root / lib / mcpu.py @ bbfa259c
History | View | Annotate | Download (26.1 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 | a8083063 | Iustin Pop | from ganeti import constants |
40 | a8083063 | Iustin Pop | from ganeti import errors |
41 | a8083063 | Iustin Pop | from ganeti import cmdlib |
42 | 04864530 | Guido Trotter | from ganeti import locking |
43 | 557838c1 | René Nussbaumer | from ganeti import utils |
44 | ebc75510 | Michael Hanselmann | from ganeti import compat |
45 | b3a6f851 | Michael Hanselmann | from ganeti import pathutils |
46 | a8083063 | Iustin Pop | |
47 | 7c0d6283 | Michael Hanselmann | |
48 | a1a7bc78 | Iustin Pop | _OP_PREFIX = "Op"
|
49 | a1a7bc78 | Iustin Pop | _LU_PREFIX = "LU"
|
50 | a1a7bc78 | Iustin Pop | |
51 | e9a81214 | Michael Hanselmann | #: LU classes which don't need to acquire the node allocation lock
|
52 | e9a81214 | Michael Hanselmann | #: (L{locking.NAL}) when they acquire all node or node resource locks
|
53 | b8028dcf | Michael Hanselmann | _NODE_ALLOC_WHITELIST = frozenset([])
|
54 | e9a81214 | Michael Hanselmann | |
55 | e9a81214 | Michael Hanselmann | #: LU classes which don't need to acquire the node allocation lock
|
56 | e9a81214 | Michael Hanselmann | #: (L{locking.NAL}) in the same mode (shared/exclusive) as the node
|
57 | e9a81214 | Michael Hanselmann | #: or node resource locks
|
58 | b8028dcf | Michael Hanselmann | _NODE_ALLOC_MODE_WHITELIST = compat.UniqueFrozenset([ |
59 | e9a81214 | Michael Hanselmann | cmdlib.LUBackupExport, |
60 | e9a81214 | Michael Hanselmann | cmdlib.LUBackupRemove, |
61 | e9a81214 | Michael Hanselmann | cmdlib.LUOobCommand, |
62 | e9a81214 | Michael Hanselmann | ]) |
63 | e9a81214 | Michael Hanselmann | |
64 | a1a7bc78 | Iustin Pop | |
65 | 831bbbc1 | Michael Hanselmann | class LockAcquireTimeout(Exception): |
66 | 831bbbc1 | Michael Hanselmann | """Exception to report timeouts on acquiring locks.
|
67 | 407339d0 | Michael Hanselmann |
|
68 | 407339d0 | Michael Hanselmann | """
|
69 | 407339d0 | Michael Hanselmann | |
70 | 407339d0 | Michael Hanselmann | |
71 | e3200b18 | Michael Hanselmann | def _CalculateLockAttemptTimeouts(): |
72 | e3200b18 | Michael Hanselmann | """Calculate timeouts for lock attempts.
|
73 | e3200b18 | Michael Hanselmann |
|
74 | e3200b18 | Michael Hanselmann | """
|
75 | d385a174 | Iustin Pop | result = [constants.LOCK_ATTEMPTS_MINWAIT] |
76 | d385a174 | Iustin Pop | running_sum = result[0]
|
77 | e3200b18 | Michael Hanselmann | |
78 | d385a174 | Iustin Pop | # Wait for a total of at least LOCK_ATTEMPTS_TIMEOUT before doing a
|
79 | d385a174 | Iustin Pop | # blocking acquire
|
80 | d385a174 | Iustin Pop | while running_sum < constants.LOCK_ATTEMPTS_TIMEOUT:
|
81 | e3200b18 | Michael Hanselmann | timeout = (result[-1] * 1.05) ** 1.25 |
82 | e3200b18 | Michael Hanselmann | |
83 | d385a174 | Iustin Pop | # Cap max timeout. This gives other jobs a chance to run even if
|
84 | d385a174 | Iustin Pop | # we're still trying to get our locks, before finally moving to a
|
85 | d385a174 | Iustin Pop | # blocking acquire.
|
86 | d385a174 | Iustin Pop | timeout = min(timeout, constants.LOCK_ATTEMPTS_MAXWAIT)
|
87 | d385a174 | Iustin Pop | # And also cap the lower boundary for safety
|
88 | d385a174 | Iustin Pop | timeout = max(timeout, constants.LOCK_ATTEMPTS_MINWAIT)
|
89 | e3200b18 | Michael Hanselmann | |
90 | e3200b18 | Michael Hanselmann | result.append(timeout) |
91 | d385a174 | Iustin Pop | running_sum += timeout |
92 | e3200b18 | Michael Hanselmann | |
93 | e3200b18 | Michael Hanselmann | return result
|
94 | e3200b18 | Michael Hanselmann | |
95 | e3200b18 | Michael Hanselmann | |
96 | a7770f03 | Michael Hanselmann | class LockAttemptTimeoutStrategy(object): |
97 | 407339d0 | Michael Hanselmann | """Class with lock acquire timeout strategy.
|
98 | 407339d0 | Michael Hanselmann |
|
99 | 407339d0 | Michael Hanselmann | """
|
100 | 407339d0 | Michael Hanselmann | __slots__ = [ |
101 | a7770f03 | Michael Hanselmann | "_timeouts",
|
102 | 407339d0 | Michael Hanselmann | "_random_fn",
|
103 | e3200b18 | Michael Hanselmann | "_time_fn",
|
104 | 407339d0 | Michael Hanselmann | ] |
105 | 407339d0 | Michael Hanselmann | |
106 | e3200b18 | Michael Hanselmann | _TIMEOUT_PER_ATTEMPT = _CalculateLockAttemptTimeouts() |
107 | 407339d0 | Michael Hanselmann | |
108 | a7770f03 | Michael Hanselmann | def __init__(self, _time_fn=time.time, _random_fn=random.random): |
109 | 407339d0 | Michael Hanselmann | """Initializes this class.
|
110 | 407339d0 | Michael Hanselmann |
|
111 | e3200b18 | Michael Hanselmann | @param _time_fn: Time function for unittests
|
112 | 407339d0 | Michael Hanselmann | @param _random_fn: Random number generator for unittests
|
113 | 407339d0 | Michael Hanselmann |
|
114 | 407339d0 | Michael Hanselmann | """
|
115 | 407339d0 | Michael Hanselmann | object.__init__(self) |
116 | 407339d0 | Michael Hanselmann | |
117 | a7770f03 | Michael Hanselmann | self._timeouts = iter(self._TIMEOUT_PER_ATTEMPT) |
118 | e3200b18 | Michael Hanselmann | self._time_fn = _time_fn
|
119 | e3200b18 | Michael Hanselmann | self._random_fn = _random_fn
|
120 | e3200b18 | Michael Hanselmann | |
121 | 407339d0 | Michael Hanselmann | def NextAttempt(self): |
122 | a7770f03 | Michael Hanselmann | """Returns the timeout for the next attempt.
|
123 | 407339d0 | Michael Hanselmann |
|
124 | 407339d0 | Michael Hanselmann | """
|
125 | a7770f03 | Michael Hanselmann | try:
|
126 | a7770f03 | Michael Hanselmann | timeout = self._timeouts.next()
|
127 | a7770f03 | Michael Hanselmann | except StopIteration: |
128 | a7770f03 | Michael Hanselmann | # No more timeouts, do blocking acquire
|
129 | a7770f03 | Michael Hanselmann | timeout = None
|
130 | 407339d0 | Michael Hanselmann | |
131 | a6db1af2 | Michael Hanselmann | if timeout is not None: |
132 | a6db1af2 | Michael Hanselmann | # Add a small variation (-/+ 5%) to timeout. This helps in situations
|
133 | a6db1af2 | Michael Hanselmann | # where two or more jobs are fighting for the same lock(s).
|
134 | a6db1af2 | Michael Hanselmann | variation_range = timeout * 0.1
|
135 | a6db1af2 | Michael Hanselmann | timeout += ((self._random_fn() * variation_range) -
|
136 | a6db1af2 | Michael Hanselmann | (variation_range * 0.5))
|
137 | 407339d0 | Michael Hanselmann | |
138 | a6db1af2 | Michael Hanselmann | return timeout
|
139 | 407339d0 | Michael Hanselmann | |
140 | 407339d0 | Michael Hanselmann | |
141 | b459a848 | Andrea Spadaccini | class OpExecCbBase: # pylint: disable=W0232 |
142 | 031a3e57 | Michael Hanselmann | """Base class for OpCode execution callbacks.
