root / lib / cmdlib / instance_migration.py @ 84ad6b78
History | View | Annotate | Download (37.5 kB)
1 | 87e25be1 | Thomas Thrainer | #
|
---|---|---|---|
2 | 87e25be1 | Thomas Thrainer | #
|
3 | 87e25be1 | Thomas Thrainer | |
4 | 87e25be1 | Thomas Thrainer | # Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013 Google Inc.
|
5 | 87e25be1 | Thomas Thrainer | #
|
6 | 87e25be1 | Thomas Thrainer | # This program is free software; you can redistribute it and/or modify
|
7 | 87e25be1 | Thomas Thrainer | # it under the terms of the GNU General Public License as published by
|
8 | 87e25be1 | Thomas Thrainer | # the Free Software Foundation; either version 2 of the License, or
|
9 | 87e25be1 | Thomas Thrainer | # (at your option) any later version.
|
10 | 87e25be1 | Thomas Thrainer | #
|
11 | 87e25be1 | Thomas Thrainer | # This program is distributed in the hope that it will be useful, but
|
12 | 87e25be1 | Thomas Thrainer | # WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 | 87e25be1 | Thomas Thrainer | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
14 | 87e25be1 | Thomas Thrainer | # General Public License for more details.
|
15 | 87e25be1 | Thomas Thrainer | #
|
16 | 87e25be1 | Thomas Thrainer | # You should have received a copy of the GNU General Public License
|
17 | 87e25be1 | Thomas Thrainer | # along with this program; if not, write to the Free Software
|
18 | 87e25be1 | Thomas Thrainer | # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
|
19 | 87e25be1 | Thomas Thrainer | # 02110-1301, USA.
|
20 | 87e25be1 | Thomas Thrainer | |
21 | 87e25be1 | Thomas Thrainer | |
22 | 87e25be1 | Thomas Thrainer | """Logical units dealing with instance migration an failover."""
|
23 | 87e25be1 | Thomas Thrainer | |
24 | 87e25be1 | Thomas Thrainer | import logging |
25 | 87e25be1 | Thomas Thrainer | import time |
26 | 87e25be1 | Thomas Thrainer | |
27 | 87e25be1 | Thomas Thrainer | from ganeti import constants |
28 | 87e25be1 | Thomas Thrainer | from ganeti import errors |
29 | 87e25be1 | Thomas Thrainer | from ganeti import locking |
30 | 87e25be1 | Thomas Thrainer | from ganeti.masterd import iallocator |
31 | 87e25be1 | Thomas Thrainer | from ganeti import utils |
32 | 87e25be1 | Thomas Thrainer | from ganeti.cmdlib.base import LogicalUnit, Tasklet |
33 | da4a52a3 | Thomas Thrainer | from ganeti.cmdlib.common import ExpandInstanceUuidAndName, \ |
34 | 1c3231aa | Thomas Thrainer | CheckIAllocatorOrNode, ExpandNodeUuidAndName |
35 | 5eacbcae | Thomas Thrainer | from ganeti.cmdlib.instance_storage import CheckDiskConsistency, \ |
36 | 5eacbcae | Thomas Thrainer | ExpandCheckDisks, ShutdownInstanceDisks, AssembleInstanceDisks |
37 | 5eacbcae | Thomas Thrainer | from ganeti.cmdlib.instance_utils import BuildInstanceHookEnvByObject, \ |
38 | 5eacbcae | Thomas Thrainer | CheckTargetNodeIPolicy, ReleaseLocks, CheckNodeNotDrained, \ |
39 | 5eacbcae | Thomas Thrainer | CopyLockList, CheckNodeFreeMemory, CheckInstanceBridgesExist |
40 | 87e25be1 | Thomas Thrainer | |
41 | 87e25be1 | Thomas Thrainer | import ganeti.masterd.instance |
42 | 87e25be1 | Thomas Thrainer | |
43 | 87e25be1 | Thomas Thrainer | |
44 | 87e25be1 | Thomas Thrainer | def _ExpandNamesForMigration(lu): |
45 | 87e25be1 | Thomas Thrainer | """Expands names for use with L{TLMigrateInstance}.
|
46 | 87e25be1 | Thomas Thrainer |
|
47 | 87e25be1 | Thomas Thrainer | @type lu: L{LogicalUnit}
|
48 | 87e25be1 | Thomas Thrainer |
|
49 | 87e25be1 | Thomas Thrainer | """
|
50 | 87e25be1 | Thomas Thrainer | if lu.op.target_node is not None: |
51 | 1c3231aa | Thomas Thrainer | (lu.op.target_node_uuid, lu.op.target_node) = \ |
52 | 1c3231aa | Thomas Thrainer | ExpandNodeUuidAndName(lu.cfg, lu.op.target_node_uuid, lu.op.target_node) |
53 | 87e25be1 | Thomas Thrainer | |
54 | 87e25be1 | Thomas Thrainer | lu.needed_locks[locking.LEVEL_NODE] = [] |
55 | 87e25be1 | Thomas Thrainer | lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE |
56 | 87e25be1 | Thomas Thrainer | |
57 | 87e25be1 | Thomas Thrainer | lu.needed_locks[locking.LEVEL_NODE_RES] = [] |
58 | 87e25be1 | Thomas Thrainer | lu.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE |
59 | 87e25be1 | Thomas Thrainer | |
60 | 87e25be1 | Thomas Thrainer | # The node allocation lock is actually only needed for externally replicated
|
61 | 87e25be1 | Thomas Thrainer | # instances (e.g. sharedfile or RBD) and if an iallocator is used.
|
62 | 87e25be1 | Thomas Thrainer | lu.needed_locks[locking.LEVEL_NODE_ALLOC] = [] |
63 | 87e25be1 | Thomas Thrainer | |
64 | 87e25be1 | Thomas Thrainer | |
65 | 87e25be1 | Thomas Thrainer | def _DeclareLocksForMigration(lu, level): |
66 | 87e25be1 | Thomas Thrainer | """Declares locks for L{TLMigrateInstance}.
|
67 | 87e25be1 | Thomas Thrainer |
|
68 | 87e25be1 | Thomas Thrainer | @type lu: L{LogicalUnit}
|
69 | 87e25be1 | Thomas Thrainer | @param level: Lock level
|
70 | 87e25be1 | Thomas Thrainer |
|
71 | 87e25be1 | Thomas Thrainer | """
|
72 | 87e25be1 | Thomas Thrainer | if level == locking.LEVEL_NODE_ALLOC:
|
73 | 87e25be1 | Thomas Thrainer | assert lu.op.instance_name in lu.owned_locks(locking.LEVEL_INSTANCE) |
74 | 87e25be1 | Thomas Thrainer | |
75 | da4a52a3 | Thomas Thrainer | instance = lu.cfg.GetInstanceInfo(lu.op.instance_uuid) |
76 | 87e25be1 | Thomas Thrainer | |
77 | 87e25be1 | Thomas Thrainer | # Node locks are already declared here rather than at LEVEL_NODE as we need
|
78 | 87e25be1 | Thomas Thrainer | # the instance object anyway to declare the node allocation lock.
|
79 | 87e25be1 | Thomas Thrainer | if instance.disk_template in constants.DTS_EXT_MIRROR: |
80 | 87e25be1 | Thomas Thrainer | if lu.op.target_node is None: |
81 | 87e25be1 | Thomas Thrainer | lu.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET |
82 | 87e25be1 | Thomas Thrainer | lu.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET |
83 | 87e25be1 | Thomas Thrainer | else:
|
84 | 87e25be1 | Thomas Thrainer | lu.needed_locks[locking.LEVEL_NODE] = [instance.primary_node, |
85 | 1c3231aa | Thomas Thrainer | lu.op.target_node_uuid] |
86 | 87e25be1 | Thomas Thrainer | del lu.recalculate_locks[locking.LEVEL_NODE]
|
87 | 87e25be1 | Thomas Thrainer | else:
|
88 | 87e25be1 | Thomas Thrainer | lu._LockInstancesNodes() # pylint: disable=W0212
|
89 | 87e25be1 | Thomas Thrainer | |
90 | 87e25be1 | Thomas Thrainer | elif level == locking.LEVEL_NODE:
|
91 | 87e25be1 | Thomas Thrainer | # Node locks are declared together with the node allocation lock
|
92 | 87e25be1 | Thomas Thrainer | assert (lu.needed_locks[locking.LEVEL_NODE] or |
93 | 87e25be1 | Thomas Thrainer | lu.needed_locks[locking.LEVEL_NODE] is locking.ALL_SET)
|
94 | 87e25be1 | Thomas Thrainer | |
95 | 87e25be1 | Thomas Thrainer | elif level == locking.LEVEL_NODE_RES:
|
96 | 87e25be1 | Thomas Thrainer | # Copy node locks
|
97 | 87e25be1 | Thomas Thrainer | lu.needed_locks[locking.LEVEL_NODE_RES] = \ |
98 | 5eacbcae | Thomas Thrainer | CopyLockList(lu.needed_locks[locking.LEVEL_NODE]) |
99 | 87e25be1 | Thomas Thrainer | |
100 | 87e25be1 | Thomas Thrainer | |
101 | 87e25be1 | Thomas Thrainer | class LUInstanceFailover(LogicalUnit): |
102 | 87e25be1 | Thomas Thrainer | """Failover an instance.
|
103 | 87e25be1 | Thomas Thrainer |
|
104 | 87e25be1 | Thomas Thrainer | """
|
105 | 87e25be1 | Thomas Thrainer | HPATH = "instance-failover"
|
106 | 87e25be1 | Thomas Thrainer | HTYPE = constants.HTYPE_INSTANCE |
107 | 87e25be1 | Thomas Thrainer | REQ_BGL = False
|
108 | 87e25be1 | Thomas Thrainer | |
109 | 87e25be1 | Thomas Thrainer | def CheckArguments(self): |
110 | 87e25be1 | Thomas Thrainer | """Check the arguments.
