root / lib / jqueue.py @ 21cc1fbd
History | View | Annotate | Download (15.8 kB)
1 | 498ae1cc | Iustin Pop | #
|
---|---|---|---|
2 | 498ae1cc | Iustin Pop | #
|
3 | 498ae1cc | Iustin Pop | |
4 | 498ae1cc | Iustin Pop | # Copyright (C) 2006, 2007 Google Inc.
|
5 | 498ae1cc | Iustin Pop | #
|
6 | 498ae1cc | Iustin Pop | # This program is free software; you can redistribute it and/or modify
|
7 | 498ae1cc | Iustin Pop | # it under the terms of the GNU General Public License as published by
|
8 | 498ae1cc | Iustin Pop | # the Free Software Foundation; either version 2 of the License, or
|
9 | 498ae1cc | Iustin Pop | # (at your option) any later version.
|
10 | 498ae1cc | Iustin Pop | #
|
11 | 498ae1cc | Iustin Pop | # This program is distributed in the hope that it will be useful, but
|
12 | 498ae1cc | Iustin Pop | # WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 | 498ae1cc | Iustin Pop | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
14 | 498ae1cc | Iustin Pop | # General Public License for more details.
|
15 | 498ae1cc | Iustin Pop | #
|
16 | 498ae1cc | Iustin Pop | # You should have received a copy of the GNU General Public License
|
17 | 498ae1cc | Iustin Pop | # along with this program; if not, write to the Free Software
|
18 | 498ae1cc | Iustin Pop | # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
|
19 | 498ae1cc | Iustin Pop | # 02110-1301, USA.
|
20 | 498ae1cc | Iustin Pop | |
21 | 498ae1cc | Iustin Pop | |
22 | 498ae1cc | Iustin Pop | """Module implementing the job queue handling."""
|
23 | 498ae1cc | Iustin Pop | |
24 | f1da30e6 | Michael Hanselmann | import os |
25 | e2715f69 | Michael Hanselmann | import logging |
26 | e2715f69 | Michael Hanselmann | import threading |
27 | f1da30e6 | Michael Hanselmann | import errno |
28 | f1da30e6 | Michael Hanselmann | import re |
29 | f1048938 | Iustin Pop | import time |
30 | 498ae1cc | Iustin Pop | |
31 | e2715f69 | Michael Hanselmann | from ganeti import constants |
32 | f1da30e6 | Michael Hanselmann | from ganeti import serializer |
33 | e2715f69 | Michael Hanselmann | from ganeti import workerpool |
34 | f1da30e6 | Michael Hanselmann | from ganeti import opcodes |
35 | 7a1ecaed | Iustin Pop | from ganeti import errors |
36 | e2715f69 | Michael Hanselmann | from ganeti import mcpu |
37 | 7996a135 | Iustin Pop | from ganeti import utils |
38 | c3f0a12f | Iustin Pop | from ganeti import rpc |
39 | e2715f69 | Michael Hanselmann | |
40 | e2715f69 | Michael Hanselmann | |
41 | e2715f69 | Michael Hanselmann | JOBQUEUE_THREADS = 5
|
42 | e2715f69 | Michael Hanselmann | |
43 | 498ae1cc | Iustin Pop | |
44 | e2715f69 | Michael Hanselmann | class _QueuedOpCode(object): |
45 | e2715f69 | Michael Hanselmann | """Encasulates an opcode object.
|
46 | e2715f69 | Michael Hanselmann |
|
47 | 307149a8 | Iustin Pop | Access is synchronized by the '_lock' attribute.
|
48 | e2715f69 | Michael Hanselmann |
|
49 | f1048938 | Iustin Pop | The 'log' attribute holds the execution log and consists of tuples
|
50 | f1048938 | Iustin Pop | of the form (timestamp, level, message).
|
51 | f1048938 | Iustin Pop |
|
52 | e2715f69 | Michael Hanselmann | """
|
53 | e2715f69 | Michael Hanselmann | def __init__(self, op): |
54 | f1048938 | Iustin Pop | self.__Setup(op, constants.OP_STATUS_QUEUED, None, []) |
55 | f1da30e6 | Michael Hanselmann | |
56 | f1048938 | Iustin Pop | def __Setup(self, input_, status, result, log): |
57 | 307149a8 | Iustin Pop | self._lock = threading.Lock()
|
58 | f1048938 | Iustin Pop | self.input = input_
|
59 | f1da30e6 | Michael Hanselmann | self.status = status
|
60 | f1da30e6 | Michael Hanselmann | self.result = result
|
61 | f1048938 | Iustin Pop | self.log = log
|
62 | f1da30e6 | Michael Hanselmann | |
63 | f1da30e6 | Michael Hanselmann | @classmethod
|
64 | f1da30e6 | Michael Hanselmann | def Restore(cls, state): |
65 | f1da30e6 | Michael Hanselmann | obj = object.__new__(cls)
|
66 | f1da30e6 | Michael Hanselmann | obj.__Setup(opcodes.OpCode.LoadOpCode(state["input"]),
|
67 | f1048938 | Iustin Pop | state["status"], state["result"], state["log"]) |
68 | f1da30e6 | Michael Hanselmann | return obj
|
69 | f1da30e6 | Michael Hanselmann | |
70 | f1da30e6 | Michael Hanselmann | @utils.LockedMethod
|
71 | f1da30e6 | Michael Hanselmann | def Serialize(self): |
72 | f1da30e6 | Michael Hanselmann | return {
|
73 | f1da30e6 | Michael Hanselmann | "input": self.input.__getstate__(), |
74 | f1da30e6 | Michael Hanselmann | "status": self.status, |
75 | f1da30e6 | Michael Hanselmann | "result": self.result, |
76 | f1048938 | Iustin Pop | "log": self.log, |
77 | f1da30e6 | Michael Hanselmann | } |
78 | 307149a8 | Iustin Pop | |
79 | 307149a8 | Iustin Pop | @utils.LockedMethod
|
80 | af30b2fd | Michael Hanselmann | def GetInput(self): |
81 | af30b2fd | Michael Hanselmann | """Returns the original opcode.
