root / lib / jqueue.py @ 5a672c30
History | View | Annotate | Download (45.9 kB)
1 | 498ae1cc | Iustin Pop | #
|
---|---|---|---|
2 | 498ae1cc | Iustin Pop | #
|
3 | 498ae1cc | Iustin Pop | |
4 | 5685c1a5 | Michael Hanselmann | # Copyright (C) 2006, 2007, 2008 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 | f1da30e6 | Michael Hanselmann | import os |
33 | e2715f69 | Michael Hanselmann | import logging |
34 | f1da30e6 | Michael Hanselmann | import errno |
35 | f1da30e6 | Michael Hanselmann | import re |
36 | f1048938 | Iustin Pop | import time |
37 | 5685c1a5 | Michael Hanselmann | import weakref |
38 | 498ae1cc | Iustin Pop | |
39 | 6c2549d6 | Guido Trotter | try:
|
40 | 6c2549d6 | Guido Trotter | # pylint: disable-msg=E0611
|
41 | 6c2549d6 | Guido Trotter | from pyinotify import pyinotify |
42 | 6c2549d6 | Guido Trotter | except ImportError: |
43 | 6c2549d6 | Guido Trotter | import pyinotify |
44 | 6c2549d6 | Guido Trotter | |
45 | 6c2549d6 | Guido Trotter | from ganeti import asyncnotifier |
46 | e2715f69 | Michael Hanselmann | from ganeti import constants |
47 | f1da30e6 | Michael Hanselmann | from ganeti import serializer |
48 | e2715f69 | Michael Hanselmann | from ganeti import workerpool |
49 | 99bd4f0a | Guido Trotter | from ganeti import locking |
50 | f1da30e6 | Michael Hanselmann | from ganeti import opcodes |
51 | 7a1ecaed | Iustin Pop | from ganeti import errors |
52 | e2715f69 | Michael Hanselmann | from ganeti import mcpu |
53 | 7996a135 | Iustin Pop | from ganeti import utils |
54 | 04ab05ce | Michael Hanselmann | from ganeti import jstore |
55 | c3f0a12f | Iustin Pop | from ganeti import rpc |
56 | e2715f69 | Michael Hanselmann | |
57 | fbf0262f | Michael Hanselmann | |
58 | 1daae384 | Iustin Pop | JOBQUEUE_THREADS = 25
|
59 | 58b22b6e | Michael Hanselmann | JOBS_PER_ARCHIVE_DIRECTORY = 10000
|
60 | e2715f69 | Michael Hanselmann | |
61 | ebb80afa | Guido Trotter | # member lock names to be passed to @ssynchronized decorator
|
62 | ebb80afa | Guido Trotter | _LOCK = "_lock"
|
63 | ebb80afa | Guido Trotter | _QUEUE = "_queue"
|
64 | 99bd4f0a | Guido Trotter | |
65 | 498ae1cc | Iustin Pop | |
66 | 9728ae5d | Iustin Pop | class CancelJob(Exception): |
67 | fbf0262f | Michael Hanselmann | """Special exception to cancel a job.
|
68 | fbf0262f | Michael Hanselmann |
|
69 | fbf0262f | Michael Hanselmann | """
|
70 | fbf0262f | Michael Hanselmann | |
71 | fbf0262f | Michael Hanselmann | |
72 | 70552c46 | Michael Hanselmann | def TimeStampNow(): |
73 | ea03467c | Iustin Pop | """Returns the current timestamp.
|
74 | ea03467c | Iustin Pop |
|
75 | ea03467c | Iustin Pop | @rtype: tuple
|
76 | ea03467c | Iustin Pop | @return: the current time in the (seconds, microseconds) format
|
77 | ea03467c | Iustin Pop |
|
78 | ea03467c | Iustin Pop | """
|
79 | 70552c46 | Michael Hanselmann | return utils.SplitTime(time.time())
|
80 | 70552c46 | Michael Hanselmann | |
81 | 70552c46 | Michael Hanselmann | |
82 | e2715f69 | Michael Hanselmann | class _QueuedOpCode(object): |
83 | 5bbd3f7f | Michael Hanselmann | """Encapsulates an opcode object.
|
84 | e2715f69 | Michael Hanselmann |
|
85 | ea03467c | Iustin Pop | @ivar log: holds the execution log and consists of tuples
|
86 | ea03467c | Iustin Pop | of the form C{(log_serial, timestamp, level, message)}
|
87 | ea03467c | Iustin Pop | @ivar input: the OpCode we encapsulate
|
88 | ea03467c | Iustin Pop | @ivar status: the current status
|
89 | ea03467c | Iustin Pop | @ivar result: the result of the LU execution
|
90 | ea03467c | Iustin Pop | @ivar start_timestamp: timestamp for the start of the execution
|
91 | b9b5abcb | Iustin Pop | @ivar exec_timestamp: timestamp for the actual LU Exec() function invocation
|
92 | ea03467c | Iustin Pop | @ivar stop_timestamp: timestamp for the end of the execution
|
93 | f1048938 | Iustin Pop |
|
94 | e2715f69 | Michael Hanselmann | """
|
95 | 66d895a8 | Iustin Pop | __slots__ = ["input", "status", "result", "log", |
96 | b9b5abcb | Iustin Pop | "start_timestamp", "exec_timestamp", "end_timestamp", |
97 | 66d895a8 | Iustin Pop | "__weakref__"]
|
98 | 66d895a8 | Iustin Pop | |
99 | 85f03e0d | Michael Hanselmann | def __init__(self, op): |
100 | ea03467c | Iustin Pop | """Constructor for the _QuededOpCode.
|
101 | ea03467c | Iustin Pop |
|
102 | ea03467c | Iustin Pop | @type op: L{opcodes.OpCode}
|
103 | ea03467c | Iustin Pop | @param op: the opcode we encapsulate
|
104 | ea03467c | Iustin Pop |
|
105 | ea03467c | Iustin Pop | """
|
106 | 85f03e0d | Michael Hanselmann | self.input = op
|
107 | 85f03e0d | Michael Hanselmann | self.status = constants.OP_STATUS_QUEUED
|
108 | 85f03e0d | Michael Hanselmann | self.result = None |
109 | 85f03e0d | Michael Hanselmann | self.log = []
|
110 | 70552c46 | Michael Hanselmann | self.start_timestamp = None |
111 | b9b5abcb | Iustin Pop | self.exec_timestamp = None |
112 | 70552c46 | Michael Hanselmann | self.end_timestamp = None |
113 | f1da30e6 | Michael Hanselmann | |
114 | f1da30e6 | Michael Hanselmann | @classmethod
|
115 | f1da30e6 | Michael Hanselmann | def Restore(cls, state): |
116 | ea03467c | Iustin Pop | """Restore the _QueuedOpCode from the serialized form.
|
117 | ea03467c | Iustin Pop |
|
118 | ea03467c | Iustin Pop | @type state: dict
|
119 | ea03467c | Iustin Pop | @param state: the serialized state
|
120 | ea03467c | Iustin Pop | @rtype: _QueuedOpCode
|
121 | ea03467c | Iustin Pop | @return: a new _QueuedOpCode instance
|
122 | ea03467c | Iustin Pop |
|
123 | ea03467c | Iustin Pop | """
|
124 | 85f03e0d | Michael Hanselmann | obj = _QueuedOpCode.__new__(cls) |
125 | 85f03e0d | Michael Hanselmann | obj.input = opcodes.OpCode.LoadOpCode(state["input"])
|
126 | 85f03e0d | Michael Hanselmann | obj.status = state["status"]
|
127 | 85f03e0d | Michael Hanselmann | obj.result = state["result"]
|
128 | 85f03e0d | Michael Hanselmann | obj.log = state["log"]
|
129 | 70552c46 | Michael Hanselmann | obj.start_timestamp = state.get("start_timestamp", None) |
130 | b9b5abcb | Iustin Pop | obj.exec_timestamp = state.get("exec_timestamp", None) |
131 | 70552c46 | Michael Hanselmann | obj.end_timestamp = state.get("end_timestamp", None) |
132 | f1da30e6 | Michael Hanselmann | return obj
|
133 | f1da30e6 | Michael Hanselmann | |
134 | f1da30e6 | Michael Hanselmann | def Serialize(self): |
135 | ea03467c | Iustin Pop | """Serializes this _QueuedOpCode.
|
136 | ea03467c | Iustin Pop |
|
137 | ea03467c | Iustin Pop | @rtype: dict
|
138 | ea03467c | Iustin Pop | @return: the dictionary holding the serialized state
|
139 | ea03467c | Iustin Pop |
|
140 | ea03467c | Iustin Pop | """
|
141 | 6c5a7090 | Michael Hanselmann | return {
|
142 | 6c5a7090 | Michael Hanselmann | "input": self.input.__getstate__(), |
143 | 6c5a7090 | Michael Hanselmann | "status": self.status, |
144 | 6c5a7090 | Michael Hanselmann | "result": self.result, |
145 | 6c5a7090 | Michael Hanselmann | "log": self.log, |
146 | 70552c46 | Michael Hanselmann | "start_timestamp": self.start_timestamp, |
147 | b9b5abcb | Iustin Pop | "exec_timestamp": self.exec_timestamp, |
148 | 70552c46 | Michael Hanselmann | "end_timestamp": self.end_timestamp, |
149 | 6c5a7090 | Michael Hanselmann | } |
150 | f1048938 | Iustin Pop | |
151 | e2715f69 | Michael Hanselmann | |
152 | e2715f69 | Michael Hanselmann | class _QueuedJob(object): |
153 | e2715f69 | Michael Hanselmann | """In-memory job representation.
|
154 | e2715f69 | Michael Hanselmann |
|
155 | ea03467c | Iustin Pop | This is what we use to track the user-submitted jobs. Locking must
|
156 | ea03467c | Iustin Pop | be taken care of by users of this class.
|
157 | ea03467c | Iustin Pop |
|
158 | ea03467c | Iustin Pop | @type queue: L{JobQueue}
|
159 | ea03467c | Iustin Pop | @ivar queue: the parent queue
|
160 | ea03467c | Iustin Pop | @ivar id: the job ID
|
161 | ea03467c | Iustin Pop | @type ops: list
|
162 | ea03467c | Iustin Pop | @ivar ops: the list of _QueuedOpCode that constitute the job
|
163 | ea03467c | Iustin Pop | @type log_serial: int
|
164 | ea03467c | Iustin Pop | @ivar log_serial: holds the index for the next log entry
|
165 | ea03467c | Iustin Pop | @ivar received_timestamp: the timestamp for when the job was received
|
166 | ea03467c | Iustin Pop | @ivar start_timestmap: the timestamp for start of execution
|
167 | ea03467c | Iustin Pop | @ivar end_timestamp: the timestamp for end of execution
|
168 | ef2df7d3 | Michael Hanselmann | @ivar lock_status: In-memory locking information for debugging
|
169 | e2715f69 | Michael Hanselmann |
|
170 | e2715f69 | Michael Hanselmann | """
|
171 | 7260cfbe | Iustin Pop | # pylint: disable-msg=W0212
|
172 | d25c1d6a | Michael Hanselmann | __slots__ = ["queue", "id", "ops", "log_serial", |
173 | 66d895a8 | Iustin Pop | "received_timestamp", "start_timestamp", "end_timestamp", |
174 | ef2df7d3 | Michael Hanselmann | "lock_status", "change", |
175 | 66d895a8 | Iustin Pop | "__weakref__"]
|
176 | 66d895a8 | Iustin Pop | |
177 | 85f03e0d | Michael Hanselmann | def __init__(self, queue, job_id, ops): |
178 | ea03467c | Iustin Pop | """Constructor for the _QueuedJob.
|
179 | ea03467c | Iustin Pop |
|
180 | ea03467c | Iustin Pop | @type queue: L{JobQueue}
|
181 | ea03467c | Iustin Pop | @param queue: our parent queue
|
182 | ea03467c | Iustin Pop | @type job_id: job_id
|
183 | ea03467c | Iustin Pop | @param job_id: our job id
|
184 | ea03467c | Iustin Pop | @type ops: list
|
185 | ea03467c | Iustin Pop | @param ops: the list of opcodes we hold, which will be encapsulated
|
186 | ea03467c | Iustin Pop | in _QueuedOpCodes
|
187 | ea03467c | Iustin Pop |
|
188 | ea03467c | Iustin Pop | """
|
189 | e2715f69 | Michael Hanselmann | if not ops: |
190 | c910bccb | Guido Trotter | raise errors.GenericError("A job needs at least one opcode") |
191 | e2715f69 | Michael Hanselmann | |
192 | 85f03e0d | Michael Hanselmann | self.queue = queue
|
193 | f1da30e6 | Michael Hanselmann | self.id = job_id
|
194 | 85f03e0d | Michael Hanselmann | self.ops = [_QueuedOpCode(op) for op in ops] |
195 | 6c5a7090 | Michael Hanselmann | self.log_serial = 0 |
196 | c56ec146 | Iustin Pop | self.received_timestamp = TimeStampNow()
|
197 | c56ec146 | Iustin Pop | self.start_timestamp = None |
198 | c56ec146 | Iustin Pop | self.end_timestamp = None |
199 | 6c5a7090 | Michael Hanselmann | |
200 | ef2df7d3 | Michael Hanselmann | # In-memory attributes
|
201 | ef2df7d3 | Michael Hanselmann | self.lock_status = None |
202 | ef2df7d3 | Michael Hanselmann | |
203 | 9fa2e150 | Michael Hanselmann | def __repr__(self): |
204 | 9fa2e150 | Michael Hanselmann | status = ["%s.%s" % (self.__class__.__module__, self.__class__.__name__), |
205 | 9fa2e150 | Michael Hanselmann | "id=%s" % self.id, |
206 | 9fa2e150 | Michael Hanselmann | "ops=%s" % ",".join([op.input.Summary() for op in self.ops])] |
207 | 9fa2e150 | Michael Hanselmann | |
208 | 9fa2e150 | Michael Hanselmann | return "<%s at %#x>" % (" ".join(status), id(self)) |
209 | 9fa2e150 | Michael Hanselmann | |
210 | f1da30e6 | Michael Hanselmann | @classmethod
|
211 | 85f03e0d | Michael Hanselmann | def Restore(cls, queue, state): |
212 | ea03467c | Iustin Pop | """Restore a _QueuedJob from serialized state:
|
213 | ea03467c | Iustin Pop |
|
214 | ea03467c | Iustin Pop | @type queue: L{JobQueue}
|
215 | ea03467c | Iustin Pop | @param queue: to which queue the restored job belongs
|
216 | ea03467c | Iustin Pop | @type state: dict
|
217 | ea03467c | Iustin Pop | @param state: the serialized state
|
218 | ea03467c | Iustin Pop | @rtype: _JobQueue
|
219 | ea03467c | Iustin Pop | @return: the restored _JobQueue instance
|
220 | ea03467c | Iustin Pop |
|
221 | ea03467c | Iustin Pop | """
|
222 | 85f03e0d | Michael Hanselmann | obj = _QueuedJob.__new__(cls) |
223 | 85f03e0d | Michael Hanselmann | obj.queue = queue |
224 | 85f03e0d | Michael Hanselmann | obj.id = state["id"]
|
225 | c56ec146 | Iustin Pop | obj.received_timestamp = state.get("received_timestamp", None) |
226 | c56ec146 | Iustin Pop | obj.start_timestamp = state.get("start_timestamp", None) |
227 | c56ec146 | Iustin Pop | obj.end_timestamp = state.get("end_timestamp", None) |
228 | 6c5a7090 | Michael Hanselmann | |
229 | ef2df7d3 | Michael Hanselmann | # In-memory attributes
|
230 | ef2df7d3 | Michael Hanselmann | obj.lock_status = None
|
231 | ef2df7d3 | Michael Hanselmann | |
232 | 6c5a7090 | Michael Hanselmann | obj.ops = [] |
233 | 6c5a7090 | Michael Hanselmann | obj.log_serial = 0
|
234 | 6c5a7090 | Michael Hanselmann | for op_state in state["ops"]: |
235 | 6c5a7090 | Michael Hanselmann | op = _QueuedOpCode.Restore(op_state) |
236 | 6c5a7090 | Michael Hanselmann | for log_entry in op.log: |
237 | 6c5a7090 | Michael Hanselmann | obj.log_serial = max(obj.log_serial, log_entry[0]) |
238 | 6c5a7090 | Michael Hanselmann | obj.ops.append(op) |
239 | 6c5a7090 | Michael Hanselmann | |
240 | f1da30e6 | Michael Hanselmann | return obj
|
241 | f1da30e6 | Michael Hanselmann | |
242 | f1da30e6 | Michael Hanselmann | def Serialize(self): |
243 | ea03467c | Iustin Pop | """Serialize the _JobQueue instance.
