root / scripts / gnt-debug @ 19b9ba9a
History | View | Annotate | Download (16.7 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 |
"""Debugging commands""" |
22 |
|
23 |
# pylint: disable-msg=W0401,W0614,C0103 |
24 |
# W0401: Wildcard import ganeti.cli |
25 |
# W0614: Unused import %s from wildcard import (since we need cli) |
26 |
# C0103: Invalid name gnt-backup |
27 |
|
28 |
import sys |
29 |
import simplejson |
30 |
import time |
31 |
import socket |
32 |
import logging |
33 |
|
34 |
from ganeti.cli import * |
35 |
from ganeti import cli |
36 |
from ganeti import constants |
37 |
from ganeti import opcodes |
38 |
from ganeti import utils |
39 |
from ganeti import errors |
40 |
|
41 |
|
42 |
#: Default fields for L{ListLocks} |
43 |
_LIST_LOCKS_DEF_FIELDS = [ |
44 |
"name", |
45 |
"mode", |
46 |
"owner", |
47 |
] |
48 |
|
49 |
|
50 |
def Delay(opts, args): |
51 |
"""Sleeps for a while |
52 |
|
53 |
@param opts: the command line options selected by the user |
54 |
@type args: list |
55 |
@param args: should contain only one element, the duration |
56 |
the sleep |
57 |
@rtype: int |
58 |
@return: the desired exit code |
59 |
|
60 |
""" |
61 |
delay = float(args[0]) |
62 |
op = opcodes.OpTestDelay(duration=delay, |
63 |
on_master=opts.on_master, |
64 |
on_nodes=opts.on_nodes, |
65 |
repeat=opts.repeat) |
66 |
SubmitOpCode(op, opts=opts) |
67 |
|
68 |
return 0 |
69 |
|
70 |
|
71 |
def GenericOpCodes(opts, args): |
72 |
"""Send any opcode to the master. |
73 |
|
74 |
@param opts: the command line options selected by the user |
75 |
@type args: list |
76 |
@param args: should contain only one element, the path of |
77 |
the file with the opcode definition |
78 |
@rtype: int |
79 |
@return: the desired exit code |
80 |
|
81 |
""" |
82 |
cl = cli.GetClient() |
83 |
jex = cli.JobExecutor(cl=cl, verbose=opts.verbose, opts=opts) |
84 |
|
85 |
job_cnt = 0 |
86 |
op_cnt = 0 |
87 |
if opts.timing_stats: |
88 |
ToStdout("Loading...") |
89 |
for job_idx in range(opts.rep_job): |
90 |
for fname in args: |
91 |
# pylint: disable-msg=W0142 |
92 |
op_data = simplejson.loads(utils.ReadFile(fname)) |
93 |
op_list = [opcodes.OpCode.LoadOpCode(val) for val in op_data] |
94 |
op_list = op_list * opts.rep_op |
95 |
jex.QueueJob("file %s/%d" % (fname, job_idx), *op_list) |
96 |
op_cnt += len(op_list) |
97 |
job_cnt += 1 |
98 |
|
99 |
if opts.timing_stats: |
100 |
t1 = time.time() |
101 |
ToStdout("Submitting...") |
102 |
|
103 |
jex.SubmitPending(each=opts.each) |
104 |
|
105 |
if opts.timing_stats: |
106 |
t2 = time.time() |
107 |
ToStdout("Executing...") |
108 |
|
109 |
jex.GetResults() |
110 |
if opts.timing_stats: |
111 |
t3 = time.time() |
112 |
ToStdout("C:op %4d" % op_cnt) |
113 |
ToStdout("C:job %4d" % job_cnt) |
114 |
ToStdout("T:submit %4.4f" % (t2-t1)) |
115 |
ToStdout("T:exec %4.4f" % (t3-t2)) |
116 |
ToStdout("T:total %4.4f" % (t3-t1)) |
117 |
return 0 |
118 |
|
119 |
|
120 |
def TestAllocator(opts, args): |
121 |
"""Runs the test allocator opcode. |
122 |
|
123 |
@param opts: the command line options selected by the user |
124 |
@type args: list |
125 |
@param args: should contain only one element, the iallocator name |
126 |
@rtype: int |
127 |
@return: the desired exit code |
128 |
|
129 |
""" |
130 |
try: |
131 |
disks = [{"size": utils.ParseUnit(val), "mode": 'w'} |
132 |
for val in opts.disks.split(",")] |
133 |
except errors.UnitParseError, err: |
