root / lib / mcpu.py @ 31d3b918
History | View | Annotate | Download (18.3 kB)
1 | 2f31098c | Iustin Pop | #
|
---|---|---|---|
2 | a8083063 | Iustin Pop | #
|
3 | a8083063 | Iustin Pop | |
4 | 5ae4945a | Iustin Pop | # Copyright (C) 2006, 2007, 2011, 2012 Google Inc.
|
5 | a8083063 | Iustin Pop | #
|
6 | a8083063 | Iustin Pop | # This program is free software; you can redistribute it and/or modify
|
7 | a8083063 | Iustin Pop | # it under the terms of the GNU General Public License as published by
|
8 | a8083063 | Iustin Pop | # the Free Software Foundation; either version 2 of the License, or
|
9 | a8083063 | Iustin Pop | # (at your option) any later version.
|
10 | a8083063 | Iustin Pop | #
|
11 | a8083063 | Iustin Pop | # This program is distributed in the hope that it will be useful, but
|
12 | a8083063 | Iustin Pop | # WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 | a8083063 | Iustin Pop | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
14 | a8083063 | Iustin Pop | # General Public License for more details.
|
15 | a8083063 | Iustin Pop | #
|
16 | a8083063 | Iustin Pop | # You should have received a copy of the GNU General Public License
|
17 | a8083063 | Iustin Pop | # along with this program; if not, write to the Free Software
|
18 | a8083063 | Iustin Pop | # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
|
19 | a8083063 | Iustin Pop | # 02110-1301, USA.
|
20 | a8083063 | Iustin Pop | |
21 | a8083063 | Iustin Pop | |
22 | a8083063 | Iustin Pop | """Module implementing the logic behind the cluster operations
|
23 | a8083063 | Iustin Pop |
|
24 | a8083063 | Iustin Pop | This module implements the logic for doing operations in the cluster. There
|
25 | a8083063 | Iustin Pop | are two kinds of classes defined:
|
26 | a8083063 | Iustin Pop | - logical units, which know how to deal with their specific opcode only
|
27 | a8083063 | Iustin Pop | - the processor, which dispatches the opcodes to their logical units
|
28 | a8083063 | Iustin Pop |
|
29 | a8083063 | Iustin Pop | """
|
30 | a8083063 | Iustin Pop | |
31 | b183c4a8 | Iustin Pop | import sys |
32 | a5eb7789 | Iustin Pop | import logging |
33 | 407339d0 | Michael Hanselmann | import random |
34 | 407339d0 | Michael Hanselmann | import time |
35 | 07923a3c | Michael Hanselmann | import itertools |
36 | b183c4a8 | Iustin Pop | import traceback |
37 | a8083063 | Iustin Pop | |
38 | a8083063 | Iustin Pop | from ganeti import opcodes |
39 | 580b1fdd | Jose A. Lopes | from ganeti import opcodes_base |
40 | a8083063 | Iustin Pop | from ganeti import constants |
41 | a8083063 | Iustin Pop | from ganeti import errors |
42 | 68d95757 | Guido Trotter | from ganeti import hooksmaster |
43 | a8083063 | Iustin Pop | from ganeti import cmdlib |
44 | 04864530 | Guido Trotter | from ganeti import locking |
45 | 557838c1 | René Nussbaumer | from ganeti import utils |
46 | ebc75510 | Michael Hanselmann | from ganeti import compat |
47 | a8083063 | Iustin Pop | |
48 | 7c0d6283 | Michael Hanselmann | |
49 | a1a7bc78 | Iustin Pop | _OP_PREFIX = "Op"
|
50 | a1a7bc78 | Iustin Pop | _LU_PREFIX = "LU"
|
51 | a1a7bc78 | Iustin Pop | |
52 | e9a81214 | Michael Hanselmann | #: LU classes which don't need to acquire the node allocation lock
|
53 | e9a81214 | Michael Hanselmann | #: (L{locking.NAL}) when they acquire all node or node resource locks
|
54 | b8028dcf | Michael Hanselmann | _NODE_ALLOC_WHITELIST = frozenset([])
|
55 | e9a81214 | Michael Hanselmann | |
56 | e9a81214 | Michael Hanselmann | #: LU classes which don't need to acquire the node allocation lock
|
57 | e9a81214 | Michael Hanselmann | #: (L{locking.NAL}) in the same mode (shared/exclusive) as the node
|
58 | e9a81214 | Michael Hanselmann | #: or node resource locks
|
59 | b8028dcf | Michael Hanselmann | _NODE_ALLOC_MODE_WHITELIST = compat.UniqueFrozenset([ |
60 | e9a81214 | Michael Hanselmann | cmdlib.LUBackupExport, |
61 | e9a81214 | Michael Hanselmann | cmdlib.LUBackupRemove, |
62 | e9a81214 | Michael Hanselmann | cmdlib.LUOobCommand, |
63 | e9a81214 | Michael Hanselmann | ]) |
64 | e9a81214 | Michael Hanselmann | |
65 | a1a7bc78 | Iustin Pop | |
66 | 831bbbc1 | Michael Hanselmann | class LockAcquireTimeout(Exception): |
67 | 831bbbc1 | Michael Hanselmann | """Exception to report timeouts on acquiring locks.
|
68 | 407339d0 | Michael Hanselmann |
|
69 | 407339d0 | Michael Hanselmann | """
|
70 | 407339d0 | Michael Hanselmann | |
71 | 407339d0 | Michael Hanselmann | |
72 | e3200b18 | Michael Hanselmann | def _CalculateLockAttemptTimeouts(): |
73 | e3200b18 | Michael Hanselmann | """Calculate timeouts for lock attempts.
|
74 | e3200b18 | Michael Hanselmann |
|
75 | e3200b18 | Michael Hanselmann | """
|
76 | d385a174 | Iustin Pop | result = [constants.LOCK_ATTEMPTS_MINWAIT] |
77 | d385a174 | Iustin Pop | running_sum = result[0]
|
78 | e3200b18 | Michael Hanselmann | |
79 | d385a174 | Iustin Pop | # Wait for a total of at least LOCK_ATTEMPTS_TIMEOUT before doing a
|
80 | d385a174 | Iustin Pop | # blocking acquire
|
81 | d385a174 | Iustin Pop | while running_sum < constants.LOCK_ATTEMPTS_TIMEOUT:
|
82 | e3200b18 | Michael Hanselmann | timeout = (result[-1] * 1.05) ** 1.25 |
83 | e3200b18 | Michael Hanselmann | |
84 | d385a174 | Iustin Pop | # Cap max timeout. This gives other jobs a chance to run even if
|
85 | d385a174 | Iustin Pop | # we're still trying to get our locks, before finally moving to a
|
86 | d385a174 | Iustin Pop | # blocking acquire.