|
143 | 031a3e57 | Michael Hanselmann |
|
144 | 031a3e57 | Michael Hanselmann | """
|
145 | 031a3e57 | Michael Hanselmann | def NotifyStart(self): |
146 | 031a3e57 | Michael Hanselmann | """Called when we are about to execute the LU.
|
147 | 031a3e57 | Michael Hanselmann |
|
148 | 031a3e57 | Michael Hanselmann | This function is called when we're about to start the lu's Exec() method,
|
149 | 031a3e57 | Michael Hanselmann | that is, after we have acquired all locks.
|
150 | 031a3e57 | Michael Hanselmann |
|
151 | 031a3e57 | Michael Hanselmann | """
|
152 | 031a3e57 | Michael Hanselmann | |
153 | 031a3e57 | Michael Hanselmann | def Feedback(self, *args): |
154 | 031a3e57 | Michael Hanselmann | """Sends feedback from the LU code to the end-user.
|
155 | 031a3e57 | Michael Hanselmann |
|
156 | 031a3e57 | Michael Hanselmann | """
|
157 | 031a3e57 | Michael Hanselmann | |
158 | e4e59de8 | Michael Hanselmann | def CurrentPriority(self): # pylint: disable=R0201 |
159 | e4e59de8 | Michael Hanselmann | """Returns current priority or C{None}.
|
160 | ef2df7d3 | Michael Hanselmann |
|
161 | ef2df7d3 | Michael Hanselmann | """
|
162 | e4e59de8 | Michael Hanselmann | return None |
163 | ef2df7d3 | Michael Hanselmann | |
164 | 6a373640 | Michael Hanselmann | def SubmitManyJobs(self, jobs): |
165 | 6a373640 | Michael Hanselmann | """Submits jobs for processing.
|
166 | 6a373640 | Michael Hanselmann |
|
167 | 6a373640 | Michael Hanselmann | See L{jqueue.JobQueue.SubmitManyJobs}.
|
168 | 6a373640 | Michael Hanselmann |
|
169 | 6a373640 | Michael Hanselmann | """
|
170 | 6a373640 | Michael Hanselmann | raise NotImplementedError |
171 | 6a373640 | Michael Hanselmann | |
172 | 031a3e57 | Michael Hanselmann | |
173 | a1a7bc78 | Iustin Pop | def _LUNameForOpName(opname): |
174 | a1a7bc78 | Iustin Pop | """Computes the LU name for a given OpCode name.
|
175 | a1a7bc78 | Iustin Pop |
|
176 | a1a7bc78 | Iustin Pop | """
|
177 | a1a7bc78 | Iustin Pop | assert opname.startswith(_OP_PREFIX), \
|
178 | a1a7bc78 | Iustin Pop | "Invalid OpCode name, doesn't start with %s: %s" % (_OP_PREFIX, opname)
|
179 | a1a7bc78 | Iustin Pop | |
180 | a1a7bc78 | Iustin Pop | return _LU_PREFIX + opname[len(_OP_PREFIX):] |
181 | a1a7bc78 | Iustin Pop | |
182 | a1a7bc78 | Iustin Pop | |
183 | a1a7bc78 | Iustin Pop | def _ComputeDispatchTable(): |
184 | a1a7bc78 | Iustin Pop | """Computes the opcode-to-lu dispatch table.
|
185 | a1a7bc78 | Iustin Pop |
|
186 | a1a7bc78 | Iustin Pop | """
|
187 | a1a7bc78 | Iustin Pop | return dict((op, getattr(cmdlib, _LUNameForOpName(op.__name__))) |
188 | a1a7bc78 | Iustin Pop | for op in opcodes.OP_MAPPING.values() |
189 | a1a7bc78 | Iustin Pop | if op.WITH_LU)
|
190 | a1a7bc78 | Iustin Pop | |
191 | a1a7bc78 | Iustin Pop | |
192 | 07923a3c | Michael Hanselmann | def _SetBaseOpParams(src, defcomment, dst): |
193 | 07923a3c | Michael Hanselmann | """Copies basic opcode parameters.
|
194 | 07923a3c | Michael Hanselmann |
|
195 | 07923a3c | Michael Hanselmann | @type src: L{opcodes.OpCode}
|
196 | 07923a3c | Michael Hanselmann | @param src: Source opcode
|
197 | 07923a3c | Michael Hanselmann | @type defcomment: string
|
198 | 07923a3c | Michael Hanselmann | @param defcomment: Comment to specify if not already given
|
199 | 07923a3c | Michael Hanselmann | @type dst: L{opcodes.OpCode}
|
200 | 07923a3c | Michael Hanselmann | @param dst: Destination opcode
|
201 | 07923a3c | Michael Hanselmann |
|
202 | 07923a3c | Michael Hanselmann | """
|
203 | 07923a3c | Michael Hanselmann | if hasattr(src, "debug_level"): |
204 | 07923a3c | Michael Hanselmann | dst.debug_level = src.debug_level |
205 | 07923a3c | Michael Hanselmann | |
206 | 07923a3c | Michael Hanselmann | if (getattr(dst, "priority", None) is None and |
207 | 07923a3c | Michael Hanselmann | hasattr(src, "priority")): |
208 | 07923a3c | Michael Hanselmann | dst.priority = src.priority |
209 | 07923a3c | Michael Hanselmann | |
210 | 07923a3c | Michael Hanselmann | if not getattr(dst, opcodes.COMMENT_ATTR, None): |
211 | 07923a3c | Michael Hanselmann | dst.comment = defcomment |
212 | 07923a3c | Michael Hanselmann | |
213 | 07923a3c | Michael Hanselmann | |
214 | 07923a3c | Michael Hanselmann | def _ProcessResult(submit_fn, op, result): |
215 | 07923a3c | Michael Hanselmann | """Examines opcode result.
|
216 | 07923a3c | Michael Hanselmann |
|
217 | 07923a3c | Michael Hanselmann | If necessary, additional processing on the result is done.
|
218 | 07923a3c | Michael Hanselmann |
|
219 | 07923a3c | Michael Hanselmann | """
|
220 | 07923a3c | Michael Hanselmann | if isinstance(result, cmdlib.ResultWithJobs): |
221 | 07923a3c | Michael Hanselmann | # Copy basic parameters (e.g. priority)
|
222 | 07923a3c | Michael Hanselmann | map(compat.partial(_SetBaseOpParams, op,
|
223 | 07923a3c | Michael Hanselmann | "Submitted by %s" % op.OP_ID),
|
224 | 07923a3c | Michael Hanselmann | itertools.chain(*result.jobs)) |
225 | 07923a3c | Michael Hanselmann | |
226 | 07923a3c | Michael Hanselmann | # Submit jobs
|
227 | 07923a3c | Michael Hanselmann | job_submission = submit_fn(result.jobs) |
228 | 07923a3c | Michael Hanselmann | |
229 | 07923a3c | Michael Hanselmann | # Build dictionary
|
230 | 07923a3c | Michael Hanselmann | result = result.other |
231 | 07923a3c | Michael Hanselmann | |
232 | 07923a3c | Michael Hanselmann | assert constants.JOB_IDS_KEY not in result, \ |
233 | 07923a3c | Michael Hanselmann | "Key '%s' found in additional return values" % constants.JOB_IDS_KEY
|
234 | 07923a3c | Michael Hanselmann | |
235 | 07923a3c | Michael Hanselmann | result[constants.JOB_IDS_KEY] = job_submission |
236 | 07923a3c | Michael Hanselmann | |
237 | 07923a3c | Michael Hanselmann | return result
|
238 | 07923a3c | Michael Hanselmann | |
239 | 07923a3c | Michael Hanselmann | |
240 | abe362d3 | Michael Hanselmann | def _FailingSubmitManyJobs(_): |
241 | abe362d3 | Michael Hanselmann | """Implementation of L{OpExecCbBase.SubmitManyJobs} to raise an exception.
|
242 | abe362d3 | Michael Hanselmann |
|
243 | abe362d3 | Michael Hanselmann | """
|
244 | abe362d3 | Michael Hanselmann | raise errors.ProgrammerError("Opcodes processed without callbacks (e.g." |
245 | abe362d3 | Michael Hanselmann | " queries) can not submit jobs")
|
246 | abe362d3 | Michael Hanselmann | |
247 | abe362d3 | Michael Hanselmann | |
248 | 949dcb1d | Andrea Spadaccini | def _RpcResultsToHooksResults(rpc_results): |
249 | 949dcb1d | Andrea Spadaccini | """Function to convert RPC results to the format expected by HooksMaster.
