root / lib / cmdlib / group.py @ 4b75f8a4
History | View | Annotate | Download (32.6 kB)
1 |
#
|
---|---|
2 |
#
|
3 |
|
4 |
# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014 Google Inc.
|
5 |
#
|
6 |
# This program is free software; you can redistribute it and/or modify
|
7 |
# it under the terms of the GNU General Public License as published by
|
8 |
# the Free Software Foundation; either version 2 of the License, or
|
9 |
# (at your option) any later version.
|
10 |
#
|
11 |
# This program is distributed in the hope that it will be useful, but
|
12 |
# WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 |
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
14 |
# General Public License for more details.
|
15 |
#
|
16 |
# You should have received a copy of the GNU General Public License
|
17 |
# along with this program; if not, write to the Free Software
|
18 |
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
|
19 |
# 02110-1301, USA.
|
20 |
|
21 |
|
22 |
"""Logical units dealing with node groups."""
|
23 |
|
24 |
import itertools |
25 |
import logging |
26 |
|
27 |
from ganeti import constants |
28 |
from ganeti import errors |
29 |
from ganeti import locking |
30 |
from ganeti import objects |
31 |
from ganeti import utils |
32 |
from ganeti.masterd import iallocator |
33 |
from ganeti.cmdlib.base import LogicalUnit, NoHooksLU, ResultWithJobs |
34 |
from ganeti.cmdlib.common import MergeAndVerifyHvState, \ |
35 |
MergeAndVerifyDiskState, GetWantedNodes, GetUpdatedParams, \ |
36 |
CheckNodeGroupInstances, GetUpdatedIPolicy, \ |
37 |
ComputeNewInstanceViolations, GetDefaultIAllocator, ShareAll, \ |
38 |
CheckInstancesNodeGroups, LoadNodeEvacResult, MapInstanceLvsToNodes, \ |
39 |
CheckIpolicyVsDiskTemplates, CheckDiskAccessModeValidity, \ |
40 |
CheckDiskAccessModeConsistency, ConnectInstanceCommunicationNetworkOp |
41 |
|
42 |
import ganeti.masterd.instance |
43 |
|
44 |
|
45 |
class LUGroupAdd(LogicalUnit): |
46 |
"""Logical unit for creating node groups.
|
47 |
|
48 |
"""
|
49 |
HPATH = "group-add"
|
50 |
HTYPE = constants.HTYPE_GROUP |
51 |
REQ_BGL = False
|
52 |
|
53 |
def ExpandNames(self): |
54 |
# We need the new group's UUID here so that we can create and acquire the
|
55 |
# corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
|
56 |
# that it should not check whether the UUID exists in the configuration.
|
57 |
self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId()) |
58 |
self.needed_locks = {}
|
59 |
self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid |
60 |
|
61 |
def _CheckIpolicy(self): |
62 |
"""Checks the group's ipolicy for consistency and validity.
|
63 |
|
64 |
"""
|
65 |
if self.op.ipolicy: |
66 |
cluster = self.cfg.GetClusterInfo()
|
67 |
full_ipolicy = cluster.SimpleFillIPolicy(self.op.ipolicy)
|
68 |
try:
|
69 |
objects.InstancePolicy.CheckParameterSyntax(full_ipolicy, False)
|
70 |
except errors.ConfigurationError, err:
|
71 |
raise errors.OpPrereqError("Invalid instance policy: %s" % err, |
72 |
errors.ECODE_INVAL) |
73 |
CheckIpolicyVsDiskTemplates(full_ipolicy, |
74 |
cluster.enabled_disk_templates) |
75 |
|
76 |
def CheckPrereq(self): |
77 |
"""Check prerequisites.
|
78 |
|
79 |
This checks that the given group name is not an existing node group
|
80 |
already.
|
81 |
|
82 |
"""
|
83 |
try:
|
84 |
existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name) |
85 |
except errors.OpPrereqError:
|
86 |
pass
|
87 |
else:
|
88 |
raise errors.OpPrereqError("Desired group name '%s' already exists as a" |
89 |
" node group (UUID: %s)" %
|
90 |
(self.op.group_name, existing_uuid),
|
91 |
errors.ECODE_EXISTS) |
92 |
|
93 |
if self.op.ndparams: |
94 |
utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
|
95 |
|
96 |
if self.op.hv_state: |
97 |
self.new_hv_state = MergeAndVerifyHvState(self.op.hv_state, None) |
98 |
else:
|
99 |
self.new_hv_state = None |
100 |
|
101 |
if self.op.disk_state: |
102 |
self.new_disk_state = MergeAndVerifyDiskState(self.op.disk_state, None) |
103 |
else:
|
104 |
self.new_disk_state = None |
105 |
|
106 |
if self.op.diskparams: |
107 |
for templ in constants.DISK_TEMPLATES: |
108 |
if templ in self.op.diskparams: |
109 |
utils.ForceDictType(self.op.diskparams[templ],
|
110 |
constants.DISK_DT_TYPES) |
111 |
self.new_diskparams = self.op.diskparams |
112 |
try:
|
113 |
utils.VerifyDictOptions(self.new_diskparams, constants.DISK_DT_DEFAULTS)
|
114 |
except errors.OpPrereqError, err:
|
115 |
raise errors.OpPrereqError("While verify diskparams options: %s" % err, |
116 |
errors.ECODE_INVAL) |
117 |
else:
|
118 |
self.new_diskparams = {}
|
119 |
|
120 |
self._CheckIpolicy()
|
121 |
|
122 |
def BuildHooksEnv(self): |
123 |
"""Build hooks env.