|
82 | af30b2fd | Michael Hanselmann |
|
83 | af30b2fd | Michael Hanselmann | """
|
84 | af30b2fd | Michael Hanselmann | return self.input |
85 | af30b2fd | Michael Hanselmann | |
86 | af30b2fd | Michael Hanselmann | @utils.LockedMethod
|
87 | 307149a8 | Iustin Pop | def SetStatus(self, status, result): |
88 | 307149a8 | Iustin Pop | """Update the opcode status and result.
|
89 | 307149a8 | Iustin Pop |
|
90 | 307149a8 | Iustin Pop | """
|
91 | 307149a8 | Iustin Pop | self.status = status
|
92 | 307149a8 | Iustin Pop | self.result = result
|
93 | 307149a8 | Iustin Pop | |
94 | 307149a8 | Iustin Pop | @utils.LockedMethod
|
95 | 307149a8 | Iustin Pop | def GetStatus(self): |
96 | 307149a8 | Iustin Pop | """Get the opcode status.
|
97 | 307149a8 | Iustin Pop |
|
98 | 307149a8 | Iustin Pop | """
|
99 | 307149a8 | Iustin Pop | return self.status |
100 | 307149a8 | Iustin Pop | |
101 | 307149a8 | Iustin Pop | @utils.LockedMethod
|
102 | 307149a8 | Iustin Pop | def GetResult(self): |
103 | 307149a8 | Iustin Pop | """Get the opcode result.
|
104 | 307149a8 | Iustin Pop |
|
105 | 307149a8 | Iustin Pop | """
|
106 | 307149a8 | Iustin Pop | return self.result |
107 | e2715f69 | Michael Hanselmann | |
108 | f1048938 | Iustin Pop | @utils.LockedMethod
|
109 | f1048938 | Iustin Pop | def Log(self, *args): |
110 | f1048938 | Iustin Pop | """Append a log entry.
|
111 | f1048938 | Iustin Pop |
|
112 | f1048938 | Iustin Pop | """
|
113 | f1048938 | Iustin Pop | assert len(args) < 2 |
114 | f1048938 | Iustin Pop | |
115 | f1048938 | Iustin Pop | if len(args) == 1: |
116 | f1048938 | Iustin Pop | log_type = constants.ELOG_MESSAGE |
117 | f1048938 | Iustin Pop | log_msg = args[0]
|
118 | f1048938 | Iustin Pop | else:
|
119 | f1048938 | Iustin Pop | log_type, log_msg = args |
120 | f1048938 | Iustin Pop | self.log.append((time.time(), log_type, log_msg))
|
121 | f1048938 | Iustin Pop | |
122 | f1048938 | Iustin Pop | @utils.LockedMethod
|
123 | f1048938 | Iustin Pop | def RetrieveLog(self, start_at=0): |
124 | f1048938 | Iustin Pop | """Retrieve (a part of) the execution log.
|
125 | f1048938 | Iustin Pop |
|
126 | f1048938 | Iustin Pop | """
|
127 | f1048938 | Iustin Pop | return self.log[start_at:] |
128 | f1048938 | Iustin Pop | |
129 | e2715f69 | Michael Hanselmann | |
130 | e2715f69 | Michael Hanselmann | class _QueuedJob(object): |
131 | e2715f69 | Michael Hanselmann | """In-memory job representation.
|
132 | e2715f69 | Michael Hanselmann |
|
133 | e2715f69 | Michael Hanselmann | This is what we use to track the user-submitted jobs.
|
134 | e2715f69 | Michael Hanselmann |
|
135 | e2715f69 | Michael Hanselmann | """
|
136 | f1da30e6 | Michael Hanselmann | def __init__(self, storage, job_id, ops): |
137 | e2715f69 | Michael Hanselmann | if not ops: |
138 | e2715f69 | Michael Hanselmann | # TODO
|
139 | e2715f69 | Michael Hanselmann | raise Exception("No opcodes") |
140 | e2715f69 | Michael Hanselmann | |
141 | f1048938 | Iustin Pop | self.__Setup(storage, job_id, [_QueuedOpCode(op) for op in ops], -1) |
142 | e2715f69 | Michael Hanselmann | |
143 | f1048938 | Iustin Pop | def __Setup(self, storage, job_id, ops, run_op_index): |
144 | f1048938 | Iustin Pop | self._lock = threading.Lock()
|
145 | f1da30e6 | Michael Hanselmann | self.storage = storage
|
146 | f1da30e6 | Michael Hanselmann | self.id = job_id
|
147 | f1da30e6 | Michael Hanselmann | self._ops = ops
|
148 | f1048938 | Iustin Pop | self.run_op_index = run_op_index
|
149 | f1da30e6 | Michael Hanselmann | |
150 | f1da30e6 | Michael Hanselmann | @classmethod
|
151 | f1da30e6 | Michael Hanselmann | def Restore(cls, storage, state): |
152 | f1da30e6 | Michael Hanselmann | obj = object.__new__(cls)
|
153 | f1048938 | Iustin Pop | op_list = [_QueuedOpCode.Restore(op_state) for op_state in state["ops"]] |
154 | f1048938 | Iustin Pop | obj.__Setup(storage, state["id"], op_list, state["run_op_index"]) |
155 | f1da30e6 | Michael Hanselmann | return obj
|
156 | f1da30e6 | Michael Hanselmann | |
157 | f1da30e6 | Michael Hanselmann | def Serialize(self): |
158 | f1da30e6 | Michael Hanselmann | return {
|
159 | f1da30e6 | Michael Hanselmann | "id": self.id, |
160 | f1da30e6 | Michael Hanselmann | "ops": [op.Serialize() for op in self._ops], |
161 | f1048938 | Iustin Pop | "run_op_index": self.run_op_index, |
162 | f1da30e6 | Michael Hanselmann | } |
163 | f1da30e6 | Michael Hanselmann | |
164 | f1da30e6 | Michael Hanselmann | def SetUnclean(self, msg): |
165 | f1da30e6 | Michael Hanselmann | try:
|
166 | f1da30e6 | Michael Hanselmann | for op in self._ops: |
167 | f1da30e6 | Michael Hanselmann | op.SetStatus(constants.OP_STATUS_ERROR, msg) |
