root / lib / jqueue.py @ 04df3c7c
History | View | Annotate | Download (58.1 kB)
1 | 498ae1cc | Iustin Pop | #
|
---|---|---|---|
2 | 498ae1cc | Iustin Pop | #
|
3 | 498ae1cc | Iustin Pop | |
4 | 7f93570a | Iustin Pop | # Copyright (C) 2006, 2007, 2008, 2009, 2010 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 | 6c5a7090 | Michael Hanselmann | """Module implementing the job queue handling.
|
23 | 6c5a7090 | Michael Hanselmann |
|
24 | ea03467c | Iustin Pop | Locking: there's a single, large lock in the L{JobQueue} class. It's
|
25 | ea03467c | Iustin Pop | used by all other classes in this module.
|
26 | ea03467c | Iustin Pop |
|
27 | ea03467c | Iustin Pop | @var JOBQUEUE_THREADS: the number of worker threads we start for
|
28 | ea03467c | Iustin Pop | processing jobs
|
29 | 6c5a7090 | Michael Hanselmann |
|
30 | 6c5a7090 | Michael Hanselmann | """
|
31 | 498ae1cc | Iustin Pop | |
32 | e2715f69 | Michael Hanselmann | import logging |
33 | f1da30e6 | Michael Hanselmann | import errno |
34 | f1da30e6 | Michael Hanselmann | import re |
35 | f1048938 | Iustin Pop | import time |
36 | 5685c1a5 | Michael Hanselmann | import weakref |
37 | 498ae1cc | Iustin Pop | |
38 | 6c2549d6 | Guido Trotter | try:
|
39 | 6c2549d6 | Guido Trotter | # pylint: disable-msg=E0611
|
40 | 6c2549d6 | Guido Trotter | from pyinotify import pyinotify |
41 | 6c2549d6 | Guido Trotter | except ImportError: |
42 | 6c2549d6 | Guido Trotter | import pyinotify |
43 | 6c2549d6 | Guido Trotter | |
44 | 6c2549d6 | Guido Trotter | from ganeti import asyncnotifier |
45 | e2715f69 | Michael Hanselmann | from ganeti import constants |
46 | f1da30e6 | Michael Hanselmann | from ganeti import serializer |
47 | e2715f69 | Michael Hanselmann | from ganeti import workerpool |
48 | 99bd4f0a | Guido Trotter | from ganeti import locking |
49 | f1da30e6 | Michael Hanselmann | from ganeti import opcodes |
50 | 7a1ecaed | Iustin Pop | from ganeti import errors |
51 | e2715f69 | Michael Hanselmann | from ganeti import mcpu |
52 | 7996a135 | Iustin Pop | from ganeti import utils |
53 | 04ab05ce | Michael Hanselmann | from ganeti import jstore |
54 | c3f0a12f | Iustin Pop | from ganeti import rpc |
55 | 82b22e19 | René Nussbaumer | from ganeti import runtime |
56 | a744b676 | Manuel Franceschini | from ganeti import netutils |
57 | 989a8bee | Michael Hanselmann | from ganeti import compat |
58 | e2715f69 | Michael Hanselmann | |
59 | fbf0262f | Michael Hanselmann | |
60 | 1daae384 | Iustin Pop | JOBQUEUE_THREADS = 25
|
61 | 58b22b6e | Michael Hanselmann | JOBS_PER_ARCHIVE_DIRECTORY = 10000
|
62 | e2715f69 | Michael Hanselmann | |
63 | ebb80afa | Guido Trotter | # member lock names to be passed to @ssynchronized decorator
|
64 | ebb80afa | Guido Trotter | _LOCK = "_lock"
|
65 | ebb80afa | Guido Trotter | _QUEUE = "_queue"
|
66 | 99bd4f0a | Guido Trotter | |
67 | 498ae1cc | Iustin Pop | |
68 | 9728ae5d | Iustin Pop | class CancelJob(Exception): |
69 | fbf0262f | Michael Hanselmann | """Special exception to cancel a job.
|
70 | fbf0262f | Michael Hanselmann |
|
71 | fbf0262f | Michael Hanselmann | """
|
72 | fbf0262f | Michael Hanselmann | |
73 | fbf0262f | Michael Hanselmann | |
74 | 70552c46 | Michael Hanselmann | def TimeStampNow(): |
75 | ea03467c | Iustin Pop | """Returns the current timestamp.
|
76 | ea03467c | Iustin Pop |
|
77 | ea03467c | Iustin Pop | @rtype: tuple
|
78 | ea03467c | Iustin Pop | @return: the current time in the (seconds, microseconds) format
|
79 | ea03467c | Iustin Pop |
|
80 | ea03467c | Iustin Pop | """
|
81 | 70552c46 | Michael Hanselmann | return utils.SplitTime(time.time())
|
82 | 70552c46 | Michael Hanselmann | |
83 | 70552c46 | Michael Hanselmann | |
84 | e2715f69 | Michael Hanselmann | class _QueuedOpCode(object): |
85 | 5bbd3f7f | Michael Hanselmann | """Encapsulates an opcode object.
|
86 | e2715f69 | Michael Hanselmann |
|
87 | ea03467c | Iustin Pop | @ivar log: holds the execution log and consists of tuples
|
88 | ea03467c | Iustin Pop | of the form C{(log_serial, timestamp, level, message)}
|
89 | ea03467c | Iustin Pop | @ivar input: the OpCode we encapsulate
|
90 | ea03467c | Iustin Pop | @ivar status: the current status
|
91 | ea03467c | Iustin Pop | @ivar result: the result of the LU execution
|
92 | ea03467c | Iustin Pop | @ivar start_timestamp: timestamp for the start of the execution
|
93 | b9b5abcb | Iustin Pop | @ivar exec_timestamp: timestamp for the actual LU Exec() function invocation
|
94 | ea03467c | Iustin Pop | @ivar stop_timestamp: timestamp for the end of the execution
|
95 | f1048938 | Iustin Pop |
|
96 | e2715f69 | Michael Hanselmann | """
|
97 | 8f5c488d | Michael Hanselmann | __slots__ = ["input", "status", "result", "log", "priority", |
98 | b9b5abcb | Iustin Pop | "start_timestamp", "exec_timestamp", "end_timestamp", |
99 | 66d895a8 | Iustin Pop | "__weakref__"]
|
100 | 66d895a8 | Iustin Pop | |
101 | 85f03e0d | Michael Hanselmann | def __init__(self, op): |
102 | ea03467c | Iustin Pop | """Constructor for the _QuededOpCode.
|
103 | ea03467c | Iustin Pop |
|
104 | ea03467c | Iustin Pop | @type op: L{opcodes.OpCode}
|
105 | ea03467c | Iustin Pop | @param op: the opcode we encapsulate
|
106 | ea03467c | Iustin Pop |
|
107 | ea03467c | Iustin Pop | """
|
108 | 85f03e0d | Michael Hanselmann | self.input = op
|
109 | 85f03e0d | Michael Hanselmann | self.status = constants.OP_STATUS_QUEUED
|
110 | 85f03e0d | Michael Hanselmann | self.result = None |
111 | 85f03e0d | Michael Hanselmann | self.log = []
|
112 | 70552c46 | Michael Hanselmann | self.start_timestamp = None |
113 | b9b5abcb | Iustin Pop | self.exec_timestamp = None |
114 | 70552c46 | Michael Hanselmann | self.end_timestamp = None |
115 | f1da30e6 | Michael Hanselmann | |
116 | 8f5c488d | Michael Hanselmann | # Get initial priority (it might change during the lifetime of this opcode)
|
117 | 8f5c488d | Michael Hanselmann | self.priority = getattr(op, "priority", constants.OP_PRIO_DEFAULT) |
118 | 8f5c488d | Michael Hanselmann | |
119 | f1da30e6 | Michael Hanselmann | @classmethod
|
120 | f1da30e6 | Michael Hanselmann | def Restore(cls, state): |
121 | ea03467c | Iustin Pop | """Restore the _QueuedOpCode from the serialized form.
|
122 | ea03467c | Iustin Pop |
|
123 | ea03467c | Iustin Pop | @type state: dict
|
124 | ea03467c | Iustin Pop | @param state: the serialized state
|
125 | ea03467c | Iustin Pop | @rtype: _QueuedOpCode
|
126 | ea03467c | Iustin Pop | @return: a new _QueuedOpCode instance
|
127 | ea03467c | Iustin Pop |
|
128 | ea03467c | Iustin Pop | """
|
129 | 85f03e0d | Michael Hanselmann | obj = _QueuedOpCode.__new__(cls) |
130 | 85f03e0d | Michael Hanselmann | obj.input = opcodes.OpCode.LoadOpCode(state["input"])
|
131 | 85f03e0d | Michael Hanselmann | obj.status = state["status"]
|
132 | 85f03e0d | Michael Hanselmann | obj.result = state["result"]
|
133 | 85f03e0d | Michael Hanselmann | obj.log = state["log"]
|
134 | 70552c46 | Michael Hanselmann | obj.start_timestamp = state.get("start_timestamp", None) |
135 | b9b5abcb | Iustin Pop | obj.exec_timestamp = state.get("exec_timestamp", None) |
136 | 70552c46 | Michael Hanselmann | obj.end_timestamp = state.get("end_timestamp", None) |
137 | 8f5c488d | Michael Hanselmann | obj.priority = state.get("priority", constants.OP_PRIO_DEFAULT)
|
138 | f1da30e6 | Michael Hanselmann | return obj
|
139 | f1da30e6 | Michael Hanselmann | |
140 | f1da30e6 | Michael Hanselmann | def Serialize(self): |
141 | ea03467c | Iustin Pop | """Serializes this _QueuedOpCode.
|
142 | ea03467c | Iustin Pop |
|
143 | ea03467c | Iustin Pop | @rtype: dict
|
144 | ea03467c | Iustin Pop | @return: the dictionary holding the serialized state
|
145 | ea03467c | Iustin Pop |
|
146 | ea03467c | Iustin Pop | """
|
147 | 6c5a7090 | Michael Hanselmann | return {
|
148 | 6c5a7090 | Michael Hanselmann | "input": self.input.__getstate__(), |
149 | 6c5a7090 | Michael Hanselmann | "status": self.status, |
150 | 6c5a7090 | Michael Hanselmann | "result": self.result, |
151 | 6c5a7090 | Michael Hanselmann | "log": self.log, |
152 | 70552c46 | Michael Hanselmann | "start_timestamp": self.start_timestamp, |
153 | b9b5abcb | Iustin Pop | "exec_timestamp": self.exec_timestamp, |
154 | 70552c46 | Michael Hanselmann | "end_timestamp": self.end_timestamp, |
155 | 8f5c488d | Michael Hanselmann | "priority": self.priority, |
156 | 6c5a7090 | Michael Hanselmann | } |
157 | f1048938 | Iustin Pop | |
158 | e2715f69 | Michael Hanselmann | |
159 | e2715f69 | Michael Hanselmann | class _QueuedJob(object): |
160 | e2715f69 | Michael Hanselmann | """In-memory job representation.
|
161 | e2715f69 | Michael Hanselmann |
|
162 | ea03467c | Iustin Pop | This is what we use to track the user-submitted jobs. Locking must
|
163 | ea03467c | Iustin Pop | be taken care of by users of this class.
|
164 | ea03467c | Iustin Pop |
|
165 | ea03467c | Iustin Pop | @type queue: L{JobQueue}
|
166 | ea03467c | Iustin Pop | @ivar queue: the parent queue
|
167 | ea03467c | Iustin Pop | @ivar id: the job ID
|
168 | ea03467c | Iustin Pop | @type ops: list
|
169 | ea03467c | Iustin Pop | @ivar ops: the list of _QueuedOpCode that constitute the job
|
170 | ea03467c | Iustin Pop | @type log_serial: int
|
171 | ea03467c | Iustin Pop | @ivar log_serial: holds the index for the next log entry
|
172 | ea03467c | Iustin Pop | @ivar received_timestamp: the timestamp for when the job was received
|
173 | ea03467c | Iustin Pop | @ivar start_timestmap: the timestamp for start of execution
|
174 | ea03467c | Iustin Pop | @ivar end_timestamp: the timestamp for end of execution
|
175 | e2715f69 | Michael Hanselmann |
|
176 | e2715f69 | Michael Hanselmann | """
|
177 | 7260cfbe | Iustin Pop | # pylint: disable-msg=W0212
|
178 | 26d3fd2f | Michael Hanselmann | __slots__ = ["queue", "id", "ops", "log_serial", "ops_iter", "cur_opctx", |
179 | 66d895a8 | Iustin Pop | "received_timestamp", "start_timestamp", "end_timestamp", |
180 | 66d895a8 | Iustin Pop | "__weakref__"]
|
181 | 66d895a8 | Iustin Pop | |
182 | 85f03e0d | Michael Hanselmann | def __init__(self, queue, job_id, ops): |
183 | ea03467c | Iustin Pop | """Constructor for the _QueuedJob.
|
184 | ea03467c | Iustin Pop |
|
185 | ea03467c | Iustin Pop | @type queue: L{JobQueue}
|
186 | ea03467c | Iustin Pop | @param queue: our parent queue
|
187 | ea03467c | Iustin Pop | @type job_id: job_id
|
188 | ea03467c | Iustin Pop | @param job_id: our job id
|
189 | ea03467c | Iustin Pop | @type ops: list
|
190 | ea03467c | Iustin Pop | @param ops: the list of opcodes we hold, which will be encapsulated
|
191 | ea03467c | Iustin Pop | in _QueuedOpCodes
|
192 | ea03467c | Iustin Pop |
|
193 | ea03467c | Iustin Pop | """
|
194 | e2715f69 | Michael Hanselmann | if not ops: |
195 | c910bccb | Guido Trotter | raise errors.GenericError("A job needs at least one opcode") |
196 | e2715f69 | Michael Hanselmann | |
197 | 85f03e0d | Michael Hanselmann | self.queue = queue
|
198 | f1da30e6 | Michael Hanselmann | self.id = job_id
|
199 | 85f03e0d | Michael Hanselmann | self.ops = [_QueuedOpCode(op) for op in ops] |
200 | 6c5a7090 | Michael Hanselmann | self.log_serial = 0 |
201 | c56ec146 | Iustin Pop | self.received_timestamp = TimeStampNow()
|
202 | c56ec146 | Iustin Pop | self.start_timestamp = None |
203 | c56ec146 | Iustin Pop | self.end_timestamp = None |
204 | 6c5a7090 | Michael Hanselmann | |
205 | fa4aa6b4 | Michael Hanselmann | self._InitInMemory(self) |
206 | fa4aa6b4 | Michael Hanselmann | |
207 | fa4aa6b4 | Michael Hanselmann | @staticmethod
|
208 | fa4aa6b4 | Michael Hanselmann | def _InitInMemory(obj): |
209 | fa4aa6b4 | Michael Hanselmann | """Initializes in-memory variables.
|
210 | fa4aa6b4 | Michael Hanselmann |
|
211 | fa4aa6b4 | Michael Hanselmann | """
|
212 | 03b63608 | Michael Hanselmann | obj.ops_iter = None
|
213 | 26d3fd2f | Michael Hanselmann | obj.cur_opctx = None
|
214 | be760ba8 | Michael Hanselmann | |
215 | 9fa2e150 | Michael Hanselmann | def __repr__(self): |
216 | 9fa2e150 | Michael Hanselmann | status = ["%s.%s" % (self.__class__.__module__, self.__class__.__name__), |
217 | 9fa2e150 | Michael Hanselmann | "id=%s" % self.id, |
218 | 9fa2e150 | Michael Hanselmann | "ops=%s" % ",".join([op.input.Summary() for op in self.ops])] |
219 | 9fa2e150 | Michael Hanselmann | |
220 | 9fa2e150 | Michael Hanselmann | return "<%s at %#x>" % (" ".join(status), id(self)) |
221 | 9fa2e150 | Michael Hanselmann | |
222 | f1da30e6 | Michael Hanselmann | @classmethod
|
223 | 85f03e0d | Michael Hanselmann | def Restore(cls, queue, state): |
224 | ea03467c | Iustin Pop | """Restore a _QueuedJob from serialized state:
|
225 | ea03467c | Iustin Pop |
|
226 | ea03467c | Iustin Pop | @type queue: L{JobQueue}
|
227 | ea03467c | Iustin Pop | @param queue: to which queue the restored job belongs
|
228 | ea03467c | Iustin Pop | @type state: dict
|
229 | ea03467c | Iustin Pop | @param state: the serialized state
|
230 | ea03467c | Iustin Pop | @rtype: _JobQueue
|
231 | ea03467c | Iustin Pop | @return: the restored _JobQueue instance
|
232 | ea03467c | Iustin Pop |
|
233 | ea03467c | Iustin Pop | """
|
234 | 85f03e0d | Michael Hanselmann | obj = _QueuedJob.__new__(cls) |
235 | 85f03e0d | Michael Hanselmann | obj.queue = queue |
236 | 85f03e0d | Michael Hanselmann | obj.id = state["id"]
|
237 | c56ec146 | Iustin Pop | obj.received_timestamp = state.get("received_timestamp", None) |
238 | c56ec146 | Iustin Pop | obj.start_timestamp = state.get("start_timestamp", None) |
239 | c56ec146 | Iustin Pop | obj.end_timestamp = state.get("end_timestamp", None) |
240 | 6c5a7090 | Michael Hanselmann | |
241 | 6c5a7090 | Michael Hanselmann | obj.ops = [] |
242 | 6c5a7090 | Michael Hanselmann | obj.log_serial = 0
|
243 | 6c5a7090 | Michael Hanselmann | for op_state in state["ops"]: |
244 | 6c5a7090 | Michael Hanselmann | op = _QueuedOpCode.Restore(op_state) |
245 | 6c5a7090 | Michael Hanselmann | for log_entry in op.log: |
246 | 6c5a7090 | Michael Hanselmann | obj.log_serial = max(obj.log_serial, log_entry[0]) |
247 | 6c5a7090 | Michael Hanselmann | obj.ops.append(op) |
248 | 6c5a7090 | Michael Hanselmann | |
249 | fa4aa6b4 | Michael Hanselmann | cls._InitInMemory(obj) |
250 | be760ba8 | Michael Hanselmann | |
251 | f1da30e6 | Michael Hanselmann | return obj
|
252 | f1da30e6 | Michael Hanselmann | |
253 | f1da30e6 | Michael Hanselmann | def Serialize(self): |
254 | ea03467c | Iustin Pop | """Serialize the _JobQueue instance.
|
255 | ea03467c | Iustin Pop |
|
256 | ea03467c | Iustin Pop | @rtype: dict
|
257 | ea03467c | Iustin Pop | @return: the serialized state
|
258 | ea03467c | Iustin Pop |
|
259 | ea03467c | Iustin Pop | """
|
260 | f1da30e6 | Michael Hanselmann | return {
|
261 | f1da30e6 | Michael Hanselmann | "id": self.id, |
262 | 85f03e0d | Michael Hanselmann | "ops": [op.Serialize() for op in self.ops], |
263 | c56ec146 | Iustin Pop | "start_timestamp": self.start_timestamp, |
264 | c56ec146 | Iustin Pop | "end_timestamp": self.end_timestamp, |
265 | c56ec146 | Iustin Pop | "received_timestamp": self.received_timestamp, |
266 | f1da30e6 | Michael Hanselmann | } |
267 | f1da30e6 | Michael Hanselmann | |
268 | 85f03e0d | Michael Hanselmann | def CalcStatus(self): |
269 | ea03467c | Iustin Pop | """Compute the status of this job.
|
270 | ea03467c | Iustin Pop |
|
271 | ea03467c | Iustin Pop | This function iterates over all the _QueuedOpCodes in the job and
|
272 | ea03467c | Iustin Pop | based on their status, computes the job status.