|
124 |
|
125 |
"""
|
126 |
return {
|
127 |
"GROUP_NAME": self.op.group_name, |
128 |
} |
129 |
|
130 |
def BuildHooksNodes(self): |
131 |
"""Build hooks nodes.
|
132 |
|
133 |
"""
|
134 |
mn = self.cfg.GetMasterNode()
|
135 |
return ([mn], [mn])
|
136 |
|
137 |
@staticmethod
|
138 |
def _ConnectInstanceCommunicationNetwork(cfg, group_uuid, network_name): |
139 |
"""Connect a node group to the instance communication network.
|
140 |
|
141 |
The group is connected to the instance communication network via
|
142 |
the Opcode 'OpNetworkConnect'.
|
143 |
|
144 |
@type cfg: L{ganeti.config.ConfigWriter}
|
145 |
@param cfg: Ganeti configuration
|
146 |
|
147 |
@type group_uuid: string
|
148 |
@param group_uuid: UUID of the group to connect
|
149 |
|
150 |
@type network_name: string
|
151 |
@param network_name: name of the network to connect to
|
152 |
|
153 |
@rtype: L{ganeti.cmdlib.ResultWithJobs} or L{None}
|
154 |
@return: L{ganeti.cmdlib.ResultWithJobs} if the group needs to be
|
155 |
connected, otherwise (the group is already connected)
|
156 |
L{None}
|
157 |
|
158 |
"""
|
159 |
try:
|
160 |
cfg.LookupNetwork(network_name) |
161 |
network_exists = True
|
162 |
except errors.OpPrereqError:
|
163 |
network_exists = False
|
164 |
|
165 |
if network_exists:
|
166 |
op = ConnectInstanceCommunicationNetworkOp(group_uuid, network_name) |
167 |
return ResultWithJobs([[op]])
|
168 |
else:
|
169 |
return None |
170 |
|
171 |
def Exec(self, feedback_fn): |
172 |
"""Add the node group to the cluster.
|
173 |
|
174 |
"""
|
175 |
group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
|
176 |
uuid=self.group_uuid,
|
177 |
alloc_policy=self.op.alloc_policy,
|
178 |
ndparams=self.op.ndparams,
|
179 |
diskparams=self.new_diskparams,
|
180 |
ipolicy=self.op.ipolicy,
|
181 |
hv_state_static=self.new_hv_state,
|
182 |
disk_state_static=self.new_disk_state)
|
183 |
|
184 |
self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False) |
185 |
del self.remove_locks[locking.LEVEL_NODEGROUP] |
186 |
|
187 |
network_name = self.cfg.GetClusterInfo().instance_communication_network
|
188 |
if network_name:
|
189 |
return self._ConnectInstanceCommunicationNetwork(self.cfg, |
190 |
self.group_uuid,
|
191 |
network_name) |
192 |
|
193 |
|
194 |
class LUGroupAssignNodes(NoHooksLU): |
195 |
"""Logical unit for assigning nodes to groups.
|
196 |
|
197 |
"""
|
198 |
REQ_BGL = False
|
199 |
|
200 |
def ExpandNames(self): |
201 |
# These raise errors.OpPrereqError on their own:
|
202 |
self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name) |
203 |
(self.op.node_uuids, self.op.nodes) = GetWantedNodes(self, self.op.nodes) |
204 |
|
205 |
# We want to lock all the affected nodes and groups. We have readily
|
206 |
# available the list of nodes, and the *destination* group. To gather the
|
207 |
# list of "source" groups, we need to fetch node information later on.
|
208 |
self.needed_locks = {
|
209 |
locking.LEVEL_NODEGROUP: set([self.group_uuid]), |
210 |
locking.LEVEL_NODE: self.op.node_uuids,
|
211 |
} |
212 |
|
213 |
def DeclareLocks(self, level): |
214 |
if level == locking.LEVEL_NODEGROUP:
|
215 |
assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1 |
216 |
|
217 |
# Try to get all affected nodes' groups without having the group or node
|
218 |
# lock yet. Needs verification later in the code flow.
|
219 |
groups = self.cfg.GetNodeGroupsFromNodes(self.op.node_uuids) |
220 |
|
221 |
self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
|
222 |
|
223 |
def CheckPrereq(self): |
224 |
"""Check prerequisites.