134 |
ToStderr("Invalid disks parameter '%s': %s", opts.disks, err) |
135 |
return 1 |
136 |
|
137 |
nics = [val.split("/") for val in opts.nics.split(",")] |
138 |
for row in nics: |
139 |
while len(row) < 3: |
140 |
row.append(None) |
141 |
for i in range(3): |
142 |
if row[i] == '': |
143 |
row[i] = None |
144 |
nic_dict = [{"mac": v[0], "ip": v[1], "bridge": v[2]} for v in nics] |
145 |
|
146 |
if opts.tags is None: |
147 |
opts.tags = [] |
148 |
else: |
149 |
opts.tags = opts.tags.split(",") |
150 |
|
151 |
op = opcodes.OpTestAllocator(mode=opts.mode, |
152 |
name=args[0], |
153 |
evac_nodes=args, |
154 |
mem_size=opts.mem, |
155 |
disks=disks, |
156 |
disk_template=opts.disk_template, |
157 |
nics=nic_dict, |
158 |
os=opts.os, |
159 |
vcpus=opts.vcpus, |
160 |
tags=opts.tags, |
161 |
direction=opts.direction, |
162 |
allocator=opts.iallocator, |
163 |
) |
164 |
result = SubmitOpCode(op, opts=opts) |
165 |
ToStdout("%s" % result) |
166 |
return 0 |
167 |
|
168 |
|
169 |
class _JobQueueTestReporter(cli.StdioJobPollReportCb): |
170 |
def __init__(self): |
171 |
"""Initializes this class. |
172 |
|
173 |
""" |
174 |
cli.StdioJobPollReportCb.__init__(self) |
175 |
self._expected_msgcount = 0 |
176 |
self._all_testmsgs = [] |
177 |
self._testmsgs = None |
178 |
self._job_id = None |
179 |
|
180 |
def GetTestMessages(self): |
181 |
"""Returns all test log messages received so far. |
182 |
|
183 |
""" |
184 |
return self._all_testmsgs |
185 |
|
186 |
def GetJobId(self): |
187 |
"""Returns the job ID. |
188 |
|
189 |
""" |
190 |
return self._job_id |
191 |
|
192 |
def ReportLogMessage(self, job_id, serial, timestamp, log_type, log_msg): |
193 |
"""Handles a log message. |
194 |
|
195 |
""" |
196 |
if self._job_id is None: |
197 |
self._job_id = job_id |
198 |
elif self._job_id != job_id: |
199 |
raise errors.ProgrammerError("The same reporter instance was used for" |
200 |
" more than one job") |
201 |
|
202 |
if log_type == constants.ELOG_JQUEUE_TEST: |
203 |
(sockname, test, arg) = log_msg |
204 |
return self._ProcessTestMessage(job_id, sockname, test, arg) |
205 |
|
206 |
elif (log_type == constants.ELOG_MESSAGE and |
207 |
log_msg.startswith(constants.JQT_MSGPREFIX)): |
208 |
if self._testmsgs is None: |
209 |
raise errors.OpExecError("Received test message without a preceding" |
210 |
" start message") |
211 |
testmsg = log_msg[len(constants.JQT_MSGPREFIX):] |
212 |
self._testmsgs.append(testmsg) |
213 |
self._all_testmsgs.append(testmsg) |
214 |
return |
215 |
|
216 |
return cli.StdioJobPollReportCb.ReportLogMessage(self, job_id, serial, |
217 |
timestamp, log_type, |
218 |
log_msg) |
219 |
|
220 |
def _ProcessTestMessage(self, job_id, sockname, test, arg): |
221 |
"""Handles a job queue test message. |
222 |
|
223 |
""" |
224 |
if test not in constants.JQT_ALL: |
225 |
raise errors.OpExecError("Received invalid test message %s" % test) |
226 |
|
227 |
sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM) |
228 |
try: |
229 |
sock.settimeout(30.0) |
230 |
|
231 |
logging.debug("Connecting to %s", sockname) |
232 |
sock.connect(sockname) |
233 |
|
234 |
logging.debug("Checking status") |
235 |
jobdetails = cli.GetClient().QueryJobs([job_id], ["status"])[0] |
236 |
if not jobdetails: |
237 |
raise errors.OpExecError("Can't find job %s" % job_id) |
238 |
|
239 |
status = jobdetails[0] |
240 |
|
241 |
logging.debug("Status of job %s is %s", job_id, status) |