|
273 | ea03467c | Iustin Pop |
|
274 | ea03467c | Iustin Pop | The algorithm is:
|
275 | ea03467c | Iustin Pop | - if we find a cancelled, or finished with error, the job
|
276 | ea03467c | Iustin Pop | status will be the same
|
277 | ea03467c | Iustin Pop | - otherwise, the last opcode with the status one of:
|
278 | ea03467c | Iustin Pop | - waitlock
|
279 | fbf0262f | Michael Hanselmann | - canceling
|
280 | ea03467c | Iustin Pop | - running
|
281 | ea03467c | Iustin Pop |
|
282 | ea03467c | Iustin Pop | will determine the job status
|
283 | ea03467c | Iustin Pop |
|
284 | ea03467c | Iustin Pop | - otherwise, it means either all opcodes are queued, or success,
|
285 | ea03467c | Iustin Pop | and the job status will be the same
|
286 | ea03467c | Iustin Pop |
|
287 | ea03467c | Iustin Pop | @return: the job status
|
288 | ea03467c | Iustin Pop |
|
289 | ea03467c | Iustin Pop | """
|
290 | e2715f69 | Michael Hanselmann | status = constants.JOB_STATUS_QUEUED |
291 | e2715f69 | Michael Hanselmann | |
292 | e2715f69 | Michael Hanselmann | all_success = True
|
293 | 85f03e0d | Michael Hanselmann | for op in self.ops: |
294 | 85f03e0d | Michael Hanselmann | if op.status == constants.OP_STATUS_SUCCESS:
|
295 | e2715f69 | Michael Hanselmann | continue
|
296 | e2715f69 | Michael Hanselmann | |
297 | e2715f69 | Michael Hanselmann | all_success = False
|
298 | e2715f69 | Michael Hanselmann | |
299 | 85f03e0d | Michael Hanselmann | if op.status == constants.OP_STATUS_QUEUED:
|
300 | e2715f69 | Michael Hanselmann | pass
|
301 | e92376d7 | Iustin Pop | elif op.status == constants.OP_STATUS_WAITLOCK:
|
302 | e92376d7 | Iustin Pop | status = constants.JOB_STATUS_WAITLOCK |
303 | 85f03e0d | Michael Hanselmann | elif op.status == constants.OP_STATUS_RUNNING:
|
304 | e2715f69 | Michael Hanselmann | status = constants.JOB_STATUS_RUNNING |
305 | fbf0262f | Michael Hanselmann | elif op.status == constants.OP_STATUS_CANCELING:
|
306 | fbf0262f | Michael Hanselmann | status = constants.JOB_STATUS_CANCELING |
307 | fbf0262f | Michael Hanselmann | break
|
308 | 85f03e0d | Michael Hanselmann | elif op.status == constants.OP_STATUS_ERROR:
|
309 | f1da30e6 | Michael Hanselmann | status = constants.JOB_STATUS_ERROR |
310 | f1da30e6 | Michael Hanselmann | # The whole job fails if one opcode failed
|
311 | f1da30e6 | Michael Hanselmann | break
|
312 | 85f03e0d | Michael Hanselmann | elif op.status == constants.OP_STATUS_CANCELED:
|
313 | 4cb1d919 | Michael Hanselmann | status = constants.OP_STATUS_CANCELED |
314 | 4cb1d919 | Michael Hanselmann | break
|
315 | e2715f69 | Michael Hanselmann | |
316 | e2715f69 | Michael Hanselmann | if all_success:
|
317 | e2715f69 | Michael Hanselmann | status = constants.JOB_STATUS_SUCCESS |
318 | e2715f69 | Michael Hanselmann | |
319 | e2715f69 | Michael Hanselmann | return status
|
320 | e2715f69 | Michael Hanselmann | |
321 | 8f5c488d | Michael Hanselmann | def CalcPriority(self): |
322 | 8f5c488d | Michael Hanselmann | """Gets the current priority for this job.
|
323 | 8f5c488d | Michael Hanselmann |
|
324 | 8f5c488d | Michael Hanselmann | Only unfinished opcodes are considered. When all are done, the default
|
325 | 8f5c488d | Michael Hanselmann | priority is used.
|
326 | 8f5c488d | Michael Hanselmann |
|
327 | 8f5c488d | Michael Hanselmann | @rtype: int
|
328 | 8f5c488d | Michael Hanselmann |
|
329 | 8f5c488d | Michael Hanselmann | """
|
330 | 8f5c488d | Michael Hanselmann | priorities = [op.priority for op in self.ops |
331 | 8f5c488d | Michael Hanselmann | if op.status not in constants.OPS_FINALIZED] |
332 | 8f5c488d | Michael Hanselmann | |
333 | 8f5c488d | Michael Hanselmann | if not priorities: |
334 | 8f5c488d | Michael Hanselmann | # All opcodes are done, assume default priority
|
335 | 8f5c488d | Michael Hanselmann | return constants.OP_PRIO_DEFAULT
|
336 | 8f5c488d | Michael Hanselmann | |
337 | 8f5c488d | Michael Hanselmann | return min(priorities) |
338 | 8f5c488d | Michael Hanselmann | |
339 | 6c5a7090 | Michael Hanselmann | def GetLogEntries(self, newer_than): |
340 | ea03467c | Iustin Pop | """Selectively returns the log entries.
|
341 | ea03467c | Iustin Pop |
|
342 | ea03467c | Iustin Pop | @type newer_than: None or int
|
343 | 5bbd3f7f | Michael Hanselmann | @param newer_than: if this is None, return all log entries,
|
344 | ea03467c | Iustin Pop | otherwise return only the log entries with serial higher
|
345 | ea03467c | Iustin Pop | than this value
|
346 | ea03467c | Iustin Pop | @rtype: list
|
347 | ea03467c | Iustin Pop | @return: the list of the log entries selected
|
348 | ea03467c | Iustin Pop |
|
349 | ea03467c | Iustin Pop | """
|
350 | 6c5a7090 | Michael Hanselmann | if newer_than is None: |
351 | 6c5a7090 | Michael Hanselmann | serial = -1
|
352 | 6c5a7090 | Michael Hanselmann | else:
|
353 | 6c5a7090 | Michael Hanselmann | serial = newer_than |
354 | 6c5a7090 | Michael Hanselmann | |
355 | 6c5a7090 | Michael Hanselmann | entries = [] |
356 | 6c5a7090 | Michael Hanselmann | for op in self.ops: |
357 | 63712a09 | Iustin Pop | entries.extend(filter(lambda entry: entry[0] > serial, op.log)) |
358 | 6c5a7090 | Michael Hanselmann | |
359 | 6c5a7090 | Michael Hanselmann | return entries
|
360 | 6c5a7090 | Michael Hanselmann | |
361 | 6a290889 | Guido Trotter | def GetInfo(self, fields): |
362 | 6a290889 | Guido Trotter | """Returns information about a job.
|
363 | 6a290889 | Guido Trotter |
|
364 | 6a290889 | Guido Trotter | @type fields: list
|
365 | 6a290889 | Guido Trotter | @param fields: names of fields to return
|
366 | 6a290889 | Guido Trotter | @rtype: list
|
367 | 6a290889 | Guido Trotter | @return: list with one element for each field
|
368 | 6a290889 | Guido Trotter | @raise errors.OpExecError: when an invalid field
|
369 | 6a290889 | Guido Trotter | has been passed
|
370 | 6a290889 | Guido Trotter |
|
371 | 6a290889 | Guido Trotter | """
|
372 | 6a290889 | Guido Trotter | row = [] |
373 | 6a290889 | Guido Trotter | for fname in fields: |
374 | 6a290889 | Guido Trotter | if fname == "id": |
375 | 6a290889 | Guido Trotter | row.append(self.id)
|
376 | 6a290889 | Guido Trotter | elif fname == "status": |
377 | 6a290889 | Guido Trotter | row.append(self.CalcStatus())
|
378 | b8802cc4 | Michael Hanselmann | elif fname == "priority": |
379 | b8802cc4 | Michael Hanselmann | row.append(self.CalcPriority())
|
380 | 6a290889 | Guido Trotter | elif fname == "ops": |
381 | 6a290889 | Guido Trotter | row.append([op.input.__getstate__() for op in self.ops]) |
382 | 6a290889 | Guido Trotter | elif fname == "opresult": |
383 | 6a290889 | Guido Trotter | row.append([op.result for op in self.ops]) |
384 | 6a290889 | Guido Trotter | elif fname == "opstatus": |
385 | 6a290889 | Guido Trotter | row.append([op.status for op in self.ops]) |
386 | 6a290889 | Guido Trotter | elif fname == "oplog": |
387 | 6a290889 | Guido Trotter | row.append([op.log for op in self.ops]) |
388 | 6a290889 | Guido Trotter | elif fname == "opstart": |
389 | 6a290889 | Guido Trotter | row.append([op.start_timestamp for op in self.ops]) |
390 | 6a290889 | Guido Trotter | elif fname == "opexec": |
391 | 6a290889 | Guido Trotter | row.append([op.exec_timestamp for op in self.ops]) |
392 | 6a290889 | Guido Trotter | elif fname == "opend": |
393 | 6a290889 | Guido Trotter | row.append([op.end_timestamp for op in self.ops]) |
394 | b8802cc4 | Michael Hanselmann | elif fname == "oppriority": |
395 | b8802cc4 | Michael Hanselmann | row.append([op.priority for op in self.ops]) |
396 | 6a290889 | Guido Trotter | elif fname == "received_ts": |
397 | 6a290889 | Guido Trotter | row.append(self.received_timestamp)
|
398 | 6a290889 | Guido Trotter | elif fname == "start_ts": |
399 | 6a290889 | Guido Trotter | row.append(self.start_timestamp)
|
400 | 6a290889 | Guido Trotter | elif fname == "end_ts": |
401 | 6a290889 | Guido Trotter | row.append(self.end_timestamp)
|
402 | 6a290889 | Guido Trotter | elif fname == "summary": |
403 | 6a290889 | Guido Trotter | row.append([op.input.Summary() for op in self.ops]) |
404 | 6a290889 | Guido Trotter | else:
|
405 | 6a290889 | Guido Trotter | raise errors.OpExecError("Invalid self query field '%s'" % fname) |
406 | 6a290889 | Guido Trotter | return row
|
407 | 6a290889 | Guido Trotter | |
408 | 34327f51 | Iustin Pop | def MarkUnfinishedOps(self, status, result): |
409 | 34327f51 | Iustin Pop | """Mark unfinished opcodes with a given status and result.
|
410 | 34327f51 | Iustin Pop |
|
411 | 34327f51 | Iustin Pop | This is an utility function for marking all running or waiting to
|
412 | 34327f51 | Iustin Pop | be run opcodes with a given status. Opcodes which are already
|
413 | 34327f51 | Iustin Pop | finalised are not changed.
|
414 | 34327f51 | Iustin Pop |
|
415 | 34327f51 | Iustin Pop | @param status: a given opcode status
|
416 | 34327f51 | Iustin Pop | @param result: the opcode result
|
417 | 34327f51 | Iustin Pop |
|
418 | 34327f51 | Iustin Pop | """
|
419 | 747f6113 | Michael Hanselmann | not_marked = True
|
420 | 747f6113 | Michael Hanselmann | for op in self.ops: |
421 | 747f6113 | Michael Hanselmann | if op.status in constants.OPS_FINALIZED: |
422 | 747f6113 | Michael Hanselmann | assert not_marked, "Finalized opcodes found after non-finalized ones" |
423 | 747f6113 | Michael Hanselmann | continue
|
424 | 747f6113 | Michael Hanselmann | op.status = status |
425 | 747f6113 | Michael Hanselmann | op.result = result |
426 | 747f6113 | Michael Hanselmann | not_marked = False
|
427 | 34327f51 | Iustin Pop | |
428 | 099b2870 | Michael Hanselmann | def Cancel(self): |
429 | a0d2fe2c | Michael Hanselmann | """Marks job as canceled/-ing if possible.
|
430 | a0d2fe2c | Michael Hanselmann |
|
431 | a0d2fe2c | Michael Hanselmann | @rtype: tuple; (bool, string)
|
432 | a0d2fe2c | Michael Hanselmann | @return: Boolean describing whether job was successfully canceled or marked
|
433 | a0d2fe2c | Michael Hanselmann | as canceling and a text message
|
434 | a0d2fe2c | Michael Hanselmann |
|
435 | a0d2fe2c | Michael Hanselmann | """
|
436 | 099b2870 | Michael Hanselmann | status = self.CalcStatus()
|
437 | 099b2870 | Michael Hanselmann | |
438 | 099b2870 | Michael Hanselmann | if status == constants.JOB_STATUS_QUEUED:
|
439 | 099b2870 | Michael Hanselmann | self.MarkUnfinishedOps(constants.OP_STATUS_CANCELED,
|
440 | 099b2870 | Michael Hanselmann | "Job canceled by request")
|
441 | 86b16e9d | Michael Hanselmann | return (True, "Job %s canceled" % self.id) |
442 | 099b2870 | Michael Hanselmann | |
443 | 099b2870 | Michael Hanselmann | elif status == constants.JOB_STATUS_WAITLOCK:
|
444 | 099b2870 | Michael Hanselmann | # The worker will notice the new status and cancel the job
|
445 | 099b2870 | Michael Hanselmann | self.MarkUnfinishedOps(constants.OP_STATUS_CANCELING, None) |
446 | 86b16e9d | Michael Hanselmann | return (True, "Job %s will be canceled" % self.id) |
447 | 099b2870 | Michael Hanselmann | |
448 | 86b16e9d | Michael Hanselmann | else:
|
449 | 86b16e9d | Michael Hanselmann | logging.debug("Job %s is no longer waiting in the queue", self.id) |
450 | 86b16e9d | Michael Hanselmann | return (False, "Job %s is no longer waiting in the queue" % self.id) |
451 | 099b2870 | Michael Hanselmann | |
452 | f1048938 | Iustin Pop | |
453 | ef2df7d3 | Michael Hanselmann | class _OpExecCallbacks(mcpu.OpExecCbBase): |
454 | 031a3e57 | Michael Hanselmann | def __init__(self, queue, job, op): |
455 | 031a3e57 | Michael Hanselmann | """Initializes this class.
|
456 | ea03467c | Iustin Pop |
|
457 | 031a3e57 | Michael Hanselmann | @type queue: L{JobQueue}
|
458 | 031a3e57 | Michael Hanselmann | @param queue: Job queue
|
459 | 031a3e57 | Michael Hanselmann | @type job: L{_QueuedJob}
|
460 | 031a3e57 | Michael Hanselmann | @param job: Job object
|
461 | 031a3e57 | Michael Hanselmann | @type op: L{_QueuedOpCode}
|
462 | 031a3e57 | Michael Hanselmann | @param op: OpCode
|
463 | 031a3e57 | Michael Hanselmann |
|
464 | 031a3e57 | Michael Hanselmann | """
|
465 | 031a3e57 | Michael Hanselmann | assert queue, "Queue is missing" |
466 | 031a3e57 | Michael Hanselmann | assert job, "Job is missing" |
467 | 031a3e57 | Michael Hanselmann | assert op, "Opcode is missing" |
468 | 031a3e57 | Michael Hanselmann | |
469 | 031a3e57 | Michael Hanselmann | self._queue = queue
|
470 | 031a3e57 | Michael Hanselmann | self._job = job
|
471 | 031a3e57 | Michael Hanselmann | self._op = op
|
472 | 031a3e57 | Michael Hanselmann | |
473 | dc1e2262 | Michael Hanselmann | def _CheckCancel(self): |
474 | dc1e2262 | Michael Hanselmann | """Raises an exception to cancel the job if asked to.
|
475 | dc1e2262 | Michael Hanselmann |
|
476 | dc1e2262 | Michael Hanselmann | """
|
477 | dc1e2262 | Michael Hanselmann | # Cancel here if we were asked to
|
478 | dc1e2262 | Michael Hanselmann | if self._op.status == constants.OP_STATUS_CANCELING: |
479 | dc1e2262 | Michael Hanselmann | logging.debug("Canceling opcode")
|
480 | dc1e2262 | Michael Hanselmann | raise CancelJob()
|
481 | dc1e2262 | Michael Hanselmann | |
482 | 271daef8 | Iustin Pop | @locking.ssynchronized(_QUEUE, shared=1) |
483 | 031a3e57 | Michael Hanselmann | def NotifyStart(self): |
484 | e92376d7 | Iustin Pop | """Mark the opcode as running, not lock-waiting.
|
485 | e92376d7 | Iustin Pop |
|
486 | 031a3e57 | Michael Hanselmann | This is called from the mcpu code as a notifier function, when the LU is
|
487 | 031a3e57 | Michael Hanselmann | finally about to start the Exec() method. Of course, to have end-user
|
488 | 031a3e57 | Michael Hanselmann | visible results, the opcode must be initially (before calling into
|
489 | 031a3e57 | Michael Hanselmann | Processor.ExecOpCode) set to OP_STATUS_WAITLOCK.
|
490 | e92376d7 | Iustin Pop |
|
491 | e92376d7 | Iustin Pop | """
|
492 | 9bdab621 | Michael Hanselmann | assert self._op in self._job.ops |
493 | 271daef8 | Iustin Pop | assert self._op.status in (constants.OP_STATUS_WAITLOCK, |
494 | 271daef8 | Iustin Pop | constants.OP_STATUS_CANCELING) |
495 | fbf0262f | Michael Hanselmann | |
496 | 271daef8 | Iustin Pop | # Cancel here if we were asked to
|
497 | dc1e2262 | Michael Hanselmann | self._CheckCancel()
|
498 | fbf0262f | Michael Hanselmann | |
499 | e35344b4 | Michael Hanselmann | logging.debug("Opcode is now running")
|
500 | 9bdab621 | Michael Hanselmann | |
501 | 271daef8 | Iustin Pop | self._op.status = constants.OP_STATUS_RUNNING
|
502 | 271daef8 | Iustin Pop | self._op.exec_timestamp = TimeStampNow()
|
503 | 271daef8 | Iustin Pop | |
504 | 271daef8 | Iustin Pop | # And finally replicate the job status
|
505 | 271daef8 | Iustin Pop | self._queue.UpdateJobUnlocked(self._job) |
506 | 031a3e57 | Michael Hanselmann | |
507 | ebb80afa | Guido Trotter | @locking.ssynchronized(_QUEUE, shared=1) |
508 | 9bf5e01f | Guido Trotter | def _AppendFeedback(self, timestamp, log_type, log_msg): |
509 | 9bf5e01f | Guido Trotter | """Internal feedback append function, with locks
|
510 | 9bf5e01f | Guido Trotter |
|
511 | 9bf5e01f | Guido Trotter | """
|
512 | 9bf5e01f | Guido Trotter | self._job.log_serial += 1 |
513 | 9bf5e01f | Guido Trotter | self._op.log.append((self._job.log_serial, timestamp, log_type, log_msg)) |
514 | 9bf5e01f | Guido Trotter | self._queue.UpdateJobUnlocked(self._job, replicate=False) |
515 | 9bf5e01f | Guido Trotter | |
516 | 031a3e57 | Michael Hanselmann | def Feedback(self, *args): |
517 | 031a3e57 | Michael Hanselmann | """Append a log entry.
|
518 | 031a3e57 | Michael Hanselmann |
|
519 | 031a3e57 | Michael Hanselmann | """
|
520 | 031a3e57 | Michael Hanselmann | assert len(args) < 3 |
521 | 031a3e57 | Michael Hanselmann | |
522 | 031a3e57 | Michael Hanselmann | if len(args) == 1: |
523 | 031a3e57 | Michael Hanselmann | log_type = constants.ELOG_MESSAGE |
524 | 031a3e57 | Michael Hanselmann | log_msg = args[0]
|
525 | 031a3e57 | Michael Hanselmann | else:
|
526 | 031a3e57 | Michael Hanselmann | (log_type, log_msg) = args |
527 | 031a3e57 | Michael Hanselmann | |
528 | 031a3e57 | Michael Hanselmann | # The time is split to make serialization easier and not lose
|
529 | 031a3e57 | Michael Hanselmann | # precision.
