Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ cd098c41

History | View | Annotate | Download (249.7 kB)

1
#
2
#
3

    
4
# Copyright (C) 2006, 2007, 2008 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=W0613,W0201
25

    
26
import os
27
import os.path
28
import time
29
import tempfile
30
import re
31
import platform
32
import logging
33
import copy
34
import random
35

    
36
from ganeti import ssh
37
from ganeti import utils
38
from ganeti import errors
39
from ganeti import hypervisor
40
from ganeti import locking
41
from ganeti import constants
42
from ganeti import objects
43
from ganeti import opcodes
44
from ganeti import serializer
45
from ganeti import ssconf
46

    
47

    
48
class LogicalUnit(object):
49
  """Logical Unit base class.
50

51
  Subclasses must follow these rules:
52
    - implement ExpandNames
53
    - implement CheckPrereq
54
    - implement Exec
55
    - implement BuildHooksEnv
56
    - redefine HPATH and HTYPE
57
    - optionally redefine their run requirements:
58
        REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
59

60
  Note that all commands require root permissions.
61

62
  """
63
  HPATH = None
64
  HTYPE = None
65
  _OP_REQP = []
66
  REQ_BGL = True
67

    
68
  def __init__(self, processor, op, context, rpc):
69
    """Constructor for LogicalUnit.
70

71
    This needs to be overriden in derived classes in order to check op
72
    validity.
73

74
    """
75
    self.proc = processor
76
    self.op = op
77
    self.cfg = context.cfg
78
    self.context = context
79
    self.rpc = rpc
80
    # Dicts used to declare locking needs to mcpu
81
    self.needed_locks = None
82
    self.acquired_locks = {}
83
    self.share_locks = dict(((i, 0) for i in locking.LEVELS))
84
    self.add_locks = {}
85
    self.remove_locks = {}
86
    # Used to force good behavior when calling helper functions
87
    self.recalculate_locks = {}
88
    self.__ssh = None
89
    # logging
90
    self.LogWarning = processor.LogWarning
91
    self.LogInfo = processor.LogInfo
92

    
93
    for attr_name in self._OP_REQP:
94
      attr_val = getattr(op, attr_name, None)
95
      if attr_val is None:
96
        raise errors.OpPrereqError("Required parameter '%s' missing" %
97
                                   attr_name)
98
    self.CheckArguments()
99

    
100
  def __GetSSH(self):
101
    """Returns the SshRunner object
102

103
    """
104
    if not self.__ssh:
105
      self.__ssh = ssh.SshRunner(self.cfg.GetClusterName())
106
    return self.__ssh
107

    
108
  ssh = property(fget=__GetSSH)
109

    
110
  def CheckArguments(self):
111
    """Check syntactic validity for the opcode arguments.
112

113
    This method is for doing a simple syntactic check and ensure
114
    validity of opcode parameters, without any cluster-related
115
    checks. While the same can be accomplished in ExpandNames and/or
116
    CheckPrereq, doing these separate is better because:
117

118
      - ExpandNames is left as as purely a lock-related function
119
      - CheckPrereq is run after we have aquired locks (and possible
120
        waited for them)
121

122
    The function is allowed to change the self.op attribute so that
123
    later methods can no longer worry about missing parameters.
124

125
    """
126
    pass
127

    
128
  def ExpandNames(self):
129
    """Expand names for this LU.
130

131
    This method is called before starting to execute the opcode, and it should
132
    update all the parameters of the opcode to their canonical form (e.g. a
133
    short node name must be fully expanded after this method has successfully
134
    completed). This way locking, hooks, logging, ecc. can work correctly.
135

136
    LUs which implement this method must also populate the self.needed_locks
137
    member, as a dict with lock levels as keys, and a list of needed lock names
138
    as values. Rules:
139

140
      - use an empty dict if you don't need any lock
141
      - if you don't need any lock at a particular level omit that level
142
      - don't put anything for the BGL level
143
      - if you want all locks at a level use locking.ALL_SET as a value
144

145
    If you need to share locks (rather than acquire them exclusively) at one
146
    level you can modify self.share_locks, setting a true value (usually 1) for
147
    that level. By default locks are not shared.
148

149
    Examples::
150

151
      # Acquire all nodes and one instance
152
      self.needed_locks = {
153
        locking.LEVEL_NODE: locking.ALL_SET,
154
        locking.LEVEL_INSTANCE: ['instance1.example.tld'],
155
      }
156
      # Acquire just two nodes
157
      self.needed_locks = {
158
        locking.LEVEL_NODE: ['node1.example.tld', 'node2.example.tld'],
159
      }
160
      # Acquire no locks
161
      self.needed_locks = {} # No, you can't leave it to the default value None
162

163
    """
164
    # The implementation of this method is mandatory only if the new LU is
165
    # concurrent, so that old LUs don't need to be changed all at the same
166
    # time.
167
    if self.REQ_BGL:
168
      self.needed_locks = {} # Exclusive LUs don't need locks.
169
    else:
170
      raise NotImplementedError
171

    
172
  def DeclareLocks(self, level):
173
    """Declare LU locking needs for a level
174

175
    While most LUs can just declare their locking needs at ExpandNames time,
176
    sometimes there's the need to calculate some locks after having acquired
177
    the ones before. This function is called just before acquiring locks at a
178
    particular level, but after acquiring the ones at lower levels, and permits
179
    such calculations. It can be used to modify self.needed_locks, and by
180
    default it does nothing.
181

182
    This function is only called if you have something already set in
183
    self.needed_locks for the level.
184

185
    @param level: Locking level which is going to be locked
186
    @type level: member of ganeti.locking.LEVELS
187

188
    """
189

    
190
  def CheckPrereq(self):
191
    """Check prerequisites for this LU.
192

193
    This method should check that the prerequisites for the execution
194
    of this LU are fulfilled. It can do internode communication, but
195
    it should be idempotent - no cluster or system changes are
196
    allowed.
197

198
    The method should raise errors.OpPrereqError in case something is
199
    not fulfilled. Its return value is ignored.
200

201
    This method should also update all the parameters of the opcode to
202
    their canonical form if it hasn't been done by ExpandNames before.
203

204
    """
205
    raise NotImplementedError
206

    
207
  def Exec(self, feedback_fn):
208
    """Execute the LU.
209

210
    This method should implement the actual work. It should raise
211
    errors.OpExecError for failures that are somewhat dealt with in
212
    code, or expected.
213

214
    """
215
    raise NotImplementedError
216

    
217
  def BuildHooksEnv(self):
218
    """Build hooks environment for this LU.
219

220
    This method should return a three-node tuple consisting of: a dict
221
    containing the environment that will be used for running the
222
    specific hook for this LU, a list of node names on which the hook
223
    should run before the execution, and a list of node names on which
224
    the hook should run after the execution.
225

226
    The keys of the dict must not have 'GANETI_' prefixed as this will
227
    be handled in the hooks runner. Also note additional keys will be
228
    added by the hooks runner. If the LU doesn't define any
229
    environment, an empty dict (and not None) should be returned.
230

231
    No nodes should be returned as an empty list (and not None).
232

233
    Note that if the HPATH for a LU class is None, this function will
234
    not be called.
235

236
    """
237
    raise NotImplementedError
238

    
239
  def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
240
    """Notify the LU about the results of its hooks.
241

242
    This method is called every time a hooks phase is executed, and notifies
243
    the Logical Unit about the hooks' result. The LU can then use it to alter
244
    its result based on the hooks.  By default the method does nothing and the
245
    previous result is passed back unchanged but any LU can define it if it
246
    wants to use the local cluster hook-scripts somehow.
247

248
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
249
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
250
    @param hook_results: the results of the multi-node hooks rpc call
251
    @param feedback_fn: function used send feedback back to the caller
252
    @param lu_result: the previous Exec result this LU had, or None
253
        in the PRE phase
254
    @return: the new Exec result, based on the previous result
255
        and hook results
256

257
    """
258
    return lu_result
259

    
260
  def _ExpandAndLockInstance(self):
261
    """Helper function to expand and lock an instance.
262

263
    Many LUs that work on an instance take its name in self.op.instance_name
264
    and need to expand it and then declare the expanded name for locking. This
265
    function does it, and then updates self.op.instance_name to the expanded
266
    name. It also initializes needed_locks as a dict, if this hasn't been done
267
    before.
268

269
    """
270
    if self.needed_locks is None:
271
      self.needed_locks = {}
272
    else:
273
      assert locking.LEVEL_INSTANCE not in self.needed_locks, \
274
        "_ExpandAndLockInstance called with instance-level locks set"
275
    expanded_name = self.cfg.ExpandInstanceName(self.op.instance_name)
276
    if expanded_name is None:
277
      raise errors.OpPrereqError("Instance '%s' not known" %
278
                                  self.op.instance_name)
279
    self.needed_locks[locking.LEVEL_INSTANCE] = expanded_name
280
    self.op.instance_name = expanded_name
281

    
282
  def _LockInstancesNodes(self, primary_only=False):
283
    """Helper function to declare instances' nodes for locking.
284

285
    This function should be called after locking one or more instances to lock
286
    their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
287
    with all primary or secondary nodes for instances already locked and
288
    present in self.needed_locks[locking.LEVEL_INSTANCE].
289

290
    It should be called from DeclareLocks, and for safety only works if
291
    self.recalculate_locks[locking.LEVEL_NODE] is set.
292

293
    In the future it may grow parameters to just lock some instance's nodes, or
294
    to just lock primaries or secondary nodes, if needed.
295

296
    If should be called in DeclareLocks in a way similar to::
297

298
      if level == locking.LEVEL_NODE:
299
        self._LockInstancesNodes()
300

301
    @type primary_only: boolean
302
    @param primary_only: only lock primary nodes of locked instances
303

304
    """
305
    assert locking.LEVEL_NODE in self.recalculate_locks, \
306
      "_LockInstancesNodes helper function called with no nodes to recalculate"
307

    
308
    # TODO: check if we're really been called with the instance locks held
309

    
310
    # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
311
    # future we might want to have different behaviors depending on the value
312
    # of self.recalculate_locks[locking.LEVEL_NODE]
313
    wanted_nodes = []
314
    for instance_name in self.acquired_locks[locking.LEVEL_INSTANCE]:
315
      instance = self.context.cfg.GetInstanceInfo(instance_name)
316
      wanted_nodes.append(instance.primary_node)
317
      if not primary_only:
318
        wanted_nodes.extend(instance.secondary_nodes)
319

    
320
    if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
321
      self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
322
    elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
323
      self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
324

    
325
    del self.recalculate_locks[locking.LEVEL_NODE]
326

    
327

    
328
class NoHooksLU(LogicalUnit):
329
  """Simple LU which runs no hooks.
330

331
  This LU is intended as a parent for other LogicalUnits which will
332
  run no hooks, in order to reduce duplicate code.
333

334
  """
335
  HPATH = None
336
  HTYPE = None
337

    
338

    
339
def _GetWantedNodes(lu, nodes):
340
  """Returns list of checked and expanded node names.
341

342
  @type lu: L{LogicalUnit}
343
  @param lu: the logical unit on whose behalf we execute
344
  @type nodes: list
345
  @param nodes: list of node names or None for all nodes
346
  @rtype: list
347
  @return: the list of nodes, sorted
348
  @raise errors.OpProgrammerError: if the nodes parameter is wrong type
349

350
  """
351
  if not isinstance(nodes, list):
352
    raise errors.OpPrereqError("Invalid argument type 'nodes'")
353

    
354
  if not nodes:
355
    raise errors.ProgrammerError("_GetWantedNodes should only be called with a"
356
      " non-empty list of nodes whose name is to be expanded.")
357

    
358
  wanted = []
359
  for name in nodes:
360
    node = lu.cfg.ExpandNodeName(name)
361
    if node is None:
362
      raise errors.OpPrereqError("No such node name '%s'" % name)
363
    wanted.append(node)
364

    
365
  return utils.NiceSort(wanted)
366

    
367

    
368
def _GetWantedInstances(lu, instances):
369
  """Returns list of checked and expanded instance names.
370

371
  @type lu: L{LogicalUnit}
372
  @param lu: the logical unit on whose behalf we execute
373
  @type instances: list
374
  @param instances: list of instance names or None for all instances
375
  @rtype: list
376
  @return: the list of instances, sorted
377
  @raise errors.OpPrereqError: if the instances parameter is wrong type
378
  @raise errors.OpPrereqError: if any of the passed instances is not found
379

380
  """
381
  if not isinstance(instances, list):
382
    raise errors.OpPrereqError("Invalid argument type 'instances'")
383

    
384
  if instances:
385
    wanted = []
386

    
387
    for name in instances:
388
      instance = lu.cfg.ExpandInstanceName(name)
389
      if instance is None:
390
        raise errors.OpPrereqError("No such instance name '%s'" % name)
391
      wanted.append(instance)
392

    
393
  else:
394
    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
395
  return wanted
396

    
397

    
398
def _CheckOutputFields(static, dynamic, selected):
399
  """Checks whether all selected fields are valid.
400

401
  @type static: L{utils.FieldSet}
402
  @param static: static fields set
403
  @type dynamic: L{utils.FieldSet}
404
  @param dynamic: dynamic fields set
405

406
  """
407
  f = utils.FieldSet()
408
  f.Extend(static)
409
  f.Extend(dynamic)
410

    
411
  delta = f.NonMatching(selected)
412
  if delta:
413
    raise errors.OpPrereqError("Unknown output fields selected: %s"
414
                               % ",".join(delta))
415

    
416

    
417
def _CheckBooleanOpField(op, name):
418
  """Validates boolean opcode parameters.
419

420
  This will ensure that an opcode parameter is either a boolean value,
421
  or None (but that it always exists).
422

423
  """
424
  val = getattr(op, name, None)
425
  if not (val is None or isinstance(val, bool)):
426
    raise errors.OpPrereqError("Invalid boolean parameter '%s' (%s)" %
427
                               (name, str(val)))
428
  setattr(op, name, val)
429

    
430

    
431
def _CheckNodeOnline(lu, node):
432
  """Ensure that a given node is online.
433

434
  @param lu: the LU on behalf of which we make the check
435
  @param node: the node to check
436
  @raise errors.OpPrereqError: if the node is offline
437

438
  """
439
  if lu.cfg.GetNodeInfo(node).offline:
440
    raise errors.OpPrereqError("Can't use offline node %s" % node)
441

    
442

    
443
def _CheckNodeNotDrained(lu, node):
444
  """Ensure that a given node is not drained.
445

446
  @param lu: the LU on behalf of which we make the check
447
  @param node: the node to check
448
  @raise errors.OpPrereqError: if the node is drained
449

450
  """
451
  if lu.cfg.GetNodeInfo(node).drained:
452
    raise errors.OpPrereqError("Can't use drained node %s" % node)
453

    
454

    
455
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
456
                          memory, vcpus, nics, disk_template, disks):
457
  """Builds instance related env variables for hooks
458

459
  This builds the hook environment from individual variables.
460

461
  @type name: string
462
  @param name: the name of the instance
463
  @type primary_node: string
464
  @param primary_node: the name of the instance's primary node
465
  @type secondary_nodes: list
466
  @param secondary_nodes: list of secondary nodes as strings
467
  @type os_type: string
468
  @param os_type: the name of the instance's OS
469
  @type status: boolean
470
  @param status: the should_run status of the instance
471
  @type memory: string
472
  @param memory: the memory size of the instance
473
  @type vcpus: string
474
  @param vcpus: the count of VCPUs the instance has
475
  @type nics: list
476
  @param nics: list of tuples (ip, bridge, mac) representing
477
      the NICs the instance  has
478
  @type disk_template: string
479
  @param disk_template: the distk template of the instance
480
  @type disks: list
481
  @param disks: the list of (size, mode) pairs
482
  @rtype: dict
483
  @return: the hook environment for this instance
484

485
  """
486
  if status:
487
    str_status = "up"
488
  else:
489
    str_status = "down"
490
  env = {
491
    "OP_TARGET": name,
492
    "INSTANCE_NAME": name,
493
    "INSTANCE_PRIMARY": primary_node,
494
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
495
    "INSTANCE_OS_TYPE": os_type,
496
    "INSTANCE_STATUS": str_status,
497
    "INSTANCE_MEMORY": memory,
498
    "INSTANCE_VCPUS": vcpus,
499
    "INSTANCE_DISK_TEMPLATE": disk_template,
500
  }
501

    
502
  if nics:
503
    nic_count = len(nics)
504
    for idx, (ip, bridge, mac) in enumerate(nics):
505
      if ip is None:
506
        ip = ""
507
      env["INSTANCE_NIC%d_IP" % idx] = ip
508
      env["INSTANCE_NIC%d_BRIDGE" % idx] = bridge
509
      env["INSTANCE_NIC%d_MAC" % idx] = mac
510
  else:
511
    nic_count = 0
512

    
513
  env["INSTANCE_NIC_COUNT"] = nic_count
514

    
515
  if disks:
516
    disk_count = len(disks)
517
    for idx, (size, mode) in enumerate(disks):
518
      env["INSTANCE_DISK%d_SIZE" % idx] = size
519
      env["INSTANCE_DISK%d_MODE" % idx] = mode
520
  else:
521
    disk_count = 0
522

    
523
  env["INSTANCE_DISK_COUNT"] = disk_count
524

    
525
  return env
526

    
527

    
528
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
529
  """Builds instance related env variables for hooks from an object.
530

531
  @type lu: L{LogicalUnit}
532
  @param lu: the logical unit on whose behalf we execute
533
  @type instance: L{objects.Instance}
534
  @param instance: the instance for which we should build the
535
      environment
536
  @type override: dict
537
  @param override: dictionary with key/values that will override
538
      our values
539
  @rtype: dict
540
  @return: the hook environment dictionary
541