|
87 | d385a174 | Iustin Pop | timeout = min(timeout, constants.LOCK_ATTEMPTS_MAXWAIT)
|
88 | d385a174 | Iustin Pop | # And also cap the lower boundary for safety
|
89 | d385a174 | Iustin Pop | timeout = max(timeout, constants.LOCK_ATTEMPTS_MINWAIT)
|
90 | e3200b18 | Michael Hanselmann | |
91 | e3200b18 | Michael Hanselmann | result.append(timeout) |
92 | d385a174 | Iustin Pop | running_sum += timeout |
93 | e3200b18 | Michael Hanselmann | |
94 | e3200b18 | Michael Hanselmann | return result
|
95 | e3200b18 | Michael Hanselmann | |
96 | e3200b18 | Michael Hanselmann | |
97 | a7770f03 | Michael Hanselmann | class LockAttemptTimeoutStrategy(object): |
98 | 407339d0 | Michael Hanselmann | """Class with lock acquire timeout strategy.
|
99 | 407339d0 | Michael Hanselmann |
|
100 | 407339d0 | Michael Hanselmann | """
|
101 | 407339d0 | Michael Hanselmann | __slots__ = [ |
102 | a7770f03 | Michael Hanselmann | "_timeouts",
|
103 | 407339d0 | Michael Hanselmann | "_random_fn",
|
104 | e3200b18 | Michael Hanselmann | "_time_fn",
|
105 | 407339d0 | Michael Hanselmann | ] |
106 | 407339d0 | Michael Hanselmann | |
107 | e3200b18 | Michael Hanselmann | _TIMEOUT_PER_ATTEMPT = _CalculateLockAttemptTimeouts() |
108 | 407339d0 | Michael Hanselmann | |
109 | a7770f03 | Michael Hanselmann | def __init__(self, _time_fn=time.time, _random_fn=random.random): |
110 | 407339d0 | Michael Hanselmann | """Initializes this class.
|
111 | 407339d0 | Michael Hanselmann |
|
112 | e3200b18 | Michael Hanselmann | @param _time_fn: Time function for unittests
|
113 | 407339d0 | Michael Hanselmann | @param _random_fn: Random number generator for unittests
|
114 | 407339d0 | Michael Hanselmann |
|
115 | 407339d0 | Michael Hanselmann | """
|
116 | 407339d0 | Michael Hanselmann | object.__init__(self) |
117 | 407339d0 | Michael Hanselmann | |
118 | a7770f03 | Michael Hanselmann | self._timeouts = iter(self._TIMEOUT_PER_ATTEMPT) |
119 | e3200b18 | Michael Hanselmann | self._time_fn = _time_fn
|
120 | e3200b18 | Michael Hanselmann | self._random_fn = _random_fn
|
121 | e3200b18 | Michael Hanselmann | |
122 | 407339d0 | Michael Hanselmann | def NextAttempt(self): |
123 | a7770f03 | Michael Hanselmann | """Returns the timeout for the next attempt.
|
124 | 407339d0 | Michael Hanselmann |
|
125 | 407339d0 | Michael Hanselmann | """
|
126 | a7770f03 | Michael Hanselmann | try:
|
127 | a7770f03 | Michael Hanselmann | timeout = self._timeouts.next()
|
128 | a7770f03 | Michael Hanselmann | except StopIteration: |
129 | a7770f03 | Michael Hanselmann | # No more timeouts, do blocking acquire
|
130 | a7770f03 | Michael Hanselmann | timeout = None
|
131 | 407339d0 | Michael Hanselmann | |
132 | a6db1af2 | Michael Hanselmann | if timeout is not None: |
133 | a6db1af2 | Michael Hanselmann | # Add a small variation (-/+ 5%) to timeout. This helps in situations
|
134 | a6db1af2 | Michael Hanselmann | # where two or more jobs are fighting for the same lock(s).
|
135 | a6db1af2 | Michael Hanselmann | variation_range = timeout * 0.1
|
136 | a6db1af2 | Michael Hanselmann | timeout += ((self._random_fn() * variation_range) -
|
137 | a6db1af2 | Michael Hanselmann | (variation_range * 0.5))
|
138 | 407339d0 | Michael Hanselmann | |
139 | a6db1af2 | Michael Hanselmann | return timeout
|
140 | 407339d0 | Michael Hanselmann | |
141 | 407339d0 | Michael Hanselmann | |
142 | b459a848 | Andrea Spadaccini | class OpExecCbBase: # pylint: disable=W0232 |
143 | 031a3e57 | Michael Hanselmann | """Base class for OpCode execution callbacks.
|
144 | 031a3e57 | Michael Hanselmann |
|
145 | 031a3e57 | Michael Hanselmann | """
|
146 | 031a3e57 | Michael Hanselmann | def NotifyStart(self): |
147 | 031a3e57 | Michael Hanselmann | """Called when we are about to execute the LU.
|
148 | 031a3e57 | Michael Hanselmann |
|
149 | 031a3e57 | Michael Hanselmann | This function is called when we're about to start the lu's Exec() method,
|
150 | 031a3e57 | Michael Hanselmann | that is, after we have acquired all locks.
|
151 | 031a3e57 | Michael Hanselmann |
|
152 | 031a3e57 | Michael Hanselmann | """
|
153 | 031a3e57 | Michael Hanselmann | |
154 | 031a3e57 | Michael Hanselmann | def Feedback(self, *args): |
155 | 031a3e57 | Michael Hanselmann | """Sends feedback from the LU code to the end-user.