168 | f1da30e6 | Michael Hanselmann | finally:
|
169 | f1da30e6 | Michael Hanselmann | self.storage.UpdateJob(self) |
170 | e2715f69 | Michael Hanselmann | |
171 | 307149a8 | Iustin Pop | def GetStatus(self): |
172 | e2715f69 | Michael Hanselmann | status = constants.JOB_STATUS_QUEUED |
173 | e2715f69 | Michael Hanselmann | |
174 | e2715f69 | Michael Hanselmann | all_success = True
|
175 | e2715f69 | Michael Hanselmann | for op in self._ops: |
176 | 307149a8 | Iustin Pop | op_status = op.GetStatus() |
177 | 307149a8 | Iustin Pop | if op_status == constants.OP_STATUS_SUCCESS:
|
178 | e2715f69 | Michael Hanselmann | continue
|
179 | e2715f69 | Michael Hanselmann | |
180 | e2715f69 | Michael Hanselmann | all_success = False
|
181 | e2715f69 | Michael Hanselmann | |
182 | 307149a8 | Iustin Pop | if op_status == constants.OP_STATUS_QUEUED:
|
183 | e2715f69 | Michael Hanselmann | pass
|
184 | 307149a8 | Iustin Pop | elif op_status == constants.OP_STATUS_RUNNING:
|
185 | e2715f69 | Michael Hanselmann | status = constants.JOB_STATUS_RUNNING |
186 | f1da30e6 | Michael Hanselmann | elif op_status == constants.OP_STATUS_ERROR:
|
187 | f1da30e6 | Michael Hanselmann | status = constants.JOB_STATUS_ERROR |
188 | f1da30e6 | Michael Hanselmann | # The whole job fails if one opcode failed
|
189 | f1da30e6 | Michael Hanselmann | break
|
190 | e2715f69 | Michael Hanselmann | |
191 | e2715f69 | Michael Hanselmann | if all_success:
|
192 | e2715f69 | Michael Hanselmann | status = constants.JOB_STATUS_SUCCESS |
193 | e2715f69 | Michael Hanselmann | |
194 | e2715f69 | Michael Hanselmann | return status
|
195 | e2715f69 | Michael Hanselmann | |
196 | f1048938 | Iustin Pop | @utils.LockedMethod
|
197 | f1048938 | Iustin Pop | def GetRunOpIndex(self): |
198 | f1048938 | Iustin Pop | return self.run_op_index |
199 | f1048938 | Iustin Pop | |
200 | e2715f69 | Michael Hanselmann | def Run(self, proc): |
201 | e2715f69 | Michael Hanselmann | """Job executor.
|
202 | e2715f69 | Michael Hanselmann |
|
203 | e2715f69 | Michael Hanselmann | This functions processes a this job in the context of given processor
|
204 | e2715f69 | Michael Hanselmann | instance.
|
205 | e2715f69 | Michael Hanselmann |
|
206 | e2715f69 | Michael Hanselmann | Args:
|
207 | e2715f69 | Michael Hanselmann | - proc: Ganeti Processor to run the job with
|
208 | e2715f69 | Michael Hanselmann |
|
209 | e2715f69 | Michael Hanselmann | """
|
210 | e2715f69 | Michael Hanselmann | try:
|
211 | c8549bfd | Michael Hanselmann | count = len(self._ops) |
212 | c8549bfd | Michael Hanselmann | for idx, op in enumerate(self._ops): |
213 | e2715f69 | Michael Hanselmann | try:
|
214 | 307149a8 | Iustin Pop | logging.debug("Op %s/%s: Starting %s", idx + 1, count, op) |
215 | f1048938 | Iustin Pop | |
216 | f1048938 | Iustin Pop | self._lock.acquire()
|
217 | f1048938 | Iustin Pop | try:
|
218 | f1048938 | Iustin Pop | self.run_op_index = idx
|
219 | f1048938 | Iustin Pop | finally:
|
220 | f1048938 | Iustin Pop | self._lock.release()
|
221 | f1048938 | Iustin Pop | |
222 | 307149a8 | Iustin Pop | op.SetStatus(constants.OP_STATUS_RUNNING, None)
|
223 | f1da30e6 | Michael Hanselmann | self.storage.UpdateJob(self) |
224 | e2715f69 | Michael Hanselmann | |
225 | f1048938 | Iustin Pop | result = proc.ExecOpCode(op.input, op.Log) |
226 | e2715f69 | Michael Hanselmann | |
227 | 307149a8 | Iustin Pop | op.SetStatus(constants.OP_STATUS_SUCCESS, result) |
228 | f1da30e6 | Michael Hanselmann | self.storage.UpdateJob(self) |
229 | 307149a8 | Iustin Pop | logging.debug("Op %s/%s: Successfully finished %s",
|
230 | 307149a8 | Iustin Pop | idx + 1, count, op)
|
231 | e2715f69 | Michael Hanselmann | except Exception, err: |
232 | f1da30e6 | Michael Hanselmann | try:
|
233 | f1da30e6 | Michael Hanselmann | op.SetStatus(constants.OP_STATUS_ERROR, str(err))
|
234 | f1da30e6 | Michael Hanselmann | logging.debug("Op %s/%s: Error in %s", idx + 1, count, op) |
235 | f1da30e6 | Michael Hanselmann | finally:
|
236 | f1da30e6 | Michael Hanselmann | self.storage.UpdateJob(self) |
237 | e2715f69 | Michael Hanselmann | raise
|
238 | e2715f69 | Michael Hanselmann | |
239 | e2715f69 | Michael Hanselmann | except errors.GenericError, err:
|
240 | e2715f69 | Michael Hanselmann | logging.error("ganeti exception %s", exc_info=err)
|
241 | e2715f69 | Michael Hanselmann | except Exception, err: |
242 | e2715f69 | Michael Hanselmann | logging.error("unhandled exception %s", exc_info=err)
|
243 | e2715f69 | Michael Hanselmann | except:
|
244 | e2715f69 | Michael Hanselmann | logging.error("unhandled unknown exception %s", exc_info=err)
|
245 | e2715f69 | Michael Hanselmann | |
246 | e2715f69 | Michael Hanselmann | |
