Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 2fa74ef4

History | View | Annotate | Download (252.1 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, mac, mode, link) in enumerate(nics):
505
      if ip is None:
506
        ip = ""
507
      env["INSTANCE_NIC%d_IP" % idx] = ip
508
      env["INSTANCE_NIC%d_MAC" % idx] = mac
509
      env["INSTANCE_NIC%d_MODE" % idx] = mode
510
      env["INSTANCE_NIC%d_LINK" % idx] = link
511
      if mode == constants.NIC_MODE_BRIDGED:
512
        env["INSTANCE_NIC%d_BRIDGE" % idx] = link
513
  else:
514
    nic_count = 0
515

    
516
  env["INSTANCE_NIC_COUNT"] = nic_count
517

    
518
  if disks:
519
    disk_count = len(disks)
520
    for idx, (size, mode) in enumerate(disks):
521
      env["INSTANCE_DISK%d_SIZE" % idx] = size
522
      env["INSTANCE_DISK%d_MODE" % idx] = mode
523
  else:
524
    disk_count = 0
525

    
526
  env["INSTANCE_DISK_COUNT"] = disk_count
527

    
528
  return env
529

    
530
def _PreBuildNICHooksList(lu, nics):
531
  """Build a list of nic information tuples.
532

533
  This list is suitable to be passed to _BuildInstanceHookEnv.
534

535
  @type lu:  L{LogicalUnit}
536
  @param lu: the logical unit on whose behalf we execute
537
  @type nics: list of L{objects.NIC}
538
  @param nics: list of nics to convert to hooks tuples
539

540
  """
541
  hooks_nics = []
542
  c_nicparams = lu.cfg.GetClusterInfo().nicparams[constants.PP_DEFAULT]
543
  for nic in nics:
544
    ip = nic.ip
545
    mac = nic.mac
546
    filled_params = objects.FillDict(c_nicparams, nic.nicparams)
547
    mode = filled_params[constants.NIC_MODE]
548
    link = filled_params[constants.NIC_LINK]
549
    hooks_nics.append((ip, mac, mode, link))
550
  return hooks_nics
551

    
552
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
553
  """Builds instance related env variables for hooks from an object.
554

555
  @type lu: L{LogicalUnit}
556
  @param lu: the logical unit on whose behalf we execute
557
  @type instance: L{objects.Instance}
558
  @param instance: the instance for which we should build the
559
      environment
560
  @type override: dict
561
  @param override: dictionary with key/values that will override
562
      our values
563
  @rtype: dict
564
  @return: the hook environment dictionary
565

566
  """
567
  bep = lu.cfg.GetClusterInfo().FillBE(instance)
568
  args = {
569
    'name': instance.name,
570
    'primary_node': instance.primary_node,
571
    'secondary_nodes': instance.secondary_nodes,
572
    'os_type': instance.os,
573
    'status': instance.admin_up,
574
    'memory': bep[constants.BE_MEMORY],
575
    'vcpus': bep[constants.BE_VCPUS],
576
    'nics': _PreBuildNICHooksList(lu, instance.nics),
577
    'disk_template': instance.disk_template,
578
    'disks': [(disk.size, disk.mode) for disk in instance.disks],
579
  }
580
  if override:
581
    args.update(override)
582
  return _BuildInstanceHookEnv(**args)
583

    
584

    
585
def _AdjustCandidatePool(lu):
586
  """Adjust the candidate pool after node operations.
587

588
  """
589
  mod_list = lu.cfg.MaintainCandidatePool()
590
  if mod_list:
591
    lu.LogInfo("Promoted nodes to master candidate role: %s",
592
               ", ".join(node.name for node in mod_list))
593
    for name in mod_list:
594
      lu.context.ReaddNode(name)
595
  mc_now, mc_max = lu.cfg.GetMasterCandidateStats()
596
  if mc_now > mc_max:
597
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
598
               (mc_now, mc_max))
599

    
600

    
601
def _CheckNicsBridgesExist(lu, target_nics, target_node,
602
                               profile=constants.PP_DEFAULT):
603
  """Check that the brigdes needed by a list of nics exist.
604

605
  """
606
  c_nicparams = lu.cfg.GetClusterInfo().nicparams[profile]
607
  paramslist = [objects.FillDict(c_nicparams, nic.nicparams)
608
                for nic in target_nics]
609
  brlist = [params[constants.NIC_LINK] for params in paramslist
610
            if params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED]
611
  if brlist:
612
    result = lu.rpc.call_bridges_exist(target_node, brlist)
613
    msg = result.RemoteFailMsg()
614
    if msg:
615
      raise errors.OpPrereqError("Error checking bridges on destination node"
616
                                 " '%s': %s" % (target_node, msg))
617

    
618

    
619
def _CheckInstanceBridgesExist(lu, instance, node=None):
620
  """Check that the brigdes needed by an instance exist.
621

622
  """
623
  if node is None:
624
    node=instance.primary_node
625
  _CheckNicsBridgesExist(lu, instance.nics, node)
626

    
627

    
628
class LUDestroyCluster(NoHooksLU):
629
  """Logical unit for destroying the cluster.
630

631
  """
632
  _OP_REQP = []
633

    
634
  def CheckPrereq(self):
635
    """Check prerequisites.
636

637
    This checks whether the cluster is empty.
638

639
    Any errors are signalled by raising errors.OpPrereqError.
640

641
    """
642
    master = self.cfg.GetMasterNode()
643

    
644
    nodelist = self.cfg.GetNodeList()
645
    if len(nodelist) != 1 or nodelist[0] != master:
646
      raise errors.OpPrereqError("There are still %d node(s) in"
647
                                 " this cluster." % (len(nodelist) - 1))
648
    instancelist = self.cfg.GetInstanceList()
649
    if instancelist:
650
      raise errors.OpPrereqError("There are still %d instance(s) in"
651
                                 " this cluster." % len(instancelist))
652

    
653
  def Exec(self, feedback_fn):
654
    """Destroys the cluster.
655

656
    """
657
    master = self.cfg.GetMasterNode()
658
    result = self.rpc.call_node_stop_master(master, False)
659
    result.Raise()
660
    if not result.data:
661
      raise errors.OpExecError("Could not disable the master role")
662
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
663
    utils.CreateBackup(priv_key)
664
    utils.CreateBackup(pub_key)
665
    return master
666

    
667

    
668
class LUVerifyCluster(LogicalUnit):
669
  """Verifies the cluster status.
670

671
  """
672
  HPATH = "cluster-verify"
673
  HTYPE = constants.HTYPE_CLUSTER
674
  _OP_REQP = ["skip_checks"]
675
  REQ_BGL = False
676

    
677
  def ExpandNames(self):
678
    self.needed_locks = {
679
      locking.LEVEL_NODE: locking.ALL_SET,
680
      locking.LEVEL_INSTANCE: locking.ALL_SET,
681
    }
682
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
683

    
684
  def _VerifyNode(self, nodeinfo, file_list, local_cksum,
685
                  node_result, feedback_fn, master_files,
686
                  drbd_map, vg_name):
687
    """Run multiple tests against a node.
688

689
    Test list:
690

691
      - compares ganeti version
692
      - checks vg existance and size > 20G
693
      - checks config file checksum
694
      - checks ssh to other nodes
695

696
    @type nodeinfo: L{objects.Node}
697
    @param nodeinfo: the node to check
698
    @param file_list: required list of files
699
    @param local_cksum: dictionary of local files and their checksums
700
    @param node_result: the results from the node
701
    @param feedback_fn: function used to accumulate results
702
    @param master_files: list of files that only masters should have
703
    @param drbd_map: the useddrbd minors for this node, in
704
        form of minor: (instance, must_exist) which correspond to instances
705
        and their running status
706
    @param vg_name: Ganeti Volume Group (result of self.cfg.GetVGName())
707

708
    """
709
    node = nodeinfo.name
710

    
711
    # main result, node_result should be a non-empty dict
712
    if not node_result or not isinstance(node_result, dict):
713
      feedback_fn("  - ERROR: unable to verify node %s." % (node,))
714
      return True
715

    
716
    # compares ganeti version
717
    local_version = constants.PROTOCOL_VERSION
718
    remote_version = node_result.get('version', None)
719
    if not (remote_version and isinstance(remote_version, (list, tuple)) and
720
            len(remote_version) == 2):
721
      feedback_fn("  - ERROR: connection to %s failed" % (node))
722
      return True
723

    
724
    if local_version != remote_version[0]:
725
      feedback_fn("  - ERROR: incompatible protocol versions: master %s,"
726
                  " node %s %s" % (local_version, node, remote_version[0]))
727
      return True
728

    
729
    # node seems compatible, we can actually try to look into its results
730

    
731
    bad = False
732

    
733
    # full package version
734
    if constants.RELEASE_VERSION != remote_version[1]:
735
      feedback_fn("  - WARNING: software version mismatch: master %s,"
736
                  " node %s %s" %
737
                  (constants.RELEASE_VERSION, node, remote_version[1]))
738

    
739
    # checks vg existence and size > 20G
740
    if vg_name is not None:
741
      vglist = node_result.get(constants.NV_VGLIST, None)
742
      if not vglist:
743
        feedback_fn("  - ERROR: unable to check volume groups on node %s." %
744
                        (node,))
745
        bad = True
746
      else:
747
        vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
748
                                              constants.MIN_VG_SIZE)
749
        if vgstatus:
750
          feedback_fn("  - ERROR: %s on node %s" % (vgstatus, node))
751
          bad = True
752

    
753
    # checks config file checksum
754

    
755
    remote_cksum = node_result.get(constants.NV_FILELIST, None)
756
    if not isinstance(remote_cksum, dict):
757
      bad = True
758
      feedback_fn("  - ERROR: node hasn't returned file checksum data")
759
    else:
760
      for file_name in file_list:
761
        node_is_mc = nodeinfo.master_candidate
762
        must_have_file = file_name not in master_files
763
        if file_name not in remote_cksum:
764
          if node_is_mc or must_have_file:
765
            bad = True
766
            feedback_fn("  - ERROR: file '%s' missing" % file_name)
767
        elif remote_cksum[file_name] != local_cksum[file_name]:
768
          if node_is_mc or must_have_file:
769
            bad = True
770
            feedback_fn("  - ERROR: file '%s' has wrong checksum" % file_name)
771
          else:
772
            # not candidate and this is not a must-have file
773
            bad = True
774
            feedback_fn("  - ERROR: non master-candidate has old/wrong file"
775
                        " '%s'" % file_name)
776
        else:
777
          # all good, except non-master/non-must have combination
778
          if not node_is_mc and not must_have_file:
779
            feedback_fn("  - ERROR: file '%s' should not exist on non master"
780
                        " candidates" % file_name)
781

    
782
    # checks ssh to any
783

    
784
    if constants.NV_NODELIST not in node_result:
785
      bad = True
786
      feedback_fn("  - ERROR: node hasn't returned node ssh connectivity data")
787
    else:
788
      if node_result[constants.NV_NODELIST]:
789
        bad = True
790
        for node in node_result[constants.NV_NODELIST]:
791
          feedback_fn("  - ERROR: ssh communication with node '%s': %s" %
792
                          (node, node_result[constants.NV_NODELIST][node]))
793

    
794
    if constants.NV_NODENETTEST not in node_result:
795
      bad = True
796
      feedback_fn("  - ERROR: node hasn't returned node tcp connectivity data")
797
    else:
798
      if node_result[constants.NV_NODENETTEST]:
799
        bad = True
800
        nlist = utils.NiceSort(node_result[constants.NV_NODENETTEST].keys())
801
        for node in nlist:
802
          feedback_fn("  - ERROR: tcp communication with node '%s': %s" %
803
                          (node, node_result[constants.NV_NODENETTEST][node]))
804

    
805
    hyp_result = node_result.get(constants.NV_HYPERVISOR, None)
806
    if isinstance(hyp_result, dict):
807
      for hv_name, hv_result in hyp_result.iteritems():
808
        if hv_result is not None:
809
          feedback_fn("  - ERROR: hypervisor %s verify failure: '%s'" %
810
                      (hv_name, hv_result))
811

    
812
    # check used drbd list
813
    if vg_name is not None:
814
      used_minors = node_result.get(constants.NV_DRBDLIST, [])
815
      if not isinstance(used_minors, (tuple, list)):
816
        feedback_fn("  - ERROR: cannot parse drbd status file: %s" %
817
                    str(used_minors))
818
      else:
819
        for minor, (iname, must_exist) in drbd_map.items():
820
          if minor not in used_minors and must_exist:
821
            feedback_fn("  - ERROR: drbd minor %d of instance %s is"
822
                        " not active" % (minor, iname))
823
            bad = True
824
        for minor in used_minors:
825
          if minor not in drbd_map:
826
            feedback_fn("  - ERROR: unallocated drbd minor %d is in use" %
827
                        minor)
828
            bad = True
829

    
830
    return bad
831

    
832
  def _VerifyInstance(self, instance, instanceconfig, node_vol_is,
833
                      node_instance, feedback_fn, n_offline):
834
    """Verify an instance.
835

836
    This function checks to see if the required block devices are
837
    available on the instance's node.
838

839
    """
840
    bad = False
841

    
842
    node_current = instanceconfig.primary_node
843

    
844
    node_vol_should = {}
845
    instanceconfig.MapLVsByNode(node_vol_should)
846

    
847
    for node in node_vol_should:
848
      if node in n_offline:
849
        # ignore missing volumes on offline nodes
850
        continue
851
      for volume in node_vol_should[node]:
852
        if node not in node_vol_is or volume not in node_vol_is[node]:
853
          feedback_fn("  - ERROR: volume %s missing on node %s" %
854
                          (volume, node))
855
          bad = True
856

    
857
    if instanceconfig.admin_up:
858
      if ((node_current not in node_instance or
859
          not instance in node_instance[node_current]) and
860
          node_current not in n_offline):
861
        feedback_fn("  - ERROR: instance %s not running on node %s" %
862
                        (instance, node_current))
863
        bad = True
864

    
865
    for node in node_instance:
866
      if (not node == node_current):
867
        if instance in node_instance[node]:
868
          feedback_fn("  - ERROR: instance %s should not run on node %s" %
869
                          (instance, node))
870
          bad = True
871

    
872
    return bad
873

    
874
  def _VerifyOrphanVolumes(self, node_vol_should, node_vol_is, feedback_fn):
875
    """Verify if there are any unknown volumes in the cluster.
876

877
    The .os, .swap and backup volumes are ignored. All other volumes are
878
    reported as unknown.
879

880
    """
881
    bad = False
882

    
883
    for node in node_vol_is:
884
      for volume in node_vol_is[node]:
885
        if node not in node_vol_should or volume not in node_vol_should[node]:
886
          feedback_fn("  - ERROR: volume %s on node %s should not exist" %
887
                      (volume, node))
888
          bad = True
889
    return bad
890

    
891
  def _VerifyOrphanInstances(self, instancelist, node_instance, feedback_fn):
892
    """Verify the list of running instances.
893

894
    This checks what instances are running but unknown to the cluster.
895

896
    """
897
    bad = False
898
    for node in node_instance:
899
      for runninginstance in node_instance[node]:
900
        if runninginstance not in instancelist:
901
          feedback_fn("  - ERROR: instance %s on node %s should not exist" %
902
                          (runninginstance, node))
903
          bad = True
904
    return bad
905

    
906
  def _VerifyNPlusOneMemory(self, node_info, instance_cfg, feedback_fn):
907
    """Verify N+1 Memory Resilience.
908

