root / lib / jqueue.py @ 21f04e5e
History | View | Annotate | Download (33.6 kB)
1 |
#
|
---|---|
2 |
#
|
3 |
|
4 |
# Copyright (C) 2006, 2007, 2008 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 job queue handling.
|
23 |
|
24 |
Locking: there's a single, large lock in the L{JobQueue} class. It's
|
25 |
used by all other classes in this module.
|
26 |
|
27 |
@var JOBQUEUE_THREADS: the number of worker threads we start for
|
28 |
processing jobs
|
29 |
|
30 |
"""
|
31 |
|
32 |
import os |
33 |
import logging |
34 |
import threading |
35 |
import errno |
36 |
import re |
37 |
import time |
38 |
import weakref |
39 |
|
40 |
from ganeti import constants |
41 |
from ganeti import serializer |
42 |
from ganeti import workerpool |
43 |
from ganeti import opcodes |
44 |
from ganeti import errors |
45 |
from ganeti import mcpu |
46 |
from ganeti import utils |
47 |
from ganeti import jstore |
48 |
from ganeti import rpc |
49 |
|
50 |
JOBQUEUE_THREADS = 25
|
51 |
|
52 |
|
53 |
def TimeStampNow(): |
54 |
"""Returns the current timestamp.
|
55 |
|
56 |
@rtype: tuple
|
57 |
@return: the current time in the (seconds, microseconds) format
|
58 |
|
59 |
"""
|
60 |
return utils.SplitTime(time.time())
|
61 |
|
62 |
|
63 |
class _QueuedOpCode(object): |
64 |
"""Encasulates an opcode object.
|
65 |
|
66 |
@ivar log: holds the execution log and consists of tuples
|
67 |
of the form C{(log_serial, timestamp, level, message)}
|
68 |
@ivar input: the OpCode we encapsulate
|
69 |
@ivar status: the current status
|
70 |
@ivar result: the result of the LU execution
|
71 |
@ivar start_timestamp: timestamp for the start of the execution
|
72 |
@ivar stop_timestamp: timestamp for the end of the execution
|
73 |
|
74 |
"""
|
75 |
def __init__(self, op): |
76 |
"""Constructor for the _QuededOpCode.
|
77 |
|
78 |
@type op: L{opcodes.OpCode}
|
79 |
@param op: the opcode we encapsulate
|
80 |
|
81 |
"""
|
82 |
self.input = op
|
83 |
self.status = constants.OP_STATUS_QUEUED
|
84 |
self.result = None |
85 |
self.log = []
|
86 |
self.start_timestamp = None |
87 |
self.end_timestamp = None |
88 |
|
89 |
@classmethod
|
90 |
def Restore(cls, state): |
91 |
"""Restore the _QueuedOpCode from the serialized form.
|
92 |
|
93 |
@type state: dict
|
94 |
@param state: the serialized state
|
95 |
@rtype: _QueuedOpCode
|
96 |
@return: a new _QueuedOpCode instance
|
97 |
|
98 |
"""
|
99 |
obj = _QueuedOpCode.__new__(cls) |
100 |
obj.input = opcodes.OpCode.LoadOpCode(state["input"])
|
101 |
obj.status = state["status"]
|
102 |
obj.result = state["result"]
|
103 |
obj.log = state["log"]
|
104 |
obj.start_timestamp = state.get("start_timestamp", None) |
105 |
obj.end_timestamp = state.get("end_timestamp", None) |
106 |
return obj
|
107 |
|
108 |
def Serialize(self): |
109 |
"""Serializes this _QueuedOpCode.
|
110 |
|
111 |
@rtype: dict
|
112 |
@return: the dictionary holding the serialized state
|
113 |
|
114 |
"""
|
115 |
return {
|
116 |
"input": self.input.__getstate__(), |
117 |
"status": self.status, |
118 |
"result": self.result, |
119 |
"log": self.log, |
120 |
"start_timestamp": self.start_timestamp, |
121 |
"end_timestamp": self.end_timestamp, |
122 |
} |
123 |
|
124 |
|
125 |
class _QueuedJob(object): |
126 |
"""In-memory job representation.
|
127 |
|
128 |
This is what we use to track the user-submitted jobs. Locking must
|
129 |
be taken care of by users of this class.
|
130 |
|
131 |
@type queue: L{JobQueue}
|
132 |
@ivar queue: the parent queue
|
133 |
@ivar id: the job ID
|
134 |
@type ops: list
|
135 |
@ivar ops: the list of _QueuedOpCode that constitute the job
|
136 |
@type run_op_index: int
|
137 |
@ivar run_op_index: the currently executing opcode, or -1 if
|
138 |
we didn't yet start executing
|
139 |
@type log_serial: int
|
140 |
@ivar log_serial: holds the index for the next log entry
|
141 |
@ivar received_timestamp: the timestamp for when the job was received
|
142 |
@ivar start_timestmap: the timestamp for start of execution
|
143 |
@ivar end_timestamp: the timestamp for end of execution
|
144 |
@ivar change: a Condition variable we use for waiting for job changes
|
145 |
|
146 |
"""
|
147 |
def __init__(self, queue, job_id, ops): |
148 |
"""Constructor for the _QueuedJob.
