root / lib / mcpu.py @ 407339d0
History | View | Annotate | Download (18.9 kB)
1 |
#
|
---|---|
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 |
"""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 rpc |
39 |
from ganeti import cmdlib |
40 |
from ganeti import locking |
41 |
from ganeti import utils |
42 |
|
43 |
|
44 |
class _LockAcquireTimeout(Exception): |
45 |
"""Internal exception to report timeouts on acquiring locks.
|
46 |
|
47 |
"""
|
48 |
|
49 |
|
50 |
class _LockTimeoutStrategy(object): |
51 |
"""Class with lock acquire timeout strategy.
|
52 |
|
53 |
"""
|
54 |
__slots__ = [ |
55 |
"_attempts",
|
56 |
"_random_fn",
|
57 |
"_start_time",
|
58 |
] |
59 |
|
60 |
_MAX_ATTEMPTS = 10
|
61 |
"""How many retries before going into blocking mode"""
|
62 |
|
63 |
_ATTEMPT_FACTOR = 1.75
|
64 |
"""Factor between attempts"""
|
65 |
|
66 |
def __init__(self, _random_fn=None): |
67 |
"""Initializes this class.
|
68 |
|
69 |
@param _random_fn: Random number generator for unittests
|
70 |
|
71 |
"""
|
72 |
object.__init__(self) |
73 |
|
74 |
self._start_time = None |
75 |
self._attempts = 0 |
76 |
|
77 |
if _random_fn is None: |
78 |
self._random_fn = random.random
|
79 |
else:
|
80 |
self._random_fn = _random_fn
|
81 |
|
82 |
def NextAttempt(self): |
83 |
"""Advances to the next attempt.
|
84 |
|
85 |
"""
|
86 |
assert self._attempts >= 0 |
87 |
self._attempts += 1 |
88 |
|
89 |
def CalcRemainingTimeout(self): |
90 |
"""Returns the remaining timeout.
|
91 |
|
92 |
"""
|
93 |
assert self._attempts >= 0 |
94 |
|
95 |
if self._attempts == self._MAX_ATTEMPTS: |
96 |
# Only blocking acquires after 10 retries
|
97 |
return None |
98 |
|
99 |
if self._attempts > self._MAX_ATTEMPTS: |
100 |
raise RuntimeError("Blocking acquire ran into timeout") |
101 |
|
102 |
# Get start time on first calculation
|
103 |
if self._start_time is None: |
104 |
self._start_time = time.time()
|
105 |
|
106 |
# Calculate remaining time for this attempt
|
107 |
timeout = (self._start_time + (self._ATTEMPT_FACTOR ** self._attempts) - |
108 |
time.time()) |
109 |
|
110 |
if timeout > 10.0: |
111 |
# Cap timeout at 10 seconds. This gives other jobs a chance to run
|
112 |
# even if we're still trying to get our locks, before finally moving
|
113 |
# to a blocking acquire.
|
114 |
timeout = 10.0
|
115 |
|
116 |
elif timeout < 0.1: |
117 |
# Lower boundary
|
118 |
timeout = 0.1
|
119 |
|
120 |
# Add a small variation (-/+ 5%) to timeouts. This helps in situations
|
121 |
# where two or more jobs are fighting for the same lock(s).
|
122 |
variation_range = timeout * 0.1
|
123 |
timeout += (self._random_fn() * variation_range) - (variation_range * 0.5) |
124 |
|
125 |
assert timeout >= 0.0, "Timeout must be positive" |
126 |
|
127 |
return timeout
|
128 |
|
129 |
|
130 |
class OpExecCbBase: |
131 |
"""Base class for OpCode execution callbacks.
|
132 |
|
133 |
"""
|
134 |
def NotifyStart(self): |
135 |
"""Called when we are about to execute the LU.
|
136 |
|
137 |
This function is called when we're about to start the lu's Exec() method,
|
138 |
that is, after we have acquired all locks.
|
139 |
|
140 |
"""
|
141 |
|
142 |
def Feedback(self, *args): |
143 |
"""Sends feedback from the LU code to the end-user.
|
144 |
|
145 |
"""
|
146 |
|
147 |
def ReportLocks(self, msg): |
148 |
"""Report lock operations.