542
  """
543
  bep = lu.cfg.GetClusterInfo().FillBE(instance)
544
  args = {
545
    'name': instance.name,
546
    'primary_node': instance.primary_node,
547
    'secondary_nodes': instance.secondary_nodes,
548
    'os_type': instance.os,
549
    'status': instance.admin_up,
550
    'memory': bep[constants.BE_MEMORY],
551
    'vcpus': bep[constants.BE_VCPUS],
552
    'nics': [(nic.ip, nic.bridge, nic.mac) for nic in instance.nics],
553
    'disk_template': instance.disk_template,
554
    'disks': [(disk.size, disk.mode) for disk in instance.disks],
555
  }
556
  if override:
557
    args.update(override)
558
  return _BuildInstanceHookEnv(**args)
559

    
560

    
561
def _AdjustCandidatePool(lu):
562
  """Adjust the candidate pool after node operations.
563

564
  """
565
  mod_list = lu.cfg.MaintainCandidatePool()
566
  if mod_list:
567
    lu.LogInfo("Promoted nodes to master candidate role: %s",
568
               ", ".join(node.name for node in mod_list))
569
    for name in mod_list:
570
      lu.context.ReaddNode(name)
571
  mc_now, mc_max = lu.cfg.GetMasterCandidateStats()
572
  if mc_now > mc_max:
573
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
574
               (mc_now, mc_max))
575

    
576

    
577
def _CheckInstanceBridgesExist(lu, instance):
578
  """Check that the brigdes needed by an instance exist.
579

580
  """
581
  # check bridges existance
582
  brlist = [nic.bridge for nic in instance.nics]
583
  result = lu.rpc.call_bridges_exist(instance.primary_node, brlist)
584
  result.Raise()
585
  if not result.data:
586
    raise errors.OpPrereqError("One or more target bridges %s does not"
587
                               " exist on destination node '%s'" %
588
                               (brlist, instance.primary_node))
589

    
590

    
591
class LUDestroyCluster(NoHooksLU):
592
  """Logical unit for destroying the cluster.
593

594
  """
595
  _OP_REQP = []
596

    
597
  def CheckPrereq(self):
598
    """Check prerequisites.
599

600
    This checks whether the cluster is empty.
601

602
    Any errors are signalled by raising errors.OpPrereqError.
603

604
    """
605
    master = self.cfg.GetMasterNode()
606

    
607
    nodelist = self.cfg.GetNodeList()
608
    if len(nodelist) != 1 or nodelist[0] != master:
609
      raise errors.OpPrereqError("There are still %d node(s) in"
610
                                 " this cluster." % (len(nodelist) - 1))
611
    instancelist = self.cfg.GetInstanceList()
612
    if instancelist:
613
      raise errors.OpPrereqError("There are still %d instance(s) in"
614
                                 " this cluster." % len(instancelist))
615

    
616
  def Exec(self, feedback_fn):
617
    """Destroys the cluster.
618

619
    """
620
    master = self.cfg.GetMasterNode()
621
    result = self.rpc.call_node_stop_master(master, False)
622
    result.Raise()
623
    if not result.data:
624
      raise errors.OpExecError("Could not disable the master role")
625
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
626
    utils.CreateBackup(priv_key)
627
    utils.CreateBackup(pub_key)
628
    return master
629

    
630

    
631
class LUVerifyCluster(LogicalUnit):
632
  """Verifies the cluster status.
633

634
  """
635
  HPATH = "cluster-verify"
636
  HTYPE = constants.HTYPE_CLUSTER
637
  _OP_REQP = ["skip_checks"]
638
  REQ_BGL = False
639

    
640
  def ExpandNames(self):
641
    self.needed_locks = {
642
      locking.LEVEL_NODE: locking.ALL_SET,
643
      locking.LEVEL_INSTANCE: locking.ALL_SET,
644
    }
645
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
646

    
647
  def _VerifyNode(self, nodeinfo, file_list, local_cksum,
648
                  node_result, feedback_fn, master_files,
649
                  drbd_map, vg_name):
650
    """Run multiple tests against a node.
651

652
    Test list:
653

654
      - compares ganeti version
655
      - checks vg existance and size > 20G
656
      - checks config file checksum
657
      - checks ssh to other nodes
658

659
    @type nodeinfo: L{objects.Node}
660
    @param nodeinfo: the node to check
661
    @param file_list: required list of files
662
    @param local_cksum: dictionary of local files and their checksums
663
    @param node_result: the results from the node
664
    @param feedback_fn: function used to accumulate results
665
    @param master_files: list of files that only masters should have
666
    @param drbd_map: the useddrbd minors for this node, in
667
        form of minor: (instance, must_exist) which correspond to instances
668
        and their running status
669
    @param vg_name: Ganeti Volume Group (result of self.cfg.GetVGName())
670

671
    """
672
    node = nodeinfo.name
673

    
674
    # main result, node_result should be a non-empty dict
675
    if not node_result or not isinstance(node_result, dict):
676
      feedback_fn("  - ERROR: unable to verify node %s." % (node,))
677
      return True
678

    
679
    # compares ganeti version
680
    local_version = constants.PROTOCOL_VERSION
681
    remote_version = node_result.get('version', None)
682
    if not (remote_version and isinstance(remote_version, (list, tuple)) and
683
            len(remote_version) == 2):
684
      feedback_fn("  - ERROR: connection to %s failed" % (node))
685
      return True
686

    
687
    if local_version != remote_version[0]:
688
      feedback_fn("  - ERROR: incompatible protocol versions: master %s,"
689
                  " node %s %s" % (local_version, node, remote_version[0]))
690
      return True
691

    
692
    # node seems compatible, we can actually try to look into its results
693

    
694
    bad = False
695

    
696
    # full package version
697
    if constants.RELEASE_VERSION != remote_version[1]:
698
      feedback_fn("  - WARNING: software version mismatch: master %s,"
699
                  " node %s %s" %
700
                  (constants.RELEASE_VERSION, node, remote_version[1]))
701

    
702
    # checks vg existence and size > 20G
703
    if vg_name is not None:
704
      vglist = node_result.get(constants.NV_VGLIST, None)
705
      if not vglist:
706
        feedback_fn("  - ERROR: unable to check volume groups on node %s." %
707
                        (node,))
708
        bad = True
709
      else:
710
        vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
711
                                              constants.MIN_VG_SIZE)
712
        if vgstatus:
713
          feedback_fn("  - ERROR: %s on node %s" % (vgstatus, node))
714
          bad = True
715

    
716
    # checks config file checksum
717

    
718
    remote_cksum = node_result.get(constants.NV_FILELIST, None)
719
    if not isinstance(remote_cksum, dict):
720
      bad = True
721
      feedback_fn("  - ERROR: node hasn't returned file checksum data")
722
    else:
723
      for file_name in file_list:
724
        node_is_mc = nodeinfo.master_candidate
725
        must_have_file = file_name not in master_files
726
        if file_name not in remote_cksum:
727
          if node_is_mc or must_have_file:
728
            bad = True
729
            feedback_fn("  - ERROR: file '%s' missing" % file_name)
730
        elif remote_cksum[file_name] != local_cksum[file_name]:
731
          if node_is_mc or must_have_file:
732
            bad = True
733
            feedback_fn("  - ERROR: file '%s' has wrong checksum" % file_name)
734
          else:
735
            # not candidate and this is not a must-have file
736
            bad = True
737
            feedback_fn("  - ERROR: non master-candidate has old/wrong file"
738
                        " '%s'" % file_name)
739
        else:
740
          # all good, except non-master/non-must have combination
741
          if not node_is_mc and not must_have_file:
742
            feedback_fn("  - ERROR: file '%s' should not exist on non master"
743
                        " candidates" % file_name)
744

    
745
    # checks ssh to any
746

    
747
    if constants.NV_NODELIST not in node_result:
748
      bad = True
749
      feedback_fn("  - ERROR: node hasn't returned node ssh connectivity data")
750
    else:
751
      if node_result[constants.NV_NODELIST]:
752
        bad = True
753
        for node in node_result[constants.NV_NODELIST]:
754
          feedback_fn("  - ERROR: ssh communication with node '%s': %s" %
755
                          (node, node_result[constants.NV_NODELIST][node]))
756

    
757
    if constants.NV_NODENETTEST not in node_result:
758
      bad = True
759
      feedback_fn("  - ERROR: node hasn't returned node tcp connectivity data")
760
    else:
761
      if node_result[constants.NV_NODENETTEST]:
762
        bad = True
763
        nlist = utils.NiceSort(node_result[constants.NV_NODENETTEST].keys())
764
        for node in nlist:
765
          feedback_fn("  - ERROR: tcp communication with node '%s': %s" %
766
                          (node, node_result[constants.NV_NODENETTEST][node]))
767

    
768
    hyp_result = node_result.get(constants.NV_HYPERVISOR, None)
769
    if isinstance(hyp_result, dict):
770
      for hv_name, hv_result in hyp_result.iteritems():
771
        if hv_result is not None:
772
          feedback_fn("  - ERROR: hypervisor %s verify failure: '%s'" %
773
                      (hv_name, hv_result))
774

    
775
    # check used drbd list
776
    if vg_name is not None:
777
      used_minors = node_result.get(constants.NV_DRBDLIST, [])
778
      if not isinstance(used_minors, (tuple, list)):
779
        feedback_fn("  - ERROR: cannot parse drbd status file: %s" %
780
                    str(used_minors))
781
      else:
782
        for minor, (iname, must_exist) in drbd_map.items():
783
          if minor not in used_minors and must_exist:
784
            feedback_fn("  - ERROR: drbd minor %d of instance %s is"
785
                        " not active" % (minor, iname))
786
            bad = True
787
        for minor in used_minors:
788
          if minor not in drbd_map:
789
            feedback_fn("  - ERROR: unallocated drbd minor %d is in use" %
790
                        minor)
791
            bad = True
792

    
793
    return bad
794

    
795
  def _VerifyInstance(self, instance, instanceconfig, node_vol_is,
796
                      node_instance, feedback_fn, n_offline):
797
    """Verify an instance.
798

799
    This function checks to see if the required block devices are
800
    available on the instance's node.
801

802
    """
803
    bad = False
804

    
805
    node_current = instanceconfig.primary_node
806

    
807
    node_vol_should = {}
808
    instanceconfig.MapLVsByNode(node_vol_should)
809

    
810
    for node in node_vol_should:
811
      if node in n_offline:
812
        # ignore missing volumes on offline nodes
813
        continue
814
      for volume in node_vol_should[node]:
815
        if node not in node_vol_is or volume not in node_vol_is[node]:
816
          feedback_fn("  - ERROR: volume %s missing on node %s" %
817
                          (volume, node))
818
          bad = True
819

    
820
    if instanceconfig.admin_up:
821
      if ((node_current not in node_instance or
822
          not instance in node_instance[node_current]) and
823
          node_current not in n_offline):
824
        feedback_fn("  - ERROR: instance %s not running on node %s" %
825
                        (instance, node_current))
826
        bad = True
827

    
828
    for node in node_instance:
829
      if (not node == node_current):
830
        if instance in node_instance[node]:
831
          feedback_fn("  - ERROR: instance %s should not run on node %s" %
832
                          (instance, node))
833
          bad = True
834

    
835
    return bad
836

    
837
  def _VerifyOrphanVolumes(self, node_vol_should, node_vol_is, feedback_fn):
838
    """Verify if there are any unknown volumes in the cluster.
839

840
    The .os, .swap and backup volumes are ignored. All other volumes are
841
    reported as unknown.
842

843
    """
844
    bad = False
845

    
846
    for node in node_vol_is:
847
      for volume in node_vol_is[node]:
848
        if node not in node_vol_should or volume not in node_vol_should[node]:
849
          feedback_fn("  - ERROR: volume %s on node %s should not exist" %
850
                      (volume, node))
851
          bad = True
852
    return bad
853

    
854
  def _VerifyOrphanInstances(self, instancelist, node_instance, feedback_fn):
855
    """Verify the list of running instances.
856

857
    This checks what instances are running but unknown to the cluster.
858

859
    """
860
    bad = False
861
    for node in node_instance:
862
      for runninginstance in node_instance[node]:
863
        if runninginstance not in instancelist:
864
          feedback_fn("  - ERROR: instance %s on node %s should not exist" %
865
                          (runninginstance, node))
866
          bad = True
867
    return bad
868

    
869
  def _VerifyNPlusOneMemory(self, node_info, instance_cfg, feedback_fn):
870
    """Verify N+1 Memory Resilience.
871

872
    Check that if one single node dies we can still start all the instances it
873
    was primary for.
874

875
    """
876
    bad = False
877

    
878
    for node, nodeinfo in node_info.iteritems():
879
      # This code checks that every node which is now listed as secondary has
880
      # enough memory to host all instances it is supposed to should a single
881
      # other node in the cluster fail.
882
      # FIXME: not ready for failover to an arbitrary node
883
      # FIXME: does not support file-backed instances
884
      # WARNING: we currently take into account down instances as well as up
885
      # ones, considering that even if they're down someone might want to start
886
      # them even in the event of a node failure.
887
      for prinode, instances in nodeinfo['sinst-by-pnode'].iteritems():
888
        needed_mem = 0
889
        for instance in instances:
890
          bep = self.cfg.GetClusterInfo().FillBE(instance_cfg[instance])
891
          if bep[constants.BE_AUTO_BALANCE]:
892
            needed_mem += bep[constants.BE_MEMORY]
893
        if nodeinfo['mfree'] < needed_mem:
894
          feedback_fn("  - ERROR: not enough memory on node %s to accomodate"
895
                      " failovers should node %s fail" % (node, prinode))
896
          bad = True
897
    return bad
898

    
899
  def CheckPrereq(self):
900
    """Check prerequisites.
901

902
    Transform the list of checks we're going to skip into a set and check that
903
    all its members are valid.
904

905
    """
906
    self.skip_set = frozenset(self.op.skip_checks)
907
    if not constants.VERIFY_OPTIONAL_CHECKS.issuperset(self.skip_set):
908
      raise errors.OpPrereqError("Invalid checks to be skipped specified")
909

    
910
  def BuildHooksEnv(self):
911
    """Build hooks env.
912

913
    Cluster-Verify hooks just rone in the post phase and their failure makes
914
    the output be logged in the verify output and the verification to fail.
915

916
    """
917
    all_nodes = self.cfg.GetNodeList()
918
    env = {
919
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
920
      }
921
    for node in self.cfg.GetAllNodesInfo().values():
922
      env["NODE_TAGS_%s" % node.name] = " ".join(node.GetTags())
923

    
924
    return env, [], all_nodes
925

    
926
  def Exec(self, feedback_fn):
927
    """Verify integrity of cluster, performing various test on nodes.
928

929
    """
930
    bad = False
931
    feedback_fn("* Verifying global settings")
932
    for msg in self.cfg.VerifyConfig():
933
      feedback_fn("  - ERROR: %s" % msg)
934

    
935
    vg_name = self.cfg.GetVGName()
936
    hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
937
    nodelist = utils.NiceSort(self.cfg.GetNodeList())
938
    nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
939
    instancelist = utils.NiceSort(self.cfg.GetInstanceList())
940
    instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
941
                        for iname in instancelist)
942
    i_non_redundant = [] # Non redundant instances
943
    i_non_a_balanced = [] # Non auto-balanced instances
944
    n_offline = [] # List of offline nodes
945
    n_drained = [] # List of nodes being drained
946
    node_volume = {}
947
    node_instance = {}
948
    node_info = {}
949
    instance_cfg = {}
950

    
951
    # FIXME: verify OS list
952
    # do local checksums
953
    master_files = [constants.CLUSTER_CONF_FILE]
954

    
955
    file_names = ssconf.SimpleStore().GetFileList()
956
    file_names.append(constants.SSL_CERT_FILE)
957
    file_names.append(constants.RAPI_CERT_FILE)
958
    file_names.extend(master_files)
959

    
960
    local_checksums = utils.FingerprintFiles(file_names)
961

    
962
    feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
963
    node_verify_param = {
964
      constants.NV_FILELIST: file_names,
965
      constants.NV_NODELIST: [node.name for node in nodeinfo
966
                              if not node.offline],
967
      constants.NV_HYPERVISOR: hypervisors,
968
      constants.NV_NODENETTEST: [(node.name, node.primary_ip,
969
                                  node.secondary_ip) for node in nodeinfo
970
                                 if not node.offline],
971
      constants.NV_INSTANCELIST: hypervisors,
972
      constants.NV_VERSION: None,
973
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
974
      }
975
    if vg_name is not None:
976
      node_verify_param[constants.NV_VGLIST] = None
977
      node_verify_param[constants.NV_LVLIST] = vg_name
978
      node_verify_param[constants.NV_DRBDLIST] = None
979
    all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
980
                                           self.cfg.GetClusterName())
981

    
982
    cluster = self.cfg.GetClusterInfo()
983
    master_node = self.cfg.GetMasterNode()
984
    all_drbd_map = self.cfg.ComputeDRBDMap()
985

    
986
    for node_i in nodeinfo:
987
      node = node_i.name
988
      nresult = all_nvinfo[node].data
989

    
990
      if node_i.offline:
991
        feedback_fn("* Skipping offline node %s" % (node,))
992
        n_offline.append(node)
993
        continue
994

    
995
      if node == master_node:
996
        ntype = "master"
997
      elif node_i.master_candidate:
998
        ntype = "master candidate"
999
      elif node_i.drained:
1000
        ntype = "drained"
1001
        n_drained.append(node)
1002
      else:
1003
        ntype = "regular"
1004
      feedback_fn("* Verifying node %s (%s)" % (node, ntype))
1005

    
1006
      if all_nvinfo[node].failed or not isinstance(nresult, dict):
1007
        feedback_fn("  - ERROR: connection to %s failed" % (node,))
1008
        bad = True
1009
        continue
1010