|
530 | 031a3e57 | Michael Hanselmann | timestamp = utils.SplitTime(time.time()) |
531 | 9bf5e01f | Guido Trotter | self._AppendFeedback(timestamp, log_type, log_msg)
|
532 | 031a3e57 | Michael Hanselmann | |
533 | acf931b7 | Michael Hanselmann | def CheckCancel(self): |
534 | acf931b7 | Michael Hanselmann | """Check whether job has been cancelled.
|
535 | ef2df7d3 | Michael Hanselmann |
|
536 | ef2df7d3 | Michael Hanselmann | """
|
537 | dc1e2262 | Michael Hanselmann | assert self._op.status in (constants.OP_STATUS_WAITLOCK, |
538 | dc1e2262 | Michael Hanselmann | constants.OP_STATUS_CANCELING) |
539 | dc1e2262 | Michael Hanselmann | |
540 | dc1e2262 | Michael Hanselmann | # Cancel here if we were asked to
|
541 | dc1e2262 | Michael Hanselmann | self._CheckCancel()
|
542 | dc1e2262 | Michael Hanselmann | |
543 | 6a373640 | Michael Hanselmann | def SubmitManyJobs(self, jobs): |
544 | 6a373640 | Michael Hanselmann | """Submits jobs for processing.
|
545 | 6a373640 | Michael Hanselmann |
|
546 | 6a373640 | Michael Hanselmann | See L{JobQueue.SubmitManyJobs}.
|
547 | 6a373640 | Michael Hanselmann |
|
548 | 6a373640 | Michael Hanselmann | """
|
549 | 6a373640 | Michael Hanselmann | # Locking is done in job queue
|
550 | 6a373640 | Michael Hanselmann | return self._queue.SubmitManyJobs(jobs) |
551 | 6a373640 | Michael Hanselmann | |
552 | 031a3e57 | Michael Hanselmann | |
553 | 989a8bee | Michael Hanselmann | class _JobChangesChecker(object): |
554 | 989a8bee | Michael Hanselmann | def __init__(self, fields, prev_job_info, prev_log_serial): |
555 | 989a8bee | Michael Hanselmann | """Initializes this class.
|
556 | 6c2549d6 | Guido Trotter |
|
557 | 989a8bee | Michael Hanselmann | @type fields: list of strings
|
558 | 989a8bee | Michael Hanselmann | @param fields: Fields requested by LUXI client
|
559 | 989a8bee | Michael Hanselmann | @type prev_job_info: string
|
560 | 989a8bee | Michael Hanselmann | @param prev_job_info: previous job info, as passed by the LUXI client
|
561 | 989a8bee | Michael Hanselmann | @type prev_log_serial: string
|
562 | 989a8bee | Michael Hanselmann | @param prev_log_serial: previous job serial, as passed by the LUXI client
|
563 | 6c2549d6 | Guido Trotter |
|
564 | 989a8bee | Michael Hanselmann | """
|
565 | 989a8bee | Michael Hanselmann | self._fields = fields
|
566 | 989a8bee | Michael Hanselmann | self._prev_job_info = prev_job_info
|
567 | 989a8bee | Michael Hanselmann | self._prev_log_serial = prev_log_serial
|
568 | 6c2549d6 | Guido Trotter | |
569 | 989a8bee | Michael Hanselmann | def __call__(self, job): |
570 | 989a8bee | Michael Hanselmann | """Checks whether job has changed.
|
571 | 6c2549d6 | Guido Trotter |
|
572 | 989a8bee | Michael Hanselmann | @type job: L{_QueuedJob}
|
573 | 989a8bee | Michael Hanselmann | @param job: Job object
|
574 | 6c2549d6 | Guido Trotter |
|
575 | 6c2549d6 | Guido Trotter | """
|
576 | 989a8bee | Michael Hanselmann | status = job.CalcStatus() |
577 | 989a8bee | Michael Hanselmann | job_info = job.GetInfo(self._fields)
|
578 | 989a8bee | Michael Hanselmann | log_entries = job.GetLogEntries(self._prev_log_serial)
|
579 | 6c2549d6 | Guido Trotter | |
580 | 6c2549d6 | Guido Trotter | # Serializing and deserializing data can cause type changes (e.g. from
|
581 | 6c2549d6 | Guido Trotter | # tuple to list) or precision loss. We're doing it here so that we get
|
582 | 6c2549d6 | Guido Trotter | # the same modifications as the data received from the client. Without
|
583 | 6c2549d6 | Guido Trotter | # this, the comparison afterwards might fail without the data being
|
584 | 6c2549d6 | Guido Trotter | # significantly different.
|
585 | 6c2549d6 | Guido Trotter | # TODO: we just deserialized from disk, investigate how to make sure that
|
586 | 6c2549d6 | Guido Trotter | # the job info and log entries are compatible to avoid this further step.
|
587 | 989a8bee | Michael Hanselmann | # TODO: Doing something like in testutils.py:UnifyValueType might be more
|
588 | 989a8bee | Michael Hanselmann | # efficient, though floats will be tricky
|
589 | 989a8bee | Michael Hanselmann | job_info = serializer.LoadJson(serializer.DumpJson(job_info)) |
590 | 989a8bee | Michael Hanselmann | log_entries = serializer.LoadJson(serializer.DumpJson(log_entries)) |
591 | 6c2549d6 | Guido Trotter | |
592 | 6c2549d6 | Guido Trotter | # Don't even try to wait if the job is no longer running, there will be
|
593 | 6c2549d6 | Guido Trotter | # no changes.
|
594 | 989a8bee | Michael Hanselmann | if (status not in (constants.JOB_STATUS_QUEUED, |
595 | 989a8bee | Michael Hanselmann | constants.JOB_STATUS_RUNNING, |
596 | 989a8bee | Michael Hanselmann | constants.JOB_STATUS_WAITLOCK) or
|
597 | 989a8bee | Michael Hanselmann | job_info != self._prev_job_info or |
598 | 989a8bee | Michael Hanselmann | (log_entries and self._prev_log_serial != log_entries[0][0])): |
599 | 989a8bee | Michael Hanselmann | logging.debug("Job %s changed", job.id)
|
600 | 989a8bee | Michael Hanselmann | return (job_info, log_entries)
|
601 | 6c2549d6 | Guido Trotter | |
602 | 989a8bee | Michael Hanselmann | return None |
603 | 989a8bee | Michael Hanselmann | |
604 | 989a8bee | Michael Hanselmann | |
605 | 989a8bee | Michael Hanselmann | class _JobFileChangesWaiter(object): |
606 | 989a8bee | Michael Hanselmann | def __init__(self, filename): |
607 | 989a8bee | Michael Hanselmann | """Initializes this class.
|
608 | 989a8bee | Michael Hanselmann |
|
609 | 989a8bee | Michael Hanselmann | @type filename: string
|
610 | 989a8bee | Michael Hanselmann | @param filename: Path to job file
|
611 | 989a8bee | Michael Hanselmann | @raises errors.InotifyError: if the notifier cannot be setup
|
612 | 6c2549d6 | Guido Trotter |
|
613 | 989a8bee | Michael Hanselmann | """
|
614 | 989a8bee | Michael Hanselmann | self._wm = pyinotify.WatchManager()
|
615 | 989a8bee | Michael Hanselmann | self._inotify_handler = \
|
616 | 989a8bee | Michael Hanselmann | asyncnotifier.SingleFileEventHandler(self._wm, self._OnInotify, filename) |
617 | 989a8bee | Michael Hanselmann | self._notifier = \
|
618 | 989a8bee | Michael Hanselmann | pyinotify.Notifier(self._wm, default_proc_fun=self._inotify_handler) |
619 | 989a8bee | Michael Hanselmann | try:
|
620 | 989a8bee | Michael Hanselmann | self._inotify_handler.enable()
|
621 | 989a8bee | Michael Hanselmann | except Exception: |
622 | 989a8bee | Michael Hanselmann | # pyinotify doesn't close file descriptors automatically
|
623 | 989a8bee | Michael Hanselmann | self._notifier.stop()
|
624 | 989a8bee | Michael Hanselmann | raise
|
625 | 989a8bee | Michael Hanselmann | |
626 | 989a8bee | Michael Hanselmann | def _OnInotify(self, notifier_enabled): |
627 | 989a8bee | Michael Hanselmann | """Callback for inotify.
|
628 | 989a8bee | Michael Hanselmann |
|
629 | 989a8bee | Michael Hanselmann | """
|
630 | 6c2549d6 | Guido Trotter | if not notifier_enabled: |
631 | 989a8bee | Michael Hanselmann | self._inotify_handler.enable()
|
632 | 989a8bee | Michael Hanselmann | |
633 | 989a8bee | Michael Hanselmann | def Wait(self, timeout): |
634 | 989a8bee | Michael Hanselmann | """Waits for the job file to change.
|
635 | 989a8bee | Michael Hanselmann |
|
636 | 989a8bee | Michael Hanselmann | @type timeout: float
|
637 | 989a8bee | Michael Hanselmann | @param timeout: Timeout in seconds
|
638 | 989a8bee | Michael Hanselmann | @return: Whether there have been events
|
639 | 989a8bee | Michael Hanselmann |
|
640 | 989a8bee | Michael Hanselmann | """
|
641 | 989a8bee | Michael Hanselmann | assert timeout >= 0 |
642 | 989a8bee | Michael Hanselmann | have_events = self._notifier.check_events(timeout * 1000) |
643 | 989a8bee | Michael Hanselmann | if have_events:
|
644 | 989a8bee | Michael Hanselmann | self._notifier.read_events()
|
645 | 989a8bee | Michael Hanselmann | self._notifier.process_events()
|
646 | 989a8bee | Michael Hanselmann | return have_events
|
647 | 989a8bee | Michael Hanselmann | |
648 | 989a8bee | Michael Hanselmann | def Close(self): |
649 | 989a8bee | Michael Hanselmann | """Closes underlying notifier and its file descriptor.
|
650 | 989a8bee | Michael Hanselmann |
|
651 | 989a8bee | Michael Hanselmann | """
|
652 | 989a8bee | Michael Hanselmann | self._notifier.stop()
|
653 | 989a8bee | Michael Hanselmann | |
654 | 989a8bee | Michael Hanselmann | |
655 | 989a8bee | Michael Hanselmann | class _JobChangesWaiter(object): |
656 | 989a8bee | Michael Hanselmann | def __init__(self, filename): |
657 | 989a8bee | Michael Hanselmann | """Initializes this class.
|
658 | 989a8bee | Michael Hanselmann |
|
659 | 989a8bee | Michael Hanselmann | @type filename: string
|
660 | 989a8bee | Michael Hanselmann | @param filename: Path to job file
|
661 | 989a8bee | Michael Hanselmann |
|
662 | 989a8bee | Michael Hanselmann | """
|
663 | 989a8bee | Michael Hanselmann | self._filewaiter = None |
664 | 989a8bee | Michael Hanselmann | self._filename = filename
|
665 | 6c2549d6 | Guido Trotter | |
666 | 989a8bee | Michael Hanselmann | def Wait(self, timeout): |
667 | 989a8bee | Michael Hanselmann | """Waits for a job to change.
|
668 | 6c2549d6 | Guido Trotter |
|
669 | 989a8bee | Michael Hanselmann | @type timeout: float
|
670 | 989a8bee | Michael Hanselmann | @param timeout: Timeout in seconds
|
671 | 989a8bee | Michael Hanselmann | @return: Whether there have been events
|
672 | 989a8bee | Michael Hanselmann |
|
673 | 989a8bee | Michael Hanselmann | """
|
674 | 989a8bee | Michael Hanselmann | if self._filewaiter: |
675 | 989a8bee | Michael Hanselmann | return self._filewaiter.Wait(timeout) |
676 | 989a8bee | Michael Hanselmann | |
677 | 989a8bee | Michael Hanselmann | # Lazy setup: Avoid inotify setup cost when job file has already changed.
|
678 | 989a8bee | Michael Hanselmann | # If this point is reached, return immediately and let caller check the job
|
679 | 989a8bee | Michael Hanselmann | # file again in case there were changes since the last check. This avoids a
|
680 | 989a8bee | Michael Hanselmann | # race condition.
|
681 | 989a8bee | Michael Hanselmann | self._filewaiter = _JobFileChangesWaiter(self._filename) |
682 | 989a8bee | Michael Hanselmann | |
683 | 989a8bee | Michael Hanselmann | return True |
684 | 989a8bee | Michael Hanselmann | |
685 | 989a8bee | Michael Hanselmann | def Close(self): |
686 | 989a8bee | Michael Hanselmann | """Closes underlying waiter.
|
687 | 989a8bee | Michael Hanselmann |
|
688 | 989a8bee | Michael Hanselmann | """
|
689 | 989a8bee | Michael Hanselmann | if self._filewaiter: |
690 | 989a8bee | Michael Hanselmann | self._filewaiter.Close()
|
691 | 989a8bee | Michael Hanselmann | |
692 | 989a8bee | Michael Hanselmann | |
693 | 989a8bee | Michael Hanselmann | class _WaitForJobChangesHelper(object): |
694 | 989a8bee | Michael Hanselmann | """Helper class using inotify to wait for changes in a job file.
|
695 | 989a8bee | Michael Hanselmann |
|
696 | 989a8bee | Michael Hanselmann | This class takes a previous job status and serial, and alerts the client when
|
697 | 989a8bee | Michael Hanselmann | the current job status has changed.
|
698 | 989a8bee | Michael Hanselmann |
|
699 | 989a8bee | Michael Hanselmann | """
|
700 | 989a8bee | Michael Hanselmann | @staticmethod
|
701 | 989a8bee | Michael Hanselmann | def _CheckForChanges(job_load_fn, check_fn): |
702 | 989a8bee | Michael Hanselmann | job = job_load_fn() |
703 | 989a8bee | Michael Hanselmann | if not job: |
704 | 989a8bee | Michael Hanselmann | raise errors.JobLost()
|
705 | 989a8bee | Michael Hanselmann | |
706 | 989a8bee | Michael Hanselmann | result = check_fn(job) |
707 | 989a8bee | Michael Hanselmann | if result is None: |
708 | 989a8bee | Michael Hanselmann | raise utils.RetryAgain()
|
709 | 989a8bee | Michael Hanselmann | |
710 | 989a8bee | Michael Hanselmann | return result
|
711 | 989a8bee | Michael Hanselmann | |
712 | 989a8bee | Michael Hanselmann | def __call__(self, filename, job_load_fn, |
713 | 989a8bee | Michael Hanselmann | fields, prev_job_info, prev_log_serial, timeout): |
714 | 989a8bee | Michael Hanselmann | """Waits for changes on a job.
|
715 | 989a8bee | Michael Hanselmann |
|
716 | 989a8bee | Michael Hanselmann | @type filename: string
|
717 | 989a8bee | Michael Hanselmann | @param filename: File on which to wait for changes
|
718 | 989a8bee | Michael Hanselmann | @type job_load_fn: callable
|
719 | 989a8bee | Michael Hanselmann | @param job_load_fn: Function to load job
|
720 | 989a8bee | Michael Hanselmann | @type fields: list of strings
|
721 | 989a8bee | Michael Hanselmann | @param fields: Which fields to check for changes
|
722 | 989a8bee | Michael Hanselmann | @type prev_job_info: list or None
|
723 | 989a8bee | Michael Hanselmann | @param prev_job_info: Last job information returned
|
724 | 989a8bee | Michael Hanselmann | @type prev_log_serial: int
|
725 | 989a8bee | Michael Hanselmann | @param prev_log_serial: Last job message serial number
|
726 | 989a8bee | Michael Hanselmann | @type timeout: float
|
727 | 989a8bee | Michael Hanselmann | @param timeout: maximum time to wait in seconds
|
728 | 989a8bee | Michael Hanselmann |
|
729 | 989a8bee | Michael Hanselmann | """
|
730 | 6c2549d6 | Guido Trotter | try:
|
731 | 989a8bee | Michael Hanselmann | check_fn = _JobChangesChecker(fields, prev_job_info, prev_log_serial) |
732 | 989a8bee | Michael Hanselmann | waiter = _JobChangesWaiter(filename) |
733 | 989a8bee | Michael Hanselmann | try:
|
734 | 989a8bee | Michael Hanselmann | return utils.Retry(compat.partial(self._CheckForChanges, |
735 | 989a8bee | Michael Hanselmann | job_load_fn, check_fn), |
736 | 989a8bee | Michael Hanselmann | utils.RETRY_REMAINING_TIME, timeout, |
737 | 989a8bee | Michael Hanselmann | wait_fn=waiter.Wait) |
738 | 989a8bee | Michael Hanselmann | finally:
|
739 | 989a8bee | Michael Hanselmann | waiter.Close() |
740 | 6c2549d6 | Guido Trotter | except (errors.InotifyError, errors.JobLost):
|
741 | 6c2549d6 | Guido Trotter | return None |
742 | 6c2549d6 | Guido Trotter | except utils.RetryTimeout:
|
743 | 6c2549d6 | Guido Trotter | return constants.JOB_NOTCHANGED
|
744 | 6c2549d6 | Guido Trotter | |
745 | 6c2549d6 | Guido Trotter | |
746 | 6760e4ed | Michael Hanselmann | def _EncodeOpError(err): |
747 | 6760e4ed | Michael Hanselmann | """Encodes an error which occurred while processing an opcode.
|
748 | 6760e4ed | Michael Hanselmann |
|
749 | 6760e4ed | Michael Hanselmann | """
|
750 | 6760e4ed | Michael Hanselmann | if isinstance(err, errors.GenericError): |
751 | 6760e4ed | Michael Hanselmann | to_encode = err |
752 | 6760e4ed | Michael Hanselmann | else:
|
753 | 6760e4ed | Michael Hanselmann | to_encode = errors.OpExecError(str(err))
|
754 | 6760e4ed | Michael Hanselmann | |
755 | 6760e4ed | Michael Hanselmann | return errors.EncodeException(to_encode)
|
756 | 6760e4ed | Michael Hanselmann | |
757 | 6760e4ed | Michael Hanselmann | |
758 | 26d3fd2f | Michael Hanselmann | class _TimeoutStrategyWrapper: |
759 | 26d3fd2f | Michael Hanselmann | def __init__(self, fn): |
760 | 26d3fd2f | Michael Hanselmann | """Initializes this class.
|
761 | 26d3fd2f | Michael Hanselmann |
|
762 | 26d3fd2f | Michael Hanselmann | """
|
763 | 26d3fd2f | Michael Hanselmann | self._fn = fn
|
764 | 26d3fd2f | Michael Hanselmann | self._next = None |
765 | 26d3fd2f | Michael Hanselmann | |
766 | 26d3fd2f | Michael Hanselmann | def _Advance(self): |
767 | 26d3fd2f | Michael Hanselmann | """Gets the next timeout if necessary.
|
768 | 26d3fd2f | Michael Hanselmann |
|
769 | 26d3fd2f | Michael Hanselmann | """
|
770 | 26d3fd2f | Michael Hanselmann | if self._next is None: |
771 | 26d3fd2f | Michael Hanselmann | self._next = self._fn() |
772 | 26d3fd2f | Michael Hanselmann | |
773 | 26d3fd2f | Michael Hanselmann | def Peek(self): |
774 | 26d3fd2f | Michael Hanselmann | """Returns the next timeout.