|
244 | ea03467c | Iustin Pop |
|
245 | ea03467c | Iustin Pop | @rtype: dict
|
246 | ea03467c | Iustin Pop | @return: the serialized state
|
247 | ea03467c | Iustin Pop |
|
248 | ea03467c | Iustin Pop | """
|
249 | f1da30e6 | Michael Hanselmann | return {
|
250 | f1da30e6 | Michael Hanselmann | "id": self.id, |
251 | 85f03e0d | Michael Hanselmann | "ops": [op.Serialize() for op in self.ops], |
252 | c56ec146 | Iustin Pop | "start_timestamp": self.start_timestamp, |
253 | c56ec146 | Iustin Pop | "end_timestamp": self.end_timestamp, |
254 | c56ec146 | Iustin Pop | "received_timestamp": self.received_timestamp, |
255 | f1da30e6 | Michael Hanselmann | } |
256 | f1da30e6 | Michael Hanselmann | |
257 | 85f03e0d | Michael Hanselmann | def CalcStatus(self): |
258 | ea03467c | Iustin Pop | """Compute the status of this job.
|
259 | ea03467c | Iustin Pop |
|
260 | ea03467c | Iustin Pop | This function iterates over all the _QueuedOpCodes in the job and
|
261 | ea03467c | Iustin Pop | based on their status, computes the job status.
|
262 | ea03467c | Iustin Pop |
|
263 | ea03467c | Iustin Pop | The algorithm is:
|
264 | ea03467c | Iustin Pop | - if we find a cancelled, or finished with error, the job
|
265 | ea03467c | Iustin Pop | status will be the same
|
266 | ea03467c | Iustin Pop | - otherwise, the last opcode with the status one of:
|
267 | ea03467c | Iustin Pop | - waitlock
|
268 | fbf0262f | Michael Hanselmann | - canceling
|
269 | ea03467c | Iustin Pop | - running
|
270 | ea03467c | Iustin Pop |
|
271 | ea03467c | Iustin Pop | will determine the job status
|
272 | ea03467c | Iustin Pop |
|
273 | ea03467c | Iustin Pop | - otherwise, it means either all opcodes are queued, or success,
|
274 | ea03467c | Iustin Pop | and the job status will be the same
|
275 | ea03467c | Iustin Pop |
|
276 | ea03467c | Iustin Pop | @return: the job status
|
277 | ea03467c | Iustin Pop |
|
278 | ea03467c | Iustin Pop | """
|
279 | e2715f69 | Michael Hanselmann | status = constants.JOB_STATUS_QUEUED |
280 | e2715f69 | Michael Hanselmann | |
281 | e2715f69 | Michael Hanselmann | all_success = True
|
282 | 85f03e0d | Michael Hanselmann | for op in self.ops: |
283 | 85f03e0d | Michael Hanselmann | if op.status == constants.OP_STATUS_SUCCESS:
|
284 | e2715f69 | Michael Hanselmann | continue
|
285 | e2715f69 | Michael Hanselmann | |
286 | e2715f69 | Michael Hanselmann | all_success = False
|
287 | e2715f69 | Michael Hanselmann | |
288 | 85f03e0d | Michael Hanselmann | if op.status == constants.OP_STATUS_QUEUED:
|
289 | e2715f69 | Michael Hanselmann | pass
|
290 | e92376d7 | Iustin Pop | elif op.status == constants.OP_STATUS_WAITLOCK:
|
291 | e92376d7 | Iustin Pop | status = constants.JOB_STATUS_WAITLOCK |
292 | 85f03e0d | Michael Hanselmann | elif op.status == constants.OP_STATUS_RUNNING:
|
293 | e2715f69 | Michael Hanselmann | status = constants.JOB_STATUS_RUNNING |
294 | fbf0262f | Michael Hanselmann | elif op.status == constants.OP_STATUS_CANCELING:
|
295 | fbf0262f | Michael Hanselmann | status = constants.JOB_STATUS_CANCELING |
296 | fbf0262f | Michael Hanselmann | break
|
297 | 85f03e0d | Michael Hanselmann | elif op.status == constants.OP_STATUS_ERROR:
|
298 | f1da30e6 | Michael Hanselmann | status = constants.JOB_STATUS_ERROR |
299 | f1da30e6 | Michael Hanselmann | # The whole job fails if one opcode failed
|
300 | f1da30e6 | Michael Hanselmann | break
|
301 | 85f03e0d | Michael Hanselmann | elif op.status == constants.OP_STATUS_CANCELED:
|
302 | 4cb1d919 | Michael Hanselmann | status = constants.OP_STATUS_CANCELED |
303 | 4cb1d919 | Michael Hanselmann | break
|
304 | e2715f69 | Michael Hanselmann | |
305 | e2715f69 | Michael Hanselmann | if all_success:
|
306 | e2715f69 | Michael Hanselmann | status = constants.JOB_STATUS_SUCCESS |
307 | e2715f69 | Michael Hanselmann | |
308 | e2715f69 | Michael Hanselmann | return status
|
309 | e2715f69 | Michael Hanselmann | |
310 | 6c5a7090 | Michael Hanselmann | def GetLogEntries(self, newer_than): |
311 | ea03467c | Iustin Pop | """Selectively returns the log entries.
|
312 | ea03467c | Iustin Pop |
|
313 | ea03467c | Iustin Pop | @type newer_than: None or int
|
314 | 5bbd3f7f | Michael Hanselmann | @param newer_than: if this is None, return all log entries,
|
315 | ea03467c | Iustin Pop | otherwise return only the log entries with serial higher
|
316 | ea03467c | Iustin Pop | than this value
|
317 | ea03467c | Iustin Pop | @rtype: list
|
318 | ea03467c | Iustin Pop | @return: the list of the log entries selected
|
319 | ea03467c | Iustin Pop |
|
320 | ea03467c | Iustin Pop | """
|
321 | 6c5a7090 | Michael Hanselmann | if newer_than is None: |
322 | 6c5a7090 | Michael Hanselmann | serial = -1
|
323 | 6c5a7090 | Michael Hanselmann | else:
|
324 | 6c5a7090 | Michael Hanselmann | serial = newer_than |
325 | 6c5a7090 | Michael Hanselmann | |
326 | 6c5a7090 | Michael Hanselmann | entries = [] |
327 | 6c5a7090 | Michael Hanselmann | for op in self.ops: |
328 | 63712a09 | Iustin Pop | entries.extend(filter(lambda entry: entry[0] > serial, op.log)) |
329 | 6c5a7090 | Michael Hanselmann | |
330 | 6c5a7090 | Michael Hanselmann | return entries
|
331 | 6c5a7090 | Michael Hanselmann | |
332 | 6a290889 | Guido Trotter | def GetInfo(self, fields): |
333 | 6a290889 | Guido Trotter | """Returns information about a job.
|
334 | 6a290889 | Guido Trotter |
|
335 | 6a290889 | Guido Trotter | @type fields: list
|
336 | 6a290889 | Guido Trotter | @param fields: names of fields to return
|
337 | 6a290889 | Guido Trotter | @rtype: list
|
338 | 6a290889 | Guido Trotter | @return: list with one element for each field
|
339 | 6a290889 | Guido Trotter | @raise errors.OpExecError: when an invalid field
|
340 | 6a290889 | Guido Trotter | has been passed
|
341 | 6a290889 | Guido Trotter |
|
342 | 6a290889 | Guido Trotter | """
|
343 | 6a290889 | Guido Trotter | row = [] |
344 | 6a290889 | Guido Trotter | for fname in fields: |
345 | 6a290889 | Guido Trotter | if fname == "id": |
346 | 6a290889 | Guido Trotter | row.append(self.id)
|
347 | 6a290889 | Guido Trotter | elif fname == "status": |
348 | 6a290889 | Guido Trotter | row.append(self.CalcStatus())
|
349 | 6a290889 | Guido Trotter | elif fname == "ops": |
350 | 6a290889 | Guido Trotter | row.append([op.input.__getstate__() for op in self.ops]) |
351 | 6a290889 | Guido Trotter | elif fname == "opresult": |
352 | 6a290889 | Guido Trotter | row.append([op.result for op in self.ops]) |
353 | 6a290889 | Guido Trotter | elif fname == "opstatus": |
354 | 6a290889 | Guido Trotter | row.append([op.status for op in self.ops]) |
355 | 6a290889 | Guido Trotter | elif fname == "oplog": |
356 | 6a290889 | Guido Trotter | row.append([op.log for op in self.ops]) |
357 | 6a290889 | Guido Trotter | elif fname == "opstart": |
358 | 6a290889 | Guido Trotter | row.append([op.start_timestamp for op in self.ops]) |
359 | 6a290889 | Guido Trotter | elif fname == "opexec": |
360 | 6a290889 | Guido Trotter | row.append([op.exec_timestamp for op in self.ops]) |
361 | 6a290889 | Guido Trotter | elif fname == "opend": |
362 | 6a290889 | Guido Trotter | row.append([op.end_timestamp for op in self.ops]) |
363 | 6a290889 | Guido Trotter | elif fname == "received_ts": |
364 | 6a290889 | Guido Trotter | row.append(self.received_timestamp)
|
365 | 6a290889 | Guido Trotter | elif fname == "start_ts": |
366 | 6a290889 | Guido Trotter | row.append(self.start_timestamp)
|
367 | 6a290889 | Guido Trotter | elif fname == "end_ts": |
368 | 6a290889 | Guido Trotter | row.append(self.end_timestamp)
|
369 | 6a290889 | Guido Trotter | elif fname == "lock_status": |
370 | 6a290889 | Guido Trotter | row.append(self.lock_status)
|
371 | 6a290889 | Guido Trotter | elif fname == "summary": |
372 | 6a290889 | Guido Trotter | row.append([op.input.Summary() for op in self.ops]) |
373 | 6a290889 | Guido Trotter | else:
|
374 | 6a290889 | Guido Trotter | raise errors.OpExecError("Invalid self query field '%s'" % fname) |
375 | 6a290889 | Guido Trotter | return row
|
376 | 6a290889 | Guido Trotter | |
377 | 34327f51 | Iustin Pop | def MarkUnfinishedOps(self, status, result): |
378 | 34327f51 | Iustin Pop | """Mark unfinished opcodes with a given status and result.
|
379 | 34327f51 | Iustin Pop |
|
380 | 34327f51 | Iustin Pop | This is an utility function for marking all running or waiting to
|
381 | 34327f51 | Iustin Pop | be run opcodes with a given status. Opcodes which are already
|
382 | 34327f51 | Iustin Pop | finalised are not changed.
|
383 | 34327f51 | Iustin Pop |
|
384 | 34327f51 | Iustin Pop | @param status: a given opcode status
|
385 | 34327f51 | Iustin Pop | @param result: the opcode result
|
386 | 34327f51 | Iustin Pop |
|
387 | 34327f51 | Iustin Pop | """
|
388 | 39ed3a98 | Guido Trotter | try:
|
389 | 39ed3a98 | Guido Trotter | not_marked = True
|
390 | 39ed3a98 | Guido Trotter | for op in self.ops: |
391 | 39ed3a98 | Guido Trotter | if op.status in constants.OPS_FINALIZED: |
392 | 39ed3a98 | Guido Trotter | assert not_marked, "Finalized opcodes found after non-finalized ones" |
393 | 39ed3a98 | Guido Trotter | continue
|
394 | 39ed3a98 | Guido Trotter | op.status = status |
395 | 39ed3a98 | Guido Trotter | op.result = result |
396 | 39ed3a98 | Guido Trotter | not_marked = False
|
397 | 39ed3a98 | Guido Trotter | finally:
|
398 | 39ed3a98 | Guido Trotter | self.queue.UpdateJobUnlocked(self) |
399 | 34327f51 | Iustin Pop | |
400 | f1048938 | Iustin Pop | |
401 | ef2df7d3 | Michael Hanselmann | class _OpExecCallbacks(mcpu.OpExecCbBase): |
402 | 031a3e57 | Michael Hanselmann | def __init__(self, queue, job, op): |
403 | 031a3e57 | Michael Hanselmann | """Initializes this class.
|
404 | ea03467c | Iustin Pop |
|
405 | 031a3e57 | Michael Hanselmann | @type queue: L{JobQueue}
|
406 | 031a3e57 | Michael Hanselmann | @param queue: Job queue
|
407 | 031a3e57 | Michael Hanselmann | @type job: L{_QueuedJob}
|
408 | 031a3e57 | Michael Hanselmann | @param job: Job object
|
409 | 031a3e57 | Michael Hanselmann | @type op: L{_QueuedOpCode}
|
410 | 031a3e57 | Michael Hanselmann | @param op: OpCode
|
411 | 031a3e57 | Michael Hanselmann |
|
412 | 031a3e57 | Michael Hanselmann | """
|
413 | 031a3e57 | Michael Hanselmann | assert queue, "Queue is missing" |
414 | 031a3e57 | Michael Hanselmann | assert job, "Job is missing" |
415 | 031a3e57 | Michael Hanselmann | assert op, "Opcode is missing" |
416 | 031a3e57 | Michael Hanselmann | |
417 | 031a3e57 | Michael Hanselmann | self._queue = queue
|
418 | 031a3e57 | Michael Hanselmann | self._job = job
|
419 | 031a3e57 | Michael Hanselmann | self._op = op
|
420 | 031a3e57 | Michael Hanselmann | |
421 | 031a3e57 | Michael Hanselmann | def NotifyStart(self): |
422 | e92376d7 | Iustin Pop | """Mark the opcode as running, not lock-waiting.