242 |
|
243 |
if test == constants.JQT_EXPANDNAMES: |
244 |
if status != constants.JOB_STATUS_WAITLOCK: |
245 |
raise errors.OpExecError("Job status while expanding names is '%s'," |
246 |
" not '%s' as expected" % |
247 |
(status, constants.JOB_STATUS_WAITLOCK)) |
248 |
elif test in (constants.JQT_EXEC, constants.JQT_LOGMSG): |
249 |
if status != constants.JOB_STATUS_RUNNING: |
250 |
raise errors.OpExecError("Job status while executing opcode is '%s'," |
251 |
" not '%s' as expected" % |
252 |
(status, constants.JOB_STATUS_RUNNING)) |
253 |
|
254 |
if test == constants.JQT_STARTMSG: |
255 |
logging.debug("Expecting %s test messages", arg) |
256 |
self._testmsgs = [] |
257 |
elif test == constants.JQT_LOGMSG: |
258 |
if len(self._testmsgs) != arg: |
259 |
raise errors.OpExecError("Received %s test messages when %s are" |
260 |
" expected" % (len(self._testmsgs), arg)) |
261 |
finally: |
262 |
logging.debug("Closing socket") |
263 |
sock.close() |
264 |
|
265 |
|
266 |
def TestJobqueue(opts, _): |
267 |
"""Runs a few tests on the job queue. |
268 |
|
269 |
""" |
270 |
(TM_SUCCESS, |
271 |
TM_MULTISUCCESS, |
272 |
TM_FAIL, |
273 |
TM_PARTFAIL) = range(4) |
274 |
TM_ALL = frozenset([TM_SUCCESS, TM_MULTISUCCESS, TM_FAIL, TM_PARTFAIL]) |
275 |
|
276 |
for mode in TM_ALL: |
277 |
test_messages = [ |
278 |
"Testing mode %s" % mode, |
279 |
"Hello World", |
280 |
"A", |
281 |
"", |
282 |
"B" |
283 |
"Foo|bar|baz", |
284 |
utils.TimestampForFilename(), |
285 |
] |
286 |
|
287 |
fail = mode in (TM_FAIL, TM_PARTFAIL) |
288 |
|
289 |
if mode == TM_PARTFAIL: |
290 |
ToStdout("Testing partial job failure") |
291 |
ops = [ |
292 |
opcodes.OpTestJobqueue(notify_waitlock=True, notify_exec=True, |
293 |
log_messages=test_messages, fail=False), |
294 |
opcodes.OpTestJobqueue(notify_waitlock=True, notify_exec=True, |
295 |
log_messages=test_messages, fail=False), |
296 |
opcodes.OpTestJobqueue(notify_waitlock=True, notify_exec=True, |
297 |
log_messages=test_messages, fail=True), |
298 |
opcodes.OpTestJobqueue(notify_waitlock=True, notify_exec=True, |
299 |
log_messages=test_messages, fail=False), |
300 |
] |
301 |
expect_messages = 3 * [test_messages] |
302 |
expect_opstatus = [ |
303 |
constants.OP_STATUS_SUCCESS, |
304 |
constants.OP_STATUS_SUCCESS, |
305 |
constants.OP_STATUS_ERROR, |
306 |
constants.OP_STATUS_ERROR, |
307 |
] |
308 |
expect_resultlen = 2 |
309 |
elif mode == TM_MULTISUCCESS: |
310 |
ToStdout("Testing multiple successful opcodes") |
311 |
ops = [ |
312 |
opcodes.OpTestJobqueue(notify_waitlock=True, notify_exec=True, |
313 |
log_messages=test_messages, fail=False), |
314 |
opcodes.OpTestJobqueue(notify_waitlock=True, notify_exec=True, |
315 |
log_messages=test_messages, fail=False), |
316 |
] |
317 |
expect_messages = 2 * [test_messages] |
318 |
expect_opstatus = [ |
319 |
constants.OP_STATUS_SUCCESS, |
320 |
constants.OP_STATUS_SUCCESS, |
321 |
] |
322 |
expect_resultlen = 2 |
323 |
else: |
324 |
if mode == TM_SUCCESS: |
325 |
ToStdout("Testing job success") |
326 |
expect_opstatus = [constants.OP_STATUS_SUCCESS] |
327 |
elif mode == TM_FAIL: |
328 |
ToStdout("Testing job failure") |
329 |
expect_opstatus = [constants.OP_STATUS_ERROR] |
330 |
else: |
331 |
raise errors.ProgrammerError("Unknown test mode %s" % mode) |
332 |
|
333 |
ops = [ |
334 |
opcodes.OpTestJobqueue(notify_waitlock=True, |
335 |
notify_exec=True, |
336 |
log_messages=test_messages, |
