Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ fe870648

History | View | Annotate | Download (441.3 kB)

1
#
2
#
3

    
4
# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 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
"""Module implementing the master-side code."""
23

    
24
# pylint: disable-msg=W0201,C0302
25

    
26
# W0201 since most LU attributes are defined in CheckPrereq or similar
27
# functions
28

    
29
# C0302: since we have waaaay to many lines in this module
30

    
31
import os
32
import os.path
33
import time
34
import re
35
import platform
36
import logging
37
import copy
38
import OpenSSL
39
import socket
40
import tempfile
41
import shutil
42
import itertools
43

    
44
from ganeti import ssh
45
from ganeti import utils
46
from ganeti import errors
47
from ganeti import hypervisor
48
from ganeti import locking
49
from ganeti import constants
50
from ganeti import objects
51
from ganeti import serializer
52
from ganeti import ssconf
53
from ganeti import uidpool
54
from ganeti import compat
55
from ganeti import masterd
56
from ganeti import netutils
57
from ganeti import query
58
from ganeti import qlang
59
from ganeti import opcodes
60
from ganeti import ht
61

    
62
import ganeti.masterd.instance # pylint: disable-msg=W0611
63

    
64

    
65
def _SupportsOob(cfg, node):
66
  """Tells if node supports OOB.
67

68
  @type cfg: L{config.ConfigWriter}
69
  @param cfg: The cluster configuration
70
  @type node: L{objects.Node}
71
  @param node: The node
72
  @return: The OOB script if supported or an empty string otherwise
73

74
  """
75
  return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM]
76

    
77

    
78
class ResultWithJobs:
79
  """Data container for LU results with jobs.
80

81
  Instances of this class returned from L{LogicalUnit.Exec} will be recognized
82
  by L{mcpu.Processor._ProcessResult}. The latter will then submit the jobs
83
  contained in the C{jobs} attribute and include the job IDs in the opcode
84
  result.
85

86
  """
87
  def __init__(self, jobs, **kwargs):
88
    """Initializes this class.
89

90
    Additional return values can be specified as keyword arguments.
91

92
    @type jobs: list of lists of L{opcode.OpCode}
93
    @param jobs: A list of lists of opcode objects
94

95
    """
96
    self.jobs = jobs
97
    self.other = kwargs
98

    
99

    
100
class LogicalUnit(object):
101
  """Logical Unit base class.
102

103
  Subclasses must follow these rules:
104
    - implement ExpandNames
105
    - implement CheckPrereq (except when tasklets are used)
106
    - implement Exec (except when tasklets are used)
107
    - implement BuildHooksEnv
108
    - implement BuildHooksNodes
109
    - redefine HPATH and HTYPE
110
    - optionally redefine their run requirements:
111
        REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
112

113
  Note that all commands require root permissions.
114

115
  @ivar dry_run_result: the value (if any) that will be returned to the caller
116
      in dry-run mode (signalled by opcode dry_run parameter)
117

118
  """
119
  HPATH = None
120
  HTYPE = None
121
  REQ_BGL = True
122

    
123
  def __init__(self, processor, op, context, rpc):
124
    """Constructor for LogicalUnit.
125

126
    This needs to be overridden in derived classes in order to check op
127
    validity.
128

129
    """
130
    self.proc = processor
131
    self.op = op
132
    self.cfg = context.cfg
133
    self.glm = context.glm
134
    self.context = context
135
    self.rpc = rpc
136
    # Dicts used to declare locking needs to mcpu
137
    self.needed_locks = None
138
    self.share_locks = dict.fromkeys(locking.LEVELS, 0)
139
    self.add_locks = {}
140
    self.remove_locks = {}
141
    # Used to force good behavior when calling helper functions
142
    self.recalculate_locks = {}
143
    # logging
144
    self.Log = processor.Log # pylint: disable-msg=C0103
145
    self.LogWarning = processor.LogWarning # pylint: disable-msg=C0103
146
    self.LogInfo = processor.LogInfo # pylint: disable-msg=C0103
147
    self.LogStep = processor.LogStep # pylint: disable-msg=C0103
148
    # support for dry-run
149
    self.dry_run_result = None
150
    # support for generic debug attribute
151
    if (not hasattr(self.op, "debug_level") or
152
        not isinstance(self.op.debug_level, int)):
153
      self.op.debug_level = 0
154

    
155
    # Tasklets
156
    self.tasklets = None
157

    
158
    # Validate opcode parameters and set defaults
159
    self.op.Validate(True)
160

    
161
    self.CheckArguments()
162

    
163
  def CheckArguments(self):
164
    """Check syntactic validity for the opcode arguments.
165

166
    This method is for doing a simple syntactic check and ensure
167
    validity of opcode parameters, without any cluster-related
168
    checks. While the same can be accomplished in ExpandNames and/or
169
    CheckPrereq, doing these separate is better because:
170

171
      - ExpandNames is left as as purely a lock-related function
172
      - CheckPrereq is run after we have acquired locks (and possible
173
        waited for them)
174

175
    The function is allowed to change the self.op attribute so that
176
    later methods can no longer worry about missing parameters.
177

178
    """
179
    pass
180

    
181
  def ExpandNames(self):
182
    """Expand names for this LU.
183

184
    This method is called before starting to execute the opcode, and it should
185
    update all the parameters of the opcode to their canonical form (e.g. a
186
    short node name must be fully expanded after this method has successfully
187
    completed). This way locking, hooks, logging, etc. can work correctly.
188

189
    LUs which implement this method must also populate the self.needed_locks
190
    member, as a dict with lock levels as keys, and a list of needed lock names
191
    as values. Rules:
192

193
      - use an empty dict if you don't need any lock
194
      - if you don't need any lock at a particular level omit that level
195
      - don't put anything for the BGL level
196
      - if you want all locks at a level use locking.ALL_SET as a value
197

198
    If you need to share locks (rather than acquire them exclusively) at one
199
    level you can modify self.share_locks, setting a true value (usually 1) for
200
    that level. By default locks are not shared.
201

202
    This function can also define a list of tasklets, which then will be
203
    executed in order instead of the usual LU-level CheckPrereq and Exec
204
    functions, if those are not defined by the LU.
205

206
    Examples::
207

208
      # Acquire all nodes and one instance
209
      self.needed_locks = {
210
        locking.LEVEL_NODE: locking.ALL_SET,
211
        locking.LEVEL_INSTANCE: ['instance1.example.com'],
212
      }
213
      # Acquire just two nodes
214
      self.needed_locks = {
215
        locking.LEVEL_NODE: ['node1.example.com', 'node2.example.com'],
216
      }
217
      # Acquire no locks
218
      self.needed_locks = {} # No, you can't leave it to the default value None
219

220
    """
221
    # The implementation of this method is mandatory only if the new LU is
222
    # concurrent, so that old LUs don't need to be changed all at the same
223
    # time.
224
    if self.REQ_BGL:
225
      self.needed_locks = {} # Exclusive LUs don't need locks.
226
    else:
227
      raise NotImplementedError
228

    
229
  def DeclareLocks(self, level):
230
    """Declare LU locking needs for a level
231

232
    While most LUs can just declare their locking needs at ExpandNames time,
233
    sometimes there's the need to calculate some locks after having acquired
234
    the ones before. This function is called just before acquiring locks at a
235
    particular level, but after acquiring the ones at lower levels, and permits
236
    such calculations. It can be used to modify self.needed_locks, and by
237
    default it does nothing.
238

239
    This function is only called if you have something already set in
240
    self.needed_locks for the level.
241

242
    @param level: Locking level which is going to be locked
243
    @type level: member of ganeti.locking.LEVELS
244

245
    """
246

    
247
  def CheckPrereq(self):
248
    """Check prerequisites for this LU.
249

250
    This method should check that the prerequisites for the execution
251
    of this LU are fulfilled. It can do internode communication, but
252
    it should be idempotent - no cluster or system changes are
253
    allowed.
254

255
    The method should raise errors.OpPrereqError in case something is
256
    not fulfilled. Its return value is ignored.
257

258
    This method should also update all the parameters of the opcode to
259
    their canonical form if it hasn't been done by ExpandNames before.
260

261
    """
262
    if self.tasklets is not None:
263
      for (idx, tl) in enumerate(self.tasklets):
264
        logging.debug("Checking prerequisites for tasklet %s/%s",
265
                      idx + 1, len(self.tasklets))
266
        tl.CheckPrereq()
267
    else:
268
      pass
269

    
270
  def Exec(self, feedback_fn):
271
    """Execute the LU.
272

273
    This method should implement the actual work. It should raise
274
    errors.OpExecError for failures that are somewhat dealt with in
275
    code, or expected.
276

277
    """
278
    if self.tasklets is not None:
279
      for (idx, tl) in enumerate(self.tasklets):
280
        logging.debug("Executing tasklet %s/%s", idx + 1, len(self.tasklets))
281
        tl.Exec(feedback_fn)
282
    else:
283
      raise NotImplementedError
284

    
285
  def BuildHooksEnv(self):
286
    """Build hooks environment for this LU.
287

288
    @rtype: dict
289
    @return: Dictionary containing the environment that will be used for
290
      running the hooks for this LU. The keys of the dict must not be prefixed
291
      with "GANETI_"--that'll be added by the hooks runner. The hooks runner
292
      will extend the environment with additional variables. If no environment
293
      should be defined, an empty dictionary should be returned (not C{None}).
294
    @note: If the C{HPATH} attribute of the LU class is C{None}, this function
295
      will not be called.
296

297
    """
298
    raise NotImplementedError
299

    
300
  def BuildHooksNodes(self):
301
    """Build list of nodes to run LU's hooks.
302

303
    @rtype: tuple; (list, list)
304
    @return: Tuple containing a list of node names on which the hook
305
      should run before the execution and a list of node names on which the
306
      hook should run after the execution. No nodes should be returned as an
307
      empty list (and not None).
308
    @note: If the C{HPATH} attribute of the LU class is C{None}, this function
309
      will not be called.
310

311
    """
312
    raise NotImplementedError
313

    
314
  def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
315
    """Notify the LU about the results of its hooks.
316

317
    This method is called every time a hooks phase is executed, and notifies
318
    the Logical Unit about the hooks' result. The LU can then use it to alter
319
    its result based on the hooks.  By default the method does nothing and the
320
    previous result is passed back unchanged but any LU can define it if it
321
    wants to use the local cluster hook-scripts somehow.
322

323
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
324
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
325
    @param hook_results: the results of the multi-node hooks rpc call
326
    @param feedback_fn: function used send feedback back to the caller
327
    @param lu_result: the previous Exec result this LU had, or None
328
        in the PRE phase
329
    @return: the new Exec result, based on the previous result
330
        and hook results
331

332
    """
333
    # API must be kept, thus we ignore the unused argument and could
334
    # be a function warnings
335
    # pylint: disable-msg=W0613,R0201
336
    return lu_result
337

    
338
  def _ExpandAndLockInstance(self):
339
    """Helper function to expand and lock an instance.
340

341
    Many LUs that work on an instance take its name in self.op.instance_name
342
    and need to expand it and then declare the expanded name for locking. This
343
    function does it, and then updates self.op.instance_name to the expanded
344
    name. It also initializes needed_locks as a dict, if this hasn't been done
345
    before.
346

347
    """
348
    if self.needed_locks is None:
349
      self.needed_locks = {}
350
    else:
351
      assert locking.LEVEL_INSTANCE not in self.needed_locks, \
352
        "_ExpandAndLockInstance called with instance-level locks set"
353
    self.op.instance_name = _ExpandInstanceName(self.cfg,
354
                                                self.op.instance_name)
355
    self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
356

    
357
  def _LockInstancesNodes(self, primary_only=False):
358
    """Helper function to declare instances' nodes for locking.
359

360
    This function should be called after locking one or more instances to lock
361
    their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
362
    with all primary or secondary nodes for instances already locked and
363
    present in self.needed_locks[locking.LEVEL_INSTANCE].
364

365
    It should be called from DeclareLocks, and for safety only works if
366
    self.recalculate_locks[locking.LEVEL_NODE] is set.
367

368
    In the future it may grow parameters to just lock some instance's nodes, or
369
    to just lock primaries or secondary nodes, if needed.
370

371
    If should be called in DeclareLocks in a way similar to::
372

373
      if level == locking.LEVEL_NODE:
374
        self._LockInstancesNodes()
375

376
    @type primary_only: boolean
377
    @param primary_only: only lock primary nodes of locked instances
378

379
    """
380
    assert locking.LEVEL_NODE in self.recalculate_locks, \
381
      "_LockInstancesNodes helper function called with no nodes to recalculate"
382

    
383
    # TODO: check if we're really been called with the instance locks held
384

    
385
    # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
386
    # future we might want to have different behaviors depending on the value
387
    # of self.recalculate_locks[locking.LEVEL_NODE]
388
    wanted_nodes = []
389
    for instance_name in self.glm.list_owned(locking.LEVEL_INSTANCE):
390
      instance = self.context.cfg.GetInstanceInfo(instance_name)
391
      wanted_nodes.append(instance.primary_node)
392
      if not primary_only:
393
        wanted_nodes.extend(instance.secondary_nodes)
394

    
395
    if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
396
      self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
397
    elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
398
      self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
399

    
400
    del self.recalculate_locks[locking.LEVEL_NODE]
401

    
402

    
403
class NoHooksLU(LogicalUnit): # pylint: disable-msg=W0223
404
  """Simple LU which runs no hooks.
405

406
  This LU is intended as a parent for other LogicalUnits which will
407
  run no hooks, in order to reduce duplicate code.
408

409
  """
410
  HPATH = None
411
  HTYPE = None
412

    
413
  def BuildHooksEnv(self):
414
    """Empty BuildHooksEnv for NoHooksLu.
415

416
    This just raises an error.
417

418
    """
419
    raise AssertionError("BuildHooksEnv called for NoHooksLUs")
420

    
421
  def BuildHooksNodes(self):
422
    """Empty BuildHooksNodes for NoHooksLU.
423

424
    """
425
    raise AssertionError("BuildHooksNodes called for NoHooksLU")
426

    
427

    
428
class Tasklet:
429
  """Tasklet base class.
430

431
  Tasklets are subcomponents for LUs. LUs can consist entirely of tasklets or
432
  they can mix legacy code with tasklets. Locking needs to be done in the LU,
433
  tasklets know nothing about locks.
434

435
  Subclasses must follow these rules:
436
    - Implement CheckPrereq
437
    - Implement Exec
438

439
  """
440
  def __init__(self, lu):
441
    self.lu = lu
442

    
443
    # Shortcuts
444
    self.cfg = lu.cfg
445
    self.rpc = lu.rpc
446

    
447
  def CheckPrereq(self):
448
    """Check prerequisites for this tasklets.
449

450
    This method should check whether the prerequisites for the execution of
451
    this tasklet are fulfilled. It can do internode communication, but it
452
    should be idempotent - no cluster or system changes are allowed.
453

454
    The method should raise errors.OpPrereqError in case something is not
455
    fulfilled. Its return value is ignored.
456

457
    This method should also update all parameters to their canonical form if it
458
    hasn't been done before.
459

460
    """
461
    pass
462

    
463
  def Exec(self, feedback_fn):
464
    """Execute the tasklet.
465

466
    This method should implement the actual work. It should raise
467
    errors.OpExecError for failures that are somewhat dealt with in code, or
468
    expected.
469

470
    """
471
    raise NotImplementedError
472

    
473

    
474
class _QueryBase:
475
  """Base for query utility classes.
476

477
  """
478
  #: Attribute holding field definitions
479
  FIELDS = None
480

    
481
  def __init__(self, filter_, fields, use_locking):
482
    """Initializes this class.
483

484
    """
485
    self.use_locking = use_locking
486

    
487
    self.query = query.Query(self.FIELDS, fields, filter_=filter_,
488
                             namefield="name")
489
    self.requested_data = self.query.RequestedData()
490
    self.names = self.query.RequestedNames()
491

    
492
    # Sort only if no names were requested
493
    self.sort_by_name = not self.names
494

    
495
    self.do_locking = None
496
    self.wanted = None
497

    
498
  def _GetNames(self, lu, all_names, lock_level):
499
    """Helper function to determine names asked for in the query.
500

501
    """
502
    if self.do_locking:
503
      names = lu.glm.list_owned(lock_level)
504
    else:
505
      names = all_names
506

    
507
    if self.wanted == locking.ALL_SET:
508
      assert not self.names
509
      # caller didn't specify names, so ordering is not important
510
      return utils.NiceSort(names)
511

    
512
    # caller specified names and we must keep the same order
513
    assert self.names
514
    assert not self.do_locking or lu.glm.is_owned(lock_level)
515

    
516
    missing = set(self.wanted).difference(names)
517
    if missing:
518
      raise errors.OpExecError("Some items were removed before retrieving"
519
                               " their data: %s" % missing)
520

    
521
    # Return expanded names
522
    return self.wanted
523

    
524
  def ExpandNames(self, lu):
525
    """Expand names for this query.
526

527
    See L{LogicalUnit.ExpandNames}.
528

529
    """
530
    raise NotImplementedError()
531

    
532
  def DeclareLocks(self, lu, level):
533
    """Declare locks for this query.
534

535
    See L{LogicalUnit.DeclareLocks}.
536

537
    """
538
    raise NotImplementedError()
539

    
540
  def _GetQueryData(self, lu):
541
    """Collects all data for this query.
542

543
    @return: Query data object
544

545
    """
546
    raise NotImplementedError()
547

    
548
  def NewStyleQuery(self, lu):
549
    """Collect data and execute query.
550

551
    """
552
    return query.GetQueryResponse(self.query, self._GetQueryData(lu),
553
                                  sort_by_name=self.sort_by_name)
554

    
555
  def OldStyleQuery(self, lu):
556
    """Collect data and execute query.
557

558
    """
559
    return self.query.OldStyleQuery(self._GetQueryData(lu),
560
                                    sort_by_name=self.sort_by_name)
561

    
562

    
563
def _GetWantedNodes(lu, nodes):
564
  """Returns list of checked and expanded node names.
565

566
  @type lu: L{LogicalUnit}
567
  @param lu: the logical unit on whose behalf we execute
568
  @type nodes: list
569
  @param nodes: list of node names or None for all nodes
570
  @rtype: list
571
  @return: the list of nodes, sorted
572
  @raise errors.ProgrammerError: if the nodes parameter is wrong type
573

574
  """
575
  if nodes:
576
    return [_ExpandNodeName(lu.cfg, name) for name in nodes]
577