|
775 | 26d3fd2f | Michael Hanselmann |
|
776 | 26d3fd2f | Michael Hanselmann | """
|
777 | 26d3fd2f | Michael Hanselmann | self._Advance()
|
778 | 26d3fd2f | Michael Hanselmann | return self._next |
779 | 26d3fd2f | Michael Hanselmann | |
780 | 26d3fd2f | Michael Hanselmann | def Next(self): |
781 | 26d3fd2f | Michael Hanselmann | """Returns the current timeout and advances the internal state.
|
782 | 26d3fd2f | Michael Hanselmann |
|
783 | 26d3fd2f | Michael Hanselmann | """
|
784 | 26d3fd2f | Michael Hanselmann | self._Advance()
|
785 | 26d3fd2f | Michael Hanselmann | result = self._next
|
786 | 26d3fd2f | Michael Hanselmann | self._next = None |
787 | 26d3fd2f | Michael Hanselmann | return result
|
788 | 26d3fd2f | Michael Hanselmann | |
789 | 26d3fd2f | Michael Hanselmann | |
790 | b80cc518 | Michael Hanselmann | class _OpExecContext: |
791 | 26d3fd2f | Michael Hanselmann | def __init__(self, op, index, log_prefix, timeout_strategy_factory): |
792 | b80cc518 | Michael Hanselmann | """Initializes this class.
|
793 | b80cc518 | Michael Hanselmann |
|
794 | b80cc518 | Michael Hanselmann | """
|
795 | b80cc518 | Michael Hanselmann | self.op = op
|
796 | b80cc518 | Michael Hanselmann | self.index = index
|
797 | b80cc518 | Michael Hanselmann | self.log_prefix = log_prefix
|
798 | b80cc518 | Michael Hanselmann | self.summary = op.input.Summary()
|
799 | b80cc518 | Michael Hanselmann | |
800 | 26d3fd2f | Michael Hanselmann | self._timeout_strategy_factory = timeout_strategy_factory
|
801 | 26d3fd2f | Michael Hanselmann | self._ResetTimeoutStrategy()
|
802 | 26d3fd2f | Michael Hanselmann | |
803 | 26d3fd2f | Michael Hanselmann | def _ResetTimeoutStrategy(self): |
804 | 26d3fd2f | Michael Hanselmann | """Creates a new timeout strategy.
|
805 | 26d3fd2f | Michael Hanselmann |
|
806 | 26d3fd2f | Michael Hanselmann | """
|
807 | 26d3fd2f | Michael Hanselmann | self._timeout_strategy = \
|
808 | 26d3fd2f | Michael Hanselmann | _TimeoutStrategyWrapper(self._timeout_strategy_factory().NextAttempt)
|
809 | 26d3fd2f | Michael Hanselmann | |
810 | 26d3fd2f | Michael Hanselmann | def CheckPriorityIncrease(self): |
811 | 26d3fd2f | Michael Hanselmann | """Checks whether priority can and should be increased.
|
812 | 26d3fd2f | Michael Hanselmann |
|
813 | 26d3fd2f | Michael Hanselmann | Called when locks couldn't be acquired.
|
814 | 26d3fd2f | Michael Hanselmann |
|
815 | 26d3fd2f | Michael Hanselmann | """
|
816 | 26d3fd2f | Michael Hanselmann | op = self.op
|
817 | 26d3fd2f | Michael Hanselmann | |
818 | 26d3fd2f | Michael Hanselmann | # Exhausted all retries and next round should not use blocking acquire
|
819 | 26d3fd2f | Michael Hanselmann | # for locks?
|
820 | 26d3fd2f | Michael Hanselmann | if (self._timeout_strategy.Peek() is None and |
821 | 26d3fd2f | Michael Hanselmann | op.priority > constants.OP_PRIO_HIGHEST): |
822 | 26d3fd2f | Michael Hanselmann | logging.debug("Increasing priority")
|
823 | 26d3fd2f | Michael Hanselmann | op.priority -= 1
|
824 | 26d3fd2f | Michael Hanselmann | self._ResetTimeoutStrategy()
|
825 | 26d3fd2f | Michael Hanselmann | return True |
826 | 26d3fd2f | Michael Hanselmann | |
827 | 26d3fd2f | Michael Hanselmann | return False |
828 | 26d3fd2f | Michael Hanselmann | |
829 | 26d3fd2f | Michael Hanselmann | def GetNextLockTimeout(self): |
830 | 26d3fd2f | Michael Hanselmann | """Returns the next lock acquire timeout.
|
831 | 26d3fd2f | Michael Hanselmann |
|
832 | 26d3fd2f | Michael Hanselmann | """
|
833 | 26d3fd2f | Michael Hanselmann | return self._timeout_strategy.Next() |
834 | 26d3fd2f | Michael Hanselmann | |
835 | b80cc518 | Michael Hanselmann | |
836 | be760ba8 | Michael Hanselmann | class _JobProcessor(object): |
837 | 26d3fd2f | Michael Hanselmann | def __init__(self, queue, opexec_fn, job, |
838 | 26d3fd2f | Michael Hanselmann | _timeout_strategy_factory=mcpu.LockAttemptTimeoutStrategy): |
839 | be760ba8 | Michael Hanselmann | """Initializes this class.
|
840 | be760ba8 | Michael Hanselmann |
|
841 | be760ba8 | Michael Hanselmann | """
|
842 | be760ba8 | Michael Hanselmann | self.queue = queue
|
843 | be760ba8 | Michael Hanselmann | self.opexec_fn = opexec_fn
|
844 | be760ba8 | Michael Hanselmann | self.job = job
|
845 | 26d3fd2f | Michael Hanselmann | self._timeout_strategy_factory = _timeout_strategy_factory
|
846 | be760ba8 | Michael Hanselmann | |
847 | be760ba8 | Michael Hanselmann | @staticmethod
|
848 | 26d3fd2f | Michael Hanselmann | def _FindNextOpcode(job, timeout_strategy_factory): |
849 | be760ba8 | Michael Hanselmann | """Locates the next opcode to run.
|
850 | be760ba8 | Michael Hanselmann |
|
851 | be760ba8 | Michael Hanselmann | @type job: L{_QueuedJob}
|
852 | be760ba8 | Michael Hanselmann | @param job: Job object
|
853 | 26d3fd2f | Michael Hanselmann | @param timeout_strategy_factory: Callable to create new timeout strategy
|
854 | be760ba8 | Michael Hanselmann |
|
855 | be760ba8 | Michael Hanselmann | """
|
856 | be760ba8 | Michael Hanselmann | # Create some sort of a cache to speed up locating next opcode for future
|
857 | be760ba8 | Michael Hanselmann | # lookups
|
858 | be760ba8 | Michael Hanselmann | # TODO: Consider splitting _QueuedJob.ops into two separate lists, one for
|
859 | be760ba8 | Michael Hanselmann | # pending and one for processed ops.
|
860 | 03b63608 | Michael Hanselmann | if job.ops_iter is None: |
861 | 03b63608 | Michael Hanselmann | job.ops_iter = enumerate(job.ops)
|
862 | be760ba8 | Michael Hanselmann | |
863 | be760ba8 | Michael Hanselmann | # Find next opcode to run
|
864 | be760ba8 | Michael Hanselmann | while True: |
865 | be760ba8 | Michael Hanselmann | try:
|
866 | 03b63608 | Michael Hanselmann | (idx, op) = job.ops_iter.next() |
867 | be760ba8 | Michael Hanselmann | except StopIteration: |
868 | be760ba8 | Michael Hanselmann | raise errors.ProgrammerError("Called for a finished job") |
869 | be760ba8 | Michael Hanselmann | |
870 | be760ba8 | Michael Hanselmann | if op.status == constants.OP_STATUS_RUNNING:
|
871 | be760ba8 | Michael Hanselmann | # Found an opcode already marked as running
|
872 | be760ba8 | Michael Hanselmann | raise errors.ProgrammerError("Called for job marked as running") |
873 | be760ba8 | Michael Hanselmann | |
874 | 26d3fd2f | Michael Hanselmann | opctx = _OpExecContext(op, idx, "Op %s/%s" % (idx + 1, len(job.ops)), |
875 | 26d3fd2f | Michael Hanselmann | timeout_strategy_factory) |
876 | be760ba8 | Michael Hanselmann | |
877 | be760ba8 | Michael Hanselmann | if op.status == constants.OP_STATUS_CANCELED:
|
878 | be760ba8 | Michael Hanselmann | # Cancelled jobs are handled by the caller
|
879 | be760ba8 | Michael Hanselmann | assert not compat.any(i.status != constants.OP_STATUS_CANCELED |
880 | be760ba8 | Michael Hanselmann | for i in job.ops[idx:]) |
881 | be760ba8 | Michael Hanselmann | |
882 | be760ba8 | Michael Hanselmann | elif op.status in constants.OPS_FINALIZED: |
883 | be760ba8 | Michael Hanselmann | # This is a job that was partially completed before master daemon
|
884 | be760ba8 | Michael Hanselmann | # shutdown, so it can be expected that some opcodes are already
|
885 | be760ba8 | Michael Hanselmann | # completed successfully (if any did error out, then the whole job
|
886 | be760ba8 | Michael Hanselmann | # should have been aborted and not resubmitted for processing).
|
887 | be760ba8 | Michael Hanselmann | logging.info("%s: opcode %s already processed, skipping",
|
888 | b80cc518 | Michael Hanselmann | opctx.log_prefix, opctx.summary) |
889 | be760ba8 | Michael Hanselmann | continue
|
890 | be760ba8 | Michael Hanselmann | |
891 | b80cc518 | Michael Hanselmann | return opctx
|
892 | be760ba8 | Michael Hanselmann | |
893 | be760ba8 | Michael Hanselmann | @staticmethod
|
894 | be760ba8 | Michael Hanselmann | def _MarkWaitlock(job, op): |
895 | be760ba8 | Michael Hanselmann | """Marks an opcode as waiting for locks.
|
896 | be760ba8 | Michael Hanselmann |
|
897 | be760ba8 | Michael Hanselmann | The job's start timestamp is also set if necessary.
|
898 | be760ba8 | Michael Hanselmann |
|
899 | be760ba8 | Michael Hanselmann | @type job: L{_QueuedJob}
|
900 | be760ba8 | Michael Hanselmann | @param job: Job object
|
901 | a38e8674 | Michael Hanselmann | @type op: L{_QueuedOpCode}
|
902 | a38e8674 | Michael Hanselmann | @param op: Opcode object
|
903 | be760ba8 | Michael Hanselmann |
|
904 | be760ba8 | Michael Hanselmann | """
|
905 | be760ba8 | Michael Hanselmann | assert op in job.ops |
906 | 5fd6b694 | Michael Hanselmann | assert op.status in (constants.OP_STATUS_QUEUED, |
907 | 5fd6b694 | Michael Hanselmann | constants.OP_STATUS_WAITLOCK) |
908 | 5fd6b694 | Michael Hanselmann | |
909 | 5fd6b694 | Michael Hanselmann | update = False
|
910 | be760ba8 | Michael Hanselmann | |
911 | be760ba8 | Michael Hanselmann | op.result = None
|
912 | 5fd6b694 | Michael Hanselmann | |
913 | 5fd6b694 | Michael Hanselmann | if op.status == constants.OP_STATUS_QUEUED:
|
914 | 5fd6b694 | Michael Hanselmann | op.status = constants.OP_STATUS_WAITLOCK |
915 | 5fd6b694 | Michael Hanselmann | update = True
|
916 | 5fd6b694 | Michael Hanselmann | |
917 | 5fd6b694 | Michael Hanselmann | if op.start_timestamp is None: |
918 | 5fd6b694 | Michael Hanselmann | op.start_timestamp = TimeStampNow() |
919 | 5fd6b694 | Michael Hanselmann | update = True
|
920 | be760ba8 | Michael Hanselmann | |
921 | be760ba8 | Michael Hanselmann | if job.start_timestamp is None: |
922 | be760ba8 | Michael Hanselmann | job.start_timestamp = op.start_timestamp |
923 | 5fd6b694 | Michael Hanselmann | update = True
|
924 | 5fd6b694 | Michael Hanselmann | |
925 | 5fd6b694 | Michael Hanselmann | assert op.status == constants.OP_STATUS_WAITLOCK
|
926 | 5fd6b694 | Michael Hanselmann | |
927 | 5fd6b694 | Michael Hanselmann | return update
|
928 | be760ba8 | Michael Hanselmann | |
929 | b80cc518 | Michael Hanselmann | def _ExecOpCodeUnlocked(self, opctx): |
930 | be760ba8 | Michael Hanselmann | """Processes one opcode and returns the result.
|
931 | be760ba8 | Michael Hanselmann |
|
932 | be760ba8 | Michael Hanselmann | """
|
933 | b80cc518 | Michael Hanselmann | op = opctx.op |
934 | b80cc518 | Michael Hanselmann | |
935 | be760ba8 | Michael Hanselmann | assert op.status == constants.OP_STATUS_WAITLOCK
|
936 | be760ba8 | Michael Hanselmann | |
937 | 26d3fd2f | Michael Hanselmann | timeout = opctx.GetNextLockTimeout() |
938 | 26d3fd2f | Michael Hanselmann | |
939 | be760ba8 | Michael Hanselmann | try:
|
940 | be760ba8 | Michael Hanselmann | # Make sure not to hold queue lock while calling ExecOpCode
|
941 | be760ba8 | Michael Hanselmann | result = self.opexec_fn(op.input,
|
942 | 26d3fd2f | Michael Hanselmann | _OpExecCallbacks(self.queue, self.job, op), |
943 | f23db633 | Michael Hanselmann | timeout=timeout, priority=op.priority) |
944 | 26d3fd2f | Michael Hanselmann | except mcpu.LockAcquireTimeout:
|
945 | 26d3fd2f | Michael Hanselmann | assert timeout is not None, "Received timeout for blocking acquire" |
946 | 26d3fd2f | Michael Hanselmann | logging.debug("Couldn't acquire locks in %0.6fs", timeout)
|
947 | 9e49dfc5 | Michael Hanselmann | |
948 | 9e49dfc5 | Michael Hanselmann | assert op.status in (constants.OP_STATUS_WAITLOCK, |
949 | 9e49dfc5 | Michael Hanselmann | constants.OP_STATUS_CANCELING) |
950 | 9e49dfc5 | Michael Hanselmann | |
951 | 9e49dfc5 | Michael Hanselmann | # Was job cancelled while we were waiting for the lock?
|
952 | 9e49dfc5 | Michael Hanselmann | if op.status == constants.OP_STATUS_CANCELING:
|
953 | 9e49dfc5 | Michael Hanselmann | return (constants.OP_STATUS_CANCELING, None) |
954 | 9e49dfc5 | Michael Hanselmann | |
955 | 5fd6b694 | Michael Hanselmann | # Stay in waitlock while trying to re-acquire lock
|
956 | 5fd6b694 | Michael Hanselmann | return (constants.OP_STATUS_WAITLOCK, None) |
957 | be760ba8 | Michael Hanselmann | except CancelJob:
|
958 | b80cc518 | Michael Hanselmann | logging.exception("%s: Canceling job", opctx.log_prefix)
|
959 | be760ba8 | Michael Hanselmann | assert op.status == constants.OP_STATUS_CANCELING
|
960 | be760ba8 | Michael Hanselmann | return (constants.OP_STATUS_CANCELING, None) |
961 | be760ba8 | Michael Hanselmann | except Exception, err: # pylint: disable-msg=W0703 |
962 | b80cc518 | Michael Hanselmann | logging.exception("%s: Caught exception in %s",
|
963 | b80cc518 | Michael Hanselmann | opctx.log_prefix, opctx.summary) |
964 | be760ba8 | Michael Hanselmann | return (constants.OP_STATUS_ERROR, _EncodeOpError(err))
|
965 | be760ba8 | Michael Hanselmann | else:
|
966 | b80cc518 | Michael Hanselmann | logging.debug("%s: %s successful",
|
967 | b80cc518 | Michael Hanselmann | opctx.log_prefix, opctx.summary) |
968 | be760ba8 | Michael Hanselmann | return (constants.OP_STATUS_SUCCESS, result)
|
969 | be760ba8 | Michael Hanselmann | |
970 | 26d3fd2f | Michael Hanselmann | def __call__(self, _nextop_fn=None): |
971 | be760ba8 | Michael Hanselmann | """Continues execution of a job.
|
972 | be760ba8 | Michael Hanselmann |
|
973 | 26d3fd2f | Michael Hanselmann | @param _nextop_fn: Callback function for tests
|
974 | be760ba8 | Michael Hanselmann | @rtype: bool
|
975 | be760ba8 | Michael Hanselmann | @return: True if job is finished, False if processor needs to be called
|
976 | be760ba8 | Michael Hanselmann | again
|
977 | be760ba8 | Michael Hanselmann |
|
978 | be760ba8 | Michael Hanselmann | """
|
979 | be760ba8 | Michael Hanselmann | queue = self.queue
|
980 | be760ba8 | Michael Hanselmann | job = self.job
|
981 | be760ba8 | Michael Hanselmann | |
982 | be760ba8 | Michael Hanselmann | logging.debug("Processing job %s", job.id)
|
983 | be760ba8 | Michael Hanselmann | |
984 | be760ba8 | Michael Hanselmann | queue.acquire(shared=1)
|
985 | be760ba8 | Michael Hanselmann | try:
|
986 | be760ba8 | Michael Hanselmann | opcount = len(job.ops)
|
987 | be760ba8 | Michael Hanselmann | |
988 | 26d3fd2f | Michael Hanselmann | # Is a previous opcode still pending?