909
    Check that if one single node dies we can still start all the instances it
910
    was primary for.
911

912
    """
913
    bad = False
914

    
915
    for node, nodeinfo in node_info.iteritems():
916
      # This code checks that every node which is now listed as secondary has
917
      # enough memory to host all instances it is supposed to should a single
918
      # other node in the cluster fail.
919
      # FIXME: not ready for failover to an arbitrary node
920
      # FIXME: does not support file-backed instances
921
      # WARNING: we currently take into account down instances as well as up
922
      # ones, considering that even if they're down someone might want to start
923
      # them even in the event of a node failure.
924
      for prinode, instances in nodeinfo['sinst-by-pnode'].iteritems():
925
        needed_mem = 0
926
        for instance in instances:
927
          bep = self.cfg.GetClusterInfo().FillBE(instance_cfg[instance])
928
          if bep[constants.BE_AUTO_BALANCE]:
929
            needed_mem += bep[constants.BE_MEMORY]
930
        if nodeinfo['mfree'] < needed_mem:
931
          feedback_fn("  - ERROR: not enough memory on node %s to accomodate"
932
                      " failovers should node %s fail" % (node, prinode))
933
          bad = True
934
    return bad
935

    
936
  def CheckPrereq(self):
937
    """Check prerequisites.
938

939
    Transform the list of checks we're going to skip into a set and check that
940
    all its members are valid.
941

942
    """
943
    self.skip_set = frozenset(self.op.skip_checks)
944
    if not constants.VERIFY_OPTIONAL_CHECKS.issuperset(self.skip_set):
945
      raise errors.OpPrereqError("Invalid checks to be skipped specified")
946

    
947
  def BuildHooksEnv(self):
948
    """Build hooks env.
949

950
    Cluster-Verify hooks just rone in the post phase and their failure makes
951
    the output be logged in the verify output and the verification to fail.
952

953
    """
954
    all_nodes = self.cfg.GetNodeList()
955
    env = {
956
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
957
      }
958
    for node in self.cfg.GetAllNodesInfo().values():
959
      env["NODE_TAGS_%s" % node.name] = " ".join(node.GetTags())
960

    
961
    return env, [], all_nodes
962

    
963
  def Exec(self, feedback_fn):
964
    """Verify integrity of cluster, performing various test on nodes.
965

966
    """
967
    bad = False
968
    feedback_fn("* Verifying global settings")
969
    for msg in self.cfg.VerifyConfig():
970
      feedback_fn("  - ERROR: %s" % msg)
971

    
972
    vg_name = self.cfg.GetVGName()
973
    hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
974
    nodelist = utils.NiceSort(self.cfg.GetNodeList())
975
    nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
976
    instancelist = utils.NiceSort(self.cfg.GetInstanceList())
977
    instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
978
                        for iname in instancelist)
979
    i_non_redundant = [] # Non redundant instances
980
    i_non_a_balanced = [] # Non auto-balanced instances
981
    n_offline = [] # List of offline nodes
982
    n_drained = [] # List of nodes being drained
983
    node_volume = {}
984
    node_instance = {}
985
    node_info = {}
986
    instance_cfg = {}
987

    
988
    # FIXME: verify OS list
989
    # do local checksums
990
    master_files = [constants.CLUSTER_CONF_FILE]
991

    
992
    file_names = ssconf.SimpleStore().GetFileList()
993
    file_names.append(constants.SSL_CERT_FILE)
994
    file_names.append(constants.RAPI_CERT_FILE)
995
    file_names.extend(master_files)
996

    
997
    local_checksums = utils.FingerprintFiles(file_names)
998

    
999
    feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
1000
    node_verify_param = {
1001
      constants.NV_FILELIST: file_names,
1002
      constants.NV_NODELIST: [node.name for node in nodeinfo
1003
                              if not node.offline],
1004
      constants.NV_HYPERVISOR: hypervisors,
1005
      constants.NV_NODENETTEST: [(node.name, node.primary_ip,
1006
                                  node.secondary_ip) for node in nodeinfo
1007
                                 if not node.offline],
1008
      constants.NV_INSTANCELIST: hypervisors,
1009
      constants.NV_VERSION: None,
1010
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
1011
      }
1012
    if vg_name is not None:
1013
      node_verify_param[constants.NV_VGLIST] = None
1014
      node_verify_param[constants.NV_LVLIST] = vg_name
1015
      node_verify_param[constants.NV_DRBDLIST] = None
1016
    all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
1017
                                           self.cfg.GetClusterName())
1018

    
1019
    cluster = self.cfg.GetClusterInfo()
1020
    master_node = self.cfg.GetMasterNode()
1021
    all_drbd_map = self.cfg.ComputeDRBDMap()
1022

    
1023
    for node_i in nodeinfo:
1024
      node = node_i.name
1025
      nresult = all_nvinfo[node].data
1026

    
1027
      if node_i.offline:
1028
        feedback_fn("* Skipping offline node %s" % (node,))
1029
        n_offline.append(node)
1030
        continue
1031

    
1032
      if node == master_node:
1033
        ntype = "master"
1034
      elif node_i.master_candidate:
1035
        ntype = "master candidate"
1036
      elif node_i.drained:
1037
        ntype = "drained"
1038
        n_drained.append(node)
1039
      else:
1040
        ntype = "regular"
1041
      feedback_fn("* Verifying node %s (%s)" % (node, ntype))
1042

    
1043
      if all_nvinfo[node].failed or not isinstance(nresult, dict):
1044
        feedback_fn("  - ERROR: connection to %s failed" % (node,))
1045
        bad = True
1046
        continue
1047

    
1048
      node_drbd = {}
1049
      for minor, instance in all_drbd_map[node].items():
1050
        if instance not in instanceinfo:
1051
          feedback_fn("  - ERROR: ghost instance '%s' in temporary DRBD map" %
1052
                      instance)
1053
          # ghost instance should not be running, but otherwise we
1054
          # don't give double warnings (both ghost instance and
1055
          # unallocated minor in use)
1056
          node_drbd[minor] = (instance, False)
1057
        else:
1058
          instance = instanceinfo[instance]
1059
          node_drbd[minor] = (instance.name, instance.admin_up)
1060
      result = self._VerifyNode(node_i, file_names, local_checksums,
1061
                                nresult, feedback_fn, master_files,
1062
                                node_drbd, vg_name)
1063
      bad = bad or result
1064

    
1065
      lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
1066
      if vg_name is None:
1067
        node_volume[node] = {}
1068
      elif isinstance(lvdata, basestring):
1069
        feedback_fn("  - ERROR: LVM problem on node %s: %s" %
1070
                    (node, utils.SafeEncode(lvdata)))
1071
        bad = True
1072
        node_volume[node] = {}
1073
      elif not isinstance(lvdata, dict):
1074
        feedback_fn("  - ERROR: connection to %s failed (lvlist)" % (node,))
1075
        bad = True
1076
        continue
1077
      else:
1078
        node_volume[node] = lvdata
1079

    
1080
      # node_instance
1081
      idata = nresult.get(constants.NV_INSTANCELIST, None)
1082
      if not isinstance(idata, list):
1083
        feedback_fn("  - ERROR: connection to %s failed (instancelist)" %
1084
                    (node,))
1085
        bad = True
1086
        continue
1087

    
1088
      node_instance[node] = idata
1089

    
1090
      # node_info
1091
      nodeinfo = nresult.get(constants.NV_HVINFO, None)
1092
      if not isinstance(nodeinfo, dict):
1093
        feedback_fn("  - ERROR: connection to %s failed (hvinfo)" % (node,))
1094
        bad = True
1095
        continue
1096

    
1097
      try:
1098
        node_info[node] = {
1099
          "mfree": int(nodeinfo['memory_free']),
1100
          "pinst": [],
1101
          "sinst": [],
1102
          # dictionary holding all instances this node is secondary for,
1103
          # grouped by their primary node. Each key is a cluster node, and each
1104
          # value is a list of instances which have the key as primary and the
1105
          # current node as secondary.  this is handy to calculate N+1 memory
1106
          # availability if you can only failover from a primary to its
1107
          # secondary.
1108
          "sinst-by-pnode": {},
1109
        }
1110
        # FIXME: devise a free space model for file based instances as well
1111
        if vg_name is not None:
1112
          if (constants.NV_VGLIST not in nresult or
1113
              vg_name not in nresult[constants.NV_VGLIST]):
1114
            feedback_fn("  - ERROR: node %s didn't return data for the"
1115
                        " volume group '%s' - it is either missing or broken" %
1116
                        (node, vg_name))
1117
            bad = True
1118
            continue
1119
          node_info[node]["dfree"] = int(nresult[constants.NV_VGLIST][vg_name])
1120
      except (ValueError, KeyError):
1121
        feedback_fn("  - ERROR: invalid nodeinfo value returned"
1122
                    " from node %s" % (node,))
1123
        bad = True
1124
        continue
1125

    
1126
    node_vol_should = {}
1127

    
1128
    for instance in instancelist:
1129
      feedback_fn("* Verifying instance %s" % instance)
1130
      inst_config = instanceinfo[instance]
1131
      result =  self._VerifyInstance(instance, inst_config, node_volume,
1132
                                     node_instance, feedback_fn, n_offline)
1133
      bad = bad or result
1134
      inst_nodes_offline = []
1135

    
1136
      inst_config.MapLVsByNode(node_vol_should)
1137

    
1138
      instance_cfg[instance] = inst_config
1139

    
1140
      pnode = inst_config.primary_node
1141
      if pnode in node_info:
1142
        node_info[pnode]['pinst'].append(instance)
1143
      elif pnode not in n_offline:
1144
        feedback_fn("  - ERROR: instance %s, connection to primary node"
1145
                    " %s failed" % (instance, pnode))
1146
        bad = True
1147

    
1148
      if pnode in n_offline:
1149
        inst_nodes_offline.append(pnode)
1150

    
1151
      # If the instance is non-redundant we cannot survive losing its primary
1152
      # node, so we are not N+1 compliant. On the other hand we have no disk
1153
      # templates with more than one secondary so that situation is not well
1154
      # supported either.
1155
      # FIXME: does not support file-backed instances
1156
      if len(inst_config.secondary_nodes) == 0:
1157
        i_non_redundant.append(instance)
1158
      elif len(inst_config.secondary_nodes) > 1:
1159
        feedback_fn("  - WARNING: multiple secondaries for instance %s"
1160
                    % instance)
1161

    
1162
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
1163
        i_non_a_balanced.append(instance)
1164

    
1165
      for snode in inst_config.secondary_nodes:
1166
        if snode in node_info:
1167
          node_info[snode]['sinst'].append(instance)
1168
          if pnode not in node_info[snode]['sinst-by-pnode']:
1169
            node_info[snode]['sinst-by-pnode'][pnode] = []
1170
          node_info[snode]['sinst-by-pnode'][pnode].append(instance)
1171
        elif snode not in n_offline:
1172
          feedback_fn("  - ERROR: instance %s, connection to secondary node"
1173
                      " %s failed" % (instance, snode))
1174
          bad = True
1175
        if snode in n_offline:
1176
          inst_nodes_offline.append(snode)
1177

    
1178
      if inst_nodes_offline:
1179
        # warn that the instance lives on offline nodes, and set bad=True
1180
        feedback_fn("  - ERROR: instance lives on offline node(s) %s" %
1181
                    ", ".join(inst_nodes_offline))
1182
        bad = True
1183

    
1184
    feedback_fn("* Verifying orphan volumes")
1185
    result = self._VerifyOrphanVolumes(node_vol_should, node_volume,
1186
                                       feedback_fn)
1187
    bad = bad or result
1188

    
1189
    feedback_fn("* Verifying remaining instances")
1190
    result = self._VerifyOrphanInstances(instancelist, node_instance,
1191
                                         feedback_fn)
1192
    bad = bad or result
1193

    
1194
    if constants.VERIFY_NPLUSONE_MEM not in self.skip_set:
1195
      feedback_fn("* Verifying N+1 Memory redundancy")
1196
      result = self._VerifyNPlusOneMemory(node_info, instance_cfg, feedback_fn)
1197
      bad = bad or result
1198

    
1199
    feedback_fn("* Other Notes")
1200
    if i_non_redundant:
1201
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
1202
                  % len(i_non_redundant))
1203

    
1204
    if i_non_a_balanced:
1205
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
1206
                  % len(i_non_a_balanced))
1207

    
1208
    if n_offline:
1209
      feedback_fn("  - NOTICE: %d offline node(s) found." % len(n_offline))
1210

    
1211
    if n_drained:
1212
      feedback_fn("  - NOTICE: %d drained node(s) found." % len(n_drained))
1213

    
1214
    return not bad
1215

    
1216
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
1217
    """Analize the post-hooks' result
1218

1219
    This method analyses the hook result, handles it, and sends some
1220
    nicely-formatted feedback back to the user.
1221

1222
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
1223
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
1224
    @param hooks_results: the results of the multi-node hooks rpc call
1225
    @param feedback_fn: function used send feedback back to the caller
1226
    @param lu_result: previous Exec result
1227
    @return: the new Exec result, based on the previous result
1228
        and hook results
1229

1230
    """
1231
    # We only really run POST phase hooks, and are only interested in
1232
    # their results
1233
    if phase == constants.HOOKS_PHASE_POST:
1234
      # Used to change hooks' output to proper indentation
1235
      indent_re = re.compile('^', re.M)
1236
      feedback_fn("* Hooks Results")
1237
      if not hooks_results:
1238
        feedback_fn("  - ERROR: general communication failure")
1239
        lu_result = 1
1240
      else:
1241
        for node_name in hooks_results:
1242
          show_node_header = True
1243
          res = hooks_results[node_name]
1244
          if res.failed or res.data is False or not isinstance(res.data, list):
1245
            if res.offline:
1246
              # no need to warn or set fail return value
1247
              continue
1248
            feedback_fn("    Communication failure in hooks execution")
1249
            lu_result = 1
1250
            continue
1251
          for script, hkr, output in res.data:
1252
            if hkr == constants.HKR_FAIL:
1253
              # The node header is only shown once, if there are
1254
              # failing hooks on that node
1255
              if show_node_header:
1256
                feedback_fn("  Node %s:" % node_name)
1257
                show_node_header = False
1258
              feedback_fn("    ERROR: Script %s failed, output:" % script)
1259
              output = indent_re.sub('      ', output)
1260
              feedback_fn("%s" % output)
1261
              lu_result = 1
1262

    
1263
      return lu_result
1264

    
1265

    
1266
class LUVerifyDisks(NoHooksLU):
1267
  """Verifies the cluster disks status.
1268

1269
  """
1270
  _OP_REQP = []
1271
  REQ_BGL = False
1272

    
1273
  def ExpandNames(self):
1274
    self.needed_locks = {
1275
      locking.LEVEL_NODE: locking.ALL_SET,
1276
      locking.LEVEL_INSTANCE: locking.ALL_SET,
1277
    }
1278
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
1279

    
1280
  def CheckPrereq(self):
1281
    """Check prerequisites.
1282

1283
    This has no prerequisites.
1284

1285
    """
1286
    pass
1287

    
1288
  def Exec(self, feedback_fn):
1289
    """Verify integrity of cluster disks.
1290

1291
    @rtype: tuple of three items
1292
    @return: a tuple of (dict of node-to-node_error, list of instances
1293
        which need activate-disks, dict of instance: (node, volume) for
1294
        missing volumes
1295

1296
    """
1297
    result = res_nodes, res_instances, res_missing = {}, [], {}
1298

    
1299
    vg_name = self.cfg.GetVGName()
1300
    nodes = utils.NiceSort(self.cfg.GetNodeList())
1301
    instances = [self.cfg.GetInstanceInfo(name)
1302
                 for name in self.cfg.GetInstanceList()]
1303

    
1304
    nv_dict = {}
1305
    for inst in instances:
1306
      inst_lvs = {}
1307
      if (not inst.admin_up or
1308
          inst.disk_template not in constants.DTS_NET_MIRROR):
1309
        continue
1310
      inst.MapLVsByNode(inst_lvs)
1311
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
1312
      for node, vol_list in inst_lvs.iteritems():
1313
        for vol in vol_list:
1314
          nv_dict[(node, vol)] = inst
1315