|
111 | 87e25be1 | Thomas Thrainer |
|
112 | 87e25be1 | Thomas Thrainer | """
|
113 | 87e25be1 | Thomas Thrainer | self.iallocator = getattr(self.op, "iallocator", None) |
114 | 87e25be1 | Thomas Thrainer | self.target_node = getattr(self.op, "target_node", None) |
115 | 87e25be1 | Thomas Thrainer | |
116 | 87e25be1 | Thomas Thrainer | def ExpandNames(self): |
117 | 87e25be1 | Thomas Thrainer | self._ExpandAndLockInstance()
|
118 | 87e25be1 | Thomas Thrainer | _ExpandNamesForMigration(self)
|
119 | 87e25be1 | Thomas Thrainer | |
120 | 87e25be1 | Thomas Thrainer | self._migrater = \
|
121 | da4a52a3 | Thomas Thrainer | TLMigrateInstance(self, self.op.instance_uuid, self.op.instance_name, |
122 | 804d72eb | Thomas Thrainer | self.op.cleanup, True, False, |
123 | 804d72eb | Thomas Thrainer | self.op.ignore_consistency, True, |
124 | 87e25be1 | Thomas Thrainer | self.op.shutdown_timeout, self.op.ignore_ipolicy) |
125 | 87e25be1 | Thomas Thrainer | |
126 | 87e25be1 | Thomas Thrainer | self.tasklets = [self._migrater] |
127 | 87e25be1 | Thomas Thrainer | |
128 | 87e25be1 | Thomas Thrainer | def DeclareLocks(self, level): |
129 | 87e25be1 | Thomas Thrainer | _DeclareLocksForMigration(self, level)
|
130 | 87e25be1 | Thomas Thrainer | |
131 | 87e25be1 | Thomas Thrainer | def BuildHooksEnv(self): |
132 | 87e25be1 | Thomas Thrainer | """Build hooks env.
|
133 | 87e25be1 | Thomas Thrainer |
|
134 | 87e25be1 | Thomas Thrainer | This runs on master, primary and secondary nodes of the instance.
|
135 | 87e25be1 | Thomas Thrainer |
|
136 | 87e25be1 | Thomas Thrainer | """
|
137 | 87e25be1 | Thomas Thrainer | instance = self._migrater.instance
|
138 | 1c3231aa | Thomas Thrainer | source_node_uuid = instance.primary_node |
139 | 87e25be1 | Thomas Thrainer | env = { |
140 | 87e25be1 | Thomas Thrainer | "IGNORE_CONSISTENCY": self.op.ignore_consistency, |
141 | 87e25be1 | Thomas Thrainer | "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout, |
142 | 1c3231aa | Thomas Thrainer | "OLD_PRIMARY": self.cfg.GetNodeName(source_node_uuid), |
143 | 1c3231aa | Thomas Thrainer | "NEW_PRIMARY": self.op.target_node, |
144 | aa7a5c90 | Michele Tartara | "FAILOVER_CLEANUP": self.op.cleanup, |
145 | 87e25be1 | Thomas Thrainer | } |
146 | 87e25be1 | Thomas Thrainer | |
147 | 87e25be1 | Thomas Thrainer | if instance.disk_template in constants.DTS_INT_MIRROR: |
148 | 1c3231aa | Thomas Thrainer | env["OLD_SECONDARY"] = self.cfg.GetNodeName(instance.secondary_nodes[0]) |
149 | 1c3231aa | Thomas Thrainer | env["NEW_SECONDARY"] = self.cfg.GetNodeName(source_node_uuid) |
150 | 87e25be1 | Thomas Thrainer | else:
|
151 | 87e25be1 | Thomas Thrainer | env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = "" |
152 | 87e25be1 | Thomas Thrainer | |
153 | 5eacbcae | Thomas Thrainer | env.update(BuildInstanceHookEnvByObject(self, instance))
|
154 | 87e25be1 | Thomas Thrainer | |
155 | 87e25be1 | Thomas Thrainer | return env
|
156 | 87e25be1 | Thomas Thrainer | |
157 | 87e25be1 | Thomas Thrainer | def BuildHooksNodes(self): |
158 | 87e25be1 | Thomas Thrainer | """Build hooks nodes.
|
159 | 87e25be1 | Thomas Thrainer |
|
160 | 87e25be1 | Thomas Thrainer | """
|
161 | 87e25be1 | Thomas Thrainer | instance = self._migrater.instance
|
162 | 87e25be1 | Thomas Thrainer | nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes) |
163 | 87e25be1 | Thomas Thrainer | return (nl, nl + [instance.primary_node])
|
164 | 87e25be1 | Thomas Thrainer | |
165 | 87e25be1 | Thomas Thrainer | |
166 | 87e25be1 | Thomas Thrainer | class LUInstanceMigrate(LogicalUnit): |
167 | 87e25be1 | Thomas Thrainer | """Migrate an instance.
|
168 | 87e25be1 | Thomas Thrainer |
|
169 | 87e25be1 | Thomas Thrainer | This is migration without shutting down, compared to the failover,
|
170 | 87e25be1 | Thomas Thrainer | which is done with shutdown.
|
171 | 87e25be1 | Thomas Thrainer |
|
172 | 87e25be1 | Thomas Thrainer | """
|
173 | 87e25be1 | Thomas Thrainer | HPATH = "instance-migrate"
|
174 | 87e25be1 | Thomas Thrainer | HTYPE = constants.HTYPE_INSTANCE |
175 | 87e25be1 | Thomas Thrainer | REQ_BGL = False
|
176 | 87e25be1 | Thomas Thrainer | |
177 | 87e25be1 | Thomas Thrainer | def ExpandNames(self): |
178 | 87e25be1 | Thomas Thrainer | self._ExpandAndLockInstance()
|
179 | 87e25be1 | Thomas Thrainer | _ExpandNamesForMigration(self)
|
180 | 87e25be1 | Thomas Thrainer | |
181 | 87e25be1 | Thomas Thrainer | self._migrater = \
|
182 | da4a52a3 | Thomas Thrainer | TLMigrateInstance(self, self.op.instance_uuid, self.op.instance_name, |
183 | da4a52a3 | Thomas Thrainer | self.op.cleanup, False, self.op.allow_failover, False, |
184 | 87e25be1 | Thomas Thrainer | self.op.allow_runtime_changes,
|
185 | 87e25be1 | Thomas Thrainer | constants.DEFAULT_SHUTDOWN_TIMEOUT, |
186 | 87e25be1 | Thomas Thrainer | self.op.ignore_ipolicy)
|
187 | 87e25be1 | Thomas Thrainer | |
188 | 87e25be1 | Thomas Thrainer | self.tasklets = [self._migrater] |
189 | 87e25be1 | Thomas Thrainer | |
190 | 87e25be1 | Thomas Thrainer | def DeclareLocks(self, level): |
191 | 87e25be1 | Thomas Thrainer | _DeclareLocksForMigration(self, level)
|
192 | 87e25be1 | Thomas Thrainer | |
193 | 87e25be1 | Thomas Thrainer | def BuildHooksEnv(self): |
194 | 87e25be1 | Thomas Thrainer | """Build hooks env.
|
195 | 87e25be1 | Thomas Thrainer |
|
196 | 87e25be1 | Thomas Thrainer | This runs on master, primary and secondary nodes of the instance.
|
197 | 87e25be1 | Thomas Thrainer |
|
198 | 87e25be1 | Thomas Thrainer | """
|
199 | 87e25be1 | Thomas Thrainer | instance = self._migrater.instance
|
200 | 1c3231aa | Thomas Thrainer | source_node_uuid = instance.primary_node |
201 | 5eacbcae | Thomas Thrainer | env = BuildInstanceHookEnvByObject(self, instance)
|
202 | 87e25be1 | Thomas Thrainer | env.update({ |
203 | 87e25be1 | Thomas Thrainer | "MIGRATE_LIVE": self._migrater.live, |
204 | 87e25be1 | Thomas Thrainer | "MIGRATE_CLEANUP": self.op.cleanup, |
205 | 1c3231aa | Thomas Thrainer | "OLD_PRIMARY": self.cfg.GetNodeName(source_node_uuid), |
206 | 1c3231aa | Thomas Thrainer | "NEW_PRIMARY": self.op.target_node, |
207 | 87e25be1 | Thomas Thrainer | "ALLOW_RUNTIME_CHANGES": self.op.allow_runtime_changes, |
208 | 87e25be1 | Thomas Thrainer | }) |
209 | 87e25be1 | Thomas Thrainer | |
210 | 87e25be1 | Thomas Thrainer | if instance.disk_template in constants.DTS_INT_MIRROR: |
211 | 1c3231aa | Thomas Thrainer | env["OLD_SECONDARY"] = self.cfg.GetNodeName(instance.secondary_nodes[0]) |
212 | 1c3231aa | Thomas Thrainer | env["NEW_SECONDARY"] = self.cfg.GetNodeName(source_node_uuid) |
213 | 87e25be1 | Thomas Thrainer | else:
|
214 | 87e25be1 | Thomas Thrainer | env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = None |
215 | 87e25be1 | Thomas Thrainer | |
216 | 87e25be1 | Thomas Thrainer | return env
|
217 | 87e25be1 | Thomas Thrainer | |
218 | 87e25be1 | Thomas Thrainer | def BuildHooksNodes(self): |
219 | 87e25be1 | Thomas Thrainer | """Build hooks nodes.
|
220 | 87e25be1 | Thomas Thrainer |
|
221 | 87e25be1 | Thomas Thrainer | """
|
222 | 87e25be1 | Thomas Thrainer | instance = self._migrater.instance
|
223 | 1c3231aa | Thomas Thrainer | snode_uuids = list(instance.secondary_nodes)
|
224 | 1c3231aa | Thomas Thrainer | nl = [self.cfg.GetMasterNode(), instance.primary_node] + snode_uuids
|
225 | 87e25be1 | Thomas Thrainer | return (nl, nl)
|
226 | 87e25be1 | Thomas Thrainer | |
227 | 87e25be1 | Thomas Thrainer | |
228 | 87e25be1 | Thomas Thrainer | class TLMigrateInstance(Tasklet): |
229 | 87e25be1 | Thomas Thrainer | """Tasklet class for instance migration.