|
989 | 26d3fd2f | Michael Hanselmann | if job.cur_opctx:
|
990 | 26d3fd2f | Michael Hanselmann | opctx = job.cur_opctx |
991 | 5fd6b694 | Michael Hanselmann | job.cur_opctx = None
|
992 | 26d3fd2f | Michael Hanselmann | else:
|
993 | 26d3fd2f | Michael Hanselmann | if __debug__ and _nextop_fn: |
994 | 26d3fd2f | Michael Hanselmann | _nextop_fn() |
995 | 26d3fd2f | Michael Hanselmann | opctx = self._FindNextOpcode(job, self._timeout_strategy_factory) |
996 | 26d3fd2f | Michael Hanselmann | |
997 | b80cc518 | Michael Hanselmann | op = opctx.op |
998 | be760ba8 | Michael Hanselmann | |
999 | be760ba8 | Michael Hanselmann | # Consistency check
|
1000 | be760ba8 | Michael Hanselmann | assert compat.all(i.status in (constants.OP_STATUS_QUEUED, |
1001 | 30c945d0 | Michael Hanselmann | constants.OP_STATUS_CANCELING, |
1002 | be760ba8 | Michael Hanselmann | constants.OP_STATUS_CANCELED) |
1003 | 5fd6b694 | Michael Hanselmann | for i in job.ops[opctx.index + 1:]) |
1004 | be760ba8 | Michael Hanselmann | |
1005 | be760ba8 | Michael Hanselmann | assert op.status in (constants.OP_STATUS_QUEUED, |
1006 | be760ba8 | Michael Hanselmann | constants.OP_STATUS_WAITLOCK, |
1007 | 30c945d0 | Michael Hanselmann | constants.OP_STATUS_CANCELING, |
1008 | be760ba8 | Michael Hanselmann | constants.OP_STATUS_CANCELED) |
1009 | be760ba8 | Michael Hanselmann | |
1010 | 26d3fd2f | Michael Hanselmann | assert (op.priority <= constants.OP_PRIO_LOWEST and |
1011 | 26d3fd2f | Michael Hanselmann | op.priority >= constants.OP_PRIO_HIGHEST) |
1012 | 26d3fd2f | Michael Hanselmann | |
1013 | 30c945d0 | Michael Hanselmann | if op.status not in (constants.OP_STATUS_CANCELING, |
1014 | 30c945d0 | Michael Hanselmann | constants.OP_STATUS_CANCELED): |
1015 | 30c945d0 | Michael Hanselmann | assert op.status in (constants.OP_STATUS_QUEUED, |
1016 | 30c945d0 | Michael Hanselmann | constants.OP_STATUS_WAITLOCK) |
1017 | 30c945d0 | Michael Hanselmann | |
1018 | be760ba8 | Michael Hanselmann | # Prepare to start opcode
|
1019 | 5fd6b694 | Michael Hanselmann | if self._MarkWaitlock(job, op): |
1020 | 5fd6b694 | Michael Hanselmann | # Write to disk
|
1021 | 5fd6b694 | Michael Hanselmann | queue.UpdateJobUnlocked(job) |
1022 | be760ba8 | Michael Hanselmann | |
1023 | be760ba8 | Michael Hanselmann | assert op.status == constants.OP_STATUS_WAITLOCK
|
1024 | be760ba8 | Michael Hanselmann | assert job.CalcStatus() == constants.JOB_STATUS_WAITLOCK
|
1025 | 5fd6b694 | Michael Hanselmann | assert job.start_timestamp and op.start_timestamp |
1026 | be760ba8 | Michael Hanselmann | |
1027 | b80cc518 | Michael Hanselmann | logging.info("%s: opcode %s waiting for locks",
|
1028 | b80cc518 | Michael Hanselmann | opctx.log_prefix, opctx.summary) |
1029 | be760ba8 | Michael Hanselmann | |
1030 | be760ba8 | Michael Hanselmann | queue.release() |
1031 | be760ba8 | Michael Hanselmann | try:
|
1032 | b80cc518 | Michael Hanselmann | (op_status, op_result) = self._ExecOpCodeUnlocked(opctx)
|
1033 | be760ba8 | Michael Hanselmann | finally:
|
1034 | be760ba8 | Michael Hanselmann | queue.acquire(shared=1)
|
1035 | be760ba8 | Michael Hanselmann | |
1036 | be760ba8 | Michael Hanselmann | op.status = op_status |
1037 | be760ba8 | Michael Hanselmann | op.result = op_result |
1038 | be760ba8 | Michael Hanselmann | |
1039 | 5fd6b694 | Michael Hanselmann | if op.status == constants.OP_STATUS_WAITLOCK:
|
1040 | 26d3fd2f | Michael Hanselmann | # Couldn't get locks in time
|
1041 | 26d3fd2f | Michael Hanselmann | assert not op.end_timestamp |
1042 | be760ba8 | Michael Hanselmann | else:
|
1043 | 26d3fd2f | Michael Hanselmann | # Finalize opcode
|
1044 | 26d3fd2f | Michael Hanselmann | op.end_timestamp = TimeStampNow() |
1045 | be760ba8 | Michael Hanselmann | |
1046 | 26d3fd2f | Michael Hanselmann | if op.status == constants.OP_STATUS_CANCELING:
|
1047 | 26d3fd2f | Michael Hanselmann | assert not compat.any(i.status != constants.OP_STATUS_CANCELING |
1048 | 26d3fd2f | Michael Hanselmann | for i in job.ops[opctx.index:]) |
1049 | 26d3fd2f | Michael Hanselmann | else:
|
1050 | 26d3fd2f | Michael Hanselmann | assert op.status in constants.OPS_FINALIZED |
1051 | be760ba8 | Michael Hanselmann | |
1052 | 5fd6b694 | Michael Hanselmann | if op.status == constants.OP_STATUS_WAITLOCK:
|
1053 | be760ba8 | Michael Hanselmann | finalize = False
|
1054 | be760ba8 | Michael Hanselmann | |
1055 | 5fd6b694 | Michael Hanselmann | if opctx.CheckPriorityIncrease():
|
1056 | 5fd6b694 | Michael Hanselmann | # Priority was changed, need to update on-disk file
|
1057 | 5fd6b694 | Michael Hanselmann | queue.UpdateJobUnlocked(job) |
1058 | be760ba8 | Michael Hanselmann | |
1059 | 26d3fd2f | Michael Hanselmann | # Keep around for another round
|
1060 | 26d3fd2f | Michael Hanselmann | job.cur_opctx = opctx |
1061 | be760ba8 | Michael Hanselmann | |
1062 | 26d3fd2f | Michael Hanselmann | assert (op.priority <= constants.OP_PRIO_LOWEST and |
1063 | 26d3fd2f | Michael Hanselmann | op.priority >= constants.OP_PRIO_HIGHEST) |
1064 | be760ba8 | Michael Hanselmann | |
1065 | 26d3fd2f | Michael Hanselmann | # In no case must the status be finalized here
|
1066 | 5fd6b694 | Michael Hanselmann | assert job.CalcStatus() == constants.JOB_STATUS_WAITLOCK
|
1067 | be760ba8 | Michael Hanselmann | |
1068 | be760ba8 | Michael Hanselmann | else:
|
1069 | 26d3fd2f | Michael Hanselmann | # Ensure all opcodes so far have been successful
|
1070 | 26d3fd2f | Michael Hanselmann | assert (opctx.index == 0 or |
1071 | 26d3fd2f | Michael Hanselmann | compat.all(i.status == constants.OP_STATUS_SUCCESS |
1072 | 26d3fd2f | Michael Hanselmann | for i in job.ops[:opctx.index])) |
1073 | 26d3fd2f | Michael Hanselmann | |
1074 | 26d3fd2f | Michael Hanselmann | # Reset context
|
1075 | 26d3fd2f | Michael Hanselmann | job.cur_opctx = None
|
1076 | 26d3fd2f | Michael Hanselmann | |
1077 | 26d3fd2f | Michael Hanselmann | if op.status == constants.OP_STATUS_SUCCESS:
|
1078 | 26d3fd2f | Michael Hanselmann | finalize = False
|
1079 | 26d3fd2f | Michael Hanselmann | |
1080 | 26d3fd2f | Michael Hanselmann | elif op.status == constants.OP_STATUS_ERROR:
|
1081 | 26d3fd2f | Michael Hanselmann | # Ensure failed opcode has an exception as its result
|
1082 | 26d3fd2f | Michael Hanselmann | assert errors.GetEncodedError(job.ops[opctx.index].result)
|
1083 | 26d3fd2f | Michael Hanselmann | |
1084 | 26d3fd2f | Michael Hanselmann | to_encode = errors.OpExecError("Preceding opcode failed")
|
1085 | 26d3fd2f | Michael Hanselmann | job.MarkUnfinishedOps(constants.OP_STATUS_ERROR, |
1086 | 26d3fd2f | Michael Hanselmann | _EncodeOpError(to_encode)) |
1087 | 26d3fd2f | Michael Hanselmann | finalize = True
|
1088 | be760ba8 | Michael Hanselmann | |
1089 | 26d3fd2f | Michael Hanselmann | # Consistency check
|
1090 | 26d3fd2f | Michael Hanselmann | assert compat.all(i.status == constants.OP_STATUS_ERROR and |
1091 | 26d3fd2f | Michael Hanselmann | errors.GetEncodedError(i.result) |
1092 | 26d3fd2f | Michael Hanselmann | for i in job.ops[opctx.index:]) |
1093 | be760ba8 | Michael Hanselmann | |
1094 | 26d3fd2f | Michael Hanselmann | elif op.status == constants.OP_STATUS_CANCELING:
|
1095 | 26d3fd2f | Michael Hanselmann | job.MarkUnfinishedOps(constants.OP_STATUS_CANCELED, |
1096 | 26d3fd2f | Michael Hanselmann | "Job canceled by request")
|
1097 | 26d3fd2f | Michael Hanselmann | finalize = True
|
1098 | 26d3fd2f | Michael Hanselmann | |
1099 | 26d3fd2f | Michael Hanselmann | elif op.status == constants.OP_STATUS_CANCELED:
|
1100 | 26d3fd2f | Michael Hanselmann | finalize = True
|
1101 | 26d3fd2f | Michael Hanselmann | |
1102 | 26d3fd2f | Michael Hanselmann | else:
|
1103 | 26d3fd2f | Michael Hanselmann | raise errors.ProgrammerError("Unknown status '%s'" % op.status) |
1104 | 26d3fd2f | Michael Hanselmann | |
1105 | 26d3fd2f | Michael Hanselmann | # Finalizing or last opcode?
|
1106 | 26d3fd2f | Michael Hanselmann | if finalize or opctx.index == (opcount - 1): |
1107 | 26d3fd2f | Michael Hanselmann | # All opcodes have been run, finalize job
|
1108 | 26d3fd2f | Michael Hanselmann | job.end_timestamp = TimeStampNow() |
1109 | 26d3fd2f | Michael Hanselmann | |
1110 | 26d3fd2f | Michael Hanselmann | # Write to disk. If the job status is final, this is the final write
|
1111 | 26d3fd2f | Michael Hanselmann | # allowed. Once the file has been written, it can be archived anytime.
|
1112 | 26d3fd2f | Michael Hanselmann | queue.UpdateJobUnlocked(job) |
1113 | be760ba8 | Michael Hanselmann | |
1114 | 26d3fd2f | Michael Hanselmann | if finalize or opctx.index == (opcount - 1): |
1115 | 26d3fd2f | Michael Hanselmann | logging.info("Finished job %s, status = %s", job.id, job.CalcStatus())
|
1116 | 26d3fd2f | Michael Hanselmann | return True |
1117 | be760ba8 | Michael Hanselmann | |
1118 | be760ba8 | Michael Hanselmann | return False |
1119 | be760ba8 | Michael Hanselmann | finally:
|
1120 | be760ba8 | Michael Hanselmann | queue.release() |
1121 | be760ba8 | Michael Hanselmann | |
1122 | be760ba8 | Michael Hanselmann | |
1123 | 031a3e57 | Michael Hanselmann | class _JobQueueWorker(workerpool.BaseWorker): |
1124 | 031a3e57 | Michael Hanselmann | """The actual job workers.
|
1125 | 031a3e57 | Michael Hanselmann |
|
1126 | 031a3e57 | Michael Hanselmann | """
|
1127 | 7260cfbe | Iustin Pop | def RunTask(self, job): # pylint: disable-msg=W0221 |
1128 | e2715f69 | Michael Hanselmann | """Job executor.
|
1129 | e2715f69 | Michael Hanselmann |
|
1130 | be760ba8 | Michael Hanselmann | This functions processes a job. It is closely tied to the L{_QueuedJob} and
|
1131 | be760ba8 | Michael Hanselmann | L{_QueuedOpCode} classes.
|
1132 | e2715f69 | Michael Hanselmann |
|
1133 | ea03467c | Iustin Pop | @type job: L{_QueuedJob}
|
1134 | ea03467c | Iustin Pop | @param job: the job to be processed
|
1135 | ea03467c | Iustin Pop |
|
1136 | e2715f69 | Michael Hanselmann | """
|
1137 | be760ba8 | Michael Hanselmann | queue = job.queue |
1138 | be760ba8 | Michael Hanselmann | assert queue == self.pool.queue |
1139 | be760ba8 | Michael Hanselmann | |
1140 | daba67c7 | Michael Hanselmann | self.SetTaskName("Job%s" % job.id) |
1141 | daba67c7 | Michael Hanselmann | |
1142 | be760ba8 | Michael Hanselmann | proc = mcpu.Processor(queue.context, job.id) |
1143 | be760ba8 | Michael Hanselmann | |
1144 | be760ba8 | Michael Hanselmann | if not _JobProcessor(queue, proc.ExecOpCode, job)(): |
1145 | be760ba8 | Michael Hanselmann | # Schedule again
|
1146 | 26d3fd2f | Michael Hanselmann | raise workerpool.DeferTask(priority=job.CalcPriority())
|
1147 | e2715f69 | Michael Hanselmann | |
1148 | e2715f69 | Michael Hanselmann | |
1149 | e2715f69 | Michael Hanselmann | class _JobQueueWorkerPool(workerpool.WorkerPool): |
1150 | ea03467c | Iustin Pop | """Simple class implementing a job-processing workerpool.
|
1151 | ea03467c | Iustin Pop |
|
1152 | ea03467c | Iustin Pop | """
|
1153 | 5bdce580 | Michael Hanselmann | def __init__(self, queue): |
1154 | 89e2b4d2 | Michael Hanselmann | super(_JobQueueWorkerPool, self).__init__("JobQueue", |
1155 | 89e2b4d2 | Michael Hanselmann | JOBQUEUE_THREADS, |
1156 | e2715f69 | Michael Hanselmann | _JobQueueWorker) |
1157 | 5bdce580 | Michael Hanselmann | self.queue = queue
|
1158 | e2715f69 | Michael Hanselmann | |
1159 | e2715f69 | Michael Hanselmann | |
1160 | 6c881c52 | Iustin Pop | def _RequireOpenQueue(fn): |
1161 | 6c881c52 | Iustin Pop | """Decorator for "public" functions.
|
1162 | ea03467c | Iustin Pop |
|
1163 | 6c881c52 | Iustin Pop | This function should be used for all 'public' functions. That is,
|
1164 | 6c881c52 | Iustin Pop | functions usually called from other classes. Note that this should
|
1165 | 6c881c52 | Iustin Pop | be applied only to methods (not plain functions), since it expects
|
1166 | 6c881c52 | Iustin Pop | that the decorated function is called with a first argument that has
|
1167 | a71f9c7d | Guido Trotter | a '_queue_filelock' argument.
|
1168 | ea03467c | Iustin Pop |
|
1169 | 99bd4f0a | Guido Trotter | @warning: Use this decorator only after locking.ssynchronized
|
1170 | f1da30e6 | Michael Hanselmann |
|
1171 | 6c881c52 | Iustin Pop | Example::
|
1172 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1173 | 6c881c52 | Iustin Pop | @_RequireOpenQueue
|
1174 | 6c881c52 | Iustin Pop | def Example(self):
|
1175 | 6c881c52 | Iustin Pop | pass
|
1176 | db37da70 | Michael Hanselmann |
|
1177 | 6c881c52 | Iustin Pop | """
|
1178 | 6c881c52 | Iustin Pop | def wrapper(self, *args, **kwargs): |
1179 | 7260cfbe | Iustin Pop | # pylint: disable-msg=W0212
|
1180 | a71f9c7d | Guido Trotter | assert self._queue_filelock is not None, "Queue should be open" |
1181 | 6c881c52 | Iustin Pop | return fn(self, *args, **kwargs) |
1182 | 6c881c52 | Iustin Pop | return wrapper
|
1183 | db37da70 | Michael Hanselmann | |
1184 | db37da70 | Michael Hanselmann | |
1185 | 6c881c52 | Iustin Pop | class JobQueue(object): |
1186 | 6c881c52 | Iustin Pop | """Queue used to manage the jobs.
|
1187 | db37da70 | Michael Hanselmann |
|
1188 | 6c881c52 | Iustin Pop | @cvar _RE_JOB_FILE: regex matching the valid job file names
|
1189 | 6c881c52 | Iustin Pop |
|
1190 | 6c881c52 | Iustin Pop | """
|
1191 | 6c881c52 | Iustin Pop | _RE_JOB_FILE = re.compile(r"^job-(%s)$" % constants.JOB_ID_TEMPLATE)
|
1192 | db37da70 | Michael Hanselmann | |
1193 | 85f03e0d | Michael Hanselmann | def __init__(self, context): |
1194 | ea03467c | Iustin Pop | """Constructor for JobQueue.
|
1195 | ea03467c | Iustin Pop |
|
1196 | ea03467c | Iustin Pop | The constructor will initialize the job queue object and then
|
1197 | ea03467c | Iustin Pop | start loading the current jobs from disk, either for starting them
|
1198 | ea03467c | Iustin Pop | (if they were queue) or for aborting them (if they were already
|
1199 | ea03467c | Iustin Pop | running).
|
1200 | ea03467c | Iustin Pop |
|
1201 | ea03467c | Iustin Pop | @type context: GanetiContext
|
1202 | ea03467c | Iustin Pop | @param context: the context object for access to the configuration
|
1203 | ea03467c | Iustin Pop | data and other ganeti objects
|
1204 | ea03467c | Iustin Pop |
|
1205 | ea03467c | Iustin Pop | """
|
1206 | 5bdce580 | Michael Hanselmann | self.context = context
|
1207 | 5685c1a5 | Michael Hanselmann | self._memcache = weakref.WeakValueDictionary()
|
1208 | b705c7a6 | Manuel Franceschini | self._my_hostname = netutils.Hostname.GetSysName()
|
1209 | f1da30e6 | Michael Hanselmann | |
1210 | ebb80afa | Guido Trotter | # The Big JobQueue lock. If a code block or method acquires it in shared
|
1211 | ebb80afa | Guido Trotter | # mode safe it must guarantee concurrency with all the code acquiring it in
|
1212 | ebb80afa | Guido Trotter | # shared mode, including itself. In order not to acquire it at all
|
1213 | ebb80afa | Guido Trotter | # concurrency must be guaranteed with all code acquiring it in shared mode
|
1214 | ebb80afa | Guido Trotter | # and all code acquiring it exclusively.
|
1215 | 7f93570a | Iustin Pop | self._lock = locking.SharedLock("JobQueue") |
1216 | ebb80afa | Guido Trotter | |
1217 | ebb80afa | Guido Trotter | self.acquire = self._lock.acquire |
1218 | ebb80afa | Guido Trotter | self.release = self._lock.release |
1219 | 85f03e0d | Michael Hanselmann | |
1220 | a71f9c7d | Guido Trotter | # Initialize the queue, and acquire the filelock.
|
1221 | a71f9c7d | Guido Trotter | # This ensures no other process is working on the job queue.
|
1222 | a71f9c7d | Guido Trotter | self._queue_filelock = jstore.InitAndVerifyQueue(must_lock=True) |
1223 | f1da30e6 | Michael Hanselmann | |
1224 | 04ab05ce | Michael Hanselmann | # Read serial file
|
1225 | 04ab05ce | Michael Hanselmann | self._last_serial = jstore.ReadSerial()
|
1226 | 04ab05ce | Michael Hanselmann | assert self._last_serial is not None, ("Serial file was modified between" |
1227 | 04ab05ce | Michael Hanselmann | " check in jstore and here")
|
1228 | c4beba1c | Iustin Pop | |
1229 | 23752136 | Michael Hanselmann | # Get initial list of nodes
|
1230 | 99aabbed | Iustin Pop | self._nodes = dict((n.name, n.primary_ip) |
1231 | 59303563 | Iustin Pop | for n in self.context.cfg.GetAllNodesInfo().values() |
1232 | 59303563 | Iustin Pop | if n.master_candidate)
|
1233 | 8e00939c | Michael Hanselmann | |
1234 | 8e00939c | Michael Hanselmann | # Remove master node
|
1235 | d8e0dc17 | Guido Trotter | self._nodes.pop(self._my_hostname, None) |
1236 | 23752136 | Michael Hanselmann | |
1237 | 23752136 | Michael Hanselmann | # TODO: Check consistency across nodes
|
1238 | 23752136 | Michael Hanselmann | |
1239 | 20571a26 | Guido Trotter | self._queue_size = 0 |
1240 | 20571a26 | Guido Trotter | self._UpdateQueueSizeUnlocked()
|
1241 | ff699aa9 | Michael Hanselmann | self._drained = jstore.CheckDrainFlag()
|
1242 | 20571a26 | Guido Trotter | |
1243 | 85f03e0d | Michael Hanselmann | # Setup worker pool
|
1244 | 5bdce580 | Michael Hanselmann | self._wpool = _JobQueueWorkerPool(self) |
1245 | 85f03e0d | Michael Hanselmann | try:
|
1246 | de9d02c7 | Michael Hanselmann | self._InspectQueue()
|
1247 | de9d02c7 | Michael Hanselmann | except:
|
1248 | de9d02c7 | Michael Hanselmann | self._wpool.TerminateWorkers()
|
1249 | de9d02c7 | Michael Hanselmann | raise
|
1250 | 711b5124 | Michael Hanselmann | |
1251 | de9d02c7 | Michael Hanselmann | @locking.ssynchronized(_LOCK)
|
1252 | de9d02c7 | Michael Hanselmann | @_RequireOpenQueue
|
1253 | de9d02c7 | Michael Hanselmann | def _InspectQueue(self): |
1254 | de9d02c7 | Michael Hanselmann | """Loads the whole job queue and resumes unfinished jobs.