|
149 |
|
150 |
@type queue: L{JobQueue}
|
151 |
@param queue: our parent queue
|
152 |
@type job_id: job_id
|
153 |
@param job_id: our job id
|
154 |
@type ops: list
|
155 |
@param ops: the list of opcodes we hold, which will be encapsulated
|
156 |
in _QueuedOpCodes
|
157 |
|
158 |
"""
|
159 |
if not ops: |
160 |
# TODO: use a better exception
|
161 |
raise Exception("No opcodes") |
162 |
|
163 |
self.queue = queue
|
164 |
self.id = job_id
|
165 |
self.ops = [_QueuedOpCode(op) for op in ops] |
166 |
self.run_op_index = -1 |
167 |
self.log_serial = 0 |
168 |
self.received_timestamp = TimeStampNow()
|
169 |
self.start_timestamp = None |
170 |
self.end_timestamp = None |
171 |
|
172 |
# Condition to wait for changes
|
173 |
self.change = threading.Condition(self.queue._lock) |
174 |
|
175 |
@classmethod
|
176 |
def Restore(cls, queue, state): |
177 |
"""Restore a _QueuedJob from serialized state:
|
178 |
|
179 |
@type queue: L{JobQueue}
|
180 |
@param queue: to which queue the restored job belongs
|
181 |
@type state: dict
|
182 |
@param state: the serialized state
|
183 |
@rtype: _JobQueue
|
184 |
@return: the restored _JobQueue instance
|
185 |
|
186 |
"""
|
187 |
obj = _QueuedJob.__new__(cls) |
188 |
obj.queue = queue |
189 |
obj.id = state["id"]
|
190 |
obj.run_op_index = state["run_op_index"]
|
191 |
obj.received_timestamp = state.get("received_timestamp", None) |
192 |
obj.start_timestamp = state.get("start_timestamp", None) |
193 |
obj.end_timestamp = state.get("end_timestamp", None) |
194 |
|
195 |
obj.ops = [] |
196 |
obj.log_serial = 0
|
197 |
for op_state in state["ops"]: |
198 |
op = _QueuedOpCode.Restore(op_state) |
199 |
for log_entry in op.log: |
200 |
obj.log_serial = max(obj.log_serial, log_entry[0]) |
201 |
obj.ops.append(op) |
202 |
|
203 |
# Condition to wait for changes
|
204 |
obj.change = threading.Condition(obj.queue._lock) |
205 |
|
206 |
return obj
|
207 |
|
208 |
def Serialize(self): |
209 |
"""Serialize the _JobQueue instance.
|
210 |
|
211 |
@rtype: dict
|
212 |
@return: the serialized state
|
213 |
|
214 |
"""
|
215 |
return {
|
216 |
"id": self.id, |
217 |
"ops": [op.Serialize() for op in self.ops], |
218 |
"run_op_index": self.run_op_index, |
219 |
"start_timestamp": self.start_timestamp, |
220 |
"end_timestamp": self.end_timestamp, |
221 |
"received_timestamp": self.received_timestamp, |
222 |
} |
223 |
|
224 |
def CalcStatus(self): |
225 |
"""Compute the status of this job.
|
226 |
|
227 |
This function iterates over all the _QueuedOpCodes in the job and
|
228 |
based on their status, computes the job status.
|
229 |
|
230 |
The algorithm is:
|
231 |
- if we find a cancelled, or finished with error, the job
|
232 |
status will be the same
|
233 |
- otherwise, the last opcode with the status one of:
|
234 |
- waitlock
|
235 |
- running
|
236 |
|
237 |
will determine the job status
|
238 |
|
239 |
- otherwise, it means either all opcodes are queued, or success,
|
240 |
and the job status will be the same
|
241 |
|
242 |
@return: the job status
|
243 |
|
244 |
"""
|
245 |
status = constants.JOB_STATUS_QUEUED |
246 |
|
247 |
all_success = True
|
248 |
for op in self.ops: |
249 |
if op.status == constants.OP_STATUS_SUCCESS:
|
250 |
continue
|
251 |
|
252 |
all_success = False
|
253 |
|
254 |
if op.status == constants.OP_STATUS_QUEUED:
|
255 |
pass
|
256 |
elif op.status == constants.OP_STATUS_WAITLOCK:
|
257 |
status = constants.JOB_STATUS_WAITLOCK |
258 |
elif op.status == constants.OP_STATUS_RUNNING:
|
259 |
status = constants.JOB_STATUS_RUNNING |
260 |
elif op.status == constants.OP_STATUS_ERROR:
|
261 |
status = constants.JOB_STATUS_ERROR |
262 |
# The whole job fails if one opcode failed
|
263 |
break
|
264 |
elif op.status == constants.OP_STATUS_CANCELED:
|
265 |
status = constants.OP_STATUS_CANCELED |
266 |
break
|
267 |
|
268 |
if all_success:
|
269 |
status = constants.JOB_STATUS_SUCCESS |
270 |
|
271 |
return status
|
272 |
|
273 |
def GetLogEntries(self, newer_than): |
274 |
"""Selectively returns the log entries.
|
275 |
|
276 |
@type newer_than: None or int
|
277 |
@param newer_than: if this is None, return all log enties,
|
278 |
otherwise return only the log entries with serial higher
|
279 |
than this value
|
280 |
@rtype: list
|
281 |
@return: the list of the log entries selected
|
282 |
|
283 |
"""
|
284 |
if newer_than is None: |
285 |
serial = -1
|
286 |
else:
|
287 |
serial = newer_than |
288 |
|
289 |
entries = [] |
290 |
for op in self.ops: |
291 |
entries.extend(filter(lambda entry: entry[0] > serial, op.log)) |
292 |
|
293 |
return entries
|
294 |
|
295 |
|
296 |
class _JobQueueWorker(workerpool.BaseWorker): |
297 |
"""The actual job workers.
|
298 |
|
299 |
"""
|
300 |
def _NotifyStart(self): |
301 |
"""Mark the opcode as running, not lock-waiting.
|
302 |
|
303 |
This is called from the mcpu code as a notifier function, when the
|
304 |
LU is finally about to start the Exec() method. Of course, to have
|
305 |
end-user visible results, the opcode must be initially (before
|
306 |
calling into Processor.ExecOpCode) set to OP_STATUS_WAITLOCK.
|
307 |
|
308 |
"""
|
309 |
assert self.queue, "Queue attribute is missing" |
310 |
assert self.opcode, "Opcode attribute is missing" |
311 |
|
312 |
self.queue.acquire()
|
313 |
try:
|
314 |
self.opcode.status = constants.OP_STATUS_RUNNING
|
315 |
finally:
|
316 |
self.queue.release()
|
317 |
|
318 |
def RunTask(self, job): |
319 |
"""Job executor.