|
250 | 949dcb1d | Andrea Spadaccini |
|
251 | 949dcb1d | Andrea Spadaccini | @type rpc_results: dict(node: L{rpc.RpcResult})
|
252 | 949dcb1d | Andrea Spadaccini | @param rpc_results: RPC results
|
253 | 949dcb1d | Andrea Spadaccini | @rtype: dict(node: (fail_msg, offline, hooks_results))
|
254 | 949dcb1d | Andrea Spadaccini | @return: RPC results unpacked according to the format expected by
|
255 | 949dcb1d | Andrea Spadaccini | L({mcpu.HooksMaster}
|
256 | 949dcb1d | Andrea Spadaccini |
|
257 | 949dcb1d | Andrea Spadaccini | """
|
258 | 949dcb1d | Andrea Spadaccini | return dict((node, (rpc_res.fail_msg, rpc_res.offline, rpc_res.payload)) |
259 | 949dcb1d | Andrea Spadaccini | for (node, rpc_res) in rpc_results.items()) |
260 | 949dcb1d | Andrea Spadaccini | |
261 | 949dcb1d | Andrea Spadaccini | |
262 | e9a81214 | Michael Hanselmann | def _VerifyLocks(lu, glm, _mode_whitelist=_NODE_ALLOC_MODE_WHITELIST, |
263 | e9a81214 | Michael Hanselmann | _nal_whitelist=_NODE_ALLOC_WHITELIST): |
264 | e9a81214 | Michael Hanselmann | """Performs consistency checks on locks acquired by a logical unit.
|
265 | e9a81214 | Michael Hanselmann |
|
266 | e9a81214 | Michael Hanselmann | @type lu: L{cmdlib.LogicalUnit}
|
267 | e9a81214 | Michael Hanselmann | @param lu: Logical unit instance
|
268 | e9a81214 | Michael Hanselmann | @type glm: L{locking.GanetiLockManager}
|
269 | e9a81214 | Michael Hanselmann | @param glm: Lock manager
|
270 | e9a81214 | Michael Hanselmann |
|
271 | e9a81214 | Michael Hanselmann | """
|
272 | e9a81214 | Michael Hanselmann | if not __debug__: |
273 | e9a81214 | Michael Hanselmann | return
|
274 | e9a81214 | Michael Hanselmann | |
275 | e9a81214 | Michael Hanselmann | have_nal = glm.check_owned(locking.LEVEL_NODE_ALLOC, locking.NAL) |
276 | e9a81214 | Michael Hanselmann | |
277 | e9a81214 | Michael Hanselmann | for level in [locking.LEVEL_NODE, locking.LEVEL_NODE_RES]: |
278 | e9a81214 | Michael Hanselmann | # TODO: Verify using actual lock mode, not using LU variables
|
279 | e9a81214 | Michael Hanselmann | if level in lu.needed_locks: |
280 | e9a81214 | Michael Hanselmann | share_node_alloc = lu.share_locks[locking.LEVEL_NODE_ALLOC] |
281 | e9a81214 | Michael Hanselmann | share_level = lu.share_locks[level] |
282 | e9a81214 | Michael Hanselmann | |
283 | e9a81214 | Michael Hanselmann | if lu.__class__ in _mode_whitelist: |
284 | e9a81214 | Michael Hanselmann | assert share_node_alloc != share_level, \
|
285 | e9a81214 | Michael Hanselmann | "LU is whitelisted to use different modes for node allocation lock"
|
286 | e9a81214 | Michael Hanselmann | else:
|
287 | e9a81214 | Michael Hanselmann | assert bool(share_node_alloc) == bool(share_level), \ |
288 | e9a81214 | Michael Hanselmann | ("Node allocation lock must be acquired using the same mode as nodes"
|
289 | e9a81214 | Michael Hanselmann | " and node resources")
|
290 | e9a81214 | Michael Hanselmann | |
291 | e9a81214 | Michael Hanselmann | if lu.__class__ in _nal_whitelist: |
292 | e9a81214 | Michael Hanselmann | assert not have_nal, \ |
293 | e9a81214 | Michael Hanselmann | "LU is whitelisted for not acquiring the node allocation lock"
|
294 | e9a81214 | Michael Hanselmann | elif lu.needed_locks[level] == locking.ALL_SET or glm.owning_all(level): |
295 | e9a81214 | Michael Hanselmann | assert have_nal, \
|
296 | e9a81214 | Michael Hanselmann | ("Node allocation lock must be used if an LU acquires all nodes"
|
297 | e9a81214 | Michael Hanselmann | " or node resources")
|
298 | e9a81214 | Michael Hanselmann | |
299 | e9a81214 | Michael Hanselmann | |
300 | a8083063 | Iustin Pop | class Processor(object): |
301 | a8083063 | Iustin Pop | """Object which runs OpCodes"""
|
302 | a1a7bc78 | Iustin Pop | DISPATCH_TABLE = _ComputeDispatchTable() |
303 | a8083063 | Iustin Pop | |
304 | dc4bdf73 | Michael Hanselmann | def __init__(self, context, ec_id, enable_locks=True): |
305 | a8083063 | Iustin Pop | """Constructor for Processor
|
306 | a8083063 | Iustin Pop |
|
307 | adfa97e3 | Guido Trotter | @type context: GanetiContext
|
308 | adfa97e3 | Guido Trotter | @param context: global Ganeti context
|
309 | adfa97e3 | Guido Trotter | @type ec_id: string
|
310 | adfa97e3 | Guido Trotter | @param ec_id: execution context identifier
|
311 | adfa97e3 | Guido Trotter |
|
312 | a8083063 | Iustin Pop | """
|
313 | 1c901d13 | Guido Trotter | self.context = context
|
314 | adfa97e3 | Guido Trotter | self._ec_id = ec_id
|
315 | 031a3e57 | Michael Hanselmann | self._cbs = None |
316 | 87b3cb26 | Michael Hanselmann | self.rpc = context.rpc
|
317 | cd46f3b4 | Luca Bigliardi | self.hmclass = HooksMaster
|
318 | dc4bdf73 | Michael Hanselmann | self._enable_locks = enable_locks
|
319 | dc4bdf73 | Michael Hanselmann | |
320 | dc4bdf73 | Michael Hanselmann | def _CheckLocksEnabled(self): |
321 | dc4bdf73 | Michael Hanselmann | """Checks if locking is enabled.
|
322 | dc4bdf73 | Michael Hanselmann |
|
323 | dc4bdf73 | Michael Hanselmann | @raise errors.ProgrammerError: In case locking is not enabled
|
324 | dc4bdf73 | Michael Hanselmann |
|
325 | dc4bdf73 | Michael Hanselmann | """
|
326 | dc4bdf73 | Michael Hanselmann | if not self._enable_locks: |
327 | dc4bdf73 | Michael Hanselmann | raise errors.ProgrammerError("Attempted to use disabled locks") |
328 | a8083063 | Iustin Pop | |
329 | 686d24f0 | Michael Hanselmann | def _AcquireLocks(self, level, names, shared, opportunistic, timeout): |
330 | 211b6132 | Michael Hanselmann | """Acquires locks via the Ganeti lock manager.