|
1255 | de9d02c7 | Michael Hanselmann |
|
1256 | de9d02c7 | Michael Hanselmann | This function needs the lock here because WorkerPool.AddTask() may start a
|
1257 | de9d02c7 | Michael Hanselmann | job while we're still doing our work.
|
1258 | 711b5124 | Michael Hanselmann |
|
1259 | de9d02c7 | Michael Hanselmann | """
|
1260 | de9d02c7 | Michael Hanselmann | logging.info("Inspecting job queue")
|
1261 | de9d02c7 | Michael Hanselmann | |
1262 | 7b5c4a69 | Michael Hanselmann | restartjobs = [] |
1263 | 7b5c4a69 | Michael Hanselmann | |
1264 | de9d02c7 | Michael Hanselmann | all_job_ids = self._GetJobIDsUnlocked()
|
1265 | de9d02c7 | Michael Hanselmann | jobs_count = len(all_job_ids)
|
1266 | de9d02c7 | Michael Hanselmann | lastinfo = time.time() |
1267 | de9d02c7 | Michael Hanselmann | for idx, job_id in enumerate(all_job_ids): |
1268 | de9d02c7 | Michael Hanselmann | # Give an update every 1000 jobs or 10 seconds
|
1269 | de9d02c7 | Michael Hanselmann | if (idx % 1000 == 0 or time.time() >= (lastinfo + 10.0) or |
1270 | de9d02c7 | Michael Hanselmann | idx == (jobs_count - 1)):
|
1271 | de9d02c7 | Michael Hanselmann | logging.info("Job queue inspection: %d/%d (%0.1f %%)",
|
1272 | de9d02c7 | Michael Hanselmann | idx, jobs_count - 1, 100.0 * (idx + 1) / jobs_count) |
1273 | 711b5124 | Michael Hanselmann | lastinfo = time.time() |
1274 | 94ed59a5 | Iustin Pop | |
1275 | de9d02c7 | Michael Hanselmann | job = self._LoadJobUnlocked(job_id)
|
1276 | 85f03e0d | Michael Hanselmann | |
1277 | de9d02c7 | Michael Hanselmann | # a failure in loading the job can cause 'None' to be returned
|
1278 | de9d02c7 | Michael Hanselmann | if job is None: |
1279 | de9d02c7 | Michael Hanselmann | continue
|
1280 | 85f03e0d | Michael Hanselmann | |
1281 | de9d02c7 | Michael Hanselmann | status = job.CalcStatus() |
1282 | 711b5124 | Michael Hanselmann | |
1283 | 320d1daf | Michael Hanselmann | if status == constants.JOB_STATUS_QUEUED:
|
1284 | 7b5c4a69 | Michael Hanselmann | restartjobs.append(job) |
1285 | de9d02c7 | Michael Hanselmann | |
1286 | de9d02c7 | Michael Hanselmann | elif status in (constants.JOB_STATUS_RUNNING, |
1287 | 5ef699a0 | Michael Hanselmann | constants.JOB_STATUS_WAITLOCK, |
1288 | de9d02c7 | Michael Hanselmann | constants.JOB_STATUS_CANCELING): |
1289 | de9d02c7 | Michael Hanselmann | logging.warning("Unfinished job %s found: %s", job.id, job)
|
1290 | 320d1daf | Michael Hanselmann | |
1291 | 320d1daf | Michael Hanselmann | if status == constants.JOB_STATUS_WAITLOCK:
|
1292 | 320d1daf | Michael Hanselmann | # Restart job
|
1293 | 320d1daf | Michael Hanselmann | job.MarkUnfinishedOps(constants.OP_STATUS_QUEUED, None)
|
1294 | 320d1daf | Michael Hanselmann | restartjobs.append(job) |
1295 | 320d1daf | Michael Hanselmann | else:
|
1296 | 320d1daf | Michael Hanselmann | job.MarkUnfinishedOps(constants.OP_STATUS_ERROR, |
1297 | 320d1daf | Michael Hanselmann | "Unclean master daemon shutdown")
|
1298 | 320d1daf | Michael Hanselmann | |
1299 | de9d02c7 | Michael Hanselmann | self.UpdateJobUnlocked(job)
|
1300 | de9d02c7 | Michael Hanselmann | |
1301 | 7b5c4a69 | Michael Hanselmann | if restartjobs:
|
1302 | 7b5c4a69 | Michael Hanselmann | logging.info("Restarting %s jobs", len(restartjobs)) |
1303 | 7b5c4a69 | Michael Hanselmann | self._EnqueueJobs(restartjobs)
|
1304 | 7b5c4a69 | Michael Hanselmann | |
1305 | de9d02c7 | Michael Hanselmann | logging.info("Job queue inspection finished")
|
1306 | 85f03e0d | Michael Hanselmann | |
1307 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1308 | d2e03a33 | Michael Hanselmann | @_RequireOpenQueue
|
1309 | 99aabbed | Iustin Pop | def AddNode(self, node): |
1310 | 99aabbed | Iustin Pop | """Register a new node with the queue.
|
1311 | 99aabbed | Iustin Pop |
|
1312 | 99aabbed | Iustin Pop | @type node: L{objects.Node}
|
1313 | 99aabbed | Iustin Pop | @param node: the node object to be added
|
1314 | 99aabbed | Iustin Pop |
|
1315 | 99aabbed | Iustin Pop | """
|
1316 | 99aabbed | Iustin Pop | node_name = node.name |
1317 | d2e03a33 | Michael Hanselmann | assert node_name != self._my_hostname |
1318 | 23752136 | Michael Hanselmann | |
1319 | 9f774ee8 | Michael Hanselmann | # Clean queue directory on added node
|
1320 | c8457ce7 | Iustin Pop | result = rpc.RpcRunner.call_jobqueue_purge(node_name) |
1321 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
1322 | c8457ce7 | Iustin Pop | if msg:
|
1323 | c8457ce7 | Iustin Pop | logging.warning("Cannot cleanup queue directory on node %s: %s",
|
1324 | c8457ce7 | Iustin Pop | node_name, msg) |
1325 | 23752136 | Michael Hanselmann | |
1326 | 59303563 | Iustin Pop | if not node.master_candidate: |
1327 | 59303563 | Iustin Pop | # remove if existing, ignoring errors
|
1328 | 59303563 | Iustin Pop | self._nodes.pop(node_name, None) |
1329 | 59303563 | Iustin Pop | # and skip the replication of the job ids
|
1330 | 59303563 | Iustin Pop | return
|
1331 | 59303563 | Iustin Pop | |
1332 | d2e03a33 | Michael Hanselmann | # Upload the whole queue excluding archived jobs
|
1333 | d2e03a33 | Michael Hanselmann | files = [self._GetJobPath(job_id) for job_id in self._GetJobIDsUnlocked()] |
1334 | 23752136 | Michael Hanselmann | |
1335 | d2e03a33 | Michael Hanselmann | # Upload current serial file
|
1336 | d2e03a33 | Michael Hanselmann | files.append(constants.JOB_QUEUE_SERIAL_FILE) |
1337 | d2e03a33 | Michael Hanselmann | |
1338 | d2e03a33 | Michael Hanselmann | for file_name in files: |
1339 | 9f774ee8 | Michael Hanselmann | # Read file content
|
1340 | 13998ef2 | Michael Hanselmann | content = utils.ReadFile(file_name) |
1341 | 9f774ee8 | Michael Hanselmann | |
1342 | a3811745 | Michael Hanselmann | result = rpc.RpcRunner.call_jobqueue_update([node_name], |
1343 | a3811745 | Michael Hanselmann | [node.primary_ip], |
1344 | a3811745 | Michael Hanselmann | file_name, content) |
1345 | 3cebe102 | Michael Hanselmann | msg = result[node_name].fail_msg |
1346 | c8457ce7 | Iustin Pop | if msg:
|
1347 | c8457ce7 | Iustin Pop | logging.error("Failed to upload file %s to node %s: %s",
|
1348 | c8457ce7 | Iustin Pop | file_name, node_name, msg) |
1349 | d2e03a33 | Michael Hanselmann | |
1350 | 99aabbed | Iustin Pop | self._nodes[node_name] = node.primary_ip
|
1351 | d2e03a33 | Michael Hanselmann | |
1352 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1353 | d2e03a33 | Michael Hanselmann | @_RequireOpenQueue
|
1354 | d2e03a33 | Michael Hanselmann | def RemoveNode(self, node_name): |
1355 | ea03467c | Iustin Pop | """Callback called when removing nodes from the cluster.
|
1356 | ea03467c | Iustin Pop |
|
1357 | ea03467c | Iustin Pop | @type node_name: str
|
1358 | ea03467c | Iustin Pop | @param node_name: the name of the node to remove
|
1359 | ea03467c | Iustin Pop |
|
1360 | ea03467c | Iustin Pop | """
|
1361 | d8e0dc17 | Guido Trotter | self._nodes.pop(node_name, None) |
1362 | 23752136 | Michael Hanselmann | |
1363 | 7e950d31 | Iustin Pop | @staticmethod
|
1364 | 7e950d31 | Iustin Pop | def _CheckRpcResult(result, nodes, failmsg): |
1365 | ea03467c | Iustin Pop | """Verifies the status of an RPC call.
|
1366 | ea03467c | Iustin Pop |
|
1367 | ea03467c | Iustin Pop | Since we aim to keep consistency should this node (the current
|
1368 | ea03467c | Iustin Pop | master) fail, we will log errors if our rpc fail, and especially
|
1369 | 5bbd3f7f | Michael Hanselmann | log the case when more than half of the nodes fails.
|
1370 | ea03467c | Iustin Pop |
|
1371 | ea03467c | Iustin Pop | @param result: the data as returned from the rpc call
|
1372 | ea03467c | Iustin Pop | @type nodes: list
|
1373 | ea03467c | Iustin Pop | @param nodes: the list of nodes we made the call to
|
1374 | ea03467c | Iustin Pop | @type failmsg: str
|
1375 | ea03467c | Iustin Pop | @param failmsg: the identifier to be used for logging
|
1376 | ea03467c | Iustin Pop |
|
1377 | ea03467c | Iustin Pop | """
|
1378 | e74798c1 | Michael Hanselmann | failed = [] |
1379 | e74798c1 | Michael Hanselmann | success = [] |
1380 | e74798c1 | Michael Hanselmann | |
1381 | e74798c1 | Michael Hanselmann | for node in nodes: |
1382 | 3cebe102 | Michael Hanselmann | msg = result[node].fail_msg |
1383 | c8457ce7 | Iustin Pop | if msg:
|
1384 | e74798c1 | Michael Hanselmann | failed.append(node) |
1385 | 45e0d704 | Iustin Pop | logging.error("RPC call %s (%s) failed on node %s: %s",
|
1386 | 45e0d704 | Iustin Pop | result[node].call, failmsg, node, msg) |
1387 | c8457ce7 | Iustin Pop | else:
|
1388 | c8457ce7 | Iustin Pop | success.append(node) |
1389 | e74798c1 | Michael Hanselmann | |
1390 | e74798c1 | Michael Hanselmann | # +1 for the master node
|
1391 | e74798c1 | Michael Hanselmann | if (len(success) + 1) < len(failed): |
1392 | e74798c1 | Michael Hanselmann | # TODO: Handle failing nodes
|
1393 | e74798c1 | Michael Hanselmann | logging.error("More than half of the nodes failed")
|
1394 | e74798c1 | Michael Hanselmann | |
1395 | 99aabbed | Iustin Pop | def _GetNodeIp(self): |
1396 | 99aabbed | Iustin Pop | """Helper for returning the node name/ip list.
|
1397 | 99aabbed | Iustin Pop |
|
1398 | ea03467c | Iustin Pop | @rtype: (list, list)
|
1399 | ea03467c | Iustin Pop | @return: a tuple of two lists, the first one with the node
|
1400 | ea03467c | Iustin Pop | names and the second one with the node addresses
|
1401 | ea03467c | Iustin Pop |
|
1402 | 99aabbed | Iustin Pop | """
|
1403 | e35344b4 | Michael Hanselmann | # TODO: Change to "tuple(map(list, zip(*self._nodes.items())))"?
|
1404 | 99aabbed | Iustin Pop | name_list = self._nodes.keys()
|
1405 | 99aabbed | Iustin Pop | addr_list = [self._nodes[name] for name in name_list] |
1406 | 99aabbed | Iustin Pop | return name_list, addr_list
|
1407 | 99aabbed | Iustin Pop | |
1408 | 4c36bdf5 | Guido Trotter | def _UpdateJobQueueFile(self, file_name, data, replicate): |
1409 | 8e00939c | Michael Hanselmann | """Writes a file locally and then replicates it to all nodes.
|
1410 | 8e00939c | Michael Hanselmann |
|
1411 | ea03467c | Iustin Pop | This function will replace the contents of a file on the local
|
1412 | ea03467c | Iustin Pop | node and then replicate it to all the other nodes we have.
|
1413 | ea03467c | Iustin Pop |
|
1414 | ea03467c | Iustin Pop | @type file_name: str
|
1415 | ea03467c | Iustin Pop | @param file_name: the path of the file to be replicated
|
1416 | ea03467c | Iustin Pop | @type data: str
|
1417 | ea03467c | Iustin Pop | @param data: the new contents of the file
|
1418 | 4c36bdf5 | Guido Trotter | @type replicate: boolean
|
1419 | 4c36bdf5 | Guido Trotter | @param replicate: whether to spread the changes to the remote nodes
|
1420 | ea03467c | Iustin Pop |
|
1421 | 8e00939c | Michael Hanselmann | """
|
1422 | 82b22e19 | René Nussbaumer | getents = runtime.GetEnts() |
1423 | 82b22e19 | René Nussbaumer | utils.WriteFile(file_name, data=data, uid=getents.masterd_uid, |
1424 | 82b22e19 | René Nussbaumer | gid=getents.masterd_gid) |
1425 | 8e00939c | Michael Hanselmann | |
1426 | 4c36bdf5 | Guido Trotter | if replicate:
|
1427 | 4c36bdf5 | Guido Trotter | names, addrs = self._GetNodeIp()
|
1428 | 4c36bdf5 | Guido Trotter | result = rpc.RpcRunner.call_jobqueue_update(names, addrs, file_name, data) |
1429 | 4c36bdf5 | Guido Trotter | self._CheckRpcResult(result, self._nodes, "Updating %s" % file_name) |
1430 | 23752136 | Michael Hanselmann | |
1431 | d7fd1f28 | Michael Hanselmann | def _RenameFilesUnlocked(self, rename): |
1432 | ea03467c | Iustin Pop | """Renames a file locally and then replicate the change.
|
1433 | ea03467c | Iustin Pop |
|
1434 | ea03467c | Iustin Pop | This function will rename a file in the local queue directory
|
1435 | ea03467c | Iustin Pop | and then replicate this rename to all the other nodes we have.
|
1436 | ea03467c | Iustin Pop |
|
1437 | d7fd1f28 | Michael Hanselmann | @type rename: list of (old, new)
|
1438 | d7fd1f28 | Michael Hanselmann | @param rename: List containing tuples mapping old to new names
|
1439 | ea03467c | Iustin Pop |
|
1440 | ea03467c | Iustin Pop | """
|
1441 | dd875d32 | Michael Hanselmann | # Rename them locally
|
1442 | d7fd1f28 | Michael Hanselmann | for old, new in rename: |
1443 | d7fd1f28 | Michael Hanselmann | utils.RenameFile(old, new, mkdir=True)
|
1444 | abc1f2ce | Michael Hanselmann | |
1445 | dd875d32 | Michael Hanselmann | # ... and on all nodes
|
1446 | dd875d32 | Michael Hanselmann | names, addrs = self._GetNodeIp()
|
1447 | dd875d32 | Michael Hanselmann | result = rpc.RpcRunner.call_jobqueue_rename(names, addrs, rename) |
1448 | dd875d32 | Michael Hanselmann | self._CheckRpcResult(result, self._nodes, "Renaming files (%r)" % rename) |
1449 | abc1f2ce | Michael Hanselmann | |
1450 | 7e950d31 | Iustin Pop | @staticmethod
|
1451 | 7e950d31 | Iustin Pop | def _FormatJobID(job_id): |
1452 | ea03467c | Iustin Pop | """Convert a job ID to string format.
|
1453 | ea03467c | Iustin Pop |
|
1454 | ea03467c | Iustin Pop | Currently this just does C{str(job_id)} after performing some
|
1455 | ea03467c | Iustin Pop | checks, but if we want to change the job id format this will
|
1456 | ea03467c | Iustin Pop | abstract this change.
|
1457 | ea03467c | Iustin Pop |
|
1458 | ea03467c | Iustin Pop | @type job_id: int or long
|
1459 | ea03467c | Iustin Pop | @param job_id: the numeric job id
|
1460 | ea03467c | Iustin Pop | @rtype: str
|
1461 | ea03467c | Iustin Pop | @return: the formatted job id
|
1462 | ea03467c | Iustin Pop |
|
1463 | ea03467c | Iustin Pop | """
|
1464 | 85f03e0d | Michael Hanselmann | if not isinstance(job_id, (int, long)): |
1465 | 85f03e0d | Michael Hanselmann | raise errors.ProgrammerError("Job ID '%s' not numeric" % job_id) |
1466 | 85f03e0d | Michael Hanselmann | if job_id < 0: |
1467 | 85f03e0d | Michael Hanselmann | raise errors.ProgrammerError("Job ID %s is negative" % job_id) |
1468 | 85f03e0d | Michael Hanselmann | |
1469 | 85f03e0d | Michael Hanselmann | return str(job_id) |
1470 | 85f03e0d | Michael Hanselmann | |
1471 | 58b22b6e | Michael Hanselmann | @classmethod
|
1472 | 58b22b6e | Michael Hanselmann | def _GetArchiveDirectory(cls, job_id): |
1473 | 58b22b6e | Michael Hanselmann | """Returns the archive directory for a job.
|
1474 | 58b22b6e | Michael Hanselmann |
|
1475 | 58b22b6e | Michael Hanselmann | @type job_id: str
|
1476 | 58b22b6e | Michael Hanselmann | @param job_id: Job identifier
|
1477 | 58b22b6e | Michael Hanselmann | @rtype: str
|
1478 | 58b22b6e | Michael Hanselmann | @return: Directory name
|
1479 | 58b22b6e | Michael Hanselmann |
|
1480 | 58b22b6e | Michael Hanselmann | """
|
1481 | 58b22b6e | Michael Hanselmann | return str(int(job_id) / JOBS_PER_ARCHIVE_DIRECTORY) |
1482 | 58b22b6e | Michael Hanselmann | |
1483 | 009e73d0 | Iustin Pop | def _NewSerialsUnlocked(self, count): |
1484 | f1da30e6 | Michael Hanselmann | """Generates a new job identifier.
|
1485 | f1da30e6 | Michael Hanselmann |
|
1486 | f1da30e6 | Michael Hanselmann | Job identifiers are unique during the lifetime of a cluster.