|
225 |
|
226 |
"""
|
227 |
assert self.needed_locks[locking.LEVEL_NODEGROUP] |
228 |
assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) == |
229 |
frozenset(self.op.node_uuids)) |
230 |
|
231 |
expected_locks = (set([self.group_uuid]) | |
232 |
self.cfg.GetNodeGroupsFromNodes(self.op.node_uuids)) |
233 |
actual_locks = self.owned_locks(locking.LEVEL_NODEGROUP)
|
234 |
if actual_locks != expected_locks:
|
235 |
raise errors.OpExecError("Nodes changed groups since locks were acquired," |
236 |
" current groups are '%s', used to be '%s'" %
|
237 |
(utils.CommaJoin(expected_locks), |
238 |
utils.CommaJoin(actual_locks))) |
239 |
|
240 |
self.node_data = self.cfg.GetAllNodesInfo() |
241 |
self.group = self.cfg.GetNodeGroup(self.group_uuid) |
242 |
instance_data = self.cfg.GetAllInstancesInfo()
|
243 |
|
244 |
if self.group is None: |
245 |
raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" % |
246 |
(self.op.group_name, self.group_uuid)) |
247 |
|
248 |
(new_splits, previous_splits) = \ |
249 |
self.CheckAssignmentForSplitInstances([(uuid, self.group_uuid) |
250 |
for uuid in self.op.node_uuids], |
251 |
self.node_data, instance_data)
|
252 |
|
253 |
if new_splits:
|
254 |
fmt_new_splits = utils.CommaJoin(utils.NiceSort( |
255 |
self.cfg.GetInstanceNames(new_splits)))
|
256 |
|
257 |
if not self.op.force: |
258 |
raise errors.OpExecError("The following instances get split by this" |
259 |
" change and --force was not given: %s" %
|
260 |
fmt_new_splits) |
261 |
else:
|
262 |
self.LogWarning("This operation will split the following instances: %s", |
263 |
fmt_new_splits) |
264 |
|
265 |
if previous_splits:
|
266 |
self.LogWarning("In addition, these already-split instances continue" |
267 |
" to be split across groups: %s",
|
268 |
utils.CommaJoin(utils.NiceSort( |
269 |
self.cfg.GetInstanceNames(previous_splits))))
|
270 |
|
271 |
def Exec(self, feedback_fn): |
272 |
"""Assign nodes to a new group.
|
273 |
|
274 |
"""
|
275 |
mods = [(node_uuid, self.group_uuid) for node_uuid in self.op.node_uuids] |
276 |
|
277 |
self.cfg.AssignGroupNodes(mods)
|
278 |
|
279 |
@staticmethod
|
280 |
def CheckAssignmentForSplitInstances(changes, node_data, instance_data): |
281 |
"""Check for split instances after a node assignment.
|
282 |
|
283 |
This method considers a series of node assignments as an atomic operation,
|
284 |
and returns information about split instances after applying the set of
|
285 |
changes.
|
286 |
|
287 |
In particular, it returns information about newly split instances, and
|
288 |
instances that were already split, and remain so after the change.
|
289 |
|
290 |
Only instances whose disk template is listed in constants.DTS_INT_MIRROR are
|
291 |
considered.
|
292 |
|
293 |
@type changes: list of (node_uuid, new_group_uuid) pairs.
|
294 |
@param changes: list of node assignments to consider.
|
295 |
@param node_data: a dict with data for all nodes
|
296 |
@param instance_data: a dict with all instances to consider
|
297 |
@rtype: a two-tuple
|
298 |
@return: a list of instances that were previously okay and result split as a
|
299 |
consequence of this change, and a list of instances that were previously
|
300 |
split and this change does not fix.
|
301 |
|
302 |
"""
|
303 |
changed_nodes = dict((uuid, group) for uuid, group in changes |
304 |
if node_data[uuid].group != group)
|
305 |
|
306 |
all_split_instances = set()
|
307 |
previously_split_instances = set()
|
308 |
|
309 |
for inst in instance_data.values(): |
310 |
if inst.disk_template not in constants.DTS_INT_MIRROR: |
311 |
continue
|
312 |
|
313 |
if len(set(node_data[node_uuid].group |
314 |
for node_uuid in inst.all_nodes)) > 1: |
315 |
previously_split_instances.add(inst.uuid) |
316 |
|
317 |
if len(set(changed_nodes.get(node_uuid, node_data[node_uuid].group) |
318 |
for node_uuid in inst.all_nodes)) > 1: |
319 |
all_split_instances.add(inst.uuid) |
320 |
|
321 |
return (list(all_split_instances - previously_split_instances), |
322 |
list(previously_split_instances & all_split_instances))
|
323 |
|
324 |
|
325 |
class LUGroupSetParams(LogicalUnit): |
326 |
"""Modifies the parameters of a node group.
|
327 |
|
328 |
"""
|
329 |
HPATH = "group-modify"
|
330 |
HTYPE = constants.HTYPE_GROUP |
331 |
REQ_BGL = False
|
332 |
|
333 |
def CheckArguments(self): |
334 |
all_changes = [ |
335 |
self.op.ndparams,
|
336 |
self.op.diskparams,
|
337 |
self.op.alloc_policy,
|
338 |
self.op.hv_state,
|
339 |
self.op.disk_state,
|
340 |
self.op.ipolicy,
|
341 |
] |
342 |
|
343 |
if all_changes.count(None) == len(all_changes): |
344 |
raise errors.OpPrereqError("Please pass at least one modification", |
345 |
errors.ECODE_INVAL) |
346 |
|
347 |
if self.op.diskparams: |
348 |
CheckDiskAccessModeValidity(self.op.diskparams)
|
349 |
|
350 |
def ExpandNames(self): |
351 |
# This raises errors.OpPrereqError on its own:
|
352 |
self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name) |
353 |
|
354 |
self.needed_locks = {
|
355 |
locking.LEVEL_INSTANCE: [], |
356 |
locking.LEVEL_NODEGROUP: [self.group_uuid],
|
357 |
} |
358 |
|
359 |
self.share_locks[locking.LEVEL_INSTANCE] = 1 |
360 |
|
361 |
def DeclareLocks(self, level): |
362 |
if level == locking.LEVEL_INSTANCE:
|
363 |
assert not self.needed_locks[locking.LEVEL_INSTANCE] |
364 |
|
365 |
# Lock instances optimistically, needs verification once group lock has
|
366 |
# been acquired
|
367 |
self.needed_locks[locking.LEVEL_INSTANCE] = \
|
368 |
self.cfg.GetInstanceNames(
|
369 |
self.cfg.GetNodeGroupInstances(self.group_uuid)) |
370 |
|
371 |
@staticmethod
|
372 |
def _UpdateAndVerifyDiskParams(old, new): |
373 |
"""Updates and verifies disk parameters.