247 | e2715f69 | Michael Hanselmann | class _JobQueueWorker(workerpool.BaseWorker): |
248 | e2715f69 | Michael Hanselmann | def RunTask(self, job): |
249 | e2715f69 | Michael Hanselmann | logging.debug("Worker %s processing job %s",
|
250 | e2715f69 | Michael Hanselmann | self.worker_id, job.id)
|
251 | e2715f69 | Michael Hanselmann | # TODO: feedback function
|
252 | f1048938 | Iustin Pop | proc = mcpu.Processor(self.pool.context)
|
253 | e2715f69 | Michael Hanselmann | try:
|
254 | e2715f69 | Michael Hanselmann | job.Run(proc) |
255 | e2715f69 | Michael Hanselmann | finally:
|
256 | e2715f69 | Michael Hanselmann | logging.debug("Worker %s finished job %s, status = %s",
|
257 | e2715f69 | Michael Hanselmann | self.worker_id, job.id, job.GetStatus())
|
258 | e2715f69 | Michael Hanselmann | |
259 | e2715f69 | Michael Hanselmann | |
260 | e2715f69 | Michael Hanselmann | class _JobQueueWorkerPool(workerpool.WorkerPool): |
261 | e2715f69 | Michael Hanselmann | def __init__(self, context): |
262 | e2715f69 | Michael Hanselmann | super(_JobQueueWorkerPool, self).__init__(JOBQUEUE_THREADS, |
263 | e2715f69 | Michael Hanselmann | _JobQueueWorker) |
264 | e2715f69 | Michael Hanselmann | self.context = context
|
265 | e2715f69 | Michael Hanselmann | |
266 | e2715f69 | Michael Hanselmann | |
267 | 21cc1fbd | Michael Hanselmann | class DiskJobStorage(object): |
268 | bac5ffc3 | Oleksiy Mishchenko | _RE_JOB_FILE = re.compile(r"^job-(%s)$" % constants.JOB_ID_TEMPLATE)
|
269 | f1da30e6 | Michael Hanselmann | |
270 | f1da30e6 | Michael Hanselmann | def __init__(self): |
271 | f1da30e6 | Michael Hanselmann | self._lock = threading.Lock()
|
272 | ac0930b9 | Iustin Pop | self._memcache = {}
|
273 | c3f0a12f | Iustin Pop | self._my_hostname = utils.HostInfo().name
|
274 | f1da30e6 | Michael Hanselmann | |
275 | f1da30e6 | Michael Hanselmann | # Make sure our directory exists
|
276 | f1da30e6 | Michael Hanselmann | try:
|
277 | f1da30e6 | Michael Hanselmann | os.mkdir(constants.QUEUE_DIR, 0700)
|
278 | f1da30e6 | Michael Hanselmann | except OSError, err: |
279 | f1da30e6 | Michael Hanselmann | if err.errno not in (errno.EEXIST, ): |
280 | f1da30e6 | Michael Hanselmann | raise
|
281 | f1da30e6 | Michael Hanselmann | |
282 | f1da30e6 | Michael Hanselmann | # Get queue lock
|
283 | f1da30e6 | Michael Hanselmann | self.lock_fd = open(constants.JOB_QUEUE_LOCK_FILE, "w") |
284 | f1da30e6 | Michael Hanselmann | try:
|
285 | f1da30e6 | Michael Hanselmann | utils.LockFile(self.lock_fd)
|
286 | f1da30e6 | Michael Hanselmann | except:
|
287 | f1da30e6 | Michael Hanselmann | self.lock_fd.close()
|
288 | f1da30e6 | Michael Hanselmann | raise
|
289 | f1da30e6 | Michael Hanselmann | |
290 | f1da30e6 | Michael Hanselmann | # Read version
|
291 | f1da30e6 | Michael Hanselmann | try:
|
292 | f1da30e6 | Michael Hanselmann | version_fd = open(constants.JOB_QUEUE_VERSION_FILE, "r") |
293 | f1da30e6 | Michael Hanselmann | except IOError, err: |
294 | f1da30e6 | Michael Hanselmann | if err.errno not in (errno.ENOENT, ): |
295 | f1da30e6 | Michael Hanselmann | raise
|
296 | f1da30e6 | Michael Hanselmann | |
297 | f1da30e6 | Michael Hanselmann | # Setup a new queue
|
298 | f1da30e6 | Michael Hanselmann | self._InitQueueUnlocked()
|
299 | f1da30e6 | Michael Hanselmann | |
300 | f1da30e6 | Michael Hanselmann | # Try to open again
|
301 | f1da30e6 | Michael Hanselmann | version_fd = open(constants.JOB_QUEUE_VERSION_FILE, "r") |
302 | f1da30e6 | Michael Hanselmann | |
303 | f1da30e6 | Michael Hanselmann | try:
|
304 | f1da30e6 | Michael Hanselmann | # Try to read version
|
305 | f1da30e6 | Michael Hanselmann | version = int(version_fd.read(128)) |
306 | f1da30e6 | Michael Hanselmann | |
307 | f1da30e6 | Michael Hanselmann | # Verify version
|
308 | f1da30e6 | Michael Hanselmann | if version != constants.JOB_QUEUE_VERSION:
|
309 | f1da30e6 | Michael Hanselmann | raise errors.JobQueueError("Found version %s, expected %s", |
310 | f1da30e6 | Michael Hanselmann | version, constants.JOB_QUEUE_VERSION) |
311 | f1da30e6 | Michael Hanselmann | finally:
|
312 | f1da30e6 | Michael Hanselmann | version_fd.close() |
313 | f1da30e6 | Michael Hanselmann | |
314 | c4beba1c | Iustin Pop | self._last_serial = self._ReadSerial() |
315 | c4beba1c | Iustin Pop | if self._last_serial is None: |
316 | c4beba1c | Iustin Pop | raise errors.ConfigurationError("Can't read/parse the job queue serial" |
317 | c4beba1c | Iustin Pop | " file")
|
318 | c4beba1c | Iustin Pop | |
319 | c4beba1c | Iustin Pop | @staticmethod
|
320 | c4beba1c | Iustin Pop | def _ReadSerial(): |
321 | c4beba1c | Iustin Pop | """Try to read the job serial file.