|
320 |
|
321 |
This functions processes a job. It is closely tied to the _QueuedJob and
|
322 |
_QueuedOpCode classes.
|
323 |
|
324 |
@type job: L{_QueuedJob}
|
325 |
@param job: the job to be processed
|
326 |
|
327 |
"""
|
328 |
logging.debug("Worker %s processing job %s",
|
329 |
self.worker_id, job.id)
|
330 |
proc = mcpu.Processor(self.pool.queue.context)
|
331 |
self.queue = queue = job.queue
|
332 |
try:
|
333 |
try:
|
334 |
count = len(job.ops)
|
335 |
for idx, op in enumerate(job.ops): |
336 |
try:
|
337 |
logging.debug("Op %s/%s: Starting %s", idx + 1, count, op) |
338 |
|
339 |
queue.acquire() |
340 |
try:
|
341 |
job.run_op_index = idx |
342 |
op.status = constants.OP_STATUS_WAITLOCK |
343 |
op.result = None
|
344 |
op.start_timestamp = TimeStampNow() |
345 |
if idx == 0: # first opcode |
346 |
job.start_timestamp = op.start_timestamp |
347 |
queue.UpdateJobUnlocked(job) |
348 |
|
349 |
input_opcode = op.input |
350 |
finally:
|
351 |
queue.release() |
352 |
|
353 |
def _Log(*args): |
354 |
"""Append a log entry.
|
355 |
|
356 |
"""
|
357 |
assert len(args) < 3 |
358 |
|
359 |
if len(args) == 1: |
360 |
log_type = constants.ELOG_MESSAGE |
361 |
log_msg = args[0]
|
362 |
else:
|
363 |
log_type, log_msg = args |
364 |
|
365 |
# The time is split to make serialization easier and not lose
|
366 |
# precision.
|
367 |
timestamp = utils.SplitTime(time.time()) |
368 |
|
369 |
queue.acquire() |
370 |
try:
|
371 |
job.log_serial += 1
|
372 |
op.log.append((job.log_serial, timestamp, log_type, log_msg)) |
373 |
|
374 |
job.change.notifyAll() |
375 |
finally:
|
376 |
queue.release() |
377 |
|
378 |
# Make sure not to hold lock while _Log is called
|
379 |
self.opcode = op
|
380 |
result = proc.ExecOpCode(input_opcode, _Log, self._NotifyStart)
|
381 |
|
382 |
queue.acquire() |
383 |
try:
|
384 |
op.status = constants.OP_STATUS_SUCCESS |
385 |
op.result = result |
386 |
op.end_timestamp = TimeStampNow() |
387 |
queue.UpdateJobUnlocked(job) |
388 |
finally:
|
389 |
queue.release() |
390 |
|
391 |
logging.debug("Op %s/%s: Successfully finished %s",
|
392 |
idx + 1, count, op)
|
393 |
except Exception, err: |
394 |
queue.acquire() |
395 |
try:
|
396 |
try:
|
397 |
op.status = constants.OP_STATUS_ERROR |
398 |
op.result = str(err)
|
399 |
op.end_timestamp = TimeStampNow() |
400 |
logging.debug("Op %s/%s: Error in %s", idx + 1, count, op) |
401 |
finally:
|
402 |
queue.UpdateJobUnlocked(job) |
403 |
finally:
|
404 |
queue.release() |
405 |
raise
|
406 |
|
407 |
except errors.GenericError, err:
|
408 |
logging.exception("Ganeti exception")
|
409 |
except:
|
410 |
logging.exception("Unhandled exception")
|
411 |
finally:
|
412 |
queue.acquire() |
413 |
try:
|
414 |
try:
|
415 |
job.run_op_idx = -1
|
416 |
job.end_timestamp = TimeStampNow() |
417 |
queue.UpdateJobUnlocked(job) |
418 |
finally:
|
419 |
job_id = job.id |
420 |
status = job.CalcStatus() |
421 |
finally:
|
422 |
queue.release() |
423 |
logging.debug("Worker %s finished job %s, status = %s",
|
424 |
self.worker_id, job_id, status)
|
425 |
|
426 |
|
427 |
class _JobQueueWorkerPool(workerpool.WorkerPool): |
428 |
"""Simple class implementing a job-processing workerpool.
|
429 |
|
430 |
"""
|
431 |
def __init__(self, queue): |
432 |
super(_JobQueueWorkerPool, self).__init__(JOBQUEUE_THREADS, |
433 |
_JobQueueWorker) |
434 |
self.queue = queue
|
435 |
|
436 |
|
437 |
class JobQueue(object): |
438 |
"""Quue used to manaage the jobs.
|
439 |
|
440 |
@cvar _RE_JOB_FILE: regex matching the valid job file names
|
441 |
|
442 |
"""
|
443 |
_RE_JOB_FILE = re.compile(r"^job-(%s)$" % constants.JOB_ID_TEMPLATE)
|
444 |
|
445 |
def _RequireOpenQueue(fn): |
446 |
"""Decorator for "public" functions.
|
447 |
|
448 |
This function should be used for all 'public' functions. That is,
|
449 |
functions usually called from other classes.
|
450 |
|
451 |
@warning: Use this decorator only after utils.LockedMethod!
|
452 |
|
453 |
Example::
|
454 |
@utils.LockedMethod
|
455 |
@_RequireOpenQueue
|
456 |
def Example(self):
|
457 |
pass
|
458 |
|
459 |
"""
|
460 |
def wrapper(self, *args, **kwargs): |
461 |
assert self._queue_lock is not None, "Queue should be open" |
462 |
return fn(self, *args, **kwargs) |
463 |
return wrapper
|
464 |
|
465 |
def __init__(self, context): |
466 |
"""Constructor for JobQueue.