    
578
  return utils.NiceSort(lu.cfg.GetNodeList())
579

    
580

    
581
def _GetWantedInstances(lu, instances):
582
  """Returns list of checked and expanded instance names.
583

584
  @type lu: L{LogicalUnit}
585
  @param lu: the logical unit on whose behalf we execute
586
  @type instances: list
587
  @param instances: list of instance names or None for all instances
588
  @rtype: list
589
  @return: the list of instances, sorted
590
  @raise errors.OpPrereqError: if the instances parameter is wrong type
591
  @raise errors.OpPrereqError: if any of the passed instances is not found
592

593
  """
594
  if instances:
595
    wanted = [_ExpandInstanceName(lu.cfg, name) for name in instances]
596
  else:
597
    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
598
  return wanted
599

    
600

    
601
def _GetUpdatedParams(old_params, update_dict,
602
                      use_default=True, use_none=False):
603
  """Return the new version of a parameter dictionary.
604

605
  @type old_params: dict
606
  @param old_params: old parameters
607
  @type update_dict: dict
608
  @param update_dict: dict containing new parameter values, or
609
      constants.VALUE_DEFAULT to reset the parameter to its default
610
      value
611
  @param use_default: boolean
612
  @type use_default: whether to recognise L{constants.VALUE_DEFAULT}
613
      values as 'to be deleted' values
614
  @param use_none: boolean
615
  @type use_none: whether to recognise C{None} values as 'to be
616
      deleted' values
617
  @rtype: dict
618
  @return: the new parameter dictionary
619

620
  """
621
  params_copy = copy.deepcopy(old_params)
622
  for key, val in update_dict.iteritems():
623
    if ((use_default and val == constants.VALUE_DEFAULT) or
624
        (use_none and val is None)):
625
      try:
626
        del params_copy[key]
627
      except KeyError:
628
        pass
629
    else:
630
      params_copy[key] = val
631
  return params_copy
632

    
633

    
634
def _ReleaseLocks(lu, level, names=None, keep=None):
635
  """Releases locks owned by an LU.
636

637
  @type lu: L{LogicalUnit}
638
  @param level: Lock level
639
  @type names: list or None
640
  @param names: Names of locks to release
641
  @type keep: list or None
642
  @param keep: Names of locks to retain
643

644
  """
645
  assert not (keep is not None and names is not None), \
646
         "Only one of the 'names' and the 'keep' parameters can be given"
647

    
648
  if names is not None:
649
    should_release = names.__contains__
650
  elif keep:
651
    should_release = lambda name: name not in keep
652
  else:
653
    should_release = None
654

    
655
  if should_release:
656
    retain = []
657
    release = []
658

    
659
    # Determine which locks to release
660
    for name in lu.glm.list_owned(level):
661
      if should_release(name):
662
        release.append(name)
663
      else:
664
        retain.append(name)
665

    
666
    assert len(lu.glm.list_owned(level)) == (len(retain) + len(release))
667

    
668
    # Release just some locks
669
    lu.glm.release(level, names=release)
670

    
671
    assert frozenset(lu.glm.list_owned(level)) == frozenset(retain)
672
  else:
673
    # Release everything
674
    lu.glm.release(level)
675

    
676
    assert not lu.glm.is_owned(level), "No locks should be owned"
677

    
678

    
679
def _RunPostHook(lu, node_name):
680
  """Runs the post-hook for an opcode on a single node.
681

682
  """
683
  hm = lu.proc.hmclass(lu.rpc.call_hooks_runner, lu)
684
  try:
685
    hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
686
  except:
687
    # pylint: disable-msg=W0702
688
    lu.LogWarning("Errors occurred running hooks on %s" % node_name)
689

    
690

    
691
def _CheckOutputFields(static, dynamic, selected):
692
  """Checks whether all selected fields are valid.
693

694
  @type static: L{utils.FieldSet}
695
  @param static: static fields set
696
  @type dynamic: L{utils.FieldSet}
697
  @param dynamic: dynamic fields set
698

699
  """
700
  f = utils.FieldSet()
701
  f.Extend(static)
702
  f.Extend(dynamic)
703

    
704
  delta = f.NonMatching(selected)
705
  if delta:
706
    raise errors.OpPrereqError("Unknown output fields selected: %s"
707
                               % ",".join(delta), errors.ECODE_INVAL)
708

    
709

    
710
def _CheckGlobalHvParams(params):
711
  """Validates that given hypervisor params are not global ones.
712

713
  This will ensure that instances don't get customised versions of
714
  global params.
715

716
  """
717
  used_globals = constants.HVC_GLOBALS.intersection(params)
718
  if used_globals:
719
    msg = ("The following hypervisor parameters are global and cannot"
720
           " be customized at instance level, please modify them at"
721
           " cluster level: %s" % utils.CommaJoin(used_globals))
722
    raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
723

    
724

    
725
def _CheckNodeOnline(lu, node, msg=None):
726
  """Ensure that a given node is online.
727

728
  @param lu: the LU on behalf of which we make the check
729
  @param node: the node to check
730
  @param msg: if passed, should be a message to replace the default one
731
  @raise errors.OpPrereqError: if the node is offline
732

733
  """
734
  if msg is None:
735
    msg = "Can't use offline node"
736
  if lu.cfg.GetNodeInfo(node).offline:
737
    raise errors.OpPrereqError("%s: %s" % (msg, node), errors.ECODE_STATE)
738

    
739

    
740
def _CheckNodeNotDrained(lu, node):
741
  """Ensure that a given node is not drained.
742

743
  @param lu: the LU on behalf of which we make the check
744
  @param node: the node to check
745
  @raise errors.OpPrereqError: if the node is drained
746

747
  """
748
  if lu.cfg.GetNodeInfo(node).drained:
749
    raise errors.OpPrereqError("Can't use drained node %s" % node,
750
                               errors.ECODE_STATE)
751

    
752

    
753
def _CheckNodeVmCapable(lu, node):
754
  """Ensure that a given node is vm capable.
755

756
  @param lu: the LU on behalf of which we make the check
757
  @param node: the node to check
758
  @raise errors.OpPrereqError: if the node is not vm capable
759

760
  """
761
  if not lu.cfg.GetNodeInfo(node).vm_capable:
762
    raise errors.OpPrereqError("Can't use non-vm_capable node %s" % node,
763
                               errors.ECODE_STATE)
764

    
765

    
766
def _CheckNodeHasOS(lu, node, os_name, force_variant):
767
  """Ensure that a node supports a given OS.
768

769
  @param lu: the LU on behalf of which we make the check
770
  @param node: the node to check
771
  @param os_name: the OS to query about
772
  @param force_variant: whether to ignore variant errors
773
  @raise errors.OpPrereqError: if the node is not supporting the OS
774

775
  """
776
  result = lu.rpc.call_os_get(node, os_name)
777
  result.Raise("OS '%s' not in supported OS list for node %s" %
778
               (os_name, node),
779
               prereq=True, ecode=errors.ECODE_INVAL)
780
  if not force_variant:
781
    _CheckOSVariant(result.payload, os_name)
782

    
783

    
784
def _CheckNodeHasSecondaryIP(lu, node, secondary_ip, prereq):
785
  """Ensure that a node has the given secondary ip.
786

787
  @type lu: L{LogicalUnit}
788
  @param lu: the LU on behalf of which we make the check
789
  @type node: string
790
  @param node: the node to check
791
  @type secondary_ip: string
792
  @param secondary_ip: the ip to check
793
  @type prereq: boolean
794
  @param prereq: whether to throw a prerequisite or an execute error
795
  @raise errors.OpPrereqError: if the node doesn't have the ip, and prereq=True
796
  @raise errors.OpExecError: if the node doesn't have the ip, and prereq=False
797

798
  """
799
  result = lu.rpc.call_node_has_ip_address(node, secondary_ip)
800
  result.Raise("Failure checking secondary ip on node %s" % node,
801
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
802
  if not result.payload:
803
    msg = ("Node claims it doesn't have the secondary ip you gave (%s),"
804
           " please fix and re-run this command" % secondary_ip)
805
    if prereq:
806
      raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
807
    else:
808
      raise errors.OpExecError(msg)
809

    
810

    
811
def _GetClusterDomainSecret():
812
  """Reads the cluster domain secret.
813

814
  """
815
  return utils.ReadOneLineFile(constants.CLUSTER_DOMAIN_SECRET_FILE,
816
                               strict=True)
817

    
818

    
819
def _CheckInstanceDown(lu, instance, reason):
820
  """Ensure that an instance is not running."""
821
  if instance.admin_up:
822
    raise errors.OpPrereqError("Instance %s is marked to be up, %s" %
823
                               (instance.name, reason), errors.ECODE_STATE)
824

    
825
  pnode = instance.primary_node
826
  ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
827
  ins_l.Raise("Can't contact node %s for instance information" % pnode,
828
              prereq=True, ecode=errors.ECODE_ENVIRON)
829

    
830
  if instance.name in ins_l.payload:
831
    raise errors.OpPrereqError("Instance %s is running, %s" %
832
                               (instance.name, reason), errors.ECODE_STATE)
833

    
834

    
835
def _ExpandItemName(fn, name, kind):
836
  """Expand an item name.
837

838
  @param fn: the function to use for expansion
839
  @param name: requested item name
840
  @param kind: text description ('Node' or 'Instance')
841
  @return: the resolved (full) name
842
  @raise errors.OpPrereqError: if the item is not found
843

844
  """
845
  full_name = fn(name)
846
  if full_name is None:
847
    raise errors.OpPrereqError("%s '%s' not known" % (kind, name),
848
                               errors.ECODE_NOENT)
849
  return full_name
850

    
851

    
852
def _ExpandNodeName(cfg, name):
853
  """Wrapper over L{_ExpandItemName} for nodes."""
854
  return _ExpandItemName(cfg.ExpandNodeName, name, "Node")
855

    
856

    
857
def _ExpandInstanceName(cfg, name):
858
  """Wrapper over L{_ExpandItemName} for instance."""
859
  return _ExpandItemName(cfg.ExpandInstanceName, name, "Instance")
860

    
861

    
862
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
863
                          memory, vcpus, nics, disk_template, disks,
864
                          bep, hvp, hypervisor_name):
865
  """Builds instance related env variables for hooks
866

867
  This builds the hook environment from individual variables.
868

869
  @type name: string
870
  @param name: the name of the instance
871
  @type primary_node: string
872
  @param primary_node: the name of the instance's primary node
873
  @type secondary_nodes: list
874
  @param secondary_nodes: list of secondary nodes as strings
875
  @type os_type: string
876
  @param os_type: the name of the instance's OS
877
  @type status: boolean
878
  @param status: the should_run status of the instance
879
  @type memory: string
880
  @param memory: the memory size of the instance
881
  @type vcpus: string
882
  @param vcpus: the count of VCPUs the instance has
883
  @type nics: list
884
  @param nics: list of tuples (ip, mac, mode, link) representing
885
      the NICs the instance has
886
  @type disk_template: string
887
  @param disk_template: the disk template of the instance
888
  @type disks: list
889
  @param disks: the list of (size, mode) pairs
890
  @type bep: dict
891
  @param bep: the backend parameters for the instance
892
  @type hvp: dict
893
  @param hvp: the hypervisor parameters for the instance
894
  @type hypervisor_name: string
895
  @param hypervisor_name: the hypervisor for the instance
896
  @rtype: dict
897
  @return: the hook environment for this instance
898

899
  """
900
  if status:
901
    str_status = "up"
902
  else:
903
    str_status = "down"
904
  env = {
905
    "OP_TARGET": name,
906
    "INSTANCE_NAME": name,
907
    "INSTANCE_PRIMARY": primary_node,
908
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
909
    "INSTANCE_OS_TYPE": os_type,
910
    "INSTANCE_STATUS": str_status,
911
    "INSTANCE_MEMORY": memory,
912
    "INSTANCE_VCPUS": vcpus,
913
    "INSTANCE_DISK_TEMPLATE": disk_template,
914
    "INSTANCE_HYPERVISOR": hypervisor_name,
915
  }
916

    
917
  if nics:
918
    nic_count = len(nics)
919
    for idx, (ip, mac, mode, link) in enumerate(nics):
920
      if ip is None:
921
        ip = ""
922
      env["INSTANCE_NIC%d_IP" % idx] = ip
923
      env["INSTANCE_NIC%d_MAC" % idx] = mac
924
      env["INSTANCE_NIC%d_MODE" % idx] = mode
925
      env["INSTANCE_NIC%d_LINK" % idx] = link
926
      if mode == constants.NIC_MODE_BRIDGED:
927
        env["INSTANCE_NIC%d_BRIDGE" % idx] = link
928
  else:
929
    nic_count = 0
930

    
931
  env["INSTANCE_NIC_COUNT"] = nic_count
932

    
933
  if disks:
934
    disk_count = len(disks)
935
    for idx, (size, mode) in enumerate(disks):
936
      env["INSTANCE_DISK%d_SIZE" % idx] = size
937
      env["INSTANCE_DISK%d_MODE" % idx] = mode
938
  else:
939
    disk_count = 0
940

    
941
  env["INSTANCE_DISK_COUNT"] = disk_count
942

    
943
  for source, kind in [(bep, "BE"), (hvp, "HV")]:
944
    for key, value in source.items():
945
      env["INSTANCE_%s_%s" % (kind, key)] = value
946

    
947
  return env
948

    
949

    
950
def _NICListToTuple(lu, nics):
951
  """Build a list of nic information tuples.
952

953
  This list is suitable to be passed to _BuildInstanceHookEnv or as a return
954
  value in LUInstanceQueryData.
955

956
  @type lu:  L{LogicalUnit}
957
  @param lu: the logical unit on whose behalf we execute
958
  @type nics: list of L{objects.NIC}
959
  @param nics: list of nics to convert to hooks tuples
960

961
  """
962
  hooks_nics = []
963
  cluster = lu.cfg.GetClusterInfo()
964
  for nic in nics:
965
    ip = nic.ip
966
    mac = nic.mac
967
    filled_params = cluster.SimpleFillNIC(nic.nicparams)
968
    mode = filled_params[constants.NIC_MODE]
969
    link = filled_params[constants.NIC_LINK]
970
    hooks_nics.append((ip, mac, mode, link))
971
  return hooks_nics
972

    
973

    
974
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
975
  """Builds instance related env variables for hooks from an object.
976

977
  @type lu: L{LogicalUnit}
978
  @param lu: the logical unit on whose behalf we execute
979
  @type instance: L{objects.Instance}
980
  @param instance: the instance for which we should build the
981
      environment
982
  @type override: dict
983
  @param override: dictionary with key/values that will override
984
      our values
985
  @rtype: dict
986
  @return: the hook environment dictionary
987

988
  """
989
  cluster = lu.cfg.GetClusterInfo()
990
  bep = cluster.FillBE(instance)
991
  hvp = cluster.FillHV(instance)
992
  args = {
993
    'name': instance.name,
994
    'primary_node': instance.primary_node,
995
    'secondary_nodes': instance.secondary_nodes,
996
    'os_type': instance.os,
997
    'status': instance.admin_up,
998
    'memory': bep[constants.BE_MEMORY],
999
    'vcpus': bep[constants.BE_VCPUS],
1000
    'nics': _NICListToTuple(lu, instance.nics),
1001
    'disk_template': instance.disk_template,
1002
    'disks': [(disk.size, disk.mode) for disk in instance.disks],
1003
    'bep': bep,
1004
    'hvp': hvp,
1005
    'hypervisor_name': instance.hypervisor,
1006
  }
1007
  if override:
1008
    args.update(override)
1009
  return _BuildInstanceHookEnv(**args) # pylint: disable-msg=W0142
1010

    
1011

    
1012
def _AdjustCandidatePool(lu, exceptions):
1013
  """Adjust the candidate pool after node operations.
1014

1015
  """
1016
  mod_list = lu.cfg.MaintainCandidatePool(exceptions)
1017
  if mod_list:
1018
    lu.LogInfo("Promoted nodes to master candidate role: %s",
1019
               utils.CommaJoin(node.name for node in mod_list))
1020
    for name in mod_list:
1021
      lu.context.ReaddNode(name)
1022
  mc_now, mc_max, _ = lu.cfg.GetMasterCandidateStats(exceptions)
1023
  if mc_now > mc_max:
1024
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
1025
               (mc_now, mc_max))
1026

    
1027

    
1028
def _DecideSelfPromotion(lu, exceptions=None):
1029
  """Decide whether I should promote myself as a master candidate.
1030

1031
  """
1032
  cp_size = lu.cfg.GetClusterInfo().candidate_pool_size
1033
  mc_now, mc_should, _ = lu.cfg.GetMasterCandidateStats(exceptions)
1034
  # the new node will increase mc_max with one, so:
1035
  mc_should = min(mc_should + 1, cp_size)
1036
  return mc_now < mc_should
1037

    
1038

    
1039
def _CheckNicsBridgesExist(lu, target_nics, target_node):
1040
  """Check that the brigdes needed by a list of nics exist.
1041

1042
  """
1043
  cluster = lu.cfg.GetClusterInfo()
1044
  paramslist = [cluster.SimpleFillNIC(nic.nicparams) for nic in target_nics]
1045
  brlist = [params[constants.NIC_LINK] for params in paramslist
1046
            if params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED]
1047
  if brlist:
1048
    result = lu.rpc.call_bridges_exist(target_node, brlist)
1049
    result.Raise("Error checking bridges on destination node '%s'" %
1050
                 target_node, prereq=True, ecode=errors.ECODE_ENVIRON)
1051

    
1052

    
1053
def _CheckInstanceBridgesExist(lu, instance, node=None):
1054
  """Check that the brigdes needed by an instance exist.
1055

1056
  """
1057
  if node is None:
1058
    node = instance.primary_node
1059
  _CheckNicsBridgesExist(lu, instance.nics, node)
1060

    
1061

    
1062
def _CheckOSVariant(os_obj, name):
1063
  """Check whether an OS name conforms to the os variants specification.
1064

1065
  @type os_obj: L{objects.OS}
1066
  @param os_obj: OS object to check
1067
  @type name: string
1068
  @param name: OS name passed by the user, to check for validity
1069

1070
  """
1071
  if not os_obj.supported_variants:
1072
    return
1073
  variant = objects.OS.GetVariant(name)
1074
  if not variant:
1075
    raise errors.OpPrereqError("OS name must include a variant",
1076
                               errors.ECODE_INVAL)
1077

    
1078
  if variant not in os_obj.supported_variants:
1079
    raise errors.OpPrereqError("Unsupported OS variant", errors.ECODE_INVAL)
1080