|
149 |
|
150 |
"""
|
151 |
|
152 |
|
153 |
class Processor(object): |
154 |
"""Object which runs OpCodes"""
|
155 |
DISPATCH_TABLE = { |
156 |
# Cluster
|
157 |
opcodes.OpPostInitCluster: cmdlib.LUPostInitCluster, |
158 |
opcodes.OpDestroyCluster: cmdlib.LUDestroyCluster, |
159 |
opcodes.OpQueryClusterInfo: cmdlib.LUQueryClusterInfo, |
160 |
opcodes.OpVerifyCluster: cmdlib.LUVerifyCluster, |
161 |
opcodes.OpQueryConfigValues: cmdlib.LUQueryConfigValues, |
162 |
opcodes.OpRenameCluster: cmdlib.LURenameCluster, |
163 |
opcodes.OpVerifyDisks: cmdlib.LUVerifyDisks, |
164 |
opcodes.OpSetClusterParams: cmdlib.LUSetClusterParams, |
165 |
opcodes.OpRedistributeConfig: cmdlib.LURedistributeConfig, |
166 |
opcodes.OpRepairDiskSizes: cmdlib.LURepairDiskSizes, |
167 |
# node lu
|
168 |
opcodes.OpAddNode: cmdlib.LUAddNode, |
169 |
opcodes.OpQueryNodes: cmdlib.LUQueryNodes, |
170 |
opcodes.OpQueryNodeVolumes: cmdlib.LUQueryNodeVolumes, |
171 |
opcodes.OpQueryNodeStorage: cmdlib.LUQueryNodeStorage, |
172 |
opcodes.OpModifyNodeStorage: cmdlib.LUModifyNodeStorage, |
173 |
opcodes.OpRepairNodeStorage: cmdlib.LURepairNodeStorage, |
174 |
opcodes.OpRemoveNode: cmdlib.LURemoveNode, |
175 |
opcodes.OpSetNodeParams: cmdlib.LUSetNodeParams, |
176 |
opcodes.OpPowercycleNode: cmdlib.LUPowercycleNode, |
177 |
opcodes.OpEvacuateNode: cmdlib.LUEvacuateNode, |
178 |
opcodes.OpMigrateNode: cmdlib.LUMigrateNode, |
179 |
# instance lu
|
180 |
opcodes.OpCreateInstance: cmdlib.LUCreateInstance, |
181 |
opcodes.OpReinstallInstance: cmdlib.LUReinstallInstance, |
182 |
opcodes.OpRemoveInstance: cmdlib.LURemoveInstance, |
183 |
opcodes.OpRenameInstance: cmdlib.LURenameInstance, |
184 |
opcodes.OpActivateInstanceDisks: cmdlib.LUActivateInstanceDisks, |
185 |
opcodes.OpShutdownInstance: cmdlib.LUShutdownInstance, |
186 |
opcodes.OpStartupInstance: cmdlib.LUStartupInstance, |
187 |
opcodes.OpRebootInstance: cmdlib.LURebootInstance, |
188 |
opcodes.OpDeactivateInstanceDisks: cmdlib.LUDeactivateInstanceDisks, |
189 |
opcodes.OpReplaceDisks: cmdlib.LUReplaceDisks, |
190 |
opcodes.OpRecreateInstanceDisks: cmdlib.LURecreateInstanceDisks, |
191 |
opcodes.OpFailoverInstance: cmdlib.LUFailoverInstance, |
192 |
opcodes.OpMigrateInstance: cmdlib.LUMigrateInstance, |
193 |
opcodes.OpMoveInstance: cmdlib.LUMoveInstance, |
194 |
opcodes.OpConnectConsole: cmdlib.LUConnectConsole, |
195 |
opcodes.OpQueryInstances: cmdlib.LUQueryInstances, |
196 |
opcodes.OpQueryInstanceData: cmdlib.LUQueryInstanceData, |
197 |
opcodes.OpSetInstanceParams: cmdlib.LUSetInstanceParams, |
198 |
opcodes.OpGrowDisk: cmdlib.LUGrowDisk, |
199 |
# os lu
|
200 |
opcodes.OpDiagnoseOS: cmdlib.LUDiagnoseOS, |
201 |
# exports lu
|
202 |
opcodes.OpQueryExports: cmdlib.LUQueryExports, |
203 |
opcodes.OpExportInstance: cmdlib.LUExportInstance, |
204 |
opcodes.OpRemoveExport: cmdlib.LURemoveExport, |
205 |
# tags lu
|
206 |
opcodes.OpGetTags: cmdlib.LUGetTags, |
207 |
opcodes.OpSearchTags: cmdlib.LUSearchTags, |
208 |
opcodes.OpAddTags: cmdlib.LUAddTags, |
209 |
opcodes.OpDelTags: cmdlib.LUDelTags, |
210 |
# test lu
|
211 |
opcodes.OpTestDelay: cmdlib.LUTestDelay, |
212 |
opcodes.OpTestAllocator: cmdlib.LUTestAllocator, |
213 |
} |
214 |
|
215 |
def __init__(self, context): |
216 |
"""Constructor for Processor
|
217 |
|
218 |
"""
|
219 |
self.context = context
|
220 |
self._cbs = None |
221 |
self.rpc = rpc.RpcRunner(context.cfg)
|
222 |
self.hmclass = HooksMaster
|
223 |
|
224 |
def _ReportLocks(self, level, names, shared, acquired): |
225 |
"""Reports lock operations.
