root / lib / mcpu.py @ ee938820
History | View | Annotate | Download (21.3 kB)
1 |
#
|
---|---|
2 |
#
|
3 |
|
4 |
# Copyright (C) 2006, 2007, 2011 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 |
"""Module implementing the logic behind the cluster operations
|
23 |
|
24 |
This module implements the logic for doing operations in the cluster. There
|
25 |
are two kinds of classes defined:
|
26 |
- logical units, which know how to deal with their specific opcode only
|
27 |
- the processor, which dispatches the opcodes to their logical units
|
28 |
|
29 |
"""
|
30 |
|
31 |
import logging |
32 |
import random |
33 |
import time |
34 |
|
35 |
from ganeti import opcodes |
36 |
from ganeti import constants |
37 |
from ganeti import errors |
38 |
from ganeti import cmdlib |
39 |
from ganeti import locking |
40 |
from ganeti import utils |
41 |
from ganeti import compat |
42 |
|
43 |
|
44 |
_OP_PREFIX = "Op"
|
45 |
_LU_PREFIX = "LU"
|
46 |
|
47 |
|
48 |
class LockAcquireTimeout(Exception): |
49 |
"""Exception to report timeouts on acquiring locks.
|
50 |
|
51 |
"""
|
52 |
|
53 |
|
54 |
def _CalculateLockAttemptTimeouts(): |
55 |
"""Calculate timeouts for lock attempts.
|
56 |
|
57 |
"""
|
58 |
result = [constants.LOCK_ATTEMPTS_MINWAIT] |
59 |
running_sum = result[0]
|
60 |
|
61 |
# Wait for a total of at least LOCK_ATTEMPTS_TIMEOUT before doing a
|
62 |
# blocking acquire
|
63 |
while running_sum < constants.LOCK_ATTEMPTS_TIMEOUT:
|
64 |
timeout = (result[-1] * 1.05) ** 1.25 |
65 |
|
66 |
# Cap max timeout. This gives other jobs a chance to run even if
|
67 |
# we're still trying to get our locks, before finally moving to a
|
68 |
# blocking acquire.
|
69 |
timeout = min(timeout, constants.LOCK_ATTEMPTS_MAXWAIT)
|
70 |
# And also cap the lower boundary for safety
|
71 |
timeout = max(timeout, constants.LOCK_ATTEMPTS_MINWAIT)
|
72 |
|
73 |
result.append(timeout) |
74 |
running_sum += timeout |
75 |
|
76 |
return result
|
77 |
|
78 |
|
79 |
class LockAttemptTimeoutStrategy(object): |
80 |
"""Class with lock acquire timeout strategy.
|
81 |
|
82 |
"""
|
83 |
__slots__ = [ |
84 |
"_timeouts",
|
85 |
"_random_fn",
|
86 |
"_time_fn",
|
87 |
] |
88 |
|
89 |
_TIMEOUT_PER_ATTEMPT = _CalculateLockAttemptTimeouts() |
90 |
|
91 |
def __init__(self, _time_fn=time.time, _random_fn=random.random): |
92 |
"""Initializes this class.
|
93 |
|
94 |
@param _time_fn: Time function for unittests
|
95 |
@param _random_fn: Random number generator for unittests
|
96 |
|
97 |
"""
|
98 |
object.__init__(self) |
99 |
|
100 |
self._timeouts = iter(self._TIMEOUT_PER_ATTEMPT) |
101 |
self._time_fn = _time_fn
|
102 |
self._random_fn = _random_fn
|
103 |
|
104 |
def NextAttempt(self): |
105 |
"""Returns the timeout for the next attempt.
|
106 |
|
107 |
"""
|
108 |
try:
|
109 |
timeout = self._timeouts.next()
|
110 |
except StopIteration: |
111 |
# No more timeouts, do blocking acquire
|
112 |
timeout = None
|
113 |
|
114 |
if timeout is not None: |
115 |
# Add a small variation (-/+ 5%) to timeout. This helps in situations
|
116 |
# where two or more jobs are fighting for the same lock(s).
|
117 |
variation_range = timeout * 0.1
|
118 |
timeout += ((self._random_fn() * variation_range) -
|
119 |
(variation_range * 0.5))
|
120 |
|
121 |
return timeout
|
122 |
|
123 |
|
124 |
class OpExecCbBase: # pylint: disable=W0232 |
125 |
"""Base class for OpCode execution callbacks.