    
1081

    
1082
def _GetNodeInstancesInner(cfg, fn):
1083
  return [i for i in cfg.GetAllInstancesInfo().values() if fn(i)]
1084

    
1085

    
1086
def _GetNodeInstances(cfg, node_name):
1087
  """Returns a list of all primary and secondary instances on a node.
1088

1089
  """
1090

    
1091
  return _GetNodeInstancesInner(cfg, lambda inst: node_name in inst.all_nodes)
1092

    
1093

    
1094
def _GetNodePrimaryInstances(cfg, node_name):
1095
  """Returns primary instances on a node.
1096

1097
  """
1098
  return _GetNodeInstancesInner(cfg,
1099
                                lambda inst: node_name == inst.primary_node)
1100

    
1101

    
1102
def _GetNodeSecondaryInstances(cfg, node_name):
1103
  """Returns secondary instances on a node.
1104

1105
  """
1106
  return _GetNodeInstancesInner(cfg,
1107
                                lambda inst: node_name in inst.secondary_nodes)
1108

    
1109

    
1110
def _GetStorageTypeArgs(cfg, storage_type):
1111
  """Returns the arguments for a storage type.
1112

1113
  """
1114
  # Special case for file storage
1115
  if storage_type == constants.ST_FILE:
1116
    # storage.FileStorage wants a list of storage directories
1117
    return [[cfg.GetFileStorageDir(), cfg.GetSharedFileStorageDir()]]
1118

    
1119
  return []
1120

    
1121

    
1122
def _FindFaultyInstanceDisks(cfg, rpc, instance, node_name, prereq):
1123
  faulty = []
1124

    
1125
  for dev in instance.disks:
1126
    cfg.SetDiskID(dev, node_name)
1127

    
1128
  result = rpc.call_blockdev_getmirrorstatus(node_name, instance.disks)
1129
  result.Raise("Failed to get disk status from node %s" % node_name,
1130
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
1131

    
1132
  for idx, bdev_status in enumerate(result.payload):
1133
    if bdev_status and bdev_status.ldisk_status == constants.LDS_FAULTY:
1134
      faulty.append(idx)
1135

    
1136
  return faulty
1137

    
1138

    
1139
def _CheckIAllocatorOrNode(lu, iallocator_slot, node_slot):
1140
  """Check the sanity of iallocator and node arguments and use the
1141
  cluster-wide iallocator if appropriate.
1142

1143
  Check that at most one of (iallocator, node) is specified. If none is
1144
  specified, then the LU's opcode's iallocator slot is filled with the
1145
  cluster-wide default iallocator.
1146

1147
  @type iallocator_slot: string
1148
  @param iallocator_slot: the name of the opcode iallocator slot
1149
  @type node_slot: string
1150
  @param node_slot: the name of the opcode target node slot
1151

1152
  """
1153
  node = getattr(lu.op, node_slot, None)
1154
  iallocator = getattr(lu.op, iallocator_slot, None)
1155

    
1156
  if node is not None and iallocator is not None:
1157
    raise errors.OpPrereqError("Do not specify both, iallocator and node",
1158
                               errors.ECODE_INVAL)
1159
  elif node is None and iallocator is None:
1160
    default_iallocator = lu.cfg.GetDefaultIAllocator()
1161
    if default_iallocator:
1162
      setattr(lu.op, iallocator_slot, default_iallocator)
1163
    else:
1164
      raise errors.OpPrereqError("No iallocator or node given and no"
1165
                                 " cluster-wide default iallocator found;"
1166
                                 " please specify either an iallocator or a"
1167
                                 " node, or set a cluster-wide default"
1168
                                 " iallocator")
1169

    
1170

    
1171
class LUClusterPostInit(LogicalUnit):
1172
  """Logical unit for running hooks after cluster initialization.
1173

1174
  """
1175
  HPATH = "cluster-init"
1176
  HTYPE = constants.HTYPE_CLUSTER
1177

    
1178
  def BuildHooksEnv(self):
1179
    """Build hooks env.
1180

1181
    """
1182
    return {
1183
      "OP_TARGET": self.cfg.GetClusterName(),
1184
      }
1185

    
1186
  def BuildHooksNodes(self):
1187
    """Build hooks nodes.
1188

1189
    """
1190
    return ([], [self.cfg.GetMasterNode()])
1191

    
1192
  def Exec(self, feedback_fn):
1193
    """Nothing to do.
1194

1195
    """
1196
    return True
1197

    
1198

    
1199
class LUClusterDestroy(LogicalUnit):
1200
  """Logical unit for destroying the cluster.
1201

1202
  """
1203
  HPATH = "cluster-destroy"
1204
  HTYPE = constants.HTYPE_CLUSTER
1205

    
1206
  def BuildHooksEnv(self):
1207
    """Build hooks env.
1208

1209
    """
1210
    return {
1211
      "OP_TARGET": self.cfg.GetClusterName(),
1212
      }
1213

    
1214
  def BuildHooksNodes(self):
1215
    """Build hooks nodes.
1216

1217
    """
1218
    return ([], [])
1219

    
1220
  def CheckPrereq(self):
1221
    """Check prerequisites.
1222

1223
    This checks whether the cluster is empty.
1224

1225
    Any errors are signaled by raising errors.OpPrereqError.
1226

1227
    """
1228
    master = self.cfg.GetMasterNode()
1229

    
1230
    nodelist = self.cfg.GetNodeList()
1231
    if len(nodelist) != 1 or nodelist[0] != master:
1232
      raise errors.OpPrereqError("There are still %d node(s) in"
1233
                                 " this cluster." % (len(nodelist) - 1),
1234
                                 errors.ECODE_INVAL)
1235
    instancelist = self.cfg.GetInstanceList()
1236
    if instancelist:
1237
      raise errors.OpPrereqError("There are still %d instance(s) in"
1238
                                 " this cluster." % len(instancelist),
1239
                                 errors.ECODE_INVAL)
1240

    
1241
  def Exec(self, feedback_fn):
1242
    """Destroys the cluster.
1243

1244
    """
1245
    master = self.cfg.GetMasterNode()
1246

    
1247
    # Run post hooks on master node before it's removed
1248
    _RunPostHook(self, master)
1249

    
1250
    result = self.rpc.call_node_stop_master(master, False)
1251
    result.Raise("Could not disable the master role")
1252

    
1253
    return master
1254

    
1255

    
1256
def _VerifyCertificate(filename):
1257
  """Verifies a certificate for LUClusterVerifyConfig.
1258

1259
  @type filename: string
1260
  @param filename: Path to PEM file
1261

1262
  """
1263
  try:
1264
    cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
1265
                                           utils.ReadFile(filename))
1266
  except Exception, err: # pylint: disable-msg=W0703
1267
    return (LUClusterVerifyConfig.ETYPE_ERROR,
1268
            "Failed to load X509 certificate %s: %s" % (filename, err))
1269

    
1270
  (errcode, msg) = \
1271
    utils.VerifyX509Certificate(cert, constants.SSL_CERT_EXPIRATION_WARN,
1272
                                constants.SSL_CERT_EXPIRATION_ERROR)
1273

    
1274
  if msg:
1275
    fnamemsg = "While verifying %s: %s" % (filename, msg)
1276
  else:
1277
    fnamemsg = None
1278

    
1279
  if errcode is None:
1280
    return (None, fnamemsg)
1281
  elif errcode == utils.CERT_WARNING:
1282
    return (LUClusterVerifyConfig.ETYPE_WARNING, fnamemsg)
1283
  elif errcode == utils.CERT_ERROR:
1284
    return (LUClusterVerifyConfig.ETYPE_ERROR, fnamemsg)
1285

    
1286
  raise errors.ProgrammerError("Unhandled certificate error code %r" % errcode)
1287

    
1288

    
1289
def _GetAllHypervisorParameters(cluster, instances):
1290
  """Compute the set of all hypervisor parameters.
1291

1292
  @type cluster: L{objects.Cluster}
1293
  @param cluster: the cluster object
1294
  @param instances: list of L{objects.Instance}
1295
  @param instances: additional instances from which to obtain parameters
1296
  @rtype: list of (origin, hypervisor, parameters)
1297
  @return: a list with all parameters found, indicating the hypervisor they
1298
       apply to, and the origin (can be "cluster", "os X", or "instance Y")
1299

1300
  """
1301
  hvp_data = []
1302

    
1303
  for hv_name in cluster.enabled_hypervisors:
1304
    hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
1305

    
1306
  for os_name, os_hvp in cluster.os_hvp.items():
1307
    for hv_name, hv_params in os_hvp.items():
1308
      if hv_params:
1309
        full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
1310
        hvp_data.append(("os %s" % os_name, hv_name, full_params))
1311

    
1312
  # TODO: collapse identical parameter values in a single one
1313
  for instance in instances:
1314
    if instance.hvparams:
1315
      hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
1316
                       cluster.FillHV(instance)))
1317

    
1318
  return hvp_data
1319

    
1320

    
1321
class _VerifyErrors(object):
1322
  """Mix-in for cluster/group verify LUs.
1323

1324
  It provides _Error and _ErrorIf, and updates the self.bad boolean. (Expects
1325
  self.op and self._feedback_fn to be available.)
1326

1327
  """
1328
  TCLUSTER = "cluster"
1329
  TNODE = "node"
1330
  TINSTANCE = "instance"
1331

    
1332
  ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
1333
  ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")
1334
  ECLUSTERFILECHECK = (TCLUSTER, "ECLUSTERFILECHECK")
1335
  EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
1336
  EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
1337
  EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
1338
  EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
1339
  EINSTANCEFAULTYDISK = (TINSTANCE, "EINSTANCEFAULTYDISK")
1340
  EINSTANCEWRONGNODE = (TINSTANCE, "EINSTANCEWRONGNODE")
1341
  EINSTANCESPLITGROUPS = (TINSTANCE, "EINSTANCESPLITGROUPS")
1342
  ENODEDRBD = (TNODE, "ENODEDRBD")
1343
  ENODEDRBDHELPER = (TNODE, "ENODEDRBDHELPER")
1344
  ENODEFILECHECK = (TNODE, "ENODEFILECHECK")
1345
  ENODEHOOKS = (TNODE, "ENODEHOOKS")
1346
  ENODEHV = (TNODE, "ENODEHV")
1347
  ENODELVM = (TNODE, "ENODELVM")
1348
  ENODEN1 = (TNODE, "ENODEN1")
1349
  ENODENET = (TNODE, "ENODENET")
1350
  ENODEOS = (TNODE, "ENODEOS")
1351
  ENODEORPHANINSTANCE = (TNODE, "ENODEORPHANINSTANCE")
1352
  ENODEORPHANLV = (TNODE, "ENODEORPHANLV")
1353
  ENODERPC = (TNODE, "ENODERPC")
1354
  ENODESSH = (TNODE, "ENODESSH")
1355
  ENODEVERSION = (TNODE, "ENODEVERSION")
1356
  ENODESETUP = (TNODE, "ENODESETUP")
1357
  ENODETIME = (TNODE, "ENODETIME")
1358
  ENODEOOBPATH = (TNODE, "ENODEOOBPATH")
1359

    
1360
  ETYPE_FIELD = "code"
1361
  ETYPE_ERROR = "ERROR"
1362
  ETYPE_WARNING = "WARNING"
1363

    
1364
  def _Error(self, ecode, item, msg, *args, **kwargs):
1365
    """Format an error message.
1366

1367
    Based on the opcode's error_codes parameter, either format a
1368
    parseable error code, or a simpler error string.
1369

1370
    This must be called only from Exec and functions called from Exec.
1371

1372
    """
1373
    ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
1374
    itype, etxt = ecode
1375
    # first complete the msg
1376
    if args:
1377
      msg = msg % args
1378
    # then format the whole message
1379
    if self.op.error_codes: # This is a mix-in. pylint: disable-msg=E1101
1380
      msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
1381
    else:
1382
      if item:
1383
        item = " " + item
1384
      else:
1385
        item = ""
1386
      msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
1387
    # and finally report it via the feedback_fn
1388
    self._feedback_fn("  - %s" % msg) # Mix-in. pylint: disable-msg=E1101
1389

    
1390
  def _ErrorIf(self, cond, *args, **kwargs):
1391
    """Log an error message if the passed condition is True.
1392

1393
    """
1394
    cond = (bool(cond)
1395
            or self.op.debug_simulate_errors) # pylint: disable-msg=E1101
1396
    if cond:
1397
      self._Error(*args, **kwargs)
1398
    # do not mark the operation as failed for WARN cases only
1399
    if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
1400
      self.bad = self.bad or cond
1401

    
1402

    
1403
class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
1404
  """Verifies the cluster config.
1405

1406
  """
1407

    
1408
  REQ_BGL = False
1409

    
1410
  def _VerifyHVP(self, hvp_data):
1411
    """Verifies locally the syntax of the hypervisor parameters.
1412

1413
    """
1414
    for item, hv_name, hv_params in hvp_data:
1415
      msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
1416
             (item, hv_name))
1417
      try:
1418
        hv_class = hypervisor.GetHypervisor(hv_name)
1419
        utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1420
        hv_class.CheckParameterSyntax(hv_params)
1421
      except errors.GenericError, err:
1422
        self._ErrorIf(True, self.ECLUSTERCFG, None, msg % str(err))
1423

    
1424
  def ExpandNames(self):
1425
    self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
1426
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1427
    self.needed_locks = {}
1428

    
1429
  def Exec(self, feedback_fn):
1430
    """Verify integrity of cluster, performing various test on nodes.
1431

1432
    """
1433
    self.bad = False
1434
    self._feedback_fn = feedback_fn
1435

    
1436
    feedback_fn("* Verifying cluster config")
1437

    
1438
    for msg in self.cfg.VerifyConfig():
1439
      self._ErrorIf(True, self.ECLUSTERCFG, None, msg)
1440

    
1441
    feedback_fn("* Verifying cluster certificate files")
1442

    
1443
    for cert_filename in constants.ALL_CERT_FILES:
1444
      (errcode, msg) = _VerifyCertificate(cert_filename)
1445
      self._ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
1446

    
1447
    feedback_fn("* Verifying hypervisor parameters")
1448

    
1449
    self._VerifyHVP(_GetAllHypervisorParameters(self.cfg.GetClusterInfo(),
1450
                                                self.all_inst_info.values()))
1451

    
1452
    return (not self.bad, [g.name for g in self.all_group_info.values()])
1453

    
1454

    
1455
class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
1456
  """Verifies the status of a node group.
1457

1458
  """
1459

    
1460
  HPATH = "cluster-verify"
1461
  HTYPE = constants.HTYPE_CLUSTER
1462
  REQ_BGL = False
1463

    
1464
  _HOOKS_INDENT_RE = re.compile("^", re.M)
1465

    
1466
  class NodeImage(object):
1467
    """A class representing the logical and physical status of a node.
1468

1469
    @type name: string
1470
    @ivar name: the node name to which this object refers
1471
    @ivar volumes: a structure as returned from
1472
        L{ganeti.backend.GetVolumeList} (runtime)
1473
    @ivar instances: a list of running instances (runtime)
1474
    @ivar pinst: list of configured primary instances (config)
1475
    @ivar sinst: list of configured secondary instances (config)
1476
    @ivar sbp: dictionary of {primary-node: list of instances} for all
1477
        instances for which this node is secondary (config)
1478
    @ivar mfree: free memory, as reported by hypervisor (runtime)
1479
    @ivar dfree: free disk, as reported by the node (runtime)
1480
    @ivar offline: the offline status (config)
1481
    @type rpc_fail: boolean
1482
    @ivar rpc_fail: whether the RPC verify call was successfull (overall,
1483
        not whether the individual keys were correct) (runtime)
1484
    @type lvm_fail: boolean
1485
    @ivar lvm_fail: whether the RPC call didn't return valid LVM data
1486
    @type hyp_fail: boolean
1487
    @ivar hyp_fail: whether the RPC call didn't return the instance list
1488
    @type ghost: boolean
1489
    @ivar ghost: whether this is a known node or not (config)
1490
    @type os_fail: boolean
1491
    @ivar os_fail: whether the RPC call didn't return valid OS data
1492
    @type oslist: list
1493
    @ivar oslist: list of OSes as diagnosed by DiagnoseOS
1494
    @type vm_capable: boolean
1495
    @ivar vm_capable: whether the node can host instances
1496

1497
    """
1498
    def __init__(self, offline=False, name=None, vm_capable=True):
1499
      self.name = name
1500
      self.volumes = {}
1501
      self.instances = []
1502
      self.pinst = []
1503
      self.sinst = []
1504
      self.sbp = {}
1505
      self.mfree = 0
1506
      self.dfree = 0
1507
      self.offline = offline
1508
      self.vm_capable = vm_capable
1509
      self.rpc_fail = False
1510
      self.lvm_fail = False
1511
      self.hyp_fail = False
1512
      self.ghost = False
1513
      self.os_fail = False
1514
      self.oslist = {}
1515

    
1516
  def ExpandNames(self):
1517
    # This raises errors.OpPrereqError on its own:
1518
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
1519

    
1520
    all_node_info = self.cfg.GetAllNodesInfo()
1521
    all_inst_info = self.cfg.GetAllInstancesInfo()
1522

    
1523
    node_names = set(node.name
1524
                     for node in all_node_info.values()
1525
                     if node.group == self.group_uuid)
1526

    
1527
    inst_names = [inst.name
1528
                  for inst in all_inst_info.values()
1529
                  if inst.primary_node in node_names]
1530

    
1531
    # In Exec(), we warn about mirrored instances that have primary and
1532
    # secondary living in separate node groups. To fully verify that
1533
    # volumes for these instances are healthy, we will need to do an
1534
    # extra call to their secondaries. We ensure here those nodes will
1535
    # be locked.
1536
    for inst in inst_names:
1537
      if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
1538
        node_names.update(all_inst_info[inst].secondary_nodes)
1539

    
1540
    self.needed_locks = {
1541
      locking.LEVEL_NODEGROUP: [self.group_uuid],
1542
      locking.LEVEL_NODE: list(node_names),
1543
      locking.LEVEL_INSTANCE: inst_names,
1544
    }
1545

    
1546
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
1547

    
1548
  def CheckPrereq(self):
1549
    self.all_node_info = self.cfg.GetAllNodesInfo()
1550
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1551

    
1552
    group_nodes = set(node.name
1553
                      for node in self.all_node_info.values()
1554
                      if node.group == self.group_uuid)
1555