|
230 | 87e25be1 | Thomas Thrainer |
|
231 | 87e25be1 | Thomas Thrainer | @type live: boolean
|
232 | 87e25be1 | Thomas Thrainer | @ivar live: whether the migration will be done live or non-live;
|
233 | 87e25be1 | Thomas Thrainer | this variable is initalized only after CheckPrereq has run
|
234 | 87e25be1 | Thomas Thrainer | @type cleanup: boolean
|
235 | 87e25be1 | Thomas Thrainer | @ivar cleanup: Wheater we cleanup from a failed migration
|
236 | 87e25be1 | Thomas Thrainer | @type iallocator: string
|
237 | 87e25be1 | Thomas Thrainer | @ivar iallocator: The iallocator used to determine target_node
|
238 | 1c3231aa | Thomas Thrainer | @type target_node_uuid: string
|
239 | 1c3231aa | Thomas Thrainer | @ivar target_node_uuid: If given, the target node UUID to reallocate the
|
240 | 1c3231aa | Thomas Thrainer | instance to
|
241 | 87e25be1 | Thomas Thrainer | @type failover: boolean
|
242 | 87e25be1 | Thomas Thrainer | @ivar failover: Whether operation results in failover or migration
|
243 | 87e25be1 | Thomas Thrainer | @type fallback: boolean
|
244 | 87e25be1 | Thomas Thrainer | @ivar fallback: Whether fallback to failover is allowed if migration not
|
245 | 87e25be1 | Thomas Thrainer | possible
|
246 | 87e25be1 | Thomas Thrainer | @type ignore_consistency: boolean
|
247 | 87e25be1 | Thomas Thrainer | @ivar ignore_consistency: Wheter we should ignore consistency between source
|
248 | 87e25be1 | Thomas Thrainer | and target node
|
249 | 87e25be1 | Thomas Thrainer | @type shutdown_timeout: int
|
250 | 87e25be1 | Thomas Thrainer | @ivar shutdown_timeout: In case of failover timeout of the shutdown
|
251 | 87e25be1 | Thomas Thrainer | @type ignore_ipolicy: bool
|
252 | 87e25be1 | Thomas Thrainer | @ivar ignore_ipolicy: If true, we can ignore instance policy when migrating
|
253 | 87e25be1 | Thomas Thrainer |
|
254 | 87e25be1 | Thomas Thrainer | """
|
255 | 87e25be1 | Thomas Thrainer | |
256 | 87e25be1 | Thomas Thrainer | # Constants
|
257 | 87e25be1 | Thomas Thrainer | _MIGRATION_POLL_INTERVAL = 1 # seconds |
258 | 87e25be1 | Thomas Thrainer | _MIGRATION_FEEDBACK_INTERVAL = 10 # seconds |
259 | 87e25be1 | Thomas Thrainer | |
260 | da4a52a3 | Thomas Thrainer | def __init__(self, lu, instance_uuid, instance_name, cleanup, failover, |
261 | da4a52a3 | Thomas Thrainer | fallback, ignore_consistency, allow_runtime_changes, |
262 | da4a52a3 | Thomas Thrainer | shutdown_timeout, ignore_ipolicy): |
263 | 87e25be1 | Thomas Thrainer | """Initializes this class.
|
264 | 87e25be1 | Thomas Thrainer |
|
265 | 87e25be1 | Thomas Thrainer | """
|
266 | 87e25be1 | Thomas Thrainer | Tasklet.__init__(self, lu)
|
267 | 87e25be1 | Thomas Thrainer | |
268 | 87e25be1 | Thomas Thrainer | # Parameters
|
269 | da4a52a3 | Thomas Thrainer | self.instance_uuid = instance_uuid
|
270 | 87e25be1 | Thomas Thrainer | self.instance_name = instance_name
|
271 | 87e25be1 | Thomas Thrainer | self.cleanup = cleanup
|
272 | 87e25be1 | Thomas Thrainer | self.live = False # will be overridden later |
273 | 87e25be1 | Thomas Thrainer | self.failover = failover
|
274 | 87e25be1 | Thomas Thrainer | self.fallback = fallback
|
275 | 87e25be1 | Thomas Thrainer | self.ignore_consistency = ignore_consistency
|
276 | 87e25be1 | Thomas Thrainer | self.shutdown_timeout = shutdown_timeout
|
277 | 87e25be1 | Thomas Thrainer | self.ignore_ipolicy = ignore_ipolicy
|
278 | 87e25be1 | Thomas Thrainer | self.allow_runtime_changes = allow_runtime_changes
|
279 | 87e25be1 | Thomas Thrainer | |
280 | 87e25be1 | Thomas Thrainer | def CheckPrereq(self): |
281 | 87e25be1 | Thomas Thrainer | """Check prerequisites.
|
282 | 87e25be1 | Thomas Thrainer |
|
283 | 87e25be1 | Thomas Thrainer | This checks that the instance is in the cluster.
|
284 | 87e25be1 | Thomas Thrainer |
|
285 | 87e25be1 | Thomas Thrainer | """
|
286 | da4a52a3 | Thomas Thrainer | (self.instance_uuid, self.instance_name) = \ |
287 | da4a52a3 | Thomas Thrainer | ExpandInstanceUuidAndName(self.lu.cfg, self.instance_uuid, |
288 | da4a52a3 | Thomas Thrainer | self.instance_name)
|
289 | da4a52a3 | Thomas Thrainer | self.instance = self.cfg.GetInstanceInfo(self.instance_uuid) |
290 | d0d7d7cf | Thomas Thrainer | assert self.instance is not None |
291 | 87e25be1 | Thomas Thrainer | cluster = self.cfg.GetClusterInfo()
|
292 | 87e25be1 | Thomas Thrainer | |
293 | 87e25be1 | Thomas Thrainer | if (not self.cleanup and |
294 | d0d7d7cf | Thomas Thrainer | not self.instance.admin_state == constants.ADMINST_UP and |
295 | 87e25be1 | Thomas Thrainer | not self.failover and self.fallback): |
296 | 87e25be1 | Thomas Thrainer | self.lu.LogInfo("Instance is marked down or offline, fallback allowed," |
297 | 87e25be1 | Thomas Thrainer | " switching to failover")
|
298 | 87e25be1 | Thomas Thrainer | self.failover = True |
299 | 87e25be1 | Thomas Thrainer | |
300 | d0d7d7cf | Thomas Thrainer | if self.instance.disk_template not in constants.DTS_MIRRORED: |
301 | 87e25be1 | Thomas Thrainer | if self.failover: |
302 | 87e25be1 | Thomas Thrainer | text = "failovers"
|
303 | 87e25be1 | Thomas Thrainer | else:
|
304 | 87e25be1 | Thomas Thrainer | text = "migrations"
|
305 | 87e25be1 | Thomas Thrainer | raise errors.OpPrereqError("Instance's disk layout '%s' does not allow" |
306 | d0d7d7cf | Thomas Thrainer | " %s" % (self.instance.disk_template, text), |
307 | 87e25be1 | Thomas Thrainer | errors.ECODE_STATE) |
308 | 87e25be1 | Thomas Thrainer | |
309 | d0d7d7cf | Thomas Thrainer | if self.instance.disk_template in constants.DTS_EXT_MIRROR: |
310 | 5eacbcae | Thomas Thrainer | CheckIAllocatorOrNode(self.lu, "iallocator", "target_node") |
311 | 87e25be1 | Thomas Thrainer | |
312 | 87e25be1 | Thomas Thrainer | if self.lu.op.iallocator: |
313 | 87e25be1 | Thomas Thrainer | assert locking.NAL in self.lu.owned_locks(locking.LEVEL_NODE_ALLOC) |
314 | 87e25be1 | Thomas Thrainer | self._RunAllocator()
|
315 | 87e25be1 | Thomas Thrainer | else:
|
316 | 1c3231aa | Thomas Thrainer | # We set set self.target_node_uuid as it is required by
|
317 | 87e25be1 | Thomas Thrainer | # BuildHooksEnv
|
318 | 1c3231aa | Thomas Thrainer | self.target_node_uuid = self.lu.op.target_node_uuid |
319 | 87e25be1 | Thomas Thrainer | |
320 | 87e25be1 | Thomas Thrainer | # Check that the target node is correct in terms of instance policy
|
321 | 1c3231aa | Thomas Thrainer | nodeinfo = self.cfg.GetNodeInfo(self.target_node_uuid) |
322 | 87e25be1 | Thomas Thrainer | group_info = self.cfg.GetNodeGroup(nodeinfo.group)
|
323 | 87e25be1 | Thomas Thrainer | ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster, |
324 | 87e25be1 | Thomas Thrainer | group_info) |
325 | d0d7d7cf | Thomas Thrainer | CheckTargetNodeIPolicy(self.lu, ipolicy, self.instance, nodeinfo, |
326 | d0d7d7cf | Thomas Thrainer | self.cfg, ignore=self.ignore_ipolicy) |
327 | 87e25be1 | Thomas Thrainer | |
328 | 87e25be1 | Thomas Thrainer | # self.target_node is already populated, either directly or by the
|
329 | 87e25be1 | Thomas Thrainer | # iallocator run
|
330 | 1c3231aa | Thomas Thrainer | target_node_uuid = self.target_node_uuid
|
331 | d0d7d7cf | Thomas Thrainer | if self.target_node_uuid == self.instance.primary_node: |
332 | 1c3231aa | Thomas Thrainer | raise errors.OpPrereqError(
|
333 | 1c3231aa | Thomas Thrainer | "Cannot migrate instance %s to its primary (%s)" %
|
334 | d0d7d7cf | Thomas Thrainer | (self.instance.name,
|
335 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.instance.primary_node)), |
336 | 1c3231aa | Thomas Thrainer | errors.ECODE_STATE) |
337 | 87e25be1 | Thomas Thrainer | |
338 | 87e25be1 | Thomas Thrainer | if len(self.lu.tasklets) == 1: |
339 | 87e25be1 | Thomas Thrainer | # It is safe to release locks only when we're the only tasklet
|
340 | 87e25be1 | Thomas Thrainer | # in the LU
|
341 | 5eacbcae | Thomas Thrainer | ReleaseLocks(self.lu, locking.LEVEL_NODE,
|
342 | d0d7d7cf | Thomas Thrainer | keep=[self.instance.primary_node, self.target_node_uuid]) |
343 | 5eacbcae | Thomas Thrainer | ReleaseLocks(self.lu, locking.LEVEL_NODE_ALLOC)
|
344 | 87e25be1 | Thomas Thrainer | |
345 | 87e25be1 | Thomas Thrainer | else:
|
346 | 87e25be1 | Thomas Thrainer | assert not self.lu.glm.is_owned(locking.LEVEL_NODE_ALLOC) |
347 | 87e25be1 | Thomas Thrainer | |
348 | d0d7d7cf | Thomas Thrainer | secondary_node_uuids = self.instance.secondary_nodes
|
349 | 1c3231aa | Thomas Thrainer | if not secondary_node_uuids: |
350 | 87e25be1 | Thomas Thrainer | raise errors.ConfigurationError("No secondary node but using" |
351 | 87e25be1 | Thomas Thrainer | " %s disk template" %
|