|
467 |
|
468 |
The constructor will initialize the job queue object and then
|
469 |
start loading the current jobs from disk, either for starting them
|
470 |
(if they were queue) or for aborting them (if they were already
|
471 |
running).
|
472 |
|
473 |
@type context: GanetiContext
|
474 |
@param context: the context object for access to the configuration
|
475 |
data and other ganeti objects
|
476 |
|
477 |
"""
|
478 |
self.context = context
|
479 |
self._memcache = weakref.WeakValueDictionary()
|
480 |
self._my_hostname = utils.HostInfo().name
|
481 |
|
482 |
# Locking
|
483 |
self._lock = threading.Lock()
|
484 |
self.acquire = self._lock.acquire |
485 |
self.release = self._lock.release |
486 |
|
487 |
# Initialize
|
488 |
self._queue_lock = jstore.InitAndVerifyQueue(must_lock=True) |
489 |
|
490 |
# Read serial file
|
491 |
self._last_serial = jstore.ReadSerial()
|
492 |
assert self._last_serial is not None, ("Serial file was modified between" |
493 |
" check in jstore and here")
|
494 |
|
495 |
# Get initial list of nodes
|
496 |
self._nodes = dict((n.name, n.primary_ip) |
497 |
for n in self.context.cfg.GetAllNodesInfo().values()) |
498 |
|
499 |
# Remove master node
|
500 |
try:
|
501 |
del self._nodes[self._my_hostname] |
502 |
except ValueError: |
503 |
pass
|
504 |
|
505 |
# TODO: Check consistency across nodes
|
506 |
|
507 |
# Setup worker pool
|
508 |
self._wpool = _JobQueueWorkerPool(self) |
509 |
|
510 |
# We need to lock here because WorkerPool.AddTask() may start a job while
|
511 |
# we're still doing our work.
|
512 |
self.acquire()
|
513 |
try:
|
514 |
for job in self._GetJobsUnlocked(None): |
515 |
# a failure in loading the job can cause 'None' to be returned
|
516 |
if job is None: |
517 |
continue
|
518 |
|
519 |
status = job.CalcStatus() |
520 |
|
521 |
if status in (constants.JOB_STATUS_QUEUED, ): |
522 |
self._wpool.AddTask(job)
|
523 |
|
524 |
elif status in (constants.JOB_STATUS_RUNNING, |
525 |
constants.JOB_STATUS_WAITLOCK): |
526 |
logging.warning("Unfinished job %s found: %s", job.id, job)
|
527 |
try:
|
528 |
for op in job.ops: |
529 |
op.status = constants.OP_STATUS_ERROR |
530 |
op.result = "Unclean master daemon shutdown"
|
531 |
finally:
|
532 |
self.UpdateJobUnlocked(job)
|
533 |
finally:
|
534 |
self.release()
|
535 |
|
536 |
@utils.LockedMethod
|
537 |
@_RequireOpenQueue
|
538 |
def AddNode(self, node): |
539 |
"""Register a new node with the queue.
|
540 |
|
541 |
@type node: L{objects.Node}
|
542 |
@param node: the node object to be added
|
543 |
|
544 |
"""
|
545 |
node_name = node.name |
546 |
assert node_name != self._my_hostname |
547 |
|
548 |
# Clean queue directory on added node
|
549 |
rpc.RpcRunner.call_jobqueue_purge(node_name) |
550 |
|
551 |
# Upload the whole queue excluding archived jobs
|
552 |
files = [self._GetJobPath(job_id) for job_id in self._GetJobIDsUnlocked()] |
553 |
|
554 |
# Upload current serial file
|
555 |
files.append(constants.JOB_QUEUE_SERIAL_FILE) |
556 |
|
557 |
for file_name in files: |
558 |
# Read file content
|
559 |
fd = open(file_name, "r") |
560 |
try:
|
561 |
content = fd.read() |
562 |
finally:
|
563 |
fd.close() |
564 |
|
565 |
result = rpc.RpcRunner.call_jobqueue_update([node_name], |
566 |
[node.primary_ip], |
567 |
file_name, content) |
568 |
if not result[node_name]: |
569 |
logging.error("Failed to upload %s to %s", file_name, node_name)
|
570 |
|
571 |
self._nodes[node_name] = node.primary_ip
|
572 |
|
573 |
@utils.LockedMethod
|
574 |
@_RequireOpenQueue
|
575 |
def RemoveNode(self, node_name): |
576 |
"""Callback called when removing nodes from the cluster.
|
577 |
|
578 |
@type node_name: str
|
579 |
@param node_name: the name of the node to remove
|
580 |
|
581 |
"""
|
582 |
try:
|
583 |
# The queue is removed by the "leave node" RPC call.
|
584 |
del self._nodes[node_name] |
585 |
except KeyError: |
586 |
pass
|
587 |
|
588 |
def _CheckRpcResult(self, result, nodes, failmsg): |
589 |
"""Verifies the status of an RPC call.
|
590 |
|
591 |
Since we aim to keep consistency should this node (the current
|
592 |
master) fail, we will log errors if our rpc fail, and especially
|
593 |
log the case when more than half of the nodes failes.
|
594 |
|
595 |
@param result: the data as returned from the rpc call
|
596 |
@type nodes: list
|
597 |
@param nodes: the list of nodes we made the call to
|
598 |
@type failmsg: str
|
599 |
@param failmsg: the identifier to be used for logging
|
600 |
|
601 |
"""
|
602 |
failed = [] |
603 |
success = [] |
604 |
|
605 |
for node in nodes: |
606 |
if result[node]:
|
607 |
success.append(node) |
608 |
else:
|
609 |
failed.append(node) |
610 |
|
611 |
if failed:
|
612 |
logging.error("%s failed on %s", failmsg, ", ".join(failed)) |
613 |
|
614 |
# +1 for the master node
|
615 |
if (len(success) + 1) < len(failed): |
616 |
# TODO: Handle failing nodes
|
617 |
logging.error("More than half of the nodes failed")
|
618 |
|
619 |
def _GetNodeIp(self): |
620 |
"""Helper for returning the node name/ip list.