|
331 | 211b6132 | Michael Hanselmann |
|
332 | 211b6132 | Michael Hanselmann | @type level: int
|
333 | 211b6132 | Michael Hanselmann | @param level: Lock level
|
334 | 211b6132 | Michael Hanselmann | @type names: list or string
|
335 | 211b6132 | Michael Hanselmann | @param names: Lock names
|
336 | 211b6132 | Michael Hanselmann | @type shared: bool
|
337 | 211b6132 | Michael Hanselmann | @param shared: Whether the locks should be acquired in shared mode
|
338 | 686d24f0 | Michael Hanselmann | @type opportunistic: bool
|
339 | 686d24f0 | Michael Hanselmann | @param opportunistic: Whether to acquire opportunistically
|
340 | 211b6132 | Michael Hanselmann | @type timeout: None or float
|
341 | 211b6132 | Michael Hanselmann | @param timeout: Timeout for acquiring the locks
|
342 | 900df6cd | Michael Hanselmann | @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
|
343 | 900df6cd | Michael Hanselmann | amount of time
|
344 | 211b6132 | Michael Hanselmann |
|
345 | 211b6132 | Michael Hanselmann | """
|
346 | dc4bdf73 | Michael Hanselmann | self._CheckLocksEnabled()
|
347 | dc4bdf73 | Michael Hanselmann | |
348 | acf931b7 | Michael Hanselmann | if self._cbs: |
349 | e4e59de8 | Michael Hanselmann | priority = self._cbs.CurrentPriority()
|
350 | e4e59de8 | Michael Hanselmann | else:
|
351 | e4e59de8 | Michael Hanselmann | priority = None
|
352 | 211b6132 | Michael Hanselmann | |
353 | 211b6132 | Michael Hanselmann | acquired = self.context.glm.acquire(level, names, shared=shared,
|
354 | 686d24f0 | Michael Hanselmann | timeout=timeout, priority=priority, |
355 | 686d24f0 | Michael Hanselmann | opportunistic=opportunistic) |
356 | 211b6132 | Michael Hanselmann | |
357 | 900df6cd | Michael Hanselmann | if acquired is None: |
358 | 900df6cd | Michael Hanselmann | raise LockAcquireTimeout()
|
359 | 900df6cd | Michael Hanselmann | |
360 | 211b6132 | Michael Hanselmann | return acquired
|
361 | 211b6132 | Michael Hanselmann | |
362 | 36c381d7 | Guido Trotter | def _ExecLU(self, lu): |
363 | 36c381d7 | Guido Trotter | """Logical Unit execution sequence.
|
364 | 36c381d7 | Guido Trotter |
|
365 | 36c381d7 | Guido Trotter | """
|
366 | 36c381d7 | Guido Trotter | write_count = self.context.cfg.write_count
|
367 | 36c381d7 | Guido Trotter | lu.CheckPrereq() |
368 | 949dcb1d | Andrea Spadaccini | |
369 | 949dcb1d | Andrea Spadaccini | hm = self.BuildHooksManager(lu)
|
370 | 36c381d7 | Guido Trotter | h_results = hm.RunPhase(constants.HOOKS_PHASE_PRE) |
371 | 36c381d7 | Guido Trotter | lu.HooksCallBack(constants.HOOKS_PHASE_PRE, h_results, |
372 | 7b4c1cb9 | Michael Hanselmann | self.Log, None) |
373 | 20777413 | Iustin Pop | |
374 | 20777413 | Iustin Pop | if getattr(lu.op, "dry_run", False): |
375 | 20777413 | Iustin Pop | # in this mode, no post-hooks are run, and the config is not
|
376 | 20777413 | Iustin Pop | # written (as it might have been modified by another LU, and we
|
377 | 20777413 | Iustin Pop | # shouldn't do writeout on behalf of other threads
|
378 | 20777413 | Iustin Pop | self.LogInfo("dry-run mode requested, not actually executing" |
379 | 20777413 | Iustin Pop | " the operation")
|
380 | 20777413 | Iustin Pop | return lu.dry_run_result
|
381 | 20777413 | Iustin Pop | |
382 | abe362d3 | Michael Hanselmann | if self._cbs: |
383 | abe362d3 | Michael Hanselmann | submit_mj_fn = self._cbs.SubmitManyJobs
|
384 | abe362d3 | Michael Hanselmann | else:
|
385 | abe362d3 | Michael Hanselmann | submit_mj_fn = _FailingSubmitManyJobs |
386 | abe362d3 | Michael Hanselmann | |
387 | 36c381d7 | Guido Trotter | try:
|
388 | abe362d3 | Michael Hanselmann | result = _ProcessResult(submit_mj_fn, lu.op, lu.Exec(self.Log))
|
389 | 36c381d7 | Guido Trotter | h_results = hm.RunPhase(constants.HOOKS_PHASE_POST) |
390 | 36c381d7 | Guido Trotter | result = lu.HooksCallBack(constants.HOOKS_PHASE_POST, h_results, |
391 | 7b4c1cb9 | Michael Hanselmann | self.Log, result)
|
392 | 36c381d7 | Guido Trotter | finally:
|
393 | 36c381d7 | Guido Trotter | # FIXME: This needs locks if not lu_class.REQ_BGL
|
394 | 36c381d7 | Guido Trotter | if write_count != self.context.cfg.write_count: |
395 | 36c381d7 | Guido Trotter | hm.RunConfigUpdate() |
396 | 36c381d7 | Guido Trotter | |
397 | 36c381d7 | Guido Trotter | return result
|
398 | 36c381d7 | Guido Trotter | |
399 | 949dcb1d | Andrea Spadaccini | def BuildHooksManager(self, lu): |
400 | 949dcb1d | Andrea Spadaccini | return self.hmclass.BuildFromLu(lu.rpc.call_hooks_runner, lu) |
401 | 949dcb1d | Andrea Spadaccini | |
402 | e4e59de8 | Michael Hanselmann | def _LockAndExecLU(self, lu, level, calc_timeout): |
403 | 68adfdb2 | Guido Trotter | """Execute a Logical Unit, with the needed locks.
|
404 | 68adfdb2 | Guido Trotter |
|
405 | 68adfdb2 | Guido Trotter | This is a recursive function that starts locking the given level, and
|
406 | 68adfdb2 | Guido Trotter | proceeds up, till there are no more locks to acquire. Then it executes the
|
407 | 68adfdb2 | Guido Trotter | given LU and its opcodes.
|
408 | 68adfdb2 | Guido Trotter |
|
409 | 68adfdb2 | Guido Trotter | """
|
410 | e9a81214 | Michael Hanselmann | glm = self.context.glm
|
411 | ca2a79e1 | Guido Trotter | adding_locks = level in lu.add_locks
|
412 | ca2a79e1 | Guido Trotter | acquiring_locks = level in lu.needed_locks
|
413 | e9a81214 | Michael Hanselmann | |
414 | 8a2941c4 | Guido Trotter | if level not in locking.LEVELS: |
415 | e9a81214 | Michael Hanselmann | _VerifyLocks(lu, glm) |
416 | e9a81214 | Michael Hanselmann | |
417 | 031a3e57 | Michael Hanselmann | if self._cbs: |
418 | 031a3e57 | Michael Hanselmann | self._cbs.NotifyStart()
|
419 | 031a3e57 | Michael Hanselmann | |
420 | b183c4a8 | Iustin Pop | try:
|
421 | b183c4a8 | Iustin Pop | result = self._ExecLU(lu)
|
422 | b183c4a8 | Iustin Pop | except AssertionError, err: |
423 | b183c4a8 | Iustin Pop | # this is a bit ugly, as we don't know from which phase
|
424 | b183c4a8 | Iustin Pop | # (prereq, exec) this comes; but it's better than an exception
|
425 | b183c4a8 | Iustin Pop | # with no information
|
426 | b183c4a8 | Iustin Pop | (_, _, tb) = sys.exc_info() |
427 | b183c4a8 | Iustin Pop | err_info = traceback.format_tb(tb) |
428 | b183c4a8 | Iustin Pop | del tb
|
429 | b183c4a8 | Iustin Pop | logging.exception("Detected AssertionError")
|
430 | b183c4a8 | Iustin Pop | raise errors.OpExecError("Internal assertion error: please report" |
431 | b183c4a8 | Iustin Pop | " this as a bug.\nError message: '%s';"
|
432 | b183c4a8 | Iustin Pop | " location:\n%s" % (str(err), err_info[-1])) |
433 | 407339d0 | Michael Hanselmann | |
434 | ca2a79e1 | Guido Trotter | elif adding_locks and acquiring_locks: |
435 | ca2a79e1 | Guido Trotter | # We could both acquire and add locks at the same level, but for now we
|
436 | ca2a79e1 | Guido Trotter | # don't need this, so we'll avoid the complicated code needed.