352 | d0d7d7cf | Thomas Thrainer | self.instance.disk_template)
|
353 | 1c3231aa | Thomas Thrainer | target_node_uuid = secondary_node_uuids[0]
|
354 | 1c3231aa | Thomas Thrainer | if self.lu.op.iallocator or \ |
355 | 1c3231aa | Thomas Thrainer | (self.lu.op.target_node_uuid and |
356 | 1c3231aa | Thomas Thrainer | self.lu.op.target_node_uuid != target_node_uuid):
|
357 | 87e25be1 | Thomas Thrainer | if self.failover: |
358 | 87e25be1 | Thomas Thrainer | text = "failed over"
|
359 | 87e25be1 | Thomas Thrainer | else:
|
360 | 87e25be1 | Thomas Thrainer | text = "migrated"
|
361 | 87e25be1 | Thomas Thrainer | raise errors.OpPrereqError("Instances with disk template %s cannot" |
362 | 87e25be1 | Thomas Thrainer | " be %s to arbitrary nodes"
|
363 | 87e25be1 | Thomas Thrainer | " (neither an iallocator nor a target"
|
364 | 87e25be1 | Thomas Thrainer | " node can be passed)" %
|
365 | d0d7d7cf | Thomas Thrainer | (self.instance.disk_template, text),
|
366 | 87e25be1 | Thomas Thrainer | errors.ECODE_INVAL) |
367 | 1c3231aa | Thomas Thrainer | nodeinfo = self.cfg.GetNodeInfo(target_node_uuid)
|
368 | 87e25be1 | Thomas Thrainer | group_info = self.cfg.GetNodeGroup(nodeinfo.group)
|
369 | 87e25be1 | Thomas Thrainer | ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster, |
370 | 87e25be1 | Thomas Thrainer | group_info) |
371 | d0d7d7cf | Thomas Thrainer | CheckTargetNodeIPolicy(self.lu, ipolicy, self.instance, nodeinfo, |
372 | d0d7d7cf | Thomas Thrainer | self.cfg, ignore=self.ignore_ipolicy) |
373 | 87e25be1 | Thomas Thrainer | |
374 | d0d7d7cf | Thomas Thrainer | i_be = cluster.FillBE(self.instance)
|
375 | 87e25be1 | Thomas Thrainer | |
376 | 87e25be1 | Thomas Thrainer | # check memory requirements on the secondary node
|
377 | 87e25be1 | Thomas Thrainer | if (not self.cleanup and |
378 | d0d7d7cf | Thomas Thrainer | (not self.failover or |
379 | d0d7d7cf | Thomas Thrainer | self.instance.admin_state == constants.ADMINST_UP)):
|
380 | a295eb80 | Helga Velroyen | self.tgt_free_mem = CheckNodeFreeMemory(
|
381 | d0d7d7cf | Thomas Thrainer | self.lu, target_node_uuid,
|
382 | d0d7d7cf | Thomas Thrainer | "migrating instance %s" % self.instance.name, |
383 | d0d7d7cf | Thomas Thrainer | i_be[constants.BE_MINMEM], self.instance.hypervisor,
|
384 | d0d7d7cf | Thomas Thrainer | self.cfg.GetClusterInfo().hvparams[self.instance.hypervisor]) |
385 | 87e25be1 | Thomas Thrainer | else:
|
386 | 87e25be1 | Thomas Thrainer | self.lu.LogInfo("Not checking memory on the secondary node as" |
387 | 87e25be1 | Thomas Thrainer | " instance will not be started")
|
388 | 87e25be1 | Thomas Thrainer | |
389 | 87e25be1 | Thomas Thrainer | # check if failover must be forced instead of migration
|
390 | 87e25be1 | Thomas Thrainer | if (not self.cleanup and not self.failover and |
391 | 87e25be1 | Thomas Thrainer | i_be[constants.BE_ALWAYS_FAILOVER]): |
392 | 87e25be1 | Thomas Thrainer | self.lu.LogInfo("Instance configured to always failover; fallback" |
393 | 87e25be1 | Thomas Thrainer | " to failover")
|
394 | 87e25be1 | Thomas Thrainer | self.failover = True |
395 | 87e25be1 | Thomas Thrainer | |
396 | 87e25be1 | Thomas Thrainer | # check bridge existance
|
397 | d0d7d7cf | Thomas Thrainer | CheckInstanceBridgesExist(self.lu, self.instance, |
398 | d0d7d7cf | Thomas Thrainer | node_uuid=target_node_uuid) |
399 | 87e25be1 | Thomas Thrainer | |
400 | 87e25be1 | Thomas Thrainer | if not self.cleanup: |
401 | 1c3231aa | Thomas Thrainer | CheckNodeNotDrained(self.lu, target_node_uuid)
|
402 | 87e25be1 | Thomas Thrainer | if not self.failover: |
403 | d0d7d7cf | Thomas Thrainer | result = self.rpc.call_instance_migratable(self.instance.primary_node, |
404 | d0d7d7cf | Thomas Thrainer | self.instance)
|
405 | 87e25be1 | Thomas Thrainer | if result.fail_msg and self.fallback: |
406 | 87e25be1 | Thomas Thrainer | self.lu.LogInfo("Can't migrate, instance offline, fallback to" |
407 | 87e25be1 | Thomas Thrainer | " failover")
|
408 | 87e25be1 | Thomas Thrainer | self.failover = True |
409 | 87e25be1 | Thomas Thrainer | else:
|
410 | 87e25be1 | Thomas Thrainer | result.Raise("Can't migrate, please use failover",
|
411 | 87e25be1 | Thomas Thrainer | prereq=True, ecode=errors.ECODE_STATE)
|
412 | 87e25be1 | Thomas Thrainer | |
413 | 87e25be1 | Thomas Thrainer | assert not (self.failover and self.cleanup) |
414 | 87e25be1 | Thomas Thrainer | |
415 | 87e25be1 | Thomas Thrainer | if not self.failover: |
416 | 87e25be1 | Thomas Thrainer | if self.lu.op.live is not None and self.lu.op.mode is not None: |
417 | 87e25be1 | Thomas Thrainer | raise errors.OpPrereqError("Only one of the 'live' and 'mode'" |
418 | 87e25be1 | Thomas Thrainer | " parameters are accepted",
|
419 | 87e25be1 | Thomas Thrainer | errors.ECODE_INVAL) |
420 | 87e25be1 | Thomas Thrainer | if self.lu.op.live is not None: |
421 | 87e25be1 | Thomas Thrainer | if self.lu.op.live: |
422 | 87e25be1 | Thomas Thrainer | self.lu.op.mode = constants.HT_MIGRATION_LIVE
|
423 | 87e25be1 | Thomas Thrainer | else:
|
424 | 87e25be1 | Thomas Thrainer | self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
|
425 | 87e25be1 | Thomas Thrainer | # reset the 'live' parameter to None so that repeated
|
426 | 87e25be1 | Thomas Thrainer | # invocations of CheckPrereq do not raise an exception
|
427 | 87e25be1 | Thomas Thrainer | self.lu.op.live = None |
428 | 87e25be1 | Thomas Thrainer | elif self.lu.op.mode is None: |
429 | 87e25be1 | Thomas Thrainer | # read the default value from the hypervisor
|
430 | 87e25be1 | Thomas Thrainer | i_hv = cluster.FillHV(self.instance, skip_globals=False) |
431 | 87e25be1 | Thomas Thrainer | self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
|
432 | 87e25be1 | Thomas Thrainer | |
433 | 87e25be1 | Thomas Thrainer | self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE |
434 | 87e25be1 | Thomas Thrainer | else:
|
435 | 87e25be1 | Thomas Thrainer | # Failover is never live
|
436 | 87e25be1 | Thomas Thrainer | self.live = False |
437 | 87e25be1 | Thomas Thrainer | |
438 | 87e25be1 | Thomas Thrainer | if not (self.failover or self.cleanup): |
439 | 0bbec3af | Helga Velroyen | remote_info = self.rpc.call_instance_info(
|
440 | d0d7d7cf | Thomas Thrainer | self.instance.primary_node, self.instance.name, |
441 | d0d7d7cf | Thomas Thrainer | self.instance.hypervisor, cluster.hvparams[self.instance.hypervisor]) |
442 | 87e25be1 | Thomas Thrainer | remote_info.Raise("Error checking instance on node %s" %
|
443 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.instance.primary_node)) |
444 | 87e25be1 | Thomas Thrainer | instance_running = bool(remote_info.payload)
|
445 | 87e25be1 | Thomas Thrainer | if instance_running:
|
446 | 87e25be1 | Thomas Thrainer | self.current_mem = int(remote_info.payload["memory"]) |
447 | 87e25be1 | Thomas Thrainer | |
448 | 87e25be1 | Thomas Thrainer | def _RunAllocator(self): |
449 | 87e25be1 | Thomas Thrainer | """Run the allocator based on input opcode.
|
450 | 87e25be1 | Thomas Thrainer |
|
451 | 87e25be1 | Thomas Thrainer | """
|
452 | 87e25be1 | Thomas Thrainer | assert locking.NAL in self.lu.owned_locks(locking.LEVEL_NODE_ALLOC) |
453 | 87e25be1 | Thomas Thrainer | |
454 | 87e25be1 | Thomas Thrainer | # FIXME: add a self.ignore_ipolicy option
|
455 | 1c3231aa | Thomas Thrainer | req = iallocator.IAReqRelocate( |
456 | da4a52a3 | Thomas Thrainer | inst_uuid=self.instance_uuid,
|
457 | 1c3231aa | Thomas Thrainer | relocate_from_node_uuids=[self.instance.primary_node])
|
458 | 87e25be1 | Thomas Thrainer | ial = iallocator.IAllocator(self.cfg, self.rpc, req) |
459 | 87e25be1 | Thomas Thrainer | |
460 | 87e25be1 | Thomas Thrainer | ial.Run(self.lu.op.iallocator)
|
461 | 87e25be1 | Thomas Thrainer | |
462 | 87e25be1 | Thomas Thrainer | if not ial.success: |
463 | 87e25be1 | Thomas Thrainer | raise errors.OpPrereqError("Can't compute nodes using" |
464 | 87e25be1 | Thomas Thrainer | " iallocator '%s': %s" %
|
465 | 87e25be1 | Thomas Thrainer | (self.lu.op.iallocator, ial.info),
|
466 | 87e25be1 | Thomas Thrainer | errors.ECODE_NORES) |
467 | 1c3231aa | Thomas Thrainer | self.target_node_uuid = self.cfg.GetNodeInfoByName(ial.result[0]).uuid |
468 | 87e25be1 | Thomas Thrainer | self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s", |
469 | 87e25be1 | Thomas Thrainer | self.instance_name, self.lu.op.iallocator, |
470 | 87e25be1 | Thomas Thrainer | utils.CommaJoin(ial.result)) |
471 | 87e25be1 | Thomas Thrainer | |
472 | 87e25be1 | Thomas Thrainer | def _WaitUntilSync(self): |
473 | 87e25be1 | Thomas Thrainer | """Poll with custom rpc for disk sync.