|
156 | 031a3e57 | Michael Hanselmann |
|
157 | 031a3e57 | Michael Hanselmann | """
|
158 | 031a3e57 | Michael Hanselmann | |
159 | e4e59de8 | Michael Hanselmann | def CurrentPriority(self): # pylint: disable=R0201 |
160 | e4e59de8 | Michael Hanselmann | """Returns current priority or C{None}.
|
161 | ef2df7d3 | Michael Hanselmann |
|
162 | ef2df7d3 | Michael Hanselmann | """
|
163 | e4e59de8 | Michael Hanselmann | return None |
164 | ef2df7d3 | Michael Hanselmann | |
165 | 6a373640 | Michael Hanselmann | def SubmitManyJobs(self, jobs): |
166 | 6a373640 | Michael Hanselmann | """Submits jobs for processing.
|
167 | 6a373640 | Michael Hanselmann |
|
168 | 6a373640 | Michael Hanselmann | See L{jqueue.JobQueue.SubmitManyJobs}.
|
169 | 6a373640 | Michael Hanselmann |
|
170 | 6a373640 | Michael Hanselmann | """
|
171 | 6a373640 | Michael Hanselmann | raise NotImplementedError |
172 | 6a373640 | Michael Hanselmann | |
173 | 031a3e57 | Michael Hanselmann | |
174 | a1a7bc78 | Iustin Pop | def _LUNameForOpName(opname): |
175 | a1a7bc78 | Iustin Pop | """Computes the LU name for a given OpCode name.
|
176 | a1a7bc78 | Iustin Pop |
|
177 | a1a7bc78 | Iustin Pop | """
|
178 | a1a7bc78 | Iustin Pop | assert opname.startswith(_OP_PREFIX), \
|
179 | a1a7bc78 | Iustin Pop | "Invalid OpCode name, doesn't start with %s: %s" % (_OP_PREFIX, opname)
|
180 | a1a7bc78 | Iustin Pop | |
181 | a1a7bc78 | Iustin Pop | return _LU_PREFIX + opname[len(_OP_PREFIX):] |
182 | a1a7bc78 | Iustin Pop | |
183 | a1a7bc78 | Iustin Pop | |
184 | a1a7bc78 | Iustin Pop | def _ComputeDispatchTable(): |
185 | a1a7bc78 | Iustin Pop | """Computes the opcode-to-lu dispatch table.
|
186 | a1a7bc78 | Iustin Pop |
|
187 | a1a7bc78 | Iustin Pop | """
|
188 | a1a7bc78 | Iustin Pop | return dict((op, getattr(cmdlib, _LUNameForOpName(op.__name__))) |
189 | a1a7bc78 | Iustin Pop | for op in opcodes.OP_MAPPING.values() |
190 | a1a7bc78 | Iustin Pop | if op.WITH_LU)
|
191 | a1a7bc78 | Iustin Pop | |
192 | a1a7bc78 | Iustin Pop | |
193 | 07923a3c | Michael Hanselmann | def _SetBaseOpParams(src, defcomment, dst): |
194 | 07923a3c | Michael Hanselmann | """Copies basic opcode parameters.
|
195 | 07923a3c | Michael Hanselmann |
|
196 | 07923a3c | Michael Hanselmann | @type src: L{opcodes.OpCode}
|
197 | 07923a3c | Michael Hanselmann | @param src: Source opcode
|
198 | 07923a3c | Michael Hanselmann | @type defcomment: string
|
199 | 07923a3c | Michael Hanselmann | @param defcomment: Comment to specify if not already given
|
200 | 07923a3c | Michael Hanselmann | @type dst: L{opcodes.OpCode}
|
201 | 07923a3c | Michael Hanselmann | @param dst: Destination opcode
|
202 | 07923a3c | Michael Hanselmann |
|
203 | 07923a3c | Michael Hanselmann | """
|
204 | 07923a3c | Michael Hanselmann | if hasattr(src, "debug_level"): |
205 | 07923a3c | Michael Hanselmann | dst.debug_level = src.debug_level |
206 | 07923a3c | Michael Hanselmann | |
207 | 07923a3c | Michael Hanselmann | if (getattr(dst, "priority", None) is None and |
208 | 07923a3c | Michael Hanselmann | hasattr(src, "priority")): |
209 | 07923a3c | Michael Hanselmann | dst.priority = src.priority |
210 | 07923a3c | Michael Hanselmann | |
211 | 580b1fdd | Jose A. Lopes | if not getattr(dst, opcodes_base.COMMENT_ATTR, None): |
212 | 07923a3c | Michael Hanselmann | dst.comment = defcomment |
213 | 07923a3c | Michael Hanselmann | |
214 | e9cf6af0 | Klaus Aehlig | # FIXME: extend reason trail, showing the derivedness
|
215 | e9cf6af0 | Klaus Aehlig | if not getattr(dst, constants.OPCODE_REASON, None): |
216 | e9cf6af0 | Klaus Aehlig | dst.reason = getattr(src, constants.OPCODE_REASON, [])
|
217 | e9cf6af0 | Klaus Aehlig | |
218 | 07923a3c | Michael Hanselmann | |
219 | 07923a3c | Michael Hanselmann | def _ProcessResult(submit_fn, op, result): |
220 | 07923a3c | Michael Hanselmann | """Examines opcode result.
|
221 | 07923a3c | Michael Hanselmann |
|
222 | 07923a3c | Michael Hanselmann | If necessary, additional processing on the result is done.
