root / lib / mcpu.py @ 211b6132
History | View | Annotate | Download (19.2 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, timeout, acquired, result): |
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 locks should be acquired in shared mode
|
233 |
@type timeout: None or float
|
234 |
@param timeout: Timeout for acquiring the locks
|
235 |
@type acquired: bool
|
236 |
@param acquired: Whether the locks have already been acquired
|
237 |
@type result: None or set
|
238 |
@param result: Result from L{locking.GanetiLockManager.acquire}
|
239 |
|
240 |
"""
|
241 |
parts = [] |
242 |
|
243 |
# Build message
|
244 |
if acquired:
|
245 |
if result is None: |
246 |
parts.append("timeout")
|
247 |
else:
|
248 |
parts.append("acquired")
|
249 |
else:
|
250 |
parts.append("waiting")
|
251 |
if timeout is None: |
252 |
parts.append("blocking")
|
253 |
else:
|
254 |
parts.append("timeout=%0.6fs" % timeout)
|
255 |
|
256 |
parts.append(locking.LEVEL_NAMES[level]) |
257 |
|
258 |
if names == locking.ALL_SET:
|
259 |
parts.append("ALL")
|
260 |
elif isinstance(names, basestring): |
261 |
parts.append(names) |
262 |
else:
|
263 |
parts.append(",".join(names))
|
264 |
|
265 |
if shared:
|
266 |
parts.append("shared")
|
267 |
else:
|
268 |
parts.append("exclusive")
|
269 |
|
270 |
msg = "/".join(parts)
|
271 |
|
272 |
logging.debug("LU locks %s", msg)
|
273 |
|
274 |
if self._cbs: |
275 |
self._cbs.ReportLocks(msg)
|
276 |
|
277 |
def _AcquireLocks(self, level, names, shared, timeout): |
278 |
"""Acquires locks via the Ganeti lock manager.
|
279 |
|
280 |
@type level: int
|
281 |
@param level: Lock level
|
282 |
@type names: list or string
|
283 |
@param names: Lock names
|
284 |
@type shared: bool
|
285 |
@param shared: Whether the locks should be acquired in shared mode
|
286 |
@type timeout: None or float
|
287 |
@param timeout: Timeout for acquiring the locks
|
288 |
|
289 |
"""
|
290 |
self._ReportLocks(level, names, shared, timeout, False, None) |
291 |
|
292 |
acquired = self.context.glm.acquire(level, names, shared=shared,
|
293 |
timeout=timeout) |
294 |
|
295 |
self._ReportLocks(level, names, shared, timeout, True, acquired) |
296 |
|
297 |
return acquired
|
298 |
|
299 |
def _ExecLU(self, lu): |
300 |
"""Logical Unit execution sequence.
|
301 |
|
302 |
"""
|
303 |
write_count = self.context.cfg.write_count
|
304 |
lu.CheckPrereq() |
305 |
hm = HooksMaster(self.rpc.call_hooks_runner, lu)
|
306 |
h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE) |
307 |
lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results, |
308 |
self._Feedback, None) |
309 |
|
310 |
if getattr(lu.op, "dry_run", False): |
311 |
# in this mode, no post-hooks are run, and the config is not
|
312 |
# written (as it might have been modified by another LU, and we
|
313 |
# shouldn't do writeout on behalf of other threads
|
314 |
self.LogInfo("dry-run mode requested, not actually executing" |
315 |
" the operation")
|
316 |
return lu.dry_run_result
|
317 |
|
318 |
try:
|
319 |
result = lu.Exec(self._Feedback)
|
320 |
h_results = hm.RunPhase(constants.HOOKS_PHASE_POST) |
321 |
result = lu.HooksCallBack(constants.HOOKS_PHASE_POST, h_results, |
322 |
self._Feedback, result)
|
323 |
finally:
|
324 |
# FIXME: This needs locks if not lu_class.REQ_BGL
|
325 |
if write_count != self.context.cfg.write_count: |
326 |
hm.RunConfigUpdate() |
327 |
|
328 |
return result
|
329 |
|
330 |
def _LockAndExecLU(self, lu, level, calc_timeout): |
331 |
"""Execute a Logical Unit, with the needed locks.