|
474 | 87e25be1 | Thomas Thrainer |
|
475 | 87e25be1 | Thomas Thrainer | This uses our own step-based rpc call.
|
476 | 87e25be1 | Thomas Thrainer |
|
477 | 87e25be1 | Thomas Thrainer | """
|
478 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* wait until resync is done") |
479 | 87e25be1 | Thomas Thrainer | all_done = False
|
480 | 87e25be1 | Thomas Thrainer | while not all_done: |
481 | 87e25be1 | Thomas Thrainer | all_done = True
|
482 | 1c3231aa | Thomas Thrainer | result = self.rpc.call_drbd_wait_sync(self.all_node_uuids, |
483 | 87e25be1 | Thomas Thrainer | (self.instance.disks,
|
484 | 87e25be1 | Thomas Thrainer | self.instance))
|
485 | 87e25be1 | Thomas Thrainer | min_percent = 100
|
486 | 1c3231aa | Thomas Thrainer | for node_uuid, nres in result.items(): |
487 | 1c3231aa | Thomas Thrainer | nres.Raise("Cannot resync disks on node %s" %
|
488 | 1c3231aa | Thomas Thrainer | self.cfg.GetNodeName(node_uuid))
|
489 | 87e25be1 | Thomas Thrainer | node_done, node_percent = nres.payload |
490 | 87e25be1 | Thomas Thrainer | all_done = all_done and node_done
|
491 | 87e25be1 | Thomas Thrainer | if node_percent is not None: |
492 | 87e25be1 | Thomas Thrainer | min_percent = min(min_percent, node_percent)
|
493 | 87e25be1 | Thomas Thrainer | if not all_done: |
494 | 87e25be1 | Thomas Thrainer | if min_percent < 100: |
495 | 87e25be1 | Thomas Thrainer | self.feedback_fn(" - progress: %.1f%%" % min_percent) |
496 | 87e25be1 | Thomas Thrainer | time.sleep(2)
|
497 | 87e25be1 | Thomas Thrainer | |
498 | 1c3231aa | Thomas Thrainer | def _EnsureSecondary(self, node_uuid): |
499 | 87e25be1 | Thomas Thrainer | """Demote a node to secondary.
|
500 | 87e25be1 | Thomas Thrainer |
|
501 | 87e25be1 | Thomas Thrainer | """
|
502 | 1c3231aa | Thomas Thrainer | self.feedback_fn("* switching node %s to secondary mode" % |
503 | 1c3231aa | Thomas Thrainer | self.cfg.GetNodeName(node_uuid))
|
504 | 87e25be1 | Thomas Thrainer | |
505 | 1c3231aa | Thomas Thrainer | result = self.rpc.call_blockdev_close(node_uuid, self.instance.name, |
506 | 0c3d9c7c | Thomas Thrainer | (self.instance.disks, self.instance)) |
507 | 1c3231aa | Thomas Thrainer | result.Raise("Cannot change disk to secondary on node %s" %
|
508 | 1c3231aa | Thomas Thrainer | self.cfg.GetNodeName(node_uuid))
|
509 | 87e25be1 | Thomas Thrainer | |
510 | 87e25be1 | Thomas Thrainer | def _GoStandalone(self): |
511 | 87e25be1 | Thomas Thrainer | """Disconnect from the network.
|
512 | 87e25be1 | Thomas Thrainer |
|
513 | 87e25be1 | Thomas Thrainer | """
|
514 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* changing into standalone mode") |
515 | 0c3d9c7c | Thomas Thrainer | result = self.rpc.call_drbd_disconnect_net(
|
516 | 0c3d9c7c | Thomas Thrainer | self.all_node_uuids, (self.instance.disks, self.instance)) |
517 | 1c3231aa | Thomas Thrainer | for node_uuid, nres in result.items(): |
518 | 1c3231aa | Thomas Thrainer | nres.Raise("Cannot disconnect disks node %s" %
|
519 | 1c3231aa | Thomas Thrainer | self.cfg.GetNodeName(node_uuid))
|
520 | 87e25be1 | Thomas Thrainer | |
521 | 87e25be1 | Thomas Thrainer | def _GoReconnect(self, multimaster): |
522 | 87e25be1 | Thomas Thrainer | """Reconnect to the network.
|
523 | 87e25be1 | Thomas Thrainer |
|
524 | 87e25be1 | Thomas Thrainer | """
|
525 | 87e25be1 | Thomas Thrainer | if multimaster:
|
526 | 87e25be1 | Thomas Thrainer | msg = "dual-master"
|
527 | 87e25be1 | Thomas Thrainer | else:
|
528 | 87e25be1 | Thomas Thrainer | msg = "single-master"
|
529 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* changing disks into %s mode" % msg) |
530 | 0c3d9c7c | Thomas Thrainer | result = self.rpc.call_drbd_attach_net(self.all_node_uuids, |
531 | 87e25be1 | Thomas Thrainer | (self.instance.disks, self.instance), |
532 | 87e25be1 | Thomas Thrainer | self.instance.name, multimaster)
|
533 | 1c3231aa | Thomas Thrainer | for node_uuid, nres in result.items(): |
534 | 1c3231aa | Thomas Thrainer | nres.Raise("Cannot change disks config on node %s" %
|
535 | 1c3231aa | Thomas Thrainer | self.cfg.GetNodeName(node_uuid))
|
536 | 87e25be1 | Thomas Thrainer | |
537 | 87e25be1 | Thomas Thrainer | def _ExecCleanup(self): |
538 | 87e25be1 | Thomas Thrainer | """Try to cleanup after a failed migration.
|
539 | 87e25be1 | Thomas Thrainer |
|
540 | 87e25be1 | Thomas Thrainer | The cleanup is done by:
|
541 | 87e25be1 | Thomas Thrainer | - check that the instance is running only on one node
|
542 | 87e25be1 | Thomas Thrainer | (and update the config if needed)
|
543 | 87e25be1 | Thomas Thrainer | - change disks on its secondary node to secondary
|
544 | 87e25be1 | Thomas Thrainer | - wait until disks are fully synchronized
|
545 | 87e25be1 | Thomas Thrainer | - disconnect from the network
|
546 | 87e25be1 | Thomas Thrainer | - change disks into single-master mode
|
547 | 87e25be1 | Thomas Thrainer | - wait again until disks are fully synchronized
|
548 | 87e25be1 | Thomas Thrainer |
|
549 | 87e25be1 | Thomas Thrainer | """
|
550 | 87e25be1 | Thomas Thrainer | # check running on only one node
|
551 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* checking where the instance actually runs" |
552 | 87e25be1 | Thomas Thrainer | " (if this hangs, the hypervisor might be in"
|
553 | 87e25be1 | Thomas Thrainer | " a bad state)")
|
554 | 8ac806e6 | Helga Velroyen | cluster_hvparams = self.cfg.GetClusterInfo().hvparams
|
555 | 1c3231aa | Thomas Thrainer | ins_l = self.rpc.call_instance_list(self.all_node_uuids, |
556 | d0d7d7cf | Thomas Thrainer | [self.instance.hypervisor],
|
557 | 8ac806e6 | Helga Velroyen | cluster_hvparams) |
558 | 1c3231aa | Thomas Thrainer | for node_uuid, result in ins_l.items(): |
559 | 1c3231aa | Thomas Thrainer | result.Raise("Can't contact node %s" % node_uuid)
|
560 | 87e25be1 | Thomas Thrainer | |
561 | d0d7d7cf | Thomas Thrainer | runningon_source = self.instance.name in \ |
562 | d0d7d7cf | Thomas Thrainer | ins_l[self.source_node_uuid].payload
|
563 | d0d7d7cf | Thomas Thrainer | runningon_target = self.instance.name in \ |
564 | d0d7d7cf | Thomas Thrainer | ins_l[self.target_node_uuid].payload
|
565 | 87e25be1 | Thomas Thrainer | |
566 | 87e25be1 | Thomas Thrainer | if runningon_source and runningon_target: |
567 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Instance seems to be running on two nodes," |
568 | 87e25be1 | Thomas Thrainer | " or the hypervisor is confused; you will have"
|
569 | 87e25be1 | Thomas Thrainer | " to ensure manually that it runs only on one"
|
570 | 87e25be1 | Thomas Thrainer | " and restart this operation")
|
571 | 87e25be1 | Thomas Thrainer | |
572 | 87e25be1 | Thomas Thrainer | if not (runningon_source or runningon_target): |
573 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Instance does not seem to be running at all;" |
574 | 87e25be1 | Thomas Thrainer | " in this case it's safer to repair by"
|
575 | 87e25be1 | Thomas Thrainer | " running 'gnt-instance stop' to ensure disk"
|
576 | 87e25be1 | Thomas Thrainer | " shutdown, and then restarting it")
|
577 | 87e25be1 | Thomas Thrainer | |
578 | 87e25be1 | Thomas Thrainer | if runningon_target:
|
579 | 87e25be1 | Thomas Thrainer | # the migration has actually succeeded, we need to update the config
|
580 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* instance running on secondary node (%s)," |
581 | 1c3231aa | Thomas Thrainer | " updating config" %
|
582 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.target_node_uuid)) |
583 | d0d7d7cf | Thomas Thrainer | self.instance.primary_node = self.target_node_uuid |
584 | d0d7d7cf | Thomas Thrainer | self.cfg.Update(self.instance, self.feedback_fn) |
585 | d0d7d7cf | Thomas Thrainer | demoted_node_uuid = self.source_node_uuid
|
586 | 87e25be1 | Thomas Thrainer | else:
|
587 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* instance confirmed to be running on its" |
588 | 1c3231aa | Thomas Thrainer | " primary node (%s)" %
|
589 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.source_node_uuid)) |
590 | d0d7d7cf | Thomas Thrainer | demoted_node_uuid = self.target_node_uuid
|
591 | 87e25be1 | Thomas Thrainer | |
592 | d0d7d7cf | Thomas Thrainer | if self.instance.disk_template in constants.DTS_INT_MIRROR: |
593 | 1c3231aa | Thomas Thrainer | self._EnsureSecondary(demoted_node_uuid)
|
594 | 87e25be1 | Thomas Thrainer | try:
|
595 | 87e25be1 | Thomas Thrainer | self._WaitUntilSync()
|
596 | 87e25be1 | Thomas Thrainer | except errors.OpExecError:
|
597 | 87e25be1 | Thomas Thrainer | # we ignore here errors, since if the device is standalone, it
|
598 | 87e25be1 | Thomas Thrainer | # won't be able to sync
|
599 | 87e25be1 | Thomas Thrainer | pass
|
600 | 87e25be1 | Thomas Thrainer | self._GoStandalone()
|
601 | 87e25be1 | Thomas Thrainer | self._GoReconnect(False) |
602 | 87e25be1 | Thomas Thrainer | self._WaitUntilSync()
|
603 | 87e25be1 | Thomas Thrainer | |
604 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* done") |
605 | 87e25be1 | Thomas Thrainer | |
606 | 87e25be1 | Thomas Thrainer | def _RevertDiskStatus(self): |
607 | 87e25be1 | Thomas Thrainer | """Try to revert the disk status after a failed migration.