337 |
fail=fail) |
338 |
] |
339 |
expect_messages = [test_messages] |
340 |
expect_resultlen = 1 |
341 |
|
342 |
cl = cli.GetClient() |
343 |
cli.SetGenericOpcodeOpts(ops, opts) |
344 |
|
345 |
# Send job to master daemon |
346 |
job_id = cli.SendJob(ops, cl=cl) |
347 |
|
348 |
reporter = _JobQueueTestReporter() |
349 |
results = None |
350 |
|
351 |
try: |
352 |
results = cli.PollJob(job_id, cl=cl, reporter=reporter) |
353 |
except errors.OpExecError, err: |
354 |
if not fail: |
355 |
raise |
356 |
ToStdout("Ignoring error: %s", err) |
357 |
else: |
358 |
if fail: |
359 |
raise errors.OpExecError("Job didn't fail when it should") |
360 |
|
361 |
# Check length of result |
362 |
if fail: |
363 |
if results is not None: |
364 |
raise errors.OpExecError("Received result from failed job") |
365 |
elif len(results) != expect_resultlen: |
366 |
raise errors.OpExecError("Received %s results (%s), expected %s" % |
367 |
(len(results), results, expect_resultlen)) |
368 |
|
369 |
# Check received log messages |
370 |
all_messages = [i for j in expect_messages for i in j] |
371 |
if reporter.GetTestMessages() != all_messages: |
372 |
raise errors.OpExecError("Received test messages don't match input" |
373 |
" (input %r, received %r)" % |
374 |
(all_messages, reporter.GetTestMessages())) |
375 |
|
376 |
# Check final status |
377 |
reported_job_id = reporter.GetJobId() |
378 |
if reported_job_id != job_id: |
379 |
raise errors.OpExecError("Reported job ID %s doesn't match" |
380 |
"submission job ID %s" % |
381 |
(reported_job_id, job_id)) |
382 |
|
383 |
jobdetails = cli.GetClient().QueryJobs([job_id], ["status", "opstatus"])[0] |
384 |
if not jobdetails: |
385 |
raise errors.OpExecError("Can't find job %s" % job_id) |
386 |
|
387 |
if fail: |
388 |
exp_status = constants.JOB_STATUS_ERROR |
389 |
else: |
390 |
exp_status = constants.JOB_STATUS_SUCCESS |
391 |
|
392 |
(final_status, final_opstatus) = jobdetails |
393 |
if final_status != exp_status: |
394 |
raise errors.OpExecError("Final job status is %s, not %s as expected" % |
395 |
(final_status, exp_status)) |
396 |
if len(final_opstatus) != len(ops): |
397 |
raise errors.OpExecError("Did not receive status for all opcodes (got %s," |
398 |
" expected %s)" % |
399 |
(len(final_opstatus), len(ops))) |
400 |
if final_opstatus != expect_opstatus: |
401 |
raise errors.OpExecError("Opcode status is %s, expected %s" % |
402 |
(final_opstatus, expect_opstatus)) |
403 |
|
404 |
ToStdout("Job queue test successful") |
405 |
|
406 |
return 0 |
407 |
|
408 |
|
409 |
def ListLocks(opts, args): # pylint: disable-msg=W0613 |
410 |
"""List all locks. |
411 |
|
412 |
@param opts: the command line options selected by the user |
413 |
@type args: list |
414 |
@param args: should be an empty list |
415 |
@rtype: int |
416 |
@return: the desired exit code |
417 |
|
418 |
""" |
419 |
selected_fields = ParseFields(opts.output, _LIST_LOCKS_DEF_FIELDS) |
420 |
|
421 |
if not opts.no_headers: |
422 |
headers = { |
423 |
"name": "Name", |
424 |
"mode": "Mode", |
425 |
"owner": "Owner", |
426 |
} |
427 |
else: |
428 |
headers = None |
429 |
|
430 |
while True: |
431 |
# Not reusing client as interval might be too long |
432 |
output = GetClient().QueryLocks(selected_fields, False) |
433 |
|
434 |
# change raw values to nicer strings |
435 |
for row in output: |
436 |
for idx, field in enumerate(selected_fields): |
437 |