|
423 | e92376d7 | Iustin Pop |
|
424 | 031a3e57 | Michael Hanselmann | This is called from the mcpu code as a notifier function, when the LU is
|
425 | 031a3e57 | Michael Hanselmann | finally about to start the Exec() method. Of course, to have end-user
|
426 | 031a3e57 | Michael Hanselmann | visible results, the opcode must be initially (before calling into
|
427 | 031a3e57 | Michael Hanselmann | Processor.ExecOpCode) set to OP_STATUS_WAITLOCK.
|
428 | e92376d7 | Iustin Pop |
|
429 | e92376d7 | Iustin Pop | """
|
430 | 3c0d60d0 | Guido Trotter | self._queue.acquire(shared=1) |
431 | e92376d7 | Iustin Pop | try:
|
432 | 031a3e57 | Michael Hanselmann | assert self._op.status in (constants.OP_STATUS_WAITLOCK, |
433 | 031a3e57 | Michael Hanselmann | constants.OP_STATUS_CANCELING) |
434 | fbf0262f | Michael Hanselmann | |
435 | ef2df7d3 | Michael Hanselmann | # All locks are acquired by now
|
436 | ef2df7d3 | Michael Hanselmann | self._job.lock_status = None |
437 | ef2df7d3 | Michael Hanselmann | |
438 | fbf0262f | Michael Hanselmann | # Cancel here if we were asked to
|
439 | 031a3e57 | Michael Hanselmann | if self._op.status == constants.OP_STATUS_CANCELING: |
440 | fbf0262f | Michael Hanselmann | raise CancelJob()
|
441 | fbf0262f | Michael Hanselmann | |
442 | 031a3e57 | Michael Hanselmann | self._op.status = constants.OP_STATUS_RUNNING
|
443 | b9b5abcb | Iustin Pop | self._op.exec_timestamp = TimeStampNow()
|
444 | e92376d7 | Iustin Pop | finally:
|
445 | 031a3e57 | Michael Hanselmann | self._queue.release()
|
446 | 031a3e57 | Michael Hanselmann | |
447 | ebb80afa | Guido Trotter | @locking.ssynchronized(_QUEUE, shared=1) |
448 | 9bf5e01f | Guido Trotter | def _AppendFeedback(self, timestamp, log_type, log_msg): |
449 | 9bf5e01f | Guido Trotter | """Internal feedback append function, with locks
|
450 | 9bf5e01f | Guido Trotter |
|
451 | 9bf5e01f | Guido Trotter | """
|
452 | 9bf5e01f | Guido Trotter | self._job.log_serial += 1 |
453 | 9bf5e01f | Guido Trotter | self._op.log.append((self._job.log_serial, timestamp, log_type, log_msg)) |
454 | 9bf5e01f | Guido Trotter | self._queue.UpdateJobUnlocked(self._job, replicate=False) |
455 | 9bf5e01f | Guido Trotter | |
456 | 031a3e57 | Michael Hanselmann | def Feedback(self, *args): |
457 | 031a3e57 | Michael Hanselmann | """Append a log entry.
|
458 | 031a3e57 | Michael Hanselmann |
|
459 | 031a3e57 | Michael Hanselmann | """
|
460 | 031a3e57 | Michael Hanselmann | assert len(args) < 3 |
461 | 031a3e57 | Michael Hanselmann | |
462 | 031a3e57 | Michael Hanselmann | if len(args) == 1: |
463 | 031a3e57 | Michael Hanselmann | log_type = constants.ELOG_MESSAGE |
464 | 031a3e57 | Michael Hanselmann | log_msg = args[0]
|
465 | 031a3e57 | Michael Hanselmann | else:
|
466 | 031a3e57 | Michael Hanselmann | (log_type, log_msg) = args |
467 | 031a3e57 | Michael Hanselmann | |
468 | 031a3e57 | Michael Hanselmann | # The time is split to make serialization easier and not lose
|
469 | 031a3e57 | Michael Hanselmann | # precision.
|
470 | 031a3e57 | Michael Hanselmann | timestamp = utils.SplitTime(time.time()) |
471 | 9bf5e01f | Guido Trotter | self._AppendFeedback(timestamp, log_type, log_msg)
|
472 | 031a3e57 | Michael Hanselmann | |
473 | ef2df7d3 | Michael Hanselmann | def ReportLocks(self, msg): |
474 | ef2df7d3 | Michael Hanselmann | """Write locking information to the job.
|
475 | ef2df7d3 | Michael Hanselmann |
|
476 | ef2df7d3 | Michael Hanselmann | Called whenever the LU processor is waiting for a lock or has acquired one.
|
477 | ef2df7d3 | Michael Hanselmann |
|
478 | ef2df7d3 | Michael Hanselmann | """
|
479 | ef2df7d3 | Michael Hanselmann | # Not getting the queue lock because this is a single assignment
|
480 | ef2df7d3 | Michael Hanselmann | self._job.lock_status = msg
|
481 | ef2df7d3 | Michael Hanselmann | |
482 | 031a3e57 | Michael Hanselmann | |
483 | 6c2549d6 | Guido Trotter | class _WaitForJobChangesHelper(object): |
484 | 6c2549d6 | Guido Trotter | """Helper class using initofy to wait for changes in a job file.
|
485 | 6c2549d6 | Guido Trotter |
|
486 | 6c2549d6 | Guido Trotter | This class takes a previous job status and serial, and alerts the client when
|
487 | 6c2549d6 | Guido Trotter | the current job status has changed.
|
488 | 6c2549d6 | Guido Trotter |
|
489 | 6c2549d6 | Guido Trotter | @type job_id: string
|
490 | 6c2549d6 | Guido Trotter | @ivar job_id: id of the job we're watching
|
491 | 6c2549d6 | Guido Trotter | @type prev_job_info: string
|
492 | 6c2549d6 | Guido Trotter | @ivar prev_job_info: previous job info, as passed by the luxi client
|
493 | 6c2549d6 | Guido Trotter | @type prev_log_serial: string
|
494 | 6c2549d6 | Guido Trotter | @ivar prev_log_serial: previous job serial, as passed by the luxi client
|
495 | 6c2549d6 | Guido Trotter | @type queue: L{JobQueue}
|
496 | 6c2549d6 | Guido Trotter | @ivar queue: job queue (used for a few utility functions)
|
497 | 6c2549d6 | Guido Trotter | @type job_path: string
|
498 | 6c2549d6 | Guido Trotter | @ivar job_path: absolute path of the job file
|
499 | 6c2549d6 | Guido Trotter | @type wm: pyinotify.WatchManager (or None)
|
500 | 6c2549d6 | Guido Trotter | @ivar wm: inotify watch manager to watch for changes
|
501 | 6c2549d6 | Guido Trotter | @type inotify_handler: L{asyncnotifier.SingleFileEventHandler}
|
502 | 6c2549d6 | Guido Trotter | @ivar inotify_handler: single file event handler, used for watching
|
503 | 6c2549d6 | Guido Trotter | @type notifier: pyinotify.Notifier
|
504 | 6c2549d6 | Guido Trotter | @ivar notifier: inotify single-threaded notifier, used for watching
|
505 | 6c2549d6 | Guido Trotter |
|
506 | 6c2549d6 | Guido Trotter | """
|
507 | 6c2549d6 | Guido Trotter | def __init__(self, job_id, fields, prev_job_info, prev_log_serial, queue): |
508 | 6c2549d6 | Guido Trotter | self.job_id = job_id
|
509 | 6c2549d6 | Guido Trotter | self.fields = fields
|
510 | 6c2549d6 | Guido Trotter | self.prev_job_info = prev_job_info
|
511 | 6c2549d6 | Guido Trotter | self.prev_log_serial = prev_log_serial
|
512 | 6c2549d6 | Guido Trotter | self.queue = queue
|
513 | 6c2549d6 | Guido Trotter | # pylint: disable-msg=W0212
|
514 | 6c2549d6 | Guido Trotter | self.job_path = self.queue._GetJobPath(self.job_id) |
515 | 6c2549d6 | Guido Trotter | self.wm = None |
516 | 6c2549d6 | Guido Trotter | self.inotify_handler = None |
517 | 6c2549d6 | Guido Trotter | self.notifier = None |
518 | 6c2549d6 | Guido Trotter | |
519 | 6c2549d6 | Guido Trotter | def _SetupInotify(self): |
520 | 6c2549d6 | Guido Trotter | """Create the inotify
|
521 | 6c2549d6 | Guido Trotter |
|
522 | 6c2549d6 | Guido Trotter | @raises errors.InotifyError: if the notifier cannot be setup
|
523 | 6c2549d6 | Guido Trotter |
|
524 | 6c2549d6 | Guido Trotter | """
|
525 | 6c2549d6 | Guido Trotter | if self.wm: |
526 | 6c2549d6 | Guido Trotter | return
|
527 | 6c2549d6 | Guido Trotter | self.wm = pyinotify.WatchManager()
|
528 | 6c2549d6 | Guido Trotter | self.inotify_handler = asyncnotifier.SingleFileEventHandler(self.wm, |
529 | 6c2549d6 | Guido Trotter | self.OnInotify,
|
530 | 6c2549d6 | Guido Trotter | self.job_path)
|
531 | 6c2549d6 | Guido Trotter | self.notifier = pyinotify.Notifier(self.wm, self.inotify_handler) |
532 | 6c2549d6 | Guido Trotter | self.inotify_handler.enable()
|
533 | 6c2549d6 | Guido Trotter | |
534 | 6c2549d6 | Guido Trotter | def _LoadDiskStatus(self): |
535 | 6c2549d6 | Guido Trotter | job = self.queue.SafeLoadJobFromDisk(self.job_id) |
536 | 6c2549d6 | Guido Trotter | if not job: |
537 | 6c2549d6 | Guido Trotter | raise errors.JobLost()
|
538 | 6c2549d6 | Guido Trotter | self.job_status = job.CalcStatus()
|
539 | 6c2549d6 | Guido Trotter | |
540 | 6c2549d6 | Guido Trotter | job_info = job.GetInfo(self.fields)
|
541 | 6c2549d6 | Guido Trotter | log_entries = job.GetLogEntries(self.prev_log_serial)
|
542 | 6c2549d6 | Guido Trotter | # Serializing and deserializing data can cause type changes (e.g. from
|
543 | 6c2549d6 | Guido Trotter | # tuple to list) or precision loss. We're doing it here so that we get
|
544 | 6c2549d6 | Guido Trotter | # the same modifications as the data received from the client. Without
|
545 | 6c2549d6 | Guido Trotter | # this, the comparison afterwards might fail without the data being
|
546 | 6c2549d6 | Guido Trotter | # significantly different.
|
547 | 6c2549d6 | Guido Trotter | # TODO: we just deserialized from disk, investigate how to make sure that
|
548 | 6c2549d6 | Guido Trotter | # the job info and log entries are compatible to avoid this further step.
|
549 | 6c2549d6 | Guido Trotter | self.job_info = serializer.LoadJson(serializer.DumpJson(job_info))
|
550 | 6c2549d6 | Guido Trotter | self.log_entries = serializer.LoadJson(serializer.DumpJson(log_entries))
|
551 | 6c2549d6 | Guido Trotter | |
552 | 6c2549d6 | Guido Trotter | def _CheckForChanges(self): |
553 | 6c2549d6 | Guido Trotter | self._LoadDiskStatus()
|
554 | 6c2549d6 | Guido Trotter | # Don't even try to wait if the job is no longer running, there will be
|
555 | 6c2549d6 | Guido Trotter | # no changes.
|
556 | 6c2549d6 | Guido Trotter | if (self.job_status not in (constants.JOB_STATUS_QUEUED, |
557 | 6c2549d6 | Guido Trotter | constants.JOB_STATUS_RUNNING, |
558 | 6c2549d6 | Guido Trotter | constants.JOB_STATUS_WAITLOCK) or
|
559 | 6c2549d6 | Guido Trotter | self.prev_job_info != self.job_info or |
560 | 6c2549d6 | Guido Trotter | (self.log_entries and self.prev_log_serial != self.log_entries[0][0])): |
561 | 6c2549d6 | Guido Trotter | logging.debug("Job %s changed", self.job_id) |
562 | 6c2549d6 | Guido Trotter | return (self.job_info, self.log_entries) |
563 | 6c2549d6 | Guido Trotter | |
564 | 6c2549d6 | Guido Trotter | raise utils.RetryAgain()
|
565 | 6c2549d6 | Guido Trotter | |
566 | 6c2549d6 | Guido Trotter | def OnInotify(self, notifier_enabled): |
567 | 6c2549d6 | Guido Trotter | if not notifier_enabled: |
568 | 6c2549d6 | Guido Trotter | self.inotify_handler.enable()
|
569 | 6c2549d6 | Guido Trotter | |
570 | 6c2549d6 | Guido Trotter | def WaitFn(self, timeout): |
571 | 6c2549d6 | Guido Trotter | self._SetupInotify()
|
572 | 6c2549d6 | Guido Trotter | if self.notifier.check_events(timeout*1000): |
573 | 6c2549d6 | Guido Trotter | self.notifier.read_events()
|
574 | 6c2549d6 | Guido Trotter | self.notifier.process_events()
|
575 | 6c2549d6 | Guido Trotter | |
576 | 6c2549d6 | Guido Trotter | def WaitForChanges(self, timeout): |
577 | 6c2549d6 | Guido Trotter | try:
|
578 | 6c2549d6 | Guido Trotter | return utils.Retry(self._CheckForChanges, |
579 | 6c2549d6 | Guido Trotter | utils.RETRY_REMAINING_TIME, |
580 | 6c2549d6 | Guido Trotter | timeout, |
581 | 6c2549d6 | Guido Trotter | wait_fn=self.WaitFn)
|
582 | 6c2549d6 | Guido Trotter | except (errors.InotifyError, errors.JobLost):
|
583 | 6c2549d6 | Guido Trotter | return None |
584 | 6c2549d6 | Guido Trotter | except utils.RetryTimeout:
|
585 | 6c2549d6 | Guido Trotter | return constants.JOB_NOTCHANGED
|
586 | 6c2549d6 | Guido Trotter | |
587 | 6c2549d6 | Guido Trotter | def Close(self): |
588 | 6c2549d6 | Guido Trotter | if self.wm: |
589 | 6c2549d6 | Guido Trotter | self.notifier.stop()
|
590 | 6c2549d6 | Guido Trotter | |
591 | 6c2549d6 | Guido Trotter | |
592 | 031a3e57 | Michael Hanselmann | class _JobQueueWorker(workerpool.BaseWorker): |
593 | 031a3e57 | Michael Hanselmann | """The actual job workers.