    
1316
    if not nv_dict:
1317
      return result
1318

    
1319
    node_lvs = self.rpc.call_volume_list(nodes, vg_name)
1320

    
1321
    to_act = set()
1322
    for node in nodes:
1323
      # node_volume
1324
      node_res = node_lvs[node]
1325
      if node_res.offline:
1326
        continue
1327
      msg = node_res.RemoteFailMsg()
1328
      if msg:
1329
        logging.warning("Error enumerating LVs on node %s: %s", node, msg)
1330
        res_nodes[node] = msg
1331
        continue
1332

    
1333
      lvs = node_res.payload
1334
      for lv_name, (_, lv_inactive, lv_online) in lvs.items():
1335
        inst = nv_dict.pop((node, lv_name), None)
1336
        if (not lv_online and inst is not None
1337
            and inst.name not in res_instances):
1338
          res_instances.append(inst.name)
1339

    
1340
    # any leftover items in nv_dict are missing LVs, let's arrange the
1341
    # data better
1342
    for key, inst in nv_dict.iteritems():
1343
      if inst.name not in res_missing:
1344
        res_missing[inst.name] = []
1345
      res_missing[inst.name].append(key)
1346

    
1347
    return result
1348

    
1349

    
1350
class LURenameCluster(LogicalUnit):
1351
  """Rename the cluster.
1352

1353
  """
1354
  HPATH = "cluster-rename"
1355
  HTYPE = constants.HTYPE_CLUSTER
1356
  _OP_REQP = ["name"]
1357

    
1358
  def BuildHooksEnv(self):
1359
    """Build hooks env.
1360

1361
    """
1362
    env = {
1363
      "OP_TARGET": self.cfg.GetClusterName(),
1364
      "NEW_NAME": self.op.name,
1365
      }
1366
    mn = self.cfg.GetMasterNode()
1367
    return env, [mn], [mn]
1368

    
1369
  def CheckPrereq(self):
1370
    """Verify that the passed name is a valid one.
1371

1372
    """
1373
    hostname = utils.HostInfo(self.op.name)
1374

    
1375
    new_name = hostname.name
1376
    self.ip = new_ip = hostname.ip
1377
    old_name = self.cfg.GetClusterName()
1378
    old_ip = self.cfg.GetMasterIP()
1379
    if new_name == old_name and new_ip == old_ip:
1380
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
1381
                                 " cluster has changed")
1382
    if new_ip != old_ip:
1383
      if utils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
1384
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
1385
                                   " reachable on the network. Aborting." %
1386
                                   new_ip)
1387

    
1388
    self.op.name = new_name
1389

    
1390
  def Exec(self, feedback_fn):
1391
    """Rename the cluster.
1392

1393
    """
1394
    clustername = self.op.name
1395
    ip = self.ip
1396

    
1397
    # shutdown the master IP
1398
    master = self.cfg.GetMasterNode()
1399
    result = self.rpc.call_node_stop_master(master, False)
1400
    if result.failed or not result.data:
1401
      raise errors.OpExecError("Could not disable the master role")
1402

    
1403
    try:
1404
      cluster = self.cfg.GetClusterInfo()
1405
      cluster.cluster_name = clustername
1406
      cluster.master_ip = ip
1407
      self.cfg.Update(cluster)
1408

    
1409
      # update the known hosts file
1410
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
1411
      node_list = self.cfg.GetNodeList()
1412
      try:
1413
        node_list.remove(master)
1414
      except ValueError:
1415
        pass
1416
      result = self.rpc.call_upload_file(node_list,
1417
                                         constants.SSH_KNOWN_HOSTS_FILE)
1418
      for to_node, to_result in result.iteritems():
1419
         msg = to_result.RemoteFailMsg()
1420
         if msg:
1421
           msg = ("Copy of file %s to node %s failed: %s" %
1422
                   (constants.SSH_KNOWN_HOSTS_FILE, to_node, msg))
1423
           self.proc.LogWarning(msg)
1424

    
1425
    finally:
1426
      result = self.rpc.call_node_start_master(master, False)
1427
      if result.failed or not result.data:
1428
        self.LogWarning("Could not re-enable the master role on"
1429
                        " the master, please restart manually.")
1430

    
1431

    
1432
def _RecursiveCheckIfLVMBased(disk):
1433
  """Check if the given disk or its children are lvm-based.
1434

1435
  @type disk: L{objects.Disk}
1436
  @param disk: the disk to check
1437
  @rtype: booleean
1438
  @return: boolean indicating whether a LD_LV dev_type was found or not
1439

1440
  """
1441
  if disk.children:
1442
    for chdisk in disk.children:
1443
      if _RecursiveCheckIfLVMBased(chdisk):
1444
        return True
1445
  return disk.dev_type == constants.LD_LV
1446

    
1447

    
1448
class LUSetClusterParams(LogicalUnit):
1449
  """Change the parameters of the cluster.
1450

1451
  """
1452
  HPATH = "cluster-modify"
1453
  HTYPE = constants.HTYPE_CLUSTER
1454
  _OP_REQP = []
1455
  REQ_BGL = False
1456

    
1457
  def CheckArguments(self):
1458
    """Check parameters
1459

1460
    """
1461
    if not hasattr(self.op, "candidate_pool_size"):
1462
      self.op.candidate_pool_size = None
1463
    if self.op.candidate_pool_size is not None:
1464
      try:
1465
        self.op.candidate_pool_size = int(self.op.candidate_pool_size)
1466
      except (ValueError, TypeError), err:
1467
        raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
1468
                                   str(err))
1469
      if self.op.candidate_pool_size < 1:
1470
        raise errors.OpPrereqError("At least one master candidate needed")
1471

    
1472
  def ExpandNames(self):
1473
    # FIXME: in the future maybe other cluster params won't require checking on
1474
    # all nodes to be modified.
1475
    self.needed_locks = {
1476
      locking.LEVEL_NODE: locking.ALL_SET,
1477
    }
1478
    self.share_locks[locking.LEVEL_NODE] = 1
1479

    
1480
  def BuildHooksEnv(self):
1481
    """Build hooks env.
1482

1483
    """
1484
    env = {
1485
      "OP_TARGET": self.cfg.GetClusterName(),
1486
      "NEW_VG_NAME": self.op.vg_name,
1487
      }
1488
    mn = self.cfg.GetMasterNode()
1489
    return env, [mn], [mn]
1490

    
1491
  def CheckPrereq(self):
1492
    """Check prerequisites.
1493

1494
    This checks whether the given params don't conflict and
1495
    if the given volume group is valid.
1496

1497
    """
1498
    if self.op.vg_name is not None and not self.op.vg_name:
1499
      instances = self.cfg.GetAllInstancesInfo().values()
1500
      for inst in instances:
1501
        for disk in inst.disks:
1502
          if _RecursiveCheckIfLVMBased(disk):
1503
            raise errors.OpPrereqError("Cannot disable lvm storage while"
1504
                                       " lvm-based instances exist")
1505

    
1506
    node_list = self.acquired_locks[locking.LEVEL_NODE]
1507

    
1508
    # if vg_name not None, checks given volume group on all nodes
1509
    if self.op.vg_name:
1510
      vglist = self.rpc.call_vg_list(node_list)
1511
      for node in node_list:
1512
        msg = vglist[node].RemoteFailMsg()
1513
        if msg:
1514
          # ignoring down node
1515
          self.LogWarning("Error while gathering data on node %s"
1516
                          " (ignoring node): %s", node, msg)
1517
          continue
1518
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
1519
                                              self.op.vg_name,
1520
                                              constants.MIN_VG_SIZE)
1521
        if vgstatus:
1522
          raise errors.OpPrereqError("Error on node '%s': %s" %
1523
                                     (node, vgstatus))
1524

    
1525
    self.cluster = cluster = self.cfg.GetClusterInfo()
1526
    # validate params changes
1527
    if self.op.beparams:
1528
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
1529
      self.new_beparams = objects.FillDict(
1530
        cluster.beparams[constants.PP_DEFAULT], self.op.beparams)
1531

    
1532
    if self.op.nicparams:
1533
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
1534
      self.new_nicparams = objects.FillDict(
1535
        cluster.nicparams[constants.PP_DEFAULT], self.op.nicparams)
1536
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
1537

    
1538
    # hypervisor list/parameters
1539
    self.new_hvparams = objects.FillDict(cluster.hvparams, {})
1540
    if self.op.hvparams:
1541
      if not isinstance(self.op.hvparams, dict):
1542
        raise errors.OpPrereqError("Invalid 'hvparams' parameter on input")
1543
      for hv_name, hv_dict in self.op.hvparams.items():
1544
        if hv_name not in self.new_hvparams:
1545
          self.new_hvparams[hv_name] = hv_dict
1546
        else:
1547
          self.new_hvparams[hv_name].update(hv_dict)
1548

    
1549
    if self.op.enabled_hypervisors is not None:
1550
      self.hv_list = self.op.enabled_hypervisors
1551
    else:
1552
      self.hv_list = cluster.enabled_hypervisors
1553

    
1554
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
1555
      # either the enabled list has changed, or the parameters have, validate
1556
      for hv_name, hv_params in self.new_hvparams.items():
1557
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
1558
            (self.op.enabled_hypervisors and
1559
             hv_name in self.op.enabled_hypervisors)):
1560
          # either this is a new hypervisor, or its parameters have changed
1561
          hv_class = hypervisor.GetHypervisor(hv_name)
1562
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1563
          hv_class.CheckParameterSyntax(hv_params)
1564
          _CheckHVParams(self, node_list, hv_name, hv_params)
1565

    
1566
  def Exec(self, feedback_fn):
1567
    """Change the parameters of the cluster.
1568

1569
    """
1570
    if self.op.vg_name is not None:
1571
      new_volume = self.op.vg_name
1572
      if not new_volume:
1573
        new_volume = None
1574
      if new_volume != self.cfg.GetVGName():
1575
        self.cfg.SetVGName(new_volume)
1576
      else:
1577
        feedback_fn("Cluster LVM configuration already in desired"
1578
                    " state, not changing")
1579
    if self.op.hvparams:
1580
      self.cluster.hvparams = self.new_hvparams
1581
    if self.op.enabled_hypervisors is not None:
1582
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
1583
    if self.op.beparams:
1584
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
1585
    if self.op.nicparams:
1586
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
1587

    
1588
    if self.op.candidate_pool_size is not None:
1589
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
1590

    
1591
    self.cfg.Update(self.cluster)
1592

    
1593
    # we want to update nodes after the cluster so that if any errors
1594
    # happen, we have recorded and saved the cluster info
1595
    if self.op.candidate_pool_size is not None:
1596
      _AdjustCandidatePool(self)
1597

    
1598

    
1599
def _RedistributeAncillaryFiles(lu, additional_nodes=None):
1600
  """Distribute additional files which are part of the cluster configuration.
1601

1602
  ConfigWriter takes care of distributing the config and ssconf files, but
1603
  there are more files which should be distributed to all nodes. This function
1604
  makes sure those are copied.
1605

1606
  @param lu: calling logical unit
1607
  @param additional_nodes: list of nodes not in the config to distribute to
1608

1609
  """
1610
  # 1. Gather target nodes
1611
  myself = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
1612
  dist_nodes = lu.cfg.GetNodeList()
1613
  if additional_nodes is not None:
1614
    dist_nodes.extend(additional_nodes)
1615
  if myself.name in dist_nodes:
1616
    dist_nodes.remove(myself.name)
1617
  # 2. Gather files to distribute
1618
  dist_files = set([constants.ETC_HOSTS,
1619
                    constants.SSH_KNOWN_HOSTS_FILE,
1620
                    constants.RAPI_CERT_FILE,
1621
                    constants.RAPI_USERS_FILE,
1622
                   ])
1623

    
1624
  enabled_hypervisors = lu.cfg.GetClusterInfo().enabled_hypervisors
1625
  for hv_name in enabled_hypervisors:
1626
    hv_class = hypervisor.GetHypervisor(hv_name)
1627
    dist_files.update(hv_class.GetAncillaryFiles())
1628

    
1629
  # 3. Perform the files upload
1630
  for fname in dist_files:
1631
    if os.path.exists(fname):
1632
      result = lu.rpc.call_upload_file(dist_nodes, fname)
1633
      for to_node, to_result in result.items():
1634
         msg = to_result.RemoteFailMsg()
1635
         if msg:
1636
           msg = ("Copy of file %s to node %s failed: %s" %
1637
                   (fname, to_node, msg))
1638
           lu.proc.LogWarning(msg)
1639

    
1640

    
1641
class LURedistributeConfig(NoHooksLU):
1642
  """Force the redistribution of cluster configuration.
1643

1644
  This is a very simple LU.
1645

1646
  """
1647
  _OP_REQP = []
1648
  REQ_BGL = False
1649

    
1650
  def ExpandNames(self):
1651
    self.needed_locks = {
1652
      locking.LEVEL_NODE: locking.ALL_SET,
1653
    }
1654
    self.share_locks[locking.LEVEL_NODE] = 1
1655

    
1656
  def CheckPrereq(self):
1657
    """Check prerequisites.
1658

1659
    """
1660

    
1661
  def Exec(self, feedback_fn):
1662
    """Redistribute the configuration.
1663

1664
    """
1665
    self.cfg.Update(self.cfg.GetClusterInfo())
1666
    _RedistributeAncillaryFiles(self)
1667

    
1668

    
1669
def _WaitForSync(lu, instance, oneshot=False, unlock=False):
1670
  """Sleep and poll for an instance's disk to sync.
1671

1672
  """
1673
  if not instance.disks:
1674
    return True
1675

    
1676
  if not oneshot:
1677
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
1678

    
1679
  node = instance.primary_node
1680

    
1681
  for dev in instance.disks:
1682
    lu.cfg.SetDiskID(dev, node)
1683

    
1684
  retries = 0
1685
  while True:
1686
    max_time = 0
1687
    done = True
1688
    cumul_degraded = False
1689
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
1690
    msg = rstats.RemoteFailMsg()
1691
    if msg:
1692
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
1693
      retries += 1
1694
      if retries >= 10:
1695
        raise errors.RemoteError("Can't contact node %s for mirror data,"
1696
                                 " aborting." % node)
1697
      time.sleep(6)
1698
      continue
1699
    rstats = rstats.payload
1700
    retries = 0
1701
    for i, mstat in enumerate(rstats):
1702
      if mstat is None:
1703
        lu.LogWarning("Can't compute data for node %s/%s",
1704
                           node, instance.disks[i].iv_name)
1705
        continue
1706
      # we ignore the ldisk parameter
1707
      perc_done, est_time, is_degraded, _ = mstat
1708
      cumul_degraded = cumul_degraded or (is_degraded and perc_done is None)
1709
      if perc_done is not None:
1710
        done = False
1711
        if est_time is not None:
1712
          rem_time = "%d estimated seconds remaining" % est_time
1713
          max_time = est_time
1714
        else:
1715
          rem_time = "no time estimate"
1716
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
1717
                        (instance.disks[i].iv_name, perc_done, rem_time))
1718
    if done or oneshot:
1719
      break
1720

    
1721
    time.sleep(min(60, max_time))
1722

    
1723
  if done:
1724
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
1725
  return not cumul_degraded
1726

    
1727

    
1728
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
1729
  """Check that mirrors are not degraded.
1730

1731
  The ldisk parameter, if True, will change the test from the
1732
  is_degraded attribute (which represents overall non-ok status for
1733
  the device(s)) to the ldisk (representing the local storage status).
1734

1735
  """
1736
  lu.cfg.SetDiskID(dev, node)
1737
  if ldisk:
1738
    idx = 6
1739
  else:
1740
    idx = 5
1741

    
1742
  result = True
1743
  if on_primary or dev.AssembleOnSecondary():
1744
    rstats = lu.rpc.call_blockdev_find(node, dev)
1745
    msg = rstats.RemoteFailMsg()
1746
    if msg:
1747
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
1748
      result = False
1749
    elif not rstats.payload:
1750
      lu.LogWarning("Can't find disk on node %s", node)
1751
      result = False
1752
    else:
1753
      result = result and (not rstats.payload[idx])
1754
  if dev.children:
1755
    for child in dev.children:
1756
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
1757