|
223 | 07923a3c | Michael Hanselmann |
|
224 | 07923a3c | Michael Hanselmann | """
|
225 | 07923a3c | Michael Hanselmann | if isinstance(result, cmdlib.ResultWithJobs): |
226 | 07923a3c | Michael Hanselmann | # Copy basic parameters (e.g. priority)
|
227 | 07923a3c | Michael Hanselmann | map(compat.partial(_SetBaseOpParams, op,
|
228 | 07923a3c | Michael Hanselmann | "Submitted by %s" % op.OP_ID),
|
229 | 07923a3c | Michael Hanselmann | itertools.chain(*result.jobs)) |
230 | 07923a3c | Michael Hanselmann | |
231 | 07923a3c | Michael Hanselmann | # Submit jobs
|
232 | 07923a3c | Michael Hanselmann | job_submission = submit_fn(result.jobs) |
233 | 07923a3c | Michael Hanselmann | |
234 | 07923a3c | Michael Hanselmann | # Build dictionary
|
235 | 07923a3c | Michael Hanselmann | result = result.other |
236 | 07923a3c | Michael Hanselmann | |
237 | 07923a3c | Michael Hanselmann | assert constants.JOB_IDS_KEY not in result, \ |
238 | 07923a3c | Michael Hanselmann | "Key '%s' found in additional return values" % constants.JOB_IDS_KEY
|
239 | 07923a3c | Michael Hanselmann | |
240 | 07923a3c | Michael Hanselmann | result[constants.JOB_IDS_KEY] = job_submission |
241 | 07923a3c | Michael Hanselmann | |
242 | 07923a3c | Michael Hanselmann | return result
|
243 | 07923a3c | Michael Hanselmann | |
244 | 07923a3c | Michael Hanselmann | |
245 | abe362d3 | Michael Hanselmann | def _FailingSubmitManyJobs(_): |
246 | abe362d3 | Michael Hanselmann | """Implementation of L{OpExecCbBase.SubmitManyJobs} to raise an exception.
|
247 | abe362d3 | Michael Hanselmann |
|
248 | abe362d3 | Michael Hanselmann | """
|
249 | abe362d3 | Michael Hanselmann | raise errors.ProgrammerError("Opcodes processed without callbacks (e.g." |
250 | abe362d3 | Michael Hanselmann | " queries) can not submit jobs")
|
251 | abe362d3 | Michael Hanselmann | |
252 | abe362d3 | Michael Hanselmann | |
253 | e9a81214 | Michael Hanselmann | def _VerifyLocks(lu, glm, _mode_whitelist=_NODE_ALLOC_MODE_WHITELIST, |
254 | e9a81214 | Michael Hanselmann | _nal_whitelist=_NODE_ALLOC_WHITELIST): |
255 | e9a81214 | Michael Hanselmann | """Performs consistency checks on locks acquired by a logical unit.
|
256 | e9a81214 | Michael Hanselmann |
|
257 | e9a81214 | Michael Hanselmann | @type lu: L{cmdlib.LogicalUnit}
|
258 | e9a81214 | Michael Hanselmann | @param lu: Logical unit instance
|
259 | e9a81214 | Michael Hanselmann | @type glm: L{locking.GanetiLockManager}
|
260 | e9a81214 | Michael Hanselmann | @param glm: Lock manager
|
261 | e9a81214 | Michael Hanselmann |
|
262 | e9a81214 | Michael Hanselmann | """
|
263 | e9a81214 | Michael Hanselmann | if not __debug__: |
264 | e9a81214 | Michael Hanselmann | return
|
265 | e9a81214 | Michael Hanselmann | |
266 | e9a81214 | Michael Hanselmann | have_nal = glm.check_owned(locking.LEVEL_NODE_ALLOC, locking.NAL) |
267 | e9a81214 | Michael Hanselmann | |
268 | e9a81214 | Michael Hanselmann | for level in [locking.LEVEL_NODE, locking.LEVEL_NODE_RES]: |
269 | e9a81214 | Michael Hanselmann | # TODO: Verify using actual lock mode, not using LU variables
|
270 | e9a81214 | Michael Hanselmann | if level in lu.needed_locks: |
271 | e9a81214 | Michael Hanselmann | share_node_alloc = lu.share_locks[locking.LEVEL_NODE_ALLOC] |
272 | e9a81214 | Michael Hanselmann | share_level = lu.share_locks[level] |
273 | e9a81214 | Michael Hanselmann | |
274 | e9a81214 | Michael Hanselmann | if lu.__class__ in _mode_whitelist: |
275 | e9a81214 | Michael Hanselmann | assert share_node_alloc != share_level, \
|
276 | e9a81214 | Michael Hanselmann | "LU is whitelisted to use different modes for node allocation lock"
|
277 | e9a81214 | Michael Hanselmann | else:
|
278 | e9a81214 | Michael Hanselmann | assert bool(share_node_alloc) == bool(share_level), \ |
279 | e9a81214 | Michael Hanselmann | ("Node allocation lock must be acquired using the same mode as nodes"
|
280 | e9a81214 | Michael Hanselmann | " and node resources")
|
281 | e9a81214 | Michael Hanselmann | |
282 | e9a81214 | Michael Hanselmann | if lu.__class__ in _nal_whitelist: |
283 | e9a81214 | Michael Hanselmann | assert not have_nal, \ |
284 | e9a81214 | Michael Hanselmann | "LU is whitelisted for not acquiring the node allocation lock"
|
285 | e9a81214 | Michael Hanselmann | elif lu.needed_locks[level] == locking.ALL_SET or glm.owning_all(level): |
286 | e9a81214 | Michael Hanselmann | assert have_nal, \
|
287 | e9a81214 | Michael Hanselmann | ("Node allocation lock must be used if an LU acquires all nodes"
|
288 | e9a81214 | Michael Hanselmann | " or node resources")
|
289 | e9a81214 | Michael Hanselmann | |
290 | e9a81214 | Michael Hanselmann | |
291 | a8083063 | Iustin Pop | class Processor(object): |
292 | a8083063 | Iustin Pop | """Object which runs OpCodes"""
|
293 | a1a7bc78 | Iustin Pop | DISPATCH_TABLE = _ComputeDispatchTable() |
294 | a8083063 | Iustin Pop | |
295 | dc4bdf73 | Michael Hanselmann | def __init__(self, context, ec_id, enable_locks=True): |
296 | a8083063 | Iustin Pop | """Constructor for Processor
|
297 | a8083063 | Iustin Pop |
|
298 | adfa97e3 | Guido Trotter | @type context: GanetiContext
|
299 | adfa97e3 | Guido Trotter | @param context: global Ganeti context
|
300 | adfa97e3 | Guido Trotter | @type ec_id: string
|
301 | adfa97e3 | Guido Trotter | @param ec_id: execution context identifier
|
302 | adfa97e3 | Guido Trotter |
|
303 | a8083063 | Iustin Pop | """
|
304 | 1c901d13 | Guido Trotter | self.context = context
|
305 | adfa97e3 | Guido Trotter | self._ec_id = ec_id
|
306 | 031a3e57 | Michael Hanselmann | self._cbs = None |
307 | 87b3cb26 | Michael Hanselmann | self.rpc = context.rpc
|
308 | 68d95757 | Guido Trotter | self.hmclass = hooksmaster.HooksMaster
|
309 | dc4bdf73 | Michael Hanselmann | self._enable_locks = enable_locks
|
310 | dc4bdf73 | Michael Hanselmann | |
311 | dc4bdf73 | Michael Hanselmann | def _CheckLocksEnabled(self): |
312 | dc4bdf73 | Michael Hanselmann | """Checks if locking is enabled.