|
1487 | f1da30e6 | Michael Hanselmann |
|
1488 | 009e73d0 | Iustin Pop | @type count: integer
|
1489 | 009e73d0 | Iustin Pop | @param count: how many serials to return
|
1490 | ea03467c | Iustin Pop | @rtype: str
|
1491 | ea03467c | Iustin Pop | @return: a string representing the job identifier.
|
1492 | f1da30e6 | Michael Hanselmann |
|
1493 | f1da30e6 | Michael Hanselmann | """
|
1494 | 009e73d0 | Iustin Pop | assert count > 0 |
1495 | f1da30e6 | Michael Hanselmann | # New number
|
1496 | 009e73d0 | Iustin Pop | serial = self._last_serial + count
|
1497 | f1da30e6 | Michael Hanselmann | |
1498 | f1da30e6 | Michael Hanselmann | # Write to file
|
1499 | 4c36bdf5 | Guido Trotter | self._UpdateJobQueueFile(constants.JOB_QUEUE_SERIAL_FILE,
|
1500 | 4c36bdf5 | Guido Trotter | "%s\n" % serial, True) |
1501 | f1da30e6 | Michael Hanselmann | |
1502 | 009e73d0 | Iustin Pop | result = [self._FormatJobID(v)
|
1503 | 009e73d0 | Iustin Pop | for v in range(self._last_serial, serial + 1)] |
1504 | f1da30e6 | Michael Hanselmann | # Keep it only if we were able to write the file
|
1505 | f1da30e6 | Michael Hanselmann | self._last_serial = serial
|
1506 | f1da30e6 | Michael Hanselmann | |
1507 | 009e73d0 | Iustin Pop | return result
|
1508 | f1da30e6 | Michael Hanselmann | |
1509 | 85f03e0d | Michael Hanselmann | @staticmethod
|
1510 | 85f03e0d | Michael Hanselmann | def _GetJobPath(job_id): |
1511 | ea03467c | Iustin Pop | """Returns the job file for a given job id.
|
1512 | ea03467c | Iustin Pop |
|
1513 | ea03467c | Iustin Pop | @type job_id: str
|
1514 | ea03467c | Iustin Pop | @param job_id: the job identifier
|
1515 | ea03467c | Iustin Pop | @rtype: str
|
1516 | ea03467c | Iustin Pop | @return: the path to the job file
|
1517 | ea03467c | Iustin Pop |
|
1518 | ea03467c | Iustin Pop | """
|
1519 | c4feafe8 | Iustin Pop | return utils.PathJoin(constants.QUEUE_DIR, "job-%s" % job_id) |
1520 | f1da30e6 | Michael Hanselmann | |
1521 | 58b22b6e | Michael Hanselmann | @classmethod
|
1522 | 58b22b6e | Michael Hanselmann | def _GetArchivedJobPath(cls, job_id): |
1523 | ea03467c | Iustin Pop | """Returns the archived job file for a give job id.
|
1524 | ea03467c | Iustin Pop |
|
1525 | ea03467c | Iustin Pop | @type job_id: str
|
1526 | ea03467c | Iustin Pop | @param job_id: the job identifier
|
1527 | ea03467c | Iustin Pop | @rtype: str
|
1528 | ea03467c | Iustin Pop | @return: the path to the archived job file
|
1529 | ea03467c | Iustin Pop |
|
1530 | ea03467c | Iustin Pop | """
|
1531 | 0411c011 | Iustin Pop | return utils.PathJoin(constants.JOB_QUEUE_ARCHIVE_DIR,
|
1532 | 0411c011 | Iustin Pop | cls._GetArchiveDirectory(job_id), "job-%s" % job_id)
|
1533 | 0cb94105 | Michael Hanselmann | |
1534 | 85a1c57d | Guido Trotter | def _GetJobIDsUnlocked(self, sort=True): |
1535 | 911a495b | Iustin Pop | """Return all known job IDs.
|
1536 | 911a495b | Iustin Pop |
|
1537 | ac0930b9 | Iustin Pop | The method only looks at disk because it's a requirement that all
|
1538 | ac0930b9 | Iustin Pop | jobs are present on disk (so in the _memcache we don't have any
|
1539 | ac0930b9 | Iustin Pop | extra IDs).
|
1540 | ac0930b9 | Iustin Pop |
|
1541 | 85a1c57d | Guido Trotter | @type sort: boolean
|
1542 | 85a1c57d | Guido Trotter | @param sort: perform sorting on the returned job ids
|
1543 | ea03467c | Iustin Pop | @rtype: list
|
1544 | ea03467c | Iustin Pop | @return: the list of job IDs
|
1545 | ea03467c | Iustin Pop |
|
1546 | 911a495b | Iustin Pop | """
|
1547 | 85a1c57d | Guido Trotter | jlist = [] |
1548 | b5b8309d | Guido Trotter | for filename in utils.ListVisibleFiles(constants.QUEUE_DIR): |
1549 | 85a1c57d | Guido Trotter | m = self._RE_JOB_FILE.match(filename)
|
1550 | 85a1c57d | Guido Trotter | if m:
|
1551 | 85a1c57d | Guido Trotter | jlist.append(m.group(1))
|
1552 | 85a1c57d | Guido Trotter | if sort:
|
1553 | 85a1c57d | Guido Trotter | jlist = utils.NiceSort(jlist) |
1554 | f0d874fe | Iustin Pop | return jlist
|
1555 | 911a495b | Iustin Pop | |
1556 | 911a495b | Iustin Pop | def _LoadJobUnlocked(self, job_id): |
1557 | ea03467c | Iustin Pop | """Loads a job from the disk or memory.
|
1558 | ea03467c | Iustin Pop |
|
1559 | ea03467c | Iustin Pop | Given a job id, this will return the cached job object if
|
1560 | ea03467c | Iustin Pop | existing, or try to load the job from the disk. If loading from
|
1561 | ea03467c | Iustin Pop | disk, it will also add the job to the cache.
|
1562 | ea03467c | Iustin Pop |
|
1563 | ea03467c | Iustin Pop | @param job_id: the job id
|
1564 | ea03467c | Iustin Pop | @rtype: L{_QueuedJob} or None
|
1565 | ea03467c | Iustin Pop | @return: either None or the job object
|
1566 | ea03467c | Iustin Pop |
|
1567 | ea03467c | Iustin Pop | """
|
1568 | 5685c1a5 | Michael Hanselmann | job = self._memcache.get(job_id, None) |
1569 | 5685c1a5 | Michael Hanselmann | if job:
|
1570 | 205d71fd | Michael Hanselmann | logging.debug("Found job %s in memcache", job_id)
|
1571 | 5685c1a5 | Michael Hanselmann | return job
|
1572 | ac0930b9 | Iustin Pop | |
1573 | 3d6c5566 | Guido Trotter | try:
|
1574 | 3d6c5566 | Guido Trotter | job = self._LoadJobFromDisk(job_id)
|
1575 | aa9f8167 | Iustin Pop | if job is None: |
1576 | aa9f8167 | Iustin Pop | return job
|
1577 | 3d6c5566 | Guido Trotter | except errors.JobFileCorrupted:
|
1578 | 3d6c5566 | Guido Trotter | old_path = self._GetJobPath(job_id)
|
1579 | 3d6c5566 | Guido Trotter | new_path = self._GetArchivedJobPath(job_id)
|
1580 | 3d6c5566 | Guido Trotter | if old_path == new_path:
|
1581 | 3d6c5566 | Guido Trotter | # job already archived (future case)
|
1582 | 3d6c5566 | Guido Trotter | logging.exception("Can't parse job %s", job_id)
|
1583 | 3d6c5566 | Guido Trotter | else:
|
1584 | 3d6c5566 | Guido Trotter | # non-archived case
|
1585 | 3d6c5566 | Guido Trotter | logging.exception("Can't parse job %s, will archive.", job_id)
|
1586 | 3d6c5566 | Guido Trotter | self._RenameFilesUnlocked([(old_path, new_path)])
|
1587 | 3d6c5566 | Guido Trotter | return None |
1588 | 162c8636 | Guido Trotter | |
1589 | 162c8636 | Guido Trotter | self._memcache[job_id] = job
|
1590 | 162c8636 | Guido Trotter | logging.debug("Added job %s to the cache", job_id)
|
1591 | 162c8636 | Guido Trotter | return job
|
1592 | 162c8636 | Guido Trotter | |
1593 | 162c8636 | Guido Trotter | def _LoadJobFromDisk(self, job_id): |
1594 | 162c8636 | Guido Trotter | """Load the given job file from disk.
|
1595 | 162c8636 | Guido Trotter |
|
1596 | 162c8636 | Guido Trotter | Given a job file, read, load and restore it in a _QueuedJob format.
|
1597 | 162c8636 | Guido Trotter |
|
1598 | 162c8636 | Guido Trotter | @type job_id: string
|
1599 | 162c8636 | Guido Trotter | @param job_id: job identifier
|
1600 | 162c8636 | Guido Trotter | @rtype: L{_QueuedJob} or None
|
1601 | 162c8636 | Guido Trotter | @return: either None or the job object
|
1602 | 162c8636 | Guido Trotter |
|
1603 | 162c8636 | Guido Trotter | """
|
1604 | 911a495b | Iustin Pop | filepath = self._GetJobPath(job_id)
|
1605 | f1da30e6 | Michael Hanselmann | logging.debug("Loading job from %s", filepath)
|
1606 | f1da30e6 | Michael Hanselmann | try:
|
1607 | 13998ef2 | Michael Hanselmann | raw_data = utils.ReadFile(filepath) |
1608 | 162c8636 | Guido Trotter | except EnvironmentError, err: |
1609 | f1da30e6 | Michael Hanselmann | if err.errno in (errno.ENOENT, ): |
1610 | f1da30e6 | Michael Hanselmann | return None |
1611 | f1da30e6 | Michael Hanselmann | raise
|
1612 | 13998ef2 | Michael Hanselmann | |
1613 | 94ed59a5 | Iustin Pop | try:
|
1614 | 162c8636 | Guido Trotter | data = serializer.LoadJson(raw_data) |
1615 | 94ed59a5 | Iustin Pop | job = _QueuedJob.Restore(self, data)
|
1616 | 7260cfbe | Iustin Pop | except Exception, err: # pylint: disable-msg=W0703 |
1617 | 3d6c5566 | Guido Trotter | raise errors.JobFileCorrupted(err)
|
1618 | 94ed59a5 | Iustin Pop | |
1619 | ac0930b9 | Iustin Pop | return job
|
1620 | f1da30e6 | Michael Hanselmann | |
1621 | 0f9c08dc | Guido Trotter | def SafeLoadJobFromDisk(self, job_id): |
1622 | 0f9c08dc | Guido Trotter | """Load the given job file from disk.
|
1623 | 0f9c08dc | Guido Trotter |
|
1624 | 0f9c08dc | Guido Trotter | Given a job file, read, load and restore it in a _QueuedJob format.
|
1625 | 0f9c08dc | Guido Trotter | In case of error reading the job, it gets returned as None, and the
|
1626 | 0f9c08dc | Guido Trotter | exception is logged.
|
1627 | 0f9c08dc | Guido Trotter |
|
1628 | 0f9c08dc | Guido Trotter | @type job_id: string
|
1629 | 0f9c08dc | Guido Trotter | @param job_id: job identifier
|
1630 | 0f9c08dc | Guido Trotter | @rtype: L{_QueuedJob} or None
|
1631 | 0f9c08dc | Guido Trotter | @return: either None or the job object
|
1632 | 0f9c08dc | Guido Trotter |
|
1633 | 0f9c08dc | Guido Trotter | """
|
1634 | 0f9c08dc | Guido Trotter | try:
|
1635 | 0f9c08dc | Guido Trotter | return self._LoadJobFromDisk(job_id) |
1636 | 0f9c08dc | Guido Trotter | except (errors.JobFileCorrupted, EnvironmentError): |
1637 | 0f9c08dc | Guido Trotter | logging.exception("Can't load/parse job %s", job_id)
|
1638 | 0f9c08dc | Guido Trotter | return None |
1639 | 0f9c08dc | Guido Trotter | |
1640 | 20571a26 | Guido Trotter | def _UpdateQueueSizeUnlocked(self): |
1641 | 20571a26 | Guido Trotter | """Update the queue size.
|
1642 | 20571a26 | Guido Trotter |
|
1643 | 20571a26 | Guido Trotter | """
|
1644 | 20571a26 | Guido Trotter | self._queue_size = len(self._GetJobIDsUnlocked(sort=False)) |
1645 | 20571a26 | Guido Trotter | |
1646 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1647 | 20571a26 | Guido Trotter | @_RequireOpenQueue
|
1648 | 20571a26 | Guido Trotter | def SetDrainFlag(self, drain_flag): |
1649 | 3ccafd0e | Iustin Pop | """Sets the drain flag for the queue.
|
1650 | 3ccafd0e | Iustin Pop |
|
1651 | ea03467c | Iustin Pop | @type drain_flag: boolean
|
1652 | 5bbd3f7f | Michael Hanselmann | @param drain_flag: Whether to set or unset the drain flag
|
1653 | ea03467c | Iustin Pop |
|
1654 | 3ccafd0e | Iustin Pop | """
|
1655 | ff699aa9 | Michael Hanselmann | jstore.SetDrainFlag(drain_flag) |
1656 | 20571a26 | Guido Trotter | |
1657 | 20571a26 | Guido Trotter | self._drained = drain_flag
|
1658 | 20571a26 | Guido Trotter | |
1659 | 3ccafd0e | Iustin Pop | return True |
1660 | 3ccafd0e | Iustin Pop | |
1661 | db37da70 | Michael Hanselmann | @_RequireOpenQueue
|
1662 | 009e73d0 | Iustin Pop | def _SubmitJobUnlocked(self, job_id, ops): |
1663 | 85f03e0d | Michael Hanselmann | """Create and store a new job.
|
1664 | f1da30e6 | Michael Hanselmann |
|
1665 | 85f03e0d | Michael Hanselmann | This enters the job into our job queue and also puts it on the new
|
1666 | 85f03e0d | Michael Hanselmann | queue, in order for it to be picked up by the queue processors.
|
1667 | c3f0a12f | Iustin Pop |
|
1668 | 009e73d0 | Iustin Pop | @type job_id: job ID
|
1669 | 69b99987 | Michael Hanselmann | @param job_id: the job ID for the new job
|
1670 | c3f0a12f | Iustin Pop | @type ops: list
|
1671 | 205d71fd | Michael Hanselmann | @param ops: The list of OpCodes that will become the new job.
|
1672 | 7beb1e53 | Guido Trotter | @rtype: L{_QueuedJob}
|
1673 | 7beb1e53 | Guido Trotter | @return: the job object to be queued
|
1674 | 7beb1e53 | Guido Trotter | @raise errors.JobQueueDrainError: if the job queue is marked for draining
|
1675 | 7beb1e53 | Guido Trotter | @raise errors.JobQueueFull: if the job queue has too many jobs in it
|
1676 | e71c8147 | Michael Hanselmann | @raise errors.GenericError: If an opcode is not valid
|
1677 | c3f0a12f | Iustin Pop |
|
1678 | c3f0a12f | Iustin Pop | """
|
1679 | 20571a26 | Guido Trotter | # Ok when sharing the big job queue lock, as the drain file is created when
|
1680 | 20571a26 | Guido Trotter | # the lock is exclusive.
|
1681 | 20571a26 | Guido Trotter | if self._drained: |
1682 | 2971c913 | Iustin Pop | raise errors.JobQueueDrainError("Job queue is drained, refusing job") |
1683 | f87b405e | Michael Hanselmann | |
1684 | 20571a26 | Guido Trotter | if self._queue_size >= constants.JOB_QUEUE_SIZE_HARD_LIMIT: |
1685 | f87b405e | Michael Hanselmann | raise errors.JobQueueFull()
|
1686 | f87b405e | Michael Hanselmann | |
1687 | f1da30e6 | Michael Hanselmann | job = _QueuedJob(self, job_id, ops)
|
1688 | f1da30e6 | Michael Hanselmann | |
1689 | e71c8147 | Michael Hanselmann | # Check priority
|
1690 | e71c8147 | Michael Hanselmann | for idx, op in enumerate(job.ops): |
1691 | e71c8147 | Michael Hanselmann | if op.priority not in constants.OP_PRIO_SUBMIT_VALID: |
1692 | e71c8147 | Michael Hanselmann | allowed = utils.CommaJoin(constants.OP_PRIO_SUBMIT_VALID) |
1693 | e71c8147 | Michael Hanselmann | raise errors.GenericError("Opcode %s has invalid priority %s, allowed" |
1694 | e71c8147 | Michael Hanselmann | " are %s" % (idx, op.priority, allowed))
|
1695 | e71c8147 | Michael Hanselmann | |
1696 | f1da30e6 | Michael Hanselmann | # Write to disk
|
1697 | 85f03e0d | Michael Hanselmann | self.UpdateJobUnlocked(job)
|
1698 | f1da30e6 | Michael Hanselmann | |
1699 | 20571a26 | Guido Trotter | self._queue_size += 1 |
1700 | 20571a26 | Guido Trotter | |
1701 | 5685c1a5 | Michael Hanselmann | logging.debug("Adding new job %s to the cache", job_id)
|
1702 | ac0930b9 | Iustin Pop | self._memcache[job_id] = job
|
1703 | ac0930b9 | Iustin Pop | |
1704 | 7beb1e53 | Guido Trotter | return job
|
1705 | f1da30e6 | Michael Hanselmann | |
1706 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1707 | 2971c913 | Iustin Pop | @_RequireOpenQueue
|
1708 | 2971c913 | Iustin Pop | def SubmitJob(self, ops): |
1709 | 2971c913 | Iustin Pop | """Create and store a new job.
|
1710 | 2971c913 | Iustin Pop |
|
1711 | 2971c913 | Iustin Pop | @see: L{_SubmitJobUnlocked}
|
1712 | 2971c913 | Iustin Pop |
|
1713 | 2971c913 | Iustin Pop | """
|
1714 | 009e73d0 | Iustin Pop | job_id = self._NewSerialsUnlocked(1)[0] |
1715 | 7b5c4a69 | Michael Hanselmann | self._EnqueueJobs([self._SubmitJobUnlocked(job_id, ops)]) |
1716 | 7beb1e53 | Guido Trotter | return job_id
|
1717 | 2971c913 | Iustin Pop | |
1718 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1719 | 2971c913 | Iustin Pop | @_RequireOpenQueue
|
1720 | 2971c913 | Iustin Pop | def SubmitManyJobs(self, jobs): |
1721 | 2971c913 | Iustin Pop | """Create and store multiple jobs.
|
1722 | 2971c913 | Iustin Pop |
|
1723 | 2971c913 | Iustin Pop | @see: L{_SubmitJobUnlocked}
|
1724 | 2971c913 | Iustin Pop |
|
1725 | 2971c913 | Iustin Pop | """
|
1726 | 2971c913 | Iustin Pop | results = [] |
1727 | 7b5c4a69 | Michael Hanselmann | added_jobs = [] |
1728 | 009e73d0 | Iustin Pop | all_job_ids = self._NewSerialsUnlocked(len(jobs)) |
1729 | 009e73d0 | Iustin Pop | for job_id, ops in zip(all_job_ids, jobs): |
1730 | 2971c913 | Iustin Pop | try:
|
1731 | 7b5c4a69 | Michael Hanselmann | added_jobs.append(self._SubmitJobUnlocked(job_id, ops))
|
1732 | 2971c913 | Iustin Pop | status = True
|
1733 | 7beb1e53 | Guido Trotter | data = job_id |
1734 | 2971c913 | Iustin Pop | except errors.GenericError, err:
|
1735 | 98ed5092 | Michael Hanselmann | data = ("%s; opcodes %s" %
|
1736 | 98ed5092 | Michael Hanselmann | (err, utils.CommaJoin(op.Summary() for op in ops))) |
1737 | 2971c913 | Iustin Pop | status = False
|
1738 | 2971c913 | Iustin Pop | results.append((status, data)) |
1739 | 7b5c4a69 | Michael Hanselmann | |
1740 | 7b5c4a69 | Michael Hanselmann | self._EnqueueJobs(added_jobs)
|
1741 | 2971c913 | Iustin Pop | |
1742 | 2971c913 | Iustin Pop | return results
|
1743 | 2971c913 | Iustin Pop | |
1744 | 7b5c4a69 | Michael Hanselmann | def _EnqueueJobs(self, jobs): |
1745 | 7b5c4a69 | Michael Hanselmann | """Helper function to add jobs to worker pool's queue.