    
1011
      node_drbd = {}
1012
      for minor, instance in all_drbd_map[node].items():
1013
        if instance not in instanceinfo:
1014
          feedback_fn("  - ERROR: ghost instance '%s' in temporary DRBD map" %
1015
                      instance)
1016
          # ghost instance should not be running, but otherwise we
1017
          # don't give double warnings (both ghost instance and
1018
          # unallocated minor in use)
1019
          node_drbd[minor] = (instance, False)
1020
        else:
1021
          instance = instanceinfo[instance]
1022
          node_drbd[minor] = (instance.name, instance.admin_up)
1023
      result = self._VerifyNode(node_i, file_names, local_checksums,
1024
                                nresult, feedback_fn, master_files,
1025
                                node_drbd, vg_name)
1026
      bad = bad or result
1027

    
1028
      lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
1029
      if vg_name is None:
1030
        node_volume[node] = {}
1031
      elif isinstance(lvdata, basestring):
1032
        feedback_fn("  - ERROR: LVM problem on node %s: %s" %
1033
                    (node, utils.SafeEncode(lvdata)))
1034
        bad = True
1035
        node_volume[node] = {}
1036
      elif not isinstance(lvdata, dict):
1037
        feedback_fn("  - ERROR: connection to %s failed (lvlist)" % (node,))
1038
        bad = True
1039
        continue
1040
      else:
1041
        node_volume[node] = lvdata
1042

    
1043
      # node_instance
1044
      idata = nresult.get(constants.NV_INSTANCELIST, None)
1045
      if not isinstance(idata, list):
1046
        feedback_fn("  - ERROR: connection to %s failed (instancelist)" %
1047
                    (node,))
1048
        bad = True
1049
        continue
1050

    
1051
      node_instance[node] = idata
1052

    
1053
      # node_info
1054
      nodeinfo = nresult.get(constants.NV_HVINFO, None)
1055
      if not isinstance(nodeinfo, dict):
1056
        feedback_fn("  - ERROR: connection to %s failed (hvinfo)" % (node,))
1057
        bad = True
1058
        continue
1059

    
1060
      try:
1061
        node_info[node] = {
1062
          "mfree": int(nodeinfo['memory_free']),
1063
          "pinst": [],
1064
          "sinst": [],
1065
          # dictionary holding all instances this node is secondary for,
1066
          # grouped by their primary node. Each key is a cluster node, and each
1067
          # value is a list of instances which have the key as primary and the
1068
          # current node as secondary.  this is handy to calculate N+1 memory
1069
          # availability if you can only failover from a primary to its
1070
          # secondary.
1071
          "sinst-by-pnode": {},
1072
        }
1073
        # FIXME: devise a free space model for file based instances as well
1074
        if vg_name is not None:
1075
          if (constants.NV_VGLIST not in nresult or
1076
              vg_name not in nresult[constants.NV_VGLIST]):
1077
            feedback_fn("  - ERROR: node %s didn't return data for the"
1078
                        " volume group '%s' - it is either missing or broken" %
1079
                        (node, vg_name))
1080
            bad = True
1081
            continue
1082
          node_info[node]["dfree"] = int(nresult[constants.NV_VGLIST][vg_name])
1083
      except (ValueError, KeyError):
1084
        feedback_fn("  - ERROR: invalid nodeinfo value returned"
1085
                    " from node %s" % (node,))
1086
        bad = True
1087
        continue
1088

    
1089
    node_vol_should = {}
1090

    
1091
    for instance in instancelist:
1092
      feedback_fn("* Verifying instance %s" % instance)
1093
      inst_config = instanceinfo[instance]
1094
      result =  self._VerifyInstance(instance, inst_config, node_volume,
1095
                                     node_instance, feedback_fn, n_offline)
1096
      bad = bad or result
1097
      inst_nodes_offline = []
1098

    
1099
      inst_config.MapLVsByNode(node_vol_should)
1100

    
1101
      instance_cfg[instance] = inst_config
1102

    
1103
      pnode = inst_config.primary_node
1104
      if pnode in node_info:
1105
        node_info[pnode]['pinst'].append(instance)
1106
      elif pnode not in n_offline:
1107
        feedback_fn("  - ERROR: instance %s, connection to primary node"
1108
                    " %s failed" % (instance, pnode))
1109
        bad = True
1110

    
1111
      if pnode in n_offline:
1112
        inst_nodes_offline.append(pnode)
1113

    
1114
      # If the instance is non-redundant we cannot survive losing its primary
1115
      # node, so we are not N+1 compliant. On the other hand we have no disk
1116
      # templates with more than one secondary so that situation is not well
1117
      # supported either.
1118
      # FIXME: does not support file-backed instances
1119
      if len(inst_config.secondary_nodes) == 0:
1120
        i_non_redundant.append(instance)
1121
      elif len(inst_config.secondary_nodes) > 1:
1122
        feedback_fn("  - WARNING: multiple secondaries for instance %s"
1123
                    % instance)
1124

    
1125
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
1126
        i_non_a_balanced.append(instance)
1127

    
1128
      for snode in inst_config.secondary_nodes:
1129
        if snode in node_info:
1130
          node_info[snode]['sinst'].append(instance)
1131
          if pnode not in node_info[snode]['sinst-by-pnode']:
1132
            node_info[snode]['sinst-by-pnode'][pnode] = []
1133
          node_info[snode]['sinst-by-pnode'][pnode].append(instance)
1134
        elif snode not in n_offline:
1135
          feedback_fn("  - ERROR: instance %s, connection to secondary node"
1136
                      " %s failed" % (instance, snode))
1137
          bad = True
1138
        if snode in n_offline:
1139
          inst_nodes_offline.append(snode)
1140

    
1141
      if inst_nodes_offline:
1142
        # warn that the instance lives on offline nodes, and set bad=True
1143
        feedback_fn("  - ERROR: instance lives on offline node(s) %s" %
1144
                    ", ".join(inst_nodes_offline))
1145
        bad = True
1146

    
1147
    feedback_fn("* Verifying orphan volumes")
1148
    result = self._VerifyOrphanVolumes(node_vol_should, node_volume,
1149
                                       feedback_fn)
1150
    bad = bad or result
1151

    
1152
    feedback_fn("* Verifying remaining instances")
1153
    result = self._VerifyOrphanInstances(instancelist, node_instance,
1154
                                         feedback_fn)
1155
    bad = bad or result
1156

    
1157
    if constants.VERIFY_NPLUSONE_MEM not in self.skip_set:
1158
      feedback_fn("* Verifying N+1 Memory redundancy")
1159
      result = self._VerifyNPlusOneMemory(node_info, instance_cfg, feedback_fn)
1160
      bad = bad or result
1161

    
1162
    feedback_fn("* Other Notes")
1163
    if i_non_redundant:
1164
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
1165
                  % len(i_non_redundant))
1166

    
1167
    if i_non_a_balanced:
1168
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
1169
                  % len(i_non_a_balanced))
1170

    
1171
    if n_offline:
1172
      feedback_fn("  - NOTICE: %d offline node(s) found." % len(n_offline))
1173

    
1174
    if n_drained:
1175
      feedback_fn("  - NOTICE: %d drained node(s) found." % len(n_drained))
1176

    
1177
    return not bad
1178

    
1179
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
1180
    """Analize the post-hooks' result
1181

1182
    This method analyses the hook result, handles it, and sends some
1183
    nicely-formatted feedback back to the user.
1184

1185
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
1186
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
1187
    @param hooks_results: the results of the multi-node hooks rpc call
1188
    @param feedback_fn: function used send feedback back to the caller
1189
    @param lu_result: previous Exec result
1190
    @return: the new Exec result, based on the previous result
1191
        and hook results
1192

1193
    """
1194
    # We only really run POST phase hooks, and are only interested in
1195
    # their results
1196
    if phase == constants.HOOKS_PHASE_POST:
1197
      # Used to change hooks' output to proper indentation
1198
      indent_re = re.compile('^', re.M)
1199
      feedback_fn("* Hooks Results")
1200
      if not hooks_results:
1201
        feedback_fn("  - ERROR: general communication failure")
1202
        lu_result = 1
1203
      else:
1204
        for node_name in hooks_results:
1205
          show_node_header = True
1206
          res = hooks_results[node_name]
1207
          if res.failed or res.data is False or not isinstance(res.data, list):
1208
            if res.offline:
1209
              # no need to warn or set fail return value
1210
              continue
1211
            feedback_fn("    Communication failure in hooks execution")
1212
            lu_result = 1
1213
            continue
1214
          for script, hkr, output in res.data:
1215
            if hkr == constants.HKR_FAIL:
1216
              # The node header is only shown once, if there are
1217
              # failing hooks on that node
1218
              if show_node_header:
1219
                feedback_fn("  Node %s:" % node_name)
1220
                show_node_header = False
1221
              feedback_fn("    ERROR: Script %s failed, output:" % script)
1222
              output = indent_re.sub('      ', output)
1223
              feedback_fn("%s" % output)
1224
              lu_result = 1
1225

    
1226
      return lu_result
1227

    
1228

    
1229
class LUVerifyDisks(NoHooksLU):
1230
  """Verifies the cluster disks status.
1231

1232
  """
1233
  _OP_REQP = []
1234
  REQ_BGL = False
1235

    
1236
  def ExpandNames(self):
1237
    self.needed_locks = {
1238
      locking.LEVEL_NODE: locking.ALL_SET,
1239
      locking.LEVEL_INSTANCE: locking.ALL_SET,
1240
    }
1241
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
1242

    
1243
  def CheckPrereq(self):
1244
    """Check prerequisites.
1245

1246
    This has no prerequisites.
1247

1248
    """
1249
    pass
1250

    
1251
  def Exec(self, feedback_fn):
1252
    """Verify integrity of cluster disks.
1253

1254
    """
1255
    result = res_nodes, res_nlvm, res_instances, res_missing = [], {}, [], {}
1256

    
1257
    vg_name = self.cfg.GetVGName()
1258
    nodes = utils.NiceSort(self.cfg.GetNodeList())
1259
    instances = [self.cfg.GetInstanceInfo(name)
1260
                 for name in self.cfg.GetInstanceList()]
1261

    
1262
    nv_dict = {}
1263
    for inst in instances:
1264
      inst_lvs = {}
1265
      if (not inst.admin_up or
1266
          inst.disk_template not in constants.DTS_NET_MIRROR):
1267
        continue
1268
      inst.MapLVsByNode(inst_lvs)
1269
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
1270
      for node, vol_list in inst_lvs.iteritems():
1271
        for vol in vol_list:
1272
          nv_dict[(node, vol)] = inst
1273

    
1274
    if not nv_dict:
1275
      return result
1276

    
1277
    node_lvs = self.rpc.call_volume_list(nodes, vg_name)
1278

    
1279
    to_act = set()
1280
    for node in nodes:
1281
      # node_volume
1282
      lvs = node_lvs[node]
1283
      if lvs.failed:
1284
        if not lvs.offline:
1285
          self.LogWarning("Connection to node %s failed: %s" %
1286
                          (node, lvs.data))
1287
        continue
1288
      lvs = lvs.data
1289
      if isinstance(lvs, basestring):
1290
        logging.warning("Error enumerating LVs on node %s: %s", node, lvs)
1291
        res_nlvm[node] = lvs
1292
        continue
1293
      elif not isinstance(lvs, dict):
1294
        logging.warning("Connection to node %s failed or invalid data"
1295
                        " returned", node)
1296
        res_nodes.append(node)
1297
        continue
1298

    
1299
      for lv_name, (_, lv_inactive, lv_online) in lvs.iteritems():
1300
        inst = nv_dict.pop((node, lv_name), None)
1301
        if (not lv_online and inst is not None
1302
            and inst.name not in res_instances):
1303
          res_instances.append(inst.name)
1304

    
1305
    # any leftover items in nv_dict are missing LVs, let's arrange the
1306
    # data better
1307
    for key, inst in nv_dict.iteritems():
1308
      if inst.name not in res_missing:
1309
        res_missing[inst.name] = []
1310
      res_missing[inst.name].append(key)
1311

    
1312
    return result
1313

    
1314

    
1315
class LURenameCluster(LogicalUnit):
1316
  """Rename the cluster.
1317

1318
  """
1319
  HPATH = "cluster-rename"
1320
  HTYPE = constants.HTYPE_CLUSTER
1321
  _OP_REQP = ["name"]
1322

    
1323
  def BuildHooksEnv(self):
1324
    """Build hooks env.
1325

1326
    """
1327
    env = {
1328
      "OP_TARGET": self.cfg.GetClusterName(),
1329
      "NEW_NAME": self.op.name,
1330
      }
1331
    mn = self.cfg.GetMasterNode()
1332
    return env, [mn], [mn]
1333

    
1334
  def CheckPrereq(self):
1335
    """Verify that the passed name is a valid one.
1336

1337
    """
1338
    hostname = utils.HostInfo(self.op.name)
1339

    
1340
    new_name = hostname.name
1341
    self.ip = new_ip = hostname.ip
1342
    old_name = self.cfg.GetClusterName()
1343
    old_ip = self.cfg.GetMasterIP()
1344
    if new_name == old_name and new_ip == old_ip:
1345
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
1346
                                 " cluster has changed")
1347
    if new_ip != old_ip:
1348
      if utils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
1349
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
1350
                                   " reachable on the network. Aborting." %
1351
                                   new_ip)
1352

    
1353
    self.op.name = new_name
1354

    
1355
  def Exec(self, feedback_fn):
1356
    """Rename the cluster.
1357

1358
    """
1359
    clustername = self.op.name
1360
    ip = self.ip
1361

    
1362
    # shutdown the master IP
1363
    master = self.cfg.GetMasterNode()
1364
    result = self.rpc.call_node_stop_master(master, False)
1365
    if result.failed or not result.data:
1366
      raise errors.OpExecError("Could not disable the master role")
1367

    
1368
    try:
1369
      cluster = self.cfg.GetClusterInfo()
1370
      cluster.cluster_name = clustername
1371
      cluster.master_ip = ip
1372
      self.cfg.Update(cluster)
1373

    
1374
      # update the known hosts file
1375
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
1376
      node_list = self.cfg.GetNodeList()
1377
      try:
1378
        node_list.remove(master)
1379
      except ValueError:
1380
        pass
1381
      result = self.rpc.call_upload_file(node_list,
1382
                                         constants.SSH_KNOWN_HOSTS_FILE)
1383
      for to_node, to_result in result.iteritems():
1384
         msg = to_result.RemoteFailMsg()
1385
         if msg:
1386
           msg = ("Copy of file %s to node %s failed: %s" %
1387
                   (constants.SSH_KNOWN_HOSTS_FILE, to_node, msg))
1388
           self.proc.LogWarning(msg)
1389

    
1390
    finally:
1391
      result = self.rpc.call_node_start_master(master, False)
1392
      if result.failed or not result.data:
1393
        self.LogWarning("Could not re-enable the master role on"
1394
                        " the master, please restart manually.")
1395

    
1396

    
1397
def _RecursiveCheckIfLVMBased(disk):
1398
  """Check if the given disk or its children are lvm-based.
1399

1400
  @type disk: L{objects.Disk}
1401
  @param disk: the disk to check
1402
  @rtype: booleean
1403
  @return: boolean indicating whether a LD_LV dev_type was found or not
1404

1405
  """
1406
  if disk.children:
1407
    for chdisk in disk.children:
1408
      if _RecursiveCheckIfLVMBased(chdisk):
1409
        return True
1410
  return disk.dev_type == constants.LD_LV
1411

    
1412

    
1413
class LUSetClusterParams(LogicalUnit):
1414
  """Change the parameters of the cluster.
1415

1416
  """
1417
  HPATH = "cluster-modify"
1418
  HTYPE = constants.HTYPE_CLUSTER
1419
  _OP_REQP = []
1420
  REQ_BGL = False
1421

    
1422
  def CheckArguments(self):
1423
    """Check parameters
1424

1425
    """
1426
    if not hasattr(self.op, "candidate_pool_size"):
1427
      self.op.candidate_pool_size = None
1428
    if self.op.candidate_pool_size is not None:
1429
      try:
1430
        self.op.candidate_pool_size = int(self.op.candidate_pool_size)
1431
      except (ValueError, TypeError), err:
1432
        raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
1433
                                   str(err))
1434
      if self.op.candidate_pool_size < 1:
1435
        raise errors.OpPrereqError("At least one master candidate needed")
1436

    
1437
  def ExpandNames(self):
1438
    # FIXME: in the future maybe other cluster params won't require checking on
1439
    # all nodes to be modified.
1440
    self.needed_locks = {
1441
      locking.LEVEL_NODE: locking.ALL_SET,
1442
    }
1443
    self.share_locks[locking.LEVEL_NODE] = 1
1444

    
1445
  def BuildHooksEnv(self):
1446
    """Build hooks env.
1447

1448
    """
1449
    env = {
1450
      "OP_TARGET": self.cfg.GetClusterName(),
1451
      "NEW_VG_NAME": self.op.vg_name,
1452
      }
1453
    mn = self.cfg.GetMasterNode()
1454
    return env, [mn], [mn]
1455

    
1456
  def CheckPrereq(self):
1457
    """Check prerequisites.
1458

1459
    This checks whether the given params don't conflict and
1460
    if the given volume group is valid.
1461

1462
    """
1463
    if self.op.vg_name is not None and not self.op.vg_name:
1464
      instances = self.cfg.GetAllInstancesInfo().values()
1465
      for inst in instances:
1466
        for disk in inst.disks:
1467
          if _RecursiveCheckIfLVMBased(disk):
1468
            raise errors.OpPrereqError("Cannot disable lvm storage while"
1469
                                       " lvm-based instances exist")
1470

    
1471
    node_list = self.acquired_locks[locking.LEVEL_NODE]
1472

    
1473
    # if vg_name not None, checks given volume group on all nodes
1474
    if self.op.vg_name:
1475
      vglist = self.rpc.call_vg_list(node_list)
1476
      for node in node_list:
1477
        if vglist[node].failed:
1478
          # ignoring down node
1479
          self.LogWarning("Node %s unreachable/error, ignoring" % node)
1480
          continue
1481
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].data,
1482
                                              self.op.vg_name,
1483
                                              constants.MIN_VG_SIZE)
1484
        if vgstatus:
1485
          raise errors.OpPrereqError("Error on node '%s': %s" %
1486
                                     (node, vgstatus))
1487