|
313 | dc4bdf73 | Michael Hanselmann |
|
314 | dc4bdf73 | Michael Hanselmann | @raise errors.ProgrammerError: In case locking is not enabled
|
315 | dc4bdf73 | Michael Hanselmann |
|
316 | dc4bdf73 | Michael Hanselmann | """
|
317 | dc4bdf73 | Michael Hanselmann | if not self._enable_locks: |
318 | dc4bdf73 | Michael Hanselmann | raise errors.ProgrammerError("Attempted to use disabled locks") |
319 | a8083063 | Iustin Pop | |
320 | 686d24f0 | Michael Hanselmann | def _AcquireLocks(self, level, names, shared, opportunistic, timeout): |
321 | 211b6132 | Michael Hanselmann | """Acquires locks via the Ganeti lock manager.
|
322 | 211b6132 | Michael Hanselmann |
|
323 | 211b6132 | Michael Hanselmann | @type level: int
|
324 | 211b6132 | Michael Hanselmann | @param level: Lock level
|
325 | 211b6132 | Michael Hanselmann | @type names: list or string
|
326 | 211b6132 | Michael Hanselmann | @param names: Lock names
|
327 | 211b6132 | Michael Hanselmann | @type shared: bool
|
328 | 211b6132 | Michael Hanselmann | @param shared: Whether the locks should be acquired in shared mode
|
329 | 686d24f0 | Michael Hanselmann | @type opportunistic: bool
|
330 | 686d24f0 | Michael Hanselmann | @param opportunistic: Whether to acquire opportunistically
|
331 | 211b6132 | Michael Hanselmann | @type timeout: None or float
|
332 | 211b6132 | Michael Hanselmann | @param timeout: Timeout for acquiring the locks
|
333 | 900df6cd | Michael Hanselmann | @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
|
334 | 900df6cd | Michael Hanselmann | amount of time
|
335 | 211b6132 | Michael Hanselmann |
|
336 | 211b6132 | Michael Hanselmann | """
|
337 | dc4bdf73 | Michael Hanselmann | self._CheckLocksEnabled()
|
338 | dc4bdf73 | Michael Hanselmann | |
339 | acf931b7 | Michael Hanselmann | if self._cbs: |
340 | e4e59de8 | Michael Hanselmann | priority = self._cbs.CurrentPriority()
|
341 | e4e59de8 | Michael Hanselmann | else:
|
342 | e4e59de8 | Michael Hanselmann | priority = None
|
343 | 211b6132 | Michael Hanselmann | |
344 | 211b6132 | Michael Hanselmann | acquired = self.context.glm.acquire(level, names, shared=shared,
|
345 | 686d24f0 | Michael Hanselmann | timeout=timeout, priority=priority, |
346 | 686d24f0 | Michael Hanselmann | opportunistic=opportunistic) |
347 | 211b6132 | Michael Hanselmann | |
348 | 900df6cd | Michael Hanselmann | if acquired is None: |
349 | 900df6cd | Michael Hanselmann | raise LockAcquireTimeout()
|
350 | 900df6cd | Michael Hanselmann | |
351 | 211b6132 | Michael Hanselmann | return acquired
|
352 | 211b6132 | Michael Hanselmann | |
353 | 36c381d7 | Guido Trotter | def _ExecLU(self, lu): |
354 | 36c381d7 | Guido Trotter | """Logical Unit execution sequence.
|
355 | 36c381d7 | Guido Trotter |
|
356 | 36c381d7 | Guido Trotter | """
|
357 | 36c381d7 | Guido Trotter | write_count = self.context.cfg.write_count
|
358 | 36c381d7 | Guido Trotter | lu.CheckPrereq() |
359 | 949dcb1d | Andrea Spadaccini | |
360 | 949dcb1d | Andrea Spadaccini | hm = self.BuildHooksManager(lu)
|
361 | 36c381d7 | Guido Trotter | h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE) |
362 | 36c381d7 | Guido Trotter | lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results, |
363 | 7b4c1cb9 | Michael Hanselmann | self.Log, None) |
364 | 20777413 | Iustin Pop | |
365 | 20777413 | Iustin Pop | if getattr(lu.op, "dry_run", False): |
366 | 20777413 | Iustin Pop | # in this mode, no post-hooks are run, and the config is not
|
367 | 20777413 | Iustin Pop | # written (as it might have been modified by another LU, and we
|
368 | 20777413 | Iustin Pop | # shouldn't do writeout on behalf of other threads
|
369 | 20777413 | Iustin Pop | self.LogInfo("dry-run mode requested, not actually executing" |
370 | 20777413 | Iustin Pop | " the operation")
|
371 | 20777413 | Iustin Pop | return lu.dry_run_result
|
372 | 20777413 | Iustin Pop | |
373 | abe362d3 | Michael Hanselmann | if self._cbs: |
374 | abe362d3 | Michael Hanselmann | submit_mj_fn = self._cbs.SubmitManyJobs
|
375 | abe362d3 | Michael Hanselmann | else:
|
376 | abe362d3 | Michael Hanselmann | submit_mj_fn = _FailingSubmitManyJobs |
377 | abe362d3 | Michael Hanselmann | |
378 | 36c381d7 | Guido Trotter | try:
|
379 | abe362d3 | Michael Hanselmann | result = _ProcessResult(submit_mj_fn, lu.op, lu.Exec(self.Log))
|
380 | 36c381d7 | Guido Trotter | h_results = hm.RunPhase(constants.HOOKS_PHASE_POST) |
381 | 36c381d7 | Guido Trotter | result = lu.HooksCallBack(constants.HOOKS_PHASE_POST, h_results, |
382 | 7b4c1cb9 | Michael Hanselmann | self.Log, result)
|
383 | 36c381d7 | Guido Trotter | finally:
|
384 | 36c381d7 | Guido Trotter | # FIXME: This needs locks if not lu_class.REQ_BGL
|
385 | 36c381d7 | Guido Trotter | if write_count != self.context.cfg.write_count: |
386 | 36c381d7 | Guido Trotter | hm.RunConfigUpdate() |
387 | 36c381d7 | Guido Trotter | |
388 | 36c381d7 | Guido Trotter | return result
|
389 | 36c381d7 | Guido Trotter | |
390 | 949dcb1d | Andrea Spadaccini | def BuildHooksManager(self, lu): |
391 | 949dcb1d | Andrea Spadaccini | return self.hmclass.BuildFromLu(lu.rpc.call_hooks_runner, lu) |
392 | 949dcb1d | Andrea Spadaccini | |
393 | e4e59de8 | Michael Hanselmann | def _LockAndExecLU(self, lu, level, calc_timeout): |
394 | 68adfdb2 | Guido Trotter | """Execute a Logical Unit, with the needed locks.