|
594 | 031a3e57 | Michael Hanselmann |
|
595 | 031a3e57 | Michael Hanselmann | """
|
596 | 7260cfbe | Iustin Pop | def RunTask(self, job): # pylint: disable-msg=W0221 |
597 | e2715f69 | Michael Hanselmann | """Job executor.
|
598 | e2715f69 | Michael Hanselmann |
|
599 | 6c5a7090 | Michael Hanselmann | This functions processes a job. It is closely tied to the _QueuedJob and
|
600 | 6c5a7090 | Michael Hanselmann | _QueuedOpCode classes.
|
601 | e2715f69 | Michael Hanselmann |
|
602 | ea03467c | Iustin Pop | @type job: L{_QueuedJob}
|
603 | ea03467c | Iustin Pop | @param job: the job to be processed
|
604 | ea03467c | Iustin Pop |
|
605 | e2715f69 | Michael Hanselmann | """
|
606 | 02fc74da | Michael Hanselmann | logging.info("Processing job %s", job.id)
|
607 | adfa97e3 | Guido Trotter | proc = mcpu.Processor(self.pool.queue.context, job.id)
|
608 | 031a3e57 | Michael Hanselmann | queue = job.queue |
609 | e2715f69 | Michael Hanselmann | try:
|
610 | 85f03e0d | Michael Hanselmann | try:
|
611 | 85f03e0d | Michael Hanselmann | count = len(job.ops)
|
612 | 85f03e0d | Michael Hanselmann | for idx, op in enumerate(job.ops): |
613 | d21d09d6 | Iustin Pop | op_summary = op.input.Summary() |
614 | f6424741 | Iustin Pop | if op.status == constants.OP_STATUS_SUCCESS:
|
615 | f6424741 | Iustin Pop | # this is a job that was partially completed before master
|
616 | f6424741 | Iustin Pop | # daemon shutdown, so it can be expected that some opcodes
|
617 | f6424741 | Iustin Pop | # are already completed successfully (if any did error
|
618 | f6424741 | Iustin Pop | # out, then the whole job should have been aborted and not
|
619 | f6424741 | Iustin Pop | # resubmitted for processing)
|
620 | f6424741 | Iustin Pop | logging.info("Op %s/%s: opcode %s already processed, skipping",
|
621 | f6424741 | Iustin Pop | idx + 1, count, op_summary)
|
622 | f6424741 | Iustin Pop | continue
|
623 | 85f03e0d | Michael Hanselmann | try:
|
624 | d21d09d6 | Iustin Pop | logging.info("Op %s/%s: Starting opcode %s", idx + 1, count, |
625 | d21d09d6 | Iustin Pop | op_summary) |
626 | 85f03e0d | Michael Hanselmann | |
627 | 3c0d60d0 | Guido Trotter | queue.acquire(shared=1)
|
628 | 85f03e0d | Michael Hanselmann | try:
|
629 | df0fb067 | Iustin Pop | if op.status == constants.OP_STATUS_CANCELED:
|
630 | df0fb067 | Iustin Pop | raise CancelJob()
|
631 | fbf0262f | Michael Hanselmann | assert op.status == constants.OP_STATUS_QUEUED
|
632 | e92376d7 | Iustin Pop | op.status = constants.OP_STATUS_WAITLOCK |
633 | 85f03e0d | Michael Hanselmann | op.result = None
|
634 | 70552c46 | Michael Hanselmann | op.start_timestamp = TimeStampNow() |
635 | c56ec146 | Iustin Pop | if idx == 0: # first opcode |
636 | c56ec146 | Iustin Pop | job.start_timestamp = op.start_timestamp |
637 | 85f03e0d | Michael Hanselmann | queue.UpdateJobUnlocked(job) |
638 | 85f03e0d | Michael Hanselmann | |
639 | 38206f3c | Iustin Pop | input_opcode = op.input |
640 | 85f03e0d | Michael Hanselmann | finally:
|
641 | 85f03e0d | Michael Hanselmann | queue.release() |
642 | 85f03e0d | Michael Hanselmann | |
643 | 031a3e57 | Michael Hanselmann | # Make sure not to hold queue lock while calling ExecOpCode
|
644 | 031a3e57 | Michael Hanselmann | result = proc.ExecOpCode(input_opcode, |
645 | ef2df7d3 | Michael Hanselmann | _OpExecCallbacks(queue, job, op)) |
646 | 85f03e0d | Michael Hanselmann | |
647 | 3c0d60d0 | Guido Trotter | queue.acquire(shared=1)
|
648 | 85f03e0d | Michael Hanselmann | try:
|
649 | 85f03e0d | Michael Hanselmann | op.status = constants.OP_STATUS_SUCCESS |
650 | 85f03e0d | Michael Hanselmann | op.result = result |
651 | 70552c46 | Michael Hanselmann | op.end_timestamp = TimeStampNow() |
652 | 85f03e0d | Michael Hanselmann | queue.UpdateJobUnlocked(job) |
653 | 85f03e0d | Michael Hanselmann | finally:
|
654 | 85f03e0d | Michael Hanselmann | queue.release() |
655 | 85f03e0d | Michael Hanselmann | |
656 | d21d09d6 | Iustin Pop | logging.info("Op %s/%s: Successfully finished opcode %s",
|
657 | d21d09d6 | Iustin Pop | idx + 1, count, op_summary)
|
658 | fbf0262f | Michael Hanselmann | except CancelJob:
|
659 | fbf0262f | Michael Hanselmann | # Will be handled further up
|
660 | fbf0262f | Michael Hanselmann | raise
|
661 | 85f03e0d | Michael Hanselmann | except Exception, err: |
662 | 3c0d60d0 | Guido Trotter | queue.acquire(shared=1)
|
663 | 85f03e0d | Michael Hanselmann | try:
|
664 | 85f03e0d | Michael Hanselmann | try:
|
665 | 85f03e0d | Michael Hanselmann | op.status = constants.OP_STATUS_ERROR |
666 | bcb66fca | Iustin Pop | if isinstance(err, errors.GenericError): |
667 | bcb66fca | Iustin Pop | op.result = errors.EncodeException(err) |
668 | bcb66fca | Iustin Pop | else:
|
669 | bcb66fca | Iustin Pop | op.result = str(err)
|
670 | 70552c46 | Michael Hanselmann | op.end_timestamp = TimeStampNow() |
671 | 0f6be82a | Iustin Pop | logging.info("Op %s/%s: Error in opcode %s: %s",
|
672 | 0f6be82a | Iustin Pop | idx + 1, count, op_summary, err)
|
673 | 85f03e0d | Michael Hanselmann | finally:
|
674 | 85f03e0d | Michael Hanselmann | queue.UpdateJobUnlocked(job) |
675 | 85f03e0d | Michael Hanselmann | finally:
|
676 | 85f03e0d | Michael Hanselmann | queue.release() |
677 | 85f03e0d | Michael Hanselmann | raise
|
678 | 85f03e0d | Michael Hanselmann | |
679 | fbf0262f | Michael Hanselmann | except CancelJob:
|
680 | 3c0d60d0 | Guido Trotter | queue.acquire(shared=1)
|
681 | fbf0262f | Michael Hanselmann | try:
|
682 | 39ed3a98 | Guido Trotter | job.MarkUnfinishedOps(constants.OP_STATUS_CANCELED, |
683 | 39ed3a98 | Guido Trotter | "Job canceled by request")
|
684 | fbf0262f | Michael Hanselmann | finally:
|
685 | fbf0262f | Michael Hanselmann | queue.release() |
686 | 85f03e0d | Michael Hanselmann | except errors.GenericError, err:
|
687 | 85f03e0d | Michael Hanselmann | logging.exception("Ganeti exception")
|
688 | 85f03e0d | Michael Hanselmann | except:
|
689 | 85f03e0d | Michael Hanselmann | logging.exception("Unhandled exception")
|
690 | e2715f69 | Michael Hanselmann | finally:
|
691 | 3c0d60d0 | Guido Trotter | queue.acquire(shared=1)
|
692 | 85f03e0d | Michael Hanselmann | try:
|
693 | 65548ed5 | Michael Hanselmann | try:
|
694 | ef2df7d3 | Michael Hanselmann | job.lock_status = None
|
695 | c56ec146 | Iustin Pop | job.end_timestamp = TimeStampNow() |
696 | 65548ed5 | Michael Hanselmann | queue.UpdateJobUnlocked(job) |
697 | 65548ed5 | Michael Hanselmann | finally:
|
698 | 65548ed5 | Michael Hanselmann | job_id = job.id |
699 | 65548ed5 | Michael Hanselmann | status = job.CalcStatus() |
700 | 85f03e0d | Michael Hanselmann | finally:
|
701 | 85f03e0d | Michael Hanselmann | queue.release() |
702 | ef2df7d3 | Michael Hanselmann | |
703 | 02fc74da | Michael Hanselmann | logging.info("Finished job %s, status = %s", job_id, status)
|
704 | e2715f69 | Michael Hanselmann | |
705 | e2715f69 | Michael Hanselmann | |
706 | e2715f69 | Michael Hanselmann | class _JobQueueWorkerPool(workerpool.WorkerPool): |
707 | ea03467c | Iustin Pop | """Simple class implementing a job-processing workerpool.
|
708 | ea03467c | Iustin Pop |
|
709 | ea03467c | Iustin Pop | """
|
710 | 5bdce580 | Michael Hanselmann | def __init__(self, queue): |
711 | 89e2b4d2 | Michael Hanselmann | super(_JobQueueWorkerPool, self).__init__("JobQueue", |
712 | 89e2b4d2 | Michael Hanselmann | JOBQUEUE_THREADS, |
713 | e2715f69 | Michael Hanselmann | _JobQueueWorker) |
714 | 5bdce580 | Michael Hanselmann | self.queue = queue
|
715 | e2715f69 | Michael Hanselmann | |
716 | e2715f69 | Michael Hanselmann | |
717 | 6c881c52 | Iustin Pop | def _RequireOpenQueue(fn): |
718 | 6c881c52 | Iustin Pop | """Decorator for "public" functions.
|
719 | ea03467c | Iustin Pop |
|
720 | 6c881c52 | Iustin Pop | This function should be used for all 'public' functions. That is,
|
721 | 6c881c52 | Iustin Pop | functions usually called from other classes. Note that this should
|
722 | 6c881c52 | Iustin Pop | be applied only to methods (not plain functions), since it expects
|
723 | 6c881c52 | Iustin Pop | that the decorated function is called with a first argument that has
|
724 | a71f9c7d | Guido Trotter | a '_queue_filelock' argument.
|
725 | ea03467c | Iustin Pop |
|
726 | 99bd4f0a | Guido Trotter | @warning: Use this decorator only after locking.ssynchronized
|
727 | f1da30e6 | Michael Hanselmann |
|
728 | 6c881c52 | Iustin Pop | Example::
|
729 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
730 | 6c881c52 | Iustin Pop | @_RequireOpenQueue
|
731 | 6c881c52 | Iustin Pop | def Example(self):
|
732 | 6c881c52 | Iustin Pop | pass
|
733 | db37da70 | Michael Hanselmann |
|
734 | 6c881c52 | Iustin Pop | """
|
735 | 6c881c52 | Iustin Pop | def wrapper(self, *args, **kwargs): |
736 | 7260cfbe | Iustin Pop | # pylint: disable-msg=W0212
|
737 | a71f9c7d | Guido Trotter | assert self._queue_filelock is not None, "Queue should be open" |
738 | 6c881c52 | Iustin Pop | return fn(self, *args, **kwargs) |
739 | 6c881c52 | Iustin Pop | return wrapper
|
740 | db37da70 | Michael Hanselmann | |
741 | db37da70 | Michael Hanselmann | |
742 | 6c881c52 | Iustin Pop | class JobQueue(object): |
743 | 6c881c52 | Iustin Pop | """Queue used to manage the jobs.
|
744 | db37da70 | Michael Hanselmann |
|
745 | 6c881c52 | Iustin Pop | @cvar _RE_JOB_FILE: regex matching the valid job file names
|
746 | 6c881c52 | Iustin Pop |
|
747 | 6c881c52 | Iustin Pop | """
|
748 | 6c881c52 | Iustin Pop | _RE_JOB_FILE = re.compile(r"^job-(%s)$" % constants.JOB_ID_TEMPLATE)
|
749 | db37da70 | Michael Hanselmann | |
750 | 85f03e0d | Michael Hanselmann | def __init__(self, context): |
751 | ea03467c | Iustin Pop | """Constructor for JobQueue.
|
752 | ea03467c | Iustin Pop |
|
753 | ea03467c | Iustin Pop | The constructor will initialize the job queue object and then
|
754 | ea03467c | Iustin Pop | start loading the current jobs from disk, either for starting them
|
755 | ea03467c | Iustin Pop | (if they were queue) or for aborting them (if they were already
|
756 | ea03467c | Iustin Pop | running).
|
757 | ea03467c | Iustin Pop |
|
758 | ea03467c | Iustin Pop | @type context: GanetiContext
|
759 | ea03467c | Iustin Pop | @param context: the context object for access to the configuration
|
760 | ea03467c | Iustin Pop | data and other ganeti objects
|
761 | ea03467c | Iustin Pop |
|
762 | ea03467c | Iustin Pop | """
|
763 | 5bdce580 | Michael Hanselmann | self.context = context
|
764 | 5685c1a5 | Michael Hanselmann | self._memcache = weakref.WeakValueDictionary()
|
765 | c3f0a12f | Iustin Pop | self._my_hostname = utils.HostInfo().name
|
766 | f1da30e6 | Michael Hanselmann | |
767 | ebb80afa | Guido Trotter | # The Big JobQueue lock. If a code block or method acquires it in shared
|
768 | ebb80afa | Guido Trotter | # mode safe it must guarantee concurrency with all the code acquiring it in
|
769 | ebb80afa | Guido Trotter | # shared mode, including itself. In order not to acquire it at all
|
770 | ebb80afa | Guido Trotter | # concurrency must be guaranteed with all code acquiring it in shared mode
|
771 | ebb80afa | Guido Trotter | # and all code acquiring it exclusively.
|
772 | ebb80afa | Guido Trotter | self._lock = locking.SharedLock()
|
773 | ebb80afa | Guido Trotter | |
774 | ebb80afa | Guido Trotter | self.acquire = self._lock.acquire |
775 | ebb80afa | Guido Trotter | self.release = self._lock.release |
776 | 85f03e0d | Michael Hanselmann | |
777 | a71f9c7d | Guido Trotter | # Initialize the queue, and acquire the filelock.
|
778 | a71f9c7d | Guido Trotter | # This ensures no other process is working on the job queue.