    
1488
    self.cluster = cluster = self.cfg.GetClusterInfo()
1489
    # validate params changes
1490
    if self.op.beparams:
1491
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
1492
      self.new_beparams = objects.FillDict(
1493
        cluster.beparams[constants.PP_DEFAULT], self.op.beparams)
1494

    
1495
    if self.op.nicparams:
1496
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
1497
      self.new_nicparams = objects.FillDict(
1498
        cluster.nicparams[constants.PP_DEFAULT], self.op.nicparams)
1499
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
1500

    
1501
    # hypervisor list/parameters
1502
    self.new_hvparams = objects.FillDict(cluster.hvparams, {})
1503
    if self.op.hvparams:
1504
      if not isinstance(self.op.hvparams, dict):
1505
        raise errors.OpPrereqError("Invalid 'hvparams' parameter on input")
1506
      for hv_name, hv_dict in self.op.hvparams.items():
1507
        if hv_name not in self.new_hvparams:
1508
          self.new_hvparams[hv_name] = hv_dict
1509
        else:
1510
          self.new_hvparams[hv_name].update(hv_dict)
1511

    
1512
    if self.op.enabled_hypervisors is not None:
1513
      self.hv_list = self.op.enabled_hypervisors
1514
    else:
1515
      self.hv_list = cluster.enabled_hypervisors
1516

    
1517
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
1518
      # either the enabled list has changed, or the parameters have, validate
1519
      for hv_name, hv_params in self.new_hvparams.items():
1520
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
1521
            (self.op.enabled_hypervisors and
1522
             hv_name in self.op.enabled_hypervisors)):
1523
          # either this is a new hypervisor, or its parameters have changed
1524
          hv_class = hypervisor.GetHypervisor(hv_name)
1525
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1526
          hv_class.CheckParameterSyntax(hv_params)
1527
          _CheckHVParams(self, node_list, hv_name, hv_params)
1528

    
1529
  def Exec(self, feedback_fn):
1530
    """Change the parameters of the cluster.
1531

1532
    """
1533
    if self.op.vg_name is not None:
1534
      new_volume = self.op.vg_name
1535
      if not new_volume:
1536
        new_volume = None
1537
      if new_volume != self.cfg.GetVGName():
1538
        self.cfg.SetVGName(new_volume)
1539
      else:
1540
        feedback_fn("Cluster LVM configuration already in desired"
1541
                    " state, not changing")
1542
    if self.op.hvparams:
1543
      self.cluster.hvparams = self.new_hvparams
1544
    if self.op.enabled_hypervisors is not None:
1545
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
1546
    if self.op.beparams:
1547
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
1548
    if self.op.nicparams:
1549
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
1550

    
1551
    if self.op.candidate_pool_size is not None:
1552
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
1553

    
1554
    self.cfg.Update(self.cluster)
1555

    
1556
    # we want to update nodes after the cluster so that if any errors
1557
    # happen, we have recorded and saved the cluster info
1558
    if self.op.candidate_pool_size is not None:
1559
      _AdjustCandidatePool(self)
1560

    
1561

    
1562
def _RedistributeAncillaryFiles(lu, additional_nodes=None):
1563
  """Distribute additional files which are part of the cluster configuration.
1564

1565
  ConfigWriter takes care of distributing the config and ssconf files, but
1566
  there are more files which should be distributed to all nodes. This function
1567
  makes sure those are copied.
1568

1569
  @param lu: calling logical unit
1570
  @param additional_nodes: list of nodes not in the config to distribute to
1571

1572
  """
1573
  # 1. Gather target nodes
1574
  myself = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
1575
  dist_nodes = lu.cfg.GetNodeList()
1576
  if additional_nodes is not None:
1577
    dist_nodes.extend(additional_nodes)
1578
  if myself.name in dist_nodes:
1579
    dist_nodes.remove(myself.name)
1580
  # 2. Gather files to distribute
1581
  dist_files = set([constants.ETC_HOSTS,
1582
                    constants.SSH_KNOWN_HOSTS_FILE,
1583
                    constants.RAPI_CERT_FILE,
1584
                    constants.RAPI_USERS_FILE,
1585
                   ])
1586

    
1587
  enabled_hypervisors = lu.cfg.GetClusterInfo().enabled_hypervisors
1588
  for hv_name in enabled_hypervisors:
1589
    hv_class = hypervisor.GetHypervisor(hv_name)
1590
    dist_files.update(hv_class.GetAncillaryFiles())
1591

    
1592
  # 3. Perform the files upload
1593
  for fname in dist_files:
1594
    if os.path.exists(fname):
1595
      result = lu.rpc.call_upload_file(dist_nodes, fname)
1596
      for to_node, to_result in result.items():
1597
         msg = to_result.RemoteFailMsg()
1598
         if msg:
1599
           msg = ("Copy of file %s to node %s failed: %s" %
1600
                   (fname, to_node, msg))
1601
           lu.proc.LogWarning(msg)
1602

    
1603

    
1604
class LURedistributeConfig(NoHooksLU):
1605
  """Force the redistribution of cluster configuration.
1606

1607
  This is a very simple LU.
1608

1609
  """
1610
  _OP_REQP = []
1611
  REQ_BGL = False
1612

    
1613
  def ExpandNames(self):
1614
    self.needed_locks = {
1615
      locking.LEVEL_NODE: locking.ALL_SET,
1616
    }
1617
    self.share_locks[locking.LEVEL_NODE] = 1
1618

    
1619
  def CheckPrereq(self):
1620
    """Check prerequisites.
1621

1622
    """
1623

    
1624
  def Exec(self, feedback_fn):
1625
    """Redistribute the configuration.
1626

1627
    """
1628
    self.cfg.Update(self.cfg.GetClusterInfo())
1629
    _RedistributeAncillaryFiles(self)
1630

    
1631

    
1632
def _WaitForSync(lu, instance, oneshot=False, unlock=False):
1633
  """Sleep and poll for an instance's disk to sync.
1634

1635
  """
1636
  if not instance.disks:
1637
    return True
1638

    
1639
  if not oneshot:
1640
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
1641

    
1642
  node = instance.primary_node
1643

    
1644
  for dev in instance.disks:
1645
    lu.cfg.SetDiskID(dev, node)
1646

    
1647
  retries = 0
1648
  while True:
1649
    max_time = 0
1650
    done = True
1651
    cumul_degraded = False
1652
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
1653
    if rstats.failed or not rstats.data:
1654
      lu.LogWarning("Can't get any data from node %s", node)
1655
      retries += 1
1656
      if retries >= 10:
1657
        raise errors.RemoteError("Can't contact node %s for mirror data,"
1658
                                 " aborting." % node)
1659
      time.sleep(6)
1660
      continue
1661
    rstats = rstats.data
1662
    retries = 0
1663
    for i, mstat in enumerate(rstats):
1664
      if mstat is None:
1665
        lu.LogWarning("Can't compute data for node %s/%s",
1666
                           node, instance.disks[i].iv_name)
1667
        continue
1668
      # we ignore the ldisk parameter
1669
      perc_done, est_time, is_degraded, _ = mstat
1670
      cumul_degraded = cumul_degraded or (is_degraded and perc_done is None)
1671
      if perc_done is not None:
1672
        done = False
1673
        if est_time is not None:
1674
          rem_time = "%d estimated seconds remaining" % est_time
1675
          max_time = est_time
1676
        else:
1677
          rem_time = "no time estimate"
1678
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
1679
                        (instance.disks[i].iv_name, perc_done, rem_time))
1680
    if done or oneshot:
1681
      break
1682

    
1683
    time.sleep(min(60, max_time))
1684

    
1685
  if done:
1686
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
1687
  return not cumul_degraded
1688

    
1689

    
1690
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
1691
  """Check that mirrors are not degraded.
1692

1693
  The ldisk parameter, if True, will change the test from the
1694
  is_degraded attribute (which represents overall non-ok status for
1695
  the device(s)) to the ldisk (representing the local storage status).
1696

1697
  """
1698
  lu.cfg.SetDiskID(dev, node)
1699
  if ldisk:
1700
    idx = 6
1701
  else:
1702
    idx = 5
1703

    
1704
  result = True
1705
  if on_primary or dev.AssembleOnSecondary():
1706
    rstats = lu.rpc.call_blockdev_find(node, dev)
1707
    msg = rstats.RemoteFailMsg()
1708
    if msg:
1709
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
1710
      result = False
1711
    elif not rstats.payload:
1712
      lu.LogWarning("Can't find disk on node %s", node)
1713
      result = False
1714
    else:
1715
      result = result and (not rstats.payload[idx])
1716
  if dev.children:
1717
    for child in dev.children:
1718
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
1719

    
1720
  return result
1721

    
1722

    
1723
class LUDiagnoseOS(NoHooksLU):
1724
  """Logical unit for OS diagnose/query.
1725

1726
  """
1727
  _OP_REQP = ["output_fields", "names"]
1728
  REQ_BGL = False
1729
  _FIELDS_STATIC = utils.FieldSet()
1730
  _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status")
1731

    
1732
  def ExpandNames(self):
1733
    if self.op.names:
1734
      raise errors.OpPrereqError("Selective OS query not supported")
1735

    
1736
    _CheckOutputFields(static=self._FIELDS_STATIC,
1737
                       dynamic=self._FIELDS_DYNAMIC,
1738
                       selected=self.op.output_fields)
1739

    
1740
    # Lock all nodes, in shared mode
1741
    # Temporary removal of locks, should be reverted later
1742
    # TODO: reintroduce locks when they are lighter-weight
1743
    self.needed_locks = {}
1744
    #self.share_locks[locking.LEVEL_NODE] = 1
1745
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
1746

    
1747
  def CheckPrereq(self):
1748
    """Check prerequisites.
1749

1750
    """
1751

    
1752
  @staticmethod
1753
  def _DiagnoseByOS(node_list, rlist):
1754
    """Remaps a per-node return list into an a per-os per-node dictionary
1755

1756
    @param node_list: a list with the names of all nodes
1757
    @param rlist: a map with node names as keys and OS objects as values
1758

1759
    @rtype: dict
1760
    @return: a dictionary with osnames as keys and as value another map, with
1761
        nodes as keys and list of OS objects as values, eg::
1762

1763
          {"debian-etch": {"node1": [<object>,...],
1764
                           "node2": [<object>,]}
1765
          }
1766

1767
    """
1768
    all_os = {}
1769
    # we build here the list of nodes that didn't fail the RPC (at RPC
1770
    # level), so that nodes with a non-responding node daemon don't
1771
    # make all OSes invalid
1772
    good_nodes = [node_name for node_name in rlist
1773
                  if not rlist[node_name].failed]
1774
    for node_name, nr in rlist.iteritems():
1775
      if nr.failed or not nr.data:
1776
        continue
1777
      for os_obj in nr.data:
1778
        if os_obj.name not in all_os:
1779
          # build a list of nodes for this os containing empty lists
1780
          # for each node in node_list
1781
          all_os[os_obj.name] = {}
1782
          for nname in good_nodes:
1783
            all_os[os_obj.name][nname] = []
1784
        all_os[os_obj.name][node_name].append(os_obj)
1785
    return all_os
1786

    
1787
  def Exec(self, feedback_fn):
1788
    """Compute the list of OSes.
1789

1790
    """
1791
    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
1792
    node_data = self.rpc.call_os_diagnose(valid_nodes)
1793
    if node_data == False:
1794
      raise errors.OpExecError("Can't gather the list of OSes")
1795
    pol = self._DiagnoseByOS(valid_nodes, node_data)
1796
    output = []
1797
    for os_name, os_data in pol.iteritems():
1798
      row = []
1799
      for field in self.op.output_fields:
1800
        if field == "name":
1801
          val = os_name
1802
        elif field == "valid":
1803
          val = utils.all([osl and osl[0] for osl in os_data.values()])
1804
        elif field == "node_status":
1805
          val = {}
1806
          for node_name, nos_list in os_data.iteritems():
1807
            val[node_name] = [(v.status, v.path) for v in nos_list]
1808
        else:
1809
          raise errors.ParameterError(field)
1810
        row.append(val)
1811
      output.append(row)
1812

    
1813
    return output
1814

    
1815

    
1816
class LURemoveNode(LogicalUnit):
1817
  """Logical unit for removing a node.
1818

1819
  """
1820
  HPATH = "node-remove"
1821
  HTYPE = constants.HTYPE_NODE
1822
  _OP_REQP = ["node_name"]
1823

    
1824
  def BuildHooksEnv(self):
1825
    """Build hooks env.
1826

1827
    This doesn't run on the target node in the pre phase as a failed
1828
    node would then be impossible to remove.
1829

1830
    """
1831
    env = {
1832
      "OP_TARGET": self.op.node_name,
1833
      "NODE_NAME": self.op.node_name,
1834
      }
1835
    all_nodes = self.cfg.GetNodeList()
1836
    all_nodes.remove(self.op.node_name)
1837
    return env, all_nodes, all_nodes
1838

    
1839
  def CheckPrereq(self):
1840
    """Check prerequisites.
1841

1842
    This checks:
1843
     - the node exists in the configuration
1844
     - it does not have primary or secondary instances
1845
     - it's not the master
1846

1847
    Any errors are signalled by raising errors.OpPrereqError.
1848

1849
    """
1850
    node = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.node_name))
1851
    if node is None:
1852
      raise errors.OpPrereqError, ("Node '%s' is unknown." % self.op.node_name)
1853

    
1854
    instance_list = self.cfg.GetInstanceList()
1855

    
1856
    masternode = self.cfg.GetMasterNode()
1857
    if node.name == masternode:
1858
      raise errors.OpPrereqError("Node is the master node,"
1859
                                 " you need to failover first.")
1860

    
1861
    for instance_name in instance_list:
1862
      instance = self.cfg.GetInstanceInfo(instance_name)
1863
      if node.name in instance.all_nodes:
1864
        raise errors.OpPrereqError("Instance %s is still running on the node,"
1865
                                   " please remove first." % instance_name)
1866
    self.op.node_name = node.name
1867
    self.node = node
1868

    
1869
  def Exec(self, feedback_fn):
1870
    """Removes the node from the cluster.
1871

1872
    """
1873
    node = self.node
1874
    logging.info("Stopping the node daemon and removing configs from node %s",
1875
                 node.name)
1876

    
1877
    self.context.RemoveNode(node.name)
1878

    
1879
    self.rpc.call_node_leave_cluster(node.name)
1880

    
1881
    # Promote nodes to master candidate as needed
1882
    _AdjustCandidatePool(self)
1883

    
1884

    
1885
class LUQueryNodes(NoHooksLU):
1886
  """Logical unit for querying nodes.
1887

1888
  """
1889
  _OP_REQP = ["output_fields", "names", "use_locking"]
1890
  REQ_BGL = False
1891
  _FIELDS_DYNAMIC = utils.FieldSet(
1892
    "dtotal", "dfree",
1893
    "mtotal", "mnode", "mfree",
1894
    "bootid",
1895
    "ctotal", "cnodes", "csockets",
1896
    )
1897

    
1898
  _FIELDS_STATIC = utils.FieldSet(
1899
    "name", "pinst_cnt", "sinst_cnt",
1900
    "pinst_list", "sinst_list",
1901
    "pip", "sip", "tags",
1902
    "serial_no",
1903
    "master_candidate",
1904
    "master",
1905
    "offline",
1906
    "drained",
1907
    )
1908

    
1909
  def ExpandNames(self):
1910
    _CheckOutputFields(static=self._FIELDS_STATIC,
1911
                       dynamic=self._FIELDS_DYNAMIC,
1912
                       selected=self.op.output_fields)
1913

    
1914
    self.needed_locks = {}
1915
    self.share_locks[locking.LEVEL_NODE] = 1
1916

    
1917
    if self.op.names:
1918
      self.wanted = _GetWantedNodes(self, self.op.names)
1919
    else:
1920
      self.wanted = locking.ALL_SET
1921

    
1922
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
1923
    self.do_locking = self.do_node_query and self.op.use_locking
1924
    if self.do_locking:
1925
      # if we don't request only static fields, we need to lock the nodes
1926
      self.needed_locks[locking.LEVEL_NODE] = self.wanted
1927

    
1928

    
1929
  def CheckPrereq(self):
1930
    """Check prerequisites.
1931

1932
    """
1933
    # The validation of the node list is done in the _GetWantedNodes,
1934
    # if non empty, and if empty, there's no validation to do
1935
    pass
1936

    
1937
  def Exec(self, feedback_fn):
1938
    """Computes the list of nodes and their attributes.
1939

1940
    """
1941
    all_info = self.cfg.GetAllNodesInfo()
1942
    if self.do_locking:
1943
      nodenames = self.acquired_locks[locking.LEVEL_NODE]
1944
    elif self.wanted != locking.ALL_SET:
1945
      nodenames = self.wanted
1946
      missing = set(nodenames).difference(all_info.keys())
1947
      if missing:
1948
        raise errors.OpExecError(
1949
          "Some nodes were removed before retrieving their data: %s" % missing)
1950
    else:
1951
      nodenames = all_info.keys()
1952

    
1953
    nodenames = utils.NiceSort(nodenames)
1954
    nodelist = [all_info[name] for name in nodenames]
1955

    
1956
    # begin data gathering
1957

    
1958
    if self.do_node_query:
1959
      live_data = {}
1960
      node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
1961
                                          self.cfg.GetHypervisorType())
1962
      for name in nodenames:
1963
        nodeinfo = node_data[name]
1964
        if not nodeinfo.failed and nodeinfo.data:
1965
          nodeinfo = nodeinfo.data
1966
          fn = utils.TryConvert
1967
          live_data[name] = {
1968
            "mtotal": fn(int, nodeinfo.get('memory_total', None)),
1969
            "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
1970
            "mfree": fn(int, nodeinfo.get('memory_free', None)),
1971
            "dtotal": fn(int, nodeinfo.get('vg_size', None)),
1972
            "dfree": fn(int, nodeinfo.get('vg_free', None)),
1973
            "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
1974
            "bootid": nodeinfo.get('bootid', None),
1975
            "cnodes": fn(int, nodeinfo.get('cpu_nodes', None)),
1976
            "csockets": fn(int, nodeinfo.get('cpu_sockets', None)),
1977
            }
1978
        else:
1979
          live_data[name] = {}
1980
    else:
1981
      live_data = dict.fromkeys(nodenames, {})
1982