|
374 |
|
375 |
"""
|
376 |
new_params = GetUpdatedParams(old, new) |
377 |
utils.ForceDictType(new_params, constants.DISK_DT_TYPES) |
378 |
return new_params
|
379 |
|
380 |
def _CheckIpolicy(self, cluster, owned_instance_names): |
381 |
"""Sanity checks for the ipolicy.
|
382 |
|
383 |
@type cluster: C{objects.Cluster}
|
384 |
@param cluster: the cluster's configuration
|
385 |
@type owned_instance_names: list of string
|
386 |
@param owned_instance_names: list of instances
|
387 |
|
388 |
"""
|
389 |
if self.op.ipolicy: |
390 |
self.new_ipolicy = GetUpdatedIPolicy(self.group.ipolicy, |
391 |
self.op.ipolicy,
|
392 |
group_policy=True)
|
393 |
|
394 |
new_ipolicy = cluster.SimpleFillIPolicy(self.new_ipolicy)
|
395 |
CheckIpolicyVsDiskTemplates(new_ipolicy, |
396 |
cluster.enabled_disk_templates) |
397 |
instances = \ |
398 |
dict(self.cfg.GetMultiInstanceInfoByName(owned_instance_names)) |
399 |
gmi = ganeti.masterd.instance |
400 |
violations = \ |
401 |
ComputeNewInstanceViolations(gmi.CalculateGroupIPolicy(cluster, |
402 |
self.group),
|
403 |
new_ipolicy, instances.values(), |
404 |
self.cfg)
|
405 |
|
406 |
if violations:
|
407 |
self.LogWarning("After the ipolicy change the following instances" |
408 |
" violate them: %s",
|
409 |
utils.CommaJoin(violations)) |
410 |
|
411 |
def CheckPrereq(self): |
412 |
"""Check prerequisites.
|
413 |
|
414 |
"""
|
415 |
owned_instance_names = frozenset(self.owned_locks(locking.LEVEL_INSTANCE)) |
416 |
|
417 |
# Check if locked instances are still correct
|
418 |
CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instance_names) |
419 |
|
420 |
self.group = self.cfg.GetNodeGroup(self.group_uuid) |
421 |
cluster = self.cfg.GetClusterInfo()
|
422 |
|
423 |
if self.group is None: |
424 |
raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" % |
425 |
(self.op.group_name, self.group_uuid)) |
426 |
|
427 |
if self.op.ndparams: |
428 |
new_ndparams = GetUpdatedParams(self.group.ndparams, self.op.ndparams) |
429 |
utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES) |
430 |
self.new_ndparams = new_ndparams
|
431 |
|
432 |
if self.op.diskparams: |
433 |
diskparams = self.group.diskparams
|
434 |
uavdp = self._UpdateAndVerifyDiskParams
|
435 |
# For each disktemplate subdict update and verify the values
|
436 |
new_diskparams = dict((dt,
|
437 |
uavdp(diskparams.get(dt, {}), |
438 |
self.op.diskparams[dt]))
|
439 |
for dt in constants.DISK_TEMPLATES |
440 |
if dt in self.op.diskparams) |
441 |
# As we've all subdicts of diskparams ready, lets merge the actual
|
442 |
# dict with all updated subdicts
|
443 |
self.new_diskparams = objects.FillDict(diskparams, new_diskparams)
|
444 |
|
445 |
try:
|
446 |
utils.VerifyDictOptions(self.new_diskparams, constants.DISK_DT_DEFAULTS)
|
447 |
CheckDiskAccessModeConsistency(self.new_diskparams, self.cfg, |
448 |
group=self.group)
|
449 |
except errors.OpPrereqError, err:
|
450 |
raise errors.OpPrereqError("While verify diskparams options: %s" % err, |
451 |
errors.ECODE_INVAL) |
452 |
|
453 |
if self.op.hv_state: |
454 |
self.new_hv_state = MergeAndVerifyHvState(self.op.hv_state, |
455 |
self.group.hv_state_static)
|
456 |
|
457 |
if self.op.disk_state: |
458 |
self.new_disk_state = \
|
459 |
MergeAndVerifyDiskState(self.op.disk_state,
|
460 |
self.group.disk_state_static)
|
461 |
|
462 |
self._CheckIpolicy(cluster, owned_instance_names)
|
463 |
|
464 |
def BuildHooksEnv(self): |
465 |
"""Build hooks env.