    
1758
  return result
1759

    
1760

    
1761
class LUDiagnoseOS(NoHooksLU):
1762
  """Logical unit for OS diagnose/query.
1763

1764
  """
1765
  _OP_REQP = ["output_fields", "names"]
1766
  REQ_BGL = False
1767
  _FIELDS_STATIC = utils.FieldSet()
1768
  _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status")
1769

    
1770
  def ExpandNames(self):
1771
    if self.op.names:
1772
      raise errors.OpPrereqError("Selective OS query not supported")
1773

    
1774
    _CheckOutputFields(static=self._FIELDS_STATIC,
1775
                       dynamic=self._FIELDS_DYNAMIC,
1776
                       selected=self.op.output_fields)
1777

    
1778
    # Lock all nodes, in shared mode
1779
    # Temporary removal of locks, should be reverted later
1780
    # TODO: reintroduce locks when they are lighter-weight
1781
    self.needed_locks = {}
1782
    #self.share_locks[locking.LEVEL_NODE] = 1
1783
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
1784

    
1785
  def CheckPrereq(self):
1786
    """Check prerequisites.
1787

1788
    """
1789

    
1790
  @staticmethod
1791
  def _DiagnoseByOS(node_list, rlist):
1792
    """Remaps a per-node return list into an a per-os per-node dictionary
1793

1794
    @param node_list: a list with the names of all nodes
1795
    @param rlist: a map with node names as keys and OS objects as values
1796

1797
    @rtype: dict
1798
    @return: a dictionary with osnames as keys and as value another map, with
1799
        nodes as keys and list of OS objects as values, eg::
1800

1801
          {"debian-etch": {"node1": [<object>,...],
1802
                           "node2": [<object>,]}
1803
          }
1804

1805
    """
1806
    all_os = {}
1807
    # we build here the list of nodes that didn't fail the RPC (at RPC
1808
    # level), so that nodes with a non-responding node daemon don't
1809
    # make all OSes invalid
1810
    good_nodes = [node_name for node_name in rlist
1811
                  if not rlist[node_name].failed]
1812
    for node_name, nr in rlist.iteritems():
1813
      if nr.failed or not nr.data:
1814
        continue
1815
      for os_obj in nr.data:
1816
        if os_obj.name not in all_os:
1817
          # build a list of nodes for this os containing empty lists
1818
          # for each node in node_list
1819
          all_os[os_obj.name] = {}
1820
          for nname in good_nodes:
1821
            all_os[os_obj.name][nname] = []
1822
        all_os[os_obj.name][node_name].append(os_obj)
1823
    return all_os
1824

    
1825
  def Exec(self, feedback_fn):
1826
    """Compute the list of OSes.
1827

1828
    """
1829
    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
1830
    node_data = self.rpc.call_os_diagnose(valid_nodes)
1831
    if node_data == False:
1832
      raise errors.OpExecError("Can't gather the list of OSes")
1833
    pol = self._DiagnoseByOS(valid_nodes, node_data)
1834
    output = []
1835
    for os_name, os_data in pol.iteritems():
1836
      row = []
1837
      for field in self.op.output_fields:
1838
        if field == "name":
1839
          val = os_name
1840
        elif field == "valid":
1841
          val = utils.all([osl and osl[0] for osl in os_data.values()])
1842
        elif field == "node_status":
1843
          val = {}
1844
          for node_name, nos_list in os_data.iteritems():
1845
            val[node_name] = [(v.status, v.path) for v in nos_list]
1846
        else:
1847
          raise errors.ParameterError(field)
1848
        row.append(val)
1849
      output.append(row)
1850

    
1851
    return output
1852

    
1853

    
1854
class LURemoveNode(LogicalUnit):
1855
  """Logical unit for removing a node.
1856

1857
  """
1858
  HPATH = "node-remove"
1859
  HTYPE = constants.HTYPE_NODE
1860
  _OP_REQP = ["node_name"]
1861

    
1862
  def BuildHooksEnv(self):
1863
    """Build hooks env.
1864

1865
    This doesn't run on the target node in the pre phase as a failed
1866
    node would then be impossible to remove.
1867

1868
    """
1869
    env = {
1870
      "OP_TARGET": self.op.node_name,
1871
      "NODE_NAME": self.op.node_name,
1872
      }
1873
    all_nodes = self.cfg.GetNodeList()
1874
    all_nodes.remove(self.op.node_name)
1875
    return env, all_nodes, all_nodes
1876

    
1877
  def CheckPrereq(self):
1878
    """Check prerequisites.
1879

1880
    This checks:
1881
     - the node exists in the configuration
1882
     - it does not have primary or secondary instances
1883
     - it's not the master
1884

1885
    Any errors are signalled by raising errors.OpPrereqError.
1886

1887
    """
1888
    node = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.node_name))
1889
    if node is None:
1890
      raise errors.OpPrereqError, ("Node '%s' is unknown." % self.op.node_name)
1891

    
1892
    instance_list = self.cfg.GetInstanceList()
1893

    
1894
    masternode = self.cfg.GetMasterNode()
1895
    if node.name == masternode:
1896
      raise errors.OpPrereqError("Node is the master node,"
1897
                                 " you need to failover first.")
1898

    
1899
    for instance_name in instance_list:
1900
      instance = self.cfg.GetInstanceInfo(instance_name)
1901
      if node.name in instance.all_nodes:
1902
        raise errors.OpPrereqError("Instance %s is still running on the node,"
1903
                                   " please remove first." % instance_name)
1904
    self.op.node_name = node.name
1905
    self.node = node
1906

    
1907
  def Exec(self, feedback_fn):
1908
    """Removes the node from the cluster.
1909

1910
    """
1911
    node = self.node
1912
    logging.info("Stopping the node daemon and removing configs from node %s",
1913
                 node.name)
1914

    
1915
    self.context.RemoveNode(node.name)
1916

    
1917
    self.rpc.call_node_leave_cluster(node.name)
1918

    
1919
    # Promote nodes to master candidate as needed
1920
    _AdjustCandidatePool(self)
1921

    
1922

    
1923
class LUQueryNodes(NoHooksLU):
1924
  """Logical unit for querying nodes.
1925

1926
  """
1927
  _OP_REQP = ["output_fields", "names", "use_locking"]
1928
  REQ_BGL = False
1929
  _FIELDS_DYNAMIC = utils.FieldSet(
1930
    "dtotal", "dfree",
1931
    "mtotal", "mnode", "mfree",
1932
    "bootid",
1933
    "ctotal", "cnodes", "csockets",
1934
    )
1935

    
1936
  _FIELDS_STATIC = utils.FieldSet(
1937
    "name", "pinst_cnt", "sinst_cnt",
1938
    "pinst_list", "sinst_list",
1939
    "pip", "sip", "tags",
1940
    "serial_no",
1941
    "master_candidate",
1942
    "master",
1943
    "offline",
1944
    "drained",
1945
    )
1946

    
1947
  def ExpandNames(self):
1948
    _CheckOutputFields(static=self._FIELDS_STATIC,
1949
                       dynamic=self._FIELDS_DYNAMIC,
1950
                       selected=self.op.output_fields)
1951

    
1952
    self.needed_locks = {}
1953
    self.share_locks[locking.LEVEL_NODE] = 1
1954

    
1955
    if self.op.names:
1956
      self.wanted = _GetWantedNodes(self, self.op.names)
1957
    else:
1958
      self.wanted = locking.ALL_SET
1959

    
1960
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
1961
    self.do_locking = self.do_node_query and self.op.use_locking
1962
    if self.do_locking:
1963
      # if we don't request only static fields, we need to lock the nodes
1964
      self.needed_locks[locking.LEVEL_NODE] = self.wanted
1965

    
1966

    
1967
  def CheckPrereq(self):
1968
    """Check prerequisites.
1969

1970
    """
1971
    # The validation of the node list is done in the _GetWantedNodes,
1972
    # if non empty, and if empty, there's no validation to do
1973
    pass
1974

    
1975
  def Exec(self, feedback_fn):
1976
    """Computes the list of nodes and their attributes.
1977

1978
    """
1979
    all_info = self.cfg.GetAllNodesInfo()
1980
    if self.do_locking:
1981
      nodenames = self.acquired_locks[locking.LEVEL_NODE]
1982
    elif self.wanted != locking.ALL_SET:
1983
      nodenames = self.wanted
1984
      missing = set(nodenames).difference(all_info.keys())
1985
      if missing:
1986
        raise errors.OpExecError(
1987
          "Some nodes were removed before retrieving their data: %s" % missing)
1988
    else:
1989
      nodenames = all_info.keys()
1990

    
1991
    nodenames = utils.NiceSort(nodenames)
1992
    nodelist = [all_info[name] for name in nodenames]
1993

    
1994
    # begin data gathering
1995

    
1996
    if self.do_node_query:
1997
      live_data = {}
1998
      node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
1999
                                          self.cfg.GetHypervisorType())
2000
      for name in nodenames:
2001
        nodeinfo = node_data[name]
2002
        if not nodeinfo.failed and nodeinfo.data:
2003
          nodeinfo = nodeinfo.data
2004
          fn = utils.TryConvert
2005
          live_data[name] = {
2006
            "mtotal": fn(int, nodeinfo.get('memory_total', None)),
2007
            "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
2008
            "mfree": fn(int, nodeinfo.get('memory_free', None)),
2009
            "dtotal": fn(int, nodeinfo.get('vg_size', None)),
2010
            "dfree": fn(int, nodeinfo.get('vg_free', None)),
2011
            "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
2012
            "bootid": nodeinfo.get('bootid', None),
2013
            "cnodes": fn(int, nodeinfo.get('cpu_nodes', None)),
2014
            "csockets": fn(int, nodeinfo.get('cpu_sockets', None)),
2015
            }
2016
        else:
2017
          live_data[name] = {}
2018
    else:
2019
      live_data = dict.fromkeys(nodenames, {})
2020

    
2021
    node_to_primary = dict([(name, set()) for name in nodenames])
2022
    node_to_secondary = dict([(name, set()) for name in nodenames])
2023

    
2024
    inst_fields = frozenset(("pinst_cnt", "pinst_list",
2025
                             "sinst_cnt", "sinst_list"))
2026
    if inst_fields & frozenset(self.op.output_fields):
2027
      instancelist = self.cfg.GetInstanceList()
2028

    
2029
      for instance_name in instancelist:
2030
        inst = self.cfg.GetInstanceInfo(instance_name)
2031
        if inst.primary_node in node_to_primary:
2032
          node_to_primary[inst.primary_node].add(inst.name)
2033
        for secnode in inst.secondary_nodes:
2034
          if secnode in node_to_secondary:
2035
            node_to_secondary[secnode].add(inst.name)
2036

    
2037
    master_node = self.cfg.GetMasterNode()
2038

    
2039
    # end data gathering
2040

    
2041
    output = []
2042
    for node in nodelist:
2043
      node_output = []
2044
      for field in self.op.output_fields:
2045
        if field == "name":
2046
          val = node.name
2047
        elif field == "pinst_list":
2048
          val = list(node_to_primary[node.name])
2049
        elif field == "sinst_list":
2050
          val = list(node_to_secondary[node.name])
2051
        elif field == "pinst_cnt":
2052
          val = len(node_to_primary[node.name])
2053
        elif field == "sinst_cnt":
2054
          val = len(node_to_secondary[node.name])
2055
        elif field == "pip":
2056
          val = node.primary_ip
2057
        elif field == "sip":
2058
          val = node.secondary_ip
2059
        elif field == "tags":
2060
          val = list(node.GetTags())
2061
        elif field == "serial_no":
2062
          val = node.serial_no
2063
        elif field == "master_candidate":
2064
          val = node.master_candidate
2065
        elif field == "master":
2066
          val = node.name == master_node
2067
        elif field == "offline":
2068
          val = node.offline
2069
        elif field == "drained":
2070
          val = node.drained
2071
        elif self._FIELDS_DYNAMIC.Matches(field):
2072
          val = live_data[node.name].get(field, None)
2073
        else:
2074
          raise errors.ParameterError(field)
2075
        node_output.append(val)
2076
      output.append(node_output)
2077

    
2078
    return output
2079

    
2080

    
2081
class LUQueryNodeVolumes(NoHooksLU):
2082
  """Logical unit for getting volumes on node(s).
2083

2084
  """
2085
  _OP_REQP = ["nodes", "output_fields"]
2086
  REQ_BGL = False
2087
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
2088
  _FIELDS_STATIC = utils.FieldSet("node")
2089

    
2090
  def ExpandNames(self):
2091
    _CheckOutputFields(static=self._FIELDS_STATIC,
2092
                       dynamic=self._FIELDS_DYNAMIC,
2093
                       selected=self.op.output_fields)
2094

    
2095
    self.needed_locks = {}
2096
    self.share_locks[locking.LEVEL_NODE] = 1
2097
    if not self.op.nodes:
2098
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
2099
    else:
2100
      self.needed_locks[locking.LEVEL_NODE] = \
2101
        _GetWantedNodes(self, self.op.nodes)
2102

    
2103
  def CheckPrereq(self):
2104
    """Check prerequisites.
2105

2106
    This checks that the fields required are valid output fields.
2107

2108
    """
2109
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
2110

    
2111
  def Exec(self, feedback_fn):
2112
    """Computes the list of nodes and their attributes.
2113

2114
    """
2115
    nodenames = self.nodes
2116
    volumes = self.rpc.call_node_volumes(nodenames)
2117

    
2118
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
2119
             in self.cfg.GetInstanceList()]
2120

    
2121
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
2122

    
2123
    output = []
2124
    for node in nodenames:
2125
      if node not in volumes or volumes[node].failed or not volumes[node].data:
2126
        continue
2127

    
2128
      node_vols = volumes[node].data[:]
2129
      node_vols.sort(key=lambda vol: vol['dev'])
2130

    
2131
      for vol in node_vols:
2132
        node_output = []
2133
        for field in self.op.output_fields:
2134
          if field == "node":
2135
            val = node
2136
          elif field == "phys":
2137
            val = vol['dev']
2138
          elif field == "vg":
2139
            val = vol['vg']
2140
          elif field == "name":
2141
            val = vol['name']
2142
          elif field == "size":
2143
            val = int(float(vol['size']))
2144
          elif field == "instance":
2145
            for inst in ilist:
2146
              if node not in lv_by_node[inst]:
2147
                continue
2148
              if vol['name'] in lv_by_node[inst][node]:
2149
                val = inst.name
2150
                break
2151
            else:
2152
              val = '-'
2153
          else:
2154
            raise errors.ParameterError(field)
2155
          node_output.append(str(val))
2156

    
2157
        output.append(node_output)
2158

    
2159
    return output
2160

    
2161

    
2162
class LUAddNode(LogicalUnit):
2163
  """Logical unit for adding node to the cluster.
2164

2165
  """
2166
  HPATH = "node-add"
2167
  HTYPE = constants.HTYPE_NODE
2168
  _OP_REQP = ["node_name"]
2169

    
2170
  def BuildHooksEnv(self):
2171
    """Build hooks env.
2172

2173
    This will run on all nodes before, and on all nodes + the new node after.
2174

2175
    """
2176
    env = {
2177
      "OP_TARGET": self.op.node_name,
2178
      "NODE_NAME": self.op.node_name,
2179
      "NODE_PIP": self.op.primary_ip,
2180
      "NODE_SIP": self.op.secondary_ip,
2181
      }
2182
    nodes_0 = self.cfg.GetNodeList()
2183
    nodes_1 = nodes_0 + [self.op.node_name, ]
2184
    return env, nodes_0, nodes_1
2185