    
1983
    node_to_primary = dict([(name, set()) for name in nodenames])
1984
    node_to_secondary = dict([(name, set()) for name in nodenames])
1985

    
1986
    inst_fields = frozenset(("pinst_cnt", "pinst_list",
1987
                             "sinst_cnt", "sinst_list"))
1988
    if inst_fields & frozenset(self.op.output_fields):
1989
      instancelist = self.cfg.GetInstanceList()
1990

    
1991
      for instance_name in instancelist:
1992
        inst = self.cfg.GetInstanceInfo(instance_name)
1993
        if inst.primary_node in node_to_primary:
1994
          node_to_primary[inst.primary_node].add(inst.name)
1995
        for secnode in inst.secondary_nodes:
1996
          if secnode in node_to_secondary:
1997
            node_to_secondary[secnode].add(inst.name)
1998

    
1999
    master_node = self.cfg.GetMasterNode()
2000

    
2001
    # end data gathering
2002

    
2003
    output = []
2004
    for node in nodelist:
2005
      node_output = []
2006
      for field in self.op.output_fields:
2007
        if field == "name":
2008
          val = node.name
2009
        elif field == "pinst_list":
2010
          val = list(node_to_primary[node.name])
2011
        elif field == "sinst_list":
2012
          val = list(node_to_secondary[node.name])
2013
        elif field == "pinst_cnt":
2014
          val = len(node_to_primary[node.name])
2015
        elif field == "sinst_cnt":
2016
          val = len(node_to_secondary[node.name])
2017
        elif field == "pip":
2018
          val = node.primary_ip
2019
        elif field == "sip":
2020
          val = node.secondary_ip
2021
        elif field == "tags":
2022
          val = list(node.GetTags())
2023
        elif field == "serial_no":
2024
          val = node.serial_no
2025
        elif field == "master_candidate":
2026
          val = node.master_candidate
2027
        elif field == "master":
2028
          val = node.name == master_node
2029
        elif field == "offline":
2030
          val = node.offline
2031
        elif field == "drained":
2032
          val = node.drained
2033
        elif self._FIELDS_DYNAMIC.Matches(field):
2034
          val = live_data[node.name].get(field, None)
2035
        else:
2036
          raise errors.ParameterError(field)
2037
        node_output.append(val)
2038
      output.append(node_output)
2039

    
2040
    return output
2041

    
2042

    
2043
class LUQueryNodeVolumes(NoHooksLU):
2044
  """Logical unit for getting volumes on node(s).
2045

2046
  """
2047
  _OP_REQP = ["nodes", "output_fields"]
2048
  REQ_BGL = False
2049
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
2050
  _FIELDS_STATIC = utils.FieldSet("node")
2051

    
2052
  def ExpandNames(self):
2053
    _CheckOutputFields(static=self._FIELDS_STATIC,
2054
                       dynamic=self._FIELDS_DYNAMIC,
2055
                       selected=self.op.output_fields)
2056

    
2057
    self.needed_locks = {}
2058
    self.share_locks[locking.LEVEL_NODE] = 1
2059
    if not self.op.nodes:
2060
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
2061
    else:
2062
      self.needed_locks[locking.LEVEL_NODE] = \
2063
        _GetWantedNodes(self, self.op.nodes)
2064

    
2065
  def CheckPrereq(self):
2066
    """Check prerequisites.
2067

2068
    This checks that the fields required are valid output fields.
2069

2070
    """
2071
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
2072

    
2073
  def Exec(self, feedback_fn):
2074
    """Computes the list of nodes and their attributes.
2075

2076
    """
2077
    nodenames = self.nodes
2078
    volumes = self.rpc.call_node_volumes(nodenames)
2079

    
2080
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
2081
             in self.cfg.GetInstanceList()]
2082

    
2083
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
2084

    
2085
    output = []
2086
    for node in nodenames:
2087
      if node not in volumes or volumes[node].failed or not volumes[node].data:
2088
        continue
2089

    
2090
      node_vols = volumes[node].data[:]
2091
      node_vols.sort(key=lambda vol: vol['dev'])
2092

    
2093
      for vol in node_vols:
2094
        node_output = []
2095
        for field in self.op.output_fields:
2096
          if field == "node":
2097
            val = node
2098
          elif field == "phys":
2099
            val = vol['dev']
2100
          elif field == "vg":
2101
            val = vol['vg']
2102
          elif field == "name":
2103
            val = vol['name']
2104
          elif field == "size":
2105
            val = int(float(vol['size']))
2106
          elif field == "instance":
2107
            for inst in ilist:
2108
              if node not in lv_by_node[inst]:
2109
                continue
2110
              if vol['name'] in lv_by_node[inst][node]:
2111
                val = inst.name
2112
                break
2113
            else:
2114
              val = '-'
2115
          else:
2116
            raise errors.ParameterError(field)
2117
          node_output.append(str(val))
2118

    
2119
        output.append(node_output)
2120

    
2121
    return output
2122

    
2123

    
2124
class LUAddNode(LogicalUnit):
2125
  """Logical unit for adding node to the cluster.
2126

2127
  """
2128
  HPATH = "node-add"
2129
  HTYPE = constants.HTYPE_NODE
2130
  _OP_REQP = ["node_name"]
2131

    
2132
  def BuildHooksEnv(self):
2133
    """Build hooks env.
2134

2135
    This will run on all nodes before, and on all nodes + the new node after.
2136

2137
    """
2138
    env = {
2139
      "OP_TARGET": self.op.node_name,
2140
      "NODE_NAME": self.op.node_name,
2141
      "NODE_PIP": self.op.primary_ip,
2142
      "NODE_SIP": self.op.secondary_ip,
2143
      }
2144
    nodes_0 = self.cfg.GetNodeList()
2145
    nodes_1 = nodes_0 + [self.op.node_name, ]
2146
    return env, nodes_0, nodes_1
2147

    
2148
  def CheckPrereq(self):
2149
    """Check prerequisites.
2150

2151
    This checks:
2152
     - the new node is not already in the config
2153
     - it is resolvable
2154
     - its parameters (single/dual homed) matches the cluster
2155

2156
    Any errors are signalled by raising errors.OpPrereqError.
2157

2158
    """
2159
    node_name = self.op.node_name
2160
    cfg = self.cfg
2161

    
2162
    dns_data = utils.HostInfo(node_name)
2163

    
2164
    node = dns_data.name
2165
    primary_ip = self.op.primary_ip = dns_data.ip
2166
    secondary_ip = getattr(self.op, "secondary_ip", None)
2167
    if secondary_ip is None:
2168
      secondary_ip = primary_ip
2169
    if not utils.IsValidIP(secondary_ip):
2170
      raise errors.OpPrereqError("Invalid secondary IP given")
2171
    self.op.secondary_ip = secondary_ip
2172

    
2173
    node_list = cfg.GetNodeList()
2174
    if not self.op.readd and node in node_list:
2175
      raise errors.OpPrereqError("Node %s is already in the configuration" %
2176
                                 node)
2177
    elif self.op.readd and node not in node_list:
2178
      raise errors.OpPrereqError("Node %s is not in the configuration" % node)
2179

    
2180
    for existing_node_name in node_list:
2181
      existing_node = cfg.GetNodeInfo(existing_node_name)
2182

    
2183
      if self.op.readd and node == existing_node_name:
2184
        if (existing_node.primary_ip != primary_ip or
2185
            existing_node.secondary_ip != secondary_ip):
2186
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
2187
                                     " address configuration as before")
2188
        continue
2189

    
2190
      if (existing_node.primary_ip == primary_ip or
2191
          existing_node.secondary_ip == primary_ip or
2192
          existing_node.primary_ip == secondary_ip or
2193
          existing_node.secondary_ip == secondary_ip):
2194
        raise errors.OpPrereqError("New node ip address(es) conflict with"
2195
                                   " existing node %s" % existing_node.name)
2196

    
2197
    # check that the type of the node (single versus dual homed) is the
2198
    # same as for the master
2199
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
2200
    master_singlehomed = myself.secondary_ip == myself.primary_ip
2201
    newbie_singlehomed = secondary_ip == primary_ip
2202
    if master_singlehomed != newbie_singlehomed:
2203
      if master_singlehomed:
2204
        raise errors.OpPrereqError("The master has no private ip but the"
2205
                                   " new node has one")
2206
      else:
2207
        raise errors.OpPrereqError("The master has a private ip but the"
2208
                                   " new node doesn't have one")
2209

    
2210
    # checks reachablity
2211
    if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
2212
      raise errors.OpPrereqError("Node not reachable by ping")
2213

    
2214
    if not newbie_singlehomed:
2215
      # check reachability from my secondary ip to newbie's secondary ip
2216
      if not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
2217
                           source=myself.secondary_ip):
2218
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
2219
                                   " based ping to noded port")
2220

    
2221
    cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2222
    mc_now, _ = self.cfg.GetMasterCandidateStats()
2223
    master_candidate = mc_now < cp_size
2224

    
2225
    self.new_node = objects.Node(name=node,
2226
                                 primary_ip=primary_ip,
2227
                                 secondary_ip=secondary_ip,
2228
                                 master_candidate=master_candidate,
2229
                                 offline=False, drained=False)
2230

    
2231
  def Exec(self, feedback_fn):
2232
    """Adds the new node to the cluster.
2233

2234
    """
2235
    new_node = self.new_node
2236
    node = new_node.name
2237

    
2238
    # check connectivity
2239
    result = self.rpc.call_version([node])[node]
2240
    result.Raise()
2241
    if result.data:
2242
      if constants.PROTOCOL_VERSION == result.data:
2243
        logging.info("Communication to node %s fine, sw version %s match",
2244
                     node, result.data)
2245
      else:
2246
        raise errors.OpExecError("Version mismatch master version %s,"
2247
                                 " node version %s" %
2248
                                 (constants.PROTOCOL_VERSION, result.data))
2249
    else:
2250
      raise errors.OpExecError("Cannot get version from the new node")
2251

    
2252
    # setup ssh on node
2253
    logging.info("Copy ssh key to node %s", node)
2254
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
2255
    keyarray = []
2256
    keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
2257
                constants.SSH_HOST_RSA_PRIV, constants.SSH_HOST_RSA_PUB,
2258
                priv_key, pub_key]
2259

    
2260
    for i in keyfiles:
2261
      f = open(i, 'r')
2262
      try:
2263
        keyarray.append(f.read())
2264
      finally:
2265
        f.close()
2266

    
2267
    result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
2268
                                    keyarray[2],
2269
                                    keyarray[3], keyarray[4], keyarray[5])
2270

    
2271
    msg = result.RemoteFailMsg()
2272
    if msg:
2273
      raise errors.OpExecError("Cannot transfer ssh keys to the"
2274
                               " new node: %s" % msg)
2275

    
2276
    # Add node to our /etc/hosts, and add key to known_hosts
2277
    if self.cfg.GetClusterInfo().modify_etc_hosts:
2278
      utils.AddHostToEtcHosts(new_node.name)
2279

    
2280
    if new_node.secondary_ip != new_node.primary_ip:
2281
      result = self.rpc.call_node_has_ip_address(new_node.name,
2282
                                                 new_node.secondary_ip)
2283
      if result.failed or not result.data:
2284
        raise errors.OpExecError("Node claims it doesn't have the secondary ip"
2285
                                 " you gave (%s). Please fix and re-run this"
2286
                                 " command." % new_node.secondary_ip)
2287

    
2288
    node_verify_list = [self.cfg.GetMasterNode()]
2289
    node_verify_param = {
2290
      'nodelist': [node],
2291
      # TODO: do a node-net-test as well?
2292
    }
2293

    
2294
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
2295
                                       self.cfg.GetClusterName())
2296
    for verifier in node_verify_list:
2297
      if result[verifier].failed or not result[verifier].data:
2298
        raise errors.OpExecError("Cannot communicate with %s's node daemon"
2299
                                 " for remote verification" % verifier)
2300
      if result[verifier].data['nodelist']:
2301
        for failed in result[verifier].data['nodelist']:
2302
          feedback_fn("ssh/hostname verification failed %s -> %s" %
2303
                      (verifier, result[verifier].data['nodelist'][failed]))
2304
        raise errors.OpExecError("ssh/hostname verification failed.")
2305

    
2306
    if self.op.readd:
2307
      _RedistributeAncillaryFiles(self)
2308
      self.context.ReaddNode(new_node)
2309
    else:
2310
      _RedistributeAncillaryFiles(self, additional_nodes=[node])
2311
      self.context.AddNode(new_node)
2312

    
2313

    
2314
class LUSetNodeParams(LogicalUnit):
2315
  """Modifies the parameters of a node.
2316

2317
  """
2318
  HPATH = "node-modify"
2319
  HTYPE = constants.HTYPE_NODE
2320
  _OP_REQP = ["node_name"]
2321
  REQ_BGL = False
2322

    
2323
  def CheckArguments(self):
2324
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2325
    if node_name is None:
2326
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2327
    self.op.node_name = node_name
2328
    _CheckBooleanOpField(self.op, 'master_candidate')
2329
    _CheckBooleanOpField(self.op, 'offline')
2330
    _CheckBooleanOpField(self.op, 'drained')
2331
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained]
2332
    if all_mods.count(None) == 3:
2333
      raise errors.OpPrereqError("Please pass at least one modification")
2334
    if all_mods.count(True) > 1:
2335
      raise errors.OpPrereqError("Can't set the node into more than one"
2336
                                 " state at the same time")
2337

    
2338
  def ExpandNames(self):
2339
    self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
2340

    
2341
  def BuildHooksEnv(self):
2342
    """Build hooks env.
2343

2344
    This runs on the master node.
2345

2346
    """
2347
    env = {
2348
      "OP_TARGET": self.op.node_name,
2349
      "MASTER_CANDIDATE": str(self.op.master_candidate),
2350
      "OFFLINE": str(self.op.offline),
2351
      "DRAINED": str(self.op.drained),
2352
      }
2353
    nl = [self.cfg.GetMasterNode(),
2354
          self.op.node_name]
2355
    return env, nl, nl
2356

    
2357
  def CheckPrereq(self):
2358
    """Check prerequisites.
2359

2360
    This only checks the instance list against the existing names.
2361

2362
    """
2363
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
2364

    
2365
    if ((self.op.master_candidate == False or self.op.offline == True or
2366
         self.op.drained == True) and node.master_candidate):
2367
      # we will demote the node from master_candidate
2368
      if self.op.node_name == self.cfg.GetMasterNode():
2369
        raise errors.OpPrereqError("The master node has to be a"
2370
                                   " master candidate, online and not drained")
2371
      cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2372
      num_candidates, _ = self.cfg.GetMasterCandidateStats()
2373
      if num_candidates <= cp_size:
2374
        msg = ("Not enough master candidates (desired"
2375
               " %d, new value will be %d)" % (cp_size, num_candidates-1))
2376
        if self.op.force:
2377
          self.LogWarning(msg)
2378
        else:
2379
          raise errors.OpPrereqError(msg)
2380

    
2381
    if (self.op.master_candidate == True and
2382
        ((node.offline and not self.op.offline == False) or
2383
         (node.drained and not self.op.drained == False))):
2384
      raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
2385
                                 " to master_candidate" % node.name)
2386

    
2387
    return
2388

    
2389
  def Exec(self, feedback_fn):
2390
    """Modifies a node.
2391

2392
    """
2393
    node = self.node
2394

    
2395
    result = []
2396
    changed_mc = False
2397

    
2398
    if self.op.offline is not None:
2399
      node.offline = self.op.offline
2400
      result.append(("offline", str(self.op.offline)))
2401
      if self.op.offline == True:
2402
        if node.master_candidate:
2403
          node.master_candidate = False
2404
          changed_mc = True
2405
          result.append(("master_candidate", "auto-demotion due to offline"))
2406
        if node.drained:
2407
          node.drained = False
2408
          result.append(("drained", "clear drained status due to offline"))
2409

    
2410
    if self.op.master_candidate is not None:
2411
      node.master_candidate = self.op.master_candidate
2412
      changed_mc = True
2413
      result.append(("master_candidate", str(self.op.master_candidate)))
2414
      if self.op.master_candidate == False:
2415
        rrc = self.rpc.call_node_demote_from_mc(node.name)
2416
        msg = rrc.RemoteFailMsg()
2417
        if msg:
2418
          self.LogWarning("Node failed to demote itself: %s" % msg)
2419

    
2420
    if self.op.drained is not None:
2421
      node.drained = self.op.drained
2422
      result.append(("drained", str(self.op.drained)))
2423
      if self.op.drained == True:
2424
        if node.master_candidate:
2425
          node.master_candidate = False
2426
          changed_mc = True
2427
          result.append(("master_candidate", "auto-demotion due to drain"))
2428
        if node.offline:
2429
          node.offline = False
2430
          result.append(("offline", "clear offline status due to drain"))
2431

    
2432
    # this will trigger configuration file update, if needed
2433
    self.cfg.Update(node)
2434
    # this will trigger job queue propagation or cleanup
2435
    if changed_mc:
2436
      self.context.ReaddNode(node)
2437

    
2438
    return result
2439

    
2440

    
2441
class LUPowercycleNode(NoHooksLU):
2442
  """Powercycles a node.
2443

2444
  """
2445
  _OP_REQP = ["node_name", "force"]
2446
  REQ_BGL = False
2447

    
2448
  def CheckArguments(self):
2449
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2450
    if node_name is None:
2451
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2452
    self.op.node_name = node_name
2453
    if node_name == self.cfg.GetMasterNode() and not self.op.force:
2454
      raise errors.OpPrereqError("The node is the master and the force"
2455
                                 " parameter was not set")
2456