|
779 | a71f9c7d | Guido Trotter | self._queue_filelock = jstore.InitAndVerifyQueue(must_lock=True) |
780 | f1da30e6 | Michael Hanselmann | |
781 | 04ab05ce | Michael Hanselmann | # Read serial file
|
782 | 04ab05ce | Michael Hanselmann | self._last_serial = jstore.ReadSerial()
|
783 | 04ab05ce | Michael Hanselmann | assert self._last_serial is not None, ("Serial file was modified between" |
784 | 04ab05ce | Michael Hanselmann | " check in jstore and here")
|
785 | c4beba1c | Iustin Pop | |
786 | 23752136 | Michael Hanselmann | # Get initial list of nodes
|
787 | 99aabbed | Iustin Pop | self._nodes = dict((n.name, n.primary_ip) |
788 | 59303563 | Iustin Pop | for n in self.context.cfg.GetAllNodesInfo().values() |
789 | 59303563 | Iustin Pop | if n.master_candidate)
|
790 | 8e00939c | Michael Hanselmann | |
791 | 8e00939c | Michael Hanselmann | # Remove master node
|
792 | d8e0dc17 | Guido Trotter | self._nodes.pop(self._my_hostname, None) |
793 | 23752136 | Michael Hanselmann | |
794 | 23752136 | Michael Hanselmann | # TODO: Check consistency across nodes
|
795 | 23752136 | Michael Hanselmann | |
796 | 20571a26 | Guido Trotter | self._queue_size = 0 |
797 | 20571a26 | Guido Trotter | self._UpdateQueueSizeUnlocked()
|
798 | 20571a26 | Guido Trotter | self._drained = self._IsQueueMarkedDrain() |
799 | 20571a26 | Guido Trotter | |
800 | 85f03e0d | Michael Hanselmann | # Setup worker pool
|
801 | 5bdce580 | Michael Hanselmann | self._wpool = _JobQueueWorkerPool(self) |
802 | 85f03e0d | Michael Hanselmann | try:
|
803 | 16714921 | Michael Hanselmann | # We need to lock here because WorkerPool.AddTask() may start a job while
|
804 | 16714921 | Michael Hanselmann | # we're still doing our work.
|
805 | 16714921 | Michael Hanselmann | self.acquire()
|
806 | 16714921 | Michael Hanselmann | try:
|
807 | 711b5124 | Michael Hanselmann | logging.info("Inspecting job queue")
|
808 | 711b5124 | Michael Hanselmann | |
809 | 711b5124 | Michael Hanselmann | all_job_ids = self._GetJobIDsUnlocked()
|
810 | b7cb9024 | Michael Hanselmann | jobs_count = len(all_job_ids)
|
811 | 711b5124 | Michael Hanselmann | lastinfo = time.time() |
812 | 711b5124 | Michael Hanselmann | for idx, job_id in enumerate(all_job_ids): |
813 | 711b5124 | Michael Hanselmann | # Give an update every 1000 jobs or 10 seconds
|
814 | b7cb9024 | Michael Hanselmann | if (idx % 1000 == 0 or time.time() >= (lastinfo + 10.0) or |
815 | b7cb9024 | Michael Hanselmann | idx == (jobs_count - 1)):
|
816 | 711b5124 | Michael Hanselmann | logging.info("Job queue inspection: %d/%d (%0.1f %%)",
|
817 | b7cb9024 | Michael Hanselmann | idx, jobs_count - 1, 100.0 * (idx + 1) / jobs_count) |
818 | 711b5124 | Michael Hanselmann | lastinfo = time.time() |
819 | 711b5124 | Michael Hanselmann | |
820 | 711b5124 | Michael Hanselmann | job = self._LoadJobUnlocked(job_id)
|
821 | 711b5124 | Michael Hanselmann | |
822 | 16714921 | Michael Hanselmann | # a failure in loading the job can cause 'None' to be returned
|
823 | 16714921 | Michael Hanselmann | if job is None: |
824 | 16714921 | Michael Hanselmann | continue
|
825 | 94ed59a5 | Iustin Pop | |
826 | 16714921 | Michael Hanselmann | status = job.CalcStatus() |
827 | 85f03e0d | Michael Hanselmann | |
828 | 16714921 | Michael Hanselmann | if status in (constants.JOB_STATUS_QUEUED, ): |
829 | 16714921 | Michael Hanselmann | self._wpool.AddTask(job)
|
830 | 85f03e0d | Michael Hanselmann | |
831 | 16714921 | Michael Hanselmann | elif status in (constants.JOB_STATUS_RUNNING, |
832 | fbf0262f | Michael Hanselmann | constants.JOB_STATUS_WAITLOCK, |
833 | fbf0262f | Michael Hanselmann | constants.JOB_STATUS_CANCELING): |
834 | 16714921 | Michael Hanselmann | logging.warning("Unfinished job %s found: %s", job.id, job)
|
835 | 39ed3a98 | Guido Trotter | job.MarkUnfinishedOps(constants.OP_STATUS_ERROR, |
836 | 39ed3a98 | Guido Trotter | "Unclean master daemon shutdown")
|
837 | 711b5124 | Michael Hanselmann | |
838 | 711b5124 | Michael Hanselmann | logging.info("Job queue inspection finished")
|
839 | 16714921 | Michael Hanselmann | finally:
|
840 | 16714921 | Michael Hanselmann | self.release()
|
841 | 16714921 | Michael Hanselmann | except:
|
842 | 16714921 | Michael Hanselmann | self._wpool.TerminateWorkers()
|
843 | 16714921 | Michael Hanselmann | raise
|
844 | 85f03e0d | Michael Hanselmann | |
845 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
846 | d2e03a33 | Michael Hanselmann | @_RequireOpenQueue
|
847 | 99aabbed | Iustin Pop | def AddNode(self, node): |
848 | 99aabbed | Iustin Pop | """Register a new node with the queue.
|
849 | 99aabbed | Iustin Pop |
|
850 | 99aabbed | Iustin Pop | @type node: L{objects.Node}
|
851 | 99aabbed | Iustin Pop | @param node: the node object to be added
|
852 | 99aabbed | Iustin Pop |
|
853 | 99aabbed | Iustin Pop | """
|
854 | 99aabbed | Iustin Pop | node_name = node.name |
855 | d2e03a33 | Michael Hanselmann | assert node_name != self._my_hostname |
856 | 23752136 | Michael Hanselmann | |
857 | 9f774ee8 | Michael Hanselmann | # Clean queue directory on added node
|
858 | c8457ce7 | Iustin Pop | result = rpc.RpcRunner.call_jobqueue_purge(node_name) |
859 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
860 | c8457ce7 | Iustin Pop | if msg:
|
861 | c8457ce7 | Iustin Pop | logging.warning("Cannot cleanup queue directory on node %s: %s",
|
862 | c8457ce7 | Iustin Pop | node_name, msg) |
863 | 23752136 | Michael Hanselmann | |
864 | 59303563 | Iustin Pop | if not node.master_candidate: |
865 | 59303563 | Iustin Pop | # remove if existing, ignoring errors
|
866 | 59303563 | Iustin Pop | self._nodes.pop(node_name, None) |
867 | 59303563 | Iustin Pop | # and skip the replication of the job ids
|
868 | 59303563 | Iustin Pop | return
|
869 | 59303563 | Iustin Pop | |
870 | d2e03a33 | Michael Hanselmann | # Upload the whole queue excluding archived jobs
|
871 | d2e03a33 | Michael Hanselmann | files = [self._GetJobPath(job_id) for job_id in self._GetJobIDsUnlocked()] |
872 | 23752136 | Michael Hanselmann | |
873 | d2e03a33 | Michael Hanselmann | # Upload current serial file
|
874 | d2e03a33 | Michael Hanselmann | files.append(constants.JOB_QUEUE_SERIAL_FILE) |
875 | d2e03a33 | Michael Hanselmann | |
876 | d2e03a33 | Michael Hanselmann | for file_name in files: |
877 | 9f774ee8 | Michael Hanselmann | # Read file content
|
878 | 13998ef2 | Michael Hanselmann | content = utils.ReadFile(file_name) |
879 | 9f774ee8 | Michael Hanselmann | |
880 | a3811745 | Michael Hanselmann | result = rpc.RpcRunner.call_jobqueue_update([node_name], |
881 | a3811745 | Michael Hanselmann | [node.primary_ip], |
882 | a3811745 | Michael Hanselmann | file_name, content) |
883 | 3cebe102 | Michael Hanselmann | msg = result[node_name].fail_msg |
884 | c8457ce7 | Iustin Pop | if msg:
|
885 | c8457ce7 | Iustin Pop | logging.error("Failed to upload file %s to node %s: %s",
|
886 | c8457ce7 | Iustin Pop | file_name, node_name, msg) |
887 | d2e03a33 | Michael Hanselmann | |
888 | 99aabbed | Iustin Pop | self._nodes[node_name] = node.primary_ip
|
889 | d2e03a33 | Michael Hanselmann | |
890 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
891 | d2e03a33 | Michael Hanselmann | @_RequireOpenQueue
|
892 | d2e03a33 | Michael Hanselmann | def RemoveNode(self, node_name): |
893 | ea03467c | Iustin Pop | """Callback called when removing nodes from the cluster.
|
894 | ea03467c | Iustin Pop |
|
895 | ea03467c | Iustin Pop | @type node_name: str
|
896 | ea03467c | Iustin Pop | @param node_name: the name of the node to remove
|
897 | ea03467c | Iustin Pop |
|
898 | ea03467c | Iustin Pop | """
|
899 | d8e0dc17 | Guido Trotter | self._nodes.pop(node_name, None) |
900 | 23752136 | Michael Hanselmann | |
901 | 7e950d31 | Iustin Pop | @staticmethod
|
902 | 7e950d31 | Iustin Pop | def _CheckRpcResult(result, nodes, failmsg): |
903 | ea03467c | Iustin Pop | """Verifies the status of an RPC call.
|
904 | ea03467c | Iustin Pop |
|
905 | ea03467c | Iustin Pop | Since we aim to keep consistency should this node (the current
|
906 | ea03467c | Iustin Pop | master) fail, we will log errors if our rpc fail, and especially
|
907 | 5bbd3f7f | Michael Hanselmann | log the case when more than half of the nodes fails.
|
908 | ea03467c | Iustin Pop |
|
909 | ea03467c | Iustin Pop | @param result: the data as returned from the rpc call
|
910 | ea03467c | Iustin Pop | @type nodes: list
|
911 | ea03467c | Iustin Pop | @param nodes: the list of nodes we made the call to
|
912 | ea03467c | Iustin Pop | @type failmsg: str
|
913 | ea03467c | Iustin Pop | @param failmsg: the identifier to be used for logging
|
914 | ea03467c | Iustin Pop |
|
915 | ea03467c | Iustin Pop | """
|
916 | e74798c1 | Michael Hanselmann | failed = [] |
917 | e74798c1 | Michael Hanselmann | success = [] |
918 | e74798c1 | Michael Hanselmann | |
919 | e74798c1 | Michael Hanselmann | for node in nodes: |
920 | 3cebe102 | Michael Hanselmann | msg = result[node].fail_msg |
921 | c8457ce7 | Iustin Pop | if msg:
|
922 | e74798c1 | Michael Hanselmann | failed.append(node) |
923 | 45e0d704 | Iustin Pop | logging.error("RPC call %s (%s) failed on node %s: %s",
|
924 | 45e0d704 | Iustin Pop | result[node].call, failmsg, node, msg) |
925 | c8457ce7 | Iustin Pop | else:
|
926 | c8457ce7 | Iustin Pop | success.append(node) |
927 | e74798c1 | Michael Hanselmann | |
928 | e74798c1 | Michael Hanselmann | # +1 for the master node
|
929 | e74798c1 | Michael Hanselmann | if (len(success) + 1) < len(failed): |
930 | e74798c1 | Michael Hanselmann | # TODO: Handle failing nodes
|
931 | e74798c1 | Michael Hanselmann | logging.error("More than half of the nodes failed")
|
932 | e74798c1 | Michael Hanselmann | |
933 | 99aabbed | Iustin Pop | def _GetNodeIp(self): |
934 | 99aabbed | Iustin Pop | """Helper for returning the node name/ip list.
|
935 | 99aabbed | Iustin Pop |
|
936 | ea03467c | Iustin Pop | @rtype: (list, list)
|
937 | ea03467c | Iustin Pop | @return: a tuple of two lists, the first one with the node
|
938 | ea03467c | Iustin Pop | names and the second one with the node addresses
|
939 | ea03467c | Iustin Pop |
|
940 | 99aabbed | Iustin Pop | """
|
941 | 99aabbed | Iustin Pop | name_list = self._nodes.keys()
|
942 | 99aabbed | Iustin Pop | addr_list = [self._nodes[name] for name in name_list] |
943 | 99aabbed | Iustin Pop | return name_list, addr_list
|
944 | 99aabbed | Iustin Pop | |
945 | 4c36bdf5 | Guido Trotter | def _UpdateJobQueueFile(self, file_name, data, replicate): |
946 | 8e00939c | Michael Hanselmann | """Writes a file locally and then replicates it to all nodes.
|
947 | 8e00939c | Michael Hanselmann |
|
948 | ea03467c | Iustin Pop | This function will replace the contents of a file on the local
|
949 | ea03467c | Iustin Pop | node and then replicate it to all the other nodes we have.
|
950 | ea03467c | Iustin Pop |
|
951 | ea03467c | Iustin Pop | @type file_name: str
|
952 | ea03467c | Iustin Pop | @param file_name: the path of the file to be replicated
|
953 | ea03467c | Iustin Pop | @type data: str
|
954 | ea03467c | Iustin Pop | @param data: the new contents of the file
|
955 | 4c36bdf5 | Guido Trotter | @type replicate: boolean
|
956 | 4c36bdf5 | Guido Trotter | @param replicate: whether to spread the changes to the remote nodes
|
957 | ea03467c | Iustin Pop |
|
958 | 8e00939c | Michael Hanselmann | """
|
959 | 8e00939c | Michael Hanselmann | utils.WriteFile(file_name, data=data) |
960 | 8e00939c | Michael Hanselmann | |
961 | 4c36bdf5 | Guido Trotter | if replicate:
|
962 | 4c36bdf5 | Guido Trotter | names, addrs = self._GetNodeIp()
|
963 | 4c36bdf5 | Guido Trotter | result = rpc.RpcRunner.call_jobqueue_update(names, addrs, file_name, data) |
964 | 4c36bdf5 | Guido Trotter | self._CheckRpcResult(result, self._nodes, "Updating %s" % file_name) |
965 | 23752136 | Michael Hanselmann | |
966 | d7fd1f28 | Michael Hanselmann | def _RenameFilesUnlocked(self, rename): |
967 | ea03467c | Iustin Pop | """Renames a file locally and then replicate the change.