|
437 | 407339d0 | Michael Hanselmann | raise NotImplementedError("Can't declare locks to acquire when adding" |
438 | 407339d0 | Michael Hanselmann | " others")
|
439 | 407339d0 | Michael Hanselmann | |
440 | ca2a79e1 | Guido Trotter | elif adding_locks or acquiring_locks: |
441 | dc4bdf73 | Michael Hanselmann | self._CheckLocksEnabled()
|
442 | dc4bdf73 | Michael Hanselmann | |
443 | fb8dcb62 | Guido Trotter | lu.DeclareLocks(level) |
444 | 3977a4c1 | Guido Trotter | share = lu.share_locks[level] |
445 | 686d24f0 | Michael Hanselmann | opportunistic = lu.opportunistic_locks[level] |
446 | 407339d0 | Michael Hanselmann | |
447 | 68adfdb2 | Guido Trotter | try:
|
448 | 407339d0 | Michael Hanselmann | assert adding_locks ^ acquiring_locks, \
|
449 | 407339d0 | Michael Hanselmann | "Locks must be either added or acquired"
|
450 | 407339d0 | Michael Hanselmann | |
451 | 407339d0 | Michael Hanselmann | if acquiring_locks:
|
452 | 407339d0 | Michael Hanselmann | # Acquiring locks
|
453 | 407339d0 | Michael Hanselmann | needed_locks = lu.needed_locks[level] |
454 | 407339d0 | Michael Hanselmann | |
455 | 686d24f0 | Michael Hanselmann | self._AcquireLocks(level, needed_locks, share, opportunistic,
|
456 | e4e59de8 | Michael Hanselmann | calc_timeout()) |
457 | 407339d0 | Michael Hanselmann | else:
|
458 | 407339d0 | Michael Hanselmann | # Adding locks
|
459 | 407339d0 | Michael Hanselmann | add_locks = lu.add_locks[level] |
460 | 407339d0 | Michael Hanselmann | lu.remove_locks[level] = add_locks |
461 | 407339d0 | Michael Hanselmann | |
462 | 407339d0 | Michael Hanselmann | try:
|
463 | e9a81214 | Michael Hanselmann | glm.add(level, add_locks, acquired=1, shared=share)
|
464 | 407339d0 | Michael Hanselmann | except errors.LockError:
|
465 | a9d40c65 | Iustin Pop | logging.exception("Detected lock error in level %s for locks"
|
466 | a9d40c65 | Iustin Pop | " %s, shared=%s", level, add_locks, share)
|
467 | 407339d0 | Michael Hanselmann | raise errors.OpPrereqError(
|
468 | a9d40c65 | Iustin Pop | "Couldn't add locks (%s), most likely because of another"
|
469 | a9d40c65 | Iustin Pop | " job who added them first" % add_locks,
|
470 | a9d40c65 | Iustin Pop | errors.ECODE_NOTUNIQUE) |
471 | 407339d0 | Michael Hanselmann | |
472 | ca2a79e1 | Guido Trotter | try:
|
473 | e4e59de8 | Michael Hanselmann | result = self._LockAndExecLU(lu, level + 1, calc_timeout) |
474 | ca2a79e1 | Guido Trotter | finally:
|
475 | ca2a79e1 | Guido Trotter | if level in lu.remove_locks: |
476 | e9a81214 | Michael Hanselmann | glm.remove(level, lu.remove_locks[level]) |
477 | 68adfdb2 | Guido Trotter | finally:
|
478 | e9a81214 | Michael Hanselmann | if glm.is_owned(level):
|
479 | e9a81214 | Michael Hanselmann | glm.release(level) |
480 | 407339d0 | Michael Hanselmann | |
481 | 68adfdb2 | Guido Trotter | else:
|
482 | e4e59de8 | Michael Hanselmann | result = self._LockAndExecLU(lu, level + 1, calc_timeout) |
483 | 68adfdb2 | Guido Trotter | |
484 | 68adfdb2 | Guido Trotter | return result
|
485 | 68adfdb2 | Guido Trotter | |
486 | e4e59de8 | Michael Hanselmann | def ExecOpCode(self, op, cbs, timeout=None): |
487 | a8083063 | Iustin Pop | """Execute an opcode.
|
488 | a8083063 | Iustin Pop |
|
489 | e92376d7 | Iustin Pop | @type op: an OpCode instance
|
490 | e92376d7 | Iustin Pop | @param op: the opcode to be executed
|
491 | 031a3e57 | Michael Hanselmann | @type cbs: L{OpExecCbBase}
|
492 | 031a3e57 | Michael Hanselmann | @param cbs: Runtime callbacks
|
493 | 831bbbc1 | Michael Hanselmann | @type timeout: float or None
|
494 | 831bbbc1 | Michael Hanselmann | @param timeout: Maximum time to acquire all locks, None for no timeout
|
495 | 831bbbc1 | Michael Hanselmann | @raise LockAcquireTimeout: In case locks couldn't be acquired in specified
|
496 | 831bbbc1 | Michael Hanselmann | amount of time
|
497 | a8083063 | Iustin Pop |
|
498 | a8083063 | Iustin Pop | """
|
499 | a8083063 | Iustin Pop | if not isinstance(op, opcodes.OpCode): |
500 | 3ecf6786 | Iustin Pop | raise errors.ProgrammerError("Non-opcode instance passed" |
501 | c7bb3095 | Michael Hanselmann | " to ExecOpcode (%s)" % type(op)) |
502 | a8083063 | Iustin Pop | |
503 | 831bbbc1 | Michael Hanselmann | lu_class = self.DISPATCH_TABLE.get(op.__class__, None) |
504 | 831bbbc1 | Michael Hanselmann | if lu_class is None: |
505 | 831bbbc1 | Michael Hanselmann | raise errors.OpCodeUnknown("Unknown opcode") |
506 | 831bbbc1 | Michael Hanselmann | |
507 | 831bbbc1 | Michael Hanselmann | if timeout is None: |
508 | 831bbbc1 | Michael Hanselmann | calc_timeout = lambda: None |
509 | 831bbbc1 | Michael Hanselmann | else:
|
510 | 557838c1 | René Nussbaumer | calc_timeout = utils.RunningTimeout(timeout, False).Remaining
|
511 | 831bbbc1 | Michael Hanselmann | |
512 | 031a3e57 | Michael Hanselmann | self._cbs = cbs
|
513 | fe482621 | Iustin Pop | try:
|
514 | dc4bdf73 | Michael Hanselmann | if self._enable_locks: |
515 | dc4bdf73 | Michael Hanselmann | # Acquire the Big Ganeti Lock exclusively if this LU requires it,
|
516 | dc4bdf73 | Michael Hanselmann | # and in a shared fashion otherwise (to prevent concurrent run with
|
517 | dc4bdf73 | Michael Hanselmann | # an exclusive LU.
|
518 | dc4bdf73 | Michael Hanselmann | self._AcquireLocks(locking.LEVEL_CLUSTER, locking.BGL,
|
519 | 686d24f0 | Michael Hanselmann | not lu_class.REQ_BGL, False, calc_timeout()) |
520 | dc4bdf73 | Michael Hanselmann | elif lu_class.REQ_BGL:
|
521 | dc4bdf73 | Michael Hanselmann | raise errors.ProgrammerError("Opcode '%s' requires BGL, but locks are" |
522 | dc4bdf73 | Michael Hanselmann | " disabled" % op.OP_ID)
|
523 | dc4bdf73 | Michael Hanselmann | |
524 | 831bbbc1 | Michael Hanselmann | try:
|
525 | 831bbbc1 | Michael Hanselmann | lu = lu_class(self, op, self.context, self.rpc) |
526 | 831bbbc1 | Michael Hanselmann | lu.ExpandNames() |
527 | 831bbbc1 | Michael Hanselmann | assert lu.needed_locks is not None, "needed_locks not set by LU" |
528 | 407339d0 | Michael Hanselmann | |
529 | 407339d0 | Michael Hanselmann | try:
|
530 | 46cde471 | Michael Hanselmann | result = self._LockAndExecLU(lu, locking.LEVEL_CLUSTER + 1, |
531 | 46cde471 | Michael Hanselmann | calc_timeout) |
532 | 831bbbc1 | Michael Hanselmann | finally:
|
533 | 831bbbc1 | Michael Hanselmann | if self._ec_id: |
534 | 831bbbc1 | Michael Hanselmann | self.context.cfg.DropECReservations(self._ec_id) |
535 | 831bbbc1 | Michael Hanselmann | finally:
|
536 | dc4bdf73 | Michael Hanselmann | # Release BGL if owned
|
537 | dc4bdf73 | Michael Hanselmann | if self.context.glm.is_owned(locking.LEVEL_CLUSTER): |
538 | dc4bdf73 | Michael Hanselmann | assert self._enable_locks |
539 | dc4bdf73 | Michael Hanselmann | self.context.glm.release(locking.LEVEL_CLUSTER)
|
540 | 04864530 | Guido Trotter | finally:
|
541 | 031a3e57 | Michael Hanselmann | self._cbs = None |
542 | 6a4aa7c1 | Iustin Pop | |
543 | 1ce03fb1 | Michael Hanselmann | resultcheck_fn = op.OP_RESULT |
544 | 1ce03fb1 | Michael Hanselmann | if not (resultcheck_fn is None or resultcheck_fn(result)): |
545 | 1ce03fb1 | Michael Hanselmann | logging.error("Expected opcode result matching %s, got %s",
|
546 | 1ce03fb1 | Michael Hanselmann | resultcheck_fn, result) |
547 | 80b207df | Iustin Pop | if not getattr(op, "dry_run", False): |
548 | 80b207df | Iustin Pop | # FIXME: LUs should still behave in dry_run mode, or
|
549 | 80b207df | Iustin Pop | # alternately we should have OP_DRYRUN_RESULT; in the
|
550 | 80b207df | Iustin Pop | # meantime, we simply skip the OP_RESULT check in dry-run mode
|
551 | 80b207df | Iustin Pop | raise errors.OpResultError("Opcode result does not match %s: %s" % |
552 | 80b207df | Iustin Pop | (resultcheck_fn, utils.Truncate(result, 80)))
|
553 | 1ce03fb1 | Michael Hanselmann | |
554 | 1ce03fb1 | Michael Hanselmann | return result
|
555 | 1ce03fb1 | Michael Hanselmann | |
556 | 7b4c1cb9 | Michael Hanselmann | def Log(self, *args): |
557 | 031a3e57 | Michael Hanselmann | """Forward call to feedback callback function.