|
126 |
|
127 |
"""
|
128 |
def NotifyStart(self): |
129 |
"""Called when we are about to execute the LU.
|
130 |
|
131 |
This function is called when we're about to start the lu's Exec() method,
|
132 |
that is, after we have acquired all locks.
|
133 |
|
134 |
"""
|
135 |
|
136 |
def Feedback(self, *args): |
137 |
"""Sends feedback from the LU code to the end-user.
|
138 |
|
139 |
"""
|
140 |
|
141 |
def CheckCancel(self): |
142 |
"""Check whether job has been cancelled.
|
143 |
|
144 |
"""
|
145 |
|
146 |
def SubmitManyJobs(self, jobs): |
147 |
"""Submits jobs for processing.
|
148 |
|
149 |
See L{jqueue.JobQueue.SubmitManyJobs}.
|
150 |
|
151 |
"""
|
152 |
raise NotImplementedError |
153 |
|
154 |
|
155 |
def _LUNameForOpName(opname): |
156 |
"""Computes the LU name for a given OpCode name.
|
157 |
|
158 |
"""
|
159 |
assert opname.startswith(_OP_PREFIX), \
|
160 |
"Invalid OpCode name, doesn't start with %s: %s" % (_OP_PREFIX, opname)
|
161 |
|
162 |
return _LU_PREFIX + opname[len(_OP_PREFIX):] |
163 |
|
164 |
|
165 |
def _ComputeDispatchTable(): |
166 |
"""Computes the opcode-to-lu dispatch table.
|
167 |
|
168 |
"""
|
169 |
return dict((op, getattr(cmdlib, _LUNameForOpName(op.__name__))) |
170 |
for op in opcodes.OP_MAPPING.values() |
171 |
if op.WITH_LU)
|
172 |
|
173 |
|
174 |
def _RpcResultsToHooksResults(rpc_results): |
175 |
"""Function to convert RPC results to the format expected by HooksMaster.
|
176 |
|
177 |
@type rpc_results: dict(node: L{rpc.RpcResult})
|
178 |
@param rpc_results: RPC results
|
179 |
@rtype: dict(node: (fail_msg, offline, hooks_results))
|
180 |
@return: RPC results unpacked according to the format expected by
|
181 |
L({mcpu.HooksMaster}
|
182 |
|
183 |
"""
|
184 |
return dict((node, (rpc_res.fail_msg, rpc_res.offline, rpc_res.payload)) |
185 |
for (node, rpc_res) in rpc_results.items()) |
186 |
|
187 |
|
188 |
class Processor(object): |
189 |
"""Object which runs OpCodes"""
|
190 |
DISPATCH_TABLE = _ComputeDispatchTable() |
191 |
|
192 |
def __init__(self, context, ec_id): |
193 |
"""Constructor for Processor
|
194 |
|
195 |
@type context: GanetiContext
|
196 |
@param context: global Ganeti context
|
197 |
@type ec_id: string
|
198 |
@param ec_id: execution context identifier
|
199 |
|
200 |
"""
|
201 |
self.context = context
|
202 |
self._ec_id = ec_id
|
203 |
self._cbs = None |
204 |
self.rpc = context.rpc
|
205 |
self.hmclass = HooksMaster
|
206 |
|
207 |
def _AcquireLocks(self, level, names, shared, timeout, priority): |
208 |
"""Acquires locks via the Ganeti lock manager.
|
209 |
|
210 |
@type level: int
|
211 |
@param level: Lock level
|
212 |
@type names: list or string
|
213 |
@param names: Lock names
|
214 |
@type shared: bool
|
215 |
@param shared: Whether the locks should be acquired in shared mode
|
216 |
@type timeout: None or float
|
217 |
@param timeout: Timeout for acquiring the locks
|
218 |
@raise LockAcquireTimeout: In case locks couldn't be acquired in specified
|
219 |
amount of time
|
220 |
|
221 |
"""
|
222 |
if self._cbs: |
223 |
self._cbs.CheckCancel()
|
224 |
|
225 |
acquired = self.context.glm.acquire(level, names, shared=shared,
|
226 |
timeout=timeout, priority=priority) |
227 |
|
228 |
if acquired is None: |
229 |
raise LockAcquireTimeout()
|
230 |
|
231 |
return acquired
|
232 |
|
233 |
def _ProcessResult(self, result): |
234 |
"""Examines opcode result.