|
968 | ea03467c | Iustin Pop |
|
969 | ea03467c | Iustin Pop | This function will rename a file in the local queue directory
|
970 | ea03467c | Iustin Pop | and then replicate this rename to all the other nodes we have.
|
971 | ea03467c | Iustin Pop |
|
972 | d7fd1f28 | Michael Hanselmann | @type rename: list of (old, new)
|
973 | d7fd1f28 | Michael Hanselmann | @param rename: List containing tuples mapping old to new names
|
974 | ea03467c | Iustin Pop |
|
975 | ea03467c | Iustin Pop | """
|
976 | dd875d32 | Michael Hanselmann | # Rename them locally
|
977 | d7fd1f28 | Michael Hanselmann | for old, new in rename: |
978 | d7fd1f28 | Michael Hanselmann | utils.RenameFile(old, new, mkdir=True)
|
979 | abc1f2ce | Michael Hanselmann | |
980 | dd875d32 | Michael Hanselmann | # ... and on all nodes
|
981 | dd875d32 | Michael Hanselmann | names, addrs = self._GetNodeIp()
|
982 | dd875d32 | Michael Hanselmann | result = rpc.RpcRunner.call_jobqueue_rename(names, addrs, rename) |
983 | dd875d32 | Michael Hanselmann | self._CheckRpcResult(result, self._nodes, "Renaming files (%r)" % rename) |
984 | abc1f2ce | Michael Hanselmann | |
985 | 7e950d31 | Iustin Pop | @staticmethod
|
986 | 7e950d31 | Iustin Pop | def _FormatJobID(job_id): |
987 | ea03467c | Iustin Pop | """Convert a job ID to string format.
|
988 | ea03467c | Iustin Pop |
|
989 | ea03467c | Iustin Pop | Currently this just does C{str(job_id)} after performing some
|
990 | ea03467c | Iustin Pop | checks, but if we want to change the job id format this will
|
991 | ea03467c | Iustin Pop | abstract this change.
|
992 | ea03467c | Iustin Pop |
|
993 | ea03467c | Iustin Pop | @type job_id: int or long
|
994 | ea03467c | Iustin Pop | @param job_id: the numeric job id
|
995 | ea03467c | Iustin Pop | @rtype: str
|
996 | ea03467c | Iustin Pop | @return: the formatted job id
|
997 | ea03467c | Iustin Pop |
|
998 | ea03467c | Iustin Pop | """
|
999 | 85f03e0d | Michael Hanselmann | if not isinstance(job_id, (int, long)): |
1000 | 85f03e0d | Michael Hanselmann | raise errors.ProgrammerError("Job ID '%s' not numeric" % job_id) |
1001 | 85f03e0d | Michael Hanselmann | if job_id < 0: |
1002 | 85f03e0d | Michael Hanselmann | raise errors.ProgrammerError("Job ID %s is negative" % job_id) |
1003 | 85f03e0d | Michael Hanselmann | |
1004 | 85f03e0d | Michael Hanselmann | return str(job_id) |
1005 | 85f03e0d | Michael Hanselmann | |
1006 | 58b22b6e | Michael Hanselmann | @classmethod
|
1007 | 58b22b6e | Michael Hanselmann | def _GetArchiveDirectory(cls, job_id): |
1008 | 58b22b6e | Michael Hanselmann | """Returns the archive directory for a job.
|
1009 | 58b22b6e | Michael Hanselmann |
|
1010 | 58b22b6e | Michael Hanselmann | @type job_id: str
|
1011 | 58b22b6e | Michael Hanselmann | @param job_id: Job identifier
|
1012 | 58b22b6e | Michael Hanselmann | @rtype: str
|
1013 | 58b22b6e | Michael Hanselmann | @return: Directory name
|
1014 | 58b22b6e | Michael Hanselmann |
|
1015 | 58b22b6e | Michael Hanselmann | """
|
1016 | 58b22b6e | Michael Hanselmann | return str(int(job_id) / JOBS_PER_ARCHIVE_DIRECTORY) |
1017 | 58b22b6e | Michael Hanselmann | |
1018 | 009e73d0 | Iustin Pop | def _NewSerialsUnlocked(self, count): |
1019 | f1da30e6 | Michael Hanselmann | """Generates a new job identifier.
|
1020 | f1da30e6 | Michael Hanselmann |
|
1021 | f1da30e6 | Michael Hanselmann | Job identifiers are unique during the lifetime of a cluster.
|
1022 | f1da30e6 | Michael Hanselmann |
|
1023 | 009e73d0 | Iustin Pop | @type count: integer
|
1024 | 009e73d0 | Iustin Pop | @param count: how many serials to return
|
1025 | ea03467c | Iustin Pop | @rtype: str
|
1026 | ea03467c | Iustin Pop | @return: a string representing the job identifier.
|
1027 | f1da30e6 | Michael Hanselmann |
|
1028 | f1da30e6 | Michael Hanselmann | """
|
1029 | 009e73d0 | Iustin Pop | assert count > 0 |
1030 | f1da30e6 | Michael Hanselmann | # New number
|
1031 | 009e73d0 | Iustin Pop | serial = self._last_serial + count
|
1032 | f1da30e6 | Michael Hanselmann | |
1033 | f1da30e6 | Michael Hanselmann | # Write to file
|
1034 | 4c36bdf5 | Guido Trotter | self._UpdateJobQueueFile(constants.JOB_QUEUE_SERIAL_FILE,
|
1035 | 4c36bdf5 | Guido Trotter | "%s\n" % serial, True) |
1036 | f1da30e6 | Michael Hanselmann | |
1037 | 009e73d0 | Iustin Pop | result = [self._FormatJobID(v)
|
1038 | 009e73d0 | Iustin Pop | for v in range(self._last_serial, serial + 1)] |
1039 | f1da30e6 | Michael Hanselmann | # Keep it only if we were able to write the file
|
1040 | f1da30e6 | Michael Hanselmann | self._last_serial = serial
|
1041 | f1da30e6 | Michael Hanselmann | |
1042 | 009e73d0 | Iustin Pop | return result
|
1043 | f1da30e6 | Michael Hanselmann | |
1044 | 85f03e0d | Michael Hanselmann | @staticmethod
|
1045 | 85f03e0d | Michael Hanselmann | def _GetJobPath(job_id): |
1046 | ea03467c | Iustin Pop | """Returns the job file for a given job id.
|
1047 | ea03467c | Iustin Pop |
|
1048 | ea03467c | Iustin Pop | @type job_id: str
|
1049 | ea03467c | Iustin Pop | @param job_id: the job identifier
|
1050 | ea03467c | Iustin Pop | @rtype: str
|
1051 | ea03467c | Iustin Pop | @return: the path to the job file
|
1052 | ea03467c | Iustin Pop |
|
1053 | ea03467c | Iustin Pop | """
|
1054 | c4feafe8 | Iustin Pop | return utils.PathJoin(constants.QUEUE_DIR, "job-%s" % job_id) |
1055 | f1da30e6 | Michael Hanselmann | |
1056 | 58b22b6e | Michael Hanselmann | @classmethod
|
1057 | 58b22b6e | Michael Hanselmann | def _GetArchivedJobPath(cls, job_id): |
1058 | ea03467c | Iustin Pop | """Returns the archived job file for a give job id.
|
1059 | ea03467c | Iustin Pop |
|
1060 | ea03467c | Iustin Pop | @type job_id: str
|
1061 | ea03467c | Iustin Pop | @param job_id: the job identifier
|
1062 | ea03467c | Iustin Pop | @rtype: str
|
1063 | ea03467c | Iustin Pop | @return: the path to the archived job file
|
1064 | ea03467c | Iustin Pop |
|
1065 | ea03467c | Iustin Pop | """
|
1066 | 0411c011 | Iustin Pop | return utils.PathJoin(constants.JOB_QUEUE_ARCHIVE_DIR,
|
1067 | 0411c011 | Iustin Pop | cls._GetArchiveDirectory(job_id), "job-%s" % job_id)
|
1068 | 0cb94105 | Michael Hanselmann | |
1069 | 85a1c57d | Guido Trotter | def _GetJobIDsUnlocked(self, sort=True): |
1070 | 911a495b | Iustin Pop | """Return all known job IDs.
|
1071 | 911a495b | Iustin Pop |
|
1072 | ac0930b9 | Iustin Pop | The method only looks at disk because it's a requirement that all
|
1073 | ac0930b9 | Iustin Pop | jobs are present on disk (so in the _memcache we don't have any
|
1074 | ac0930b9 | Iustin Pop | extra IDs).
|
1075 | ac0930b9 | Iustin Pop |
|
1076 | 85a1c57d | Guido Trotter | @type sort: boolean
|
1077 | 85a1c57d | Guido Trotter | @param sort: perform sorting on the returned job ids
|
1078 | ea03467c | Iustin Pop | @rtype: list
|
1079 | ea03467c | Iustin Pop | @return: the list of job IDs
|
1080 | ea03467c | Iustin Pop |
|
1081 | 911a495b | Iustin Pop | """
|
1082 | 85a1c57d | Guido Trotter | jlist = [] |
1083 | b5b8309d | Guido Trotter | for filename in utils.ListVisibleFiles(constants.QUEUE_DIR): |
1084 | 85a1c57d | Guido Trotter | m = self._RE_JOB_FILE.match(filename)
|
1085 | 85a1c57d | Guido Trotter | if m:
|
1086 | 85a1c57d | Guido Trotter | jlist.append(m.group(1))
|
1087 | 85a1c57d | Guido Trotter | if sort:
|
1088 | 85a1c57d | Guido Trotter | jlist = utils.NiceSort(jlist) |
1089 | f0d874fe | Iustin Pop | return jlist
|
1090 | 911a495b | Iustin Pop | |
1091 | 911a495b | Iustin Pop | def _LoadJobUnlocked(self, job_id): |
1092 | ea03467c | Iustin Pop | """Loads a job from the disk or memory.
|
1093 | ea03467c | Iustin Pop |
|
1094 | ea03467c | Iustin Pop | Given a job id, this will return the cached job object if
|
1095 | ea03467c | Iustin Pop | existing, or try to load the job from the disk. If loading from
|
1096 | ea03467c | Iustin Pop | disk, it will also add the job to the cache.
|
1097 | ea03467c | Iustin Pop |
|
1098 | ea03467c | Iustin Pop | @param job_id: the job id
|
1099 | ea03467c | Iustin Pop | @rtype: L{_QueuedJob} or None
|
1100 | ea03467c | Iustin Pop | @return: either None or the job object
|
1101 | ea03467c | Iustin Pop |
|
1102 | ea03467c | Iustin Pop | """
|
1103 | 5685c1a5 | Michael Hanselmann | job = self._memcache.get(job_id, None) |
1104 | 5685c1a5 | Michael Hanselmann | if job:
|
1105 | 205d71fd | Michael Hanselmann | logging.debug("Found job %s in memcache", job_id)
|
1106 | 5685c1a5 | Michael Hanselmann | return job
|
1107 | ac0930b9 | Iustin Pop | |
1108 | 3d6c5566 | Guido Trotter | try:
|
1109 | 3d6c5566 | Guido Trotter | job = self._LoadJobFromDisk(job_id)
|
1110 | 3d6c5566 | Guido Trotter | except errors.JobFileCorrupted:
|
1111 | 3d6c5566 | Guido Trotter | old_path = self._GetJobPath(job_id)
|
1112 | 3d6c5566 | Guido Trotter | new_path = self._GetArchivedJobPath(job_id)
|
1113 | 3d6c5566 | Guido Trotter | if old_path == new_path:
|
1114 | 3d6c5566 | Guido Trotter | # job already archived (future case)
|
1115 | 3d6c5566 | Guido Trotter | logging.exception("Can't parse job %s", job_id)
|
1116 | 3d6c5566 | Guido Trotter | else:
|
1117 | 3d6c5566 | Guido Trotter | # non-archived case
|
1118 | 3d6c5566 | Guido Trotter | logging.exception("Can't parse job %s, will archive.", job_id)
|
1119 | 3d6c5566 | Guido Trotter | self._RenameFilesUnlocked([(old_path, new_path)])
|
1120 | 3d6c5566 | Guido Trotter | return None |
1121 | 162c8636 | Guido Trotter | |
1122 | 162c8636 | Guido Trotter | self._memcache[job_id] = job
|
1123 | 162c8636 | Guido Trotter | logging.debug("Added job %s to the cache", job_id)
|
1124 | 162c8636 | Guido Trotter | return job
|
1125 | 162c8636 | Guido Trotter | |
1126 | 162c8636 | Guido Trotter | def _LoadJobFromDisk(self, job_id): |
1127 | 162c8636 | Guido Trotter | """Load the given job file from disk.
|
1128 | 162c8636 | Guido Trotter |
|
1129 | 162c8636 | Guido Trotter | Given a job file, read, load and restore it in a _QueuedJob format.
|
1130 | 162c8636 | Guido Trotter |
|
1131 | 162c8636 | Guido Trotter | @type job_id: string
|
1132 | 162c8636 | Guido Trotter | @param job_id: job identifier
|
1133 | 162c8636 | Guido Trotter | @rtype: L{_QueuedJob} or None
|
1134 | 162c8636 | Guido Trotter | @return: either None or the job object
|
1135 | 162c8636 | Guido Trotter |
|
1136 | 162c8636 | Guido Trotter | """
|
1137 | 911a495b | Iustin Pop | filepath = self._GetJobPath(job_id)
|
1138 | f1da30e6 | Michael Hanselmann | logging.debug("Loading job from %s", filepath)
|
1139 | f1da30e6 | Michael Hanselmann | try:
|
1140 | 13998ef2 | Michael Hanselmann | raw_data = utils.ReadFile(filepath) |
1141 | 162c8636 | Guido Trotter | except EnvironmentError, err: |
1142 | f1da30e6 | Michael Hanselmann | if err.errno in (errno.ENOENT, ): |
1143 | f1da30e6 | Michael Hanselmann | return None |
1144 | f1da30e6 | Michael Hanselmann | raise
|
1145 | 13998ef2 | Michael Hanselmann | |
1146 | 94ed59a5 | Iustin Pop | try:
|
1147 | 162c8636 | Guido Trotter | data = serializer.LoadJson(raw_data) |
1148 | 94ed59a5 | Iustin Pop | job = _QueuedJob.Restore(self, data)
|
1149 | 7260cfbe | Iustin Pop | except Exception, err: # pylint: disable-msg=W0703 |
1150 | 3d6c5566 | Guido Trotter | raise errors.JobFileCorrupted(err)
|
1151 | 94ed59a5 | Iustin Pop | |
1152 | ac0930b9 | Iustin Pop | return job
|
1153 | f1da30e6 | Michael Hanselmann | |
1154 | 0f9c08dc | Guido Trotter | def SafeLoadJobFromDisk(self, job_id): |
1155 | 0f9c08dc | Guido Trotter | """Load the given job file from disk.