|
322 | c4beba1c | Iustin Pop |
|
323 | c4beba1c | Iustin Pop | @rtype: None or int
|
324 | c4beba1c | Iustin Pop | @return: If the serial can be read, then it is returned. Otherwise None
|
325 | c4beba1c | Iustin Pop | is returned.
|
326 | c4beba1c | Iustin Pop |
|
327 | c4beba1c | Iustin Pop | """
|
328 | f1da30e6 | Michael Hanselmann | try:
|
329 | c4beba1c | Iustin Pop | serial_fd = open(constants.JOB_QUEUE_SERIAL_FILE, "r") |
330 | c4beba1c | Iustin Pop | try:
|
331 | c4beba1c | Iustin Pop | # Read last serial
|
332 | c4beba1c | Iustin Pop | serial = int(serial_fd.read(1024).strip()) |
333 | c4beba1c | Iustin Pop | finally:
|
334 | c4beba1c | Iustin Pop | serial_fd.close() |
335 | c4beba1c | Iustin Pop | except (ValueError, EnvironmentError): |
336 | c4beba1c | Iustin Pop | serial = None
|
337 | c4beba1c | Iustin Pop | |
338 | c4beba1c | Iustin Pop | return serial
|
339 | f1da30e6 | Michael Hanselmann | |
340 | f1da30e6 | Michael Hanselmann | def Close(self): |
341 | f1da30e6 | Michael Hanselmann | assert self.lock_fd, "Queue should be open" |
342 | f1da30e6 | Michael Hanselmann | |
343 | f1da30e6 | Michael Hanselmann | self.lock_fd.close()
|
344 | f1da30e6 | Michael Hanselmann | self.lock_fd = None |
345 | f1da30e6 | Michael Hanselmann | |
346 | f1da30e6 | Michael Hanselmann | def _InitQueueUnlocked(self): |
347 | f1da30e6 | Michael Hanselmann | assert self.lock_fd, "Queue should be open" |
348 | f1da30e6 | Michael Hanselmann | |
349 | f1da30e6 | Michael Hanselmann | utils.WriteFile(constants.JOB_QUEUE_VERSION_FILE, |
350 | f1da30e6 | Michael Hanselmann | data="%s\n" % constants.JOB_QUEUE_VERSION)
|
351 | c4beba1c | Iustin Pop | if self._ReadSerial() is None: |
352 | c4beba1c | Iustin Pop | utils.WriteFile(constants.JOB_QUEUE_SERIAL_FILE, |
353 | c4beba1c | Iustin Pop | data="%s\n" % 0) |
354 | f1da30e6 | Michael Hanselmann | |
355 | c3f0a12f | Iustin Pop | def _NewSerialUnlocked(self, nodes): |
356 | f1da30e6 | Michael Hanselmann | """Generates a new job identifier.
|
357 | f1da30e6 | Michael Hanselmann |
|
358 | f1da30e6 | Michael Hanselmann | Job identifiers are unique during the lifetime of a cluster.
|
359 | f1da30e6 | Michael Hanselmann |
|
360 | f1da30e6 | Michael Hanselmann | Returns: A string representing the job identifier.
|
361 | f1da30e6 | Michael Hanselmann |
|
362 | f1da30e6 | Michael Hanselmann | """
|
363 | f1da30e6 | Michael Hanselmann | assert self.lock_fd, "Queue should be open" |
364 | f1da30e6 | Michael Hanselmann | |
365 | f1da30e6 | Michael Hanselmann | # New number
|
366 | f1da30e6 | Michael Hanselmann | serial = self._last_serial + 1 |
367 | f1da30e6 | Michael Hanselmann | |
368 | f1da30e6 | Michael Hanselmann | # Write to file
|
369 | f1da30e6 | Michael Hanselmann | utils.WriteFile(constants.JOB_QUEUE_SERIAL_FILE, |
370 | f1da30e6 | Michael Hanselmann | data="%s\n" % serial)
|
371 | f1da30e6 | Michael Hanselmann | |
372 | f1da30e6 | Michael Hanselmann | # Keep it only if we were able to write the file
|
373 | f1da30e6 | Michael Hanselmann | self._last_serial = serial
|
374 | f1da30e6 | Michael Hanselmann | |
375 | c3f0a12f | Iustin Pop | # Distribute the serial to the other nodes
|
376 | c3f0a12f | Iustin Pop | try:
|
377 | c3f0a12f | Iustin Pop | nodes.remove(self._my_hostname)
|
378 | c3f0a12f | Iustin Pop | except ValueError: |
379 | c3f0a12f | Iustin Pop | pass
|
380 | c3f0a12f | Iustin Pop | |
381 | c3f0a12f | Iustin Pop | result = rpc.call_upload_file(nodes, constants.JOB_QUEUE_SERIAL_FILE) |
382 | c3f0a12f | Iustin Pop | for node in nodes: |
383 | c3f0a12f | Iustin Pop | if not result[node]: |
384 | c3f0a12f | Iustin Pop | logging.error("copy of job queue file to node %s failed", node)
|
385 | c3f0a12f | Iustin Pop | |
386 | 3be9a705 | Michael Hanselmann | return str(serial) |
387 | f1da30e6 | Michael Hanselmann | |
388 | f1da30e6 | Michael Hanselmann | def _GetJobPath(self, job_id): |
389 | f1da30e6 | Michael Hanselmann | return os.path.join(constants.QUEUE_DIR, "job-%s" % job_id) |
390 | f1da30e6 | Michael Hanselmann | |
391 | 911a495b | Iustin Pop | def _GetJobIDsUnlocked(self, archived=False): |
392 | 911a495b | Iustin Pop | """Return all known job IDs.