|
226 |
|
227 |
@type level: int
|
228 |
@param level: Lock level
|
229 |
@type names: list or string
|
230 |
@param names: Lock names
|
231 |
@type shared: bool
|
232 |
@param shared: Whether the lock should be acquired in shared mode
|
233 |
@type acquired: bool
|
234 |
@param acquired: Whether the lock has already been acquired
|
235 |
|
236 |
"""
|
237 |
parts = [] |
238 |
|
239 |
# Build message
|
240 |
if acquired:
|
241 |
parts.append("acquired")
|
242 |
else:
|
243 |
parts.append("waiting")
|
244 |
|
245 |
parts.append(locking.LEVEL_NAMES[level]) |
246 |
|
247 |
if names == locking.ALL_SET:
|
248 |
parts.append("ALL")
|
249 |
elif isinstance(names, basestring): |
250 |
parts.append(names) |
251 |
else:
|
252 |
parts.append(",".join(names))
|
253 |
|
254 |
if shared:
|
255 |
parts.append("shared")
|
256 |
else:
|
257 |
parts.append("exclusive")
|
258 |
|
259 |
msg = "/".join(parts)
|
260 |
|
261 |
logging.debug("LU locks %s", msg)
|
262 |
|
263 |
if self._cbs: |
264 |
self._cbs.ReportLocks(msg)
|
265 |
|
266 |
def _ExecLU(self, lu): |
267 |
"""Logical Unit execution sequence.
|
268 |
|
269 |
"""
|
270 |
write_count = self.context.cfg.write_count
|
271 |
lu.CheckPrereq() |
272 |
hm = HooksMaster(self.rpc.call_hooks_runner, lu)
|
273 |
h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE) |
274 |
lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results, |
275 |
self._Feedback, None) |
276 |
|
277 |
if getattr(lu.op, "dry_run", False): |
278 |
# in this mode, no post-hooks are run, and the config is not
|
279 |
# written (as it might have been modified by another LU, and we
|
280 |
# shouldn't do writeout on behalf of other threads
|
281 |
self.LogInfo("dry-run mode requested, not actually executing" |
282 |
" the operation")
|
283 |
return lu.dry_run_result
|
284 |
|
285 |
try:
|
286 |
result = lu.Exec(self._Feedback)
|
287 |
h_results = hm.RunPhase(constants.HOOKS_PHASE_POST) |
288 |
result = lu.HooksCallBack(constants.HOOKS_PHASE_POST, h_results, |
289 |
self._Feedback, result)
|
290 |
finally:
|
291 |
# FIXME: This needs locks if not lu_class.REQ_BGL
|
292 |
if write_count != self.context.cfg.write_count: |
293 |
hm.RunConfigUpdate() |
294 |
|
295 |
return result
|
296 |
|
297 |
def _LockAndExecLU(self, lu, level, calc_timeout): |
298 |
"""Execute a Logical Unit, with the needed locks.
|
299 |
|
300 |
This is a recursive function that starts locking the given level, and
|
301 |
proceeds up, till there are no more locks to acquire. Then it executes the
|
302 |
given LU and its opcodes.