|
395 | 68adfdb2 | Guido Trotter |
|
396 | 68adfdb2 | Guido Trotter | This is a recursive function that starts locking the given level, and
|
397 | 68adfdb2 | Guido Trotter | proceeds up, till there are no more locks to acquire. Then it executes the
|
398 | 68adfdb2 | Guido Trotter | given LU and its opcodes.
|
399 | 68adfdb2 | Guido Trotter |
|
400 | 68adfdb2 | Guido Trotter | """
|
401 | e9a81214 | Michael Hanselmann | glm = self.context.glm
|
402 | ca2a79e1 | Guido Trotter | adding_locks = level in lu.add_locks
|
403 | ca2a79e1 | Guido Trotter | acquiring_locks = level in lu.needed_locks
|
404 | e9a81214 | Michael Hanselmann | |
405 | 8a2941c4 | Guido Trotter | if level not in locking.LEVELS: |
406 | e9a81214 | Michael Hanselmann | _VerifyLocks(lu, glm) |
407 | e9a81214 | Michael Hanselmann | |
408 | 031a3e57 | Michael Hanselmann | if self._cbs: |
409 | 031a3e57 | Michael Hanselmann | self._cbs.NotifyStart()
|
410 | 031a3e57 | Michael Hanselmann | |
411 | b183c4a8 | Iustin Pop | try:
|
412 | b183c4a8 | Iustin Pop | result = self._ExecLU(lu)
|
413 | b183c4a8 | Iustin Pop | except AssertionError, err: |
414 | b183c4a8 | Iustin Pop | # this is a bit ugly, as we don't know from which phase
|
415 | b183c4a8 | Iustin Pop | # (prereq, exec) this comes; but it's better than an exception
|
416 | b183c4a8 | Iustin Pop | # with no information
|
417 | b183c4a8 | Iustin Pop | (_, _, tb) = sys.exc_info() |
418 | b183c4a8 | Iustin Pop | err_info = traceback.format_tb(tb) |
419 | b183c4a8 | Iustin Pop | del tb
|
420 | b183c4a8 | Iustin Pop | logging.exception("Detected AssertionError")
|
421 | b183c4a8 | Iustin Pop | raise errors.OpExecError("Internal assertion error: please report" |
422 | b183c4a8 | Iustin Pop | " this as a bug.\nError message: '%s';"
|
423 | b183c4a8 | Iustin Pop | " location:\n%s" % (str(err), err_info[-1])) |
424 | 407339d0 | Michael Hanselmann | |
425 | ca2a79e1 | Guido Trotter | elif adding_locks and acquiring_locks: |
426 | ca2a79e1 | Guido Trotter | # We could both acquire and add locks at the same level, but for now we
|
427 | ca2a79e1 | Guido Trotter | # don't need this, so we'll avoid the complicated code needed.
|
428 | 407339d0 | Michael Hanselmann | raise NotImplementedError("Can't declare locks to acquire when adding" |
429 | 407339d0 | Michael Hanselmann | " others")
|
430 | 407339d0 | Michael Hanselmann | |
431 | ca2a79e1 | Guido Trotter | elif adding_locks or acquiring_locks: |
432 | dc4bdf73 | Michael Hanselmann | self._CheckLocksEnabled()
|
433 | dc4bdf73 | Michael Hanselmann | |
434 | fb8dcb62 | Guido Trotter | lu.DeclareLocks(level) |
435 | 3977a4c1 | Guido Trotter | share = lu.share_locks[level] |
436 | 686d24f0 | Michael Hanselmann | opportunistic = lu.opportunistic_locks[level] |
437 | 407339d0 | Michael Hanselmann | |
438 | 68adfdb2 | Guido Trotter | try:
|
439 | 407339d0 | Michael Hanselmann | assert adding_locks ^ acquiring_locks, \
|
440 | 407339d0 | Michael Hanselmann | "Locks must be either added or acquired"
|
441 | 407339d0 | Michael Hanselmann | |
442 | 407339d0 | Michael Hanselmann | if acquiring_locks:
|
443 | 407339d0 | Michael Hanselmann | # Acquiring locks
|
444 | 407339d0 | Michael Hanselmann | needed_locks = lu.needed_locks[level] |
445 | 407339d0 | Michael Hanselmann | |
446 | 686d24f0 | Michael Hanselmann | self._AcquireLocks(level, needed_locks, share, opportunistic,
|
447 | e4e59de8 | Michael Hanselmann | calc_timeout()) |
448 | 407339d0 | Michael Hanselmann | else:
|
449 | 407339d0 | Michael Hanselmann | # Adding locks
|
450 | 407339d0 | Michael Hanselmann | add_locks = lu.add_locks[level] |
451 | 407339d0 | Michael Hanselmann | lu.remove_locks[level] = add_locks |
452 | 407339d0 | Michael Hanselmann | |
453 | 407339d0 | Michael Hanselmann | try:
|
454 | e9a81214 | Michael Hanselmann | glm.add(level, add_locks, acquired=1, shared=share)
|
455 | 407339d0 | Michael Hanselmann | except errors.LockError:
|
456 | a9d40c65 | Iustin Pop | logging.exception("Detected lock error in level %s for locks"
|
457 | a9d40c65 | Iustin Pop | " %s, shared=%s", level, add_locks, share)
|
458 | 407339d0 | Michael Hanselmann | raise errors.OpPrereqError(
|
459 | a9d40c65 | Iustin Pop | "Couldn't add locks (%s), most likely because of another"
|
460 | a9d40c65 | Iustin Pop | " job who added them first" % add_locks,
|
461 | a9d40c65 | Iustin Pop | errors.ECODE_NOTUNIQUE) |
462 | 407339d0 | Michael Hanselmann | |
463 | ca2a79e1 | Guido Trotter | try:
|
464 | e4e59de8 | Michael Hanselmann | result = self._LockAndExecLU(lu, level + 1, calc_timeout) |
465 | ca2a79e1 | Guido Trotter | finally:
|
466 | ca2a79e1 | Guido Trotter | if level in lu.remove_locks: |
467 | e9a81214 | Michael Hanselmann | glm.remove(level, lu.remove_locks[level]) |
468 | 68adfdb2 | Guido Trotter | finally:
|
469 | e9a81214 | Michael Hanselmann | if glm.is_owned(level):
|
470 | e9a81214 | Michael Hanselmann | glm.release(level) |
471 | 407339d0 | Michael Hanselmann | |
472 | 68adfdb2 | Guido Trotter | else:
|
473 | e4e59de8 | Michael Hanselmann | result = self._LockAndExecLU(lu, level + 1, calc_timeout) |
474 | 68adfdb2 | Guido Trotter | |
475 | 68adfdb2 | Guido Trotter | return result
|
476 | 68adfdb2 | Guido Trotter | |
477 | fdfa63cb | Thomas Thrainer | # pylint: disable=R0201
|
478 | b8f45292 | Thomas Thrainer | def _CheckLUResult(self, op, result): |
479 | b8f45292 | Thomas Thrainer | """Check the LU result against the contract in the opcode.