    
1556
    group_instances = set(inst.name
1557
                          for inst in self.all_inst_info.values()
1558
                          if inst.primary_node in group_nodes)
1559

    
1560
    unlocked_nodes = \
1561
        group_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
1562

    
1563
    unlocked_instances = \
1564
        group_instances.difference(self.glm.list_owned(locking.LEVEL_INSTANCE))
1565

    
1566
    if unlocked_nodes:
1567
      raise errors.OpPrereqError("missing lock for nodes: %s" %
1568
                                 utils.CommaJoin(unlocked_nodes))
1569

    
1570
    if unlocked_instances:
1571
      raise errors.OpPrereqError("missing lock for instances: %s" %
1572
                                 utils.CommaJoin(unlocked_instances))
1573

    
1574
    self.my_node_names = utils.NiceSort(group_nodes)
1575
    self.my_inst_names = utils.NiceSort(group_instances)
1576

    
1577
    self.my_node_info = dict((name, self.all_node_info[name])
1578
                             for name in self.my_node_names)
1579

    
1580
    self.my_inst_info = dict((name, self.all_inst_info[name])
1581
                             for name in self.my_inst_names)
1582

    
1583
    # We detect here the nodes that will need the extra RPC calls for verifying
1584
    # split LV volumes; they should be locked.
1585
    extra_lv_nodes = set()
1586

    
1587
    for inst in self.my_inst_info.values():
1588
      if inst.disk_template in constants.DTS_INT_MIRROR:
1589
        group = self.my_node_info[inst.primary_node].group
1590
        for nname in inst.secondary_nodes:
1591
          if self.all_node_info[nname].group != group:
1592
            extra_lv_nodes.add(nname)
1593

    
1594
    unlocked_lv_nodes = \
1595
        extra_lv_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
1596

    
1597
    if unlocked_lv_nodes:
1598
      raise errors.OpPrereqError("these nodes could be locked: %s" %
1599
                                 utils.CommaJoin(unlocked_lv_nodes))
1600
    self.extra_lv_nodes = list(extra_lv_nodes)
1601

    
1602
  def _VerifyNode(self, ninfo, nresult):
1603
    """Perform some basic validation on data returned from a node.
1604

1605
      - check the result data structure is well formed and has all the
1606
        mandatory fields
1607
      - check ganeti version
1608

1609
    @type ninfo: L{objects.Node}
1610
    @param ninfo: the node to check
1611
    @param nresult: the results from the node
1612
    @rtype: boolean
1613
    @return: whether overall this call was successful (and we can expect
1614
         reasonable values in the respose)
1615

1616
    """
1617
    node = ninfo.name
1618
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1619

    
1620
    # main result, nresult should be a non-empty dict
1621
    test = not nresult or not isinstance(nresult, dict)
1622
    _ErrorIf(test, self.ENODERPC, node,
1623
                  "unable to verify node: no data returned")
1624
    if test:
1625
      return False
1626

    
1627
    # compares ganeti version
1628
    local_version = constants.PROTOCOL_VERSION
1629
    remote_version = nresult.get("version", None)
1630
    test = not (remote_version and
1631
                isinstance(remote_version, (list, tuple)) and
1632
                len(remote_version) == 2)
1633
    _ErrorIf(test, self.ENODERPC, node,
1634
             "connection to node returned invalid data")
1635
    if test:
1636
      return False
1637

    
1638
    test = local_version != remote_version[0]
1639
    _ErrorIf(test, self.ENODEVERSION, node,
1640
             "incompatible protocol versions: master %s,"
1641
             " node %s", local_version, remote_version[0])
1642
    if test:
1643
      return False
1644

    
1645
    # node seems compatible, we can actually try to look into its results
1646

    
1647
    # full package version
1648
    self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
1649
                  self.ENODEVERSION, node,
1650
                  "software version mismatch: master %s, node %s",
1651
                  constants.RELEASE_VERSION, remote_version[1],
1652
                  code=self.ETYPE_WARNING)
1653

    
1654
    hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
1655
    if ninfo.vm_capable and isinstance(hyp_result, dict):
1656
      for hv_name, hv_result in hyp_result.iteritems():
1657
        test = hv_result is not None
1658
        _ErrorIf(test, self.ENODEHV, node,
1659
                 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
1660

    
1661
    hvp_result = nresult.get(constants.NV_HVPARAMS, None)
1662
    if ninfo.vm_capable and isinstance(hvp_result, list):
1663
      for item, hv_name, hv_result in hvp_result:
1664
        _ErrorIf(True, self.ENODEHV, node,
1665
                 "hypervisor %s parameter verify failure (source %s): %s",
1666
                 hv_name, item, hv_result)
1667

    
1668
    test = nresult.get(constants.NV_NODESETUP,
1669
                       ["Missing NODESETUP results"])
1670
    _ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
1671
             "; ".join(test))
1672

    
1673
    return True
1674

    
1675
  def _VerifyNodeTime(self, ninfo, nresult,
1676
                      nvinfo_starttime, nvinfo_endtime):
1677
    """Check the node time.
1678

1679
    @type ninfo: L{objects.Node}
1680
    @param ninfo: the node to check
1681
    @param nresult: the remote results for the node
1682
    @param nvinfo_starttime: the start time of the RPC call
1683
    @param nvinfo_endtime: the end time of the RPC call
1684

1685
    """
1686
    node = ninfo.name
1687
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1688

    
1689
    ntime = nresult.get(constants.NV_TIME, None)
1690
    try:
1691
      ntime_merged = utils.MergeTime(ntime)
1692
    except (ValueError, TypeError):
1693
      _ErrorIf(True, self.ENODETIME, node, "Node returned invalid time")
1694
      return
1695

    
1696
    if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
1697
      ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
1698
    elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
1699
      ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
1700
    else:
1701
      ntime_diff = None
1702

    
1703
    _ErrorIf(ntime_diff is not None, self.ENODETIME, node,
1704
             "Node time diverges by at least %s from master node time",
1705
             ntime_diff)
1706

    
1707
  def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
1708
    """Check the node LVM results.
1709

1710
    @type ninfo: L{objects.Node}
1711
    @param ninfo: the node to check
1712
    @param nresult: the remote results for the node
1713
    @param vg_name: the configured VG name
1714

1715
    """
1716
    if vg_name is None:
1717
      return
1718

    
1719
    node = ninfo.name
1720
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1721

    
1722
    # checks vg existence and size > 20G
1723
    vglist = nresult.get(constants.NV_VGLIST, None)
1724
    test = not vglist
1725
    _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
1726
    if not test:
1727
      vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
1728
                                            constants.MIN_VG_SIZE)
1729
      _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
1730

    
1731
    # check pv names
1732
    pvlist = nresult.get(constants.NV_PVLIST, None)
1733
    test = pvlist is None
1734
    _ErrorIf(test, self.ENODELVM, node, "Can't get PV list from node")
1735
    if not test:
1736
      # check that ':' is not present in PV names, since it's a
1737
      # special character for lvcreate (denotes the range of PEs to
1738
      # use on the PV)
1739
      for _, pvname, owner_vg in pvlist:
1740
        test = ":" in pvname
1741
        _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
1742
                 " '%s' of VG '%s'", pvname, owner_vg)
1743

    
1744
  def _VerifyNodeBridges(self, ninfo, nresult, bridges):
1745
    """Check the node bridges.
1746

1747
    @type ninfo: L{objects.Node}
1748
    @param ninfo: the node to check
1749
    @param nresult: the remote results for the node
1750
    @param bridges: the expected list of bridges
1751

1752
    """
1753
    if not bridges:
1754
      return
1755

    
1756
    node = ninfo.name
1757
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1758

    
1759
    missing = nresult.get(constants.NV_BRIDGES, None)
1760
    test = not isinstance(missing, list)
1761
    _ErrorIf(test, self.ENODENET, node,
1762
             "did not return valid bridge information")
1763
    if not test:
1764
      _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
1765
               utils.CommaJoin(sorted(missing)))
1766

    
1767
  def _VerifyNodeNetwork(self, ninfo, nresult):
1768
    """Check the node network connectivity results.
1769

1770
    @type ninfo: L{objects.Node}
1771
    @param ninfo: the node to check
1772
    @param nresult: the remote results for the node
1773

1774
    """
1775
    node = ninfo.name
1776
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1777

    
1778
    test = constants.NV_NODELIST not in nresult
1779
    _ErrorIf(test, self.ENODESSH, node,
1780
             "node hasn't returned node ssh connectivity data")
1781
    if not test:
1782
      if nresult[constants.NV_NODELIST]:
1783
        for a_node, a_msg in nresult[constants.NV_NODELIST].items():
1784
          _ErrorIf(True, self.ENODESSH, node,
1785
                   "ssh communication with node '%s': %s", a_node, a_msg)
1786

    
1787
    test = constants.NV_NODENETTEST not in nresult
1788
    _ErrorIf(test, self.ENODENET, node,
1789
             "node hasn't returned node tcp connectivity data")
1790
    if not test:
1791
      if nresult[constants.NV_NODENETTEST]:
1792
        nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
1793
        for anode in nlist:
1794
          _ErrorIf(True, self.ENODENET, node,
1795
                   "tcp communication with node '%s': %s",
1796
                   anode, nresult[constants.NV_NODENETTEST][anode])
1797

    
1798
    test = constants.NV_MASTERIP not in nresult
1799
    _ErrorIf(test, self.ENODENET, node,
1800
             "node hasn't returned node master IP reachability data")
1801
    if not test:
1802
      if not nresult[constants.NV_MASTERIP]:
1803
        if node == self.master_node:
1804
          msg = "the master node cannot reach the master IP (not configured?)"
1805
        else:
1806
          msg = "cannot reach the master IP"
1807
        _ErrorIf(True, self.ENODENET, node, msg)
1808

    
1809
  def _VerifyInstance(self, instance, instanceconfig, node_image,
1810
                      diskstatus):
1811
    """Verify an instance.
1812

1813
    This function checks to see if the required block devices are
1814
    available on the instance's node.
1815

1816
    """
1817
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1818
    node_current = instanceconfig.primary_node
1819

    
1820
    node_vol_should = {}
1821
    instanceconfig.MapLVsByNode(node_vol_should)
1822

    
1823
    for node in node_vol_should:
1824
      n_img = node_image[node]
1825
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
1826
        # ignore missing volumes on offline or broken nodes
1827
        continue
1828
      for volume in node_vol_should[node]:
1829
        test = volume not in n_img.volumes
1830
        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
1831
                 "volume %s missing on node %s", volume, node)
1832

    
1833
    if instanceconfig.admin_up:
1834
      pri_img = node_image[node_current]
1835
      test = instance not in pri_img.instances and not pri_img.offline
1836
      _ErrorIf(test, self.EINSTANCEDOWN, instance,
1837
               "instance not running on its primary node %s",
1838
               node_current)
1839

    
1840
    diskdata = [(nname, success, status, idx)
1841
                for (nname, disks) in diskstatus.items()
1842
                for idx, (success, status) in enumerate(disks)]
1843

    
1844
    for nname, success, bdev_status, idx in diskdata:
1845
      # the 'ghost node' construction in Exec() ensures that we have a
1846
      # node here
1847
      snode = node_image[nname]
1848
      bad_snode = snode.ghost or snode.offline
1849
      _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
1850
               self.EINSTANCEFAULTYDISK, instance,
1851
               "couldn't retrieve status for disk/%s on %s: %s",
1852
               idx, nname, bdev_status)
1853
      _ErrorIf((instanceconfig.admin_up and success and
1854
                bdev_status.ldisk_status == constants.LDS_FAULTY),
1855
               self.EINSTANCEFAULTYDISK, instance,
1856
               "disk/%s on %s is faulty", idx, nname)
1857

    
1858
  def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
1859
    """Verify if there are any unknown volumes in the cluster.
1860

1861
    The .os, .swap and backup volumes are ignored. All other volumes are
1862
    reported as unknown.
1863

1864
    @type reserved: L{ganeti.utils.FieldSet}
1865
    @param reserved: a FieldSet of reserved volume names
1866

1867
    """
1868
    for node, n_img in node_image.items():
1869
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
1870
        # skip non-healthy nodes
1871
        continue
1872
      for volume in n_img.volumes:
1873
        test = ((node not in node_vol_should or
1874
                volume not in node_vol_should[node]) and
1875
                not reserved.Matches(volume))
1876
        self._ErrorIf(test, self.ENODEORPHANLV, node,
1877
                      "volume %s is unknown", volume)
1878

    
1879
  def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
1880
    """Verify N+1 Memory Resilience.
1881

1882
    Check that if one single node dies we can still start all the
1883
    instances it was primary for.
1884

1885
    """
1886
    cluster_info = self.cfg.GetClusterInfo()
1887
    for node, n_img in node_image.items():
1888
      # This code checks that every node which is now listed as
1889
      # secondary has enough memory to host all instances it is
1890
      # supposed to should a single other node in the cluster fail.
1891
      # FIXME: not ready for failover to an arbitrary node
1892
      # FIXME: does not support file-backed instances
1893
      # WARNING: we currently take into account down instances as well
1894
      # as up ones, considering that even if they're down someone
1895
      # might want to start them even in the event of a node failure.
1896
      if n_img.offline:
1897
        # we're skipping offline nodes from the N+1 warning, since
1898
        # most likely we don't have good memory infromation from them;
1899
        # we already list instances living on such nodes, and that's
1900
        # enough warning
1901
        continue
1902
      for prinode, instances in n_img.sbp.items():
1903
        needed_mem = 0
1904
        for instance in instances:
1905
          bep = cluster_info.FillBE(instance_cfg[instance])
1906
          if bep[constants.BE_AUTO_BALANCE]:
1907
            needed_mem += bep[constants.BE_MEMORY]
1908
        test = n_img.mfree < needed_mem
1909
        self._ErrorIf(test, self.ENODEN1, node,
1910
                      "not enough memory to accomodate instance failovers"
1911
                      " should node %s fail (%dMiB needed, %dMiB available)",
1912
                      prinode, needed_mem, n_img.mfree)
1913

    
1914
  @classmethod
1915
  def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
1916
                   (files_all, files_all_opt, files_mc, files_vm)):
1917
    """Verifies file checksums collected from all nodes.
1918

1919
    @param errorif: Callback for reporting errors
1920
    @param nodeinfo: List of L{objects.Node} objects
1921
    @param master_node: Name of master node
1922
    @param all_nvinfo: RPC results
1923

1924
    """
1925
    node_names = frozenset(node.name for node in nodeinfo)
1926

    
1927
    assert master_node in node_names
1928
    assert (len(files_all | files_all_opt | files_mc | files_vm) ==
1929
            sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
1930
           "Found file listed in more than one file list"
1931

    
1932
    # Define functions determining which nodes to consider for a file
1933
    file2nodefn = dict([(filename, fn)
1934
      for (files, fn) in [(files_all, None),
1935
                          (files_all_opt, None),
1936
                          (files_mc, lambda node: (node.master_candidate or
1937
                                                   node.name == master_node)),
1938
                          (files_vm, lambda node: node.vm_capable)]
1939
      for filename in files])
1940

    
1941
    fileinfo = dict((filename, {}) for filename in file2nodefn.keys())
1942

    
1943
    for node in nodeinfo:
1944
      nresult = all_nvinfo[node.name]
1945

    
1946
      if nresult.fail_msg or not nresult.payload:
1947
        node_files = None
1948
      else:
1949
        node_files = nresult.payload.get(constants.NV_FILELIST, None)
1950

    
1951
      test = not (node_files and isinstance(node_files, dict))
1952
      errorif(test, cls.ENODEFILECHECK, node.name,
1953
              "Node did not return file checksum data")
1954
      if test:
1955
        continue
1956

    
1957
      for (filename, checksum) in node_files.items():
1958
        # Check if the file should be considered for a node
1959
        fn = file2nodefn[filename]
1960
        if fn is None or fn(node):
1961
          fileinfo[filename].setdefault(checksum, set()).add(node.name)
1962

    
1963
    for (filename, checksums) in fileinfo.items():
1964
      assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
1965

    
1966
      # Nodes having the file
1967
      with_file = frozenset(node_name
1968
                            for nodes in fileinfo[filename].values()
1969
                            for node_name in nodes)
1970

    
1971
      # Nodes missing file
1972
      missing_file = node_names - with_file
1973

    
1974
      if filename in files_all_opt:
1975
        # All or no nodes
1976
        errorif(missing_file and missing_file != node_names,
1977
                cls.ECLUSTERFILECHECK, None,
1978
                "File %s is optional, but it must exist on all or no nodes (not"
1979
                " found on %s)",
1980
                filename, utils.CommaJoin(utils.NiceSort(missing_file)))
1981
      else:
1982
        errorif(missing_file, cls.ECLUSTERFILECHECK, None,
1983
                "File %s is missing from node(s) %s", filename,
1984
                utils.CommaJoin(utils.NiceSort(missing_file)))
1985

    
1986
      # See if there are multiple versions of the file
1987
      test = len(checksums) > 1
1988
      if test:
1989
        variants = ["variant %s on %s" %
1990
                    (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
1991
                    for (idx, (checksum, nodes)) in
1992
                      enumerate(sorted(checksums.items()))]
1993
      else:
1994
        variants = []
1995

    
1996
      errorif(test, cls.ECLUSTERFILECHECK, None,
1997
              "File %s found with %s different checksums (%s)",
1998
              filename, len(checksums), "; ".join(variants))
1999

    
2000
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2001
                      drbd_map):
2002
    """Verifies and the node DRBD status.
2003

2004
    @type ninfo: L{objects.Node}
2005
    @param ninfo: the node to check
2006
    @param nresult: the remote results for the node
2007
    @param instanceinfo: the dict of instances
2008
    @param drbd_helper: the configured DRBD usermode helper
2009
    @param drbd_map: the DRBD map as returned by
2010
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2011

2012
    """
2013
    node = ninfo.name
2014
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2015

    
2016
    if drbd_helper:
2017
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2018
      test = (helper_result == None)
2019
      _ErrorIf(test, self.ENODEDRBDHELPER, node,
2020
               "no drbd usermode helper returned")
2021
      if helper_result:
2022
        status, payload = helper_result
2023
        test = not status
2024
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2025
                 "drbd usermode helper check unsuccessful: %s", payload)
2026
        test = status and (payload != drbd_helper)
2027
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2028
                 "wrong drbd usermode helper: %s", payload)
2029