|
235 |
|
236 |
If necessary, additional processing on the result is done.
|
237 |
|
238 |
"""
|
239 |
if isinstance(result, cmdlib.ResultWithJobs): |
240 |
# Submit jobs
|
241 |
job_submission = self._cbs.SubmitManyJobs(result.jobs)
|
242 |
|
243 |
# Build dictionary
|
244 |
result = result.other |
245 |
|
246 |
assert constants.JOB_IDS_KEY not in result, \ |
247 |
"Key '%s' found in additional return values" % constants.JOB_IDS_KEY
|
248 |
|
249 |
result[constants.JOB_IDS_KEY] = job_submission |
250 |
|
251 |
return result
|
252 |
|
253 |
def _ExecLU(self, lu): |
254 |
"""Logical Unit execution sequence.
|
255 |
|
256 |
"""
|
257 |
write_count = self.context.cfg.write_count
|
258 |
lu.CheckPrereq() |
259 |
|
260 |
hm = self.BuildHooksManager(lu)
|
261 |
h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE) |
262 |
lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results, |
263 |
self.Log, None) |
264 |
|
265 |
if getattr(lu.op, "dry_run", False): |
266 |
# in this mode, no post-hooks are run, and the config is not
|
267 |
# written (as it might have been modified by another LU, and we
|
268 |
# shouldn't do writeout on behalf of other threads
|
269 |
self.LogInfo("dry-run mode requested, not actually executing" |
270 |
" the operation")
|
271 |
return lu.dry_run_result
|
272 |
|
273 |
try:
|
274 |
result = self._ProcessResult(lu.Exec(self.Log)) |
275 |
h_results = hm.RunPhase(constants.HOOKS_PHASE_POST) |
276 |
result = lu.HooksCallBack(constants.HOOKS_PHASE_POST, h_results, |
277 |
self.Log, result)
|
278 |
finally:
|
279 |
# FIXME: This needs locks if not lu_class.REQ_BGL
|
280 |
if write_count != self.context.cfg.write_count: |
281 |
hm.RunConfigUpdate() |
282 |
|
283 |
return result
|
284 |
|
285 |
def BuildHooksManager(self, lu): |
286 |
return self.hmclass.BuildFromLu(lu.rpc.call_hooks_runner, lu) |
287 |
|
288 |
def _LockAndExecLU(self, lu, level, calc_timeout, priority): |
289 |
"""Execute a Logical Unit, with the needed locks.
|
290 |
|
291 |
This is a recursive function that starts locking the given level, and
|
292 |
proceeds up, till there are no more locks to acquire. Then it executes the
|
293 |
given LU and its opcodes.
|
294 |
|
295 |
"""
|
296 |
adding_locks = level in lu.add_locks
|
297 |
acquiring_locks = level in lu.needed_locks
|
298 |
if level not in locking.LEVELS: |
299 |
if self._cbs: |
300 |
self._cbs.NotifyStart()
|
301 |
|
302 |
result = self._ExecLU(lu)