|
393 | 911a495b | Iustin Pop |
|
394 | 911a495b | Iustin Pop | If the parameter archived is True, archived jobs IDs will be
|
395 | 911a495b | Iustin Pop | included. Currently this argument is unused.
|
396 | 911a495b | Iustin Pop |
|
397 | ac0930b9 | Iustin Pop | The method only looks at disk because it's a requirement that all
|
398 | ac0930b9 | Iustin Pop | jobs are present on disk (so in the _memcache we don't have any
|
399 | ac0930b9 | Iustin Pop | extra IDs).
|
400 | ac0930b9 | Iustin Pop |
|
401 | 911a495b | Iustin Pop | """
|
402 | 911a495b | Iustin Pop | jfiles = self._ListJobFiles()
|
403 | f0d874fe | Iustin Pop | jlist = [int(m.group(1)) for m in |
404 | f0d874fe | Iustin Pop | [self._RE_JOB_FILE.match(name) for name in jfiles]] |
405 | f0d874fe | Iustin Pop | jlist.sort() |
406 | f0d874fe | Iustin Pop | return jlist
|
407 | 911a495b | Iustin Pop | |
408 | f1da30e6 | Michael Hanselmann | def _ListJobFiles(self): |
409 | f1da30e6 | Michael Hanselmann | assert self.lock_fd, "Queue should be open" |
410 | f1da30e6 | Michael Hanselmann | |
411 | f1da30e6 | Michael Hanselmann | return [name for name in utils.ListVisibleFiles(constants.QUEUE_DIR) |
412 | f1da30e6 | Michael Hanselmann | if self._RE_JOB_FILE.match(name)] |
413 | f1da30e6 | Michael Hanselmann | |
414 | 911a495b | Iustin Pop | def _LoadJobUnlocked(self, job_id): |
415 | f1da30e6 | Michael Hanselmann | assert self.lock_fd, "Queue should be open" |
416 | f1da30e6 | Michael Hanselmann | |
417 | ac0930b9 | Iustin Pop | if job_id in self._memcache: |
418 | 205d71fd | Michael Hanselmann | logging.debug("Found job %s in memcache", job_id)
|
419 | ac0930b9 | Iustin Pop | return self._memcache[job_id] |
420 | ac0930b9 | Iustin Pop | |
421 | 911a495b | Iustin Pop | filepath = self._GetJobPath(job_id)
|
422 | f1da30e6 | Michael Hanselmann | logging.debug("Loading job from %s", filepath)
|
423 | f1da30e6 | Michael Hanselmann | try:
|
424 | f1da30e6 | Michael Hanselmann | fd = open(filepath, "r") |
425 | f1da30e6 | Michael Hanselmann | except IOError, err: |
426 | f1da30e6 | Michael Hanselmann | if err.errno in (errno.ENOENT, ): |
427 | f1da30e6 | Michael Hanselmann | return None |
428 | f1da30e6 | Michael Hanselmann | raise
|
429 | f1da30e6 | Michael Hanselmann | try:
|
430 | f1da30e6 | Michael Hanselmann | data = serializer.LoadJson(fd.read()) |
431 | f1da30e6 | Michael Hanselmann | finally:
|
432 | f1da30e6 | Michael Hanselmann | fd.close() |
433 | f1da30e6 | Michael Hanselmann | |
434 | ac0930b9 | Iustin Pop | job = _QueuedJob.Restore(self, data)
|
435 | ac0930b9 | Iustin Pop | self._memcache[job_id] = job
|
436 | 205d71fd | Michael Hanselmann | logging.debug("Added job %s to the cache", job_id)
|
437 | ac0930b9 | Iustin Pop | return job
|
438 | f1da30e6 | Michael Hanselmann | |
439 | f1da30e6 | Michael Hanselmann | def _GetJobsUnlocked(self, job_ids): |
440 | 911a495b | Iustin Pop | if not job_ids: |
441 | 911a495b | Iustin Pop | job_ids = self._GetJobIDsUnlocked()
|
442 | f1da30e6 | Michael Hanselmann | |
443 | 911a495b | Iustin Pop | return [self._LoadJobUnlocked(job_id) for job_id in job_ids] |
444 | f1da30e6 | Michael Hanselmann | |
445 | f1da30e6 | Michael Hanselmann | @utils.LockedMethod
|
446 | f1da30e6 | Michael Hanselmann | def GetJobs(self, job_ids): |
447 | f1da30e6 | Michael Hanselmann | return self._GetJobsUnlocked(job_ids) |
448 | f1da30e6 | Michael Hanselmann | |
449 | f1da30e6 | Michael Hanselmann | @utils.LockedMethod
|
450 | c3f0a12f | Iustin Pop | def AddJob(self, ops, nodes): |
451 | c3f0a12f | Iustin Pop | """Create and store on disk a new job.
|
452 | c3f0a12f | Iustin Pop |
|
453 | c3f0a12f | Iustin Pop | @type ops: list
|
454 | 205d71fd | Michael Hanselmann | @param ops: The list of OpCodes that will become the new job.
|
455 | c3f0a12f | Iustin Pop | @type nodes: list
|
456 | c3f0a12f | Iustin Pop | @param nodes: The list of nodes to which the new job serial will be
|
457 | c3f0a12f | Iustin Pop | distributed.