|
1156 | 0f9c08dc | Guido Trotter |
|
1157 | 0f9c08dc | Guido Trotter | Given a job file, read, load and restore it in a _QueuedJob format.
|
1158 | 0f9c08dc | Guido Trotter | In case of error reading the job, it gets returned as None, and the
|
1159 | 0f9c08dc | Guido Trotter | exception is logged.
|
1160 | 0f9c08dc | Guido Trotter |
|
1161 | 0f9c08dc | Guido Trotter | @type job_id: string
|
1162 | 0f9c08dc | Guido Trotter | @param job_id: job identifier
|
1163 | 0f9c08dc | Guido Trotter | @rtype: L{_QueuedJob} or None
|
1164 | 0f9c08dc | Guido Trotter | @return: either None or the job object
|
1165 | 0f9c08dc | Guido Trotter |
|
1166 | 0f9c08dc | Guido Trotter | """
|
1167 | 0f9c08dc | Guido Trotter | try:
|
1168 | 0f9c08dc | Guido Trotter | return self._LoadJobFromDisk(job_id) |
1169 | 0f9c08dc | Guido Trotter | except (errors.JobFileCorrupted, EnvironmentError): |
1170 | 0f9c08dc | Guido Trotter | logging.exception("Can't load/parse job %s", job_id)
|
1171 | 0f9c08dc | Guido Trotter | return None |
1172 | 0f9c08dc | Guido Trotter | |
1173 | 686d7433 | Iustin Pop | @staticmethod
|
1174 | 686d7433 | Iustin Pop | def _IsQueueMarkedDrain(): |
1175 | 686d7433 | Iustin Pop | """Check if the queue is marked from drain.
|
1176 | 686d7433 | Iustin Pop |
|
1177 | 686d7433 | Iustin Pop | This currently uses the queue drain file, which makes it a
|
1178 | 686d7433 | Iustin Pop | per-node flag. In the future this can be moved to the config file.
|
1179 | 686d7433 | Iustin Pop |
|
1180 | ea03467c | Iustin Pop | @rtype: boolean
|
1181 | ea03467c | Iustin Pop | @return: True of the job queue is marked for draining
|
1182 | ea03467c | Iustin Pop |
|
1183 | 686d7433 | Iustin Pop | """
|
1184 | 686d7433 | Iustin Pop | return os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
|
1185 | 686d7433 | Iustin Pop | |
1186 | 20571a26 | Guido Trotter | def _UpdateQueueSizeUnlocked(self): |
1187 | 20571a26 | Guido Trotter | """Update the queue size.
|
1188 | 20571a26 | Guido Trotter |
|
1189 | 20571a26 | Guido Trotter | """
|
1190 | 20571a26 | Guido Trotter | self._queue_size = len(self._GetJobIDsUnlocked(sort=False)) |
1191 | 20571a26 | Guido Trotter | |
1192 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1193 | 20571a26 | Guido Trotter | @_RequireOpenQueue
|
1194 | 20571a26 | Guido Trotter | def SetDrainFlag(self, drain_flag): |
1195 | 3ccafd0e | Iustin Pop | """Sets the drain flag for the queue.
|
1196 | 3ccafd0e | Iustin Pop |
|
1197 | ea03467c | Iustin Pop | @type drain_flag: boolean
|
1198 | 5bbd3f7f | Michael Hanselmann | @param drain_flag: Whether to set or unset the drain flag
|
1199 | ea03467c | Iustin Pop |
|
1200 | 3ccafd0e | Iustin Pop | """
|
1201 | 3ccafd0e | Iustin Pop | if drain_flag:
|
1202 | 3ccafd0e | Iustin Pop | utils.WriteFile(constants.JOB_QUEUE_DRAIN_FILE, data="", close=True) |
1203 | 3ccafd0e | Iustin Pop | else:
|
1204 | 3ccafd0e | Iustin Pop | utils.RemoveFile(constants.JOB_QUEUE_DRAIN_FILE) |
1205 | 20571a26 | Guido Trotter | |
1206 | 20571a26 | Guido Trotter | self._drained = drain_flag
|
1207 | 20571a26 | Guido Trotter | |
1208 | 3ccafd0e | Iustin Pop | return True |
1209 | 3ccafd0e | Iustin Pop | |
1210 | db37da70 | Michael Hanselmann | @_RequireOpenQueue
|
1211 | 009e73d0 | Iustin Pop | def _SubmitJobUnlocked(self, job_id, ops): |
1212 | 85f03e0d | Michael Hanselmann | """Create and store a new job.
|
1213 | f1da30e6 | Michael Hanselmann |
|
1214 | 85f03e0d | Michael Hanselmann | This enters the job into our job queue and also puts it on the new
|
1215 | 85f03e0d | Michael Hanselmann | queue, in order for it to be picked up by the queue processors.
|
1216 | c3f0a12f | Iustin Pop |
|
1217 | 009e73d0 | Iustin Pop | @type job_id: job ID
|
1218 | 69b99987 | Michael Hanselmann | @param job_id: the job ID for the new job
|
1219 | c3f0a12f | Iustin Pop | @type ops: list
|
1220 | 205d71fd | Michael Hanselmann | @param ops: The list of OpCodes that will become the new job.
|
1221 | 7beb1e53 | Guido Trotter | @rtype: L{_QueuedJob}
|
1222 | 7beb1e53 | Guido Trotter | @return: the job object to be queued
|
1223 | 7beb1e53 | Guido Trotter | @raise errors.JobQueueDrainError: if the job queue is marked for draining
|
1224 | 7beb1e53 | Guido Trotter | @raise errors.JobQueueFull: if the job queue has too many jobs in it
|
1225 | c3f0a12f | Iustin Pop |
|
1226 | c3f0a12f | Iustin Pop | """
|
1227 | 20571a26 | Guido Trotter | # Ok when sharing the big job queue lock, as the drain file is created when
|
1228 | 20571a26 | Guido Trotter | # the lock is exclusive.
|
1229 | 20571a26 | Guido Trotter | if self._drained: |
1230 | 2971c913 | Iustin Pop | raise errors.JobQueueDrainError("Job queue is drained, refusing job") |
1231 | f87b405e | Michael Hanselmann | |
1232 | 20571a26 | Guido Trotter | if self._queue_size >= constants.JOB_QUEUE_SIZE_HARD_LIMIT: |
1233 | f87b405e | Michael Hanselmann | raise errors.JobQueueFull()
|
1234 | f87b405e | Michael Hanselmann | |
1235 | f1da30e6 | Michael Hanselmann | job = _QueuedJob(self, job_id, ops)
|
1236 | f1da30e6 | Michael Hanselmann | |
1237 | f1da30e6 | Michael Hanselmann | # Write to disk
|
1238 | 85f03e0d | Michael Hanselmann | self.UpdateJobUnlocked(job)
|
1239 | f1da30e6 | Michael Hanselmann | |
1240 | 20571a26 | Guido Trotter | self._queue_size += 1 |
1241 | 20571a26 | Guido Trotter | |
1242 | 5685c1a5 | Michael Hanselmann | logging.debug("Adding new job %s to the cache", job_id)
|
1243 | ac0930b9 | Iustin Pop | self._memcache[job_id] = job
|
1244 | ac0930b9 | Iustin Pop | |
1245 | 7beb1e53 | Guido Trotter | return job
|
1246 | f1da30e6 | Michael Hanselmann | |
1247 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1248 | 2971c913 | Iustin Pop | @_RequireOpenQueue
|
1249 | 2971c913 | Iustin Pop | def SubmitJob(self, ops): |
1250 | 2971c913 | Iustin Pop | """Create and store a new job.
|
1251 | 2971c913 | Iustin Pop |
|
1252 | 2971c913 | Iustin Pop | @see: L{_SubmitJobUnlocked}
|
1253 | 2971c913 | Iustin Pop |
|
1254 | 2971c913 | Iustin Pop | """
|
1255 | 009e73d0 | Iustin Pop | job_id = self._NewSerialsUnlocked(1)[0] |
1256 | 7beb1e53 | Guido Trotter | self._wpool.AddTask(self._SubmitJobUnlocked(job_id, ops)) |
1257 | 7beb1e53 | Guido Trotter | return job_id
|
1258 | 2971c913 | Iustin Pop | |
1259 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1260 | 2971c913 | Iustin Pop | @_RequireOpenQueue
|
1261 | 2971c913 | Iustin Pop | def SubmitManyJobs(self, jobs): |
1262 | 2971c913 | Iustin Pop | """Create and store multiple jobs.
|
1263 | 2971c913 | Iustin Pop |
|
1264 | 2971c913 | Iustin Pop | @see: L{_SubmitJobUnlocked}
|
1265 | 2971c913 | Iustin Pop |
|
1266 | 2971c913 | Iustin Pop | """
|
1267 | 2971c913 | Iustin Pop | results = [] |
1268 | 7beb1e53 | Guido Trotter | tasks = [] |
1269 | 009e73d0 | Iustin Pop | all_job_ids = self._NewSerialsUnlocked(len(jobs)) |
1270 | 009e73d0 | Iustin Pop | for job_id, ops in zip(all_job_ids, jobs): |
1271 | 2971c913 | Iustin Pop | try:
|
1272 | 7beb1e53 | Guido Trotter | tasks.append((self._SubmitJobUnlocked(job_id, ops), ))
|
1273 | 2971c913 | Iustin Pop | status = True
|
1274 | 7beb1e53 | Guido Trotter | data = job_id |
1275 | 2971c913 | Iustin Pop | except errors.GenericError, err:
|
1276 | 2971c913 | Iustin Pop | data = str(err)
|
1277 | 2971c913 | Iustin Pop | status = False
|
1278 | 2971c913 | Iustin Pop | results.append((status, data)) |
1279 | 7beb1e53 | Guido Trotter | self._wpool.AddManyTasks(tasks)
|
1280 | 2971c913 | Iustin Pop | |
1281 | 2971c913 | Iustin Pop | return results
|
1282 | 2971c913 | Iustin Pop | |
1283 | db37da70 | Michael Hanselmann | @_RequireOpenQueue
|
1284 | 4c36bdf5 | Guido Trotter | def UpdateJobUnlocked(self, job, replicate=True): |
1285 | ea03467c | Iustin Pop | """Update a job's on disk storage.
|
1286 | ea03467c | Iustin Pop |
|
1287 | ea03467c | Iustin Pop | After a job has been modified, this function needs to be called in
|
1288 | ea03467c | Iustin Pop | order to write the changes to disk and replicate them to the other
|
1289 | ea03467c | Iustin Pop | nodes.
|
1290 | ea03467c | Iustin Pop |
|
1291 | ea03467c | Iustin Pop | @type job: L{_QueuedJob}
|
1292 | ea03467c | Iustin Pop | @param job: the changed job
|
1293 | 4c36bdf5 | Guido Trotter | @type replicate: boolean
|
1294 | 4c36bdf5 | Guido Trotter | @param replicate: whether to replicate the change to remote nodes
|
1295 | ea03467c | Iustin Pop |
|
1296 | ea03467c | Iustin Pop | """
|
1297 | f1da30e6 | Michael Hanselmann | filename = self._GetJobPath(job.id)
|
1298 | 23752136 | Michael Hanselmann | data = serializer.DumpJson(job.Serialize(), indent=False)
|
1299 | f1da30e6 | Michael Hanselmann | logging.debug("Writing job %s to %s", job.id, filename)
|
1300 | 4c36bdf5 | Guido Trotter | self._UpdateJobQueueFile(filename, data, replicate)
|
1301 | ac0930b9 | Iustin Pop | |
1302 | 5c735209 | Iustin Pop | def WaitForJobChanges(self, job_id, fields, prev_job_info, prev_log_serial, |
1303 | 5c735209 | Iustin Pop | timeout): |
1304 | 6c5a7090 | Michael Hanselmann | """Waits for changes in a job.
|
1305 | 6c5a7090 | Michael Hanselmann |
|
1306 | 6c5a7090 | Michael Hanselmann | @type job_id: string
|
1307 | 6c5a7090 | Michael Hanselmann | @param job_id: Job identifier
|
1308 | 6c5a7090 | Michael Hanselmann | @type fields: list of strings
|
1309 | 6c5a7090 | Michael Hanselmann | @param fields: Which fields to check for changes
|
1310 | 6c5a7090 | Michael Hanselmann | @type prev_job_info: list or None
|
1311 | 6c5a7090 | Michael Hanselmann | @param prev_job_info: Last job information returned
|
1312 | 6c5a7090 | Michael Hanselmann | @type prev_log_serial: int
|
1313 | 6c5a7090 | Michael Hanselmann | @param prev_log_serial: Last job message serial number
|
1314 | 5c735209 | Iustin Pop | @type timeout: float
|
1315 | 5c735209 | Iustin Pop | @param timeout: maximum time to wait
|
1316 | ea03467c | Iustin Pop | @rtype: tuple (job info, log entries)
|
1317 | ea03467c | Iustin Pop | @return: a tuple of the job information as required via
|
1318 | ea03467c | Iustin Pop | the fields parameter, and the log entries as a list
|
1319 | ea03467c | Iustin Pop |
|
1320 | ea03467c | Iustin Pop | if the job has not changed and the timeout has expired,
|
1321 | ea03467c | Iustin Pop | we instead return a special value,
|
1322 | ea03467c | Iustin Pop | L{constants.JOB_NOTCHANGED}, which should be interpreted
|
1323 | ea03467c | Iustin Pop | as such by the clients
|
1324 | 6c5a7090 | Michael Hanselmann |
|
1325 | 6c5a7090 | Michael Hanselmann | """
|
1326 | 6c2549d6 | Guido Trotter | helper = _WaitForJobChangesHelper(job_id, fields, prev_job_info, |
1327 | 6c2549d6 | Guido Trotter | prev_log_serial, self)
|
1328 | 6bcb1446 | Michael Hanselmann | try:
|
1329 | 6c2549d6 | Guido Trotter | return helper.WaitForChanges(timeout)
|
1330 | 6c2549d6 | Guido Trotter | finally:
|
1331 | 6c2549d6 | Guido Trotter | helper.Close() |
1332 | dfe57c22 | Michael Hanselmann | |
1333 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1334 | db37da70 | Michael Hanselmann | @_RequireOpenQueue
|
1335 | 188c5e0a | Michael Hanselmann | def CancelJob(self, job_id): |
1336 | 188c5e0a | Michael Hanselmann | """Cancels a job.
|
1337 | 188c5e0a | Michael Hanselmann |
|
1338 | ea03467c | Iustin Pop | This will only succeed if the job has not started yet.
|
1339 | ea03467c | Iustin Pop |
|
1340 | 188c5e0a | Michael Hanselmann | @type job_id: string
|
1341 | ea03467c | Iustin Pop | @param job_id: job ID of job to be cancelled.