|
466 |
|
467 |
"""
|
468 |
return {
|
469 |
"GROUP_NAME": self.op.group_name, |
470 |
"NEW_ALLOC_POLICY": self.op.alloc_policy, |
471 |
} |
472 |
|
473 |
def BuildHooksNodes(self): |
474 |
"""Build hooks nodes.
|
475 |
|
476 |
"""
|
477 |
mn = self.cfg.GetMasterNode()
|
478 |
return ([mn], [mn])
|
479 |
|
480 |
def Exec(self, feedback_fn): |
481 |
"""Modifies the node group.
|
482 |
|
483 |
"""
|
484 |
result = [] |
485 |
|
486 |
if self.op.ndparams: |
487 |
self.group.ndparams = self.new_ndparams |
488 |
result.append(("ndparams", str(self.group.ndparams))) |
489 |
|
490 |
if self.op.diskparams: |
491 |
self.group.diskparams = self.new_diskparams |
492 |
result.append(("diskparams", str(self.group.diskparams))) |
493 |
|
494 |
if self.op.alloc_policy: |
495 |
self.group.alloc_policy = self.op.alloc_policy |
496 |
|
497 |
if self.op.hv_state: |
498 |
self.group.hv_state_static = self.new_hv_state |
499 |
|
500 |
if self.op.disk_state: |
501 |
self.group.disk_state_static = self.new_disk_state |
502 |
|
503 |
if self.op.ipolicy: |
504 |
self.group.ipolicy = self.new_ipolicy |
505 |
|
506 |
self.cfg.Update(self.group, feedback_fn) |
507 |
return result
|
508 |
|
509 |
|
510 |
class LUGroupRemove(LogicalUnit): |
511 |
HPATH = "group-remove"
|
512 |
HTYPE = constants.HTYPE_GROUP |
513 |
REQ_BGL = False
|
514 |
|
515 |
def ExpandNames(self): |
516 |
# This will raises errors.OpPrereqError on its own:
|
517 |
self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name) |
518 |
self.needed_locks = {
|
519 |
locking.LEVEL_NODEGROUP: [self.group_uuid],
|
520 |
} |
521 |
|
522 |
def CheckPrereq(self): |
523 |
"""Check prerequisites.
|
524 |
|
525 |
This checks that the given group name exists as a node group, that is
|
526 |
empty (i.e., contains no nodes), and that is not the last group of the
|
527 |
cluster.
|
528 |
|
529 |
"""
|
530 |
# Verify that the group is empty.
|
531 |
group_nodes = [node.uuid |
532 |
for node in self.cfg.GetAllNodesInfo().values() |
533 |
if node.group == self.group_uuid] |
534 |
|
535 |
if group_nodes:
|
536 |
raise errors.OpPrereqError("Group '%s' not empty, has the following" |
537 |
" nodes: %s" %
|
538 |
(self.op.group_name,
|
539 |
utils.CommaJoin(utils.NiceSort(group_nodes))), |
540 |
errors.ECODE_STATE) |
541 |
|
542 |
# Verify the cluster would not be left group-less.
|
543 |
if len(self.cfg.GetNodeGroupList()) == 1: |
544 |
raise errors.OpPrereqError("Group '%s' is the only group, cannot be" |
545 |
" removed" % self.op.group_name, |
546 |
errors.ECODE_STATE) |
547 |
|
548 |
def BuildHooksEnv(self): |
549 |
"""Build hooks env.
|
550 |
|
551 |
"""
|
552 |
return {
|
553 |
"GROUP_NAME": self.op.group_name, |
554 |
} |
555 |
|
556 |
def BuildHooksNodes(self): |
557 |
"""Build hooks nodes.
|
558 |
|
559 |
"""
|
560 |
mn = self.cfg.GetMasterNode()
|
561 |
return ([mn], [mn])
|
562 |
|
563 |
def Exec(self, feedback_fn): |
564 |
"""Remove the node group.
|
565 |
|
566 |
"""
|
567 |
try:
|
568 |
self.cfg.RemoveNodeGroup(self.group_uuid) |
569 |
except errors.ConfigurationError:
|
570 |
raise errors.OpExecError("Group '%s' with UUID %s disappeared" % |
571 |
(self.op.group_name, self.group_uuid)) |
572 |
|
573 |
self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid |
574 |
|
575 |
|
576 |
class LUGroupRename(LogicalUnit): |
577 |
HPATH = "group-rename"
|
578 |
HTYPE = constants.HTYPE_GROUP |
579 |
REQ_BGL = False
|
580 |
|
581 |
def ExpandNames(self): |
582 |
# This raises errors.OpPrereqError on its own:
|
583 |
self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name) |
584 |
|
585 |
self.needed_locks = {
|
586 |
locking.LEVEL_NODEGROUP: [self.group_uuid],
|
587 |
} |
588 |
|
589 |
def CheckPrereq(self): |
590 |
"""Check prerequisites.
|
591 |
|
592 |
Ensures requested new name is not yet used.
|
593 |
|
594 |
"""
|
595 |
try:
|
596 |
new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name) |
597 |
except errors.OpPrereqError:
|
598 |
pass
|
599 |
else:
|
600 |
raise errors.OpPrereqError("Desired new name '%s' clashes with existing" |
601 |
" node group (UUID: %s)" %
|
602 |
(self.op.new_name, new_name_uuid),
|
603 |
errors.ECODE_EXISTS) |
604 |
|
605 |
def BuildHooksEnv(self): |
606 |
"""Build hooks env.