|
558 | 031a3e57 | Michael Hanselmann |
|
559 | 031a3e57 | Michael Hanselmann | """
|
560 | 031a3e57 | Michael Hanselmann | if self._cbs: |
561 | 031a3e57 | Michael Hanselmann | self._cbs.Feedback(*args)
|
562 | 031a3e57 | Michael Hanselmann | |
563 | 0fbbf897 | Iustin Pop | def LogStep(self, current, total, message): |
564 | 0fbbf897 | Iustin Pop | """Log a change in LU execution progress.
|
565 | 0fbbf897 | Iustin Pop |
|
566 | 0fbbf897 | Iustin Pop | """
|
567 | a5eb7789 | Iustin Pop | logging.debug("Step %d/%d %s", current, total, message)
|
568 | 7b4c1cb9 | Michael Hanselmann | self.Log("STEP %d/%d %s" % (current, total, message)) |
569 | 0fbbf897 | Iustin Pop | |
570 | c0088fb9 | Iustin Pop | def LogWarning(self, message, *args, **kwargs): |
571 | 0fbbf897 | Iustin Pop | """Log a warning to the logs and the user.
|
572 | 0fbbf897 | Iustin Pop |
|
573 | c0088fb9 | Iustin Pop | The optional keyword argument is 'hint' and can be used to show a
|
574 | c0088fb9 | Iustin Pop | hint to the user (presumably related to the warning). If the
|
575 | c0088fb9 | Iustin Pop | message is empty, it will not be printed at all, allowing one to
|
576 | c0088fb9 | Iustin Pop | show only a hint.
|
577 | 0fbbf897 | Iustin Pop |
|
578 | c0088fb9 | Iustin Pop | """
|
579 | c0088fb9 | Iustin Pop | assert not kwargs or (len(kwargs) == 1 and "hint" in kwargs), \ |
580 | c0088fb9 | Iustin Pop | "Invalid keyword arguments for LogWarning (%s)" % str(kwargs) |
581 | c0088fb9 | Iustin Pop | if args:
|
582 | c0088fb9 | Iustin Pop | message = message % tuple(args)
|
583 | c0088fb9 | Iustin Pop | if message:
|
584 | c0088fb9 | Iustin Pop | logging.warning(message) |
585 | 7b4c1cb9 | Michael Hanselmann | self.Log(" - WARNING: %s" % message) |
586 | c0088fb9 | Iustin Pop | if "hint" in kwargs: |
587 | 7b4c1cb9 | Michael Hanselmann | self.Log(" Hint: %s" % kwargs["hint"]) |
588 | c0088fb9 | Iustin Pop | |
589 | c0088fb9 | Iustin Pop | def LogInfo(self, message, *args): |
590 | 0fbbf897 | Iustin Pop | """Log an informational message to the logs and the user.
|
591 | 0fbbf897 | Iustin Pop |
|
592 | 0fbbf897 | Iustin Pop | """
|
593 | c0088fb9 | Iustin Pop | if args:
|
594 | c0088fb9 | Iustin Pop | message = message % tuple(args)
|
595 | a5eb7789 | Iustin Pop | logging.info(message) |
596 | 7b4c1cb9 | Michael Hanselmann | self.Log(" - INFO: %s" % message) |
597 | 0fbbf897 | Iustin Pop | |
598 | adfa97e3 | Guido Trotter | def GetECId(self): |
599 | 3ae70d76 | Michael Hanselmann | """Returns the current execution context ID.
|
600 | 3ae70d76 | Michael Hanselmann |
|
601 | 3ae70d76 | Michael Hanselmann | """
|
602 | adfa97e3 | Guido Trotter | if not self._ec_id: |
603 | 3ae70d76 | Michael Hanselmann | raise errors.ProgrammerError("Tried to use execution context id when" |
604 | 3ae70d76 | Michael Hanselmann | " not set")
|
605 | adfa97e3 | Guido Trotter | return self._ec_id |
606 | adfa97e3 | Guido Trotter | |
607 | a8083063 | Iustin Pop | |
608 | a8083063 | Iustin Pop | class HooksMaster(object): |
609 | 949dcb1d | Andrea Spadaccini | def __init__(self, opcode, hooks_path, nodes, hooks_execution_fn, |
610 | 5ae4945a | Iustin Pop | hooks_results_adapt_fn, build_env_fn, log_fn, htype=None,
|
611 | 5ae4945a | Iustin Pop | cluster_name=None, master_name=None): |
612 | 949dcb1d | Andrea Spadaccini | """Base class for hooks masters.
|
613 | 949dcb1d | Andrea Spadaccini |
|
614 | 949dcb1d | Andrea Spadaccini | This class invokes the execution of hooks according to the behaviour
|
615 | 949dcb1d | Andrea Spadaccini | specified by its parameters.
|
616 | 949dcb1d | Andrea Spadaccini |
|
617 | 949dcb1d | Andrea Spadaccini | @type opcode: string
|
618 | 949dcb1d | Andrea Spadaccini | @param opcode: opcode of the operation to which the hooks are tied
|
619 | 949dcb1d | Andrea Spadaccini | @type hooks_path: string
|
620 | 949dcb1d | Andrea Spadaccini | @param hooks_path: prefix of the hooks directories
|
621 | 949dcb1d | Andrea Spadaccini | @type nodes: 2-tuple of lists
|
622 | 949dcb1d | Andrea Spadaccini | @param nodes: 2-tuple of lists containing nodes on which pre-hooks must be
|
623 | 949dcb1d | Andrea Spadaccini | run and nodes on which post-hooks must be run
|
624 | 949dcb1d | Andrea Spadaccini | @type hooks_execution_fn: function that accepts the following parameters:
|
625 | 949dcb1d | Andrea Spadaccini | (node_list, hooks_path, phase, environment)
|
626 | 949dcb1d | Andrea Spadaccini | @param hooks_execution_fn: function that will execute the hooks; can be
|
627 | 949dcb1d | Andrea Spadaccini | None, indicating that no conversion is necessary.