|
303 |
|
304 |
"""
|
305 |
adding_locks = level in lu.add_locks
|
306 |
acquiring_locks = level in lu.needed_locks
|
307 |
if level not in locking.LEVELS: |
308 |
if self._cbs: |
309 |
self._cbs.NotifyStart()
|
310 |
|
311 |
result = self._ExecLU(lu)
|
312 |
|
313 |
elif adding_locks and acquiring_locks: |
314 |
# We could both acquire and add locks at the same level, but for now we
|
315 |
# don't need this, so we'll avoid the complicated code needed.
|
316 |
raise NotImplementedError("Can't declare locks to acquire when adding" |
317 |
" others")
|
318 |
|
319 |
elif adding_locks or acquiring_locks: |
320 |
lu.DeclareLocks(level) |
321 |
share = lu.share_locks[level] |
322 |
|
323 |
try:
|
324 |
assert adding_locks ^ acquiring_locks, \
|
325 |
"Locks must be either added or acquired"
|
326 |
|
327 |
if acquiring_locks:
|
328 |
# Acquiring locks
|
329 |
needed_locks = lu.needed_locks[level] |
330 |
|
331 |
self._ReportLocks(level, needed_locks, share, False) |
332 |
acquired = self.context.glm.acquire(level,
|
333 |
needed_locks, |
334 |
shared=share, |
335 |
timeout=calc_timeout()) |
336 |
# TODO: Report timeout
|
337 |
self._ReportLocks(level, needed_locks, share, True) |
338 |
|
339 |
if acquired is None: |
340 |
raise _LockAcquireTimeout()
|
341 |
|
342 |
lu.acquired_locks[level] = acquired |
343 |
|
344 |
else:
|
345 |
# Adding locks
|
346 |
add_locks = lu.add_locks[level] |
347 |
lu.remove_locks[level] = add_locks |
348 |
|
349 |
try:
|
350 |
self.context.glm.add(level, add_locks, acquired=1, shared=share) |
351 |
except errors.LockError:
|
352 |
raise errors.OpPrereqError(
|
353 |
"Couldn't add locks (%s), probably because of a race condition"
|
354 |
" with another job, who added them first" % add_locks)
|
355 |
|
356 |
lu.acquired_locks[level] = add_locks |
357 |
try:
|
358 |
result = self._LockAndExecLU(lu, level + 1, calc_timeout) |
359 |
finally:
|
360 |
if level in lu.remove_locks: |
361 |
self.context.glm.remove(level, lu.remove_locks[level])
|
362 |
finally:
|
363 |
if self.context.glm.is_owned(level): |
364 |
self.context.glm.release(level)
|
365 |
|
366 |
else:
|
367 |
result = self._LockAndExecLU(lu, level + 1, calc_timeout) |
368 |
|
369 |
return result
|
370 |
|
371 |
def ExecOpCode(self, op, cbs): |
372 |
"""Execute an opcode.
|
373 |
|
374 |
@type op: an OpCode instance
|
375 |
@param op: the opcode to be executed
|
376 |
@type cbs: L{OpExecCbBase}
|
377 |
@param cbs: Runtime callbacks
|
378 |
|
379 |
"""
|
380 |
if not isinstance(op, opcodes.OpCode): |
381 |
raise errors.ProgrammerError("Non-opcode instance passed" |
382 |
" to ExecOpcode")
|
383 |
|
384 |
self._cbs = cbs
|
385 |
try:
|
386 |
lu_class = self.DISPATCH_TABLE.get(op.__class__, None) |
387 |
if lu_class is None: |
388 |
raise errors.OpCodeUnknown("Unknown opcode") |
389 |
|
390 |
timeout_strategy = _LockTimeoutStrategy() |
391 |
calc_timeout = timeout_strategy.CalcRemainingTimeout |
392 |
|
393 |
while True: |
394 |
try:
|
395 |
self._ReportLocks(locking.LEVEL_CLUSTER, [locking.BGL],
|
396 |
not lu_class.REQ_BGL, False) |
397 |
try:
|
398 |
# Acquire the Big Ganeti Lock exclusively if this LU requires it,
|
399 |
# and in a shared fashion otherwise (to prevent concurrent run with
|
400 |
# an exclusive LU.
|
401 |
acquired_bgl = self.context.glm.acquire(locking.LEVEL_CLUSTER,
|
402 |
[locking.BGL], |
403 |
shared=not lu_class.REQ_BGL,
|
404 |
timeout=calc_timeout()) |
405 |
finally:
|
406 |
# TODO: Report timeout
|
407 |
self._ReportLocks(locking.LEVEL_CLUSTER, [locking.BGL],
|
408 |
not lu_class.REQ_BGL, True) |
409 |
|
410 |
if acquired_bgl is None: |
411 |
raise _LockAcquireTimeout()
|
412 |
|
413 |
try:
|
414 |
lu = lu_class(self, op, self.context, self.rpc) |
415 |
lu.ExpandNames() |
416 |
assert lu.needed_locks is not None, "needed_locks not set by LU" |
417 |
|
418 |
return self._LockAndExecLU(lu, locking.LEVEL_INSTANCE, calc_timeout) |
419 |
finally:
|
420 |
self.context.glm.release(locking.LEVEL_CLUSTER)
|
421 |
|
422 |
except _LockAcquireTimeout:
|
423 |
# Timeout while waiting for lock, try again
|
424 |
pass
|
425 |
|
426 |
timeout_strategy.NextAttempt() |
427 |
|
428 |
finally:
|
429 |
self._cbs = None |
430 |
|
431 |
def _Feedback(self, *args): |
432 |
"""Forward call to feedback callback function.
|
433 |
|
434 |
"""
|
435 |
if self._cbs: |
436 |
self._cbs.Feedback(*args)
|
437 |
|
438 |
def LogStep(self, current, total, message): |
439 |
"""Log a change in LU execution progress.
|
440 |
|
441 |
"""
|
442 |
logging.debug("Step %d/%d %s", current, total, message)
|
443 |
self._Feedback("STEP %d/%d %s" % (current, total, message)) |
444 |
|
445 |
def LogWarning(self, message, *args, **kwargs): |
446 |
"""Log a warning to the logs and the user.
|
447 |
|
448 |
The optional keyword argument is 'hint' and can be used to show a
|
449 |
hint to the user (presumably related to the warning). If the
|
450 |
message is empty, it will not be printed at all, allowing one to
|
451 |
show only a hint.
|
452 |
|
453 |
"""
|
454 |
assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \ |
455 |
"Invalid keyword arguments for LogWarning (%s)" % str(kwargs) |
456 |
if args:
|
457 |
message = message % tuple(args)
|
458 |
if message:
|
459 |
logging.warning(message) |
460 |
self._Feedback(" - WARNING: %s" % message) |
461 |
if "hint" in kwargs: |
462 |
self._Feedback(" Hint: %s" % kwargs["hint"]) |
463 |
|
464 |
def LogInfo(self, message, *args): |
465 |
"""Log an informational message to the logs and the user.
|
466 |
|
467 |
"""
|
468 |
if args:
|
469 |
message = message % tuple(args)
|
470 |
logging.info(message) |
471 |
self._Feedback(" - INFO: %s" % message) |
472 |
|
473 |
|
474 |
class HooksMaster(object): |
475 |
"""Hooks master.
|
476 |
|
477 |
This class distributes the run commands to the nodes based on the
|
478 |
specific LU class.
|
479 |
|
480 |
In order to remove the direct dependency on the rpc module, the
|
481 |
constructor needs a function which actually does the remote
|
482 |
call. This will usually be rpc.call_hooks_runner, but any function
|
483 |
which behaves the same works.
|
484 |
|
485 |
"""
|
486 |
def __init__(self, callfn, lu): |
487 |
self.callfn = callfn
|
488 |
self.lu = lu
|
489 |
self.op = lu.op
|
490 |
self.env, node_list_pre, node_list_post = self._BuildEnv() |
491 |
self.node_list = {constants.HOOKS_PHASE_PRE: node_list_pre,
|
492 |
constants.HOOKS_PHASE_POST: node_list_post} |
493 |
|
494 |
def _BuildEnv(self): |
495 |
"""Compute the environment and the target nodes.
|
496 |
|
497 |
Based on the opcode and the current node list, this builds the
|
498 |
environment for the hooks and the target node list for the run.