|
480 | b8f45292 | Thomas Thrainer |
|
481 | b8f45292 | Thomas Thrainer | """
|
482 | b8f45292 | Thomas Thrainer | resultcheck_fn = op.OP_RESULT |
483 | b8f45292 | Thomas Thrainer | if not (resultcheck_fn is None or resultcheck_fn(result)): |
484 | b8f45292 | Thomas Thrainer | logging.error("Expected opcode result matching %s, got %s",
|
485 | b8f45292 | Thomas Thrainer | resultcheck_fn, result) |
486 | b8f45292 | Thomas Thrainer | if not getattr(op, "dry_run", False): |
487 | b8f45292 | Thomas Thrainer | # FIXME: LUs should still behave in dry_run mode, or
|
488 | b8f45292 | Thomas Thrainer | # alternately we should have OP_DRYRUN_RESULT; in the
|
489 | b8f45292 | Thomas Thrainer | # meantime, we simply skip the OP_RESULT check in dry-run mode
|
490 | b8f45292 | Thomas Thrainer | raise errors.OpResultError("Opcode result does not match %s: %s" % |
491 | b8f45292 | Thomas Thrainer | (resultcheck_fn, utils.Truncate(result, 80)))
|
492 | b8f45292 | Thomas Thrainer | |
493 | e4e59de8 | Michael Hanselmann | def ExecOpCode(self, op, cbs, timeout=None): |
494 | a8083063 | Iustin Pop | """Execute an opcode.
|
495 | a8083063 | Iustin Pop |
|
496 | e92376d7 | Iustin Pop | @type op: an OpCode instance
|
497 | e92376d7 | Iustin Pop | @param op: the opcode to be executed
|
498 | 031a3e57 | Michael Hanselmann | @type cbs: L{OpExecCbBase}
|
499 | 031a3e57 | Michael Hanselmann | @param cbs: Runtime callbacks
|
500 | 831bbbc1 | Michael Hanselmann | @type timeout: float or None
|
501 | 831bbbc1 | Michael Hanselmann | @param timeout: Maximum time to acquire all locks, None for no timeout
|
502 | 831bbbc1 | Michael Hanselmann | @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
|
503 | 831bbbc1 | Michael Hanselmann | amount of time
|
504 | a8083063 | Iustin Pop |
|
505 | a8083063 | Iustin Pop | """
|
506 | a8083063 | Iustin Pop | if not isinstance(op, opcodes.OpCode): |
507 | 3ecf6786 | Iustin Pop | raise errors.ProgrammerError("Non-opcode instance passed" |
508 | c7bb3095 | Michael Hanselmann | " to ExecOpcode (%s)" % type(op)) |
509 | a8083063 | Iustin Pop | |
510 | 831bbbc1 | Michael Hanselmann | lu_class = self.DISPATCH_TABLE.get(op.__class__, None) |
511 | 831bbbc1 | Michael Hanselmann | if lu_class is None: |
512 | 831bbbc1 | Michael Hanselmann | raise errors.OpCodeUnknown("Unknown opcode") |
513 | 831bbbc1 | Michael Hanselmann | |
514 | 831bbbc1 | Michael Hanselmann | if timeout is None: |
515 | 831bbbc1 | Michael Hanselmann | calc_timeout = lambda: None |
516 | 831bbbc1 | Michael Hanselmann | else:
|
517 | 557838c1 | René Nussbaumer | calc_timeout = utils.RunningTimeout(timeout, False).Remaining
|
518 | 831bbbc1 | Michael Hanselmann | |
519 | 031a3e57 | Michael Hanselmann | self._cbs = cbs
|
520 | fe482621 | Iustin Pop | try:
|
521 | dc4bdf73 | Michael Hanselmann | if self._enable_locks: |
522 | dc4bdf73 | Michael Hanselmann | # Acquire the Big Ganeti Lock exclusively if this LU requires it,
|
523 | dc4bdf73 | Michael Hanselmann | # and in a shared fashion otherwise (to prevent concurrent run with
|
524 | dc4bdf73 | Michael Hanselmann | # an exclusive LU.