    
2030
    # compute the DRBD minors
2031
    node_drbd = {}
2032
    for minor, instance in drbd_map[node].items():
2033
      test = instance not in instanceinfo
2034
      _ErrorIf(test, self.ECLUSTERCFG, None,
2035
               "ghost instance '%s' in temporary DRBD map", instance)
2036
        # ghost instance should not be running, but otherwise we
2037
        # don't give double warnings (both ghost instance and
2038
        # unallocated minor in use)
2039
      if test:
2040
        node_drbd[minor] = (instance, False)
2041
      else:
2042
        instance = instanceinfo[instance]
2043
        node_drbd[minor] = (instance.name, instance.admin_up)
2044

    
2045
    # and now check them
2046
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
2047
    test = not isinstance(used_minors, (tuple, list))
2048
    _ErrorIf(test, self.ENODEDRBD, node,
2049
             "cannot parse drbd status file: %s", str(used_minors))
2050
    if test:
2051
      # we cannot check drbd status
2052
      return
2053

    
2054
    for minor, (iname, must_exist) in node_drbd.items():
2055
      test = minor not in used_minors and must_exist
2056
      _ErrorIf(test, self.ENODEDRBD, node,
2057
               "drbd minor %d of instance %s is not active", minor, iname)
2058
    for minor in used_minors:
2059
      test = minor not in node_drbd
2060
      _ErrorIf(test, self.ENODEDRBD, node,
2061
               "unallocated drbd minor %d is in use", minor)
2062

    
2063
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
2064
    """Builds the node OS structures.
2065

2066
    @type ninfo: L{objects.Node}
2067
    @param ninfo: the node to check
2068
    @param nresult: the remote results for the node
2069
    @param nimg: the node image object
2070

2071
    """
2072
    node = ninfo.name
2073
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2074

    
2075
    remote_os = nresult.get(constants.NV_OSLIST, None)
2076
    test = (not isinstance(remote_os, list) or
2077
            not compat.all(isinstance(v, list) and len(v) == 7
2078
                           for v in remote_os))
2079

    
2080
    _ErrorIf(test, self.ENODEOS, node,
2081
             "node hasn't returned valid OS data")
2082

    
2083
    nimg.os_fail = test
2084

    
2085
    if test:
2086
      return
2087

    
2088
    os_dict = {}
2089

    
2090
    for (name, os_path, status, diagnose,
2091
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2092

    
2093
      if name not in os_dict:
2094
        os_dict[name] = []
2095

    
2096
      # parameters is a list of lists instead of list of tuples due to
2097
      # JSON lacking a real tuple type, fix it:
2098
      parameters = [tuple(v) for v in parameters]
2099
      os_dict[name].append((os_path, status, diagnose,
2100
                            set(variants), set(parameters), set(api_ver)))
2101

    
2102
    nimg.oslist = os_dict
2103

    
2104
  def _VerifyNodeOS(self, ninfo, nimg, base):
2105
    """Verifies the node OS list.
2106

2107
    @type ninfo: L{objects.Node}
2108
    @param ninfo: the node to check
2109
    @param nimg: the node image object
2110
    @param base: the 'template' node we match against (e.g. from the master)
2111

2112
    """
2113
    node = ninfo.name
2114
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2115

    
2116
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2117

    
2118
    beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2119
    for os_name, os_data in nimg.oslist.items():
2120
      assert os_data, "Empty OS status for OS %s?!" % os_name
2121
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2122
      _ErrorIf(not f_status, self.ENODEOS, node,
2123
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2124
      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
2125
               "OS '%s' has multiple entries (first one shadows the rest): %s",
2126
               os_name, utils.CommaJoin([v[0] for v in os_data]))
2127
      # this will catched in backend too
2128
      _ErrorIf(compat.any(v >= constants.OS_API_V15 for v in f_api)
2129
               and not f_var, self.ENODEOS, node,
2130
               "OS %s with API at least %d does not declare any variant",
2131
               os_name, constants.OS_API_V15)
2132
      # comparisons with the 'base' image
2133
      test = os_name not in base.oslist
2134
      _ErrorIf(test, self.ENODEOS, node,
2135
               "Extra OS %s not present on reference node (%s)",
2136
               os_name, base.name)
2137
      if test:
2138
        continue
2139
      assert base.oslist[os_name], "Base node has empty OS status?"
2140
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2141
      if not b_status:
2142
        # base OS is invalid, skipping
2143
        continue
2144
      for kind, a, b in [("API version", f_api, b_api),
2145
                         ("variants list", f_var, b_var),
2146
                         ("parameters", beautify_params(f_param),
2147
                          beautify_params(b_param))]:
2148
        _ErrorIf(a != b, self.ENODEOS, node,
2149
                 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2150
                 kind, os_name, base.name,
2151
                 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2152

    
2153
    # check any missing OSes
2154
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2155
    _ErrorIf(missing, self.ENODEOS, node,
2156
             "OSes present on reference node %s but missing on this node: %s",
2157
             base.name, utils.CommaJoin(missing))
2158

    
2159
  def _VerifyOob(self, ninfo, nresult):
2160
    """Verifies out of band functionality of a node.
2161

2162
    @type ninfo: L{objects.Node}
2163
    @param ninfo: the node to check
2164
    @param nresult: the remote results for the node
2165

2166
    """
2167
    node = ninfo.name
2168
    # We just have to verify the paths on master and/or master candidates
2169
    # as the oob helper is invoked on the master
2170
    if ((ninfo.master_candidate or ninfo.master_capable) and
2171
        constants.NV_OOB_PATHS in nresult):
2172
      for path_result in nresult[constants.NV_OOB_PATHS]:
2173
        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
2174

    
2175
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2176
    """Verifies and updates the node volume data.
2177

2178
    This function will update a L{NodeImage}'s internal structures
2179
    with data from the remote call.
2180

2181
    @type ninfo: L{objects.Node}
2182
    @param ninfo: the node to check
2183
    @param nresult: the remote results for the node
2184
    @param nimg: the node image object
2185
    @param vg_name: the configured VG name
2186

2187
    """
2188
    node = ninfo.name
2189
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2190

    
2191
    nimg.lvm_fail = True
2192
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2193
    if vg_name is None:
2194
      pass
2195
    elif isinstance(lvdata, basestring):
2196
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
2197
               utils.SafeEncode(lvdata))
2198
    elif not isinstance(lvdata, dict):
2199
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
2200
    else:
2201
      nimg.volumes = lvdata
2202
      nimg.lvm_fail = False
2203

    
2204
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2205
    """Verifies and updates the node instance list.
2206

2207
    If the listing was successful, then updates this node's instance
2208
    list. Otherwise, it marks the RPC call as failed for the instance
2209
    list key.
2210

2211
    @type ninfo: L{objects.Node}
2212
    @param ninfo: the node to check
2213
    @param nresult: the remote results for the node
2214
    @param nimg: the node image object
2215

2216
    """
2217
    idata = nresult.get(constants.NV_INSTANCELIST, None)
2218
    test = not isinstance(idata, list)
2219
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
2220
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
2221
    if test:
2222
      nimg.hyp_fail = True
2223
    else:
2224
      nimg.instances = idata
2225

    
2226
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2227
    """Verifies and computes a node information map
2228

2229
    @type ninfo: L{objects.Node}
2230
    @param ninfo: the node to check
2231
    @param nresult: the remote results for the node
2232
    @param nimg: the node image object
2233
    @param vg_name: the configured VG name
2234

2235
    """
2236
    node = ninfo.name
2237
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2238

    
2239
    # try to read free memory (from the hypervisor)
2240
    hv_info = nresult.get(constants.NV_HVINFO, None)
2241
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2242
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
2243
    if not test:
2244
      try:
2245
        nimg.mfree = int(hv_info["memory_free"])
2246
      except (ValueError, TypeError):
2247
        _ErrorIf(True, self.ENODERPC, node,
2248
                 "node returned invalid nodeinfo, check hypervisor")
2249

    
2250
    # FIXME: devise a free space model for file based instances as well
2251
    if vg_name is not None:
2252
      test = (constants.NV_VGLIST not in nresult or
2253
              vg_name not in nresult[constants.NV_VGLIST])
2254
      _ErrorIf(test, self.ENODELVM, node,
2255
               "node didn't return data for the volume group '%s'"
2256
               " - it is either missing or broken", vg_name)
2257
      if not test:
2258
        try:
2259
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2260
        except (ValueError, TypeError):
2261
          _ErrorIf(True, self.ENODERPC, node,
2262
                   "node returned invalid LVM info, check LVM status")
2263

    
2264
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2265
    """Gets per-disk status information for all instances.
2266

2267
    @type nodelist: list of strings
2268
    @param nodelist: Node names
2269
    @type node_image: dict of (name, L{objects.Node})
2270
    @param node_image: Node objects
2271
    @type instanceinfo: dict of (name, L{objects.Instance})
2272
    @param instanceinfo: Instance objects
2273
    @rtype: {instance: {node: [(succes, payload)]}}
2274
    @return: a dictionary of per-instance dictionaries with nodes as
2275
        keys and disk information as values; the disk information is a
2276
        list of tuples (success, payload)
2277

2278
    """
2279
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2280

    
2281
    node_disks = {}
2282
    node_disks_devonly = {}
2283
    diskless_instances = set()
2284
    diskless = constants.DT_DISKLESS
2285

    
2286
    for nname in nodelist:
2287
      node_instances = list(itertools.chain(node_image[nname].pinst,
2288
                                            node_image[nname].sinst))
2289
      diskless_instances.update(inst for inst in node_instances
2290
                                if instanceinfo[inst].disk_template == diskless)
2291
      disks = [(inst, disk)
2292
               for inst in node_instances
2293
               for disk in instanceinfo[inst].disks]
2294

    
2295
      if not disks:
2296
        # No need to collect data
2297
        continue
2298

    
2299
      node_disks[nname] = disks
2300

    
2301
      # Creating copies as SetDiskID below will modify the objects and that can
2302
      # lead to incorrect data returned from nodes
2303
      devonly = [dev.Copy() for (_, dev) in disks]
2304

    
2305
      for dev in devonly:
2306
        self.cfg.SetDiskID(dev, nname)
2307

    
2308
      node_disks_devonly[nname] = devonly
2309

    
2310
    assert len(node_disks) == len(node_disks_devonly)
2311

    
2312
    # Collect data from all nodes with disks
2313
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2314
                                                          node_disks_devonly)
2315

    
2316
    assert len(result) == len(node_disks)
2317

    
2318
    instdisk = {}
2319

    
2320
    for (nname, nres) in result.items():
2321
      disks = node_disks[nname]
2322

    
2323
      if nres.offline:
2324
        # No data from this node
2325
        data = len(disks) * [(False, "node offline")]
2326
      else:
2327
        msg = nres.fail_msg
2328
        _ErrorIf(msg, self.ENODERPC, nname,
2329
                 "while getting disk information: %s", msg)
2330
        if msg:
2331
          # No data from this node
2332
          data = len(disks) * [(False, msg)]
2333
        else:
2334
          data = []
2335
          for idx, i in enumerate(nres.payload):
2336
            if isinstance(i, (tuple, list)) and len(i) == 2:
2337
              data.append(i)
2338
            else:
2339
              logging.warning("Invalid result from node %s, entry %d: %s",
2340
                              nname, idx, i)
2341
              data.append((False, "Invalid result from the remote node"))
2342

    
2343
      for ((inst, _), status) in zip(disks, data):
2344
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2345

    
2346
    # Add empty entries for diskless instances.
2347
    for inst in diskless_instances:
2348
      assert inst not in instdisk
2349
      instdisk[inst] = {}
2350

    
2351
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2352
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2353
                      compat.all(isinstance(s, (tuple, list)) and
2354
                                 len(s) == 2 for s in statuses)
2355
                      for inst, nnames in instdisk.items()
2356
                      for nname, statuses in nnames.items())
2357
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2358

    
2359
    return instdisk
2360

    
2361
  def BuildHooksEnv(self):
2362
    """Build hooks env.
2363

2364
    Cluster-Verify hooks just ran in the post phase and their failure makes
2365
    the output be logged in the verify output and the verification to fail.
2366

2367
    """
2368
    env = {
2369
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2370
      }
2371

    
2372
    env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
2373
               for node in self.my_node_info.values())
2374

    
2375
    return env
2376

    
2377
  def BuildHooksNodes(self):
2378
    """Build hooks nodes.
2379

2380
    """
2381
    assert self.my_node_names, ("Node list not gathered,"
2382
      " has CheckPrereq been executed?")
2383
    return ([], self.my_node_names)
2384

    
2385
  def Exec(self, feedback_fn):
2386
    """Verify integrity of the node group, performing various test on nodes.
2387

2388
    """
2389
    # This method has too many local variables. pylint: disable-msg=R0914
2390
    self.bad = False
2391
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2392
    verbose = self.op.verbose
2393
    self._feedback_fn = feedback_fn
2394

    
2395
    vg_name = self.cfg.GetVGName()
2396
    drbd_helper = self.cfg.GetDRBDHelper()
2397
    cluster = self.cfg.GetClusterInfo()
2398
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2399
    hypervisors = cluster.enabled_hypervisors
2400
    node_data_list = [self.my_node_info[name] for name in self.my_node_names]
2401

    
2402
    i_non_redundant = [] # Non redundant instances
2403
    i_non_a_balanced = [] # Non auto-balanced instances
2404
    n_offline = 0 # Count of offline nodes
2405
    n_drained = 0 # Count of nodes being drained
2406
    node_vol_should = {}
2407

    
2408
    # FIXME: verify OS list
2409

    
2410
    # File verification
2411
    filemap = _ComputeAncillaryFiles(cluster, False)
2412

    
2413
    # do local checksums
2414
    master_node = self.master_node = self.cfg.GetMasterNode()
2415
    master_ip = self.cfg.GetMasterIP()
2416

    
2417
    feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
2418

    
2419
    # We will make nodes contact all nodes in their group, and one node from
2420
    # every other group.
2421
    # TODO: should it be a *random* node, different every time?
2422
    online_nodes = [node.name for node in node_data_list if not node.offline]
2423
    other_group_nodes = {}
2424

    
2425
    for name in sorted(self.all_node_info):
2426
      node = self.all_node_info[name]
2427
      if (node.group not in other_group_nodes
2428
          and node.group != self.group_uuid
2429
          and not node.offline):
2430
        other_group_nodes[node.group] = node.name
2431

    
2432
    node_verify_param = {
2433
      constants.NV_FILELIST:
2434
        utils.UniqueSequence(filename
2435
                             for files in filemap
2436
                             for filename in files),
2437
      constants.NV_NODELIST: online_nodes + other_group_nodes.values(),
2438
      constants.NV_HYPERVISOR: hypervisors,
2439
      constants.NV_HVPARAMS:
2440
        _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
2441
      constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
2442
                                 for node in node_data_list
2443
                                 if not node.offline],
2444
      constants.NV_INSTANCELIST: hypervisors,
2445
      constants.NV_VERSION: None,
2446
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2447
      constants.NV_NODESETUP: None,
2448
      constants.NV_TIME: None,
2449
      constants.NV_MASTERIP: (master_node, master_ip),
2450
      constants.NV_OSLIST: None,
2451
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2452
      }
2453

    
2454
    if vg_name is not None:
2455
      node_verify_param[constants.NV_VGLIST] = None
2456
      node_verify_param[constants.NV_LVLIST] = vg_name
2457
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2458
      node_verify_param[constants.NV_DRBDLIST] = None
2459

    
2460
    if drbd_helper:
2461
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2462

    
2463
    # bridge checks
2464
    # FIXME: this needs to be changed per node-group, not cluster-wide
2465
    bridges = set()
2466
    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
2467
    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2468
      bridges.add(default_nicpp[constants.NIC_LINK])
2469
    for instance in self.my_inst_info.values():
2470
      for nic in instance.nics:
2471
        full_nic = cluster.SimpleFillNIC(nic.nicparams)
2472
        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2473
          bridges.add(full_nic[constants.NIC_LINK])
2474

    
2475
    if bridges:
2476
      node_verify_param[constants.NV_BRIDGES] = list(bridges)
2477

    
2478
    # Build our expected cluster state
2479
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2480
                                                 name=node.name,
2481
                                                 vm_capable=node.vm_capable))
2482
                      for node in node_data_list)
2483

    
2484
    # Gather OOB paths
2485
    oob_paths = []
2486
    for node in self.all_node_info.values():
2487
      path = _SupportsOob(self.cfg, node)
2488
      if path and path not in oob_paths:
2489
        oob_paths.append(path)
2490

    
2491
    if oob_paths:
2492
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2493

    
2494
    for instance in self.my_inst_names:
2495
      inst_config = self.my_inst_info[instance]
2496

    
2497
      for nname in inst_config.all_nodes:
2498
        if nname not in node_image:
2499
          gnode = self.NodeImage(name=nname)
2500
          gnode.ghost = (nname not in self.all_node_info)
2501
          node_image[nname] = gnode
2502

    
2503
      inst_config.MapLVsByNode(node_vol_should)
2504

    
2505
      pnode = inst_config.primary_node
2506
      node_image[pnode].pinst.append(instance)
2507

    
2508
      for snode in inst_config.secondary_nodes:
2509
        nimg = node_image[snode]
2510
        nimg.sinst.append(instance)
2511
        if pnode not in nimg.sbp:
2512
          nimg.sbp[pnode] = []
2513
        nimg.sbp[pnode].append(instance)
2514

    
2515
    # At this point, we have the in-memory data structures complete,
2516
    # except for the runtime information, which we'll gather next
2517

    
2518
    # Due to the way our RPC system works, exact response times cannot be
2519
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2520
    # time before and after executing the request, we can at least have a time
2521
    # window.
2522
    nvinfo_starttime = time.time()
2523
    all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
2524
                                           node_verify_param,
2525
                                           self.cfg.GetClusterName())
2526
    if self.extra_lv_nodes and vg_name is not None:
2527
      extra_lv_nvinfo = \
2528
          self.rpc.call_node_verify(self.extra_lv_nodes,
2529
                                    {constants.NV_LVLIST: vg_name},
2530
                                    self.cfg.GetClusterName())
2531
    else:
2532
      extra_lv_nvinfo = {}
2533
    nvinfo_endtime = time.time()
2534

    
2535
    all_drbd_map = self.cfg.ComputeDRBDMap()
2536

    
2537
    feedback_fn("* Gathering disk information (%s nodes)" %
2538
                len(self.my_node_names))
2539
    instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
2540
                                     self.my_inst_info)
2541