|
499 |
|
500 |
"""
|
501 |
env = { |
502 |
"PATH": "/sbin:/bin:/usr/sbin:/usr/bin", |
503 |
"GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
|
504 |
"GANETI_OP_CODE": self.op.OP_ID, |
505 |
"GANETI_OBJECT_TYPE": self.lu.HTYPE, |
506 |
"GANETI_DATA_DIR": constants.DATA_DIR,
|
507 |
} |
508 |
|
509 |
if self.lu.HPATH is not None: |
510 |
lu_env, lu_nodes_pre, lu_nodes_post = self.lu.BuildHooksEnv()
|
511 |
if lu_env:
|
512 |
for key in lu_env: |
513 |
env["GANETI_" + key] = lu_env[key]
|
514 |
else:
|
515 |
lu_nodes_pre = lu_nodes_post = [] |
516 |
|
517 |
return env, frozenset(lu_nodes_pre), frozenset(lu_nodes_post) |
518 |
|
519 |
def _RunWrapper(self, node_list, hpath, phase): |
520 |
"""Simple wrapper over self.callfn.
|
521 |
|
522 |
This method fixes the environment before doing the rpc call.
|
523 |
|
524 |
"""
|
525 |
env = self.env.copy()
|
526 |
env["GANETI_HOOKS_PHASE"] = phase
|
527 |
env["GANETI_HOOKS_PATH"] = hpath
|
528 |
if self.lu.cfg is not None: |
529 |
env["GANETI_CLUSTER"] = self.lu.cfg.GetClusterName() |
530 |
env["GANETI_MASTER"] = self.lu.cfg.GetMasterNode() |
531 |
|
532 |
env = dict([(str(key), str(val)) for key, val in env.iteritems()]) |
533 |
|
534 |
return self.callfn(node_list, hpath, phase, env) |
535 |
|
536 |
def RunPhase(self, phase, nodes=None): |
537 |
"""Run all the scripts for a phase.
|
538 |
|
539 |
This is the main function of the HookMaster.
|
540 |
|
541 |
@param phase: one of L{constants.HOOKS_PHASE_POST} or
|
542 |
L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
|
543 |
@param nodes: overrides the predefined list of nodes for the given phase
|
544 |
@return: the processed results of the hooks multi-node rpc call
|
545 |
@raise errors.HooksFailure: on communication failure to the nodes
|
546 |
@raise errors.HooksAbort: on failure of one of the hooks
|
547 |
|
548 |
"""
|
549 |
if not self.node_list[phase] and not nodes: |
550 |
# empty node list, we should not attempt to run this as either
|
551 |
# we're in the cluster init phase and the rpc client part can't
|
552 |
# even attempt to run, or this LU doesn't do hooks at all
|
553 |
return
|
554 |
hpath = self.lu.HPATH
|
555 |
if nodes is not None: |
556 |
results = self._RunWrapper(nodes, hpath, phase)
|
557 |
else:
|
558 |
results = self._RunWrapper(self.node_list[phase], hpath, phase) |
559 |
errs = [] |
560 |
if not results: |
561 |
msg = "Communication Failure"
|
562 |
if phase == constants.HOOKS_PHASE_PRE:
|
563 |
raise errors.HooksFailure(msg)
|
564 |
else:
|
565 |
self.lu.LogWarning(msg)
|
566 |
return results
|
567 |
for node_name in results: |
568 |
res = results[node_name] |
569 |
if res.offline:
|
570 |
continue
|
571 |
msg = res.fail_msg |
572 |
if msg:
|
573 |
self.lu.LogWarning("Communication failure to node %s: %s", |
574 |
node_name, msg) |
575 |
continue
|
576 |
for script, hkr, output in res.payload: |
577 |
if hkr == constants.HKR_FAIL:
|
578 |
if phase == constants.HOOKS_PHASE_PRE:
|
579 |
errs.append((node_name, script, output)) |
580 |
else:
|
581 |
if not output: |
582 |
output = "(no output)"
|
583 |
self.lu.LogWarning("On %s script %s failed, output: %s" % |
584 |
(node_name, script, output)) |
585 |
if errs and phase == constants.HOOKS_PHASE_PRE: |
586 |
raise errors.HooksAbort(errs)
|
587 |
return results
|
588 |
|
589 |
def RunConfigUpdate(self): |
590 |
"""Run the special configuration update hook
|
591 |
|
592 |
This is a special hook that runs only on the master after each
|
593 |
top-level LI if the configuration has been updated.
|
594 |
|
595 |
"""
|
596 |
phase = constants.HOOKS_PHASE_POST |
597 |
hpath = constants.HOOKS_NAME_CFGUPDATE |
598 |
nodes = [self.lu.cfg.GetMasterNode()]
|
599 |
self._RunWrapper(nodes, hpath, phase)
|