|
1746 | 7b5c4a69 | Michael Hanselmann |
|
1747 | 7b5c4a69 | Michael Hanselmann | @type jobs: list
|
1748 | 7b5c4a69 | Michael Hanselmann | @param jobs: List of all jobs
|
1749 | 7b5c4a69 | Michael Hanselmann |
|
1750 | 7b5c4a69 | Michael Hanselmann | """
|
1751 | 7b5c4a69 | Michael Hanselmann | self._wpool.AddManyTasks([(job, ) for job in jobs], |
1752 | 7b5c4a69 | Michael Hanselmann | priority=[job.CalcPriority() for job in jobs]) |
1753 | 7b5c4a69 | Michael Hanselmann | |
1754 | db37da70 | Michael Hanselmann | @_RequireOpenQueue
|
1755 | 4c36bdf5 | Guido Trotter | def UpdateJobUnlocked(self, job, replicate=True): |
1756 | ea03467c | Iustin Pop | """Update a job's on disk storage.
|
1757 | ea03467c | Iustin Pop |
|
1758 | ea03467c | Iustin Pop | After a job has been modified, this function needs to be called in
|
1759 | ea03467c | Iustin Pop | order to write the changes to disk and replicate them to the other
|
1760 | ea03467c | Iustin Pop | nodes.
|
1761 | ea03467c | Iustin Pop |
|
1762 | ea03467c | Iustin Pop | @type job: L{_QueuedJob}
|
1763 | ea03467c | Iustin Pop | @param job: the changed job
|
1764 | 4c36bdf5 | Guido Trotter | @type replicate: boolean
|
1765 | 4c36bdf5 | Guido Trotter | @param replicate: whether to replicate the change to remote nodes
|
1766 | ea03467c | Iustin Pop |
|
1767 | ea03467c | Iustin Pop | """
|
1768 | f1da30e6 | Michael Hanselmann | filename = self._GetJobPath(job.id)
|
1769 | 23752136 | Michael Hanselmann | data = serializer.DumpJson(job.Serialize(), indent=False)
|
1770 | f1da30e6 | Michael Hanselmann | logging.debug("Writing job %s to %s", job.id, filename)
|
1771 | 4c36bdf5 | Guido Trotter | self._UpdateJobQueueFile(filename, data, replicate)
|
1772 | ac0930b9 | Iustin Pop | |
1773 | 5c735209 | Iustin Pop | def WaitForJobChanges(self, job_id, fields, prev_job_info, prev_log_serial, |
1774 | 5c735209 | Iustin Pop | timeout): |
1775 | 6c5a7090 | Michael Hanselmann | """Waits for changes in a job.
|
1776 | 6c5a7090 | Michael Hanselmann |
|
1777 | 6c5a7090 | Michael Hanselmann | @type job_id: string
|
1778 | 6c5a7090 | Michael Hanselmann | @param job_id: Job identifier
|
1779 | 6c5a7090 | Michael Hanselmann | @type fields: list of strings
|
1780 | 6c5a7090 | Michael Hanselmann | @param fields: Which fields to check for changes
|
1781 | 6c5a7090 | Michael Hanselmann | @type prev_job_info: list or None
|
1782 | 6c5a7090 | Michael Hanselmann | @param prev_job_info: Last job information returned
|
1783 | 6c5a7090 | Michael Hanselmann | @type prev_log_serial: int
|
1784 | 6c5a7090 | Michael Hanselmann | @param prev_log_serial: Last job message serial number
|
1785 | 5c735209 | Iustin Pop | @type timeout: float
|
1786 | 989a8bee | Michael Hanselmann | @param timeout: maximum time to wait in seconds
|
1787 | ea03467c | Iustin Pop | @rtype: tuple (job info, log entries)
|
1788 | ea03467c | Iustin Pop | @return: a tuple of the job information as required via
|
1789 | ea03467c | Iustin Pop | the fields parameter, and the log entries as a list
|
1790 | ea03467c | Iustin Pop |
|
1791 | ea03467c | Iustin Pop | if the job has not changed and the timeout has expired,
|
1792 | ea03467c | Iustin Pop | we instead return a special value,
|
1793 | ea03467c | Iustin Pop | L{constants.JOB_NOTCHANGED}, which should be interpreted
|
1794 | ea03467c | Iustin Pop | as such by the clients
|
1795 | 6c5a7090 | Michael Hanselmann |
|
1796 | 6c5a7090 | Michael Hanselmann | """
|
1797 | 989a8bee | Michael Hanselmann | load_fn = compat.partial(self.SafeLoadJobFromDisk, job_id)
|
1798 | 989a8bee | Michael Hanselmann | |
1799 | 989a8bee | Michael Hanselmann | helper = _WaitForJobChangesHelper() |
1800 | 989a8bee | Michael Hanselmann | |
1801 | 989a8bee | Michael Hanselmann | return helper(self._GetJobPath(job_id), load_fn, |
1802 | 989a8bee | Michael Hanselmann | fields, prev_job_info, prev_log_serial, timeout) |
1803 | dfe57c22 | Michael Hanselmann | |
1804 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1805 | db37da70 | Michael Hanselmann | @_RequireOpenQueue
|
1806 | 188c5e0a | Michael Hanselmann | def CancelJob(self, job_id): |
1807 | 188c5e0a | Michael Hanselmann | """Cancels a job.
|
1808 | 188c5e0a | Michael Hanselmann |
|
1809 | ea03467c | Iustin Pop | This will only succeed if the job has not started yet.
|
1810 | ea03467c | Iustin Pop |
|
1811 | 188c5e0a | Michael Hanselmann | @type job_id: string
|
1812 | ea03467c | Iustin Pop | @param job_id: job ID of job to be cancelled.
|
1813 | 188c5e0a | Michael Hanselmann |
|
1814 | 188c5e0a | Michael Hanselmann | """
|
1815 | fbf0262f | Michael Hanselmann | logging.info("Cancelling job %s", job_id)
|
1816 | 188c5e0a | Michael Hanselmann | |
1817 | 85f03e0d | Michael Hanselmann | job = self._LoadJobUnlocked(job_id)
|
1818 | 188c5e0a | Michael Hanselmann | if not job: |
1819 | 188c5e0a | Michael Hanselmann | logging.debug("Job %s not found", job_id)
|
1820 | fbf0262f | Michael Hanselmann | return (False, "Job %s not found" % job_id) |
1821 | fbf0262f | Michael Hanselmann | |
1822 | 099b2870 | Michael Hanselmann | (success, msg) = job.Cancel() |
1823 | 188c5e0a | Michael Hanselmann | |
1824 | 099b2870 | Michael Hanselmann | if success:
|
1825 | 099b2870 | Michael Hanselmann | self.UpdateJobUnlocked(job)
|
1826 | fbf0262f | Michael Hanselmann | |
1827 | 099b2870 | Michael Hanselmann | return (success, msg)
|
1828 | fbf0262f | Michael Hanselmann | |
1829 | fbf0262f | Michael Hanselmann | @_RequireOpenQueue
|
1830 | d7fd1f28 | Michael Hanselmann | def _ArchiveJobsUnlocked(self, jobs): |
1831 | d7fd1f28 | Michael Hanselmann | """Archives jobs.
|
1832 | c609f802 | Michael Hanselmann |
|
1833 | d7fd1f28 | Michael Hanselmann | @type jobs: list of L{_QueuedJob}
|
1834 | 25e7b43f | Iustin Pop | @param jobs: Job objects
|
1835 | d7fd1f28 | Michael Hanselmann | @rtype: int
|
1836 | d7fd1f28 | Michael Hanselmann | @return: Number of archived jobs
|
1837 | c609f802 | Michael Hanselmann |
|
1838 | c609f802 | Michael Hanselmann | """
|
1839 | d7fd1f28 | Michael Hanselmann | archive_jobs = [] |
1840 | d7fd1f28 | Michael Hanselmann | rename_files = [] |
1841 | d7fd1f28 | Michael Hanselmann | for job in jobs: |
1842 | 989a8bee | Michael Hanselmann | if job.CalcStatus() not in constants.JOBS_FINALIZED: |
1843 | d7fd1f28 | Michael Hanselmann | logging.debug("Job %s is not yet done", job.id)
|
1844 | d7fd1f28 | Michael Hanselmann | continue
|
1845 | c609f802 | Michael Hanselmann | |
1846 | d7fd1f28 | Michael Hanselmann | archive_jobs.append(job) |
1847 | c609f802 | Michael Hanselmann | |
1848 | d7fd1f28 | Michael Hanselmann | old = self._GetJobPath(job.id)
|
1849 | d7fd1f28 | Michael Hanselmann | new = self._GetArchivedJobPath(job.id)
|
1850 | d7fd1f28 | Michael Hanselmann | rename_files.append((old, new)) |
1851 | c609f802 | Michael Hanselmann | |
1852 | d7fd1f28 | Michael Hanselmann | # TODO: What if 1..n files fail to rename?
|
1853 | d7fd1f28 | Michael Hanselmann | self._RenameFilesUnlocked(rename_files)
|
1854 | f1da30e6 | Michael Hanselmann | |
1855 | d7fd1f28 | Michael Hanselmann | logging.debug("Successfully archived job(s) %s",
|
1856 | 1f864b60 | Iustin Pop | utils.CommaJoin(job.id for job in archive_jobs)) |
1857 | d7fd1f28 | Michael Hanselmann | |
1858 | 20571a26 | Guido Trotter | # Since we haven't quite checked, above, if we succeeded or failed renaming
|
1859 | 20571a26 | Guido Trotter | # the files, we update the cached queue size from the filesystem. When we
|
1860 | 20571a26 | Guido Trotter | # get around to fix the TODO: above, we can use the number of actually
|
1861 | 20571a26 | Guido Trotter | # archived jobs to fix this.
|
1862 | 20571a26 | Guido Trotter | self._UpdateQueueSizeUnlocked()
|
1863 | d7fd1f28 | Michael Hanselmann | return len(archive_jobs) |
1864 | 78d12585 | Michael Hanselmann | |
1865 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1866 | 07cd723a | Iustin Pop | @_RequireOpenQueue
|
1867 | 07cd723a | Iustin Pop | def ArchiveJob(self, job_id): |
1868 | 07cd723a | Iustin Pop | """Archives a job.
|
1869 | 07cd723a | Iustin Pop |
|
1870 | 25e7b43f | Iustin Pop | This is just a wrapper over L{_ArchiveJobsUnlocked}.
|
1871 | ea03467c | Iustin Pop |
|
1872 | 07cd723a | Iustin Pop | @type job_id: string
|
1873 | 07cd723a | Iustin Pop | @param job_id: Job ID of job to be archived.
|
1874 | 78d12585 | Michael Hanselmann | @rtype: bool
|
1875 | 78d12585 | Michael Hanselmann | @return: Whether job was archived
|
1876 | 07cd723a | Iustin Pop |
|
1877 | 07cd723a | Iustin Pop | """
|
1878 | 78d12585 | Michael Hanselmann | logging.info("Archiving job %s", job_id)
|
1879 | 78d12585 | Michael Hanselmann | |
1880 | 78d12585 | Michael Hanselmann | job = self._LoadJobUnlocked(job_id)
|
1881 | 78d12585 | Michael Hanselmann | if not job: |
1882 | 78d12585 | Michael Hanselmann | logging.debug("Job %s not found", job_id)
|
1883 | 78d12585 | Michael Hanselmann | return False |
1884 | 78d12585 | Michael Hanselmann | |
1885 | 5278185a | Iustin Pop | return self._ArchiveJobsUnlocked([job]) == 1 |
1886 | 07cd723a | Iustin Pop | |
1887 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1888 | 07cd723a | Iustin Pop | @_RequireOpenQueue
|
1889 | f8ad5591 | Michael Hanselmann | def AutoArchiveJobs(self, age, timeout): |
1890 | 07cd723a | Iustin Pop | """Archives all jobs based on age.
|
1891 | 07cd723a | Iustin Pop |
|
1892 | 07cd723a | Iustin Pop | The method will archive all jobs which are older than the age
|
1893 | 07cd723a | Iustin Pop | parameter. For jobs that don't have an end timestamp, the start
|
1894 | 07cd723a | Iustin Pop | timestamp will be considered. The special '-1' age will cause
|
1895 | 07cd723a | Iustin Pop | archival of all jobs (that are not running or queued).
|
1896 | 07cd723a | Iustin Pop |
|
1897 | 07cd723a | Iustin Pop | @type age: int
|
1898 | 07cd723a | Iustin Pop | @param age: the minimum age in seconds
|
1899 | 07cd723a | Iustin Pop |
|
1900 | 07cd723a | Iustin Pop | """
|
1901 | 07cd723a | Iustin Pop | logging.info("Archiving jobs with age more than %s seconds", age)
|
1902 | 07cd723a | Iustin Pop | |
1903 | 07cd723a | Iustin Pop | now = time.time() |
1904 | f8ad5591 | Michael Hanselmann | end_time = now + timeout |
1905 | f8ad5591 | Michael Hanselmann | archived_count = 0
|
1906 | f8ad5591 | Michael Hanselmann | last_touched = 0
|
1907 | f8ad5591 | Michael Hanselmann | |
1908 | 69b03fd7 | Guido Trotter | all_job_ids = self._GetJobIDsUnlocked()
|
1909 | d7fd1f28 | Michael Hanselmann | pending = [] |
1910 | f8ad5591 | Michael Hanselmann | for idx, job_id in enumerate(all_job_ids): |
1911 | d2c8afb1 | Michael Hanselmann | last_touched = idx + 1
|
1912 | f8ad5591 | Michael Hanselmann | |
1913 | d7fd1f28 | Michael Hanselmann | # Not optimal because jobs could be pending
|
1914 | d7fd1f28 | Michael Hanselmann | # TODO: Measure average duration for job archival and take number of
|
1915 | d7fd1f28 | Michael Hanselmann | # pending jobs into account.
|
1916 | f8ad5591 | Michael Hanselmann | if time.time() > end_time:
|
1917 | f8ad5591 | Michael Hanselmann | break
|
1918 | f8ad5591 | Michael Hanselmann | |
1919 | 78d12585 | Michael Hanselmann | # Returns None if the job failed to load
|
1920 | 78d12585 | Michael Hanselmann | job = self._LoadJobUnlocked(job_id)
|
1921 | f8ad5591 | Michael Hanselmann | if job:
|
1922 | f8ad5591 | Michael Hanselmann | if job.end_timestamp is None: |
1923 | f8ad5591 | Michael Hanselmann | if job.start_timestamp is None: |
1924 | f8ad5591 | Michael Hanselmann | job_age = job.received_timestamp |
1925 | f8ad5591 | Michael Hanselmann | else:
|
1926 | f8ad5591 | Michael Hanselmann | job_age = job.start_timestamp |
1927 | 07cd723a | Iustin Pop | else:
|
1928 | f8ad5591 | Michael Hanselmann | job_age = job.end_timestamp |
1929 | f8ad5591 | Michael Hanselmann | |
1930 | f8ad5591 | Michael Hanselmann | if age == -1 or now - job_age[0] > age: |
1931 | d7fd1f28 | Michael Hanselmann | pending.append(job) |
1932 | d7fd1f28 | Michael Hanselmann | |
1933 | d7fd1f28 | Michael Hanselmann | # Archive 10 jobs at a time
|
1934 | d7fd1f28 | Michael Hanselmann | if len(pending) >= 10: |
1935 | d7fd1f28 | Michael Hanselmann | archived_count += self._ArchiveJobsUnlocked(pending)
|
1936 | d7fd1f28 | Michael Hanselmann | pending = [] |
1937 | f8ad5591 | Michael Hanselmann | |
1938 | d7fd1f28 | Michael Hanselmann | if pending:
|
1939 | d7fd1f28 | Michael Hanselmann | archived_count += self._ArchiveJobsUnlocked(pending)
|
1940 | 07cd723a | Iustin Pop | |
1941 | d2c8afb1 | Michael Hanselmann | return (archived_count, len(all_job_ids) - last_touched) |
1942 | 07cd723a | Iustin Pop | |
1943 | e2715f69 | Michael Hanselmann | def QueryJobs(self, job_ids, fields): |
1944 | e2715f69 | Michael Hanselmann | """Returns a list of jobs in queue.
|
1945 | e2715f69 | Michael Hanselmann |
|
1946 | ea03467c | Iustin Pop | @type job_ids: list
|
1947 | ea03467c | Iustin Pop | @param job_ids: sequence of job identifiers or None for all
|
1948 | ea03467c | Iustin Pop | @type fields: list
|
1949 | ea03467c | Iustin Pop | @param fields: names of fields to return
|
1950 | ea03467c | Iustin Pop | @rtype: list
|
1951 | ea03467c | Iustin Pop | @return: list one element per job, each element being list with
|
1952 | ea03467c | Iustin Pop | the requested fields
|
1953 | e2715f69 | Michael Hanselmann |
|
1954 | e2715f69 | Michael Hanselmann | """
|
1955 | 85f03e0d | Michael Hanselmann | jobs = [] |
1956 | 9f7b4967 | Guido Trotter | list_all = False
|
1957 | 9f7b4967 | Guido Trotter | if not job_ids: |
1958 | 9f7b4967 | Guido Trotter | # Since files are added to/removed from the queue atomically, there's no
|
1959 | 9f7b4967 | Guido Trotter | # risk of getting the job ids in an inconsistent state.
|
1960 | 9f7b4967 | Guido Trotter | job_ids = self._GetJobIDsUnlocked()
|
1961 | 9f7b4967 | Guido Trotter | list_all = True
|
1962 | e2715f69 | Michael Hanselmann | |
1963 | 9f7b4967 | Guido Trotter | for job_id in job_ids: |
1964 | 9f7b4967 | Guido Trotter | job = self.SafeLoadJobFromDisk(job_id)
|
1965 | 9f7b4967 | Guido Trotter | if job is not None: |
1966 | 6a290889 | Guido Trotter | jobs.append(job.GetInfo(fields)) |
1967 | 9f7b4967 | Guido Trotter | elif not list_all: |
1968 | 9f7b4967 | Guido Trotter | jobs.append(None)
|
1969 | e2715f69 | Michael Hanselmann | |
1970 | 85f03e0d | Michael Hanselmann | return jobs
|
1971 | e2715f69 | Michael Hanselmann | |
1972 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1973 | db37da70 | Michael Hanselmann | @_RequireOpenQueue
|
1974 | e2715f69 | Michael Hanselmann | def Shutdown(self): |
1975 | e2715f69 | Michael Hanselmann | """Stops the job queue.
|
1976 | e2715f69 | Michael Hanselmann |
|
1977 | ea03467c | Iustin Pop | This shutdowns all the worker threads an closes the queue.
|
1978 | ea03467c | Iustin Pop |
|
1979 | e2715f69 | Michael Hanselmann | """
|
1980 | e2715f69 | Michael Hanselmann | self._wpool.TerminateWorkers()
|
1981 | 85f03e0d | Michael Hanselmann | |
1982 | a71f9c7d | Guido Trotter | self._queue_filelock.Close()
|
1983 | a71f9c7d | Guido Trotter | self._queue_filelock = None |