root / daemons / ganeti-watcher @ 28cc354f
History | View | Annotate | Download (8.7 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 |
from optparse import OptionParser |
43 |
|
44 |
|
45 |
from ganeti import utils |
46 |
from ganeti import constants |
47 |
|
48 |
|
49 |
class Error(Exception): |
50 |
"""Generic custom error class.""" |
51 |
pass |
52 |
|
53 |
|
54 |
def Indent(s, prefix='| '): |
55 |
"""Indent a piece of text with a given prefix before each line. |
56 |
|
57 |
Args: |
58 |
s: The string to indent |
59 |
prefix: The string to prepend each line. |
60 |
""" |
61 |
return "%s%s\n" % (prefix, ('\n' + prefix).join(s.splitlines())) |
62 |
|
63 |
|
64 |
def DoCmd(cmd): |
65 |
"""Run a shell command. |
66 |
|
67 |
Args: |
68 |
cmd: the command to run. |
69 |
|
70 |
Raises CommandError with verbose commentary on error. |
71 |
""" |
72 |
res = utils.RunCmd(cmd) |
73 |
|
74 |
if res.failed: |
75 |
raise Error("Command %s failed:\n%s\nstdout:\n%sstderr:\n%s" % |
76 |
(repr(cmd), |
77 |
Indent(res.fail_reason), |
78 |
Indent(res.stdout), |
79 |
Indent(res.stderr))) |
80 |
|
81 |
return res |
82 |
|
83 |
|
84 |
class RestarterState(object): |
85 |
"""Interface to a state file recording restart attempts. |
86 |
|
87 |
Methods: |
88 |
Open(): open, lock, read and parse the file. |
89 |
Raises StandardError on lock contention. |
90 |
|
91 |
NumberOfAttempts(name): returns the number of times in succession |
92 |
a restart has been attempted of the named instance. |
93 |
|
94 |
RecordAttempt(name, when): records one restart attempt of name at |
95 |
time in when. |
96 |
|
97 |
Remove(name): remove record given by name, if exists. |
98 |
|
99 |
Save(name): saves all records to file, releases lock and closes file. |
100 |
""" |
101 |
def __init__(self): |
102 |
# The two-step dance below is necessary to allow both opening existing |
103 |
# file read/write and creating if not existing. Vanilla open will truncate |
104 |
# an existing file -or- allow creating if not existing. |
105 |
f = os.open(constants.WATCHER_STATEFILE, os.O_RDWR | os.O_CREAT) |
106 |
f = os.fdopen(f, 'w+') |
107 |
|
108 |
try: |
109 |
fcntl.flock(f.fileno(), fcntl.LOCK_EX|fcntl.LOCK_NB) |
110 |
except IOError, x: |
111 |
if x.errno == errno.EAGAIN: |
112 |
raise StandardError('State file already locked') |
113 |
raise |
114 |
|
115 |
self.statefile = f |
116 |
self.inst_map = {} |
117 |
|
118 |
for line in f: |
119 |
name, when, count = line.rstrip().split(':') |
120 |
|
121 |
when = int(when) |
122 |
count = int(count) |
123 |
|
124 |
self.inst_map[name] = (when, count) |
125 |
|
126 |
def NumberOfAttempts(self, instance): |
127 |
"""Returns number of previous restart attempts. |
128 |
|
129 |
Args: |
130 |
instance - the instance to look up. |
131 |
""" |
132 |
assert self.statefile |
133 |
|
134 |
if instance.name in self.inst_map: |
135 |
return self.inst_map[instance.name][1] |
136 |
|
137 |
return 0 |
138 |
|
139 |
def RecordAttempt(self, instance): |
140 |
"""Record a restart attempt. |
141 |
|
142 |
Args: |
143 |
instance - the instance being restarted |
144 |
""" |
145 |
assert self.statefile |
146 |
|
147 |
when = time.time() |
148 |
|
149 |
self.inst_map[instance.name] = (when, 1 + self.NumberOfAttempts(instance)) |
150 |
|
151 |
def Remove(self, instance): |
152 |
"""Update state to reflect that a machine is running, i.e. remove record |
153 |
|
154 |
Args: |
155 |
instance - the instance to remove from books |
156 |
|
157 |
This method removes the record for a named instance |
158 |
""" |
159 |
assert self.statefile |
160 |
|
161 |
if instance.name in self.inst_map: |
162 |
del self.inst_map[instance.name] |
163 |
|
164 |
def Save(self): |
165 |
"""Save records to file, then unlock and close file. |
166 |
""" |
167 |
assert self.statefile |
168 |
|
169 |
self.statefile.seek(0) |
170 |
self.statefile.truncate() |
171 |
|
172 |
for name in self.inst_map: |
173 |
print >> self.statefile, "%s:%d:%d" % ((name,) + self.inst_map[name]) |
174 |
|
175 |
fcntl.flock(self.statefile.fileno(), fcntl.LOCK_UN) |
176 |
|
177 |
self.statefile.close() |
178 |
self.statefile = None |
179 |
|
180 |
|
181 |
class Instance(object): |
182 |
"""Abstraction for a Virtual Machine instance. |
183 |
|
184 |
Methods: |
185 |
Restart(): issue a command to restart the represented machine. |
186 |
""" |
187 |
def __init__(self, name, state): |