|
628 | 949dcb1d | Andrea Spadaccini | @type hooks_results_adapt_fn: function
|
629 | 949dcb1d | Andrea Spadaccini | @param hooks_results_adapt_fn: function that will adapt the return value of
|
630 | 949dcb1d | Andrea Spadaccini | hooks_execution_fn to the format expected by RunPhase
|
631 | 949dcb1d | Andrea Spadaccini | @type build_env_fn: function that returns a dictionary having strings as
|
632 | 949dcb1d | Andrea Spadaccini | keys
|
633 | 949dcb1d | Andrea Spadaccini | @param build_env_fn: function that builds the environment for the hooks
|
634 | 949dcb1d | Andrea Spadaccini | @type log_fn: function that accepts a string
|
635 | 949dcb1d | Andrea Spadaccini | @param log_fn: logging function
|
636 | 949dcb1d | Andrea Spadaccini | @type htype: string or None
|
637 | 949dcb1d | Andrea Spadaccini | @param htype: None or one of L{constants.HTYPE_CLUSTER},
|
638 | 949dcb1d | Andrea Spadaccini | L{constants.HTYPE_NODE}, L{constants.HTYPE_INSTANCE}
|
639 | 949dcb1d | Andrea Spadaccini | @type cluster_name: string
|
640 | 949dcb1d | Andrea Spadaccini | @param cluster_name: name of the cluster
|
641 | 949dcb1d | Andrea Spadaccini | @type master_name: string
|
642 | 949dcb1d | Andrea Spadaccini | @param master_name: name of the master
|
643 | a8083063 | Iustin Pop |
|
644 | 949dcb1d | Andrea Spadaccini | """
|
645 | 949dcb1d | Andrea Spadaccini | self.opcode = opcode
|
646 | 949dcb1d | Andrea Spadaccini | self.hooks_path = hooks_path
|
647 | 949dcb1d | Andrea Spadaccini | self.hooks_execution_fn = hooks_execution_fn
|
648 | 949dcb1d | Andrea Spadaccini | self.hooks_results_adapt_fn = hooks_results_adapt_fn
|
649 | 949dcb1d | Andrea Spadaccini | self.build_env_fn = build_env_fn
|
650 | 949dcb1d | Andrea Spadaccini | self.log_fn = log_fn
|
651 | 949dcb1d | Andrea Spadaccini | self.htype = htype
|
652 | 949dcb1d | Andrea Spadaccini | self.cluster_name = cluster_name
|
653 | 949dcb1d | Andrea Spadaccini | self.master_name = master_name
|
654 | a8083063 | Iustin Pop | |
655 | 07e0896f | Michael Hanselmann | self.pre_env = self._BuildEnv(constants.HOOKS_PHASE_PRE) |
656 | 07e0896f | Michael Hanselmann | (self.pre_nodes, self.post_nodes) = nodes |
657 | a8083063 | Iustin Pop | |
658 | dd7f6776 | Michael Hanselmann | def _BuildEnv(self, phase): |
659 | a8083063 | Iustin Pop | """Compute the environment and the target nodes.
|
660 | a8083063 | Iustin Pop |
|
661 | a8083063 | Iustin Pop | Based on the opcode and the current node list, this builds the
|
662 | a8083063 | Iustin Pop | environment for the hooks and the target node list for the run.
|
663 | a8083063 | Iustin Pop |
|
664 | a8083063 | Iustin Pop | """
|
665 | dd7f6776 | Michael Hanselmann | if phase == constants.HOOKS_PHASE_PRE:
|
666 | dd7f6776 | Michael Hanselmann | prefix = "GANETI_"
|
667 | dd7f6776 | Michael Hanselmann | elif phase == constants.HOOKS_PHASE_POST:
|
668 | dd7f6776 | Michael Hanselmann | prefix = "GANETI_POST_"
|
669 | dd7f6776 | Michael Hanselmann | else:
|
670 | dd7f6776 | Michael Hanselmann | raise AssertionError("Unknown phase '%s'" % phase) |
671 | dd7f6776 | Michael Hanselmann | |
672 | dd7f6776 | Michael Hanselmann | env = {} |
673 | a8083063 | Iustin Pop | |
674 | 949dcb1d | Andrea Spadaccini | if self.hooks_path is not None: |
675 | 949dcb1d | Andrea Spadaccini | phase_env = self.build_env_fn()
|
676 | 949dcb1d | Andrea Spadaccini | if phase_env:
|
677 | 949dcb1d | Andrea Spadaccini | assert not compat.any(key.upper().startswith(prefix) |
678 | 949dcb1d | Andrea Spadaccini | for key in phase_env) |
679 | dd7f6776 | Michael Hanselmann | env.update(("%s%s" % (prefix, key), value)
|
680 | 949dcb1d | Andrea Spadaccini | for (key, value) in phase_env.items()) |
681 | a8083063 | Iustin Pop | |
682 | dd7f6776 | Michael Hanselmann | if phase == constants.HOOKS_PHASE_PRE:
|
683 | dd7f6776 | Michael Hanselmann | assert compat.all((key.startswith("GANETI_") and |
684 | dd7f6776 | Michael Hanselmann | not key.startswith("GANETI_POST_")) |
685 | dd7f6776 | Michael Hanselmann | for key in env) |
686 | dd7f6776 | Michael Hanselmann | |
687 | dd7f6776 | Michael Hanselmann | elif phase == constants.HOOKS_PHASE_POST:
|
688 | dd7f6776 | Michael Hanselmann | assert compat.all(key.startswith("GANETI_POST_") for key in env) |
689 | 07e0896f | Michael Hanselmann | assert isinstance(self.pre_env, dict) |
690 | dd7f6776 | Michael Hanselmann | |
691 | 07e0896f | Michael Hanselmann | # Merge with pre-phase environment
|
692 | 07e0896f | Michael Hanselmann | assert not compat.any(key.startswith("GANETI_POST_") |
693 | 07e0896f | Michael Hanselmann | for key in self.pre_env) |
694 | 07e0896f | Michael Hanselmann | env.update(self.pre_env)
|
695 | dd7f6776 | Michael Hanselmann | else:
|
696 | dd7f6776 | Michael Hanselmann | raise AssertionError("Unknown phase '%s'" % phase) |
697 | dd7f6776 | Michael Hanselmann | |
698 | 07e0896f | Michael Hanselmann | return env
|
699 | 4167825b | Iustin Pop | |
700 | dd7f6776 | Michael Hanselmann | def _RunWrapper(self, node_list, hpath, phase, phase_env): |
701 | 4167825b | Iustin Pop | """Simple wrapper over self.callfn.
|
702 | 4167825b | Iustin Pop |
|
703 | 949dcb1d | Andrea Spadaccini | This method fixes the environment before executing the hooks.
|
704 | 4167825b | Iustin Pop |
|
705 | 4167825b | Iustin Pop | """
|
706 | dd7f6776 | Michael Hanselmann | env = { |
707 | fe5ca2bb | Andrea Spadaccini | "PATH": constants.HOOKS_PATH,
|
708 | dd7f6776 | Michael Hanselmann | "GANETI_HOOKS_VERSION": constants.HOOKS_VERSION,
|
709 | 949dcb1d | Andrea Spadaccini | "GANETI_OP_CODE": self.opcode, |
710 | b3a6f851 | Michael Hanselmann | "GANETI_DATA_DIR": pathutils.DATA_DIR,
|
711 | dd7f6776 | Michael Hanselmann | "GANETI_HOOKS_PHASE": phase,
|
712 | dd7f6776 | Michael Hanselmann | "GANETI_HOOKS_PATH": hpath,
|
713 | dd7f6776 | Michael Hanselmann | } |
714 | dd7f6776 | Michael Hanselmann | |
715 | 949dcb1d | Andrea Spadaccini | if self.htype: |
716 | 949dcb1d | Andrea Spadaccini | env["GANETI_OBJECT_TYPE"] = self.htype |
717 | 949dcb1d | Andrea Spadaccini | |
718 | 949dcb1d | Andrea Spadaccini | if self.cluster_name is not None: |
719 | 949dcb1d | Andrea Spadaccini | env["GANETI_CLUSTER"] = self.cluster_name |
720 | 07e0896f | Michael Hanselmann | |
721 | 949dcb1d | Andrea Spadaccini | if self.master_name is not None: |
722 | 949dcb1d | Andrea Spadaccini | env["GANETI_MASTER"] = self.master_name |
723 | dd7f6776 | Michael Hanselmann | |
724 | dd7f6776 | Michael Hanselmann | if phase_env:
|
725 | af5af644 | Andrea Spadaccini | env = utils.algo.JoinDisjointDicts(env, phase_env) |
726 | a8083063 | Iustin Pop | |
727 | dd7f6776 | Michael Hanselmann | # Convert everything to strings
|
728 | 4167825b | Iustin Pop | env = dict([(str(key), str(val)) for key, val in env.iteritems()]) |
729 | a8083063 | Iustin Pop | |
730 | dd7f6776 | Michael Hanselmann | assert compat.all(key == "PATH" or key.startswith("GANETI_") |
731 | ebc75510 | Michael Hanselmann | for key in env) |
732 | ebc75510 | Michael Hanselmann | |
733 | 949dcb1d | Andrea Spadaccini | return self.hooks_execution_fn(node_list, hpath, phase, env) |
734 | a8083063 | Iustin Pop | |
735 | 17e82923 | Luca Bigliardi | def RunPhase(self, phase, nodes=None): |
736 | a8083063 | Iustin Pop | """Run all the scripts for a phase.