|
621 |
|
622 |
@rtype: (list, list)
|
623 |
@return: a tuple of two lists, the first one with the node
|
624 |
names and the second one with the node addresses
|
625 |
|
626 |
"""
|
627 |
name_list = self._nodes.keys()
|
628 |
addr_list = [self._nodes[name] for name in name_list] |
629 |
return name_list, addr_list
|
630 |
|
631 |
def _WriteAndReplicateFileUnlocked(self, file_name, data): |
632 |
"""Writes a file locally and then replicates it to all nodes.
|
633 |
|
634 |
This function will replace the contents of a file on the local
|
635 |
node and then replicate it to all the other nodes we have.
|
636 |
|
637 |
@type file_name: str
|
638 |
@param file_name: the path of the file to be replicated
|
639 |
@type data: str
|
640 |
@param data: the new contents of the file
|
641 |
|
642 |
"""
|
643 |
utils.WriteFile(file_name, data=data) |
644 |
|
645 |
names, addrs = self._GetNodeIp()
|
646 |
result = rpc.RpcRunner.call_jobqueue_update(names, addrs, file_name, data) |
647 |
self._CheckRpcResult(result, self._nodes, |
648 |
"Updating %s" % file_name)
|
649 |
|
650 |
def _RenameFileUnlocked(self, old, new): |
651 |
"""Renames a file locally and then replicate the change.
|
652 |
|
653 |
This function will rename a file in the local queue directory
|
654 |
and then replicate this rename to all the other nodes we have.
|
655 |
|
656 |
@type old: str
|
657 |
@param old: the current name of the file
|
658 |
@type new: str
|
659 |
@param new: the new name of the file
|
660 |
|
661 |
"""
|
662 |
os.rename(old, new) |
663 |
|
664 |
names, addrs = self._GetNodeIp()
|
665 |
result = rpc.RpcRunner.call_jobqueue_rename(names, addrs, old, new) |
666 |
self._CheckRpcResult(result, self._nodes, |
667 |
"Moving %s to %s" % (old, new))
|
668 |
|
669 |
def _FormatJobID(self, job_id): |
670 |
"""Convert a job ID to string format.
|
671 |
|
672 |
Currently this just does C{str(job_id)} after performing some
|
673 |
checks, but if we want to change the job id format this will
|
674 |
abstract this change.
|
675 |
|
676 |
@type job_id: int or long
|
677 |
@param job_id: the numeric job id
|
678 |
@rtype: str
|
679 |
@return: the formatted job id
|
680 |
|
681 |
"""
|
682 |
if not isinstance(job_id, (int, long)): |
683 |
raise errors.ProgrammerError("Job ID '%s' not numeric" % job_id) |
684 |
if job_id < 0: |
685 |
raise errors.ProgrammerError("Job ID %s is negative" % job_id) |
686 |
|
687 |
return str(job_id) |
688 |
|
689 |
def _NewSerialUnlocked(self): |
690 |
"""Generates a new job identifier.
|
691 |
|
692 |
Job identifiers are unique during the lifetime of a cluster.
|
693 |
|
694 |
@rtype: str
|
695 |
@return: a string representing the job identifier.
|
696 |
|
697 |
"""
|
698 |
# New number
|
699 |
serial = self._last_serial + 1 |
700 |
|
701 |
# Write to file
|
702 |
self._WriteAndReplicateFileUnlocked(constants.JOB_QUEUE_SERIAL_FILE,
|
703 |
"%s\n" % serial)
|
704 |
|
705 |
# Keep it only if we were able to write the file
|
706 |
self._last_serial = serial
|
707 |
|
708 |
return self._FormatJobID(serial) |
709 |
|
710 |
@staticmethod
|
711 |
def _GetJobPath(job_id): |
712 |
"""Returns the job file for a given job id.
|
713 |
|
714 |
@type job_id: str
|
715 |
@param job_id: the job identifier
|
716 |
@rtype: str
|
717 |
@return: the path to the job file
|
718 |
|
719 |
"""
|
720 |
return os.path.join(constants.QUEUE_DIR, "job-%s" % job_id) |
721 |
|
722 |
@staticmethod
|
723 |
def _GetArchivedJobPath(job_id): |
724 |
"""Returns the archived job file for a give job id.
|
725 |
|
726 |
@type job_id: str
|
727 |
@param job_id: the job identifier
|
728 |
@rtype: str
|
729 |
@return: the path to the archived job file
|
730 |
|
731 |
"""
|
732 |
return os.path.join(constants.JOB_QUEUE_ARCHIVE_DIR, "job-%s" % job_id) |
733 |
|
734 |
@classmethod
|
735 |
def _ExtractJobID(cls, name): |
736 |
"""Extract the job id from a filename.
|
737 |
|
738 |
@type name: str
|
739 |
@param name: the job filename
|
740 |
@rtype: job id or None
|
741 |
@return: the job id corresponding to the given filename,
|
742 |
or None if the filename does not represent a valid
|
743 |
job file
|
744 |
|
745 |
"""
|
746 |
m = cls._RE_JOB_FILE.match(name) |
747 |
if m:
|
748 |
return m.group(1) |
749 |
else:
|
750 |
return None |
751 |
|
752 |
def _GetJobIDsUnlocked(self, archived=False): |
753 |
"""Return all known job IDs.
|
754 |
|
755 |
If the parameter archived is True, archived jobs IDs will be
|
756 |
included. Currently this argument is unused.
|
757 |
|
758 |
The method only looks at disk because it's a requirement that all
|
759 |
jobs are present on disk (so in the _memcache we don't have any
|
760 |
extra IDs).