|
458 | c3f0a12f | Iustin Pop |
|
459 | c3f0a12f | Iustin Pop | """
|
460 | f1da30e6 | Michael Hanselmann | assert self.lock_fd, "Queue should be open" |
461 | f1da30e6 | Michael Hanselmann | |
462 | f1da30e6 | Michael Hanselmann | # Get job identifier
|
463 | c3f0a12f | Iustin Pop | job_id = self._NewSerialUnlocked(nodes)
|
464 | f1da30e6 | Michael Hanselmann | job = _QueuedJob(self, job_id, ops)
|
465 | f1da30e6 | Michael Hanselmann | |
466 | f1da30e6 | Michael Hanselmann | # Write to disk
|
467 | f1da30e6 | Michael Hanselmann | self._UpdateJobUnlocked(job)
|
468 | f1da30e6 | Michael Hanselmann | |
469 | 205d71fd | Michael Hanselmann | logging.debug("Added new job %s to the cache", job_id)
|
470 | ac0930b9 | Iustin Pop | self._memcache[job_id] = job
|
471 | ac0930b9 | Iustin Pop | |
472 | f1da30e6 | Michael Hanselmann | return job
|
473 | f1da30e6 | Michael Hanselmann | |
474 | f1da30e6 | Michael Hanselmann | def _UpdateJobUnlocked(self, job): |
475 | f1da30e6 | Michael Hanselmann | assert self.lock_fd, "Queue should be open" |
476 | f1da30e6 | Michael Hanselmann | |
477 | f1da30e6 | Michael Hanselmann | filename = self._GetJobPath(job.id)
|
478 | f1da30e6 | Michael Hanselmann | logging.debug("Writing job %s to %s", job.id, filename)
|
479 | f1da30e6 | Michael Hanselmann | utils.WriteFile(filename, |
480 | f1da30e6 | Michael Hanselmann | data=serializer.DumpJson(job.Serialize(), indent=False))
|
481 | 57f8615f | Michael Hanselmann | self._CleanCacheUnlocked([job.id])
|
482 | ac0930b9 | Iustin Pop | |
483 | 57f8615f | Michael Hanselmann | def _CleanCacheUnlocked(self, exclude): |
484 | ac0930b9 | Iustin Pop | """Clean the memory cache.
|
485 | ac0930b9 | Iustin Pop |
|
486 | ac0930b9 | Iustin Pop | The exceptions argument contains job IDs that should not be
|
487 | ac0930b9 | Iustin Pop | cleaned.
|
488 | ac0930b9 | Iustin Pop |
|
489 | ac0930b9 | Iustin Pop | """
|
490 | 57f8615f | Michael Hanselmann | assert isinstance(exclude, list) |
491 | ac0930b9 | Iustin Pop | for job in self._memcache.values(): |
492 | 57f8615f | Michael Hanselmann | if job.id in exclude: |
493 | ac0930b9 | Iustin Pop | continue
|
494 | ac0930b9 | Iustin Pop | if job.GetStatus() not in (constants.JOB_STATUS_QUEUED, |
495 | ac0930b9 | Iustin Pop | constants.JOB_STATUS_RUNNING): |
496 | 205d71fd | Michael Hanselmann | logging.debug("Cleaning job %s from the cache", job.id)
|
497 | ac0930b9 | Iustin Pop | try:
|
498 | ac0930b9 | Iustin Pop | del self._memcache[job.id] |
499 | ac0930b9 | Iustin Pop | except KeyError: |
500 | ac0930b9 | Iustin Pop | pass
|
501 | f1da30e6 | Michael Hanselmann | |
502 | f1da30e6 | Michael Hanselmann | @utils.LockedMethod
|
503 | f1da30e6 | Michael Hanselmann | def UpdateJob(self, job): |
504 | f1da30e6 | Michael Hanselmann | return self._UpdateJobUnlocked(job) |
505 | f1da30e6 | Michael Hanselmann | |
506 | f1da30e6 | Michael Hanselmann | def ArchiveJob(self, job_id): |
507 | f1da30e6 | Michael Hanselmann | raise NotImplementedError() |
508 | f1da30e6 | Michael Hanselmann | |
509 | f1da30e6 | Michael Hanselmann | |
510 | e2715f69 | Michael Hanselmann | class JobQueue: |
511 | e2715f69 | Michael Hanselmann | """The job queue.
|
512 | e2715f69 | Michael Hanselmann |
|
513 | e2715f69 | Michael Hanselmann | """
|
514 | e2715f69 | Michael Hanselmann | def __init__(self, context): |
515 | e2715f69 | Michael Hanselmann | self._lock = threading.Lock()
|
516 | 21cc1fbd | Michael Hanselmann | self._jobs = DiskJobStorage()
|
517 | e2715f69 | Michael Hanselmann | self._wpool = _JobQueueWorkerPool(context)
|
518 | e2715f69 | Michael Hanselmann | |
519 | f1da30e6 | Michael Hanselmann | for job in self._jobs.GetJobs(None): |
520 | f1da30e6 | Michael Hanselmann | status = job.GetStatus() |
521 | f1da30e6 | Michael Hanselmann | if status in (constants.JOB_STATUS_QUEUED, ): |
522 | f1da30e6 | Michael Hanselmann | self._wpool.AddTask(job)
|
523 | e2715f69 | Michael Hanselmann | |
524 | f1da30e6 | Michael Hanselmann | elif status in (constants.JOB_STATUS_RUNNING, ): |
525 | f1da30e6 | Michael Hanselmann | logging.warning("Unfinished job %s found: %s", job.id, job)
|
526 | f1da30e6 | Michael Hanselmann | job.SetUnclean("Unclean master daemon shutdown")
|
527 | e2715f69 | Michael Hanselmann | |
528 | f1da30e6 | Michael Hanselmann | @utils.LockedMethod
|
529 | c3f0a12f | Iustin Pop | def SubmitJob(self, ops, nodes): |
530 | e2715f69 | Michael Hanselmann | """Add a new job to the queue.
|
531 | e2715f69 | Michael Hanselmann |
|
532 | e2715f69 | Michael Hanselmann | This enters the job into our job queue and also puts it on the new
|
533 | e2715f69 | Michael Hanselmann | queue, in order for it to be picked up by the queue processors.