|
737 | a8083063 | Iustin Pop |
|
738 | a8083063 | Iustin Pop | This is the main function of the HookMaster.
|
739 | 949dcb1d | Andrea Spadaccini | It executes self.hooks_execution_fn, and after running
|
740 | 949dcb1d | Andrea Spadaccini | self.hooks_results_adapt_fn on its results it expects them to be in the form
|
741 | 949dcb1d | Andrea Spadaccini | {node_name: (fail_msg, [(script, result, output), ...]}).
|
742 | a8083063 | Iustin Pop |
|
743 | 8dca23a3 | Iustin Pop | @param phase: one of L{constants.HOOKS_PHASE_POST} or
|
744 | 8dca23a3 | Iustin Pop | L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
|
745 | 17e82923 | Luca Bigliardi | @param nodes: overrides the predefined list of nodes for the given phase
|
746 | 8dca23a3 | Iustin Pop | @return: the processed results of the hooks multi-node rpc call
|
747 | 8dca23a3 | Iustin Pop | @raise errors.HooksFailure: on communication failure to the nodes
|
748 | 6ef2dc74 | Luca Bigliardi | @raise errors.HooksAbort: on failure of one of the hooks
|
749 | b07a6922 | Guido Trotter |
|
750 | a8083063 | Iustin Pop | """
|
751 | 07e0896f | Michael Hanselmann | if phase == constants.HOOKS_PHASE_PRE:
|
752 | 07e0896f | Michael Hanselmann | if nodes is None: |
753 | 07e0896f | Michael Hanselmann | nodes = self.pre_nodes
|
754 | 07e0896f | Michael Hanselmann | env = self.pre_env
|
755 | 07e0896f | Michael Hanselmann | elif phase == constants.HOOKS_PHASE_POST:
|
756 | 07e0896f | Michael Hanselmann | if nodes is None: |
757 | b423c513 | Michael Hanselmann | nodes = self.post_nodes
|
758 | 07e0896f | Michael Hanselmann | env = self._BuildEnv(phase)
|
759 | 07e0896f | Michael Hanselmann | else:
|
760 | 07e0896f | Michael Hanselmann | raise AssertionError("Unknown phase '%s'" % phase) |
761 | 0306ff62 | Michael Hanselmann | |
762 | 0306ff62 | Michael Hanselmann | if not nodes: |
763 | 9a395a76 | Iustin Pop | # empty node list, we should not attempt to run this as either
|
764 | 9a395a76 | Iustin Pop | # we're in the cluster init phase and the rpc client part can't
|
765 | 9a395a76 | Iustin Pop | # even attempt to run, or this LU doesn't do hooks at all
|
766 | a8083063 | Iustin Pop | return
|
767 | 0306ff62 | Michael Hanselmann | |
768 | 949dcb1d | Andrea Spadaccini | results = self._RunWrapper(nodes, self.hooks_path, phase, env) |
769 | 8c4b9364 | Luca Bigliardi | if not results: |
770 | 8c4b9364 | Luca Bigliardi | msg = "Communication Failure"
|
771 | 8c4b9364 | Luca Bigliardi | if phase == constants.HOOKS_PHASE_PRE:
|
772 | 8c4b9364 | Luca Bigliardi | raise errors.HooksFailure(msg)
|
773 | 8c4b9364 | Luca Bigliardi | else:
|
774 | 949dcb1d | Andrea Spadaccini | self.log_fn(msg)
|
775 | 640b961e | Luca Bigliardi | return results
|
776 | 0306ff62 | Michael Hanselmann | |
777 | 949dcb1d | Andrea Spadaccini | converted_res = results |
778 | 949dcb1d | Andrea Spadaccini | if self.hooks_results_adapt_fn: |
779 | 949dcb1d | Andrea Spadaccini | converted_res = self.hooks_results_adapt_fn(results)
|
780 | 949dcb1d | Andrea Spadaccini | |
781 | 0306ff62 | Michael Hanselmann | errs = [] |
782 | 949dcb1d | Andrea Spadaccini | for node_name, (fail_msg, offline, hooks_results) in converted_res.items(): |
783 | 949dcb1d | Andrea Spadaccini | if offline:
|
784 | 8c4b9364 | Luca Bigliardi | continue
|
785 | 0306ff62 | Michael Hanselmann | |
786 | 949dcb1d | Andrea Spadaccini | if fail_msg:
|
787 | 949dcb1d | Andrea Spadaccini | self.log_fn("Communication failure to node %s: %s", node_name, fail_msg) |
788 | 8c4b9364 | Luca Bigliardi | continue
|
789 | 0306ff62 | Michael Hanselmann | |
790 | 949dcb1d | Andrea Spadaccini | for script, hkr, output in hooks_results: |
791 | 8c4b9364 | Luca Bigliardi | if hkr == constants.HKR_FAIL:
|
792 | 8c4b9364 | Luca Bigliardi | if phase == constants.HOOKS_PHASE_PRE:
|
793 | a8083063 | Iustin Pop | errs.append((node_name, script, output)) |
794 | 8c4b9364 | Luca Bigliardi | else:
|
795 | 8c4b9364 | Luca Bigliardi | if not output: |
796 | 640b961e | Luca Bigliardi | output = "(no output)"
|
797 | 949dcb1d | Andrea Spadaccini | self.log_fn("On %s script %s failed, output: %s" % |
798 | 949dcb1d | Andrea Spadaccini | (node_name, script, output)) |
799 | 0306ff62 | Michael Hanselmann | |
800 | 8c4b9364 | Luca Bigliardi | if errs and phase == constants.HOOKS_PHASE_PRE: |
801 | 8c4b9364 | Luca Bigliardi | raise errors.HooksAbort(errs)
|
802 | 0306ff62 | Michael Hanselmann | |
803 | b07a6922 | Guido Trotter | return results
|
804 | 6a4aa7c1 | Iustin Pop | |
805 | 6a4aa7c1 | Iustin Pop | def RunConfigUpdate(self): |
806 | 6a4aa7c1 | Iustin Pop | """Run the special configuration update hook
|
807 | 6a4aa7c1 | Iustin Pop |
|
808 | 6a4aa7c1 | Iustin Pop | This is a special hook that runs only on the master after each
|
809 | 6a4aa7c1 | Iustin Pop | top-level LI if the configuration has been updated.
|
810 | 6a4aa7c1 | Iustin Pop |
|
811 | 6a4aa7c1 | Iustin Pop | """
|
812 | 6a4aa7c1 | Iustin Pop | phase = constants.HOOKS_PHASE_POST |
813 | 6a4aa7c1 | Iustin Pop | hpath = constants.HOOKS_NAME_CFGUPDATE |
814 | 949dcb1d | Andrea Spadaccini | nodes = [self.master_name]
|
815 | dd7f6776 | Michael Hanselmann | self._RunWrapper(nodes, hpath, phase, self.pre_env) |
816 | 949dcb1d | Andrea Spadaccini | |
817 | 949dcb1d | Andrea Spadaccini | @staticmethod
|
818 | 949dcb1d | Andrea Spadaccini | def BuildFromLu(hooks_execution_fn, lu): |
819 | 949dcb1d | Andrea Spadaccini | if lu.HPATH is None: |
820 | 949dcb1d | Andrea Spadaccini | nodes = (None, None) |
821 | 949dcb1d | Andrea Spadaccini | else:
|
822 | 949dcb1d | Andrea Spadaccini | nodes = map(frozenset, lu.BuildHooksNodes()) |
823 | 949dcb1d | Andrea Spadaccini | |
824 | 949dcb1d | Andrea Spadaccini | master_name = cluster_name = None
|
825 | 949dcb1d | Andrea Spadaccini | if lu.cfg:
|
826 | 949dcb1d | Andrea Spadaccini | master_name = lu.cfg.GetMasterNode() |
827 | 949dcb1d | Andrea Spadaccini | cluster_name = lu.cfg.GetClusterName() |
828 | 949dcb1d | Andrea Spadaccini | |
829 | 949dcb1d | Andrea Spadaccini | return HooksMaster(lu.op.OP_ID, lu.HPATH, nodes, hooks_execution_fn,
|
830 | 949dcb1d | Andrea Spadaccini | _RpcResultsToHooksResults, lu.BuildHooksEnv, |
831 | 949dcb1d | Andrea Spadaccini | lu.LogWarning, lu.HTYPE, cluster_name, master_name) |