|
761 |
|
762 |
@rtype: list
|
763 |
@return: the list of job IDs
|
764 |
|
765 |
"""
|
766 |
jlist = [self._ExtractJobID(name) for name in self._ListJobFiles()] |
767 |
jlist = utils.NiceSort(jlist) |
768 |
return jlist
|
769 |
|
770 |
def _ListJobFiles(self): |
771 |
"""Returns the list of current job files.
|
772 |
|
773 |
@rtype: list
|
774 |
@return: the list of job file names
|
775 |
|
776 |
"""
|
777 |
return [name for name in utils.ListVisibleFiles(constants.QUEUE_DIR) |
778 |
if self._RE_JOB_FILE.match(name)] |
779 |
|
780 |
def _LoadJobUnlocked(self, job_id): |
781 |
"""Loads a job from the disk or memory.
|
782 |
|
783 |
Given a job id, this will return the cached job object if
|
784 |
existing, or try to load the job from the disk. If loading from
|
785 |
disk, it will also add the job to the cache.
|
786 |
|
787 |
@param job_id: the job id
|
788 |
@rtype: L{_QueuedJob} or None
|
789 |
@return: either None or the job object
|
790 |
|
791 |
"""
|
792 |
job = self._memcache.get(job_id, None) |
793 |
if job:
|
794 |
logging.debug("Found job %s in memcache", job_id)
|
795 |
return job
|
796 |
|
797 |
filepath = self._GetJobPath(job_id)
|
798 |
logging.debug("Loading job from %s", filepath)
|
799 |
try:
|
800 |
fd = open(filepath, "r") |
801 |
except IOError, err: |
802 |
if err.errno in (errno.ENOENT, ): |
803 |
return None |
804 |
raise
|
805 |
try:
|
806 |
data = serializer.LoadJson(fd.read()) |
807 |
finally:
|
808 |
fd.close() |
809 |
|
810 |
try:
|
811 |
job = _QueuedJob.Restore(self, data)
|
812 |
except Exception, err: |
813 |
new_path = self._GetArchivedJobPath(job_id)
|
814 |
if filepath == new_path:
|
815 |
# job already archived (future case)
|
816 |
logging.exception("Can't parse job %s", job_id)
|
817 |
else:
|
818 |
# non-archived case
|
819 |
logging.exception("Can't parse job %s, will archive.", job_id)
|
820 |
self._RenameFileUnlocked(filepath, new_path)
|
821 |
return None |
822 |
|
823 |
self._memcache[job_id] = job
|
824 |
logging.debug("Added job %s to the cache", job_id)
|
825 |
return job
|
826 |
|
827 |
def _GetJobsUnlocked(self, job_ids): |
828 |
"""Return a list of jobs based on their IDs.
|
829 |
|
830 |
@type job_ids: list
|
831 |
@param job_ids: either an empty list (meaning all jobs),
|
832 |
or a list of job IDs
|
833 |
@rtype: list
|
834 |
@return: the list of job objects
|
835 |
|
836 |
"""
|
837 |
if not job_ids: |
838 |
job_ids = self._GetJobIDsUnlocked()
|
839 |
|
840 |
return [self._LoadJobUnlocked(job_id) for job_id in job_ids] |
841 |
|
842 |
@staticmethod
|
843 |
def _IsQueueMarkedDrain(): |
844 |
"""Check if the queue is marked from drain.
|
845 |
|
846 |
This currently uses the queue drain file, which makes it a
|
847 |
per-node flag. In the future this can be moved to the config file.
|
848 |
|
849 |
@rtype: boolean
|
850 |
@return: True of the job queue is marked for draining
|
851 |
|
852 |
"""
|
853 |
return os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
|
854 |
|
855 |
@staticmethod
|
856 |
def SetDrainFlag(drain_flag): |
857 |
"""Sets the drain flag for the queue.
|
858 |
|
859 |
This is similar to the function L{backend.JobQueueSetDrainFlag},
|
860 |
and in the future we might merge them.
|
861 |
|
862 |
@type drain_flag: boolean
|
863 |
@param drain_flag: wheter to set or unset the drain flag
|
864 |
|
865 |
"""
|
866 |
if drain_flag:
|
867 |
utils.WriteFile(constants.JOB_QUEUE_DRAIN_FILE, data="", close=True) |
868 |
else:
|
869 |
utils.RemoveFile(constants.JOB_QUEUE_DRAIN_FILE) |
870 |
return True |
871 |
|
872 |
@utils.LockedMethod
|
873 |
@_RequireOpenQueue
|
874 |
def SubmitJob(self, ops): |
875 |
"""Create and store a new job.
|
876 |
|
877 |
This enters the job into our job queue and also puts it on the new
|
878 |
queue, in order for it to be picked up by the queue processors.
|
879 |
|
880 |
@type ops: list
|
881 |
@param ops: The list of OpCodes that will become the new job.
|
882 |
@rtype: job ID
|
883 |
@return: the job ID of the newly created job
|
884 |
@raise errors.JobQueueDrainError: if the job is marked for draining
|
885 |
|
886 |
"""
|
887 |
if self._IsQueueMarkedDrain(): |
888 |
raise errors.JobQueueDrainError()
|
889 |
# Get job identifier
|
890 |
job_id = self._NewSerialUnlocked()
|
891 |
job = _QueuedJob(self, job_id, ops)
|
892 |
|
893 |
# Write to disk
|
894 |
self.UpdateJobUnlocked(job)
|
895 |
|
896 |
logging.debug("Adding new job %s to the cache", job_id)
|
897 |
self._memcache[job_id] = job
|
898 |
|
899 |
# Add to worker pool
|
900 |
self._wpool.AddTask(job)
|
901 |
|
902 |
return job.id
|
903 |
|
904 |
@_RequireOpenQueue
|
905 |
def UpdateJobUnlocked(self, job): |
906 |
"""Update a job's on disk storage.
|
907 |
|
908 |
After a job has been modified, this function needs to be called in
|
909 |
order to write the changes to disk and replicate them to the other
|
910 |
nodes.
