
182 |
self.statefile.truncate() |
183 |
|
184 |
for name in self.inst_map: |
185 |
print >> self.statefile, "%s:%d:%d" % ((name,) + self.inst_map[name]) |
186 |
|
187 |
fcntl.flock(self.statefile.fileno(), fcntl.LOCK_UN) |
188 |
|
189 |
self.statefile.close() |
190 |
self.statefile = None |
191 |
|
192 |
|
193 |
class Instance(object): |
194 |
"""Abstraction for a Virtual Machine instance. |
195 |
|
196 |
Methods: |
197 |
Restart(): issue a command to restart the represented machine. |
198 |
|
199 |
""" |
200 |
def __init__(self, name, state): |
201 |
self.name = name |
202 |
self.state = state |
203 |
|
204 |
def Restart(self): |
205 |
"""Encapsulates the start of an instance. |
206 |
|
207 |
This is currently done using the command line interface and not |
208 |
the Ganeti modules. |
209 |
|
210 |
""" |
211 |
DoCmd(['gnt-instance', 'startup', '--lock-retries=15', self.name]) |
212 |
|
213 |
|
214 |
class InstanceList(object): |
215 |
"""The set of Virtual Machine instances on a cluster. |
216 |
|
217 |
""" |
218 |
cmd = ['gnt-instance', 'list', '--lock-retries=15', |
219 |
'-o', 'name,admin_state,oper_state', '--no-headers', '--separator=:'] |
220 |
|
221 |
def __init__(self): |
222 |
res = DoCmd(self.cmd) |
223 |
|
224 |
lines = res.stdout.splitlines() |
225 |
|
226 |
self.instances = [] |
227 |
for line in lines: |
228 |
fields = [fld.strip() for fld in line.split(':')] |
229 |
|
230 |
if len(fields) != 3: |
231 |
continue |
232 |
if fields[1] == "no": #no autostart, we don't care about this instance |
233 |
continue |
234 |
name, status = fields[0], fields[2] |
235 |
|
236 |
self.instances.append(Instance(name, status)) |
237 |
|
238 |
def __iter__(self): |
239 |
return self.instances.__iter__() |
240 |
|
241 |
|
242 |
class Message(object): |
243 |
"""Encapsulation of a notice or error message. |
244 |
|
245 |
""" |
246 |
def __init__(self, level, msg): |
247 |
self.level = level |
248 |
self.msg = msg |
249 |
self.when = time.time() |
250 |
|
251 |
def __str__(self): |
252 |
return self.level + ' ' + time.ctime(self.when) + '\n' + Indent(self.msg) |
253 |
|
254 |
|
255 |
class Restarter(object): |
256 |
"""Encapsulate the logic for restarting erronously halted virtual machines. |
257 |
|
258 |
The calling program should periodically instantiate me and call Run(). |
259 |
This will traverse the list of instances, and make up to MAXTRIES attempts |
260 |
to restart machines that are down. |
261 |
|
262 |
""" |
263 |
def __init__(self): |
264 |
sstore = ssconf.SimpleStore() |
265 |
master = sstore.GetMasterNode() |
266 |
if master != socket.gethostname(): |
267 |
raise NotMasterError("This is not the master node") |
268 |
self.instances = InstanceList() |
269 |
self.messages = [] |
270 |
|
271 |
def Run(self): |
272 |
"""Make a pass over the list of instances, restarting downed ones. |
273 |
|
274 |
""" |
275 |
notepad = RestarterState() |
276 |
|
277 |
for instance in self.instances: |
278 |
if instance.state in BAD_STATES: |
279 |
n = notepad.NumberOfAttempts(instance) |
280 |
|
281 |
if n > MAXTRIES: |
282 |
# stay quiet. |
283 |
continue |
284 |
elif n < MAXTRIES: |
285 |
last = " (Attempt #%d)" % (n + 1) |
286 |
else: |
287 |
notepad.RecordAttempt(instance) |
288 |
self.messages.append(Message(ERROR, "Could not restart %s for %d" |
289 |
" times, giving up..." % |
290 |
(instance.name, MAXTRIES))) |
291 |
continue |
292 |
try: |
293 |
self.messages.append(Message(NOTICE, |
294 |
"Restarting %s%s." % |
295 |
(instance.name, last))) |
296 |
instance.Restart() |
297 |
except Error, x: |
298 |
self.messages.append(Message(ERROR, str(x))) |
299 |
|
300 |
notepad.RecordAttempt(instance) |
301 |
elif instance.state in HELPLESS_STATES: |
302 |
if notepad.NumberOfAttempts(instance): |
303 |
notepad.Remove(instance) |
304 |
else: |
305 |
if notepad.NumberOfAttempts(instance): |
306 |
notepad.Remove(instance) |
307 |
msg = Message(NOTICE, |
308 |
"Restart of %s succeeded." % instance.name) |
309 |
self.messages.append(msg) |
310 |
|
311 |
notepad.Save() |
312 |
|
313 |
def WriteReport(self, logfile): |
314 |
"""Log all messages to file. |
315 |
|
316 |
Args: |
317 |
logfile: file object open for writing (the log file) |
318 |
|
319 |
""" |
320 |
for msg in self.messages: |
321 |
print >> logfile, str(msg) |
322 |
|
323 |
|
324 |
def ParseOptions(): |
325 |
"""Parse the command line options. |
326 |
|
327 |
Returns: |
328 |
(options, args) as from OptionParser.parse_args() |
329 |
|
330 |
""" |
331 |
parser = OptionParser(description="Ganeti cluster watcher", |
332 |
usage="%prog [-d]", |
333 |
version="%%prog (ganeti) %s" % |
334 |
constants.RELEASE_VERSION) |
335 |
|
336 |
parser.add_option("-d", "--debug", dest="debug", |
337 |
help="Don't redirect messages to the log file", |
338 |
default=False, action="store_true") |
339 |
options, args = parser.parse_args() |
340 |
return options, args |
341 |
|
342 |
|
343 |
def main(): |
344 |
"""Main function. |
345 |
|
346 |
""" |
347 |
options, args = ParseOptions() |
348 |
|
349 |
if not options.debug: |
350 |
sys.stderr = sys.stdout = open(LOGFILE, 'a') |
351 |
|
352 |
try: |
353 |
restarter = Restarter() |
354 |
restarter.Run() |
355 |
restarter.WriteReport(sys.stdout) |
356 |
except NotMasterError: |
357 |
if options.debug: |
358 |
sys.stderr.write("Not master, exiting.\n") |
359 |
sys.exit(constants.EXIT_NOTMASTER) |
360 |
except Error, err: |
361 |
print err |
362 |
|
363 |
if __name__ == '__main__': |
364 |
main() |