    
2457
  def ExpandNames(self):
2458
    """Locking for PowercycleNode.
2459

2460
    This is a last-resource option and shouldn't block on other
2461
    jobs. Therefore, we grab no locks.
2462

2463
    """
2464
    self.needed_locks = {}
2465

    
2466
  def CheckPrereq(self):
2467
    """Check prerequisites.
2468

2469
    This LU has no prereqs.
2470

2471
    """
2472
    pass
2473

    
2474
  def Exec(self, feedback_fn):
2475
    """Reboots a node.
2476

2477
    """
2478
    result = self.rpc.call_node_powercycle(self.op.node_name,
2479
                                           self.cfg.GetHypervisorType())
2480
    msg = result.RemoteFailMsg()
2481
    if msg:
2482
      raise errors.OpExecError("Failed to schedule the reboot: %s" % msg)
2483
    return result.payload
2484

    
2485

    
2486
class LUQueryClusterInfo(NoHooksLU):
2487
  """Query cluster configuration.
2488

2489
  """
2490
  _OP_REQP = []
2491
  REQ_BGL = False
2492

    
2493
  def ExpandNames(self):
2494
    self.needed_locks = {}
2495

    
2496
  def CheckPrereq(self):
2497
    """No prerequsites needed for this LU.
2498

2499
    """
2500
    pass
2501

    
2502
  def Exec(self, feedback_fn):
2503
    """Return cluster config.
2504

2505
    """
2506
    cluster = self.cfg.GetClusterInfo()
2507
    result = {
2508
      "software_version": constants.RELEASE_VERSION,
2509
      "protocol_version": constants.PROTOCOL_VERSION,
2510
      "config_version": constants.CONFIG_VERSION,
2511
      "os_api_version": constants.OS_API_VERSION,
2512
      "export_version": constants.EXPORT_VERSION,
2513
      "architecture": (platform.architecture()[0], platform.machine()),
2514
      "name": cluster.cluster_name,
2515
      "master": cluster.master_node,
2516
      "default_hypervisor": cluster.default_hypervisor,
2517
      "enabled_hypervisors": cluster.enabled_hypervisors,
2518
      "hvparams": dict([(hypervisor, cluster.hvparams[hypervisor])
2519
                        for hypervisor in cluster.enabled_hypervisors]),
2520
      "beparams": cluster.beparams,
2521
      "nicparams": cluster.nicparams,
2522
      "candidate_pool_size": cluster.candidate_pool_size,
2523
      "default_bridge": cluster.default_bridge,
2524
      "master_netdev": cluster.master_netdev,
2525
      "volume_group_name": cluster.volume_group_name,
2526
      "file_storage_dir": cluster.file_storage_dir,
2527
      }
2528

    
2529
    return result
2530

    
2531

    
2532
class LUQueryConfigValues(NoHooksLU):
2533
  """Return configuration values.
2534

2535
  """
2536
  _OP_REQP = []
2537
  REQ_BGL = False
2538
  _FIELDS_DYNAMIC = utils.FieldSet()
2539
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag")
2540

    
2541
  def ExpandNames(self):
2542
    self.needed_locks = {}
2543

    
2544
    _CheckOutputFields(static=self._FIELDS_STATIC,
2545
                       dynamic=self._FIELDS_DYNAMIC,
2546
                       selected=self.op.output_fields)
2547

    
2548
  def CheckPrereq(self):
2549
    """No prerequisites.
2550

2551
    """
2552
    pass
2553

    
2554
  def Exec(self, feedback_fn):
2555
    """Dump a representation of the cluster config to the standard output.
2556

2557
    """
2558
    values = []
2559
    for field in self.op.output_fields:
2560
      if field == "cluster_name":
2561
        entry = self.cfg.GetClusterName()
2562
      elif field == "master_node":
2563
        entry = self.cfg.GetMasterNode()
2564
      elif field == "drain_flag":
2565
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
2566
      else:
2567
        raise errors.ParameterError(field)
2568
      values.append(entry)
2569
    return values
2570

    
2571

    
2572
class LUActivateInstanceDisks(NoHooksLU):
2573
  """Bring up an instance's disks.
2574

2575
  """
2576
  _OP_REQP = ["instance_name"]
2577
  REQ_BGL = False
2578

    
2579
  def ExpandNames(self):
2580
    self._ExpandAndLockInstance()
2581
    self.needed_locks[locking.LEVEL_NODE] = []
2582
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2583

    
2584
  def DeclareLocks(self, level):
2585
    if level == locking.LEVEL_NODE:
2586
      self._LockInstancesNodes()
2587

    
2588
  def CheckPrereq(self):
2589
    """Check prerequisites.
2590

2591
    This checks that the instance is in the cluster.
2592

2593
    """
2594
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2595
    assert self.instance is not None, \
2596
      "Cannot retrieve locked instance %s" % self.op.instance_name
2597
    _CheckNodeOnline(self, self.instance.primary_node)
2598

    
2599
  def Exec(self, feedback_fn):
2600
    """Activate the disks.
2601

2602
    """
2603
    disks_ok, disks_info = _AssembleInstanceDisks(self, self.instance)
2604
    if not disks_ok:
2605
      raise errors.OpExecError("Cannot activate block devices")
2606

    
2607
    return disks_info
2608

    
2609

    
2610
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False):
2611
  """Prepare the block devices for an instance.
2612

2613
  This sets up the block devices on all nodes.
2614

2615
  @type lu: L{LogicalUnit}
2616
  @param lu: the logical unit on whose behalf we execute
2617
  @type instance: L{objects.Instance}
2618
  @param instance: the instance for whose disks we assemble
2619
  @type ignore_secondaries: boolean
2620
  @param ignore_secondaries: if true, errors on secondary nodes
2621
      won't result in an error return from the function
2622
  @return: False if the operation failed, otherwise a list of
2623
      (host, instance_visible_name, node_visible_name)
2624
      with the mapping from node devices to instance devices
2625

2626
  """
2627
  device_info = []
2628
  disks_ok = True
2629
  iname = instance.name
2630
  # With the two passes mechanism we try to reduce the window of
2631
  # opportunity for the race condition of switching DRBD to primary
2632
  # before handshaking occured, but we do not eliminate it
2633

    
2634
  # The proper fix would be to wait (with some limits) until the
2635
  # connection has been made and drbd transitions from WFConnection
2636
  # into any other network-connected state (Connected, SyncTarget,
2637
  # SyncSource, etc.)
2638

    
2639
  # 1st pass, assemble on all nodes in secondary mode
2640
  for inst_disk in instance.disks:
2641
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2642
      lu.cfg.SetDiskID(node_disk, node)
2643
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
2644
      msg = result.RemoteFailMsg()
2645
      if msg:
2646
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2647
                           " (is_primary=False, pass=1): %s",
2648
                           inst_disk.iv_name, node, msg)
2649
        if not ignore_secondaries:
2650
          disks_ok = False
2651

    
2652
  # FIXME: race condition on drbd migration to primary
2653

    
2654
  # 2nd pass, do only the primary node
2655
  for inst_disk in instance.disks:
2656
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2657
      if node != instance.primary_node:
2658
        continue
2659
      lu.cfg.SetDiskID(node_disk, node)
2660
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
2661
      msg = result.RemoteFailMsg()
2662
      if msg:
2663
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2664
                           " (is_primary=True, pass=2): %s",
2665
                           inst_disk.iv_name, node, msg)
2666
        disks_ok = False
2667
    device_info.append((instance.primary_node, inst_disk.iv_name,
2668
                        result.payload))
2669

    
2670
  # leave the disks configured for the primary node
2671
  # this is a workaround that would be fixed better by
2672
  # improving the logical/physical id handling
2673
  for disk in instance.disks:
2674
    lu.cfg.SetDiskID(disk, instance.primary_node)
2675

    
2676
  return disks_ok, device_info
2677

    
2678

    
2679
def _StartInstanceDisks(lu, instance, force):
2680
  """Start the disks of an instance.
2681

2682
  """
2683
  disks_ok, dummy = _AssembleInstanceDisks(lu, instance,
2684
                                           ignore_secondaries=force)
2685
  if not disks_ok:
2686
    _ShutdownInstanceDisks(lu, instance)
2687
    if force is not None and not force:
2688
      lu.proc.LogWarning("", hint="If the message above refers to a"
2689
                         " secondary node,"
2690
                         " you can retry the operation using '--force'.")
2691
    raise errors.OpExecError("Disk consistency error")
2692

    
2693

    
2694
class LUDeactivateInstanceDisks(NoHooksLU):
2695
  """Shutdown an instance's disks.
2696

2697
  """
2698
  _OP_REQP = ["instance_name"]
2699
  REQ_BGL = False
2700

    
2701
  def ExpandNames(self):
2702
    self._ExpandAndLockInstance()
2703
    self.needed_locks[locking.LEVEL_NODE] = []
2704
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2705

    
2706
  def DeclareLocks(self, level):
2707
    if level == locking.LEVEL_NODE:
2708
      self._LockInstancesNodes()
2709

    
2710
  def CheckPrereq(self):
2711
    """Check prerequisites.
2712

2713
    This checks that the instance is in the cluster.
2714

2715
    """
2716
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2717
    assert self.instance is not None, \
2718
      "Cannot retrieve locked instance %s" % self.op.instance_name
2719

    
2720
  def Exec(self, feedback_fn):
2721
    """Deactivate the disks
2722

2723
    """
2724
    instance = self.instance
2725
    _SafeShutdownInstanceDisks(self, instance)
2726

    
2727

    
2728
def _SafeShutdownInstanceDisks(lu, instance):
2729
  """Shutdown block devices of an instance.
2730

2731
  This function checks if an instance is running, before calling
2732
  _ShutdownInstanceDisks.
2733

2734
  """
2735
  ins_l = lu.rpc.call_instance_list([instance.primary_node],
2736
                                      [instance.hypervisor])
2737
  ins_l = ins_l[instance.primary_node]
2738
  if ins_l.failed or not isinstance(ins_l.data, list):
2739
    raise errors.OpExecError("Can't contact node '%s'" %
2740
                             instance.primary_node)
2741

    
2742
  if instance.name in ins_l.data:
2743
    raise errors.OpExecError("Instance is running, can't shutdown"
2744
                             " block devices.")
2745

    
2746
  _ShutdownInstanceDisks(lu, instance)
2747

    
2748

    
2749
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
2750
  """Shutdown block devices of an instance.
2751

2752
  This does the shutdown on all nodes of the instance.
2753

2754
  If the ignore_primary is false, errors on the primary node are
2755
  ignored.
2756

2757
  """
2758
  all_result = True
2759
  for disk in instance.disks:
2760
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
2761
      lu.cfg.SetDiskID(top_disk, node)
2762
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
2763
      msg = result.RemoteFailMsg()
2764
      if msg:
2765
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
2766
                      disk.iv_name, node, msg)
2767
        if not ignore_primary or node != instance.primary_node:
2768
          all_result = False
2769
  return all_result
2770

    
2771

    
2772
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
2773
  """Checks if a node has enough free memory.
2774

2775
  This function check if a given node has the needed amount of free
2776
  memory. In case the node has less memory or we cannot get the
2777
  information from the node, this function raise an OpPrereqError
2778
  exception.
2779

2780
  @type lu: C{LogicalUnit}
2781
  @param lu: a logical unit from which we get configuration data
2782
  @type node: C{str}
2783
  @param node: the node to check
2784
  @type reason: C{str}
2785
  @param reason: string to use in the error message
2786
  @type requested: C{int}
2787
  @param requested: the amount of memory in MiB to check for
2788
  @type hypervisor_name: C{str}
2789
  @param hypervisor_name: the hypervisor to ask for memory stats
2790
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
2791
      we cannot check the node
2792

2793
  """
2794
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
2795
  nodeinfo[node].Raise()
2796
  free_mem = nodeinfo[node].data.get('memory_free')
2797
  if not isinstance(free_mem, int):
2798
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
2799
                             " was '%s'" % (node, free_mem))
2800
  if requested > free_mem:
2801
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
2802
                             " needed %s MiB, available %s MiB" %
2803
                             (node, reason, requested, free_mem))
2804

    
2805

    
2806
class LUStartupInstance(LogicalUnit):
2807
  """Starts an instance.
2808

2809
  """
2810
  HPATH = "instance-start"
2811
  HTYPE = constants.HTYPE_INSTANCE
2812
  _OP_REQP = ["instance_name", "force"]
2813
  REQ_BGL = False
2814

    
2815
  def ExpandNames(self):
2816
    self._ExpandAndLockInstance()
2817

    
2818
  def BuildHooksEnv(self):
2819
    """Build hooks env.
2820

2821
    This runs on master, primary and secondary nodes of the instance.
2822

2823
    """
2824
    env = {
2825
      "FORCE": self.op.force,
2826
      }
2827
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2828
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2829
    return env, nl, nl
2830

    
2831
  def CheckPrereq(self):
2832
    """Check prerequisites.
2833

2834
    This checks that the instance is in the cluster.
2835

2836
    """
2837
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2838
    assert self.instance is not None, \
2839
      "Cannot retrieve locked instance %s" % self.op.instance_name
2840

    
2841
    # extra beparams
2842
    self.beparams = getattr(self.op, "beparams", {})
2843
    if self.beparams:
2844
      if not isinstance(self.beparams, dict):
2845
        raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
2846
                                   " dict" % (type(self.beparams), ))
2847
      # fill the beparams dict
2848
      utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
2849
      self.op.beparams = self.beparams
2850

    
2851
    # extra hvparams
2852
    self.hvparams = getattr(self.op, "hvparams", {})
2853
    if self.hvparams:
2854
      if not isinstance(self.hvparams, dict):
2855
        raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
2856
                                   " dict" % (type(self.hvparams), ))
2857

    
2858
      # check hypervisor parameter syntax (locally)
2859
      cluster = self.cfg.GetClusterInfo()
2860
      utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
2861
      filled_hvp = objects.FillDict(cluster.hvparams[instance.hypervisor],
2862
                                    instance.hvparams)
2863
      filled_hvp.update(self.hvparams)
2864
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
2865
      hv_type.CheckParameterSyntax(filled_hvp)
2866
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
2867
      self.op.hvparams = self.hvparams
2868

    
2869
    _CheckNodeOnline(self, instance.primary_node)
2870

    
2871
    bep = self.cfg.GetClusterInfo().FillBE(instance)
2872
    # check bridges existance
2873
    _CheckInstanceBridgesExist(self, instance)
2874

    
2875
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2876
                                              instance.name,
2877
                                              instance.hypervisor)
2878
    remote_info.Raise()
2879
    if not remote_info.data:
2880
      _CheckNodeFreeMemory(self, instance.primary_node,
2881
                           "starting instance %s" % instance.name,
2882
                           bep[constants.BE_MEMORY], instance.hypervisor)
2883

    
2884
  def Exec(self, feedback_fn):
2885
    """Start the instance.
2886

2887
    """
2888
    instance = self.instance
2889
    force = self.op.force
2890

    
2891
    self.cfg.MarkInstanceUp(instance.name)
2892

    
2893
    node_current = instance.primary_node
2894

    
2895
    _StartInstanceDisks(self, instance, force)
2896

    
2897
    result = self.rpc.call_instance_start(node_current, instance,
2898
                                          self.hvparams, self.beparams)
2899
    msg = result.RemoteFailMsg()
2900
    if msg:
2901
      _ShutdownInstanceDisks(self, instance)
2902
      raise errors.OpExecError("Could not start instance: %s" % msg)
2903

    
2904

    
2905
class LURebootInstance(LogicalUnit):
2906
  """Reboot an instance.
2907

2908
  """
2909
  HPATH = "instance-reboot"
2910
  HTYPE = constants.HTYPE_INSTANCE
2911
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
2912
  REQ_BGL = False
2913

    
2914
  def ExpandNames(self):
2915
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
2916
                                   constants.INSTANCE_REBOOT_HARD,
2917
                                   constants.INSTANCE_REBOOT_FULL]:
2918
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
2919
                                  (constants.INSTANCE_REBOOT_SOFT,
2920
                                   constants.INSTANCE_REBOOT_HARD,
2921
                                   constants.INSTANCE_REBOOT_FULL))
2922
    self._ExpandAndLockInstance()
2923

    
2924
  def BuildHooksEnv(self):
2925
    """Build hooks env.
2926

2927
    This runs on master, primary and secondary nodes of the instance.
2928

2929
    """
2930
    env = {
2931
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
2932
      "REBOOT_TYPE": self.op.reboot_type,
2933
      }
2934
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2935
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2936
    return env, nl, nl
2937

    
2938
  def CheckPrereq(self):
2939
    """Check prerequisites.
2940

2941
    This checks that the instance is in the cluster.
2942

2943
    """
2944
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2945
    assert self.instance is not None, \
2946
      "Cannot retrieve locked instance %s" % self.op.instance_name
2947

    
2948
    _CheckNodeOnline(self, instance.primary_node)
2949

    
2950
    # check bridges existance
2951
    _CheckInstanceBridgesExist(self, instance)
2952

    
2953
  def Exec(self, feedback_fn):
2954
    """Reboot the instance.
2955

2956
    """
2957
    instance = self.instance
2958
    ignore_secondaries = self.op.ignore_secondaries
2959
    reboot_type = self.op.reboot_type
2960