    
2186
  def CheckPrereq(self):
2187
    """Check prerequisites.
2188

2189
    This checks:
2190
     - the new node is not already in the config
2191
     - it is resolvable
2192
     - its parameters (single/dual homed) matches the cluster
2193

2194
    Any errors are signalled by raising errors.OpPrereqError.
2195

2196
    """
2197
    node_name = self.op.node_name
2198
    cfg = self.cfg
2199

    
2200
    dns_data = utils.HostInfo(node_name)
2201

    
2202
    node = dns_data.name
2203
    primary_ip = self.op.primary_ip = dns_data.ip
2204
    secondary_ip = getattr(self.op, "secondary_ip", None)
2205
    if secondary_ip is None:
2206
      secondary_ip = primary_ip
2207
    if not utils.IsValidIP(secondary_ip):
2208
      raise errors.OpPrereqError("Invalid secondary IP given")
2209
    self.op.secondary_ip = secondary_ip
2210

    
2211
    node_list = cfg.GetNodeList()
2212
    if not self.op.readd and node in node_list:
2213
      raise errors.OpPrereqError("Node %s is already in the configuration" %
2214
                                 node)
2215
    elif self.op.readd and node not in node_list:
2216
      raise errors.OpPrereqError("Node %s is not in the configuration" % node)
2217

    
2218
    for existing_node_name in node_list:
2219
      existing_node = cfg.GetNodeInfo(existing_node_name)
2220

    
2221
      if self.op.readd and node == existing_node_name:
2222
        if (existing_node.primary_ip != primary_ip or
2223
            existing_node.secondary_ip != secondary_ip):
2224
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
2225
                                     " address configuration as before")
2226
        continue
2227

    
2228
      if (existing_node.primary_ip == primary_ip or
2229
          existing_node.secondary_ip == primary_ip or
2230
          existing_node.primary_ip == secondary_ip or
2231
          existing_node.secondary_ip == secondary_ip):
2232
        raise errors.OpPrereqError("New node ip address(es) conflict with"
2233
                                   " existing node %s" % existing_node.name)
2234

    
2235
    # check that the type of the node (single versus dual homed) is the
2236
    # same as for the master
2237
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
2238
    master_singlehomed = myself.secondary_ip == myself.primary_ip
2239
    newbie_singlehomed = secondary_ip == primary_ip
2240
    if master_singlehomed != newbie_singlehomed:
2241
      if master_singlehomed:
2242
        raise errors.OpPrereqError("The master has no private ip but the"
2243
                                   " new node has one")
2244
      else:
2245
        raise errors.OpPrereqError("The master has a private ip but the"
2246
                                   " new node doesn't have one")
2247

    
2248
    # checks reachablity
2249
    if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
2250
      raise errors.OpPrereqError("Node not reachable by ping")
2251

    
2252
    if not newbie_singlehomed:
2253
      # check reachability from my secondary ip to newbie's secondary ip
2254
      if not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
2255
                           source=myself.secondary_ip):
2256
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
2257
                                   " based ping to noded port")
2258

    
2259
    cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2260
    mc_now, _ = self.cfg.GetMasterCandidateStats()
2261
    master_candidate = mc_now < cp_size
2262

    
2263
    self.new_node = objects.Node(name=node,
2264
                                 primary_ip=primary_ip,
2265
                                 secondary_ip=secondary_ip,
2266
                                 master_candidate=master_candidate,
2267
                                 offline=False, drained=False)
2268

    
2269
  def Exec(self, feedback_fn):
2270
    """Adds the new node to the cluster.
2271

2272
    """
2273
    new_node = self.new_node
2274
    node = new_node.name
2275

    
2276
    # check connectivity
2277
    result = self.rpc.call_version([node])[node]
2278
    result.Raise()
2279
    if result.data:
2280
      if constants.PROTOCOL_VERSION == result.data:
2281
        logging.info("Communication to node %s fine, sw version %s match",
2282
                     node, result.data)
2283
      else:
2284
        raise errors.OpExecError("Version mismatch master version %s,"
2285
                                 " node version %s" %
2286
                                 (constants.PROTOCOL_VERSION, result.data))
2287
    else:
2288
      raise errors.OpExecError("Cannot get version from the new node")
2289

    
2290
    # setup ssh on node
2291
    logging.info("Copy ssh key to node %s", node)
2292
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
2293
    keyarray = []
2294
    keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
2295
                constants.SSH_HOST_RSA_PRIV, constants.SSH_HOST_RSA_PUB,
2296
                priv_key, pub_key]
2297

    
2298
    for i in keyfiles:
2299
      f = open(i, 'r')
2300
      try:
2301
        keyarray.append(f.read())
2302
      finally:
2303
        f.close()
2304

    
2305
    result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
2306
                                    keyarray[2],
2307
                                    keyarray[3], keyarray[4], keyarray[5])
2308

    
2309
    msg = result.RemoteFailMsg()
2310
    if msg:
2311
      raise errors.OpExecError("Cannot transfer ssh keys to the"
2312
                               " new node: %s" % msg)
2313

    
2314
    # Add node to our /etc/hosts, and add key to known_hosts
2315
    if self.cfg.GetClusterInfo().modify_etc_hosts:
2316
      utils.AddHostToEtcHosts(new_node.name)
2317

    
2318
    if new_node.secondary_ip != new_node.primary_ip:
2319
      result = self.rpc.call_node_has_ip_address(new_node.name,
2320
                                                 new_node.secondary_ip)
2321
      if result.failed or not result.data:
2322
        raise errors.OpExecError("Node claims it doesn't have the secondary ip"
2323
                                 " you gave (%s). Please fix and re-run this"
2324
                                 " command." % new_node.secondary_ip)
2325

    
2326
    node_verify_list = [self.cfg.GetMasterNode()]
2327
    node_verify_param = {
2328
      'nodelist': [node],
2329
      # TODO: do a node-net-test as well?
2330
    }
2331

    
2332
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
2333
                                       self.cfg.GetClusterName())
2334
    for verifier in node_verify_list:
2335
      if result[verifier].failed or not result[verifier].data:
2336
        raise errors.OpExecError("Cannot communicate with %s's node daemon"
2337
                                 " for remote verification" % verifier)
2338
      if result[verifier].data['nodelist']:
2339
        for failed in result[verifier].data['nodelist']:
2340
          feedback_fn("ssh/hostname verification failed %s -> %s" %
2341
                      (verifier, result[verifier].data['nodelist'][failed]))
2342
        raise errors.OpExecError("ssh/hostname verification failed.")
2343

    
2344
    if self.op.readd:
2345
      _RedistributeAncillaryFiles(self)
2346
      self.context.ReaddNode(new_node)
2347
    else:
2348
      _RedistributeAncillaryFiles(self, additional_nodes=[node])
2349
      self.context.AddNode(new_node)
2350

    
2351

    
2352
class LUSetNodeParams(LogicalUnit):
2353
  """Modifies the parameters of a node.
2354

2355
  """
2356
  HPATH = "node-modify"
2357
  HTYPE = constants.HTYPE_NODE
2358
  _OP_REQP = ["node_name"]
2359
  REQ_BGL = False
2360

    
2361
  def CheckArguments(self):
2362
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2363
    if node_name is None:
2364
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2365
    self.op.node_name = node_name
2366
    _CheckBooleanOpField(self.op, 'master_candidate')
2367
    _CheckBooleanOpField(self.op, 'offline')
2368
    _CheckBooleanOpField(self.op, 'drained')
2369
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained]
2370
    if all_mods.count(None) == 3:
2371
      raise errors.OpPrereqError("Please pass at least one modification")
2372
    if all_mods.count(True) > 1:
2373
      raise errors.OpPrereqError("Can't set the node into more than one"
2374
                                 " state at the same time")
2375

    
2376
  def ExpandNames(self):
2377
    self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
2378

    
2379
  def BuildHooksEnv(self):
2380
    """Build hooks env.
2381

2382
    This runs on the master node.
2383

2384
    """
2385
    env = {
2386
      "OP_TARGET": self.op.node_name,
2387
      "MASTER_CANDIDATE": str(self.op.master_candidate),
2388
      "OFFLINE": str(self.op.offline),
2389
      "DRAINED": str(self.op.drained),
2390
      }
2391
    nl = [self.cfg.GetMasterNode(),
2392
          self.op.node_name]
2393
    return env, nl, nl
2394

    
2395
  def CheckPrereq(self):
2396
    """Check prerequisites.
2397

2398
    This only checks the instance list against the existing names.
2399

2400
    """
2401
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
2402

    
2403
    if ((self.op.master_candidate == False or self.op.offline == True or
2404
         self.op.drained == True) and node.master_candidate):
2405
      # we will demote the node from master_candidate
2406
      if self.op.node_name == self.cfg.GetMasterNode():
2407
        raise errors.OpPrereqError("The master node has to be a"
2408
                                   " master candidate, online and not drained")
2409
      cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2410
      num_candidates, _ = self.cfg.GetMasterCandidateStats()
2411
      if num_candidates <= cp_size:
2412
        msg = ("Not enough master candidates (desired"
2413
               " %d, new value will be %d)" % (cp_size, num_candidates-1))
2414
        if self.op.force:
2415
          self.LogWarning(msg)
2416
        else:
2417
          raise errors.OpPrereqError(msg)
2418

    
2419
    if (self.op.master_candidate == True and
2420
        ((node.offline and not self.op.offline == False) or
2421
         (node.drained and not self.op.drained == False))):
2422
      raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
2423
                                 " to master_candidate" % node.name)
2424

    
2425
    return
2426

    
2427
  def Exec(self, feedback_fn):
2428
    """Modifies a node.
2429

2430
    """
2431
    node = self.node
2432

    
2433
    result = []
2434
    changed_mc = False
2435

    
2436
    if self.op.offline is not None:
2437
      node.offline = self.op.offline
2438
      result.append(("offline", str(self.op.offline)))
2439
      if self.op.offline == True:
2440
        if node.master_candidate:
2441
          node.master_candidate = False
2442
          changed_mc = True
2443
          result.append(("master_candidate", "auto-demotion due to offline"))
2444
        if node.drained:
2445
          node.drained = False
2446
          result.append(("drained", "clear drained status due to offline"))
2447

    
2448
    if self.op.master_candidate is not None:
2449
      node.master_candidate = self.op.master_candidate
2450
      changed_mc = True
2451
      result.append(("master_candidate", str(self.op.master_candidate)))
2452
      if self.op.master_candidate == False:
2453
        rrc = self.rpc.call_node_demote_from_mc(node.name)
2454
        msg = rrc.RemoteFailMsg()
2455
        if msg:
2456
          self.LogWarning("Node failed to demote itself: %s" % msg)
2457

    
2458
    if self.op.drained is not None:
2459
      node.drained = self.op.drained
2460
      result.append(("drained", str(self.op.drained)))
2461
      if self.op.drained == True:
2462
        if node.master_candidate:
2463
          node.master_candidate = False
2464
          changed_mc = True
2465
          result.append(("master_candidate", "auto-demotion due to drain"))
2466
        if node.offline:
2467
          node.offline = False
2468
          result.append(("offline", "clear offline status due to drain"))
2469

    
2470
    # this will trigger configuration file update, if needed
2471
    self.cfg.Update(node)
2472
    # this will trigger job queue propagation or cleanup
2473
    if changed_mc:
2474
      self.context.ReaddNode(node)
2475

    
2476
    return result
2477

    
2478

    
2479
class LUPowercycleNode(NoHooksLU):
2480
  """Powercycles a node.
2481

2482
  """
2483
  _OP_REQP = ["node_name", "force"]
2484
  REQ_BGL = False
2485

    
2486
  def CheckArguments(self):
2487
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2488
    if node_name is None:
2489
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2490
    self.op.node_name = node_name
2491
    if node_name == self.cfg.GetMasterNode() and not self.op.force:
2492
      raise errors.OpPrereqError("The node is the master and the force"
2493
                                 " parameter was not set")
2494

    
2495
  def ExpandNames(self):
2496
    """Locking for PowercycleNode.
2497

2498
    This is a last-resource option and shouldn't block on other
2499
    jobs. Therefore, we grab no locks.
2500

2501
    """
2502
    self.needed_locks = {}
2503

    
2504
  def CheckPrereq(self):
2505
    """Check prerequisites.
2506

2507
    This LU has no prereqs.
2508

2509
    """
2510
    pass
2511

    
2512
  def Exec(self, feedback_fn):
2513
    """Reboots a node.
2514

2515
    """
2516
    result = self.rpc.call_node_powercycle(self.op.node_name,
2517
                                           self.cfg.GetHypervisorType())
2518
    msg = result.RemoteFailMsg()
2519
    if msg:
2520
      raise errors.OpExecError("Failed to schedule the reboot: %s" % msg)
2521
    return result.payload
2522

    
2523

    
2524
class LUQueryClusterInfo(NoHooksLU):
2525
  """Query cluster configuration.
2526

2527
  """
2528
  _OP_REQP = []
2529
  REQ_BGL = False
2530

    
2531
  def ExpandNames(self):
2532
    self.needed_locks = {}
2533

    
2534
  def CheckPrereq(self):
2535
    """No prerequsites needed for this LU.
2536

2537
    """
2538
    pass
2539

    
2540
  def Exec(self, feedback_fn):
2541
    """Return cluster config.
2542

2543
    """
2544
    cluster = self.cfg.GetClusterInfo()
2545
    result = {
2546
      "software_version": constants.RELEASE_VERSION,
2547
      "protocol_version": constants.PROTOCOL_VERSION,
2548
      "config_version": constants.CONFIG_VERSION,
2549
      "os_api_version": constants.OS_API_VERSION,
2550
      "export_version": constants.EXPORT_VERSION,
2551
      "architecture": (platform.architecture()[0], platform.machine()),
2552
      "name": cluster.cluster_name,
2553
      "master": cluster.master_node,
2554
      "default_hypervisor": cluster.default_hypervisor,
2555
      "enabled_hypervisors": cluster.enabled_hypervisors,
2556
      "hvparams": dict([(hypervisor, cluster.hvparams[hypervisor])
2557
                        for hypervisor in cluster.enabled_hypervisors]),
2558
      "beparams": cluster.beparams,
2559
      "nicparams": cluster.nicparams,
2560
      "candidate_pool_size": cluster.candidate_pool_size,
2561
      "master_netdev": cluster.master_netdev,
2562
      "volume_group_name": cluster.volume_group_name,
2563
      "file_storage_dir": cluster.file_storage_dir,
2564
      }
2565

    
2566
    return result
2567

    
2568

    
2569
class LUQueryConfigValues(NoHooksLU):
2570
  """Return configuration values.
2571

2572
  """
2573
  _OP_REQP = []
2574
  REQ_BGL = False
2575
  _FIELDS_DYNAMIC = utils.FieldSet()
2576
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag")
2577

    
2578
  def ExpandNames(self):
2579
    self.needed_locks = {}
2580

    
2581
    _CheckOutputFields(static=self._FIELDS_STATIC,
2582
                       dynamic=self._FIELDS_DYNAMIC,
2583
                       selected=self.op.output_fields)
2584

    
2585
  def CheckPrereq(self):
2586
    """No prerequisites.
2587

2588
    """
2589
    pass
2590

    
2591
  def Exec(self, feedback_fn):
2592
    """Dump a representation of the cluster config to the standard output.
2593

2594
    """
2595
    values = []
2596
    for field in self.op.output_fields:
2597
      if field == "cluster_name":
2598
        entry = self.cfg.GetClusterName()
2599
      elif field == "master_node":
2600
        entry = self.cfg.GetMasterNode()
2601
      elif field == "drain_flag":
2602
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
2603
      else:
2604
        raise errors.ParameterError(field)
2605
      values.append(entry)
2606
    return values
2607

    
2608

    
2609
class LUActivateInstanceDisks(NoHooksLU):
2610
  """Bring up an instance's disks.
2611