|
525 | dc4bdf73 | Michael Hanselmann | self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL,
|
526 | 686d24f0 | Michael Hanselmann | not lu_class.REQ_BGL, False, calc_timeout()) |
527 | dc4bdf73 | Michael Hanselmann | elif lu_class.REQ_BGL:
|
528 | dc4bdf73 | Michael Hanselmann | raise errors.ProgrammerError("Opcode '%s' requires BGL, but locks are" |
529 | dc4bdf73 | Michael Hanselmann | " disabled" % op.OP_ID)
|
530 | dc4bdf73 | Michael Hanselmann | |
531 | 831bbbc1 | Michael Hanselmann | try:
|
532 | 831bbbc1 | Michael Hanselmann | lu = lu_class(self, op, self.context, self.rpc) |
533 | 831bbbc1 | Michael Hanselmann | lu.ExpandNames() |
534 | 831bbbc1 | Michael Hanselmann | assert lu.needed_locks is not None, "needed_locks not set by LU" |
535 | 407339d0 | Michael Hanselmann | |
536 | 407339d0 | Michael Hanselmann | try:
|
537 | 46cde471 | Michael Hanselmann | result = self._LockAndExecLU(lu, locking.LEVEL_CLUSTER + 1, |
538 | 46cde471 | Michael Hanselmann | calc_timeout) |
539 | 831bbbc1 | Michael Hanselmann | finally:
|
540 | 831bbbc1 | Michael Hanselmann | if self._ec_id: |
541 | 831bbbc1 | Michael Hanselmann | self.context.cfg.DropECReservations(self._ec_id) |
542 | 831bbbc1 | Michael Hanselmann | finally:
|
543 | dc4bdf73 | Michael Hanselmann | # Release BGL if owned
|
544 | dc4bdf73 | Michael Hanselmann | if self.context.glm.is_owned(locking.LEVEL_CLUSTER): |
545 | dc4bdf73 | Michael Hanselmann | assert self._enable_locks |
546 | dc4bdf73 | Michael Hanselmann | self.context.glm.release(locking.LEVEL_CLUSTER)
|
547 | 04864530 | Guido Trotter | finally:
|
548 | 031a3e57 | Michael Hanselmann | self._cbs = None |
549 | 6a4aa7c1 | Iustin Pop | |
550 | b8f45292 | Thomas Thrainer | self._CheckLUResult(op, result)
|
551 | 1ce03fb1 | Michael Hanselmann | |
552 | 1ce03fb1 | Michael Hanselmann | return result
|
553 | 1ce03fb1 | Michael Hanselmann | |
554 | 7b4c1cb9 | Michael Hanselmann | def Log(self, *args): |
555 | 031a3e57 | Michael Hanselmann | """Forward call to feedback callback function.
|
556 | 031a3e57 | Michael Hanselmann |
|
557 | 031a3e57 | Michael Hanselmann | """
|
558 | 031a3e57 | Michael Hanselmann | if self._cbs: |
559 | 031a3e57 | Michael Hanselmann | self._cbs.Feedback(*args)
|
560 | 031a3e57 | Michael Hanselmann | |
561 | 0fbbf897 | Iustin Pop | def LogStep(self, current, total, message): |
562 | 0fbbf897 | Iustin Pop | """Log a change in LU execution progress.
|
563 | 0fbbf897 | Iustin Pop |
|
564 | 0fbbf897 | Iustin Pop | """
|
565 | a5eb7789 | Iustin Pop | logging.debug("Step %d/%d %s", current, total, message)
|
566 | 7b4c1cb9 | Michael Hanselmann | self.Log("STEP %d/%d %s" % (current, total, message)) |
567 | 0fbbf897 | Iustin Pop | |
568 | c0088fb9 | Iustin Pop | def LogWarning(self, message, *args, **kwargs): |
569 | 0fbbf897 | Iustin Pop | """Log a warning to the logs and the user.
|
570 | 0fbbf897 | Iustin Pop |
|
571 | c0088fb9 | Iustin Pop | The optional keyword argument is 'hint' and can be used to show a
|
572 | c0088fb9 | Iustin Pop | hint to the user (presumably related to the warning). If the
|
573 | c0088fb9 | Iustin Pop | message is empty, it will not be printed at all, allowing one to
|
574 | c0088fb9 | Iustin Pop | show only a hint.
|
575 | 0fbbf897 | Iustin Pop |
|
576 | c0088fb9 | Iustin Pop | """
|
577 | c0088fb9 | Iustin Pop | assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \ |
578 | c0088fb9 | Iustin Pop | "Invalid keyword arguments for LogWarning (%s)" % str(kwargs) |
579 | c0088fb9 | Iustin Pop | if args:
|
580 | c0088fb9 | Iustin Pop | message = message % tuple(args)
|
581 | c0088fb9 | Iustin Pop | if message:
|
582 | c0088fb9 | Iustin Pop | logging.warning(message) |
583 | 7b4c1cb9 | Michael Hanselmann | self.Log(" - WARNING: %s" % message) |
584 | c0088fb9 | Iustin Pop | if "hint" in kwargs: |
585 | 7b4c1cb9 | Michael Hanselmann | self.Log(" Hint: %s" % kwargs["hint"]) |
586 | c0088fb9 | Iustin Pop | |
587 | c0088fb9 | Iustin Pop | def LogInfo(self, message, *args): |
588 | 0fbbf897 | Iustin Pop | """Log an informational message to the logs and the user.
|
589 | 0fbbf897 | Iustin Pop |
|
590 | 0fbbf897 | Iustin Pop | """
|
591 | c0088fb9 | Iustin Pop | if args:
|
592 | c0088fb9 | Iustin Pop | message = message % tuple(args)
|
593 | a5eb7789 | Iustin Pop | logging.info(message) |
594 | 7b4c1cb9 | Michael Hanselmann | self.Log(" - INFO: %s" % message) |
595 | 0fbbf897 | Iustin Pop | |
596 | adfa97e3 | Guido Trotter | def GetECId(self): |
597 | 3ae70d76 | Michael Hanselmann | """Returns the current execution context ID.
|
598 | 3ae70d76 | Michael Hanselmann |
|
599 | 3ae70d76 | Michael Hanselmann | """
|
600 | adfa97e3 | Guido Trotter | if not self._ec_id: |
601 | 3ae70d76 | Michael Hanselmann | raise errors.ProgrammerError("Tried to use execution context id when" |
602 | 3ae70d76 | Michael Hanselmann | " not set")
|
603 | adfa97e3 | Guido Trotter | return self._ec_id |