    
2542
    feedback_fn("* Verifying configuration file consistency")
2543

    
2544
    # If not all nodes are being checked, we need to make sure the master node
2545
    # and a non-checked vm_capable node are in the list.
2546
    absent_nodes = set(self.all_node_info).difference(self.my_node_info)
2547
    if absent_nodes:
2548
      vf_nvinfo = all_nvinfo.copy()
2549
      vf_node_info = list(self.my_node_info.values())
2550
      additional_nodes = []
2551
      if master_node not in self.my_node_info:
2552
        additional_nodes.append(master_node)
2553
        vf_node_info.append(self.all_node_info[master_node])
2554
      # Add the first vm_capable node we find which is not included
2555
      for node in absent_nodes:
2556
        nodeinfo = self.all_node_info[node]
2557
        if nodeinfo.vm_capable and not nodeinfo.offline:
2558
          additional_nodes.append(node)
2559
          vf_node_info.append(self.all_node_info[node])
2560
          break
2561
      key = constants.NV_FILELIST
2562
      vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
2563
                                                 {key: node_verify_param[key]},
2564
                                                 self.cfg.GetClusterName()))
2565
    else:
2566
      vf_nvinfo = all_nvinfo
2567
      vf_node_info = self.my_node_info.values()
2568

    
2569
    self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
2570

    
2571
    feedback_fn("* Verifying node status")
2572

    
2573
    refos_img = None
2574

    
2575
    for node_i in node_data_list:
2576
      node = node_i.name
2577
      nimg = node_image[node]
2578

    
2579
      if node_i.offline:
2580
        if verbose:
2581
          feedback_fn("* Skipping offline node %s" % (node,))
2582
        n_offline += 1
2583
        continue
2584

    
2585
      if node == master_node:
2586
        ntype = "master"
2587
      elif node_i.master_candidate:
2588
        ntype = "master candidate"
2589
      elif node_i.drained:
2590
        ntype = "drained"
2591
        n_drained += 1
2592
      else:
2593
        ntype = "regular"
2594
      if verbose:
2595
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2596

    
2597
      msg = all_nvinfo[node].fail_msg
2598
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
2599
      if msg:
2600
        nimg.rpc_fail = True
2601
        continue
2602

    
2603
      nresult = all_nvinfo[node].payload
2604

    
2605
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2606
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2607
      self._VerifyNodeNetwork(node_i, nresult)
2608
      self._VerifyOob(node_i, nresult)
2609

    
2610
      if nimg.vm_capable:
2611
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2612
        self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
2613
                             all_drbd_map)
2614

    
2615
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2616
        self._UpdateNodeInstances(node_i, nresult, nimg)
2617
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2618
        self._UpdateNodeOS(node_i, nresult, nimg)
2619

    
2620
        if not nimg.os_fail:
2621
          if refos_img is None:
2622
            refos_img = nimg
2623
          self._VerifyNodeOS(node_i, nimg, refos_img)
2624
        self._VerifyNodeBridges(node_i, nresult, bridges)
2625

    
2626
        # Check whether all running instancies are primary for the node. (This
2627
        # can no longer be done from _VerifyInstance below, since some of the
2628
        # wrong instances could be from other node groups.)
2629
        non_primary_inst = set(nimg.instances).difference(nimg.pinst)
2630

    
2631
        for inst in non_primary_inst:
2632
          test = inst in self.all_inst_info
2633
          _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
2634
                   "instance should not run on node %s", node_i.name)
2635
          _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
2636
                   "node is running unknown instance %s", inst)
2637

    
2638
    for node, result in extra_lv_nvinfo.items():
2639
      self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
2640
                              node_image[node], vg_name)
2641

    
2642
    feedback_fn("* Verifying instance status")
2643
    for instance in self.my_inst_names:
2644
      if verbose:
2645
        feedback_fn("* Verifying instance %s" % instance)
2646
      inst_config = self.my_inst_info[instance]
2647
      self._VerifyInstance(instance, inst_config, node_image,
2648
                           instdisk[instance])
2649
      inst_nodes_offline = []
2650

    
2651
      pnode = inst_config.primary_node
2652
      pnode_img = node_image[pnode]
2653
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
2654
               self.ENODERPC, pnode, "instance %s, connection to"
2655
               " primary node failed", instance)
2656

    
2657
      _ErrorIf(inst_config.admin_up and pnode_img.offline,
2658
               self.EINSTANCEBADNODE, instance,
2659
               "instance is marked as running and lives on offline node %s",
2660
               inst_config.primary_node)
2661

    
2662
      # If the instance is non-redundant we cannot survive losing its primary
2663
      # node, so we are not N+1 compliant. On the other hand we have no disk
2664
      # templates with more than one secondary so that situation is not well
2665
      # supported either.
2666
      # FIXME: does not support file-backed instances
2667
      if not inst_config.secondary_nodes:
2668
        i_non_redundant.append(instance)
2669

    
2670
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
2671
               instance, "instance has multiple secondary nodes: %s",
2672
               utils.CommaJoin(inst_config.secondary_nodes),
2673
               code=self.ETYPE_WARNING)
2674

    
2675
      if inst_config.disk_template in constants.DTS_INT_MIRROR:
2676
        pnode = inst_config.primary_node
2677
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
2678
        instance_groups = {}
2679

    
2680
        for node in instance_nodes:
2681
          instance_groups.setdefault(self.all_node_info[node].group,
2682
                                     []).append(node)
2683

    
2684
        pretty_list = [
2685
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
2686
          # Sort so that we always list the primary node first.
2687
          for group, nodes in sorted(instance_groups.items(),
2688
                                     key=lambda (_, nodes): pnode in nodes,
2689
                                     reverse=True)]
2690

    
2691
        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
2692
                      instance, "instance has primary and secondary nodes in"
2693
                      " different groups: %s", utils.CommaJoin(pretty_list),
2694
                      code=self.ETYPE_WARNING)
2695

    
2696
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
2697
        i_non_a_balanced.append(instance)
2698

    
2699
      for snode in inst_config.secondary_nodes:
2700
        s_img = node_image[snode]
2701
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
2702
                 "instance %s, connection to secondary node failed", instance)
2703

    
2704
        if s_img.offline:
2705
          inst_nodes_offline.append(snode)
2706

    
2707
      # warn that the instance lives on offline nodes
2708
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
2709
               "instance has offline secondary node(s) %s",
2710
               utils.CommaJoin(inst_nodes_offline))
2711
      # ... or ghost/non-vm_capable nodes
2712
      for node in inst_config.all_nodes:
2713
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
2714
                 "instance lives on ghost node %s", node)
2715
        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
2716
                 instance, "instance lives on non-vm_capable node %s", node)
2717

    
2718
    feedback_fn("* Verifying orphan volumes")
2719
    reserved = utils.FieldSet(*cluster.reserved_lvs)
2720

    
2721
    # We will get spurious "unknown volume" warnings if any node of this group
2722
    # is secondary for an instance whose primary is in another group. To avoid
2723
    # them, we find these instances and add their volumes to node_vol_should.
2724
    for inst in self.all_inst_info.values():
2725
      for secondary in inst.secondary_nodes:
2726
        if (secondary in self.my_node_info
2727
            and inst.name not in self.my_inst_info):
2728
          inst.MapLVsByNode(node_vol_should)
2729
          break
2730

    
2731
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
2732

    
2733
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
2734
      feedback_fn("* Verifying N+1 Memory redundancy")
2735
      self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
2736

    
2737
    feedback_fn("* Other Notes")
2738
    if i_non_redundant:
2739
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
2740
                  % len(i_non_redundant))
2741

    
2742
    if i_non_a_balanced:
2743
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
2744
                  % len(i_non_a_balanced))
2745

    
2746
    if n_offline:
2747
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
2748

    
2749
    if n_drained:
2750
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
2751

    
2752
    return not self.bad
2753

    
2754
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
2755
    """Analyze the post-hooks' result
2756

2757
    This method analyses the hook result, handles it, and sends some
2758
    nicely-formatted feedback back to the user.
2759

2760
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
2761
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
2762
    @param hooks_results: the results of the multi-node hooks rpc call
2763
    @param feedback_fn: function used send feedback back to the caller
2764
    @param lu_result: previous Exec result
2765
    @return: the new Exec result, based on the previous result
2766
        and hook results
2767

2768
    """
2769
    # We only really run POST phase hooks, and are only interested in
2770
    # their results
2771
    if phase == constants.HOOKS_PHASE_POST:
2772
      # Used to change hooks' output to proper indentation
2773
      feedback_fn("* Hooks Results")
2774
      assert hooks_results, "invalid result from hooks"
2775

    
2776
      for node_name in hooks_results:
2777
        res = hooks_results[node_name]
2778
        msg = res.fail_msg
2779
        test = msg and not res.offline
2780
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
2781
                      "Communication failure in hooks execution: %s", msg)
2782
        if res.offline or msg:
2783
          # No need to investigate payload if node is offline or gave an error.
2784
          # override manually lu_result here as _ErrorIf only
2785
          # overrides self.bad
2786
          lu_result = 1
2787
          continue
2788
        for script, hkr, output in res.payload:
2789
          test = hkr == constants.HKR_FAIL
2790
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
2791
                        "Script %s failed, output:", script)
2792
          if test:
2793
            output = self._HOOKS_INDENT_RE.sub('      ', output)
2794
            feedback_fn("%s" % output)
2795
            lu_result = 0
2796

    
2797
      return lu_result
2798

    
2799

    
2800
class LUClusterVerifyDisks(NoHooksLU):
2801
  """Verifies the cluster disks status.
2802

2803
  """
2804
  REQ_BGL = False
2805

    
2806
  def ExpandNames(self):
2807
    self.needed_locks = {
2808
      locking.LEVEL_NODE: locking.ALL_SET,
2809
      locking.LEVEL_INSTANCE: locking.ALL_SET,
2810
    }
2811
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
2812

    
2813
  def Exec(self, feedback_fn):
2814
    """Verify integrity of cluster disks.
2815

2816
    @rtype: tuple of three items
2817
    @return: a tuple of (dict of node-to-node_error, list of instances
2818
        which need activate-disks, dict of instance: (node, volume) for
2819
        missing volumes
2820

2821
    """
2822
    result = res_nodes, res_instances, res_missing = {}, [], {}
2823

    
2824
    nodes = utils.NiceSort(self.cfg.GetVmCapableNodeList())
2825
    instances = self.cfg.GetAllInstancesInfo().values()
2826

    
2827
    nv_dict = {}
2828
    for inst in instances:
2829
      inst_lvs = {}
2830
      if not inst.admin_up:
2831
        continue
2832
      inst.MapLVsByNode(inst_lvs)
2833
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
2834
      for node, vol_list in inst_lvs.iteritems():
2835
        for vol in vol_list:
2836
          nv_dict[(node, vol)] = inst
2837

    
2838
    if not nv_dict:
2839
      return result
2840

    
2841
    node_lvs = self.rpc.call_lv_list(nodes, [])
2842
    for node, node_res in node_lvs.items():
2843
      if node_res.offline:
2844
        continue
2845
      msg = node_res.fail_msg
2846
      if msg:
2847
        logging.warning("Error enumerating LVs on node %s: %s", node, msg)
2848
        res_nodes[node] = msg
2849
        continue
2850

    
2851
      lvs = node_res.payload
2852
      for lv_name, (_, _, lv_online) in lvs.items():
2853
        inst = nv_dict.pop((node, lv_name), None)
2854
        if (not lv_online and inst is not None
2855
            and inst.name not in res_instances):
2856
          res_instances.append(inst.name)
2857

    
2858
    # any leftover items in nv_dict are missing LVs, let's arrange the
2859
    # data better
2860
    for key, inst in nv_dict.iteritems():
2861
      if inst.name not in res_missing:
2862
        res_missing[inst.name] = []
2863
      res_missing[inst.name].append(key)
2864

    
2865
    return result
2866

    
2867

    
2868
class LUClusterRepairDiskSizes(NoHooksLU):
2869
  """Verifies the cluster disks sizes.
2870

2871
  """
2872
  REQ_BGL = False
2873

    
2874
  def ExpandNames(self):
2875
    if self.op.instances:
2876
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
2877
      self.needed_locks = {
2878
        locking.LEVEL_NODE: [],
2879
        locking.LEVEL_INSTANCE: self.wanted_names,
2880
        }
2881
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2882
    else:
2883
      self.wanted_names = None
2884
      self.needed_locks = {
2885
        locking.LEVEL_NODE: locking.ALL_SET,
2886
        locking.LEVEL_INSTANCE: locking.ALL_SET,
2887
        }
2888
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
2889

    
2890
  def DeclareLocks(self, level):
2891
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
2892
      self._LockInstancesNodes(primary_only=True)
2893

    
2894
  def CheckPrereq(self):
2895
    """Check prerequisites.
2896

2897
    This only checks the optional instance list against the existing names.
2898

2899
    """
2900
    if self.wanted_names is None:
2901
      self.wanted_names = self.glm.list_owned(locking.LEVEL_INSTANCE)
2902

    
2903
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
2904
                             in self.wanted_names]
2905

    
2906
  def _EnsureChildSizes(self, disk):
2907
    """Ensure children of the disk have the needed disk size.
2908

2909
    This is valid mainly for DRBD8 and fixes an issue where the
2910
    children have smaller disk size.
2911

2912
    @param disk: an L{ganeti.objects.Disk} object
2913

2914
    """
2915
    if disk.dev_type == constants.LD_DRBD8:
2916
      assert disk.children, "Empty children for DRBD8?"
2917
      fchild = disk.children[0]
2918
      mismatch = fchild.size < disk.size
2919
      if mismatch:
2920
        self.LogInfo("Child disk has size %d, parent %d, fixing",
2921
                     fchild.size, disk.size)
2922
        fchild.size = disk.size
2923

    
2924
      # and we recurse on this child only, not on the metadev
2925
      return self._EnsureChildSizes(fchild) or mismatch
2926
    else:
2927
      return False
2928

    
2929
  def Exec(self, feedback_fn):
2930
    """Verify the size of cluster disks.
2931

2932
    """
2933
    # TODO: check child disks too
2934
    # TODO: check differences in size between primary/secondary nodes
2935
    per_node_disks = {}
2936
    for instance in self.wanted_instances:
2937
      pnode = instance.primary_node
2938
      if pnode not in per_node_disks:
2939
        per_node_disks[pnode] = []
2940
      for idx, disk in enumerate(instance.disks):
2941
        per_node_disks[pnode].append((instance, idx, disk))
2942

    
2943
    changed = []
2944
    for node, dskl in per_node_disks.items():
2945
      newl = [v[2].Copy() for v in dskl]
2946
      for dsk in newl:
2947
        self.cfg.SetDiskID(dsk, node)
2948
      result = self.rpc.call_blockdev_getsize(node, newl)
2949
      if result.fail_msg:
2950
        self.LogWarning("Failure in blockdev_getsize call to node"
2951
                        " %s, ignoring", node)
2952
        continue
2953
      if len(result.payload) != len(dskl):
2954
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
2955
                        " result.payload=%s", node, len(dskl), result.payload)
2956
        self.LogWarning("Invalid result from node %s, ignoring node results",
2957
                        node)
2958
        continue
2959
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
2960
        if size is None:
2961
          self.LogWarning("Disk %d of instance %s did not return size"
2962
                          " information, ignoring", idx, instance.name)
2963
          continue
2964
        if not isinstance(size, (int, long)):
2965
          self.LogWarning("Disk %d of instance %s did not return valid"
2966
                          " size information, ignoring", idx, instance.name)
2967
          continue
2968
        size = size >> 20
2969
        if size != disk.size:
2970
          self.LogInfo("Disk %d of instance %s has mismatched size,"
2971
                       " correcting: recorded %d, actual %d", idx,
2972
                       instance.name, disk.size, size)
2973
          disk.size = size
2974
          self.cfg.Update(instance, feedback_fn)
2975
          changed.append((instance.name, idx, size))
2976
        if self._EnsureChildSizes(disk):
2977
          self.cfg.Update(instance, feedback_fn)
2978
          changed.append((instance.name, idx, disk.size))
2979
    return changed
2980

    
2981

    
2982
class LUClusterRename(LogicalUnit):
2983
  """Rename the cluster.
2984

2985
  """
2986
  HPATH = "cluster-rename"
2987
  HTYPE = constants.HTYPE_CLUSTER
2988

    
2989
  def BuildHooksEnv(self):
2990
    """Build hooks env.
2991

2992
    """
2993
    return {
2994
      "OP_TARGET": self.cfg.GetClusterName(),
2995
      "NEW_NAME": self.op.name,
2996
      }
2997

    
2998
  def BuildHooksNodes(self):
2999
    """Build hooks nodes.
3000

3001
    """
3002
    return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3003

    
3004
  def CheckPrereq(self):
3005
    """Verify that the passed name is a valid one.
3006

3007
    """
3008
    hostname = netutils.GetHostname(name=self.op.name,
3009
                                    family=self.cfg.GetPrimaryIPFamily())
3010

    
3011
    new_name = hostname.name
3012
    self.ip = new_ip = hostname.ip
3013
    old_name = self.cfg.GetClusterName()
3014
    old_ip = self.cfg.GetMasterIP()
3015
    if new_name == old_name and new_ip == old_ip:
3016
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
3017
                                 " cluster has changed",
3018
                                 errors.ECODE_INVAL)
3019
    if new_ip != old_ip:
3020
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3021
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
3022
                                   " reachable on the network" %
3023
                                   new_ip, errors.ECODE_NOTUNIQUE)
3024

    
3025
    self.op.name = new_name
3026

    
3027
  def Exec(self, feedback_fn):
3028
    """Rename the cluster.
3029

3030
    """
3031
    clustername = self.op.name
3032
    ip = self.ip
3033

    
3034
    # shutdown the master IP
3035
    master = self.cfg.GetMasterNode()
3036
    result = self.rpc.call_node_stop_master(master, False)
3037
    result.Raise("Could not disable the master role")
3038

    
3039
    try:
3040
      cluster = self.cfg.GetClusterInfo()
3041
      cluster.cluster_name = clustername
3042
      cluster.master_ip = ip
3043
      self.cfg.Update(cluster, feedback_fn)
3044

    
3045
      # update the known hosts file
3046
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3047
      node_list = self.cfg.GetOnlineNodeList()
3048
      try:
3049
        node_list.remove(master)
3050
      except ValueError:
3051
        pass
3052
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3053
    finally:
3054
      result = self.rpc.call_node_start_master(master, False, False)
3055
      msg = result.fail_msg
3056
      if msg:
3057
        self.LogWarning("Could not re-enable the master role on"
3058
                        " the master, please restart manually: %s", msg)
3059