val = row[idx] |
438 |
|
439 |
if field in ("mode", "owner") and val is None: |
440 |
val = "-" |
441 |
elif field == "owner": |
442 |
val = utils.CommaJoin(val) |
443 |
|
444 |
row[idx] = str(val) |
445 |
|
446 |
data = GenerateTable(separator=opts.separator, headers=headers, |
447 |
fields=selected_fields, data=output) |
448 |
for line in data: |
449 |
ToStdout(line) |
450 |
|
451 |
if not opts.interval: |
452 |
break |
453 |
|
454 |
ToStdout("") |
455 |
time.sleep(opts.interval) |
456 |
|
457 |
return 0 |
458 |
|
459 |
|
460 |
commands = { |
461 |
'delay': ( |
462 |
Delay, [ArgUnknown(min=1, max=1)], |
463 |
[cli_option("--no-master", dest="on_master", default=True, |
464 |
action="store_false", help="Do not sleep in the master code"), |
465 |
cli_option("-n", dest="on_nodes", default=[], |
466 |
action="append", help="Select nodes to sleep on"), |
467 |
cli_option("-r", "--repeat", type="int", default="0", dest="repeat", |
468 |
help="Number of times to repeat the sleep"), |
469 |
DRY_RUN_OPT, |
470 |
], |
471 |
"[opts...] <duration>", "Executes a TestDelay OpCode"), |
472 |
'submit-job': ( |
473 |
GenericOpCodes, [ArgFile(min=1)], |
474 |
[VERBOSE_OPT, |
475 |
cli_option("--op-repeat", type="int", default="1", dest="rep_op", |
476 |
help="Repeat the opcode sequence this number of times"), |
477 |
cli_option("--job-repeat", type="int", default="1", dest="rep_job", |
478 |
help="Repeat the job this number of times"), |
479 |
cli_option("--timing-stats", default=False, |
480 |
action="store_true", help="Show timing stats"), |
481 |
cli_option("--each", default=False, action="store_true", |
482 |
help="Submit each job separately"), |
483 |
DRY_RUN_OPT, |
484 |
], |
485 |
"<op_list_file...>", "Submits jobs built from json files" |
486 |
" containing a list of serialized opcodes"), |
487 |
'allocator': ( |
488 |
TestAllocator, [ArgUnknown(min=1)], |
489 |
[cli_option("--dir", dest="direction", |
490 |
default="in", choices=["in", "out"], |
491 |
help="Show allocator input (in) or allocator" |
492 |
" results (out)"), |
493 |
IALLOCATOR_OPT, |
494 |
cli_option("-m", "--mode", default="relocate", |
495 |
choices=["relocate", "allocate", "multi-evacuate"], |
496 |
help="Request mode, either allocate or relocate"), |
497 |
cli_option("--mem", default=128, type="unit", |
498 |
help="Memory size for the instance (MiB)"), |
499 |
cli_option("--disks", default="4096,4096", |
500 |
help="Comma separated list of disk sizes (MiB)"), |
501 |
DISK_TEMPLATE_OPT, |
502 |
cli_option("--nics", default="00:11:22:33:44:55", |
503 |
help="Comma separated list of nics, each nic" |
504 |
" definition is of form mac/ip/bridge, if" |
505 |
" missing values are replace by None"), |
506 |
OS_OPT, |
507 |
cli_option("-p", "--vcpus", default=1, type="int", |
508 |
help="Select number of VCPUs for the instance"), |
509 |
cli_option("--tags", default=None, |
510 |
help="Comma separated list of tags"), |
511 |
DRY_RUN_OPT, |
512 |
], |
513 |
"{opts...} <instance>", "Executes a TestAllocator OpCode"), |
514 |
"test-jobqueue": ( |
515 |
TestJobqueue, ARGS_NONE, [], |
516 |
"", "Test a few aspects of the job queue"), |
517 |
"locks": ( |
518 |
ListLocks, ARGS_NONE, [NOHDR_OPT, SEP_OPT, FIELDS_OPT, INTERVAL_OPT], |
519 |
"[--interval N]", "Show a list of locks in the master daemon"), |
520 |
} |
521 |
|
522 |
|
523 |
if __name__ == '__main__': |
524 |
sys.exit(GenericMain(commands)) |