2612
  """
2613
  _OP_REQP = ["instance_name"]
2614
  REQ_BGL = False
2615

    
2616
  def ExpandNames(self):
2617
    self._ExpandAndLockInstance()
2618
    self.needed_locks[locking.LEVEL_NODE] = []
2619
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2620

    
2621
  def DeclareLocks(self, level):
2622
    if level == locking.LEVEL_NODE:
2623
      self._LockInstancesNodes()
2624

    
2625
  def CheckPrereq(self):
2626
    """Check prerequisites.
2627

2628
    This checks that the instance is in the cluster.
2629

2630
    """
2631
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2632
    assert self.instance is not None, \
2633
      "Cannot retrieve locked instance %s" % self.op.instance_name
2634
    _CheckNodeOnline(self, self.instance.primary_node)
2635

    
2636
  def Exec(self, feedback_fn):
2637
    """Activate the disks.
2638

2639
    """
2640
    disks_ok, disks_info = _AssembleInstanceDisks(self, self.instance)
2641
    if not disks_ok:
2642
      raise errors.OpExecError("Cannot activate block devices")
2643

    
2644
    return disks_info
2645

    
2646

    
2647
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False):
2648
  """Prepare the block devices for an instance.
2649

2650
  This sets up the block devices on all nodes.
2651

2652
  @type lu: L{LogicalUnit}
2653
  @param lu: the logical unit on whose behalf we execute
2654
  @type instance: L{objects.Instance}
2655
  @param instance: the instance for whose disks we assemble
2656
  @type ignore_secondaries: boolean
2657
  @param ignore_secondaries: if true, errors on secondary nodes
2658
      won't result in an error return from the function
2659
  @return: False if the operation failed, otherwise a list of
2660
      (host, instance_visible_name, node_visible_name)
2661
      with the mapping from node devices to instance devices
2662

2663
  """
2664
  device_info = []
2665
  disks_ok = True
2666
  iname = instance.name
2667
  # With the two passes mechanism we try to reduce the window of
2668
  # opportunity for the race condition of switching DRBD to primary
2669
  # before handshaking occured, but we do not eliminate it
2670

    
2671
  # The proper fix would be to wait (with some limits) until the
2672
  # connection has been made and drbd transitions from WFConnection
2673
  # into any other network-connected state (Connected, SyncTarget,
2674
  # SyncSource, etc.)
2675

    
2676
  # 1st pass, assemble on all nodes in secondary mode
2677
  for inst_disk in instance.disks:
2678
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2679
      lu.cfg.SetDiskID(node_disk, node)
2680
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
2681
      msg = result.RemoteFailMsg()
2682
      if msg:
2683
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2684
                           " (is_primary=False, pass=1): %s",
2685
                           inst_disk.iv_name, node, msg)
2686
        if not ignore_secondaries:
2687
          disks_ok = False
2688

    
2689
  # FIXME: race condition on drbd migration to primary
2690

    
2691
  # 2nd pass, do only the primary node
2692
  for inst_disk in instance.disks:
2693
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2694
      if node != instance.primary_node:
2695
        continue
2696
      lu.cfg.SetDiskID(node_disk, node)
2697
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
2698
      msg = result.RemoteFailMsg()
2699
      if msg:
2700
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2701
                           " (is_primary=True, pass=2): %s",
2702
                           inst_disk.iv_name, node, msg)
2703
        disks_ok = False
2704
    device_info.append((instance.primary_node, inst_disk.iv_name,
2705
                        result.payload))
2706

    
2707
  # leave the disks configured for the primary node
2708
  # this is a workaround that would be fixed better by
2709
  # improving the logical/physical id handling
2710
  for disk in instance.disks:
2711
    lu.cfg.SetDiskID(disk, instance.primary_node)
2712

    
2713
  return disks_ok, device_info
2714

    
2715

    
2716
def _StartInstanceDisks(lu, instance, force):
2717
  """Start the disks of an instance.
2718

2719
  """
2720
  disks_ok, dummy = _AssembleInstanceDisks(lu, instance,
2721
                                           ignore_secondaries=force)
2722
  if not disks_ok:
2723
    _ShutdownInstanceDisks(lu, instance)
2724
    if force is not None and not force:
2725
      lu.proc.LogWarning("", hint="If the message above refers to a"
2726
                         " secondary node,"
2727
                         " you can retry the operation using '--force'.")
2728
    raise errors.OpExecError("Disk consistency error")
2729

    
2730

    
2731
class LUDeactivateInstanceDisks(NoHooksLU):
2732
  """Shutdown an instance's disks.
2733

2734
  """
2735
  _OP_REQP = ["instance_name"]
2736
  REQ_BGL = False
2737

    
2738
  def ExpandNames(self):
2739
    self._ExpandAndLockInstance()
2740
    self.needed_locks[locking.LEVEL_NODE] = []
2741
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2742

    
2743
  def DeclareLocks(self, level):
2744
    if level == locking.LEVEL_NODE:
2745
      self._LockInstancesNodes()
2746

    
2747
  def CheckPrereq(self):
2748
    """Check prerequisites.
2749

2750
    This checks that the instance is in the cluster.
2751

2752
    """
2753
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2754
    assert self.instance is not None, \
2755
      "Cannot retrieve locked instance %s" % self.op.instance_name
2756

    
2757
  def Exec(self, feedback_fn):
2758
    """Deactivate the disks
2759

2760
    """
2761
    instance = self.instance
2762
    _SafeShutdownInstanceDisks(self, instance)
2763

    
2764

    
2765
def _SafeShutdownInstanceDisks(lu, instance):
2766
  """Shutdown block devices of an instance.
2767

2768
  This function checks if an instance is running, before calling
2769
  _ShutdownInstanceDisks.
2770

2771
  """
2772
  ins_l = lu.rpc.call_instance_list([instance.primary_node],
2773
                                      [instance.hypervisor])
2774
  ins_l = ins_l[instance.primary_node]
2775
  if ins_l.failed or not isinstance(ins_l.data, list):
2776
    raise errors.OpExecError("Can't contact node '%s'" %
2777
                             instance.primary_node)
2778

    
2779
  if instance.name in ins_l.data:
2780
    raise errors.OpExecError("Instance is running, can't shutdown"
2781
                             " block devices.")
2782

    
2783
  _ShutdownInstanceDisks(lu, instance)
2784

    
2785

    
2786
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
2787
  """Shutdown block devices of an instance.
2788

2789
  This does the shutdown on all nodes of the instance.
2790

2791
  If the ignore_primary is false, errors on the primary node are
2792
  ignored.
2793

2794
  """
2795
  all_result = True
2796
  for disk in instance.disks:
2797
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
2798
      lu.cfg.SetDiskID(top_disk, node)
2799
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
2800
      msg = result.RemoteFailMsg()
2801
      if msg:
2802
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
2803
                      disk.iv_name, node, msg)
2804
        if not ignore_primary or node != instance.primary_node:
2805
          all_result = False
2806
  return all_result
2807

    
2808

    
2809
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
2810
  """Checks if a node has enough free memory.
2811

2812
  This function check if a given node has the needed amount of free
2813
  memory. In case the node has less memory or we cannot get the
2814
  information from the node, this function raise an OpPrereqError
2815
  exception.
2816

2817
  @type lu: C{LogicalUnit}
2818
  @param lu: a logical unit from which we get configuration data
2819
  @type node: C{str}
2820
  @param node: the node to check
2821
  @type reason: C{str}
2822
  @param reason: string to use in the error message
2823
  @type requested: C{int}
2824
  @param requested: the amount of memory in MiB to check for
2825
  @type hypervisor_name: C{str}
2826
  @param hypervisor_name: the hypervisor to ask for memory stats
2827
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
2828
      we cannot check the node
2829

2830
  """
2831
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
2832
  nodeinfo[node].Raise()
2833
  free_mem = nodeinfo[node].data.get('memory_free')
2834
  if not isinstance(free_mem, int):
2835
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
2836
                             " was '%s'" % (node, free_mem))
2837
  if requested > free_mem:
2838
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
2839
                             " needed %s MiB, available %s MiB" %
2840
                             (node, reason, requested, free_mem))
2841

    
2842

    
2843
class LUStartupInstance(LogicalUnit):
2844
  """Starts an instance.
2845

2846
  """
2847
  HPATH = "instance-start"
2848
  HTYPE = constants.HTYPE_INSTANCE
2849
  _OP_REQP = ["instance_name", "force"]
2850
  REQ_BGL = False
2851

    
2852
  def ExpandNames(self):
2853
    self._ExpandAndLockInstance()
2854

    
2855
  def BuildHooksEnv(self):
2856
    """Build hooks env.
2857

2858
    This runs on master, primary and secondary nodes of the instance.
2859

2860
    """
2861
    env = {
2862
      "FORCE": self.op.force,
2863
      }
2864
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2865
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2866
    return env, nl, nl
2867

    
2868
  def CheckPrereq(self):
2869
    """Check prerequisites.
2870

2871
    This checks that the instance is in the cluster.
2872

2873
    """
2874
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2875
    assert self.instance is not None, \
2876
      "Cannot retrieve locked instance %s" % self.op.instance_name
2877

    
2878
    # extra beparams
2879
    self.beparams = getattr(self.op, "beparams", {})
2880
    if self.beparams:
2881
      if not isinstance(self.beparams, dict):
2882
        raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
2883
                                   " dict" % (type(self.beparams), ))
2884
      # fill the beparams dict
2885
      utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
2886
      self.op.beparams = self.beparams
2887

    
2888
    # extra hvparams
2889
    self.hvparams = getattr(self.op, "hvparams", {})
2890
    if self.hvparams:
2891
      if not isinstance(self.hvparams, dict):
2892
        raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
2893
                                   " dict" % (type(self.hvparams), ))
2894

    
2895
      # check hypervisor parameter syntax (locally)
2896
      cluster = self.cfg.GetClusterInfo()
2897
      utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
2898
      filled_hvp = objects.FillDict(cluster.hvparams[instance.hypervisor],
2899
                                    instance.hvparams)
2900
      filled_hvp.update(self.hvparams)
2901
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
2902
      hv_type.CheckParameterSyntax(filled_hvp)
2903
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
2904
      self.op.hvparams = self.hvparams
2905

    
2906
    _CheckNodeOnline(self, instance.primary_node)
2907

    
2908
    bep = self.cfg.GetClusterInfo().FillBE(instance)
2909
    # check bridges existance
2910
    _CheckInstanceBridgesExist(self, instance)
2911

    
2912
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2913
                                              instance.name,
2914
                                              instance.hypervisor)
2915
    msg = remote_info.RemoteFailMsg()
2916
    if msg:
2917
      raise errors.OpPrereqError("Error checking node %s: %s" %
2918
                                 (instance.primary_node, msg))
2919
    if not remote_info.payload: # not running already
2920
      _CheckNodeFreeMemory(self, instance.primary_node,
2921
                           "starting instance %s" % instance.name,
2922
                           bep[constants.BE_MEMORY], instance.hypervisor)
2923

    
2924
  def Exec(self, feedback_fn):
2925
    """Start the instance.
2926

2927
    """
2928
    instance = self.instance
2929
    force = self.op.force
2930

    
2931
    self.cfg.MarkInstanceUp(instance.name)
2932

    
2933
    node_current = instance.primary_node
2934

    
2935
    _StartInstanceDisks(self, instance, force)
2936

    
2937
    result = self.rpc.call_instance_start(node_current, instance,
2938
                                          self.hvparams, self.beparams)
2939
    msg = result.RemoteFailMsg()
2940
    if msg:
2941
      _ShutdownInstanceDisks(self, instance)
2942
      raise errors.OpExecError("Could not start instance: %s" % msg)
2943

    
2944

    
2945
class LURebootInstance(LogicalUnit):
2946
  """Reboot an instance.
2947

2948
  """
2949
  HPATH = "instance-reboot"
2950
  HTYPE = constants.HTYPE_INSTANCE
2951
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
2952
  REQ_BGL = False
2953

    
2954
  def ExpandNames(self):
2955
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
2956
                                   constants.INSTANCE_REBOOT_HARD,
2957
                                   constants.INSTANCE_REBOOT_FULL]:
2958
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
2959
                                  (constants.INSTANCE_REBOOT_SOFT,
2960
                                   constants.INSTANCE_REBOOT_HARD,
2961
                                   constants.INSTANCE_REBOOT_FULL))
2962
    self._ExpandAndLockInstance()
2963

    
2964
  def BuildHooksEnv(self):
2965
    """Build hooks env.
2966

2967
    This runs on master, primary and secondary nodes of the instance.
2968

2969
    """
2970
    env = {
2971
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
2972
      "REBOOT_TYPE": self.op.reboot_type,
2973
      }
2974
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2975
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2976
    return env, nl, nl
2977

    
2978
  def CheckPrereq(self):
2979
    """Check prerequisites.
2980

2981
    This checks that the instance is in the cluster.
2982

2983
    """
2984
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2985
    assert self.instance is not None, \
2986
      "Cannot retrieve locked instance %s" % self.op.instance_name
2987

    
2988
    _CheckNodeOnline(self, instance.primary_node)
2989

    
2990
    # check bridges existance
2991
    _CheckInstanceBridgesExist(self, instance)
2992

    
2993
  def Exec(self, feedback_fn):
2994
    """Reboot the instance.
2995

2996
    """
2997
    instance = self.instance
2998
    ignore_secondaries = self.op.ignore_secondaries
2999
    reboot_type = self.op.reboot_type
3000

    
3001
    node_current = instance.primary_node
3002

    
3003
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
3004
                       constants.INSTANCE_REBOOT_HARD]:
3005
      for disk in instance.disks:
3006
        self.cfg.SetDiskID(disk, node_current)
3007
      result = self.rpc.call_instance_reboot(node_current, instance,
3008
                                             reboot_type)
3009
      msg = result.RemoteFailMsg()
3010
      if msg:
3011
        raise errors.OpExecError("Could not reboot instance: %s" % msg)
3012
    else:
3013
      result = self.rpc.call_instance_shutdown(node_current, instance)
3014
      msg = result.RemoteFailMsg()
3015
      if msg:
3016
        raise errors.OpExecError("Could not shutdown instance for"
3017
                                 " full reboot: %s" % msg)
3018
      _ShutdownInstanceDisks(self, instance)
3019
      _StartInstanceDisks(self, instance, ignore_secondaries)
3020
      result = self.rpc.call_instance_start(node_current, instance, None, None)
3021
      msg = result.RemoteFailMsg()
3022
      if msg:
3023
        _ShutdownInstanceDisks(self, instance)
3024
        raise errors.OpExecError("Could not start instance for"
3025
                                 " full reboot: %s" % msg)
3026

    
3027
    self.cfg.MarkInstanceUp(instance.name)
3028

    
3029

    
3030
class LUShutdownInstance(LogicalUnit):
3031
  """Shutdown an instance.
3032

3033
  """
3034
  HPATH = "instance-stop"
3035
  HTYPE = constants.HTYPE_INSTANCE
3036
  _OP_REQP = ["instance_name"]
3037
  REQ_BGL = False
3038

    
3039
  def ExpandNames(self):
3040
    self._ExpandAndLockInstance()
3041

    
3042
  def BuildHooksEnv(self):
3043
    """Build hooks env.
3044

3045
    This runs on master, primary and secondary nodes of the instance.
3046

3047
    """
3048
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3049
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3050
    return env, nl, nl
3051

    
3052
  def CheckPrereq(self):
3053
    """Check prerequisites.
3054

3055
    This checks that the instance is in the cluster.
3056

3057
    """
3058
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3059
    assert self.instance is not None, \
3060
      "Cannot retrieve locked instance %s" % self.op.instance_name
3061
    _CheckNodeOnline(self, self.instance.primary_node)
3062