    
3060
    return clustername
3061

    
3062

    
3063
class LUClusterSetParams(LogicalUnit):
3064
  """Change the parameters of the cluster.
3065

3066
  """
3067
  HPATH = "cluster-modify"
3068
  HTYPE = constants.HTYPE_CLUSTER
3069
  REQ_BGL = False
3070

    
3071
  def CheckArguments(self):
3072
    """Check parameters
3073

3074
    """
3075
    if self.op.uid_pool:
3076
      uidpool.CheckUidPool(self.op.uid_pool)
3077

    
3078
    if self.op.add_uids:
3079
      uidpool.CheckUidPool(self.op.add_uids)
3080

    
3081
    if self.op.remove_uids:
3082
      uidpool.CheckUidPool(self.op.remove_uids)
3083

    
3084
  def ExpandNames(self):
3085
    # FIXME: in the future maybe other cluster params won't require checking on
3086
    # all nodes to be modified.
3087
    self.needed_locks = {
3088
      locking.LEVEL_NODE: locking.ALL_SET,
3089
    }
3090
    self.share_locks[locking.LEVEL_NODE] = 1
3091

    
3092
  def BuildHooksEnv(self):
3093
    """Build hooks env.
3094

3095
    """
3096
    return {
3097
      "OP_TARGET": self.cfg.GetClusterName(),
3098
      "NEW_VG_NAME": self.op.vg_name,
3099
      }
3100

    
3101
  def BuildHooksNodes(self):
3102
    """Build hooks nodes.
3103

3104
    """
3105
    mn = self.cfg.GetMasterNode()
3106
    return ([mn], [mn])
3107

    
3108
  def CheckPrereq(self):
3109
    """Check prerequisites.
3110

3111
    This checks whether the given params don't conflict and
3112
    if the given volume group is valid.
3113

3114
    """
3115
    if self.op.vg_name is not None and not self.op.vg_name:
3116
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3117
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3118
                                   " instances exist", errors.ECODE_INVAL)
3119

    
3120
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
3121
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3122
        raise errors.OpPrereqError("Cannot disable drbd helper while"
3123
                                   " drbd-based instances exist",
3124
                                   errors.ECODE_INVAL)
3125

    
3126
    node_list = self.glm.list_owned(locking.LEVEL_NODE)
3127

    
3128
    # if vg_name not None, checks given volume group on all nodes
3129
    if self.op.vg_name:
3130
      vglist = self.rpc.call_vg_list(node_list)
3131
      for node in node_list:
3132
        msg = vglist[node].fail_msg
3133
        if msg:
3134
          # ignoring down node
3135
          self.LogWarning("Error while gathering data on node %s"
3136
                          " (ignoring node): %s", node, msg)
3137
          continue
3138
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3139
                                              self.op.vg_name,
3140
                                              constants.MIN_VG_SIZE)
3141
        if vgstatus:
3142
          raise errors.OpPrereqError("Error on node '%s': %s" %
3143
                                     (node, vgstatus), errors.ECODE_ENVIRON)
3144

    
3145
    if self.op.drbd_helper:
3146
      # checks given drbd helper on all nodes
3147
      helpers = self.rpc.call_drbd_helper(node_list)
3148
      for node in node_list:
3149
        ninfo = self.cfg.GetNodeInfo(node)
3150
        if ninfo.offline:
3151
          self.LogInfo("Not checking drbd helper on offline node %s", node)
3152
          continue
3153
        msg = helpers[node].fail_msg
3154
        if msg:
3155
          raise errors.OpPrereqError("Error checking drbd helper on node"
3156
                                     " '%s': %s" % (node, msg),
3157
                                     errors.ECODE_ENVIRON)
3158
        node_helper = helpers[node].payload
3159
        if node_helper != self.op.drbd_helper:
3160
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3161
                                     (node, node_helper), errors.ECODE_ENVIRON)
3162

    
3163
    self.cluster = cluster = self.cfg.GetClusterInfo()
3164
    # validate params changes
3165
    if self.op.beparams:
3166
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3167
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3168

    
3169
    if self.op.ndparams:
3170
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3171
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3172

    
3173
      # TODO: we need a more general way to handle resetting
3174
      # cluster-level parameters to default values
3175
      if self.new_ndparams["oob_program"] == "":
3176
        self.new_ndparams["oob_program"] = \
3177
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3178

    
3179
    if self.op.nicparams:
3180
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
3181
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
3182
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
3183
      nic_errors = []
3184

    
3185
      # check all instances for consistency
3186
      for instance in self.cfg.GetAllInstancesInfo().values():
3187
        for nic_idx, nic in enumerate(instance.nics):
3188
          params_copy = copy.deepcopy(nic.nicparams)
3189
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
3190

    
3191
          # check parameter syntax
3192
          try:
3193
            objects.NIC.CheckParameterSyntax(params_filled)
3194
          except errors.ConfigurationError, err:
3195
            nic_errors.append("Instance %s, nic/%d: %s" %
3196
                              (instance.name, nic_idx, err))
3197

    
3198
          # if we're moving instances to routed, check that they have an ip
3199
          target_mode = params_filled[constants.NIC_MODE]
3200
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
3201
            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
3202
                              " address" % (instance.name, nic_idx))
3203
      if nic_errors:
3204
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
3205
                                   "\n".join(nic_errors))
3206

    
3207
    # hypervisor list/parameters
3208
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
3209
    if self.op.hvparams:
3210
      for hv_name, hv_dict in self.op.hvparams.items():
3211
        if hv_name not in self.new_hvparams:
3212
          self.new_hvparams[hv_name] = hv_dict
3213
        else:
3214
          self.new_hvparams[hv_name].update(hv_dict)
3215

    
3216
    # os hypervisor parameters
3217
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
3218
    if self.op.os_hvp:
3219
      for os_name, hvs in self.op.os_hvp.items():
3220
        if os_name not in self.new_os_hvp:
3221
          self.new_os_hvp[os_name] = hvs
3222
        else:
3223
          for hv_name, hv_dict in hvs.items():
3224
            if hv_name not in self.new_os_hvp[os_name]:
3225
              self.new_os_hvp[os_name][hv_name] = hv_dict
3226
            else:
3227
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
3228

    
3229
    # os parameters
3230
    self.new_osp = objects.FillDict(cluster.osparams, {})
3231
    if self.op.osparams:
3232
      for os_name, osp in self.op.osparams.items():
3233
        if os_name not in self.new_osp:
3234
          self.new_osp[os_name] = {}
3235

    
3236
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
3237
                                                  use_none=True)
3238

    
3239
        if not self.new_osp[os_name]:
3240
          # we removed all parameters
3241
          del self.new_osp[os_name]
3242
        else:
3243
          # check the parameter validity (remote check)
3244
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
3245
                         os_name, self.new_osp[os_name])
3246

    
3247
    # changes to the hypervisor list
3248
    if self.op.enabled_hypervisors is not None:
3249
      self.hv_list = self.op.enabled_hypervisors
3250
      for hv in self.hv_list:
3251
        # if the hypervisor doesn't already exist in the cluster
3252
        # hvparams, we initialize it to empty, and then (in both
3253
        # cases) we make sure to fill the defaults, as we might not
3254
        # have a complete defaults list if the hypervisor wasn't
3255
        # enabled before
3256
        if hv not in new_hvp:
3257
          new_hvp[hv] = {}
3258
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
3259
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
3260
    else:
3261
      self.hv_list = cluster.enabled_hypervisors
3262

    
3263
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
3264
      # either the enabled list has changed, or the parameters have, validate
3265
      for hv_name, hv_params in self.new_hvparams.items():
3266
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
3267
            (self.op.enabled_hypervisors and
3268
             hv_name in self.op.enabled_hypervisors)):
3269
          # either this is a new hypervisor, or its parameters have changed
3270
          hv_class = hypervisor.GetHypervisor(hv_name)
3271
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3272
          hv_class.CheckParameterSyntax(hv_params)
3273
          _CheckHVParams(self, node_list, hv_name, hv_params)
3274

    
3275
    if self.op.os_hvp:
3276
      # no need to check any newly-enabled hypervisors, since the
3277
      # defaults have already been checked in the above code-block
3278
      for os_name, os_hvp in self.new_os_hvp.items():
3279
        for hv_name, hv_params in os_hvp.items():
3280
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3281
          # we need to fill in the new os_hvp on top of the actual hv_p
3282
          cluster_defaults = self.new_hvparams.get(hv_name, {})
3283
          new_osp = objects.FillDict(cluster_defaults, hv_params)
3284
          hv_class = hypervisor.GetHypervisor(hv_name)
3285
          hv_class.CheckParameterSyntax(new_osp)
3286
          _CheckHVParams(self, node_list, hv_name, new_osp)
3287

    
3288
    if self.op.default_iallocator:
3289
      alloc_script = utils.FindFile(self.op.default_iallocator,
3290
                                    constants.IALLOCATOR_SEARCH_PATH,
3291
                                    os.path.isfile)
3292
      if alloc_script is None:
3293
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
3294
                                   " specified" % self.op.default_iallocator,
3295
                                   errors.ECODE_INVAL)
3296

    
3297
  def Exec(self, feedback_fn):
3298
    """Change the parameters of the cluster.
3299

3300
    """
3301
    if self.op.vg_name is not None:
3302
      new_volume = self.op.vg_name
3303
      if not new_volume:
3304
        new_volume = None
3305
      if new_volume != self.cfg.GetVGName():
3306
        self.cfg.SetVGName(new_volume)
3307
      else:
3308
        feedback_fn("Cluster LVM configuration already in desired"
3309
                    " state, not changing")
3310
    if self.op.drbd_helper is not None:
3311
      new_helper = self.op.drbd_helper
3312
      if not new_helper:
3313
        new_helper = None
3314
      if new_helper != self.cfg.GetDRBDHelper():
3315
        self.cfg.SetDRBDHelper(new_helper)
3316
      else:
3317
        feedback_fn("Cluster DRBD helper already in desired state,"
3318
                    " not changing")
3319
    if self.op.hvparams:
3320
      self.cluster.hvparams = self.new_hvparams
3321
    if self.op.os_hvp:
3322
      self.cluster.os_hvp = self.new_os_hvp
3323
    if self.op.enabled_hypervisors is not None:
3324
      self.cluster.hvparams = self.new_hvparams
3325
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
3326
    if self.op.beparams:
3327
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
3328
    if self.op.nicparams:
3329
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
3330
    if self.op.osparams:
3331
      self.cluster.osparams = self.new_osp
3332
    if self.op.ndparams:
3333
      self.cluster.ndparams = self.new_ndparams
3334

    
3335
    if self.op.candidate_pool_size is not None:
3336
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
3337
      # we need to update the pool size here, otherwise the save will fail
3338
      _AdjustCandidatePool(self, [])
3339

    
3340
    if self.op.maintain_node_health is not None:
3341
      self.cluster.maintain_node_health = self.op.maintain_node_health
3342

    
3343
    if self.op.prealloc_wipe_disks is not None:
3344
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
3345

    
3346
    if self.op.add_uids is not None:
3347
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
3348

    
3349
    if self.op.remove_uids is not None:
3350
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
3351

    
3352
    if self.op.uid_pool is not None:
3353
      self.cluster.uid_pool = self.op.uid_pool
3354

    
3355
    if self.op.default_iallocator is not None:
3356
      self.cluster.default_iallocator = self.op.default_iallocator
3357

    
3358
    if self.op.reserved_lvs is not None:
3359
      self.cluster.reserved_lvs = self.op.reserved_lvs
3360

    
3361
    def helper_os(aname, mods, desc):
3362
      desc += " OS list"
3363
      lst = getattr(self.cluster, aname)
3364
      for key, val in mods:
3365
        if key == constants.DDM_ADD:
3366
          if val in lst:
3367
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
3368
          else:
3369
            lst.append(val)
3370
        elif key == constants.DDM_REMOVE:
3371
          if val in lst:
3372
            lst.remove(val)
3373
          else:
3374
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3375
        else:
3376
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3377

    
3378
    if self.op.hidden_os:
3379
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3380

    
3381
    if self.op.blacklisted_os:
3382
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3383

    
3384
    if self.op.master_netdev:
3385
      master = self.cfg.GetMasterNode()
3386
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3387
                  self.cluster.master_netdev)
3388
      result = self.rpc.call_node_stop_master(master, False)
3389
      result.Raise("Could not disable the master ip")
3390
      feedback_fn("Changing master_netdev from %s to %s" %
3391
                  (self.cluster.master_netdev, self.op.master_netdev))
3392
      self.cluster.master_netdev = self.op.master_netdev
3393

    
3394
    self.cfg.Update(self.cluster, feedback_fn)
3395

    
3396
    if self.op.master_netdev:
3397
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3398
                  self.op.master_netdev)
3399
      result = self.rpc.call_node_start_master(master, False, False)
3400
      if result.fail_msg:
3401
        self.LogWarning("Could not re-enable the master ip on"
3402
                        " the master, please restart manually: %s",
3403
                        result.fail_msg)
3404

    
3405

    
3406
def _UploadHelper(lu, nodes, fname):
3407
  """Helper for uploading a file and showing warnings.
3408

3409
  """
3410
  if os.path.exists(fname):
3411
    result = lu.rpc.call_upload_file(nodes, fname)
3412
    for to_node, to_result in result.items():
3413
      msg = to_result.fail_msg
3414
      if msg:
3415
        msg = ("Copy of file %s to node %s failed: %s" %
3416
               (fname, to_node, msg))
3417
        lu.proc.LogWarning(msg)
3418

    
3419

    
3420
def _ComputeAncillaryFiles(cluster, redist):
3421
  """Compute files external to Ganeti which need to be consistent.
3422

3423
  @type redist: boolean
3424
  @param redist: Whether to include files which need to be redistributed
3425

3426
  """
3427
  # Compute files for all nodes
3428
  files_all = set([
3429
    constants.SSH_KNOWN_HOSTS_FILE,
3430
    constants.CONFD_HMAC_KEY,
3431
    constants.CLUSTER_DOMAIN_SECRET_FILE,
3432
    ])
3433

    
3434
  if not redist:
3435
    files_all.update(constants.ALL_CERT_FILES)
3436
    files_all.update(ssconf.SimpleStore().GetFileList())
3437

    
3438
  if cluster.modify_etc_hosts:
3439
    files_all.add(constants.ETC_HOSTS)
3440

    
3441
  # Files which must either exist on all nodes or on none
3442
  files_all_opt = set([
3443
    constants.RAPI_USERS_FILE,
3444
    ])
3445

    
3446
  # Files which should only be on master candidates
3447
  files_mc = set()
3448
  if not redist:
3449
    files_mc.add(constants.CLUSTER_CONF_FILE)
3450

    
3451
  # Files which should only be on VM-capable nodes
3452
  files_vm = set(filename
3453
    for hv_name in cluster.enabled_hypervisors
3454
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
3455

    
3456
  # Filenames must be unique
3457
  assert (len(files_all | files_all_opt | files_mc | files_vm) ==
3458
          sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
3459
         "Found file listed in more than one file list"
3460

    
3461
  return (files_all, files_all_opt, files_mc, files_vm)
3462

    
3463

    
3464
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
3465
  """Distribute additional files which are part of the cluster configuration.
3466

3467
  ConfigWriter takes care of distributing the config and ssconf files, but
3468
  there are more files which should be distributed to all nodes. This function
3469
  makes sure those are copied.
3470

3471
  @param lu: calling logical unit
3472
  @param additional_nodes: list of nodes not in the config to distribute to
3473
  @type additional_vm: boolean
3474
  @param additional_vm: whether the additional nodes are vm-capable or not
3475

3476
  """
3477
  # Gather target nodes
3478
  cluster = lu.cfg.GetClusterInfo()
3479
  master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
3480

    
3481
  online_nodes = lu.cfg.GetOnlineNodeList()
3482
  vm_nodes = lu.cfg.GetVmCapableNodeList()
3483

    
3484
  if additional_nodes is not None:
3485
    online_nodes.extend(additional_nodes)
3486
    if additional_vm:
3487
      vm_nodes.extend(additional_nodes)
3488

    
3489
  # Never distribute to master node
3490
  for nodelist in [online_nodes, vm_nodes]:
3491
    if master_info.name in nodelist:
3492
      nodelist.remove(master_info.name)
3493

    
3494
  # Gather file lists
3495
  (files_all, files_all_opt, files_mc, files_vm) = \
3496
    _ComputeAncillaryFiles(cluster, True)
3497

    
3498
  # Never re-distribute configuration file from here
3499
  assert not (constants.CLUSTER_CONF_FILE in files_all or
3500
              constants.CLUSTER_CONF_FILE in files_vm)
3501
  assert not files_mc, "Master candidates not handled in this function"
3502

    
3503
  filemap = [
3504
    (online_nodes, files_all),
3505
    (online_nodes, files_all_opt),
3506
    (vm_nodes, files_vm),
3507
    ]
3508

    
3509
  # Upload the files
3510
  for (node_list, files) in filemap:
3511
    for fname in files:
3512
      _UploadHelper(lu, node_list, fname)
3513

    
3514

    
3515
class LUClusterRedistConf(NoHooksLU):
3516
  """Force the redistribution of cluster configuration.
3517

3518
  This is a very simple LU.
3519

3520
  """
3521
  REQ_BGL = False
3522

    
3523
  def ExpandNames(self):
3524
    self.needed_locks = {
3525
      locking.LEVEL_NODE: locking.ALL_SET,
3526
    }
3527
    self.share_locks[locking.LEVEL_NODE] = 1
3528

    
3529
  def Exec(self, feedback_fn):
3530
    """Redistribute the configuration.
3531

3532
    """
3533
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
3534
    _RedistributeAncillaryFiles(self)
3535

    
3536

    
3537
def _WaitForSync(lu, instance, disks=None, oneshot=False):
3538
  """Sleep and poll for an instance's disk to sync.
3539

3540
  """
3541
  if not instance.disks or disks is not None and not disks:
3542
    return True
3543

    
3544
  disks = _ExpandCheckDisks(instance, disks)
3545

    
3546
  if not oneshot:
3547
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
3548