|
1342 | 188c5e0a | Michael Hanselmann |
|
1343 | 188c5e0a | Michael Hanselmann | """
|
1344 | fbf0262f | Michael Hanselmann | logging.info("Cancelling job %s", job_id)
|
1345 | 188c5e0a | Michael Hanselmann | |
1346 | 85f03e0d | Michael Hanselmann | job = self._LoadJobUnlocked(job_id)
|
1347 | 188c5e0a | Michael Hanselmann | if not job: |
1348 | 188c5e0a | Michael Hanselmann | logging.debug("Job %s not found", job_id)
|
1349 | fbf0262f | Michael Hanselmann | return (False, "Job %s not found" % job_id) |
1350 | fbf0262f | Michael Hanselmann | |
1351 | fbf0262f | Michael Hanselmann | job_status = job.CalcStatus() |
1352 | 188c5e0a | Michael Hanselmann | |
1353 | fbf0262f | Michael Hanselmann | if job_status not in (constants.JOB_STATUS_QUEUED, |
1354 | fbf0262f | Michael Hanselmann | constants.JOB_STATUS_WAITLOCK): |
1355 | a9e97393 | Michael Hanselmann | logging.debug("Job %s is no longer waiting in the queue", job.id)
|
1356 | a9e97393 | Michael Hanselmann | return (False, "Job %s is no longer waiting in the queue" % job.id) |
1357 | fbf0262f | Michael Hanselmann | |
1358 | fbf0262f | Michael Hanselmann | if job_status == constants.JOB_STATUS_QUEUED:
|
1359 | 39ed3a98 | Guido Trotter | job.MarkUnfinishedOps(constants.OP_STATUS_CANCELED, |
1360 | 39ed3a98 | Guido Trotter | "Job canceled by request")
|
1361 | fbf0262f | Michael Hanselmann | return (True, "Job %s canceled" % job.id) |
1362 | 188c5e0a | Michael Hanselmann | |
1363 | fbf0262f | Michael Hanselmann | elif job_status == constants.JOB_STATUS_WAITLOCK:
|
1364 | fbf0262f | Michael Hanselmann | # The worker will notice the new status and cancel the job
|
1365 | 39ed3a98 | Guido Trotter | job.MarkUnfinishedOps(constants.OP_STATUS_CANCELING, None)
|
1366 | fbf0262f | Michael Hanselmann | return (True, "Job %s will be canceled" % job.id) |
1367 | fbf0262f | Michael Hanselmann | |
1368 | fbf0262f | Michael Hanselmann | @_RequireOpenQueue
|
1369 | d7fd1f28 | Michael Hanselmann | def _ArchiveJobsUnlocked(self, jobs): |
1370 | d7fd1f28 | Michael Hanselmann | """Archives jobs.
|
1371 | c609f802 | Michael Hanselmann |
|
1372 | d7fd1f28 | Michael Hanselmann | @type jobs: list of L{_QueuedJob}
|
1373 | 25e7b43f | Iustin Pop | @param jobs: Job objects
|
1374 | d7fd1f28 | Michael Hanselmann | @rtype: int
|
1375 | d7fd1f28 | Michael Hanselmann | @return: Number of archived jobs
|
1376 | c609f802 | Michael Hanselmann |
|
1377 | c609f802 | Michael Hanselmann | """
|
1378 | d7fd1f28 | Michael Hanselmann | archive_jobs = [] |
1379 | d7fd1f28 | Michael Hanselmann | rename_files = [] |
1380 | d7fd1f28 | Michael Hanselmann | for job in jobs: |
1381 | d7fd1f28 | Michael Hanselmann | if job.CalcStatus() not in (constants.JOB_STATUS_CANCELED, |
1382 | d7fd1f28 | Michael Hanselmann | constants.JOB_STATUS_SUCCESS, |
1383 | d7fd1f28 | Michael Hanselmann | constants.JOB_STATUS_ERROR): |
1384 | d7fd1f28 | Michael Hanselmann | logging.debug("Job %s is not yet done", job.id)
|
1385 | d7fd1f28 | Michael Hanselmann | continue
|
1386 | c609f802 | Michael Hanselmann | |
1387 | d7fd1f28 | Michael Hanselmann | archive_jobs.append(job) |
1388 | c609f802 | Michael Hanselmann | |
1389 | d7fd1f28 | Michael Hanselmann | old = self._GetJobPath(job.id)
|
1390 | d7fd1f28 | Michael Hanselmann | new = self._GetArchivedJobPath(job.id)
|
1391 | d7fd1f28 | Michael Hanselmann | rename_files.append((old, new)) |
1392 | c609f802 | Michael Hanselmann | |
1393 | d7fd1f28 | Michael Hanselmann | # TODO: What if 1..n files fail to rename?
|
1394 | d7fd1f28 | Michael Hanselmann | self._RenameFilesUnlocked(rename_files)
|
1395 | f1da30e6 | Michael Hanselmann | |
1396 | d7fd1f28 | Michael Hanselmann | logging.debug("Successfully archived job(s) %s",
|
1397 | 1f864b60 | Iustin Pop | utils.CommaJoin(job.id for job in archive_jobs)) |
1398 | d7fd1f28 | Michael Hanselmann | |
1399 | 20571a26 | Guido Trotter | # Since we haven't quite checked, above, if we succeeded or failed renaming
|
1400 | 20571a26 | Guido Trotter | # the files, we update the cached queue size from the filesystem. When we
|
1401 | 20571a26 | Guido Trotter | # get around to fix the TODO: above, we can use the number of actually
|
1402 | 20571a26 | Guido Trotter | # archived jobs to fix this.
|
1403 | 20571a26 | Guido Trotter | self._UpdateQueueSizeUnlocked()
|
1404 | d7fd1f28 | Michael Hanselmann | return len(archive_jobs) |
1405 | 78d12585 | Michael Hanselmann | |
1406 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1407 | 07cd723a | Iustin Pop | @_RequireOpenQueue
|
1408 | 07cd723a | Iustin Pop | def ArchiveJob(self, job_id): |
1409 | 07cd723a | Iustin Pop | """Archives a job.
|
1410 | 07cd723a | Iustin Pop |
|
1411 | 25e7b43f | Iustin Pop | This is just a wrapper over L{_ArchiveJobsUnlocked}.
|
1412 | ea03467c | Iustin Pop |
|
1413 | 07cd723a | Iustin Pop | @type job_id: string
|
1414 | 07cd723a | Iustin Pop | @param job_id: Job ID of job to be archived.
|
1415 | 78d12585 | Michael Hanselmann | @rtype: bool
|
1416 | 78d12585 | Michael Hanselmann | @return: Whether job was archived
|
1417 | 07cd723a | Iustin Pop |
|
1418 | 07cd723a | Iustin Pop | """
|
1419 | 78d12585 | Michael Hanselmann | logging.info("Archiving job %s", job_id)
|
1420 | 78d12585 | Michael Hanselmann | |
1421 | 78d12585 | Michael Hanselmann | job = self._LoadJobUnlocked(job_id)
|
1422 | 78d12585 | Michael Hanselmann | if not job: |
1423 | 78d12585 | Michael Hanselmann | logging.debug("Job %s not found", job_id)
|
1424 | 78d12585 | Michael Hanselmann | return False |
1425 | 78d12585 | Michael Hanselmann | |
1426 | 5278185a | Iustin Pop | return self._ArchiveJobsUnlocked([job]) == 1 |
1427 | 07cd723a | Iustin Pop | |
1428 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1429 | 07cd723a | Iustin Pop | @_RequireOpenQueue
|
1430 | f8ad5591 | Michael Hanselmann | def AutoArchiveJobs(self, age, timeout): |
1431 | 07cd723a | Iustin Pop | """Archives all jobs based on age.
|
1432 | 07cd723a | Iustin Pop |
|
1433 | 07cd723a | Iustin Pop | The method will archive all jobs which are older than the age
|
1434 | 07cd723a | Iustin Pop | parameter. For jobs that don't have an end timestamp, the start
|
1435 | 07cd723a | Iustin Pop | timestamp will be considered. The special '-1' age will cause
|
1436 | 07cd723a | Iustin Pop | archival of all jobs (that are not running or queued).
|
1437 | 07cd723a | Iustin Pop |
|
1438 | 07cd723a | Iustin Pop | @type age: int
|
1439 | 07cd723a | Iustin Pop | @param age: the minimum age in seconds
|
1440 | 07cd723a | Iustin Pop |
|
1441 | 07cd723a | Iustin Pop | """
|
1442 | 07cd723a | Iustin Pop | logging.info("Archiving jobs with age more than %s seconds", age)
|
1443 | 07cd723a | Iustin Pop | |
1444 | 07cd723a | Iustin Pop | now = time.time() |
1445 | f8ad5591 | Michael Hanselmann | end_time = now + timeout |
1446 | f8ad5591 | Michael Hanselmann | archived_count = 0
|
1447 | f8ad5591 | Michael Hanselmann | last_touched = 0
|
1448 | f8ad5591 | Michael Hanselmann | |
1449 | 69b03fd7 | Guido Trotter | all_job_ids = self._GetJobIDsUnlocked()
|
1450 | d7fd1f28 | Michael Hanselmann | pending = [] |
1451 | f8ad5591 | Michael Hanselmann | for idx, job_id in enumerate(all_job_ids): |
1452 | d2c8afb1 | Michael Hanselmann | last_touched = idx + 1
|
1453 | f8ad5591 | Michael Hanselmann | |
1454 | d7fd1f28 | Michael Hanselmann | # Not optimal because jobs could be pending
|
1455 | d7fd1f28 | Michael Hanselmann | # TODO: Measure average duration for job archival and take number of
|
1456 | d7fd1f28 | Michael Hanselmann | # pending jobs into account.
|
1457 | f8ad5591 | Michael Hanselmann | if time.time() > end_time:
|
1458 | f8ad5591 | Michael Hanselmann | break
|
1459 | f8ad5591 | Michael Hanselmann | |
1460 | 78d12585 | Michael Hanselmann | # Returns None if the job failed to load
|
1461 | 78d12585 | Michael Hanselmann | job = self._LoadJobUnlocked(job_id)
|
1462 | f8ad5591 | Michael Hanselmann | if job:
|
1463 | f8ad5591 | Michael Hanselmann | if job.end_timestamp is None: |
1464 | f8ad5591 | Michael Hanselmann | if job.start_timestamp is None: |
1465 | f8ad5591 | Michael Hanselmann | job_age = job.received_timestamp |
1466 | f8ad5591 | Michael Hanselmann | else:
|
1467 | f8ad5591 | Michael Hanselmann | job_age = job.start_timestamp |
1468 | 07cd723a | Iustin Pop | else:
|
1469 | f8ad5591 | Michael Hanselmann | job_age = job.end_timestamp |
1470 | f8ad5591 | Michael Hanselmann | |
1471 | f8ad5591 | Michael Hanselmann | if age == -1 or now - job_age[0] > age: |
1472 | d7fd1f28 | Michael Hanselmann | pending.append(job) |
1473 | d7fd1f28 | Michael Hanselmann | |
1474 | d7fd1f28 | Michael Hanselmann | # Archive 10 jobs at a time
|
1475 | d7fd1f28 | Michael Hanselmann | if len(pending) >= 10: |
1476 | d7fd1f28 | Michael Hanselmann | archived_count += self._ArchiveJobsUnlocked(pending)
|
1477 | d7fd1f28 | Michael Hanselmann | pending = [] |
1478 | f8ad5591 | Michael Hanselmann | |
1479 | d7fd1f28 | Michael Hanselmann | if pending:
|
1480 | d7fd1f28 | Michael Hanselmann | archived_count += self._ArchiveJobsUnlocked(pending)
|
1481 | 07cd723a | Iustin Pop | |
1482 | d2c8afb1 | Michael Hanselmann | return (archived_count, len(all_job_ids) - last_touched) |
1483 | 07cd723a | Iustin Pop | |
1484 | e2715f69 | Michael Hanselmann | def QueryJobs(self, job_ids, fields): |
1485 | e2715f69 | Michael Hanselmann | """Returns a list of jobs in queue.
|
1486 | e2715f69 | Michael Hanselmann |
|
1487 | ea03467c | Iustin Pop | @type job_ids: list
|
1488 | ea03467c | Iustin Pop | @param job_ids: sequence of job identifiers or None for all
|
1489 | ea03467c | Iustin Pop | @type fields: list
|
1490 | ea03467c | Iustin Pop | @param fields: names of fields to return
|
1491 | ea03467c | Iustin Pop | @rtype: list
|
1492 | ea03467c | Iustin Pop | @return: list one element per job, each element being list with
|
1493 | ea03467c | Iustin Pop | the requested fields
|
1494 | e2715f69 | Michael Hanselmann |
|
1495 | e2715f69 | Michael Hanselmann | """
|
1496 | 85f03e0d | Michael Hanselmann | jobs = [] |
1497 | 9f7b4967 | Guido Trotter | list_all = False
|
1498 | 9f7b4967 | Guido Trotter | if not job_ids: |
1499 | 9f7b4967 | Guido Trotter | # Since files are added to/removed from the queue atomically, there's no
|
1500 | 9f7b4967 | Guido Trotter | # risk of getting the job ids in an inconsistent state.
|
1501 | 9f7b4967 | Guido Trotter | job_ids = self._GetJobIDsUnlocked()
|
1502 | 9f7b4967 | Guido Trotter | list_all = True
|
1503 | e2715f69 | Michael Hanselmann | |
1504 | 9f7b4967 | Guido Trotter | for job_id in job_ids: |
1505 | 9f7b4967 | Guido Trotter | job = self.SafeLoadJobFromDisk(job_id)
|
1506 | 9f7b4967 | Guido Trotter | if job is not None: |
1507 | 6a290889 | Guido Trotter | jobs.append(job.GetInfo(fields)) |
1508 | 9f7b4967 | Guido Trotter | elif not list_all: |
1509 | 9f7b4967 | Guido Trotter | jobs.append(None)
|
1510 | e2715f69 | Michael Hanselmann | |
1511 | 85f03e0d | Michael Hanselmann | return jobs
|
1512 | e2715f69 | Michael Hanselmann | |
1513 | ebb80afa | Guido Trotter | @locking.ssynchronized(_LOCK)
|
1514 | db37da70 | Michael Hanselmann | @_RequireOpenQueue
|
1515 | e2715f69 | Michael Hanselmann | def Shutdown(self): |
1516 | e2715f69 | Michael Hanselmann | """Stops the job queue.
|
1517 | e2715f69 | Michael Hanselmann |
|
1518 | ea03467c | Iustin Pop | This shutdowns all the worker threads an closes the queue.
|
1519 | ea03467c | Iustin Pop |
|
1520 | e2715f69 | Michael Hanselmann | """
|
1521 | e2715f69 | Michael Hanselmann | self._wpool.TerminateWorkers()
|
1522 | 85f03e0d | Michael Hanselmann | |
1523 | a71f9c7d | Guido Trotter | self._queue_filelock.Close()
|
1524 | a71f9c7d | Guido Trotter | self._queue_filelock = None |