    
2961
    node_current = instance.primary_node
2962

    
2963
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
2964
                       constants.INSTANCE_REBOOT_HARD]:
2965
      for disk in instance.disks:
2966
        self.cfg.SetDiskID(disk, node_current)
2967
      result = self.rpc.call_instance_reboot(node_current, instance,
2968
                                             reboot_type)
2969
      msg = result.RemoteFailMsg()
2970
      if msg:
2971
        raise errors.OpExecError("Could not reboot instance: %s" % msg)
2972
    else:
2973
      result = self.rpc.call_instance_shutdown(node_current, instance)
2974
      msg = result.RemoteFailMsg()
2975
      if msg:
2976
        raise errors.OpExecError("Could not shutdown instance for"
2977
                                 " full reboot: %s" % msg)
2978
      _ShutdownInstanceDisks(self, instance)
2979
      _StartInstanceDisks(self, instance, ignore_secondaries)
2980
      result = self.rpc.call_instance_start(node_current, instance, None, None)
2981
      msg = result.RemoteFailMsg()
2982
      if msg:
2983
        _ShutdownInstanceDisks(self, instance)
2984
        raise errors.OpExecError("Could not start instance for"
2985
                                 " full reboot: %s" % msg)
2986

    
2987
    self.cfg.MarkInstanceUp(instance.name)
2988

    
2989

    
2990
class LUShutdownInstance(LogicalUnit):
2991
  """Shutdown an instance.
2992

2993
  """
2994
  HPATH = "instance-stop"
2995
  HTYPE = constants.HTYPE_INSTANCE
2996
  _OP_REQP = ["instance_name"]
2997
  REQ_BGL = False
2998

    
2999
  def ExpandNames(self):
3000
    self._ExpandAndLockInstance()
3001

    
3002
  def BuildHooksEnv(self):
3003
    """Build hooks env.
3004

3005
    This runs on master, primary and secondary nodes of the instance.
3006

3007
    """
3008
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3009
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3010
    return env, nl, nl
3011

    
3012
  def CheckPrereq(self):
3013
    """Check prerequisites.
3014

3015
    This checks that the instance is in the cluster.
3016

3017
    """
3018
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3019
    assert self.instance is not None, \
3020
      "Cannot retrieve locked instance %s" % self.op.instance_name
3021
    _CheckNodeOnline(self, self.instance.primary_node)
3022

    
3023
  def Exec(self, feedback_fn):
3024
    """Shutdown the instance.
3025

3026
    """
3027
    instance = self.instance
3028
    node_current = instance.primary_node
3029
    self.cfg.MarkInstanceDown(instance.name)
3030
    result = self.rpc.call_instance_shutdown(node_current, instance)
3031
    msg = result.RemoteFailMsg()
3032
    if msg:
3033
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
3034

    
3035
    _ShutdownInstanceDisks(self, instance)
3036

    
3037

    
3038
class LUReinstallInstance(LogicalUnit):
3039
  """Reinstall an instance.
3040

3041
  """
3042
  HPATH = "instance-reinstall"
3043
  HTYPE = constants.HTYPE_INSTANCE
3044
  _OP_REQP = ["instance_name"]
3045
  REQ_BGL = False
3046

    
3047
  def ExpandNames(self):
3048
    self._ExpandAndLockInstance()
3049

    
3050
  def BuildHooksEnv(self):
3051
    """Build hooks env.
3052

3053
    This runs on master, primary and secondary nodes of the instance.
3054

3055
    """
3056
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3057
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3058
    return env, nl, nl
3059

    
3060
  def CheckPrereq(self):
3061
    """Check prerequisites.
3062

3063
    This checks that the instance is in the cluster and is not running.
3064

3065
    """
3066
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3067
    assert instance is not None, \
3068
      "Cannot retrieve locked instance %s" % self.op.instance_name
3069
    _CheckNodeOnline(self, instance.primary_node)
3070

    
3071
    if instance.disk_template == constants.DT_DISKLESS:
3072
      raise errors.OpPrereqError("Instance '%s' has no disks" %
3073
                                 self.op.instance_name)
3074
    if instance.admin_up:
3075
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3076
                                 self.op.instance_name)
3077
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3078
                                              instance.name,
3079
                                              instance.hypervisor)
3080
    remote_info.Raise()
3081
    if remote_info.data:
3082
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3083
                                 (self.op.instance_name,
3084
                                  instance.primary_node))
3085

    
3086
    self.op.os_type = getattr(self.op, "os_type", None)
3087
    if self.op.os_type is not None:
3088
      # OS verification
3089
      pnode = self.cfg.GetNodeInfo(
3090
        self.cfg.ExpandNodeName(instance.primary_node))
3091
      if pnode is None:
3092
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
3093
                                   self.op.pnode)
3094
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
3095
      result.Raise()
3096
      if not isinstance(result.data, objects.OS):
3097
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
3098
                                   " primary node"  % self.op.os_type)
3099

    
3100
    self.instance = instance
3101

    
3102
  def Exec(self, feedback_fn):
3103
    """Reinstall the instance.
3104

3105
    """
3106
    inst = self.instance
3107

    
3108
    if self.op.os_type is not None:
3109
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
3110
      inst.os = self.op.os_type
3111
      self.cfg.Update(inst)
3112

    
3113
    _StartInstanceDisks(self, inst, None)
3114
    try:
3115
      feedback_fn("Running the instance OS create scripts...")
3116
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True)
3117
      msg = result.RemoteFailMsg()
3118
      if msg:
3119
        raise errors.OpExecError("Could not install OS for instance %s"
3120
                                 " on node %s: %s" %
3121
                                 (inst.name, inst.primary_node, msg))
3122
    finally:
3123
      _ShutdownInstanceDisks(self, inst)
3124

    
3125

    
3126
class LURenameInstance(LogicalUnit):
3127
  """Rename an instance.
3128

3129
  """
3130
  HPATH = "instance-rename"
3131
  HTYPE = constants.HTYPE_INSTANCE
3132
  _OP_REQP = ["instance_name", "new_name"]
3133

    
3134
  def BuildHooksEnv(self):
3135
    """Build hooks env.
3136

3137
    This runs on master, primary and secondary nodes of the instance.
3138

3139
    """
3140
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3141
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3142
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3143
    return env, nl, nl
3144

    
3145
  def CheckPrereq(self):
3146
    """Check prerequisites.
3147

3148
    This checks that the instance is in the cluster and is not running.
3149

3150
    """
3151
    instance = self.cfg.GetInstanceInfo(
3152
      self.cfg.ExpandInstanceName(self.op.instance_name))
3153
    if instance is None:
3154
      raise errors.OpPrereqError("Instance '%s' not known" %
3155
                                 self.op.instance_name)
3156
    _CheckNodeOnline(self, instance.primary_node)
3157

    
3158
    if instance.admin_up:
3159
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3160
                                 self.op.instance_name)
3161
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3162
                                              instance.name,
3163
                                              instance.hypervisor)
3164
    remote_info.Raise()
3165
    if remote_info.data:
3166
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3167
                                 (self.op.instance_name,
3168
                                  instance.primary_node))
3169
    self.instance = instance
3170

    
3171
    # new name verification
3172
    name_info = utils.HostInfo(self.op.new_name)
3173

    
3174
    self.op.new_name = new_name = name_info.name
3175
    instance_list = self.cfg.GetInstanceList()
3176
    if new_name in instance_list:
3177
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3178
                                 new_name)
3179

    
3180
    if not getattr(self.op, "ignore_ip", False):
3181
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3182
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3183
                                   (name_info.ip, new_name))
3184

    
3185

    
3186
  def Exec(self, feedback_fn):
3187
    """Reinstall the instance.
3188

3189
    """
3190
    inst = self.instance
3191
    old_name = inst.name
3192

    
3193
    if inst.disk_template == constants.DT_FILE:
3194
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3195

    
3196
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3197
    # Change the instance lock. This is definitely safe while we hold the BGL
3198
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3199
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3200

    
3201
    # re-read the instance from the configuration after rename
3202
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3203

    
3204
    if inst.disk_template == constants.DT_FILE:
3205
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3206
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3207
                                                     old_file_storage_dir,
3208
                                                     new_file_storage_dir)
3209
      result.Raise()
3210
      if not result.data:
3211
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3212
                                 " directory '%s' to '%s' (but the instance"
3213
                                 " has been renamed in Ganeti)" % (
3214
                                 inst.primary_node, old_file_storage_dir,
3215
                                 new_file_storage_dir))
3216

    
3217
      if not result.data[0]:
3218
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3219
                                 " (but the instance has been renamed in"
3220
                                 " Ganeti)" % (old_file_storage_dir,
3221
                                               new_file_storage_dir))
3222

    
3223
    _StartInstanceDisks(self, inst, None)
3224
    try:
3225
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3226
                                                 old_name)
3227
      msg = result.RemoteFailMsg()
3228
      if msg:
3229
        msg = ("Could not run OS rename script for instance %s on node %s"
3230
               " (but the instance has been renamed in Ganeti): %s" %
3231
               (inst.name, inst.primary_node, msg))
3232
        self.proc.LogWarning(msg)
3233
    finally:
3234
      _ShutdownInstanceDisks(self, inst)
3235

    
3236

    
3237
class LURemoveInstance(LogicalUnit):
3238
  """Remove an instance.
3239

3240
  """
3241
  HPATH = "instance-remove"
3242
  HTYPE = constants.HTYPE_INSTANCE
3243
  _OP_REQP = ["instance_name", "ignore_failures"]
3244
  REQ_BGL = False
3245

    
3246
  def ExpandNames(self):
3247
    self._ExpandAndLockInstance()
3248
    self.needed_locks[locking.LEVEL_NODE] = []
3249
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3250

    
3251
  def DeclareLocks(self, level):
3252
    if level == locking.LEVEL_NODE:
3253
      self._LockInstancesNodes()
3254

    
3255
  def BuildHooksEnv(self):
3256
    """Build hooks env.
3257

3258
    This runs on master, primary and secondary nodes of the instance.
3259

3260
    """
3261
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3262
    nl = [self.cfg.GetMasterNode()]
3263
    return env, nl, nl
3264

    
3265
  def CheckPrereq(self):
3266
    """Check prerequisites.
3267

3268
    This checks that the instance is in the cluster.
3269

3270
    """
3271
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3272
    assert self.instance is not None, \
3273
      "Cannot retrieve locked instance %s" % self.op.instance_name
3274

    
3275
  def Exec(self, feedback_fn):
3276
    """Remove the instance.
3277

3278
    """
3279
    instance = self.instance
3280
    logging.info("Shutting down instance %s on node %s",
3281
                 instance.name, instance.primary_node)
3282

    
3283
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3284
    msg = result.RemoteFailMsg()
3285
    if msg:
3286
      if self.op.ignore_failures:
3287
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3288
      else:
3289
        raise errors.OpExecError("Could not shutdown instance %s on"
3290
                                 " node %s: %s" %
3291
                                 (instance.name, instance.primary_node, msg))
3292

    
3293
    logging.info("Removing block devices for instance %s", instance.name)
3294

    
3295
    if not _RemoveDisks(self, instance):
3296
      if self.op.ignore_failures:
3297
        feedback_fn("Warning: can't remove instance's disks")
3298
      else:
3299
        raise errors.OpExecError("Can't remove instance's disks")
3300

    
3301
    logging.info("Removing instance %s out of cluster config", instance.name)
3302

    
3303
    self.cfg.RemoveInstance(instance.name)
3304
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3305

    
3306

    
3307
class LUQueryInstances(NoHooksLU):
3308
  """Logical unit for querying instances.
3309

3310
  """
3311
  _OP_REQP = ["output_fields", "names", "use_locking"]
3312
  REQ_BGL = False
3313
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3314
                                    "admin_state",
3315
                                    "disk_template", "ip", "mac", "bridge",
3316
                                    "sda_size", "sdb_size", "vcpus", "tags",
3317
                                    "network_port", "beparams",
3318
                                    r"(disk)\.(size)/([0-9]+)",
3319
                                    r"(disk)\.(sizes)", "disk_usage",
3320
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3321
                                    r"(nic)\.(macs|ips|bridges)",
3322
                                    r"(disk|nic)\.(count)",
3323
                                    "serial_no", "hypervisor", "hvparams",] +
3324
                                  ["hv/%s" % name
3325
                                   for name in constants.HVS_PARAMETERS] +
3326
                                  ["be/%s" % name
3327
                                   for name in constants.BES_PARAMETERS])
3328
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3329

    
3330

    
3331
  def ExpandNames(self):
3332
    _CheckOutputFields(static=self._FIELDS_STATIC,
3333
                       dynamic=self._FIELDS_DYNAMIC,
3334
                       selected=self.op.output_fields)
3335

    
3336
    self.needed_locks = {}
3337
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3338
    self.share_locks[locking.LEVEL_NODE] = 1
3339

    
3340
    if self.op.names:
3341
      self.wanted = _GetWantedInstances(self, self.op.names)
3342
    else:
3343
      self.wanted = locking.ALL_SET
3344

    
3345
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3346
    self.do_locking = self.do_node_query and self.op.use_locking
3347
    if self.do_locking:
3348
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3349
      self.needed_locks[locking.LEVEL_NODE] = []
3350
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3351

    
3352
  def DeclareLocks(self, level):
3353
    if level == locking.LEVEL_NODE and self.do_locking:
3354
      self._LockInstancesNodes()
3355

    
3356
  def CheckPrereq(self):
3357
    """Check prerequisites.
3358

3359
    """
3360
    pass
3361

    
3362
  def Exec(self, feedback_fn):
3363
    """Computes the list of nodes and their attributes.
3364

3365
    """
3366
    all_info = self.cfg.GetAllInstancesInfo()
3367
    if self.wanted == locking.ALL_SET:
3368
      # caller didn't specify instance names, so ordering is not important
3369
      if self.do_locking:
3370
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3371
      else:
3372
        instance_names = all_info.keys()
3373
      instance_names = utils.NiceSort(instance_names)
3374
    else:
3375
      # caller did specify names, so we must keep the ordering
3376
      if self.do_locking:
3377
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3378
      else:
3379
        tgt_set = all_info.keys()
3380
      missing = set(self.wanted).difference(tgt_set)
3381
      if missing:
3382
        raise errors.OpExecError("Some instances were removed before"
3383
                                 " retrieving their data: %s" % missing)
3384
      instance_names = self.wanted
3385

    
3386
    instance_list = [all_info[iname] for iname in instance_names]
3387

    
3388
    # begin data gathering
3389

    
3390
    nodes = frozenset([inst.primary_node for inst in instance_list])
3391
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3392

    
3393
    bad_nodes = []
3394
    off_nodes = []
3395
    if self.do_node_query:
3396
      live_data = {}
3397
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3398
      for name in nodes:
3399
        result = node_data[name]
3400
        if result.offline:
3401
          # offline nodes will be in both lists
3402
          off_nodes.append(name)
3403
        if result.failed:
3404
          bad_nodes.append(name)
3405
        else:
3406
          if result.data:
3407
            live_data.update(result.data)
3408
            # else no instance is alive
3409
    else:
3410
      live_data = dict([(name, {}) for name in instance_names])
3411

    
3412
    # end data gathering
3413

    
3414
    HVPREFIX = "hv/"
3415
    BEPREFIX = "be/"
3416
    output = []
3417
    for instance in instance_list:
3418
      iout = []
3419
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3420
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3421
      for field in self.op.output_fields:
3422
        st_match = self._FIELDS_STATIC.Matches(field)
3423
        if field == "name":
3424
          val = instance.name
3425
        elif field == "os":
3426
          val = instance.os
3427
        elif field == "pnode":
3428
          val = instance.primary_node
3429
        elif field == "snodes":
3430
          val = list(instance.secondary_nodes)
3431
        elif field == "admin_state":
3432
          val = instance.admin_up
3433
        elif field == "oper_state":
3434
          if instance.primary_node in bad_nodes:
3435
            val = None
3436
          else:
3437
            val = bool(live_data.get(instance.name))
3438
        elif field == "status":
3439
          if instance.primary_node in off_nodes:
3440
            val = "ERROR_nodeoffline"
3441
          elif instance.primary_node in bad_nodes:
3442
            val = "ERROR_nodedown"
3443
          else:
3444
            running = bool(live_data.get(instance.name))
3445
            if running:
3446
              if instance.admin_up:
3447
                val = "running"
3448
              else:
3449
                val = "ERROR_up"
3450
            else:
3451
              if instance.admin_up:
3452
                val = "ERROR_down"
3453
              else:
3454
                val = "ADMIN_down"
3455
        elif field == "oper_ram":
3456
          if instance.primary_node in bad_nodes:
3457
            val = None
3458
          elif instance.name in live_data:
3459
            val = live_data[instance.name].get("memory", "?")
3460
          else:
3461
            val = "-"
3462
        elif field == "disk_template":
3463
          val = instance.disk_template
3464
        elif field == "ip":
3465
          val = instance.nics[0].ip
3466
        elif field == "bridge":
3467
          val = instance.nics[0].bridge
3468
        elif field == "mac":
3469
          val = instance.nics[0].mac
3470
        elif field == "sda_size" or field == "sdb_size":
3471
          idx = ord(field[2]) - ord('a')
3472
          try:
3473
            val = instance.FindDisk(idx).size
3474
          except errors.OpPrereqError:
3475
            val = None
3476
        elif field == "disk_usage": # total disk usage per node
3477
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3478
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3479
        elif field == "tags":
3480
          val = list(instance.GetTags())
3481
        elif field == "serial_no":
3482
          val = instance.serial_no
3483
        elif field == "network_port":
3484
          val = instance.network_port
3485
        elif field == "hypervisor":
3486
          val = instance.hypervisor
3487
        elif field == "hvparams":
3488
          val = i_hv
3489
        elif (field.startswith(HVPREFIX) and
3490
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3491
          val = i_hv.get(field[len(HVPREFIX):], None)
3492
        elif field == "beparams":
3493
          val = i_be
3494
        elif (field.startswith(BEPREFIX) and
3495
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3496
          val = i_be.get(field[len(BEPREFIX):], None)
3497
        elif st_match and st_match.groups():
3498
          # matches a variable list
3499
          st_groups = st_match.groups()
3500
          if st_groups and st_groups[0] == "disk":
3501
            if st_groups[1] == "count":
3502
              val = len(instance.disks)
3503
            elif st_groups[1] == "sizes":
3504
              val = [disk.size for disk in instance.disks]
3505
            elif st_groups[1] == "size":
3506
              try:
3507
                val = instance.FindDisk(st_groups[2]).size
3508
              except errors.OpPrereqError:
3509
                val = None
3510
            else:
3511
              assert False, "Unhandled disk parameter"
3512
          elif st_groups[0] == "nic":
3513
            if st_groups[1] == "count":
3514
              val = len(instance.nics)
3515
            elif st_groups[1] == "macs":
3516
              val = [nic.mac for nic in instance.nics]
3517
            elif st_groups[1] == "ips":
3518
              val = [nic.ip for nic in instance.nics]
3519
            elif st_groups[1] == "bridges":
3520
              val = [nic.bridge for nic in instance.nics]
3521
            else:
3522
              # index-based item
3523
              nic_idx = int(st_groups[2])
3524
              if nic_idx >= len(instance.nics):
3525
                val = None
3526
              else:
3527
                if st_groups[1] == "mac":
3528
                  val = instance.nics[nic_idx].mac
3529
                elif st_groups[1] == "ip":
3530
                  val = instance.nics[nic_idx].ip
3531
                elif st_groups[1] == "bridge":
3532
                  val = instance.nics[nic_idx].bridge
3533
                else:
3534
                  assert False, "Unhandled NIC parameter"
3535
          else:
3536
            assert False, "Unhandled variable parameter"
3537
        else:
3538
          raise errors.ParameterError(field)
3539
        iout.append(val)
3540
      output.append(iout)
3541