|
911 |
|
912 |
@type job: L{_QueuedJob}
|
913 |
@param job: the changed job
|
914 |
|
915 |
"""
|
916 |
filename = self._GetJobPath(job.id)
|
917 |
data = serializer.DumpJson(job.Serialize(), indent=False)
|
918 |
logging.debug("Writing job %s to %s", job.id, filename)
|
919 |
self._WriteAndReplicateFileUnlocked(filename, data)
|
920 |
|
921 |
# Notify waiters about potential changes
|
922 |
job.change.notifyAll() |
923 |
|
924 |
@utils.LockedMethod
|
925 |
@_RequireOpenQueue
|
926 |
def WaitForJobChanges(self, job_id, fields, prev_job_info, prev_log_serial, |
927 |
timeout): |
928 |
"""Waits for changes in a job.
|
929 |
|
930 |
@type job_id: string
|
931 |
@param job_id: Job identifier
|
932 |
@type fields: list of strings
|
933 |
@param fields: Which fields to check for changes
|
934 |
@type prev_job_info: list or None
|
935 |
@param prev_job_info: Last job information returned
|
936 |
@type prev_log_serial: int
|
937 |
@param prev_log_serial: Last job message serial number
|
938 |
@type timeout: float
|
939 |
@param timeout: maximum time to wait
|
940 |
@rtype: tuple (job info, log entries)
|
941 |
@return: a tuple of the job information as required via
|
942 |
the fields parameter, and the log entries as a list
|
943 |
|
944 |
if the job has not changed and the timeout has expired,
|
945 |
we instead return a special value,
|
946 |
L{constants.JOB_NOTCHANGED}, which should be interpreted
|
947 |
as such by the clients
|
948 |
|
949 |
"""
|
950 |
logging.debug("Waiting for changes in job %s", job_id)
|
951 |
end_time = time.time() + timeout |
952 |
while True: |
953 |
delta_time = end_time - time.time() |
954 |
if delta_time < 0: |
955 |
return constants.JOB_NOTCHANGED
|
956 |
|
957 |
job = self._LoadJobUnlocked(job_id)
|
958 |
if not job: |
959 |
logging.debug("Job %s not found", job_id)
|
960 |
break
|
961 |
|
962 |
status = job.CalcStatus() |
963 |
job_info = self._GetJobInfoUnlocked(job, fields)
|
964 |
log_entries = job.GetLogEntries(prev_log_serial) |
965 |
|
966 |
# Serializing and deserializing data can cause type changes (e.g. from
|
967 |
# tuple to list) or precision loss. We're doing it here so that we get
|
968 |
# the same modifications as the data received from the client. Without
|
969 |
# this, the comparison afterwards might fail without the data being
|
970 |
# significantly different.
|
971 |
job_info = serializer.LoadJson(serializer.DumpJson(job_info)) |
972 |
log_entries = serializer.LoadJson(serializer.DumpJson(log_entries)) |
973 |
|
974 |
if status not in (constants.JOB_STATUS_QUEUED, |
975 |
constants.JOB_STATUS_RUNNING, |
976 |
constants.JOB_STATUS_WAITLOCK): |
977 |
# Don't even try to wait if the job is no longer running, there will be
|
978 |
# no changes.
|
979 |
break
|
980 |
|
981 |
if (prev_job_info != job_info or |
982 |
(log_entries and prev_log_serial != log_entries[0][0])): |
983 |
break
|
984 |
|
985 |
logging.debug("Waiting again")
|
986 |
|
987 |
# Release the queue lock while waiting
|
988 |
job.change.wait(delta_time) |
989 |
|
990 |
logging.debug("Job %s changed", job_id)
|
991 |
|
992 |
return (job_info, log_entries)
|
993 |
|
994 |
@utils.LockedMethod
|
995 |
@_RequireOpenQueue
|
996 |
def CancelJob(self, job_id): |
997 |
"""Cancels a job.
|
998 |
|
999 |
This will only succeed if the job has not started yet.
|
1000 |
|
1001 |
@type job_id: string
|
1002 |
@param job_id: job ID of job to be cancelled.
|
1003 |
|
1004 |
"""
|
1005 |
logging.debug("Cancelling job %s", job_id)
|
1006 |
|
1007 |
job = self._LoadJobUnlocked(job_id)
|
1008 |
if not job: |
1009 |
logging.debug("Job %s not found", job_id)
|
1010 |
return
|
1011 |
|
1012 |
if job.CalcStatus() not in (constants.JOB_STATUS_QUEUED,): |
1013 |
logging.debug("Job %s is no longer in the queue", job.id)
|
1014 |
return
|
1015 |
|
1016 |
try:
|
1017 |
for op in job.ops: |
1018 |
op.status = constants.OP_STATUS_ERROR |
1019 |
op.result = "Job cancelled by request"
|
1020 |
finally:
|
1021 |
self.UpdateJobUnlocked(job)
|
1022 |
|
1023 |
@_RequireOpenQueue
|
1024 |
def _ArchiveJobUnlocked(self, job_id): |
1025 |
"""Archives a job.
|
1026 |
|
1027 |
@type job_id: string
|
1028 |
@param job_id: Job ID of job to be archived.
|
1029 |
|
1030 |
"""
|
1031 |
logging.info("Archiving job %s", job_id)
|
1032 |
|
1033 |
job = self._LoadJobUnlocked(job_id)
|
1034 |
if not job: |
1035 |
logging.debug("Job %s not found", job_id)
|
1036 |
return
|
1037 |
|
1038 |
if job.CalcStatus() not in (constants.JOB_STATUS_CANCELED, |
1039 |
constants.JOB_STATUS_SUCCESS, |
1040 |
constants.JOB_STATUS_ERROR): |
1041 |
logging.debug("Job %s is not yet done", job.id)
|
1042 |
return
|
1043 |
|
1044 |
old = self._GetJobPath(job.id)
|
1045 |
new = self._GetArchivedJobPath(job.id)
|
1046 |
|
1047 |
self._RenameFileUnlocked(old, new)
|
1048 |
|
1049 |
logging.debug("Successfully archived job %s", job.id)
|
1050 |
|
1051 |
@utils.LockedMethod
|
1052 |
@_RequireOpenQueue
|
1053 |
def ArchiveJob(self, job_id): |
1054 |
"""Archives a job.