|
608 | 87e25be1 | Thomas Thrainer |
|
609 | 87e25be1 | Thomas Thrainer | """
|
610 | 87e25be1 | Thomas Thrainer | if self.instance.disk_template in constants.DTS_EXT_MIRROR: |
611 | 87e25be1 | Thomas Thrainer | return
|
612 | 87e25be1 | Thomas Thrainer | |
613 | 87e25be1 | Thomas Thrainer | try:
|
614 | 1c3231aa | Thomas Thrainer | self._EnsureSecondary(self.target_node_uuid) |
615 | 87e25be1 | Thomas Thrainer | self._GoStandalone()
|
616 | 87e25be1 | Thomas Thrainer | self._GoReconnect(False) |
617 | 87e25be1 | Thomas Thrainer | self._WaitUntilSync()
|
618 | 87e25be1 | Thomas Thrainer | except errors.OpExecError, err:
|
619 | 87e25be1 | Thomas Thrainer | self.lu.LogWarning("Migration failed and I can't reconnect the drives," |
620 | 87e25be1 | Thomas Thrainer | " please try to recover the instance manually;"
|
621 | 87e25be1 | Thomas Thrainer | " error '%s'" % str(err)) |
622 | 87e25be1 | Thomas Thrainer | |
623 | 87e25be1 | Thomas Thrainer | def _AbortMigration(self): |
624 | 87e25be1 | Thomas Thrainer | """Call the hypervisor code to abort a started migration.
|
625 | 87e25be1 | Thomas Thrainer |
|
626 | 87e25be1 | Thomas Thrainer | """
|
627 | 1c3231aa | Thomas Thrainer | abort_result = self.rpc.call_instance_finalize_migration_dst(
|
628 | d0d7d7cf | Thomas Thrainer | self.target_node_uuid, self.instance, self.migration_info, |
629 | d0d7d7cf | Thomas Thrainer | False)
|
630 | 87e25be1 | Thomas Thrainer | abort_msg = abort_result.fail_msg |
631 | 87e25be1 | Thomas Thrainer | if abort_msg:
|
632 | 87e25be1 | Thomas Thrainer | logging.error("Aborting migration failed on target node %s: %s",
|
633 | 1c3231aa | Thomas Thrainer | self.cfg.GetNodeName(self.target_node_uuid), abort_msg) |
634 | 87e25be1 | Thomas Thrainer | # Don't raise an exception here, as we stil have to try to revert the
|
635 | 87e25be1 | Thomas Thrainer | # disk status, even if this step failed.
|
636 | 87e25be1 | Thomas Thrainer | |
637 | 87e25be1 | Thomas Thrainer | abort_result = self.rpc.call_instance_finalize_migration_src(
|
638 | d0d7d7cf | Thomas Thrainer | self.source_node_uuid, self.instance, False, self.live) |
639 | 87e25be1 | Thomas Thrainer | abort_msg = abort_result.fail_msg |
640 | 87e25be1 | Thomas Thrainer | if abort_msg:
|
641 | 87e25be1 | Thomas Thrainer | logging.error("Aborting migration failed on source node %s: %s",
|
642 | 1c3231aa | Thomas Thrainer | self.cfg.GetNodeName(self.source_node_uuid), abort_msg) |
643 | 87e25be1 | Thomas Thrainer | |
644 | 87e25be1 | Thomas Thrainer | def _ExecMigration(self): |
645 | 87e25be1 | Thomas Thrainer | """Migrate an instance.
|
646 | 87e25be1 | Thomas Thrainer |
|
647 | 87e25be1 | Thomas Thrainer | The migrate is done by:
|
648 | 87e25be1 | Thomas Thrainer | - change the disks into dual-master mode
|
649 | 87e25be1 | Thomas Thrainer | - wait until disks are fully synchronized again
|
650 | 87e25be1 | Thomas Thrainer | - migrate the instance
|
651 | 87e25be1 | Thomas Thrainer | - change disks on the new secondary node (the old primary) to secondary
|
652 | 87e25be1 | Thomas Thrainer | - wait until disks are fully synchronized
|
653 | 87e25be1 | Thomas Thrainer | - change disks into single-master mode
|
654 | 87e25be1 | Thomas Thrainer |
|
655 | 87e25be1 | Thomas Thrainer | """
|
656 | 87e25be1 | Thomas Thrainer | # Check for hypervisor version mismatch and warn the user.
|
657 | d0d7d7cf | Thomas Thrainer | hvspecs = [(self.instance.hypervisor,
|
658 | d0d7d7cf | Thomas Thrainer | self.cfg.GetClusterInfo().hvparams[self.instance.hypervisor])] |
659 | d0d7d7cf | Thomas Thrainer | nodeinfo = self.rpc.call_node_info(
|
660 | da803ff1 | Helga Velroyen | [self.source_node_uuid, self.target_node_uuid], None, hvspecs) |
661 | 87e25be1 | Thomas Thrainer | for ninfo in nodeinfo.values(): |
662 | 87e25be1 | Thomas Thrainer | ninfo.Raise("Unable to retrieve node information from node '%s'" %
|
663 | 87e25be1 | Thomas Thrainer | ninfo.node) |
664 | d0d7d7cf | Thomas Thrainer | (_, _, (src_info, )) = nodeinfo[self.source_node_uuid].payload
|
665 | d0d7d7cf | Thomas Thrainer | (_, _, (dst_info, )) = nodeinfo[self.target_node_uuid].payload
|
666 | 87e25be1 | Thomas Thrainer | |
667 | 87e25be1 | Thomas Thrainer | if ((constants.HV_NODEINFO_KEY_VERSION in src_info) and |
668 | 87e25be1 | Thomas Thrainer | (constants.HV_NODEINFO_KEY_VERSION in dst_info)):
|
669 | 87e25be1 | Thomas Thrainer | src_version = src_info[constants.HV_NODEINFO_KEY_VERSION] |
670 | 87e25be1 | Thomas Thrainer | dst_version = dst_info[constants.HV_NODEINFO_KEY_VERSION] |
671 | 87e25be1 | Thomas Thrainer | if src_version != dst_version:
|
672 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* warning: hypervisor version mismatch between" |
673 | 87e25be1 | Thomas Thrainer | " source (%s) and target (%s) node" %
|
674 | 87e25be1 | Thomas Thrainer | (src_version, dst_version)) |
675 | 87e25be1 | Thomas Thrainer | |
676 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* checking disk consistency between source and target") |
677 | d0d7d7cf | Thomas Thrainer | for (idx, dev) in enumerate(self.instance.disks): |
678 | d0d7d7cf | Thomas Thrainer | if not CheckDiskConsistency(self.lu, self.instance, dev, |
679 | d0d7d7cf | Thomas Thrainer | self.target_node_uuid,
|
680 | 1c3231aa | Thomas Thrainer | False):
|
681 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Disk %s is degraded or not fully" |
682 | 87e25be1 | Thomas Thrainer | " synchronized on target node,"
|
683 | 87e25be1 | Thomas Thrainer | " aborting migration" % idx)
|
684 | 87e25be1 | Thomas Thrainer | |
685 | 87e25be1 | Thomas Thrainer | if self.current_mem > self.tgt_free_mem: |
686 | 87e25be1 | Thomas Thrainer | if not self.allow_runtime_changes: |
687 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Memory ballooning not allowed and not enough" |
688 | 87e25be1 | Thomas Thrainer | " free memory to fit instance %s on target"
|
689 | 87e25be1 | Thomas Thrainer | " node %s (have %dMB, need %dMB)" %
|
690 | d0d7d7cf | Thomas Thrainer | (self.instance.name,
|
691 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.target_node_uuid), |
692 | 87e25be1 | Thomas Thrainer | self.tgt_free_mem, self.current_mem)) |
693 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* setting instance memory to %s" % self.tgt_free_mem) |
694 | d0d7d7cf | Thomas Thrainer | rpcres = self.rpc.call_instance_balloon_memory(self.instance.primary_node, |
695 | d0d7d7cf | Thomas Thrainer | self.instance,
|
696 | 87e25be1 | Thomas Thrainer | self.tgt_free_mem)
|
697 | 87e25be1 | Thomas Thrainer | rpcres.Raise("Cannot modify instance runtime memory")
|
698 | 87e25be1 | Thomas Thrainer | |
699 | 87e25be1 | Thomas Thrainer | # First get the migration information from the remote node
|
700 | d0d7d7cf | Thomas Thrainer | result = self.rpc.call_migration_info(self.source_node_uuid, self.instance) |
701 | 87e25be1 | Thomas Thrainer | msg = result.fail_msg |
702 | 87e25be1 | Thomas Thrainer | if msg:
|
703 | 87e25be1 | Thomas Thrainer | log_err = ("Failed fetching source migration information from %s: %s" %
|
704 | d0d7d7cf | Thomas Thrainer | (self.cfg.GetNodeName(self.source_node_uuid), msg)) |
705 | 87e25be1 | Thomas Thrainer | logging.error(log_err) |
706 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError(log_err)
|
707 | 87e25be1 | Thomas Thrainer | |
708 | 87e25be1 | Thomas Thrainer | self.migration_info = migration_info = result.payload