|
534 | e2715f69 | Michael Hanselmann |
|
535 | c3f0a12f | Iustin Pop | @type ops: list
|
536 | c3f0a12f | Iustin Pop | @param ops: the sequence of opcodes that will become the new job
|
537 | c3f0a12f | Iustin Pop | @type nodes: list
|
538 | c3f0a12f | Iustin Pop | @param nodes: the list of nodes to which the queue should be
|
539 | c3f0a12f | Iustin Pop | distributed
|
540 | e2715f69 | Michael Hanselmann |
|
541 | e2715f69 | Michael Hanselmann | """
|
542 | c3f0a12f | Iustin Pop | job = self._jobs.AddJob(ops, nodes)
|
543 | e2715f69 | Michael Hanselmann | |
544 | e2715f69 | Michael Hanselmann | # Add to worker pool
|
545 | e2715f69 | Michael Hanselmann | self._wpool.AddTask(job)
|
546 | e2715f69 | Michael Hanselmann | |
547 | f1da30e6 | Michael Hanselmann | return job.id
|
548 | e2715f69 | Michael Hanselmann | |
549 | e2715f69 | Michael Hanselmann | def ArchiveJob(self, job_id): |
550 | e2715f69 | Michael Hanselmann | raise NotImplementedError() |
551 | e2715f69 | Michael Hanselmann | |
552 | e2715f69 | Michael Hanselmann | def CancelJob(self, job_id): |
553 | e2715f69 | Michael Hanselmann | raise NotImplementedError() |
554 | e2715f69 | Michael Hanselmann | |
555 | e2715f69 | Michael Hanselmann | def _GetJobInfo(self, job, fields): |
556 | e2715f69 | Michael Hanselmann | row = [] |
557 | e2715f69 | Michael Hanselmann | for fname in fields: |
558 | e2715f69 | Michael Hanselmann | if fname == "id": |
559 | e2715f69 | Michael Hanselmann | row.append(job.id) |
560 | e2715f69 | Michael Hanselmann | elif fname == "status": |
561 | e2715f69 | Michael Hanselmann | row.append(job.GetStatus()) |
562 | af30b2fd | Michael Hanselmann | elif fname == "ops": |
563 | af30b2fd | Michael Hanselmann | row.append([op.GetInput().__getstate__() for op in job._ops]) |
564 | af30b2fd | Michael Hanselmann | elif fname == "opresult": |
565 | 307149a8 | Iustin Pop | row.append([op.GetResult() for op in job._ops]) |
566 | af30b2fd | Michael Hanselmann | elif fname == "opstatus": |
567 | af30b2fd | Michael Hanselmann | row.append([op.GetStatus() for op in job._ops]) |
568 | f1048938 | Iustin Pop | elif fname == "ticker": |
569 | f1048938 | Iustin Pop | ji = job.GetRunOpIndex() |
570 | f1048938 | Iustin Pop | if ji < 0: |
571 | f1048938 | Iustin Pop | lmsg = None
|
572 | f1048938 | Iustin Pop | else:
|
573 | f1048938 | Iustin Pop | lmsg = job._ops[ji].RetrieveLog(-1)
|
574 | f1048938 | Iustin Pop | # message might be empty here
|
575 | f1048938 | Iustin Pop | if lmsg:
|
576 | f1048938 | Iustin Pop | lmsg = lmsg[0]
|
577 | f1048938 | Iustin Pop | else:
|
578 | f1048938 | Iustin Pop | lmsg = None
|
579 | f1048938 | Iustin Pop | row.append(lmsg) |
580 | e2715f69 | Michael Hanselmann | else:
|
581 | e2715f69 | Michael Hanselmann | raise errors.OpExecError("Invalid job query field '%s'" % fname) |
582 | e2715f69 | Michael Hanselmann | return row
|
583 | e2715f69 | Michael Hanselmann | |
584 | e2715f69 | Michael Hanselmann | def QueryJobs(self, job_ids, fields): |
585 | e2715f69 | Michael Hanselmann | """Returns a list of jobs in queue.
|
586 | e2715f69 | Michael Hanselmann |
|
587 | e2715f69 | Michael Hanselmann | Args:
|
588 | e2715f69 | Michael Hanselmann | - job_ids: Sequence of job identifiers or None for all
|
589 | e2715f69 | Michael Hanselmann | - fields: Names of fields to return
|
590 | e2715f69 | Michael Hanselmann |
|
591 | e2715f69 | Michael Hanselmann | """
|
592 | e2715f69 | Michael Hanselmann | self._lock.acquire()
|
593 | e2715f69 | Michael Hanselmann | try:
|
594 | e2715f69 | Michael Hanselmann | jobs = [] |
595 | e2715f69 | Michael Hanselmann | |
596 | f1da30e6 | Michael Hanselmann | for job in self._jobs.GetJobs(job_ids): |
597 | e2715f69 | Michael Hanselmann | if job is None: |
598 | e2715f69 | Michael Hanselmann | jobs.append(None)
|
599 | e2715f69 | Michael Hanselmann | else:
|
600 | e2715f69 | Michael Hanselmann | jobs.append(self._GetJobInfo(job, fields))
|
601 | e2715f69 | Michael Hanselmann | |
602 | e2715f69 | Michael Hanselmann | return jobs
|
603 | e2715f69 | Michael Hanselmann | finally:
|
604 | e2715f69 | Michael Hanselmann | self._lock.release()
|
605 | e2715f69 | Michael Hanselmann | |
606 | f1da30e6 | Michael Hanselmann | @utils.LockedMethod
|
607 | e2715f69 | Michael Hanselmann | def Shutdown(self): |
608 | e2715f69 | Michael Hanselmann | """Stops the job queue.
|
609 | e2715f69 | Michael Hanselmann |
|
610 | e2715f69 | Michael Hanselmann | """
|
611 | e2715f69 | Michael Hanselmann | self._wpool.TerminateWorkers()
|
612 | f1da30e6 | Michael Hanselmann | self._jobs.Close() |