|
607 |
|
608 |
"""
|
609 |
return {
|
610 |
"OLD_NAME": self.op.group_name, |
611 |
"NEW_NAME": self.op.new_name, |
612 |
} |
613 |
|
614 |
def BuildHooksNodes(self): |
615 |
"""Build hooks nodes.
|
616 |
|
617 |
"""
|
618 |
mn = self.cfg.GetMasterNode()
|
619 |
|
620 |
all_nodes = self.cfg.GetAllNodesInfo()
|
621 |
all_nodes.pop(mn, None)
|
622 |
|
623 |
run_nodes = [mn] |
624 |
run_nodes.extend(node.uuid for node in all_nodes.values() |
625 |
if node.group == self.group_uuid) |
626 |
|
627 |
return (run_nodes, run_nodes)
|
628 |
|
629 |
def Exec(self, feedback_fn): |
630 |
"""Rename the node group.
|
631 |
|
632 |
"""
|
633 |
group = self.cfg.GetNodeGroup(self.group_uuid) |
634 |
|
635 |
if group is None: |
636 |
raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" % |
637 |
(self.op.group_name, self.group_uuid)) |
638 |
|
639 |
group.name = self.op.new_name
|
640 |
self.cfg.Update(group, feedback_fn)
|
641 |
|
642 |
return self.op.new_name |
643 |
|
644 |
|
645 |
class LUGroupEvacuate(LogicalUnit): |
646 |
HPATH = "group-evacuate"
|
647 |
HTYPE = constants.HTYPE_GROUP |
648 |
REQ_BGL = False
|
649 |
|
650 |
def ExpandNames(self): |
651 |
# This raises errors.OpPrereqError on its own:
|
652 |
self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name) |
653 |
|
654 |
if self.op.target_groups: |
655 |
self.req_target_uuids = map(self.cfg.LookupNodeGroup, |
656 |
self.op.target_groups)
|
657 |
else:
|
658 |
self.req_target_uuids = []
|
659 |
|
660 |
if self.group_uuid in self.req_target_uuids: |
661 |
raise errors.OpPrereqError("Group to be evacuated (%s) can not be used" |
662 |
" as a target group (targets are %s)" %
|
663 |
(self.group_uuid,
|
664 |
utils.CommaJoin(self.req_target_uuids)),
|
665 |
errors.ECODE_INVAL) |
666 |
|
667 |
self.op.iallocator = GetDefaultIAllocator(self.cfg, self.op.iallocator) |
668 |
|
669 |
self.share_locks = ShareAll()
|
670 |
self.needed_locks = {
|
671 |
locking.LEVEL_INSTANCE: [], |
672 |
locking.LEVEL_NODEGROUP: [], |
673 |
locking.LEVEL_NODE: [], |
674 |
} |
675 |
|
676 |
def DeclareLocks(self, level): |
677 |
if level == locking.LEVEL_INSTANCE:
|
678 |
assert not self.needed_locks[locking.LEVEL_INSTANCE] |
679 |
|
680 |
# Lock instances optimistically, needs verification once node and group
|
681 |
# locks have been acquired
|
682 |
self.needed_locks[locking.LEVEL_INSTANCE] = \
|
683 |
self.cfg.GetInstanceNames(
|
684 |
self.cfg.GetNodeGroupInstances(self.group_uuid)) |
685 |
|
686 |
elif level == locking.LEVEL_NODEGROUP:
|
687 |
assert not self.needed_locks[locking.LEVEL_NODEGROUP] |
688 |
|
689 |
if self.req_target_uuids: |
690 |
lock_groups = set([self.group_uuid] + self.req_target_uuids) |
691 |
|
692 |
# Lock all groups used by instances optimistically; this requires going
|
693 |
# via the node before it's locked, requiring verification later on
|
694 |
lock_groups.update(group_uuid |
695 |
for instance_name in |
696 |
self.owned_locks(locking.LEVEL_INSTANCE)
|
697 |
for group_uuid in |
698 |
self.cfg.GetInstanceNodeGroups(
|
699 |
self.cfg.GetInstanceInfoByName(instance_name)
|
700 |
.uuid)) |
701 |
else:
|
702 |
# No target groups, need to lock all of them
|
703 |
lock_groups = locking.ALL_SET |
704 |
|
705 |
self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
|
706 |
|
707 |
elif level == locking.LEVEL_NODE:
|
708 |
# This will only lock the nodes in the group to be evacuated which
|
709 |
# contain actual instances
|
710 |
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
|
711 |
self._LockInstancesNodes()
|
712 |
|
713 |
# Lock all nodes in group to be evacuated and target groups
|
714 |
owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) |
715 |
assert self.group_uuid in owned_groups |
716 |
member_node_uuids = [node_uuid |
717 |
for group in owned_groups |
718 |
for node_uuid in |
719 |
self.cfg.GetNodeGroup(group).members]
|
720 |
self.needed_locks[locking.LEVEL_NODE].extend(member_node_uuids)
|
721 |
|
722 |
def CheckPrereq(self): |
723 |
owned_instance_names = frozenset(self.owned_locks(locking.LEVEL_INSTANCE)) |
724 |
owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) |
725 |
owned_node_uuids = frozenset(self.owned_locks(locking.LEVEL_NODE)) |
726 |
|
727 |
assert owned_groups.issuperset(self.req_target_uuids) |
728 |
assert self.group_uuid in owned_groups |
729 |
|
730 |
# Check if locked instances are still correct
|
731 |
CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instance_names) |
732 |
|
733 |
# Get instance information
|
734 |
self.instances = \
|
735 |
dict(self.cfg.GetMultiInstanceInfoByName(owned_instance_names)) |
736 |
|
737 |
# Check if node groups for locked instances are still correct
|
738 |
CheckInstancesNodeGroups(self.cfg, self.instances, |
739 |
owned_groups, owned_node_uuids, self.group_uuid)
|
740 |
|
741 |
if self.req_target_uuids: |
742 |
# User requested specific target groups
|
743 |
self.target_uuids = self.req_target_uuids |
744 |
else:
|
745 |
# All groups except the one to be evacuated are potential targets
|
746 |
self.target_uuids = [group_uuid for group_uuid in owned_groups |
747 |
if group_uuid != self.group_uuid] |
748 |
|
749 |
if not self.target_uuids: |
750 |
raise errors.OpPrereqError("There are no possible target groups", |
751 |
errors.ECODE_INVAL) |
752 |
|
753 |
def BuildHooksEnv(self): |
754 |
"""Build hooks env.