|
709 | 87e25be1 | Thomas Thrainer | |
710 | 87e25be1 | Thomas Thrainer | if self.instance.disk_template not in constants.DTS_EXT_MIRROR: |
711 | 87e25be1 | Thomas Thrainer | # Then switch the disks to master/master mode
|
712 | d0d7d7cf | Thomas Thrainer | self._EnsureSecondary(self.target_node_uuid) |
713 | 87e25be1 | Thomas Thrainer | self._GoStandalone()
|
714 | 87e25be1 | Thomas Thrainer | self._GoReconnect(True) |
715 | 87e25be1 | Thomas Thrainer | self._WaitUntilSync()
|
716 | 87e25be1 | Thomas Thrainer | |
717 | 1c3231aa | Thomas Thrainer | self.feedback_fn("* preparing %s to accept the instance" % |
718 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.target_node_uuid)) |
719 | d0d7d7cf | Thomas Thrainer | result = self.rpc.call_accept_instance(self.target_node_uuid, |
720 | d0d7d7cf | Thomas Thrainer | self.instance,
|
721 | 87e25be1 | Thomas Thrainer | migration_info, |
722 | d0d7d7cf | Thomas Thrainer | self.nodes_ip[self.target_node_uuid]) |
723 | 87e25be1 | Thomas Thrainer | |
724 | 87e25be1 | Thomas Thrainer | msg = result.fail_msg |
725 | 87e25be1 | Thomas Thrainer | if msg:
|
726 | 87e25be1 | Thomas Thrainer | logging.error("Instance pre-migration failed, trying to revert"
|
727 | 87e25be1 | Thomas Thrainer | " disk status: %s", msg)
|
728 | 87e25be1 | Thomas Thrainer | self.feedback_fn("Pre-migration failed, aborting") |
729 | 87e25be1 | Thomas Thrainer | self._AbortMigration()
|
730 | 87e25be1 | Thomas Thrainer | self._RevertDiskStatus()
|
731 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Could not pre-migrate instance %s: %s" % |
732 | d0d7d7cf | Thomas Thrainer | (self.instance.name, msg))
|
733 | 87e25be1 | Thomas Thrainer | |
734 | 1c3231aa | Thomas Thrainer | self.feedback_fn("* migrating instance to %s" % |
735 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.target_node_uuid)) |
736 | bc0a2284 | Helga Velroyen | cluster = self.cfg.GetClusterInfo()
|
737 | bc0a2284 | Helga Velroyen | result = self.rpc.call_instance_migrate(
|
738 | d0d7d7cf | Thomas Thrainer | self.source_node_uuid, cluster.cluster_name, self.instance, |
739 | d0d7d7cf | Thomas Thrainer | self.nodes_ip[self.target_node_uuid], self.live) |
740 | 87e25be1 | Thomas Thrainer | msg = result.fail_msg |
741 | 87e25be1 | Thomas Thrainer | if msg:
|
742 | 87e25be1 | Thomas Thrainer | logging.error("Instance migration failed, trying to revert"
|
743 | 87e25be1 | Thomas Thrainer | " disk status: %s", msg)
|
744 | 87e25be1 | Thomas Thrainer | self.feedback_fn("Migration failed, aborting") |
745 | 87e25be1 | Thomas Thrainer | self._AbortMigration()
|
746 | 87e25be1 | Thomas Thrainer | self._RevertDiskStatus()
|
747 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Could not migrate instance %s: %s" % |
748 | d0d7d7cf | Thomas Thrainer | (self.instance.name, msg))
|
749 | 87e25be1 | Thomas Thrainer | |
750 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* starting memory transfer") |
751 | 87e25be1 | Thomas Thrainer | last_feedback = time.time() |
752 | 87e25be1 | Thomas Thrainer | while True: |
753 | d0d7d7cf | Thomas Thrainer | result = self.rpc.call_instance_get_migration_status(
|
754 | d0d7d7cf | Thomas Thrainer | self.source_node_uuid, self.instance) |
755 | 87e25be1 | Thomas Thrainer | msg = result.fail_msg |
756 | 87e25be1 | Thomas Thrainer | ms = result.payload # MigrationStatus instance
|
757 | 87e25be1 | Thomas Thrainer | if msg or (ms.status in constants.HV_MIGRATION_FAILED_STATUSES): |
758 | 87e25be1 | Thomas Thrainer | logging.error("Instance migration failed, trying to revert"
|
759 | 87e25be1 | Thomas Thrainer | " disk status: %s", msg)
|
760 | 87e25be1 | Thomas Thrainer | self.feedback_fn("Migration failed, aborting") |
761 | 87e25be1 | Thomas Thrainer | self._AbortMigration()
|
762 | 87e25be1 | Thomas Thrainer | self._RevertDiskStatus()
|
763 | 87e25be1 | Thomas Thrainer | if not msg: |
764 | 87e25be1 | Thomas Thrainer | msg = "hypervisor returned failure"
|
765 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Could not migrate instance %s: %s" % |
766 | d0d7d7cf | Thomas Thrainer | (self.instance.name, msg))
|
767 | 87e25be1 | Thomas Thrainer | |
768 | 87e25be1 | Thomas Thrainer | if result.payload.status != constants.HV_MIGRATION_ACTIVE:
|
769 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* memory transfer complete") |
770 | 87e25be1 | Thomas Thrainer | break
|
771 | 87e25be1 | Thomas Thrainer | |
772 | 87e25be1 | Thomas Thrainer | if (utils.TimeoutExpired(last_feedback,
|
773 | 87e25be1 | Thomas Thrainer | self._MIGRATION_FEEDBACK_INTERVAL) and |
774 | 87e25be1 | Thomas Thrainer | ms.transferred_ram is not None): |
775 | 87e25be1 | Thomas Thrainer | mem_progress = 100 * float(ms.transferred_ram) / float(ms.total_ram) |
776 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* memory transfer progress: %.2f %%" % mem_progress) |
777 | 87e25be1 | Thomas Thrainer | last_feedback = time.time() |
778 | 87e25be1 | Thomas Thrainer | |
779 | 87e25be1 | Thomas Thrainer | time.sleep(self._MIGRATION_POLL_INTERVAL)
|
780 | 87e25be1 | Thomas Thrainer | |
781 | d0d7d7cf | Thomas Thrainer | result = self.rpc.call_instance_finalize_migration_src(
|
782 | d0d7d7cf | Thomas Thrainer | self.source_node_uuid, self.instance, True, self.live) |
783 | 87e25be1 | Thomas Thrainer | msg = result.fail_msg |
784 | 87e25be1 | Thomas Thrainer | if msg:
|
785 | 87e25be1 | Thomas Thrainer | logging.error("Instance migration succeeded, but finalization failed"
|
786 | 87e25be1 | Thomas Thrainer | " on the source node: %s", msg)
|
787 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Could not finalize instance migration: %s" % |
788 | 87e25be1 | Thomas Thrainer | msg) |
789 | 87e25be1 | Thomas Thrainer | |
790 | d0d7d7cf | Thomas Thrainer | self.instance.primary_node = self.target_node_uuid |
791 | 87e25be1 | Thomas Thrainer | |
792 | 87e25be1 | Thomas Thrainer | # distribute new instance config to the other nodes
|
793 | d0d7d7cf | Thomas Thrainer | self.cfg.Update(self.instance, self.feedback_fn) |
794 | 87e25be1 | Thomas Thrainer | |
795 | d0d7d7cf | Thomas Thrainer | result = self.rpc.call_instance_finalize_migration_dst(
|
796 | d0d7d7cf | Thomas Thrainer | self.target_node_uuid, self.instance, migration_info, True) |
797 | 87e25be1 | Thomas Thrainer | msg = result.fail_msg |
798 | 87e25be1 | Thomas Thrainer | if msg:
|
799 | 87e25be1 | Thomas Thrainer | logging.error("Instance migration succeeded, but finalization failed"
|
800 | 87e25be1 | Thomas Thrainer | " on the target node: %s", msg)
|
801 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Could not finalize instance migration: %s" % |
802 | 87e25be1 | Thomas Thrainer | msg) |
803 | 87e25be1 | Thomas Thrainer | |
804 | 87e25be1 | Thomas Thrainer | if self.instance.disk_template not in constants.DTS_EXT_MIRROR: |
805 | d0d7d7cf | Thomas Thrainer | self._EnsureSecondary(self.source_node_uuid) |
806 | 87e25be1 | Thomas Thrainer | self._WaitUntilSync()
|
807 | 87e25be1 | Thomas Thrainer | self._GoStandalone()
|
808 | 87e25be1 | Thomas Thrainer | self._GoReconnect(False) |
809 | 87e25be1 | Thomas Thrainer | self._WaitUntilSync()
|
810 | 87e25be1 | Thomas Thrainer | |
811 | 87e25be1 | Thomas Thrainer | # If the instance's disk template is `rbd' or `ext' and there was a
|
812 | 87e25be1 | Thomas Thrainer | # successful migration, unmap the device from the source node.