    
3063
  def Exec(self, feedback_fn):
3064
    """Shutdown the instance.
3065

3066
    """
3067
    instance = self.instance
3068
    node_current = instance.primary_node
3069
    self.cfg.MarkInstanceDown(instance.name)
3070
    result = self.rpc.call_instance_shutdown(node_current, instance)
3071
    msg = result.RemoteFailMsg()
3072
    if msg:
3073
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
3074

    
3075
    _ShutdownInstanceDisks(self, instance)
3076

    
3077

    
3078
class LUReinstallInstance(LogicalUnit):
3079
  """Reinstall an instance.
3080

3081
  """
3082
  HPATH = "instance-reinstall"
3083
  HTYPE = constants.HTYPE_INSTANCE
3084
  _OP_REQP = ["instance_name"]
3085
  REQ_BGL = False
3086

    
3087
  def ExpandNames(self):
3088
    self._ExpandAndLockInstance()
3089

    
3090
  def BuildHooksEnv(self):
3091
    """Build hooks env.
3092

3093
    This runs on master, primary and secondary nodes of the instance.
3094

3095
    """
3096
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3097
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3098
    return env, nl, nl
3099

    
3100
  def CheckPrereq(self):
3101
    """Check prerequisites.
3102

3103
    This checks that the instance is in the cluster and is not running.
3104

3105
    """
3106
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3107
    assert instance is not None, \
3108
      "Cannot retrieve locked instance %s" % self.op.instance_name
3109
    _CheckNodeOnline(self, instance.primary_node)
3110

    
3111
    if instance.disk_template == constants.DT_DISKLESS:
3112
      raise errors.OpPrereqError("Instance '%s' has no disks" %
3113
                                 self.op.instance_name)
3114
    if instance.admin_up:
3115
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3116
                                 self.op.instance_name)
3117
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3118
                                              instance.name,
3119
                                              instance.hypervisor)
3120
    msg = remote_info.RemoteFailMsg()
3121
    if msg:
3122
      raise errors.OpPrereqError("Error checking node %s: %s" %
3123
                                 (instance.primary_node, msg))
3124
    if remote_info.payload:
3125
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3126
                                 (self.op.instance_name,
3127
                                  instance.primary_node))
3128

    
3129
    self.op.os_type = getattr(self.op, "os_type", None)
3130
    if self.op.os_type is not None:
3131
      # OS verification
3132
      pnode = self.cfg.GetNodeInfo(
3133
        self.cfg.ExpandNodeName(instance.primary_node))
3134
      if pnode is None:
3135
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
3136
                                   self.op.pnode)
3137
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
3138
      result.Raise()
3139
      if not isinstance(result.data, objects.OS):
3140
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
3141
                                   " primary node"  % self.op.os_type)
3142

    
3143
    self.instance = instance
3144

    
3145
  def Exec(self, feedback_fn):
3146
    """Reinstall the instance.
3147

3148
    """
3149
    inst = self.instance
3150

    
3151
    if self.op.os_type is not None:
3152
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
3153
      inst.os = self.op.os_type
3154
      self.cfg.Update(inst)
3155

    
3156
    _StartInstanceDisks(self, inst, None)
3157
    try:
3158
      feedback_fn("Running the instance OS create scripts...")
3159
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True)
3160
      msg = result.RemoteFailMsg()
3161
      if msg:
3162
        raise errors.OpExecError("Could not install OS for instance %s"
3163
                                 " on node %s: %s" %
3164
                                 (inst.name, inst.primary_node, msg))
3165
    finally:
3166
      _ShutdownInstanceDisks(self, inst)
3167

    
3168

    
3169
class LURenameInstance(LogicalUnit):
3170
  """Rename an instance.
3171

3172
  """
3173
  HPATH = "instance-rename"
3174
  HTYPE = constants.HTYPE_INSTANCE
3175
  _OP_REQP = ["instance_name", "new_name"]
3176

    
3177
  def BuildHooksEnv(self):
3178
    """Build hooks env.
3179

3180
    This runs on master, primary and secondary nodes of the instance.
3181

3182
    """
3183
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3184
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3185
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3186
    return env, nl, nl
3187

    
3188
  def CheckPrereq(self):
3189
    """Check prerequisites.
3190

3191
    This checks that the instance is in the cluster and is not running.
3192

3193
    """
3194
    instance = self.cfg.GetInstanceInfo(
3195
      self.cfg.ExpandInstanceName(self.op.instance_name))
3196
    if instance is None:
3197
      raise errors.OpPrereqError("Instance '%s' not known" %
3198
                                 self.op.instance_name)
3199
    _CheckNodeOnline(self, instance.primary_node)
3200

    
3201
    if instance.admin_up:
3202
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3203
                                 self.op.instance_name)
3204
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3205
                                              instance.name,
3206
                                              instance.hypervisor)
3207
    msg = remote_info.RemoteFailMsg()
3208
    if msg:
3209
      raise errors.OpPrereqError("Error checking node %s: %s" %
3210
                                 (instance.primary_node, msg))
3211
    if remote_info.payload:
3212
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3213
                                 (self.op.instance_name,
3214
                                  instance.primary_node))
3215
    self.instance = instance
3216

    
3217
    # new name verification
3218
    name_info = utils.HostInfo(self.op.new_name)
3219

    
3220
    self.op.new_name = new_name = name_info.name
3221
    instance_list = self.cfg.GetInstanceList()
3222
    if new_name in instance_list:
3223
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3224
                                 new_name)
3225

    
3226
    if not getattr(self.op, "ignore_ip", False):
3227
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3228
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3229
                                   (name_info.ip, new_name))
3230

    
3231

    
3232
  def Exec(self, feedback_fn):
3233
    """Reinstall the instance.
3234

3235
    """
3236
    inst = self.instance
3237
    old_name = inst.name
3238

    
3239
    if inst.disk_template == constants.DT_FILE:
3240
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3241

    
3242
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3243
    # Change the instance lock. This is definitely safe while we hold the BGL
3244
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3245
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3246

    
3247
    # re-read the instance from the configuration after rename
3248
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3249

    
3250
    if inst.disk_template == constants.DT_FILE:
3251
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3252
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3253
                                                     old_file_storage_dir,
3254
                                                     new_file_storage_dir)
3255
      result.Raise()
3256
      if not result.data:
3257
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3258
                                 " directory '%s' to '%s' (but the instance"
3259
                                 " has been renamed in Ganeti)" % (
3260
                                 inst.primary_node, old_file_storage_dir,
3261
                                 new_file_storage_dir))
3262

    
3263
      if not result.data[0]:
3264
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3265
                                 " (but the instance has been renamed in"
3266
                                 " Ganeti)" % (old_file_storage_dir,
3267
                                               new_file_storage_dir))
3268

    
3269
    _StartInstanceDisks(self, inst, None)
3270
    try:
3271
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3272
                                                 old_name)
3273
      msg = result.RemoteFailMsg()
3274
      if msg:
3275
        msg = ("Could not run OS rename script for instance %s on node %s"
3276
               " (but the instance has been renamed in Ganeti): %s" %
3277
               (inst.name, inst.primary_node, msg))
3278
        self.proc.LogWarning(msg)
3279
    finally:
3280
      _ShutdownInstanceDisks(self, inst)
3281

    
3282

    
3283
class LURemoveInstance(LogicalUnit):
3284
  """Remove an instance.
3285

3286
  """
3287
  HPATH = "instance-remove"
3288
  HTYPE = constants.HTYPE_INSTANCE
3289
  _OP_REQP = ["instance_name", "ignore_failures"]
3290
  REQ_BGL = False
3291

    
3292
  def ExpandNames(self):
3293
    self._ExpandAndLockInstance()
3294
    self.needed_locks[locking.LEVEL_NODE] = []
3295
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3296

    
3297
  def DeclareLocks(self, level):
3298
    if level == locking.LEVEL_NODE:
3299
      self._LockInstancesNodes()
3300

    
3301
  def BuildHooksEnv(self):
3302
    """Build hooks env.
3303

3304
    This runs on master, primary and secondary nodes of the instance.
3305

3306
    """
3307
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3308
    nl = [self.cfg.GetMasterNode()]
3309
    return env, nl, nl
3310

    
3311
  def CheckPrereq(self):
3312
    """Check prerequisites.
3313

3314
    This checks that the instance is in the cluster.
3315

3316
    """
3317
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3318
    assert self.instance is not None, \
3319
      "Cannot retrieve locked instance %s" % self.op.instance_name
3320

    
3321
  def Exec(self, feedback_fn):
3322
    """Remove the instance.
3323

3324
    """
3325
    instance = self.instance
3326
    logging.info("Shutting down instance %s on node %s",
3327
                 instance.name, instance.primary_node)
3328

    
3329
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3330
    msg = result.RemoteFailMsg()
3331
    if msg:
3332
      if self.op.ignore_failures:
3333
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3334
      else:
3335
        raise errors.OpExecError("Could not shutdown instance %s on"
3336
                                 " node %s: %s" %
3337
                                 (instance.name, instance.primary_node, msg))
3338

    
3339
    logging.info("Removing block devices for instance %s", instance.name)
3340

    
3341
    if not _RemoveDisks(self, instance):
3342
      if self.op.ignore_failures:
3343
        feedback_fn("Warning: can't remove instance's disks")
3344
      else:
3345
        raise errors.OpExecError("Can't remove instance's disks")
3346

    
3347
    logging.info("Removing instance %s out of cluster config", instance.name)
3348

    
3349
    self.cfg.RemoveInstance(instance.name)
3350
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3351

    
3352

    
3353
class LUQueryInstances(NoHooksLU):
3354
  """Logical unit for querying instances.
3355

3356
  """
3357
  _OP_REQP = ["output_fields", "names", "use_locking"]
3358
  REQ_BGL = False
3359
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3360
                                    "admin_state",
3361
                                    "disk_template", "ip", "mac", "bridge",
3362
                                    "sda_size", "sdb_size", "vcpus", "tags",
3363
                                    "network_port", "beparams",
3364
                                    r"(disk)\.(size)/([0-9]+)",
3365
                                    r"(disk)\.(sizes)", "disk_usage",
3366
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3367
                                    r"(nic)\.(macs|ips|bridges)",
3368
                                    r"(disk|nic)\.(count)",
3369
                                    "serial_no", "hypervisor", "hvparams",] +
3370
                                  ["hv/%s" % name
3371
                                   for name in constants.HVS_PARAMETERS] +
3372
                                  ["be/%s" % name
3373
                                   for name in constants.BES_PARAMETERS])
3374
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3375

    
3376

    
3377
  def ExpandNames(self):
3378
    _CheckOutputFields(static=self._FIELDS_STATIC,
3379
                       dynamic=self._FIELDS_DYNAMIC,
3380
                       selected=self.op.output_fields)
3381

    
3382
    self.needed_locks = {}
3383
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3384
    self.share_locks[locking.LEVEL_NODE] = 1
3385

    
3386
    if self.op.names:
3387
      self.wanted = _GetWantedInstances(self, self.op.names)
3388
    else:
3389
      self.wanted = locking.ALL_SET
3390

    
3391
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3392
    self.do_locking = self.do_node_query and self.op.use_locking
3393
    if self.do_locking:
3394
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3395
      self.needed_locks[locking.LEVEL_NODE] = []
3396
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3397

    
3398
  def DeclareLocks(self, level):
3399
    if level == locking.LEVEL_NODE and self.do_locking:
3400
      self._LockInstancesNodes()
3401

    
3402
  def CheckPrereq(self):
3403
    """Check prerequisites.
3404

3405
    """
3406
    pass
3407

    
3408
  def Exec(self, feedback_fn):
3409
    """Computes the list of nodes and their attributes.
3410

3411
    """
3412
    all_info = self.cfg.GetAllInstancesInfo()
3413
    if self.wanted == locking.ALL_SET:
3414
      # caller didn't specify instance names, so ordering is not important
3415
      if self.do_locking:
3416
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3417
      else:
3418
        instance_names = all_info.keys()
3419
      instance_names = utils.NiceSort(instance_names)
3420
    else:
3421
      # caller did specify names, so we must keep the ordering
3422
      if self.do_locking:
3423
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3424
      else:
3425
        tgt_set = all_info.keys()
3426
      missing = set(self.wanted).difference(tgt_set)
3427
      if missing:
3428
        raise errors.OpExecError("Some instances were removed before"
3429
                                 " retrieving their data: %s" % missing)
3430
      instance_names = self.wanted
3431

    
3432
    instance_list = [all_info[iname] for iname in instance_names]
3433

    
3434
    # begin data gathering
3435

    
3436
    nodes = frozenset([inst.primary_node for inst in instance_list])
3437
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3438

    
3439
    bad_nodes = []
3440
    off_nodes = []
3441
    if self.do_node_query:
3442
      live_data = {}
3443
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3444
      for name in nodes:
3445
        result = node_data[name]
3446
        if result.offline:
3447
          # offline nodes will be in both lists
3448
          off_nodes.append(name)
3449
        if result.failed or result.RemoteFailMsg():
3450
          bad_nodes.append(name)
3451
        else:
3452
          if result.payload:
3453
            live_data.update(result.payload)
3454
          # else no instance is alive
3455
    else:
3456
      live_data = dict([(name, {}) for name in instance_names])
3457

    
3458
    # end data gathering
3459

    
3460
    HVPREFIX = "hv/"
3461
    BEPREFIX = "be/"
3462
    output = []
3463
    for instance in instance_list:
3464
      iout = []
3465
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3466
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3467
      for field in self.op.output_fields:
3468
        st_match = self._FIELDS_STATIC.Matches(field)
3469
        if field == "name":
3470
          val = instance.name
3471
        elif field == "os":
3472
          val = instance.os
3473
        elif field == "pnode":
3474
          val = instance.primary_node
3475
        elif field == "snodes":
3476
          val = list(instance.secondary_nodes)
3477
        elif field == "admin_state":
3478
          val = instance.admin_up
3479
        elif field == "oper_state":
3480
          if instance.primary_node in bad_nodes:
3481
            val = None
3482
          else:
3483
            val = bool(live_data.get(instance.name))
3484
        elif field == "status":
3485
          if instance.primary_node in off_nodes:
3486
            val = "ERROR_nodeoffline"
3487
          elif instance.primary_node in bad_nodes:
3488
            val = "ERROR_nodedown"
3489
          else:
3490
            running = bool(live_data.get(instance.name))
3491
            if running:
3492
              if instance.admin_up:
3493
                val = "running"
3494
              else:
3495
                val = "ERROR_up"
3496
            else:
3497
              if instance.admin_up:
3498
                val = "ERROR_down"
3499
              else:
3500
                val = "ADMIN_down"
3501
        elif field == "oper_ram":
3502
          if instance.primary_node in bad_nodes:
3503
            val = None
3504
          elif instance.name in live_data:
3505
            val = live_data[instance.name].get("memory", "?")
3506
          else:
3507
            val = "-"
3508
        elif field == "disk_template":
3509
          val = instance.disk_template
3510
        elif field == "ip":
3511
          val = instance.nics[0].ip
3512
        elif field == "bridge":
3513
          val = instance.nics[0].bridge
3514
        elif field == "mac":
3515
          val = instance.nics[0].mac
3516
        elif field == "sda_size" or field == "sdb_size":
3517
          idx = ord(field[2]) - ord('a')
3518
          try:
3519
            val = instance.FindDisk(idx).size
3520
          except errors.OpPrereqError:
3521
            val = None
3522
        elif field == "disk_usage": # total disk usage per node
3523
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3524
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3525
        elif field == "tags":
3526
          val = list(instance.GetTags())
3527
        elif field == "serial_no":
3528
          val = instance.serial_no
3529
        elif field == "network_port":
3530
          val = instance.network_port
3531
        elif field == "hypervisor":
3532
          val = instance.hypervisor
3533
        elif field == "hvparams":
3534
          val = i_hv
3535
        elif (field.startswith(HVPREFIX) and
3536
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3537
          val = i_hv.get(field[len(HVPREFIX):], None)
3538
        elif field == "beparams":
3539
          val = i_be
3540
        elif (field.startswith(BEPREFIX) and
3541
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3542
          val = i_be.get(field[len(BEPREFIX):], None)
3543
        elif st_match and st_match.groups():
3544
          # matches a variable list
3545
          st_groups = st_match.groups()
3546
          if st_groups and st_groups[0] == "disk":
3547
            if st_groups[1] == "count":
3548
              val = len(instance.disks)
3549
            elif st_groups[1] == "sizes":
3550
              val = [disk.size for disk in instance.disks]
3551
            elif st_groups[1] == "size":
3552
              try:
3553
                val = instance.FindDisk(st_groups[2]).size
3554
              except errors.OpPrereqError:
3555
                val = None
3556
            else:
3557
              assert False, "Unhandled disk parameter"
3558
          elif st_groups[0] == "nic":
3559
            if st_groups[1] == "count":
3560
              val = len(instance.nics)
3561
            elif st_groups[1] == "macs":
3562
              val = [nic.mac for nic in instance.nics]
3563
            elif st_groups[1] == "ips":
3564
              val = [nic.ip for nic in instance.nics]
3565
            elif st_groups[1] == "bridges":
3566
              val = [nic.bridge for nic in instance.nics]
3567
            else:
3568
              # index-based item
3569
              nic_idx = int(st_groups[2])
3570
              if nic_idx >= len(instance.nics):
3571
                val = None
3572
              else:
3573
                if st_groups[1] == "mac":
3574
                  val = instance.nics[nic_idx].mac
3575
                elif st_groups[1] == "ip":
3576
                  val = instance.nics[nic_idx].ip
3577
                elif st_groups[1] == "bridge":
3578
                  val = instance.nics[nic_idx].bridge
3579
                else:
3580
                  assert False, "Unhandled NIC parameter"
3581
          else:
3582
            assert False, "Unhandled variable parameter"
3583
        else:
3584
          raise errors.ParameterError(field)
3585
        iout.append(val)
3586
      output.append(iout)
3587