|
303 |
|
304 |
elif adding_locks and acquiring_locks: |
305 |
# We could both acquire and add locks at the same level, but for now we
|
306 |
# don't need this, so we'll avoid the complicated code needed.
|
307 |
raise NotImplementedError("Can't declare locks to acquire when adding" |
308 |
" others")
|
309 |
|
310 |
elif adding_locks or acquiring_locks: |
311 |
lu.DeclareLocks(level) |
312 |
share = lu.share_locks[level] |
313 |
|
314 |
try:
|
315 |
assert adding_locks ^ acquiring_locks, \
|
316 |
"Locks must be either added or acquired"
|
317 |
|
318 |
if acquiring_locks:
|
319 |
# Acquiring locks
|
320 |
needed_locks = lu.needed_locks[level] |
321 |
|
322 |
self._AcquireLocks(level, needed_locks, share,
|
323 |
calc_timeout(), priority) |
324 |
else:
|
325 |
# Adding locks
|
326 |
add_locks = lu.add_locks[level] |
327 |
lu.remove_locks[level] = add_locks |
328 |
|
329 |
try:
|
330 |
self.context.glm.add(level, add_locks, acquired=1, shared=share) |
331 |
except errors.LockError:
|
332 |
raise errors.OpPrereqError(
|
333 |
"Couldn't add locks (%s), probably because of a race condition"
|
334 |
" with another job, who added them first" % add_locks,
|
335 |
errors.ECODE_FAULT) |
336 |
|
337 |
try:
|
338 |
result = self._LockAndExecLU(lu, level + 1, calc_timeout, priority) |
339 |
finally:
|
340 |
if level in lu.remove_locks: |
341 |
self.context.glm.remove(level, lu.remove_locks[level])
|
342 |
finally:
|
343 |
if self.context.glm.is_owned(level): |
344 |
self.context.glm.release(level)
|
345 |
|
346 |
else:
|
347 |
result = self._LockAndExecLU(lu, level + 1, calc_timeout, priority) |
348 |
|
349 |
return result
|
350 |
|
351 |
def ExecOpCode(self, op, cbs, timeout=None, priority=None): |
352 |
"""Execute an opcode.
|
353 |
|
354 |
@type op: an OpCode instance
|
355 |
@param op: the opcode to be executed
|
356 |
@type cbs: L{OpExecCbBase}
|
357 |
@param cbs: Runtime callbacks
|
358 |
@type timeout: float or None
|
359 |
@param timeout: Maximum time to acquire all locks, None for no timeout
|
360 |
@type priority: number or None
|
361 |
@param priority: Priority for acquiring lock(s)
|
362 |
@raise LockAcquireTimeout: In case locks couldn't be acquired in specified
|
363 |
amount of time
|
364 |
|
365 |
"""
|
366 |
if not isinstance(op, opcodes.OpCode): |
367 |
raise errors.ProgrammerError("Non-opcode instance passed" |
368 |
" to ExecOpcode (%s)" % type(op)) |
369 |
|
370 |
lu_class = self.DISPATCH_TABLE.get(op.__class__, None) |
371 |
if lu_class is None: |
372 |
raise errors.OpCodeUnknown("Unknown opcode") |
373 |
|
374 |
if timeout is None: |
375 |
calc_timeout = lambda: None |
376 |
else:
|
377 |
calc_timeout = utils.RunningTimeout(timeout, False).Remaining
|
378 |
|
379 |
self._cbs = cbs
|
380 |
try:
|
381 |
# Acquire the Big Ganeti Lock exclusively if this LU requires it,
|
382 |
# and in a shared fashion otherwise (to prevent concurrent run with
|
383 |
# an exclusive LU.
|
384 |
self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL,
|
385 |
not lu_class.REQ_BGL, calc_timeout(),
|
386 |
priority) |
387 |
try:
|
388 |
lu = lu_class(self, op, self.context, self.rpc) |
389 |
lu.ExpandNames() |
390 |
assert lu.needed_locks is not None, "needed_locks not set by LU" |
391 |
|
392 |
try:
|
393 |
result = self._LockAndExecLU(lu, locking.LEVEL_INSTANCE, calc_timeout,
|
394 |
priority) |
395 |
finally:
|
396 |
if self._ec_id: |
397 |
self.context.cfg.DropECReservations(self._ec_id) |
398 |
finally:
|
399 |
self.context.glm.release(locking.LEVEL_CLUSTER)
|
400 |
finally:
|
401 |
self._cbs = None |
402 |
|
403 |
resultcheck_fn = op.OP_RESULT |
404 |
if not (resultcheck_fn is None or resultcheck_fn(result)): |
405 |
logging.error("Expected opcode result matching %s, got %s",
|
406 |
resultcheck_fn, result) |
407 |
raise errors.OpResultError("Opcode result does not match %s, got %s" % |
408 |
(resultcheck_fn, result[:80]))
|
409 |
|
410 |
return result
|
411 |
|
412 |
def Log(self, *args): |
413 |
"""Forward call to feedback callback function.