|
332 |
|
333 |
This is a recursive function that starts locking the given level, and
|
334 |
proceeds up, till there are no more locks to acquire. Then it executes the
|
335 |
given LU and its opcodes.
|
336 |
|
337 |
"""
|
338 |
adding_locks = level in lu.add_locks
|
339 |
acquiring_locks = level in lu.needed_locks
|
340 |
if level not in locking.LEVELS: |
341 |
if self._cbs: |
342 |
self._cbs.NotifyStart()
|
343 |
|
344 |
result = self._ExecLU(lu)
|
345 |
|
346 |
elif adding_locks and acquiring_locks: |
347 |
# We could both acquire and add locks at the same level, but for now we
|
348 |
# don't need this, so we'll avoid the complicated code needed.
|
349 |
raise NotImplementedError("Can't declare locks to acquire when adding" |
350 |
" others")
|
351 |
|
352 |
elif adding_locks or acquiring_locks: |
353 |
lu.DeclareLocks(level) |
354 |
share = lu.share_locks[level] |
355 |
|
356 |
try:
|
357 |
assert adding_locks ^ acquiring_locks, \
|
358 |
"Locks must be either added or acquired"
|
359 |
|
360 |
if acquiring_locks:
|
361 |
# Acquiring locks
|
362 |
needed_locks = lu.needed_locks[level] |
363 |
|
364 |
acquired = self._AcquireLocks(level, needed_locks, share,
|
365 |
calc_timeout()) |
366 |
|
367 |
if acquired is None: |
368 |
raise _LockAcquireTimeout()
|
369 |
|
370 |
lu.acquired_locks[level] = acquired |
371 |
|
372 |
else:
|
373 |
# Adding locks
|
374 |
add_locks = lu.add_locks[level] |
375 |
lu.remove_locks[level] = add_locks |
376 |
|
377 |
try:
|
378 |
self.context.glm.add(level, add_locks, acquired=1, shared=share) |
379 |
except errors.LockError:
|
380 |
raise errors.OpPrereqError(
|
381 |
"Couldn't add locks (%s), probably because of a race condition"
|
382 |
" with another job, who added them first" % add_locks)
|
383 |
|
384 |
lu.acquired_locks[level] = add_locks |
385 |
try:
|
386 |
result = self._LockAndExecLU(lu, level + 1, calc_timeout) |
387 |
finally:
|
388 |
if level in lu.remove_locks: |
389 |
self.context.glm.remove(level, lu.remove_locks[level])
|
390 |
finally:
|
391 |
if self.context.glm.is_owned(level): |
392 |
self.context.glm.release(level)
|
393 |
|
394 |
else:
|
395 |
result = self._LockAndExecLU(lu, level + 1, calc_timeout) |
396 |
|
397 |
return result
|
398 |
|
399 |
def ExecOpCode(self, op, cbs): |
400 |
"""Execute an opcode.
|
401 |
|
402 |
@type op: an OpCode instance
|
403 |
@param op: the opcode to be executed
|
404 |
@type cbs: L{OpExecCbBase}
|
405 |
@param cbs: Runtime callbacks
|
406 |
|
407 |
"""
|
408 |
if not isinstance(op, opcodes.OpCode): |
409 |
raise errors.ProgrammerError("Non-opcode instance passed" |
410 |
" to ExecOpcode")
|
411 |
|
412 |
self._cbs = cbs
|
413 |
try:
|
414 |
lu_class = self.DISPATCH_TABLE.get(op.__class__, None) |
415 |
if lu_class is None: |
416 |
raise errors.OpCodeUnknown("Unknown opcode") |
417 |
|
418 |
timeout_strategy = _LockTimeoutStrategy() |
419 |
calc_timeout = timeout_strategy.CalcRemainingTimeout |
420 |
|
421 |
while True: |
422 |
try:
|
423 |
# Acquire the Big Ganeti Lock exclusively if this LU requires it,
|
424 |
# and in a shared fashion otherwise (to prevent concurrent run with
|
425 |
# an exclusive LU.
|
426 |
if self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL, |
427 |
not lu_class.REQ_BGL, calc_timeout()) is None: |
428 |
raise _LockAcquireTimeout()
|
429 |
|
430 |
try:
|
431 |
lu = lu_class(self, op, self.context, self.rpc) |
432 |
lu.ExpandNames() |
433 |
assert lu.needed_locks is not None, "needed_locks not set by LU" |
434 |
|
435 |
return self._LockAndExecLU(lu, locking.LEVEL_INSTANCE, calc_timeout) |
436 |
finally:
|
437 |
self.context.glm.release(locking.LEVEL_CLUSTER)
|
438 |
|
439 |
except _LockAcquireTimeout:
|
440 |
# Timeout while waiting for lock, try again
|
441 |
pass
|
442 |
|
443 |
timeout_strategy.NextAttempt() |
444 |
|
445 |
finally:
|
446 |
self._cbs = None |
447 |
|
448 |
def _Feedback(self, *args): |
449 |
"""Forward call to feedback callback function.
|
450 |
|
451 |
"""
|
452 |
if self._cbs: |
453 |
self._cbs.Feedback(*args)
|
454 |
|
455 |
def LogStep(self, current, total, message): |
456 |
"""Log a change in LU execution progress.
|
457 |
|
458 |
"""
|
459 |
logging.debug("Step %d/%d %s", current, total, message)
|
460 |
self._Feedback("STEP %d/%d %s" % (current, total, message)) |
461 |
|
462 |
def LogWarning(self, message, *args, **kwargs): |
463 |
"""Log a warning to the logs and the user.
|
464 |
|
465 |
The optional keyword argument is 'hint' and can be used to show a
|
466 |
hint to the user (presumably related to the warning). If the
|
467 |
message is empty, it will not be printed at all, allowing one to
|
468 |
show only a hint.
|
469 |
|
470 |
"""
|
471 |
assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \ |
472 |
"Invalid keyword arguments for LogWarning (%s)" % str(kwargs) |
473 |
if args:
|
474 |
message = message % tuple(args)
|
475 |
if message:
|
476 |
logging.warning(message) |
477 |
self._Feedback(" - WARNING: %s" % message) |
478 |
if "hint" in kwargs: |
479 |
self._Feedback(" Hint: %s" % kwargs["hint"]) |
480 |
|
481 |
def LogInfo(self, message, *args): |
482 |
"""Log an informational message to the logs and the user.
|
483 |
|
484 |
"""
|
485 |
if args:
|
486 |
message = message % tuple(args)
|
487 |
logging.info(message) |
488 |
self._Feedback(" - INFO: %s" % message) |
489 |
|
490 |
|
491 |
class HooksMaster(object): |
492 |
"""Hooks master.
|
493 |
|
494 |
This class distributes the run commands to the nodes based on the
|
495 |
specific LU class.
|
496 |
|
497 |
In order to remove the direct dependency on the rpc module, the
|
498 |
constructor needs a function which actually does the remote
|
499 |
call. This will usually be rpc.call_hooks_runner, but any function
|
500 |
which behaves the same works.
|
501 |
|
502 |
"""
|
503 |
def __init__(self, callfn, lu): |
504 |
self.callfn = callfn
|
505 |
self.lu = lu
|
506 |
self.op = lu.op
|
507 |
self.env, node_list_pre, node_list_post = self._BuildEnv() |
508 |
self.node_list = {constants.HOOKS_PHASE_PRE: node_list_pre,
|
509 |
constants.HOOKS_PHASE_POST: node_list_post} |
510 |
|
511 |
def _BuildEnv(self): |
512 |
"""Compute the environment and the target nodes.
|
513 |
|
514 |
Based on the opcode and the current node list, this builds the
|
515 |
environment for the hooks and the target node list for the run.
|
516 |
|
517 |
"""
|
518 |
env = { |
519 |
"PATH": "/sbin:/bin:/usr/sbin:/usr/bin", |
520 |
"GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
|
521 |
"GANETI_OP_CODE": self.op.OP_ID, |
522 |
"GANETI_OBJECT_TYPE": self.lu.HTYPE, |
523 |
"GANETI_DATA_DIR": constants.DATA_DIR,
|
524 |
} |
525 |
|
526 |
if self.lu.HPATH is not None: |
527 |
lu_env, lu_nodes_pre, lu_nodes_post = self.lu.BuildHooksEnv()
|
528 |
if lu_env:
|
529 |
for key in lu_env: |
530 |
env["GANETI_" + key] = lu_env[key]
|
531 |
else:
|
532 |
lu_nodes_pre = lu_nodes_post = [] |
533 |
|
534 |
return env, frozenset(lu_nodes_pre), frozenset(lu_nodes_post) |
535 |
|
536 |
def _RunWrapper(self, node_list, hpath, phase): |
537 |
"""Simple wrapper over self.callfn.