|
813 | 87e25be1 | Thomas Thrainer | if self.instance.disk_template in (constants.DT_RBD, constants.DT_EXT): |
814 | d0d7d7cf | Thomas Thrainer | disks = ExpandCheckDisks(self.instance, self.instance.disks) |
815 | 1c3231aa | Thomas Thrainer | self.feedback_fn("* unmapping instance's disks from %s" % |
816 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.source_node_uuid)) |
817 | 87e25be1 | Thomas Thrainer | for disk in disks: |
818 | d0d7d7cf | Thomas Thrainer | result = self.rpc.call_blockdev_shutdown(self.source_node_uuid, |
819 | d0d7d7cf | Thomas Thrainer | (disk, self.instance))
|
820 | 87e25be1 | Thomas Thrainer | msg = result.fail_msg |
821 | 87e25be1 | Thomas Thrainer | if msg:
|
822 | 87e25be1 | Thomas Thrainer | logging.error("Migration was successful, but couldn't unmap the"
|
823 | 87e25be1 | Thomas Thrainer | " block device %s on source node %s: %s",
|
824 | d0d7d7cf | Thomas Thrainer | disk.iv_name, |
825 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.source_node_uuid), msg) |
826 | 87e25be1 | Thomas Thrainer | logging.error("You need to unmap the device %s manually on %s",
|
827 | d0d7d7cf | Thomas Thrainer | disk.iv_name, |
828 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.source_node_uuid)) |
829 | 87e25be1 | Thomas Thrainer | |
830 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* done") |
831 | 87e25be1 | Thomas Thrainer | |
832 | 87e25be1 | Thomas Thrainer | def _ExecFailover(self): |
833 | 87e25be1 | Thomas Thrainer | """Failover an instance.
|
834 | 87e25be1 | Thomas Thrainer |
|
835 | 87e25be1 | Thomas Thrainer | The failover is done by shutting it down on its present node and
|
836 | 87e25be1 | Thomas Thrainer | starting it on the secondary.
|
837 | 87e25be1 | Thomas Thrainer |
|
838 | 87e25be1 | Thomas Thrainer | """
|
839 | d0d7d7cf | Thomas Thrainer | primary_node = self.cfg.GetNodeInfo(self.instance.primary_node) |
840 | 87e25be1 | Thomas Thrainer | |
841 | d0d7d7cf | Thomas Thrainer | source_node_uuid = self.instance.primary_node
|
842 | 87e25be1 | Thomas Thrainer | |
843 | d0d7d7cf | Thomas Thrainer | if self.instance.disks_active: |
844 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* checking disk consistency between source and target") |
845 | d0d7d7cf | Thomas Thrainer | for (idx, dev) in enumerate(self.instance.disks): |
846 | 87e25be1 | Thomas Thrainer | # for drbd, these are drbd over lvm
|
847 | d0d7d7cf | Thomas Thrainer | if not CheckDiskConsistency(self.lu, self.instance, dev, |
848 | d0d7d7cf | Thomas Thrainer | self.target_node_uuid, False): |
849 | 87e25be1 | Thomas Thrainer | if primary_node.offline:
|
850 | 87e25be1 | Thomas Thrainer | self.feedback_fn("Node %s is offline, ignoring degraded disk %s on" |
851 | 87e25be1 | Thomas Thrainer | " target node %s" %
|
852 | 1c3231aa | Thomas Thrainer | (primary_node.name, idx, |
853 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.target_node_uuid))) |
854 | 87e25be1 | Thomas Thrainer | elif not self.ignore_consistency: |
855 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Disk %s is degraded on target node," |
856 | 87e25be1 | Thomas Thrainer | " aborting failover" % idx)
|
857 | 87e25be1 | Thomas Thrainer | else:
|
858 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* not checking disk consistency as instance is not" |
859 | 87e25be1 | Thomas Thrainer | " running")
|
860 | 87e25be1 | Thomas Thrainer | |
861 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* shutting down instance on source node") |
862 | 87e25be1 | Thomas Thrainer | logging.info("Shutting down instance %s on node %s",
|
863 | d0d7d7cf | Thomas Thrainer | self.instance.name, self.cfg.GetNodeName(source_node_uuid)) |
864 | 87e25be1 | Thomas Thrainer | |
865 | d0d7d7cf | Thomas Thrainer | result = self.rpc.call_instance_shutdown(source_node_uuid, self.instance, |
866 | 87e25be1 | Thomas Thrainer | self.shutdown_timeout,
|
867 | 87e25be1 | Thomas Thrainer | self.lu.op.reason)
|
868 | 87e25be1 | Thomas Thrainer | msg = result.fail_msg |
869 | 87e25be1 | Thomas Thrainer | if msg:
|
870 | 87e25be1 | Thomas Thrainer | if self.ignore_consistency or primary_node.offline: |
871 | 87e25be1 | Thomas Thrainer | self.lu.LogWarning("Could not shutdown instance %s on node %s," |
872 | 87e25be1 | Thomas Thrainer | " proceeding anyway; please make sure node"
|
873 | 87e25be1 | Thomas Thrainer | " %s is down; error details: %s",
|
874 | d0d7d7cf | Thomas Thrainer | self.instance.name,
|
875 | 1c3231aa | Thomas Thrainer | self.cfg.GetNodeName(source_node_uuid),
|
876 | 1c3231aa | Thomas Thrainer | self.cfg.GetNodeName(source_node_uuid), msg)
|
877 | 87e25be1 | Thomas Thrainer | else:
|
878 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Could not shutdown instance %s on" |
879 | 87e25be1 | Thomas Thrainer | " node %s: %s" %
|
880 | d0d7d7cf | Thomas Thrainer | (self.instance.name,
|
881 | 1c3231aa | Thomas Thrainer | self.cfg.GetNodeName(source_node_uuid), msg))
|
882 | 87e25be1 | Thomas Thrainer | |
883 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* deactivating the instance's disks on source node") |
884 | d0d7d7cf | Thomas Thrainer | if not ShutdownInstanceDisks(self.lu, self.instance, ignore_primary=True): |
885 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Can't shut down the instance's disks") |
886 | 87e25be1 | Thomas Thrainer | |
887 | d0d7d7cf | Thomas Thrainer | self.instance.primary_node = self.target_node_uuid |
888 | 87e25be1 | Thomas Thrainer | # distribute new instance config to the other nodes
|
889 | d0d7d7cf | Thomas Thrainer | self.cfg.Update(self.instance, self.feedback_fn) |
890 | 87e25be1 | Thomas Thrainer | |
891 | 87e25be1 | Thomas Thrainer | # Only start the instance if it's marked as up
|
892 | d0d7d7cf | Thomas Thrainer | if self.instance.admin_state == constants.ADMINST_UP: |
893 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* activating the instance's disks on target node %s" % |
894 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.target_node_uuid)) |
895 | d0d7d7cf | Thomas Thrainer | logging.info("Starting instance %s on node %s", self.instance.name, |
896 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.target_node_uuid)) |
897 | 87e25be1 | Thomas Thrainer | |
898 | d0d7d7cf | Thomas Thrainer | disks_ok, _ = AssembleInstanceDisks(self.lu, self.instance, |
899 | 5eacbcae | Thomas Thrainer | ignore_secondaries=True)
|
900 | 87e25be1 | Thomas Thrainer | if not disks_ok: |
901 | d0d7d7cf | Thomas Thrainer | ShutdownInstanceDisks(self.lu, self.instance) |
902 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Can't activate the instance's disks") |
903 | 87e25be1 | Thomas Thrainer | |
904 | 87e25be1 | Thomas Thrainer | self.feedback_fn("* starting the instance on the target node %s" % |
905 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.target_node_uuid)) |
906 | d0d7d7cf | Thomas Thrainer | result = self.rpc.call_instance_start(self.target_node_uuid, |
907 | d0d7d7cf | Thomas Thrainer | (self.instance, None, None), False, |
908 | 1c3231aa | Thomas Thrainer | self.lu.op.reason)
|
909 | 87e25be1 | Thomas Thrainer | msg = result.fail_msg |
910 | 87e25be1 | Thomas Thrainer | if msg:
|
911 | d0d7d7cf | Thomas Thrainer | ShutdownInstanceDisks(self.lu, self.instance) |
912 | 87e25be1 | Thomas Thrainer | raise errors.OpExecError("Could not start instance %s on node %s: %s" % |
913 | d0d7d7cf | Thomas Thrainer | (self.instance.name,
|
914 | d0d7d7cf | Thomas Thrainer | self.cfg.GetNodeName(self.target_node_uuid), |
915 | d0d7d7cf | Thomas Thrainer | msg)) |
916 | 87e25be1 | Thomas Thrainer | |
917 | 87e25be1 | Thomas Thrainer | def Exec(self, feedback_fn): |
918 | 87e25be1 | Thomas Thrainer | """Perform the migration.
|
919 | 87e25be1 | Thomas Thrainer |
|
920 | 87e25be1 | Thomas Thrainer | """
|
921 | 87e25be1 | Thomas Thrainer | self.feedback_fn = feedback_fn
|
922 | 1c3231aa | Thomas Thrainer | self.source_node_uuid = self.instance.primary_node |
923 | 87e25be1 | Thomas Thrainer | |
924 | 87e25be1 | Thomas Thrainer | # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
|
925 | 87e25be1 | Thomas Thrainer | if self.instance.disk_template in constants.DTS_INT_MIRROR: |
926 | 1c3231aa | Thomas Thrainer | self.target_node_uuid = self.instance.secondary_nodes[0] |
927 | 87e25be1 | Thomas Thrainer | # Otherwise self.target_node has been populated either
|
928 | 87e25be1 | Thomas Thrainer | # directly, or through an iallocator.
|
929 | 87e25be1 | Thomas Thrainer | |
930 | 1c3231aa | Thomas Thrainer | self.all_node_uuids = [self.source_node_uuid, self.target_node_uuid] |
931 | 1c3231aa | Thomas Thrainer | self.nodes_ip = dict((uuid, node.secondary_ip) for (uuid, node) |
932 | 1c3231aa | Thomas Thrainer | in self.cfg.GetMultiNodeInfo(self.all_node_uuids)) |
933 | 87e25be1 | Thomas Thrainer | |
934 | 87e25be1 | Thomas Thrainer | if self.failover: |
935 | 87e25be1 | Thomas Thrainer | feedback_fn("Failover instance %s" % self.instance.name) |
936 | 87e25be1 | Thomas Thrainer | self._ExecFailover()
|
937 | 87e25be1 | Thomas Thrainer | else:
|
938 | 87e25be1 | Thomas Thrainer | feedback_fn("Migrating instance %s" % self.instance.name) |
939 | 87e25be1 | Thomas Thrainer | |
940 | 87e25be1 | Thomas Thrainer | if self.cleanup: |
941 | 87e25be1 | Thomas Thrainer | return self._ExecCleanup() |
942 | 87e25be1 | Thomas Thrainer | else:
|
943 | 87e25be1 | Thomas Thrainer | return self._ExecMigration() |