    
3549
  node = instance.primary_node
3550

    
3551
  for dev in disks:
3552
    lu.cfg.SetDiskID(dev, node)
3553

    
3554
  # TODO: Convert to utils.Retry
3555

    
3556
  retries = 0
3557
  degr_retries = 10 # in seconds, as we sleep 1 second each time
3558
  while True:
3559
    max_time = 0
3560
    done = True
3561
    cumul_degraded = False
3562
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
3563
    msg = rstats.fail_msg
3564
    if msg:
3565
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
3566
      retries += 1
3567
      if retries >= 10:
3568
        raise errors.RemoteError("Can't contact node %s for mirror data,"
3569
                                 " aborting." % node)
3570
      time.sleep(6)
3571
      continue
3572
    rstats = rstats.payload
3573
    retries = 0
3574
    for i, mstat in enumerate(rstats):
3575
      if mstat is None:
3576
        lu.LogWarning("Can't compute data for node %s/%s",
3577
                           node, disks[i].iv_name)
3578
        continue
3579

    
3580
      cumul_degraded = (cumul_degraded or
3581
                        (mstat.is_degraded and mstat.sync_percent is None))
3582
      if mstat.sync_percent is not None:
3583
        done = False
3584
        if mstat.estimated_time is not None:
3585
          rem_time = ("%s remaining (estimated)" %
3586
                      utils.FormatSeconds(mstat.estimated_time))
3587
          max_time = mstat.estimated_time
3588
        else:
3589
          rem_time = "no time estimate"
3590
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
3591
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
3592

    
3593
    # if we're done but degraded, let's do a few small retries, to
3594
    # make sure we see a stable and not transient situation; therefore
3595
    # we force restart of the loop
3596
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
3597
      logging.info("Degraded disks found, %d retries left", degr_retries)
3598
      degr_retries -= 1
3599
      time.sleep(1)
3600
      continue
3601

    
3602
    if done or oneshot:
3603
      break
3604

    
3605
    time.sleep(min(60, max_time))
3606

    
3607
  if done:
3608
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
3609
  return not cumul_degraded
3610

    
3611

    
3612
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
3613
  """Check that mirrors are not degraded.
3614

3615
  The ldisk parameter, if True, will change the test from the
3616
  is_degraded attribute (which represents overall non-ok status for
3617
  the device(s)) to the ldisk (representing the local storage status).
3618

3619
  """
3620
  lu.cfg.SetDiskID(dev, node)
3621

    
3622
  result = True
3623

    
3624
  if on_primary or dev.AssembleOnSecondary():
3625
    rstats = lu.rpc.call_blockdev_find(node, dev)
3626
    msg = rstats.fail_msg
3627
    if msg:
3628
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
3629
      result = False
3630
    elif not rstats.payload:
3631
      lu.LogWarning("Can't find disk on node %s", node)
3632
      result = False
3633
    else:
3634
      if ldisk:
3635
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
3636
      else:
3637
        result = result and not rstats.payload.is_degraded
3638

    
3639
  if dev.children:
3640
    for child in dev.children:
3641
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
3642

    
3643
  return result
3644

    
3645

    
3646
class LUOobCommand(NoHooksLU):
3647
  """Logical unit for OOB handling.
3648

3649
  """
3650
  REG_BGL = False
3651
  _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
3652

    
3653
  def ExpandNames(self):
3654
    """Gather locks we need.
3655

3656
    """
3657
    if self.op.node_names:
3658
      self.op.node_names = _GetWantedNodes(self, self.op.node_names)
3659
      lock_names = self.op.node_names
3660
    else:
3661
      lock_names = locking.ALL_SET
3662

    
3663
    self.needed_locks = {
3664
      locking.LEVEL_NODE: lock_names,
3665
      }
3666

    
3667
  def CheckPrereq(self):
3668
    """Check prerequisites.
3669

3670
    This checks:
3671
     - the node exists in the configuration
3672
     - OOB is supported
3673

3674
    Any errors are signaled by raising errors.OpPrereqError.
3675

3676
    """
3677
    self.nodes = []
3678
    self.master_node = self.cfg.GetMasterNode()
3679

    
3680
    assert self.op.power_delay >= 0.0
3681

    
3682
    if self.op.node_names:
3683
      if (self.op.command in self._SKIP_MASTER and
3684
          self.master_node in self.op.node_names):
3685
        master_node_obj = self.cfg.GetNodeInfo(self.master_node)
3686
        master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
3687

    
3688
        if master_oob_handler:
3689
          additional_text = ("run '%s %s %s' if you want to operate on the"
3690
                             " master regardless") % (master_oob_handler,
3691
                                                      self.op.command,
3692
                                                      self.master_node)
3693
        else:
3694
          additional_text = "it does not support out-of-band operations"
3695

    
3696
        raise errors.OpPrereqError(("Operating on the master node %s is not"
3697
                                    " allowed for %s; %s") %
3698
                                   (self.master_node, self.op.command,
3699
                                    additional_text), errors.ECODE_INVAL)
3700
    else:
3701
      self.op.node_names = self.cfg.GetNodeList()
3702
      if self.op.command in self._SKIP_MASTER:
3703
        self.op.node_names.remove(self.master_node)
3704

    
3705
    if self.op.command in self._SKIP_MASTER:
3706
      assert self.master_node not in self.op.node_names
3707

    
3708
    for node_name in self.op.node_names:
3709
      node = self.cfg.GetNodeInfo(node_name)
3710

    
3711
      if node is None:
3712
        raise errors.OpPrereqError("Node %s not found" % node_name,
3713
                                   errors.ECODE_NOENT)
3714
      else:
3715
        self.nodes.append(node)
3716

    
3717
      if (not self.op.ignore_status and
3718
          (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
3719
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
3720
                                    " not marked offline") % node_name,
3721
                                   errors.ECODE_STATE)
3722

    
3723
  def Exec(self, feedback_fn):
3724
    """Execute OOB and return result if we expect any.
3725

3726
    """
3727
    master_node = self.master_node
3728
    ret = []
3729

    
3730
    for idx, node in enumerate(utils.NiceSort(self.nodes,
3731
                                              key=lambda node: node.name)):
3732
      node_entry = [(constants.RS_NORMAL, node.name)]
3733
      ret.append(node_entry)
3734

    
3735
      oob_program = _SupportsOob(self.cfg, node)
3736

    
3737
      if not oob_program:
3738
        node_entry.append((constants.RS_UNAVAIL, None))
3739
        continue
3740

    
3741
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
3742
                   self.op.command, oob_program, node.name)
3743
      result = self.rpc.call_run_oob(master_node, oob_program,
3744
                                     self.op.command, node.name,
3745
                                     self.op.timeout)
3746

    
3747
      if result.fail_msg:
3748
        self.LogWarning("Out-of-band RPC failed on node '%s': %s",
3749
                        node.name, result.fail_msg)
3750
        node_entry.append((constants.RS_NODATA, None))
3751
      else:
3752
        try:
3753
          self._CheckPayload(result)
3754
        except errors.OpExecError, err:
3755
          self.LogWarning("Payload returned by node '%s' is not valid: %s",
3756
                          node.name, err)
3757
          node_entry.append((constants.RS_NODATA, None))
3758
        else:
3759
          if self.op.command == constants.OOB_HEALTH:
3760
            # For health we should log important events
3761
            for item, status in result.payload:
3762
              if status in [constants.OOB_STATUS_WARNING,
3763
                            constants.OOB_STATUS_CRITICAL]:
3764
                self.LogWarning("Item '%s' on node '%s' has status '%s'",
3765
                                item, node.name, status)
3766

    
3767
          if self.op.command == constants.OOB_POWER_ON:
3768
            node.powered = True
3769
          elif self.op.command == constants.OOB_POWER_OFF:
3770
            node.powered = False
3771
          elif self.op.command == constants.OOB_POWER_STATUS:
3772
            powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
3773
            if powered != node.powered:
3774
              logging.warning(("Recorded power state (%s) of node '%s' does not"
3775
                               " match actual power state (%s)"), node.powered,
3776
                              node.name, powered)
3777

    
3778
          # For configuration changing commands we should update the node
3779
          if self.op.command in (constants.OOB_POWER_ON,
3780
                                 constants.OOB_POWER_OFF):
3781
            self.cfg.Update(node, feedback_fn)
3782

    
3783
          node_entry.append((constants.RS_NORMAL, result.payload))
3784

    
3785
          if (self.op.command == constants.OOB_POWER_ON and
3786
              idx < len(self.nodes) - 1):
3787
            time.sleep(self.op.power_delay)
3788

    
3789
    return ret
3790

    
3791
  def _CheckPayload(self, result):
3792
    """Checks if the payload is valid.
3793

3794
    @param result: RPC result
3795
    @raises errors.OpExecError: If payload is not valid
3796

3797
    """
3798
    errs = []
3799
    if self.op.command == constants.OOB_HEALTH:
3800
      if not isinstance(result.payload, list):
3801
        errs.append("command 'health' is expected to return a list but got %s" %
3802
                    type(result.payload))
3803
      else:
3804
        for item, status in result.payload:
3805
          if status not in constants.OOB_STATUSES:
3806
            errs.append("health item '%s' has invalid status '%s'" %
3807
                        (item, status))
3808

    
3809
    if self.op.command == constants.OOB_POWER_STATUS:
3810
      if not isinstance(result.payload, dict):
3811
        errs.append("power-status is expected to return a dict but got %s" %
3812
                    type(result.payload))
3813

    
3814
    if self.op.command in [
3815
        constants.OOB_POWER_ON,
3816
        constants.OOB_POWER_OFF,
3817
        constants.OOB_POWER_CYCLE,
3818
        ]:
3819
      if result.payload is not None:
3820
        errs.append("%s is expected to not return payload but got '%s'" %
3821
                    (self.op.command, result.payload))
3822

    
3823
    if errs:
3824
      raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
3825
                               utils.CommaJoin(errs))
3826

    
3827
class _OsQuery(_QueryBase):
3828
  FIELDS = query.OS_FIELDS
3829

    
3830
  def ExpandNames(self, lu):
3831
    # Lock all nodes in shared mode
3832
    # Temporary removal of locks, should be reverted later
3833
    # TODO: reintroduce locks when they are lighter-weight
3834
    lu.needed_locks = {}
3835
    #self.share_locks[locking.LEVEL_NODE] = 1
3836
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
3837

    
3838
    # The following variables interact with _QueryBase._GetNames
3839
    if self.names:
3840
      self.wanted = self.names
3841
    else:
3842
      self.wanted = locking.ALL_SET
3843

    
3844
    self.do_locking = self.use_locking
3845

    
3846
  def DeclareLocks(self, lu, level):
3847
    pass
3848

    
3849
  @staticmethod
3850
  def _DiagnoseByOS(rlist):
3851
    """Remaps a per-node return list into an a per-os per-node dictionary
3852

3853
    @param rlist: a map with node names as keys and OS objects as values
3854

3855
    @rtype: dict
3856
    @return: a dictionary with osnames as keys and as value another
3857
        map, with nodes as keys and tuples of (path, status, diagnose,
3858
        variants, parameters, api_versions) as values, eg::
3859

3860
          {"debian-etch": {"node1": [(/usr/lib/..., True, "", [], []),
3861
                                     (/srv/..., False, "invalid api")],
3862
                           "node2": [(/srv/..., True, "", [], [])]}
3863
          }
3864

3865
    """
3866
    all_os = {}
3867
    # we build here the list of nodes that didn't fail the RPC (at RPC
3868
    # level), so that nodes with a non-responding node daemon don't
3869
    # make all OSes invalid
3870
    good_nodes = [node_name for node_name in rlist
3871
                  if not rlist[node_name].fail_msg]
3872
    for node_name, nr in rlist.items():
3873
      if nr.fail_msg or not nr.payload:
3874
        continue
3875
      for (name, path, status, diagnose, variants,
3876
           params, api_versions) in nr.payload:
3877
        if name not in all_os:
3878
          # build a list of nodes for this os containing empty lists
3879
          # for each node in node_list
3880
          all_os[name] = {}
3881
          for nname in good_nodes:
3882
            all_os[name][nname] = []
3883
        # convert params from [name, help] to (name, help)
3884
        params = [tuple(v) for v in params]
3885
        all_os[name][node_name].append((path, status, diagnose,
3886
                                        variants, params, api_versions))
3887
    return all_os
3888

    
3889
  def _GetQueryData(self, lu):
3890
    """Computes the list of nodes and their attributes.
3891

3892
    """
3893
    # Locking is not used
3894
    assert not (compat.any(lu.glm.is_owned(level)
3895
                           for level in locking.LEVELS
3896
                           if level != locking.LEVEL_CLUSTER) or
3897
                self.do_locking or self.use_locking)
3898

    
3899
    valid_nodes = [node.name
3900
                   for node in lu.cfg.GetAllNodesInfo().values()
3901
                   if not node.offline and node.vm_capable]
3902
    pol = self._DiagnoseByOS(lu.rpc.call_os_diagnose(valid_nodes))
3903
    cluster = lu.cfg.GetClusterInfo()
3904

    
3905
    data = {}
3906

    
3907
    for (os_name, os_data) in pol.items():
3908
      info = query.OsInfo(name=os_name, valid=True, node_status=os_data,
3909
                          hidden=(os_name in cluster.hidden_os),
3910
                          blacklisted=(os_name in cluster.blacklisted_os))
3911

    
3912
      variants = set()
3913
      parameters = set()
3914
      api_versions = set()
3915

    
3916
      for idx, osl in enumerate(os_data.values()):
3917
        info.valid = bool(info.valid and osl and osl[0][1])
3918
        if not info.valid:
3919
          break
3920

    
3921
        (node_variants, node_params, node_api) = osl[0][3:6]
3922
        if idx == 0:
3923
          # First entry
3924
          variants.update(node_variants)
3925
          parameters.update(node_params)
3926
          api_versions.update(node_api)
3927
        else:
3928
          # Filter out inconsistent values
3929
          variants.intersection_update(node_variants)
3930
          parameters.intersection_update(node_params)
3931
          api_versions.intersection_update(node_api)
3932

    
3933
      info.variants = list(variants)
3934
      info.parameters = list(parameters)
3935
      info.api_versions = list(api_versions)
3936

    
3937
      data[os_name] = info
3938

    
3939
    # Prepare data in requested order
3940
    return [data[name] for name in self._GetNames(lu, pol.keys(), None)
3941
            if name in data]
3942

    
3943

    
3944
class LUOsDiagnose(NoHooksLU):
3945
  """Logical unit for OS diagnose/query.
3946

3947
  """
3948
  REQ_BGL = False
3949

    
3950
  @staticmethod
3951
  def _BuildFilter(fields, names):
3952
    """Builds a filter for querying OSes.
3953

3954
    """
3955
    name_filter = qlang.MakeSimpleFilter("name", names)
3956

    
3957
    # Legacy behaviour: Hide hidden, blacklisted or invalid OSes if the
3958
    # respective field is not requested
3959
    status_filter = [[qlang.OP_NOT, [qlang.OP_TRUE, fname]]
3960
                     for fname in ["hidden", "blacklisted"]
3961
                     if fname not in fields]
3962
    if "valid" not in fields:
3963
      status_filter.append([qlang.OP_TRUE, "valid"])
3964

    
3965
    if status_filter:
3966
      status_filter.insert(0, qlang.OP_AND)
3967
    else:
3968
      status_filter = None
3969

    
3970
    if name_filter and status_filter:
3971
      return [qlang.OP_AND, name_filter, status_filter]
3972
    elif name_filter:
3973
      return name_filter
3974
    else:
3975
      return status_filter
3976

    
3977
  def CheckArguments(self):
3978
    self.oq = _OsQuery(self._BuildFilter(self.op.output_fields, self.op.names),
3979
                       self.op.output_fields, False)
3980

    
3981
  def ExpandNames(self):
3982
    self.oq.ExpandNames(self)
3983

    
3984
  def Exec(self, feedback_fn):
3985
    return self.oq.OldStyleQuery(self)
3986

    
3987

    
3988
class LUNodeRemove(LogicalUnit):
3989
  """Logical unit for removing a node.
3990

3991
  """
3992
  HPATH = "node-remove"
3993
  HTYPE = constants.HTYPE_NODE
3994

    
3995
  def BuildHooksEnv(self):
3996
    """Build hooks env.
3997

3998
    This doesn't run on the target node in the pre phase as a failed
3999
    node would then be impossible to remove.
4000

4001
    """
4002
    return {
4003
      "OP_TARGET": self.op.node_name,
4004
      "NODE_NAME": self.op.node_name,
4005
      }
4006

    
4007
  def BuildHooksNodes(self):
4008
    """Build hooks nodes.
4009

4010
    """
4011
    all_nodes = self.cfg.GetNodeList()
4012
    try:
4013
      all_nodes.remove(self.op.node_name)
4014
    except ValueError:
4015
      logging.warning("Node '%s', which is about to be removed, was not found"
4016
                      " in the list of all nodes", self.op.node_name)
4017
    return (all_nodes, all_nodes)
4018

    
4019
  def CheckPrereq(self):
4020
    """Check prerequisites.
4021

4022
    This checks:
4023
     - the node exists in the configuration
4024
     - it does not have primary or secondary instances
4025
     - it's not the master
4026

4027
    Any errors are signaled by raising errors.OpPrereqError.
4028

4029
    """
4030
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4031
    node = self.cfg.GetNodeInfo(self.op.node_name)
4032
    assert node is not None
4033

    
4034
    instance_list = self.cfg.GetInstanceList()
4035

    
4036
    masternode = self.cfg.GetMasterNode()
4037
    if node.name == masternode:
4038
      raise errors.OpPrereqError("Node is the master node, failover to another"
4039
                                 " node is required", errors.ECODE_INVAL)
4040

    
4041
    for instance_name in instance_list:
4042
      instance = self.cfg.GetInstanceInfo(instance_name)
4043
      if node.name in instance.all_nodes:
4044
        raise errors.OpPrereqError("Instance %s is still running on the node,"
4045
                                   " please remove first" % instance_name,
4046
                                   errors.ECODE_INVAL)
4047
    self.op.node_name = node.name
4048
    self.node = node
4049

    
4050
  def Exec(self, feedback_fn):
4051
    """Removes the node from the cluster.
4052

4053
    """
4054
    node = self.node
4055
    logging.info("Stopping the node daemon and removing configs from node %s",
4056
                 node.name)
4057

    
4058
    modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
4059

    
4060
    # Promote nodes to master candidate as needed
4061
    _AdjustCandidatePool(self, exceptions=[node.name])
4062
    self.context.RemoveNode(node.name)
4063

    
4064
    # Run post hooks on the node before it's removed
4065
    _RunPostHook(self, node.name)
4066

    
4067
    result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
4068
    msg = result.fail_msg
4069
    if msg:
4070
      self.LogWarning("Errors encountered on the remote node while leaving"
4071
                      " the cluster: %s", msg)
4072