188 |
self.name = name |
189 |
self.state = state |
190 |
|
191 |
def Restart(self): |
192 |
DoCmd(['gnt-instance', 'startup', '--lock-retries=15', self.name]) |
193 |
|
194 |
|
195 |
class InstanceList(object): |
196 |
"""The set of Virtual Machine instances on a cluster. |
197 |
""" |
198 |
cmd = ['gnt-instance', 'list', '--lock-retries=15', |
199 |
'-o', 'name,admin_state,oper_state', '--no-headers', '--separator=:'] |
200 |
|
201 |
def __init__(self): |
202 |
res = DoCmd(self.cmd) |
203 |
|
204 |
lines = res.stdout.splitlines() |
205 |
|
206 |
self.instances = [] |
207 |
for line in lines: |
208 |
fields = [fld.strip() for fld in line.split(':')] |
209 |
|
210 |
if len(fields) != 3: |
211 |
continue |
212 |
if fields[1] == "no": #no autostart, we don't care about this instance |
213 |
continue |
214 |
name, status = fields[0], fields[2] |
215 |
|
216 |
self.instances.append(Instance(name, status)) |
217 |
|
218 |
def __iter__(self): |
219 |
return self.instances.__iter__() |
220 |
|
221 |
|
222 |
class Message(object): |
223 |
"""Encapsulation of a notice or error message. |
224 |
""" |
225 |
def __init__(self, level, msg): |
226 |
self.level = level |
227 |
self.msg = msg |
228 |
self.when = time.time() |
229 |
|
230 |
def __str__(self): |
231 |
return self.level + ' ' + time.ctime(self.when) + '\n' + Indent(self.msg) |
232 |
|
233 |
|
234 |
class Restarter(object): |
235 |
"""Encapsulate the logic for restarting erronously halted virtual machines. |
236 |
|
237 |
The calling program should periodically instantiate me and call Run(). |
238 |
This will traverse the list of instances, and make up to MAXTRIES attempts |
239 |
to restart machines that are down. |
240 |
""" |
241 |
def __init__(self): |
242 |
self.instances = InstanceList() |
243 |
self.messages = [] |
244 |
|
245 |
def Run(self): |
246 |
"""Make a pass over the list of instances, restarting downed ones. |
247 |
""" |
248 |
notepad = RestarterState() |
249 |
|
250 |
for instance in self.instances: |
251 |
if instance.state in BAD_STATES: |
252 |
n = notepad.NumberOfAttempts(instance) |
253 |
|
254 |
if n > MAXTRIES: |
255 |
# stay quiet. |
256 |
continue |
257 |
elif n < MAXTRIES: |
258 |
last = " (Attempt #%d)" % (n + 1) |
259 |
else: |
260 |
notepad.RecordAttempt(instance) |
261 |
self.messages.append(Message(ERROR, "Could not restart %s for %d" |
262 |
" times, giving up..." % |
263 |
(instance.name, MAXTRIES))) |
264 |
continue |
265 |
try: |
266 |
self.messages.append(Message(NOTICE, |
267 |
"Restarting %s%s." % |
268 |
(instance.name, last))) |
269 |
instance.Restart() |
270 |
except Error, x: |
271 |
self.messages.append(Message(ERROR, str(x))) |
272 |
|
273 |
notepad.RecordAttempt(instance) |
274 |
elif instance.state in HELPLESS_STATES: |
275 |
if notepad.NumberOfAttempts(instance): |
276 |
notepad.Remove(instance) |
277 |
else: |
278 |
if notepad.NumberOfAttempts(instance): |
279 |
notepad.Remove(instance) |
280 |
msg = Message(NOTICE, |
281 |
"Restart of %s succeeded." % instance.name) |
282 |
self.messages.append(msg) |
283 |
|
284 |
notepad.Save() |
285 |
|
286 |
def WriteReport(self, logfile): |
287 |
""" |
288 |
Log all messages to file. |
289 |
|
290 |
Args: |
291 |
logfile: file object open for writing (the log file) |
292 |
""" |
293 |
for msg in self.messages: |
294 |
print >> logfile, str(msg) |
295 |
|
296 |
|
297 |
def ParseOptions(): |
298 |
"""Parse the command line options. |
299 |
|
300 |
Returns: |
301 |
(options, args) as from OptionParser.parse_args() |
302 |
|
303 |
""" |
304 |
parser = OptionParser(description="Ganeti cluster watcher", |
305 |
usage="%prog [-d]", |
306 |
version="%%prog (ganeti) %s" % |
307 |
constants.RELEASE_VERSION) |
308 |
|
309 |
parser.add_option("-d", "--debug", dest="debug", |
310 |
help="Don't redirect messages to the log file", |
311 |
default=False, action="store_true") |
312 |
options, args = parser.parse_args() |
313 |
return options, args |
314 |
|
315 |
|
316 |
def main(): |
317 |
"""Main function. |
318 |
|
319 |
""" |
320 |
options, args = ParseOptions() |
321 |
|
322 |
if not options.debug: |
323 |
sys.stderr = sys.stdout = open(LOGFILE, 'a') |
324 |
|
325 |
try: |
326 |
restarter = Restarter() |
327 |
restarter.Run() |
328 |
restarter.WriteReport(sys.stdout) |
329 |
except Error, err: |
330 |
print err |
331 |
|
332 |
if __name__ == '__main__': |
333 |
main() |