|
414 |
|
415 |
"""
|
416 |
if self._cbs: |
417 |
self._cbs.Feedback(*args)
|
418 |
|
419 |
def LogStep(self, current, total, message): |
420 |
"""Log a change in LU execution progress.
|
421 |
|
422 |
"""
|
423 |
logging.debug("Step %d/%d %s", current, total, message)
|
424 |
self.Log("STEP %d/%d %s" % (current, total, message)) |
425 |
|
426 |
def LogWarning(self, message, *args, **kwargs): |
427 |
"""Log a warning to the logs and the user.
|
428 |
|
429 |
The optional keyword argument is 'hint' and can be used to show a
|
430 |
hint to the user (presumably related to the warning). If the
|
431 |
message is empty, it will not be printed at all, allowing one to
|
432 |
show only a hint.
|
433 |
|
434 |
"""
|
435 |
assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \ |
436 |
"Invalid keyword arguments for LogWarning (%s)" % str(kwargs) |
437 |
if args:
|
438 |
message = message % tuple(args)
|
439 |
if message:
|
440 |
logging.warning(message) |
441 |
self.Log(" - WARNING: %s" % message) |
442 |
if "hint" in kwargs: |
443 |
self.Log(" Hint: %s" % kwargs["hint"]) |
444 |
|
445 |
def LogInfo(self, message, *args): |
446 |
"""Log an informational message to the logs and the user.
|
447 |
|
448 |
"""
|
449 |
if args:
|
450 |
message = message % tuple(args)
|
451 |
logging.info(message) |
452 |
self.Log(" - INFO: %s" % message) |
453 |
|
454 |
def GetECId(self): |
455 |
"""Returns the current execution context ID.
|
456 |
|
457 |
"""
|
458 |
if not self._ec_id: |
459 |
raise errors.ProgrammerError("Tried to use execution context id when" |
460 |
" not set")
|
461 |
return self._ec_id |
462 |
|
463 |
|
464 |
class HooksMaster(object): |
465 |
def __init__(self, opcode, hooks_path, nodes, hooks_execution_fn, |
466 |
hooks_results_adapt_fn, build_env_fn, log_fn, htype=None, cluster_name=None, |
467 |
master_name=None):
|
468 |
"""Base class for hooks masters.
|
469 |
|
470 |
This class invokes the execution of hooks according to the behaviour
|
471 |
specified by its parameters.
|
472 |
|
473 |
@type opcode: string
|
474 |
@param opcode: opcode of the operation to which the hooks are tied
|
475 |
@type hooks_path: string
|
476 |
@param hooks_path: prefix of the hooks directories
|
477 |
@type nodes: 2-tuple of lists
|
478 |
@param nodes: 2-tuple of lists containing nodes on which pre-hooks must be
|
479 |
run and nodes on which post-hooks must be run
|
480 |
@type hooks_execution_fn: function that accepts the following parameters:
|
481 |
(node_list, hooks_path, phase, environment)
|
482 |
@param hooks_execution_fn: function that will execute the hooks; can be
|
483 |
None, indicating that no conversion is necessary.
|
484 |
@type hooks_results_adapt_fn: function
|
485 |
@param hooks_results_adapt_fn: function that will adapt the return value of
|
486 |
hooks_execution_fn to the format expected by RunPhase
|
487 |
@type build_env_fn: function that returns a dictionary having strings as
|
488 |
keys
|
489 |
@param build_env_fn: function that builds the environment for the hooks
|
490 |
@type log_fn: function that accepts a string
|
491 |
@param log_fn: logging function
|
492 |
@type htype: string or None
|
493 |
@param htype: None or one of L{constants.HTYPE_CLUSTER},
|
494 |
L{constants.HTYPE_NODE}, L{constants.HTYPE_INSTANCE}
|
495 |
@type cluster_name: string
|
496 |
@param cluster_name: name of the cluster
|
497 |
@type master_name: string
|
498 |
@param master_name: name of the master
|
499 |
|
500 |
"""
|
501 |
self.opcode = opcode
|
502 |
self.hooks_path = hooks_path
|
503 |
self.hooks_execution_fn = hooks_execution_fn
|
504 |
self.hooks_results_adapt_fn = hooks_results_adapt_fn
|
505 |
self.build_env_fn = build_env_fn
|
506 |
self.log_fn = log_fn
|
507 |
self.htype = htype
|
508 |
self.cluster_name = cluster_name
|
509 |
self.master_name = master_name
|
510 |
|
511 |
self.pre_env = self._BuildEnv(constants.HOOKS_PHASE_PRE) |
512 |
(self.pre_nodes, self.post_nodes) = nodes |
513 |
|
514 |
def _BuildEnv(self, phase): |
515 |
"""Compute the environment and the target nodes.