|
538 |
|
539 |
This method fixes the environment before doing the rpc call.
|
540 |
|
541 |
"""
|
542 |
env = self.env.copy()
|
543 |
env["GANETI_HOOKS_PHASE"] = phase
|
544 |
env["GANETI_HOOKS_PATH"] = hpath
|
545 |
if self.lu.cfg is not None: |
546 |
env["GANETI_CLUSTER"] = self.lu.cfg.GetClusterName() |
547 |
env["GANETI_MASTER"] = self.lu.cfg.GetMasterNode() |
548 |
|
549 |
env = dict([(str(key), str(val)) for key, val in env.iteritems()]) |
550 |
|
551 |
return self.callfn(node_list, hpath, phase, env) |
552 |
|
553 |
def RunPhase(self, phase, nodes=None): |
554 |
"""Run all the scripts for a phase.
|
555 |
|
556 |
This is the main function of the HookMaster.
|
557 |
|
558 |
@param phase: one of L{constants.HOOKS_PHASE_POST} or
|
559 |
L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
|
560 |
@param nodes: overrides the predefined list of nodes for the given phase
|
561 |
@return: the processed results of the hooks multi-node rpc call
|
562 |
@raise errors.HooksFailure: on communication failure to the nodes
|
563 |
@raise errors.HooksAbort: on failure of one of the hooks
|
564 |
|
565 |
"""
|
566 |
if not self.node_list[phase] and not nodes: |
567 |
# empty node list, we should not attempt to run this as either
|
568 |
# we're in the cluster init phase and the rpc client part can't
|
569 |
# even attempt to run, or this LU doesn't do hooks at all
|
570 |
return
|
571 |
hpath = self.lu.HPATH
|
572 |
if nodes is not None: |
573 |
results = self._RunWrapper(nodes, hpath, phase)
|
574 |
else:
|
575 |
results = self._RunWrapper(self.node_list[phase], hpath, phase) |
576 |
errs = [] |
577 |
if not results: |
578 |
msg = "Communication Failure"
|
579 |
if phase == constants.HOOKS_PHASE_PRE:
|
580 |
raise errors.HooksFailure(msg)
|
581 |
else:
|
582 |
self.lu.LogWarning(msg)
|
583 |
return results
|
584 |
for node_name in results: |
585 |
res = results[node_name] |
586 |
if res.offline:
|
587 |
continue
|
588 |
msg = res.fail_msg |
589 |
if msg:
|
590 |
self.lu.LogWarning("Communication failure to node %s: %s", |
591 |
node_name, msg) |
592 |
continue
|
593 |
for script, hkr, output in res.payload: |
594 |
if hkr == constants.HKR_FAIL:
|
595 |
if phase == constants.HOOKS_PHASE_PRE:
|
596 |
errs.append((node_name, script, output)) |
597 |
else:
|
598 |
if not output: |
599 |
output = "(no output)"
|
600 |
self.lu.LogWarning("On %s script %s failed, output: %s" % |
601 |
(node_name, script, output)) |
602 |
if errs and phase == constants.HOOKS_PHASE_PRE: |
603 |
raise errors.HooksAbort(errs)
|
604 |
return results
|
605 |
|
606 |
def RunConfigUpdate(self): |
607 |
"""Run the special configuration update hook
|
608 |
|
609 |
This is a special hook that runs only on the master after each
|
610 |
top-level LI if the configuration has been updated.
|
611 |
|
612 |
"""
|
613 |
phase = constants.HOOKS_PHASE_POST |
614 |
hpath = constants.HOOKS_NAME_CFGUPDATE |
615 |
nodes = [self.lu.cfg.GetMasterNode()]
|
616 |
self._RunWrapper(nodes, hpath, phase)
|