    
3542
    return output
3543

    
3544

    
3545
class LUFailoverInstance(LogicalUnit):
3546
  """Failover an instance.
3547

3548
  """
3549
  HPATH = "instance-failover"
3550
  HTYPE = constants.HTYPE_INSTANCE
3551
  _OP_REQP = ["instance_name", "ignore_consistency"]
3552
  REQ_BGL = False
3553

    
3554
  def ExpandNames(self):
3555
    self._ExpandAndLockInstance()
3556
    self.needed_locks[locking.LEVEL_NODE] = []
3557
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3558

    
3559
  def DeclareLocks(self, level):
3560
    if level == locking.LEVEL_NODE:
3561
      self._LockInstancesNodes()
3562

    
3563
  def BuildHooksEnv(self):
3564
    """Build hooks env.
3565

3566
    This runs on master, primary and secondary nodes of the instance.
3567

3568
    """
3569
    env = {
3570
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3571
      }
3572
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3573
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3574
    return env, nl, nl
3575

    
3576
  def CheckPrereq(self):
3577
    """Check prerequisites.
3578

3579
    This checks that the instance is in the cluster.
3580

3581
    """
3582
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3583
    assert self.instance is not None, \
3584
      "Cannot retrieve locked instance %s" % self.op.instance_name
3585

    
3586
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3587
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3588
      raise errors.OpPrereqError("Instance's disk layout is not"
3589
                                 " network mirrored, cannot failover.")
3590

    
3591
    secondary_nodes = instance.secondary_nodes
3592
    if not secondary_nodes:
3593
      raise errors.ProgrammerError("no secondary node but using "
3594
                                   "a mirrored disk template")
3595

    
3596
    target_node = secondary_nodes[0]
3597
    _CheckNodeOnline(self, target_node)
3598
    _CheckNodeNotDrained(self, target_node)
3599
    # check memory requirements on the secondary node
3600
    _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3601
                         instance.name, bep[constants.BE_MEMORY],
3602
                         instance.hypervisor)
3603

    
3604
    # check bridge existance
3605
    brlist = [nic.bridge for nic in instance.nics]
3606
    result = self.rpc.call_bridges_exist(target_node, brlist)
3607
    result.Raise()
3608
    if not result.data:
3609
      raise errors.OpPrereqError("One or more target bridges %s does not"
3610
                                 " exist on destination node '%s'" %
3611
                                 (brlist, target_node))
3612

    
3613
  def Exec(self, feedback_fn):
3614
    """Failover an instance.
3615

3616
    The failover is done by shutting it down on its present node and
3617
    starting it on the secondary.
3618

3619
    """
3620
    instance = self.instance
3621

    
3622
    source_node = instance.primary_node
3623
    target_node = instance.secondary_nodes[0]
3624

    
3625
    feedback_fn("* checking disk consistency between source and target")
3626
    for dev in instance.disks:
3627
      # for drbd, these are drbd over lvm
3628
      if not _CheckDiskConsistency(self, dev, target_node, False):
3629
        if instance.admin_up and not self.op.ignore_consistency:
3630
          raise errors.OpExecError("Disk %s is degraded on target node,"
3631
                                   " aborting failover." % dev.iv_name)
3632

    
3633
    feedback_fn("* shutting down instance on source node")
3634
    logging.info("Shutting down instance %s on node %s",
3635
                 instance.name, source_node)
3636

    
3637
    result = self.rpc.call_instance_shutdown(source_node, instance)
3638
    msg = result.RemoteFailMsg()
3639
    if msg:
3640
      if self.op.ignore_consistency:
3641
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3642
                             " Proceeding anyway. Please make sure node"
3643
                             " %s is down. Error details: %s",
3644
                             instance.name, source_node, source_node, msg)
3645
      else:
3646
        raise errors.OpExecError("Could not shutdown instance %s on"
3647
                                 " node %s: %s" %
3648
                                 (instance.name, source_node, msg))
3649

    
3650
    feedback_fn("* deactivating the instance's disks on source node")
3651
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3652
      raise errors.OpExecError("Can't shut down the instance's disks.")
3653

    
3654
    instance.primary_node = target_node
3655
    # distribute new instance config to the other nodes
3656
    self.cfg.Update(instance)
3657

    
3658
    # Only start the instance if it's marked as up
3659
    if instance.admin_up:
3660
      feedback_fn("* activating the instance's disks on target node")
3661
      logging.info("Starting instance %s on node %s",
3662
                   instance.name, target_node)
3663

    
3664
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3665
                                               ignore_secondaries=True)
3666
      if not disks_ok:
3667
        _ShutdownInstanceDisks(self, instance)
3668
        raise errors.OpExecError("Can't activate the instance's disks")
3669

    
3670
      feedback_fn("* starting the instance on the target node")
3671
      result = self.rpc.call_instance_start(target_node, instance, None, None)
3672
      msg = result.RemoteFailMsg()
3673
      if msg:
3674
        _ShutdownInstanceDisks(self, instance)
3675
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3676
                                 (instance.name, target_node, msg))
3677

    
3678

    
3679
class LUMigrateInstance(LogicalUnit):
3680
  """Migrate an instance.
3681

3682
  This is migration without shutting down, compared to the failover,
3683
  which is done with shutdown.
3684

3685
  """
3686
  HPATH = "instance-migrate"
3687
  HTYPE = constants.HTYPE_INSTANCE
3688
  _OP_REQP = ["instance_name", "live", "cleanup"]
3689

    
3690
  REQ_BGL = False
3691

    
3692
  def ExpandNames(self):
3693
    self._ExpandAndLockInstance()
3694
    self.needed_locks[locking.LEVEL_NODE] = []
3695
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3696

    
3697
  def DeclareLocks(self, level):
3698
    if level == locking.LEVEL_NODE:
3699
      self._LockInstancesNodes()
3700

    
3701
  def BuildHooksEnv(self):
3702
    """Build hooks env.
3703

3704
    This runs on master, primary and secondary nodes of the instance.
3705

3706
    """
3707
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3708
    env["MIGRATE_LIVE"] = self.op.live
3709
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3710
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3711
    return env, nl, nl
3712

    
3713
  def CheckPrereq(self):
3714
    """Check prerequisites.
3715

3716
    This checks that the instance is in the cluster.
3717

3718
    """
3719
    instance = self.cfg.GetInstanceInfo(
3720
      self.cfg.ExpandInstanceName(self.op.instance_name))
3721
    if instance is None:
3722
      raise errors.OpPrereqError("Instance '%s' not known" %
3723
                                 self.op.instance_name)
3724

    
3725
    if instance.disk_template != constants.DT_DRBD8:
3726
      raise errors.OpPrereqError("Instance's disk layout is not"
3727
                                 " drbd8, cannot migrate.")
3728

    
3729
    secondary_nodes = instance.secondary_nodes
3730
    if not secondary_nodes:
3731
      raise errors.ConfigurationError("No secondary node but using"
3732
                                      " drbd8 disk template")
3733

    
3734
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3735

    
3736
    target_node = secondary_nodes[0]
3737
    # check memory requirements on the secondary node
3738
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3739
                         instance.name, i_be[constants.BE_MEMORY],
3740
                         instance.hypervisor)
3741

    
3742
    # check bridge existance
3743
    brlist = [nic.bridge for nic in instance.nics]
3744
    result = self.rpc.call_bridges_exist(target_node, brlist)
3745
    if result.failed or not result.data:
3746
      raise errors.OpPrereqError("One or more target bridges %s does not"
3747
                                 " exist on destination node '%s'" %
3748
                                 (brlist, target_node))
3749

    
3750
    if not self.op.cleanup:
3751
      _CheckNodeNotDrained(self, target_node)
3752
      result = self.rpc.call_instance_migratable(instance.primary_node,
3753
                                                 instance)
3754
      msg = result.RemoteFailMsg()
3755
      if msg:
3756
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3757
                                   msg)
3758

    
3759
    self.instance = instance
3760

    
3761
  def _WaitUntilSync(self):
3762
    """Poll with custom rpc for disk sync.
3763

3764
    This uses our own step-based rpc call.
3765

3766
    """
3767
    self.feedback_fn("* wait until resync is done")
3768
    all_done = False
3769
    while not all_done:
3770
      all_done = True
3771
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3772
                                            self.nodes_ip,
3773
                                            self.instance.disks)
3774
      min_percent = 100
3775
      for node, nres in result.items():
3776
        msg = nres.RemoteFailMsg()
3777
        if msg:
3778
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3779
                                   (node, msg))
3780
        node_done, node_percent = nres.payload
3781
        all_done = all_done and node_done
3782
        if node_percent is not None:
3783
          min_percent = min(min_percent, node_percent)
3784
      if not all_done:
3785
        if min_percent < 100:
3786
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3787
        time.sleep(2)
3788

    
3789
  def _EnsureSecondary(self, node):
3790
    """Demote a node to secondary.
3791

3792
    """
3793
    self.feedback_fn("* switching node %s to secondary mode" % node)
3794

    
3795
    for dev in self.instance.disks:
3796
      self.cfg.SetDiskID(dev, node)
3797

    
3798
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3799
                                          self.instance.disks)
3800
    msg = result.RemoteFailMsg()
3801
    if msg:
3802
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3803
                               " error %s" % (node, msg))
3804

    
3805
  def _GoStandalone(self):
3806
    """Disconnect from the network.
3807

3808
    """
3809
    self.feedback_fn("* changing into standalone mode")
3810
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3811
                                               self.instance.disks)
3812
    for node, nres in result.items():
3813
      msg = nres.RemoteFailMsg()
3814
      if msg:
3815
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3816
                                 " error %s" % (node, msg))
3817

    
3818
  def _GoReconnect(self, multimaster):
3819
    """Reconnect to the network.
3820

3821
    """
3822
    if multimaster:
3823
      msg = "dual-master"
3824
    else:
3825
      msg = "single-master"
3826
    self.feedback_fn("* changing disks into %s mode" % msg)
3827
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3828
                                           self.instance.disks,
3829
                                           self.instance.name, multimaster)
3830
    for node, nres in result.items():
3831
      msg = nres.RemoteFailMsg()
3832
      if msg:
3833
        raise errors.OpExecError("Cannot change disks config on node %s,"
3834
                                 " error: %s" % (node, msg))
3835

    
3836
  def _ExecCleanup(self):
3837
    """Try to cleanup after a failed migration.
3838

3839
    The cleanup is done by:
3840
      - check that the instance is running only on one node
3841
        (and update the config if needed)
3842
      - change disks on its secondary node to secondary
3843
      - wait until disks are fully synchronized
3844
      - disconnect from the network
3845
      - change disks into single-master mode
3846
      - wait again until disks are fully synchronized
3847

3848
    """
3849
    instance = self.instance
3850
    target_node = self.target_node
3851
    source_node = self.source_node
3852

    
3853
    # check running on only one node
3854
    self.feedback_fn("* checking where the instance actually runs"
3855
                     " (if this hangs, the hypervisor might be in"
3856
                     " a bad state)")
3857
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3858
    for node, result in ins_l.items():
3859
      result.Raise()
3860
      if not isinstance(result.data, list):
3861
        raise errors.OpExecError("Can't contact node '%s'" % node)
3862

    
3863
    runningon_source = instance.name in ins_l[source_node].data
3864
    runningon_target = instance.name in ins_l[target_node].data
3865

    
3866
    if runningon_source and runningon_target:
3867
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3868
                               " or the hypervisor is confused. You will have"
3869
                               " to ensure manually that it runs only on one"
3870
                               " and restart this operation.")
3871

    
3872
    if not (runningon_source or runningon_target):
3873
      raise errors.OpExecError("Instance does not seem to be running at all."
3874
                               " In this case, it's safer to repair by"
3875
                               " running 'gnt-instance stop' to ensure disk"
3876
                               " shutdown, and then restarting it.")
3877

    
3878
    if runningon_target:
3879
      # the migration has actually succeeded, we need to update the config
3880
      self.feedback_fn("* instance running on secondary node (%s),"
3881
                       " updating config" % target_node)
3882
      instance.primary_node = target_node
3883
      self.cfg.Update(instance)
3884
      demoted_node = source_node
3885
    else:
3886
      self.feedback_fn("* instance confirmed to be running on its"
3887
                       " primary node (%s)" % source_node)
3888
      demoted_node = target_node
3889

    
3890
    self._EnsureSecondary(demoted_node)
3891
    try:
3892
      self._WaitUntilSync()
3893
    except errors.OpExecError:
3894
      # we ignore here errors, since if the device is standalone, it
3895
      # won't be able to sync
3896
      pass
3897
    self._GoStandalone()
3898
    self._GoReconnect(False)
3899
    self._WaitUntilSync()
3900

    
3901
    self.feedback_fn("* done")
3902

    
3903
  def _RevertDiskStatus(self):
3904
    """Try to revert the disk status after a failed migration.
3905

3906
    """
3907
    target_node = self.target_node
3908
    try:
3909
      self._EnsureSecondary(target_node)
3910
      self._GoStandalone()
3911
      self._GoReconnect(False)
3912
      self._WaitUntilSync()
3913
    except errors.OpExecError, err:
3914
      self.LogWarning("Migration failed and I can't reconnect the"
3915
                      " drives: error '%s'\n"
3916
                      "Please look and recover the instance status" %
3917
                      str(err))
3918

    
3919
  def _AbortMigration(self):
3920
    """Call the hypervisor code to abort a started migration.
3921

3922
    """
3923
    instance = self.instance
3924
    target_node = self.target_node
3925
    migration_info = self.migration_info
3926

    
3927
    abort_result = self.rpc.call_finalize_migration(target_node,
3928
                                                    instance,
3929
                                                    migration_info,
3930
                                                    False)
3931
    abort_msg = abort_result.RemoteFailMsg()
3932
    if abort_msg:
3933
      logging.error("Aborting migration failed on target node %s: %s" %
3934
                    (target_node, abort_msg))
3935
      # Don't raise an exception here, as we stil have to try to revert the
3936
      # disk status, even if this step failed.
3937

    
3938
  def _ExecMigration(self):
3939
    """Migrate an instance.
3940

3941
    The migrate is done by:
3942
      - change the disks into dual-master mode
3943
      - wait until disks are fully synchronized again
3944
      - migrate the instance
3945
      - change disks on the new secondary node (the old primary) to secondary
3946
      - wait until disks are fully synchronized
3947
      - change disks into single-master mode
3948

3949
    """
3950
    instance = self.instance
3951
    target_node = self.target_node
3952
    source_node = self.source_node
3953

    
3954
    self.feedback_fn("* checking disk consistency between source and target")
3955
    for dev in instance.disks:
3956
      if not _CheckDiskConsistency(self, dev, target_node, False):
3957
        raise errors.OpExecError("Disk %s is degraded or not fully"
3958
                                 " synchronized on target node,"
3959
                                 " aborting migrate." % dev.iv_name)
3960

    
3961
    # First get the migration information from the remote node
3962
    result = self.rpc.call_migration_info(source_node, instance)
3963
    msg = result.RemoteFailMsg()
3964
    if msg:
3965
      log_err = ("Failed fetching source migration information from %s: %s" %
3966
                 (source_node, msg))
3967
      logging.error(log_err)
3968
      raise errors.OpExecError(log_err)
3969

    
3970
    self.migration_info = migration_info = result.payload
3971

    
3972
    # Then switch the disks to master/master mode
3973
    self._EnsureSecondary(target_node)
3974
    self._GoStandalone()
3975
    self._GoReconnect(True)
3976
    self._WaitUntilSync()
3977

    
3978
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
3979
    result = self.rpc.call_accept_instance(target_node,
3980
                                           instance,
3981
                                           migration_info,
3982
                                           self.nodes_ip[target_node])
3983

    
3984
    msg = result.RemoteFailMsg()
3985
    if msg:
3986
      logging.error("Instance pre-migration failed, trying to revert"
3987
                    " disk status: %s", msg)
3988
      self._AbortMigration()
3989
      self._RevertDiskStatus()
3990
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
3991
                               (instance.name, msg))
3992

    
3993
    self.feedback_fn("* migrating instance to %s" % target_node)
3994
    time.sleep(10)
3995
    result = self.rpc.call_instance_migrate(source_node, instance,
3996
                                            self.nodes_ip[target_node],
3997
                                            self.op.live)
3998
    msg = result.RemoteFailMsg()
3999
    if msg:
4000
      logging.error("Instance migration failed, trying to revert"
4001
                    " disk status: %s", msg)
4002
      self._AbortMigration()
4003
      self._RevertDiskStatus()
4004
      raise errors.OpExecError("Could not migrate instance %s: %s" %
4005
                               (instance.name, msg))
4006
    time.sleep(10)
4007