|
516 |
|
517 |
Based on the opcode and the current node list, this builds the
|
518 |
environment for the hooks and the target node list for the run.
|
519 |
|
520 |
"""
|
521 |
if phase == constants.HOOKS_PHASE_PRE:
|
522 |
prefix = "GANETI_"
|
523 |
elif phase == constants.HOOKS_PHASE_POST:
|
524 |
prefix = "GANETI_POST_"
|
525 |
else:
|
526 |
raise AssertionError("Unknown phase '%s'" % phase) |
527 |
|
528 |
env = {} |
529 |
|
530 |
if self.hooks_path is not None: |
531 |
phase_env = self.build_env_fn()
|
532 |
if phase_env:
|
533 |
assert not compat.any(key.upper().startswith(prefix) |
534 |
for key in phase_env) |
535 |
env.update(("%s%s" % (prefix, key), value)
|
536 |
for (key, value) in phase_env.items()) |
537 |
|
538 |
if phase == constants.HOOKS_PHASE_PRE:
|
539 |
assert compat.all((key.startswith("GANETI_") and |
540 |
not key.startswith("GANETI_POST_")) |
541 |
for key in env) |
542 |
|
543 |
elif phase == constants.HOOKS_PHASE_POST:
|
544 |
assert compat.all(key.startswith("GANETI_POST_") for key in env) |
545 |
assert isinstance(self.pre_env, dict) |
546 |
|
547 |
# Merge with pre-phase environment
|
548 |
assert not compat.any(key.startswith("GANETI_POST_") |
549 |
for key in self.pre_env) |
550 |
env.update(self.pre_env)
|
551 |
else:
|
552 |
raise AssertionError("Unknown phase '%s'" % phase) |
553 |
|
554 |
return env
|
555 |
|
556 |
def _RunWrapper(self, node_list, hpath, phase, phase_env): |
557 |
"""Simple wrapper over self.callfn.
|
558 |
|
559 |
This method fixes the environment before executing the hooks.
|
560 |
|
561 |
"""
|
562 |
env = { |
563 |
"PATH": constants.HOOKS_PATH,
|
564 |
"GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
|
565 |
"GANETI_OP_CODE": self.opcode, |
566 |
"GANETI_DATA_DIR": constants.DATA_DIR,
|
567 |
"GANETI_HOOKS_PHASE": phase,
|
568 |
"GANETI_HOOKS_PATH": hpath,
|
569 |
} |
570 |
|
571 |
if self.htype: |
572 |
env["GANETI_OBJECT_TYPE"] = self.htype |
573 |
|
574 |
if self.cluster_name is not None: |
575 |
env["GANETI_CLUSTER"] = self.cluster_name |
576 |
|
577 |
if self.master_name is not None: |
578 |
env["GANETI_MASTER"] = self.master_name |
579 |
|
580 |
if phase_env:
|
581 |
env = utils.algo.JoinDisjointDicts(env, phase_env) |
582 |
|
583 |
# Convert everything to strings
|
584 |
env = dict([(str(key), str(val)) for key, val in env.iteritems()]) |
585 |
|
586 |
assert compat.all(key == "PATH" or key.startswith("GANETI_") |
587 |
for key in env) |
588 |
|
589 |
return self.hooks_execution_fn(node_list, hpath, phase, env) |
590 |
|
591 |
def RunPhase(self, phase, nodes=None): |
592 |
"""Run all the scripts for a phase.
|
593 |
|
594 |
This is the main function of the HookMaster.
|
595 |
It executes self.hooks_execution_fn, and after running
|
596 |
self.hooks_results_adapt_fn on its results it expects them to be in the form
|
597 |
{node_name: (fail_msg, [(script, result, output), ...]}).