    
3588
    return output
3589

    
3590

    
3591
class LUFailoverInstance(LogicalUnit):
3592
  """Failover an instance.
3593

3594
  """
3595
  HPATH = "instance-failover"
3596
  HTYPE = constants.HTYPE_INSTANCE
3597
  _OP_REQP = ["instance_name", "ignore_consistency"]
3598
  REQ_BGL = False
3599

    
3600
  def ExpandNames(self):
3601
    self._ExpandAndLockInstance()
3602
    self.needed_locks[locking.LEVEL_NODE] = []
3603
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3604

    
3605
  def DeclareLocks(self, level):
3606
    if level == locking.LEVEL_NODE:
3607
      self._LockInstancesNodes()
3608

    
3609
  def BuildHooksEnv(self):
3610
    """Build hooks env.
3611

3612
    This runs on master, primary and secondary nodes of the instance.
3613

3614
    """
3615
    env = {
3616
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3617
      }
3618
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3619
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3620
    return env, nl, nl
3621

    
3622
  def CheckPrereq(self):
3623
    """Check prerequisites.
3624

3625
    This checks that the instance is in the cluster.
3626

3627
    """
3628
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3629
    assert self.instance is not None, \
3630
      "Cannot retrieve locked instance %s" % self.op.instance_name
3631

    
3632
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3633
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3634
      raise errors.OpPrereqError("Instance's disk layout is not"
3635
                                 " network mirrored, cannot failover.")
3636

    
3637
    secondary_nodes = instance.secondary_nodes
3638
    if not secondary_nodes:
3639
      raise errors.ProgrammerError("no secondary node but using "
3640
                                   "a mirrored disk template")
3641

    
3642
    target_node = secondary_nodes[0]
3643
    _CheckNodeOnline(self, target_node)
3644
    _CheckNodeNotDrained(self, target_node)
3645
    # check memory requirements on the secondary node
3646
    _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3647
                         instance.name, bep[constants.BE_MEMORY],
3648
                         instance.hypervisor)
3649
    # check bridge existance
3650
    _CheckInstanceBridgesExist(self, instance, node=target_node)
3651

    
3652
  def Exec(self, feedback_fn):
3653
    """Failover an instance.
3654

3655
    The failover is done by shutting it down on its present node and
3656
    starting it on the secondary.
3657

3658
    """
3659
    instance = self.instance
3660

    
3661
    source_node = instance.primary_node
3662
    target_node = instance.secondary_nodes[0]
3663

    
3664
    feedback_fn("* checking disk consistency between source and target")
3665
    for dev in instance.disks:
3666
      # for drbd, these are drbd over lvm
3667
      if not _CheckDiskConsistency(self, dev, target_node, False):
3668
        if instance.admin_up and not self.op.ignore_consistency:
3669
          raise errors.OpExecError("Disk %s is degraded on target node,"
3670
                                   " aborting failover." % dev.iv_name)
3671

    
3672
    feedback_fn("* shutting down instance on source node")
3673
    logging.info("Shutting down instance %s on node %s",
3674
                 instance.name, source_node)
3675

    
3676
    result = self.rpc.call_instance_shutdown(source_node, instance)
3677
    msg = result.RemoteFailMsg()
3678
    if msg:
3679
      if self.op.ignore_consistency:
3680
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3681
                             " Proceeding anyway. Please make sure node"
3682
                             " %s is down. Error details: %s",
3683
                             instance.name, source_node, source_node, msg)
3684
      else:
3685
        raise errors.OpExecError("Could not shutdown instance %s on"
3686
                                 " node %s: %s" %
3687
                                 (instance.name, source_node, msg))
3688

    
3689
    feedback_fn("* deactivating the instance's disks on source node")
3690
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3691
      raise errors.OpExecError("Can't shut down the instance's disks.")
3692

    
3693
    instance.primary_node = target_node
3694
    # distribute new instance config to the other nodes
3695
    self.cfg.Update(instance)
3696

    
3697
    # Only start the instance if it's marked as up
3698
    if instance.admin_up:
3699
      feedback_fn("* activating the instance's disks on target node")
3700
      logging.info("Starting instance %s on node %s",
3701
                   instance.name, target_node)
3702

    
3703
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3704
                                               ignore_secondaries=True)
3705
      if not disks_ok:
3706
        _ShutdownInstanceDisks(self, instance)
3707
        raise errors.OpExecError("Can't activate the instance's disks")
3708

    
3709
      feedback_fn("* starting the instance on the target node")
3710
      result = self.rpc.call_instance_start(target_node, instance, None, None)
3711
      msg = result.RemoteFailMsg()
3712
      if msg:
3713
        _ShutdownInstanceDisks(self, instance)
3714
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3715
                                 (instance.name, target_node, msg))
3716

    
3717

    
3718
class LUMigrateInstance(LogicalUnit):
3719
  """Migrate an instance.
3720

3721
  This is migration without shutting down, compared to the failover,
3722
  which is done with shutdown.
3723

3724
  """
3725
  HPATH = "instance-migrate"
3726
  HTYPE = constants.HTYPE_INSTANCE
3727
  _OP_REQP = ["instance_name", "live", "cleanup"]
3728

    
3729
  REQ_BGL = False
3730

    
3731
  def ExpandNames(self):
3732
    self._ExpandAndLockInstance()
3733
    self.needed_locks[locking.LEVEL_NODE] = []
3734
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3735

    
3736
  def DeclareLocks(self, level):
3737
    if level == locking.LEVEL_NODE:
3738
      self._LockInstancesNodes()
3739

    
3740
  def BuildHooksEnv(self):
3741
    """Build hooks env.
3742

3743
    This runs on master, primary and secondary nodes of the instance.
3744

3745
    """
3746
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3747
    env["MIGRATE_LIVE"] = self.op.live
3748
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3749
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3750
    return env, nl, nl
3751

    
3752
  def CheckPrereq(self):
3753
    """Check prerequisites.
3754

3755
    This checks that the instance is in the cluster.
3756

3757
    """
3758
    instance = self.cfg.GetInstanceInfo(
3759
      self.cfg.ExpandInstanceName(self.op.instance_name))
3760
    if instance is None:
3761
      raise errors.OpPrereqError("Instance '%s' not known" %
3762
                                 self.op.instance_name)
3763

    
3764
    if instance.disk_template != constants.DT_DRBD8:
3765
      raise errors.OpPrereqError("Instance's disk layout is not"
3766
                                 " drbd8, cannot migrate.")
3767

    
3768
    secondary_nodes = instance.secondary_nodes
3769
    if not secondary_nodes:
3770
      raise errors.ConfigurationError("No secondary node but using"
3771
                                      " drbd8 disk template")
3772

    
3773
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3774

    
3775
    target_node = secondary_nodes[0]
3776
    # check memory requirements on the secondary node
3777
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3778
                         instance.name, i_be[constants.BE_MEMORY],
3779
                         instance.hypervisor)
3780

    
3781
    # check bridge existance
3782
    _CheckInstanceBridgesExist(self, instance, node=target_node)
3783

    
3784
    if not self.op.cleanup:
3785
      _CheckNodeNotDrained(self, target_node)
3786
      result = self.rpc.call_instance_migratable(instance.primary_node,
3787
                                                 instance)
3788
      msg = result.RemoteFailMsg()
3789
      if msg:
3790
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3791
                                   msg)
3792

    
3793
    self.instance = instance
3794

    
3795
  def _WaitUntilSync(self):
3796
    """Poll with custom rpc for disk sync.
3797

3798
    This uses our own step-based rpc call.
3799

3800
    """
3801
    self.feedback_fn("* wait until resync is done")
3802
    all_done = False
3803
    while not all_done:
3804
      all_done = True
3805
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3806
                                            self.nodes_ip,
3807
                                            self.instance.disks)
3808
      min_percent = 100
3809
      for node, nres in result.items():
3810
        msg = nres.RemoteFailMsg()
3811
        if msg:
3812
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3813
                                   (node, msg))
3814
        node_done, node_percent = nres.payload
3815
        all_done = all_done and node_done
3816
        if node_percent is not None:
3817
          min_percent = min(min_percent, node_percent)
3818
      if not all_done:
3819
        if min_percent < 100:
3820
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3821
        time.sleep(2)
3822

    
3823
  def _EnsureSecondary(self, node):
3824
    """Demote a node to secondary.
3825

3826
    """
3827
    self.feedback_fn("* switching node %s to secondary mode" % node)
3828

    
3829
    for dev in self.instance.disks:
3830
      self.cfg.SetDiskID(dev, node)
3831

    
3832
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3833
                                          self.instance.disks)
3834
    msg = result.RemoteFailMsg()
3835
    if msg:
3836
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3837
                               " error %s" % (node, msg))
3838

    
3839
  def _GoStandalone(self):
3840
    """Disconnect from the network.
3841

3842
    """
3843
    self.feedback_fn("* changing into standalone mode")
3844
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3845
                                               self.instance.disks)
3846
    for node, nres in result.items():
3847
      msg = nres.RemoteFailMsg()
3848
      if msg:
3849
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3850
                                 " error %s" % (node, msg))
3851

    
3852
  def _GoReconnect(self, multimaster):
3853
    """Reconnect to the network.
3854

3855
    """
3856
    if multimaster:
3857
      msg = "dual-master"
3858
    else:
3859
      msg = "single-master"
3860
    self.feedback_fn("* changing disks into %s mode" % msg)
3861
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3862
                                           self.instance.disks,
3863
                                           self.instance.name, multimaster)
3864
    for node, nres in result.items():
3865
      msg = nres.RemoteFailMsg()
3866
      if msg:
3867
        raise errors.OpExecError("Cannot change disks config on node %s,"
3868
                                 " error: %s" % (node, msg))
3869

    
3870
  def _ExecCleanup(self):
3871
    """Try to cleanup after a failed migration.
3872

3873
    The cleanup is done by:
3874
      - check that the instance is running only on one node
3875
        (and update the config if needed)
3876
      - change disks on its secondary node to secondary
3877
      - wait until disks are fully synchronized
3878
      - disconnect from the network
3879
      - change disks into single-master mode
3880
      - wait again until disks are fully synchronized
3881

3882
    """
3883
    instance = self.instance
3884
    target_node = self.target_node
3885
    source_node = self.source_node
3886

    
3887
    # check running on only one node
3888
    self.feedback_fn("* checking where the instance actually runs"
3889
                     " (if this hangs, the hypervisor might be in"
3890
                     " a bad state)")
3891
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3892
    for node, result in ins_l.items():
3893
      result.Raise()
3894
      if not isinstance(result.data, list):
3895
        raise errors.OpExecError("Can't contact node '%s'" % node)
3896

    
3897
    runningon_source = instance.name in ins_l[source_node].data
3898
    runningon_target = instance.name in ins_l[target_node].data
3899

    
3900
    if runningon_source and runningon_target:
3901
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3902
                               " or the hypervisor is confused. You will have"
3903
                               " to ensure manually that it runs only on one"
3904
                               " and restart this operation.")
3905

    
3906
    if not (runningon_source or runningon_target):
3907
      raise errors.OpExecError("Instance does not seem to be running at all."
3908
                               " In this case, it's safer to repair by"
3909
                               " running 'gnt-instance stop' to ensure disk"
3910
                               " shutdown, and then restarting it.")
3911

    
3912
    if runningon_target:
3913
      # the migration has actually succeeded, we need to update the config
3914
      self.feedback_fn("* instance running on secondary node (%s),"
3915
                       " updating config" % target_node)
3916
      instance.primary_node = target_node
3917
      self.cfg.Update(instance)
3918
      demoted_node = source_node
3919
    else:
3920
      self.feedback_fn("* instance confirmed to be running on its"
3921
                       " primary node (%s)" % source_node)
3922
      demoted_node = target_node
3923

    
3924
    self._EnsureSecondary(demoted_node)
3925
    try:
3926
      self._WaitUntilSync()
3927
    except errors.OpExecError:
3928
      # we ignore here errors, since if the device is standalone, it
3929
      # won't be able to sync
3930
      pass
3931
    self._GoStandalone()
3932
    self._GoReconnect(False)
3933
    self._WaitUntilSync()
3934

    
3935
    self.feedback_fn("* done")
3936

    
3937
  def _RevertDiskStatus(self):
3938
    """Try to revert the disk status after a failed migration.
3939

3940
    """
3941
    target_node = self.target_node
3942
    try:
3943
      self._EnsureSecondary(target_node)
3944
      self._GoStandalone()
3945
      self._GoReconnect(False)
3946
      self._WaitUntilSync()
3947
    except errors.OpExecError, err:
3948
      self.LogWarning("Migration failed and I can't reconnect the"
3949
                      " drives: error '%s'\n"
3950
                      "Please look and recover the instance status" %
3951
                      str(err))
3952

    
3953
  def _AbortMigration(self):
3954
    """Call the hypervisor code to abort a started migration.
3955

3956
    """
3957
    instance = self.instance
3958
    target_node = self.target_node
3959
    migration_info = self.migration_info
3960

    
3961
    abort_result = self.rpc.call_finalize_migration(target_node,
3962
                                                    instance,
3963
                                                    migration_info,
3964
                                                    False)
3965
    abort_msg = abort_result.RemoteFailMsg()
3966
    if abort_msg:
3967
      logging.error("Aborting migration failed on target node %s: %s" %
3968
                    (target_node, abort_msg))
3969
      # Don't raise an exception here, as we stil have to try to revert the
3970
      # disk status, even if this step failed.
3971

    
3972
  def _ExecMigration(self):
3973
    """Migrate an instance.
3974

3975
    The migrate is done by:
3976
      - change the disks into dual-master mode
3977
      - wait until disks are fully synchronized again
3978
      - migrate the instance
3979
      - change disks on the new secondary node (the old primary) to secondary
3980
      - wait until disks are fully synchronized
3981
      - change disks into single-master mode
3982

3983
    """
3984
    instance = self.instance
3985
    target_node = self.target_node
3986
    source_node = self.source_node
3987