|
755 |
|
756 |
"""
|
757 |
return {
|
758 |
"GROUP_NAME": self.op.group_name, |
759 |
"TARGET_GROUPS": " ".join(self.target_uuids), |
760 |
} |
761 |
|
762 |
def BuildHooksNodes(self): |
763 |
"""Build hooks nodes.
|
764 |
|
765 |
"""
|
766 |
mn = self.cfg.GetMasterNode()
|
767 |
|
768 |
assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP) |
769 |
|
770 |
run_nodes = [mn] + self.cfg.GetNodeGroup(self.group_uuid).members |
771 |
|
772 |
return (run_nodes, run_nodes)
|
773 |
|
774 |
def Exec(self, feedback_fn): |
775 |
inst_names = list(self.owned_locks(locking.LEVEL_INSTANCE)) |
776 |
|
777 |
assert self.group_uuid not in self.target_uuids |
778 |
|
779 |
req = iallocator.IAReqGroupChange(instances=inst_names, |
780 |
target_groups=self.target_uuids)
|
781 |
ial = iallocator.IAllocator(self.cfg, self.rpc, req) |
782 |
|
783 |
ial.Run(self.op.iallocator)
|
784 |
|
785 |
if not ial.success: |
786 |
raise errors.OpPrereqError("Can't compute group evacuation using" |
787 |
" iallocator '%s': %s" %
|
788 |
(self.op.iallocator, ial.info),
|
789 |
errors.ECODE_NORES) |
790 |
|
791 |
jobs = LoadNodeEvacResult(self, ial.result, self.op.early_release, False) |
792 |
|
793 |
self.LogInfo("Iallocator returned %s job(s) for evacuating node group %s", |
794 |
len(jobs), self.op.group_name) |
795 |
|
796 |
return ResultWithJobs(jobs)
|
797 |
|
798 |
|
799 |
class LUGroupVerifyDisks(NoHooksLU): |
800 |
"""Verifies the status of all disks in a node group.
|
801 |
|
802 |
"""
|
803 |
REQ_BGL = False
|
804 |
|
805 |
def ExpandNames(self): |
806 |
# Raises errors.OpPrereqError on its own if group can't be found
|
807 |
self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name) |
808 |
|
809 |
self.share_locks = ShareAll()