|
598 |
|
599 |
@param phase: one of L{constants.HOOKS_PHASE_POST} or
|
600 |
L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
|
601 |
@param nodes: overrides the predefined list of nodes for the given phase
|
602 |
@return: the processed results of the hooks multi-node rpc call
|
603 |
@raise errors.HooksFailure: on communication failure to the nodes
|
604 |
@raise errors.HooksAbort: on failure of one of the hooks
|
605 |
|
606 |
"""
|
607 |
if phase == constants.HOOKS_PHASE_PRE:
|
608 |
if nodes is None: |
609 |
nodes = self.pre_nodes
|
610 |
env = self.pre_env
|
611 |
elif phase == constants.HOOKS_PHASE_POST:
|
612 |
if nodes is None: |
613 |
nodes = self.post_nodes
|
614 |
env = self._BuildEnv(phase)
|
615 |
else:
|
616 |
raise AssertionError("Unknown phase '%s'" % phase) |
617 |
|
618 |
if not nodes: |
619 |
# empty node list, we should not attempt to run this as either
|
620 |
# we're in the cluster init phase and the rpc client part can't
|
621 |
# even attempt to run, or this LU doesn't do hooks at all
|
622 |
return
|
623 |
|
624 |
results = self._RunWrapper(nodes, self.hooks_path, phase, env) |
625 |
if not results: |
626 |
msg = "Communication Failure"
|
627 |
if phase == constants.HOOKS_PHASE_PRE:
|
628 |
raise errors.HooksFailure(msg)
|
629 |
else:
|
630 |
self.log_fn(msg)
|
631 |
return results
|
632 |
|
633 |
converted_res = results |
634 |
if self.hooks_results_adapt_fn: |
635 |
converted_res = self.hooks_results_adapt_fn(results)
|
636 |
|
637 |
errs = [] |
638 |
for node_name, (fail_msg, offline, hooks_results) in converted_res.items(): |
639 |
if offline:
|
640 |
continue
|
641 |
|
642 |
if fail_msg:
|
643 |
self.log_fn("Communication failure to node %s: %s", node_name, fail_msg) |
644 |
continue
|
645 |
|
646 |
for script, hkr, output in hooks_results: |
647 |
if hkr == constants.HKR_FAIL:
|
648 |
if phase == constants.HOOKS_PHASE_PRE:
|
649 |
errs.append((node_name, script, output)) |
650 |
else:
|
651 |
if not output: |
652 |
output = "(no output)"
|
653 |
self.log_fn("On %s script %s failed, output: %s" % |
654 |
(node_name, script, output)) |
655 |
|
656 |
if errs and phase == constants.HOOKS_PHASE_PRE: |
657 |
raise errors.HooksAbort(errs)
|
658 |
|
659 |
return results
|
660 |
|
661 |
def RunConfigUpdate(self): |
662 |
"""Run the special configuration update hook
|
663 |
|
664 |
This is a special hook that runs only on the master after each
|
665 |
top-level LI if the configuration has been updated.
|
666 |
|
667 |
"""
|
668 |
phase = constants.HOOKS_PHASE_POST |
669 |
hpath = constants.HOOKS_NAME_CFGUPDATE |
670 |
nodes = [self.master_name]
|
671 |
self._RunWrapper(nodes, hpath, phase, self.pre_env) |
672 |
|
673 |
@staticmethod
|
674 |
def BuildFromLu(hooks_execution_fn, lu): |
675 |
if lu.HPATH is None: |
676 |
nodes = (None, None) |
677 |
else:
|
678 |
nodes = map(frozenset, lu.BuildHooksNodes()) |
679 |
|
680 |
master_name = cluster_name = None
|
681 |
if lu.cfg:
|
682 |
master_name = lu.cfg.GetMasterNode() |
683 |
cluster_name = lu.cfg.GetClusterName() |
684 |
|
685 |
return HooksMaster(lu.op.OP_ID, lu.HPATH, nodes, hooks_execution_fn,
|
686 |
_RpcResultsToHooksResults, lu.BuildHooksEnv, |
687 |
lu.LogWarning, lu.HTYPE, cluster_name, master_name) |