|
1055 |
|
1056 |
This is just a wrapper over L{_ArchiveJobUnlocked}.
|
1057 |
|
1058 |
@type job_id: string
|
1059 |
@param job_id: Job ID of job to be archived.
|
1060 |
|
1061 |
"""
|
1062 |
return self._ArchiveJobUnlocked(job_id) |
1063 |
|
1064 |
@utils.LockedMethod
|
1065 |
@_RequireOpenQueue
|
1066 |
def AutoArchiveJobs(self, age): |
1067 |
"""Archives all jobs based on age.
|
1068 |
|
1069 |
The method will archive all jobs which are older than the age
|
1070 |
parameter. For jobs that don't have an end timestamp, the start
|
1071 |
timestamp will be considered. The special '-1' age will cause
|
1072 |
archival of all jobs (that are not running or queued).
|
1073 |
|
1074 |
@type age: int
|
1075 |
@param age: the minimum age in seconds
|
1076 |
|
1077 |
"""
|
1078 |
logging.info("Archiving jobs with age more than %s seconds", age)
|
1079 |
|
1080 |
now = time.time() |
1081 |
for jid in self._GetJobIDsUnlocked(archived=False): |
1082 |
job = self._LoadJobUnlocked(jid)
|
1083 |
if job.CalcStatus() not in (constants.OP_STATUS_SUCCESS, |
1084 |
constants.OP_STATUS_ERROR, |
1085 |
constants.OP_STATUS_CANCELED): |
1086 |
continue
|
1087 |
if job.end_timestamp is None: |
1088 |
if job.start_timestamp is None: |
1089 |
job_age = job.received_timestamp |
1090 |
else:
|
1091 |
job_age = job.start_timestamp |
1092 |
else:
|
1093 |
job_age = job.end_timestamp |
1094 |
|
1095 |
if age == -1 or now - job_age[0] > age: |
1096 |
self._ArchiveJobUnlocked(jid)
|
1097 |
|
1098 |
def _GetJobInfoUnlocked(self, job, fields): |
1099 |
"""Returns information about a job.
|
1100 |
|
1101 |
@type job: L{_QueuedJob}
|
1102 |
@param job: the job which we query
|
1103 |
@type fields: list
|
1104 |
@param fields: names of fields to return
|
1105 |
@rtype: list
|
1106 |
@return: list with one element for each field
|
1107 |
@raise errors.OpExecError: when an invalid field
|
1108 |
has been passed
|
1109 |
|
1110 |
"""
|
1111 |
row = [] |
1112 |
for fname in fields: |
1113 |
if fname == "id": |
1114 |
row.append(job.id) |
1115 |
elif fname == "status": |
1116 |
row.append(job.CalcStatus()) |
1117 |
elif fname == "ops": |
1118 |
row.append([op.input.__getstate__() for op in job.ops]) |
1119 |
elif fname == "opresult": |
1120 |
row.append([op.result for op in job.ops]) |
1121 |
elif fname == "opstatus": |
1122 |
row.append([op.status for op in job.ops]) |
1123 |
elif fname == "oplog": |
1124 |
row.append([op.log for op in job.ops]) |
1125 |
elif fname == "opstart": |
1126 |
row.append([op.start_timestamp for op in job.ops]) |
1127 |
elif fname == "opend": |
1128 |
row.append([op.end_timestamp for op in job.ops]) |
1129 |
elif fname == "received_ts": |
1130 |
row.append(job.received_timestamp) |
1131 |
elif fname == "start_ts": |
1132 |
row.append(job.start_timestamp) |
1133 |
elif fname == "end_ts": |
1134 |
row.append(job.end_timestamp) |
1135 |
elif fname == "summary": |
1136 |
row.append([op.input.Summary() for op in job.ops]) |
1137 |
else:
|
1138 |
raise errors.OpExecError("Invalid job query field '%s'" % fname) |
1139 |
return row
|
1140 |
|
1141 |
@utils.LockedMethod
|
1142 |
@_RequireOpenQueue
|
1143 |
def QueryJobs(self, job_ids, fields): |
1144 |
"""Returns a list of jobs in queue.
|
1145 |
|
1146 |
This is a wrapper of L{_GetJobsUnlocked}, which actually does the
|
1147 |
processing for each job.
|
1148 |
|
1149 |
@type job_ids: list
|
1150 |
@param job_ids: sequence of job identifiers or None for all
|
1151 |
@type fields: list
|
1152 |
@param fields: names of fields to return
|
1153 |
@rtype: list
|
1154 |
@return: list one element per job, each element being list with
|
1155 |
the requested fields
|
1156 |
|
1157 |
"""
|
1158 |
jobs = [] |
1159 |
|
1160 |
for job in self._GetJobsUnlocked(job_ids): |
1161 |
if job is None: |
1162 |
jobs.append(None)
|
1163 |
else:
|
1164 |
jobs.append(self._GetJobInfoUnlocked(job, fields))
|
1165 |
|
1166 |
return jobs
|
1167 |
|
1168 |
@utils.LockedMethod
|
1169 |
@_RequireOpenQueue
|
1170 |
def Shutdown(self): |
1171 |
"""Stops the job queue.
|
1172 |
|
1173 |
This shutdowns all the worker threads an closes the queue.
|
1174 |
|
1175 |
"""
|
1176 |
self._wpool.TerminateWorkers()
|
1177 |
|
1178 |
self._queue_lock.Close()
|
1179 |
self._queue_lock = None |