|
810 |
self.needed_locks = {
|
811 |
locking.LEVEL_INSTANCE: [], |
812 |
locking.LEVEL_NODEGROUP: [], |
813 |
locking.LEVEL_NODE: [], |
814 |
|
815 |
# This opcode is acquires all node locks in a group. LUClusterVerifyDisks
|
816 |
# starts one instance of this opcode for every group, which means all
|
817 |
# nodes will be locked for a short amount of time, so it's better to
|
818 |
# acquire the node allocation lock as well.
|
819 |
locking.LEVEL_NODE_ALLOC: locking.ALL_SET, |
820 |
} |
821 |
|
822 |
def DeclareLocks(self, level): |
823 |
if level == locking.LEVEL_INSTANCE:
|
824 |
assert not self.needed_locks[locking.LEVEL_INSTANCE] |
825 |
|
826 |
# Lock instances optimistically, needs verification once node and group
|
827 |
# locks have been acquired
|
828 |
self.needed_locks[locking.LEVEL_INSTANCE] = \
|
829 |
self.cfg.GetInstanceNames(
|
830 |
self.cfg.GetNodeGroupInstances(self.group_uuid)) |
831 |
|
832 |
elif level == locking.LEVEL_NODEGROUP:
|
833 |
assert not self.needed_locks[locking.LEVEL_NODEGROUP] |
834 |
|
835 |
self.needed_locks[locking.LEVEL_NODEGROUP] = \
|
836 |
set([self.group_uuid] + |
837 |
# Lock all groups used by instances optimistically; this requires
|
838 |
# going via the node before it's locked, requiring verification
|
839 |
# later on
|
840 |
[group_uuid |
841 |
for instance_name in self.owned_locks(locking.LEVEL_INSTANCE) |
842 |
for group_uuid in |
843 |
self.cfg.GetInstanceNodeGroups(
|
844 |
self.cfg.GetInstanceInfoByName(instance_name).uuid)])
|
845 |
|
846 |
elif level == locking.LEVEL_NODE:
|
847 |
# This will only lock the nodes in the group to be verified which contain
|
848 |
# actual instances
|
849 |
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
|
850 |
self._LockInstancesNodes()
|
851 |
|
852 |
# Lock all nodes in group to be verified
|
853 |
assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP) |
854 |
member_node_uuids = self.cfg.GetNodeGroup(self.group_uuid).members |
855 |
self.needed_locks[locking.LEVEL_NODE].extend(member_node_uuids)
|
856 |
|
857 |
def CheckPrereq(self): |
858 |
owned_inst_names = frozenset(self.owned_locks(locking.LEVEL_INSTANCE)) |
859 |
owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) |
860 |
owned_node_uuids = frozenset(self.owned_locks(locking.LEVEL_NODE)) |
861 |
|
862 |
assert self.group_uuid in owned_groups |
863 |
|
864 |
# Check if locked instances are still correct
|
865 |
CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_inst_names) |
866 |
|
867 |
# Get instance information
|
868 |
self.instances = dict(self.cfg.GetMultiInstanceInfoByName(owned_inst_names)) |
869 |
|
870 |
# Check if node groups for locked instances are still correct
|
871 |
CheckInstancesNodeGroups(self.cfg, self.instances, |
872 |
owned_groups, owned_node_uuids, self.group_uuid)
|
873 |
|
874 |
def _VerifyInstanceLvs(self, node_errors, offline_disk_instance_names, |
875 |
missing_disks): |
876 |
node_lv_to_inst = MapInstanceLvsToNodes( |
877 |
[inst for inst in self.instances.values() if inst.disks_active]) |
878 |
if node_lv_to_inst:
|
879 |
node_uuids = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) & |
880 |
set(self.cfg.GetVmCapableNodeList())) |
881 |
|
882 |
node_lvs = self.rpc.call_lv_list(node_uuids, [])
|
883 |
|
884 |
for (node_uuid, node_res) in node_lvs.items(): |
885 |
if node_res.offline:
|
886 |
continue
|
887 |
|
888 |
msg = node_res.fail_msg |
889 |
if msg:
|
890 |
logging.warning("Error enumerating LVs on node %s: %s",
|
891 |
self.cfg.GetNodeName(node_uuid), msg)
|
892 |
node_errors[node_uuid] = msg |
893 |
continue
|
894 |
|
895 |
for lv_name, (_, _, lv_online) in node_res.payload.items(): |
896 |
inst = node_lv_to_inst.pop((node_uuid, lv_name), None)
|
897 |
if not lv_online and inst is not None: |
898 |
offline_disk_instance_names.add(inst.name) |
899 |
|
900 |
# any leftover items in nv_dict are missing LVs, let's arrange the data
|
901 |
# better
|
902 |
for key, inst in node_lv_to_inst.iteritems(): |
903 |
missing_disks.setdefault(inst.name, []).append(list(key))
|
904 |
|
905 |
def _VerifyDrbdStates(self, node_errors, offline_disk_instance_names): |
906 |
node_to_inst = {} |
907 |
for inst in self.instances.values(): |
908 |
if not inst.disks_active or inst.disk_template != constants.DT_DRBD8: |
909 |
continue
|
910 |
|
911 |
for node_uuid in itertools.chain([inst.primary_node], |
912 |
inst.secondary_nodes): |
913 |
node_to_inst.setdefault(node_uuid, []).append(inst) |
914 |
|
915 |
for (node_uuid, insts) in node_to_inst.items(): |
916 |
node_disks = [(inst.disks, inst) for inst in insts] |
917 |
node_res = self.rpc.call_drbd_needs_activation(node_uuid, node_disks)
|
918 |
msg = node_res.fail_msg |
919 |
if msg:
|
920 |
logging.warning("Error getting DRBD status on node %s: %s",
|
921 |
self.cfg.GetNodeName(node_uuid), msg)
|
922 |
node_errors[node_uuid] = msg |
923 |
continue
|
924 |
|
925 |
faulty_disk_uuids = set(node_res.payload)
|
926 |
for inst in self.instances.values(): |
927 |
inst_disk_uuids = set([disk.uuid for disk in inst.disks]) |
928 |
if inst_disk_uuids.intersection(faulty_disk_uuids):
|
929 |
offline_disk_instance_names.add(inst.name) |
930 |
|
931 |
def Exec(self, feedback_fn): |
932 |
"""Verify integrity of cluster disks.
|
933 |
|
934 |
@rtype: tuple of three items
|
935 |
@return: a tuple of (dict of node-to-node_error, list of instances
|
936 |
which need activate-disks, dict of instance: (node, volume) for
|
937 |
missing volumes
|
938 |
|
939 |
"""
|
940 |
node_errors = {} |
941 |
offline_disk_instance_names = set()
|
942 |
missing_disks = {} |
943 |
|
944 |
self._VerifyInstanceLvs(node_errors, offline_disk_instance_names,
|
945 |
missing_disks) |
946 |
self._VerifyDrbdStates(node_errors, offline_disk_instance_names)
|
947 |
|
948 |
return (node_errors, list(offline_disk_instance_names), missing_disks) |