Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 0623d351

History | View | Annotate | Download (253.2 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
    msg = result.RemoteFailMsg()
660
    if msg:
661
      raise errors.OpExecError("Could not disable the master role: %s" % msg)
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

    
1026
      if node_i.offline:
1027
        feedback_fn("* Skipping offline node %s" % (node,))
1028
        n_offline.append(node)
1029
        continue
1030

    
1031
      if node == master_node:
1032
        ntype = "master"
1033
      elif node_i.master_candidate:
1034
        ntype = "master candidate"
1035
      elif node_i.drained:
1036
        ntype = "drained"
1037
        n_drained.append(node)
1038
      else:
1039
        ntype = "regular"
1040
      feedback_fn("* Verifying node %s (%s)" % (node, ntype))
1041

    
1042
      msg = all_nvinfo[node].RemoteFailMsg()
1043
      if msg:
1044
        feedback_fn("  - ERROR: while contacting node %s: %s" % (node, msg))
1045
        bad = True
1046
        continue
1047

    
1048
      nresult = all_nvinfo[node].payload
1049
      node_drbd = {}
1050
      for minor, instance in all_drbd_map[node].items():
1051
        if instance not in instanceinfo:
1052
          feedback_fn("  - ERROR: ghost instance '%s' in temporary DRBD map" %
1053
                      instance)
1054
          # ghost instance should not be running, but otherwise we
1055
          # don't give double warnings (both ghost instance and
1056
          # unallocated minor in use)
1057
          node_drbd[minor] = (instance, False)
1058
        else:
1059
          instance = instanceinfo[instance]
1060
          node_drbd[minor] = (instance.name, instance.admin_up)
1061
      result = self._VerifyNode(node_i, file_names, local_checksums,
1062
                                nresult, feedback_fn, master_files,
1063
                                node_drbd, vg_name)
1064
      bad = bad or result
1065

    
1066
      lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
1067
      if vg_name is None:
1068
        node_volume[node] = {}
1069
      elif isinstance(lvdata, basestring):
1070
        feedback_fn("  - ERROR: LVM problem on node %s: %s" %
1071
                    (node, utils.SafeEncode(lvdata)))
1072
        bad = True
1073
        node_volume[node] = {}
1074
      elif not isinstance(lvdata, dict):
1075
        feedback_fn("  - ERROR: connection to %s failed (lvlist)" % (node,))
1076
        bad = True
1077
        continue
1078
      else:
1079
        node_volume[node] = lvdata
1080

    
1081
      # node_instance
1082
      idata = nresult.get(constants.NV_INSTANCELIST, None)
1083
      if not isinstance(idata, list):
1084
        feedback_fn("  - ERROR: connection to %s failed (instancelist)" %
1085
                    (node,))
1086
        bad = True
1087
        continue
1088

    
1089
      node_instance[node] = idata
1090

    
1091
      # node_info
1092
      nodeinfo = nresult.get(constants.NV_HVINFO, None)
1093
      if not isinstance(nodeinfo, dict):
1094
        feedback_fn("  - ERROR: connection to %s failed (hvinfo)" % (node,))
1095
        bad = True
1096
        continue
1097

    
1098
      try:
1099
        node_info[node] = {
1100
          "mfree": int(nodeinfo['memory_free']),
1101
          "pinst": [],
1102
          "sinst": [],
1103
          # dictionary holding all instances this node is secondary for,
1104
          # grouped by their primary node. Each key is a cluster node, and each
1105
          # value is a list of instances which have the key as primary and the
1106
          # current node as secondary.  this is handy to calculate N+1 memory
1107
          # availability if you can only failover from a primary to its
1108
          # secondary.
1109
          "sinst-by-pnode": {},
1110
        }
1111
        # FIXME: devise a free space model for file based instances as well
1112
        if vg_name is not None:
1113
          if (constants.NV_VGLIST not in nresult or
1114
              vg_name not in nresult[constants.NV_VGLIST]):
1115
            feedback_fn("  - ERROR: node %s didn't return data for the"
1116
                        " volume group '%s' - it is either missing or broken" %
1117
                        (node, vg_name))
1118
            bad = True
1119
            continue
1120
          node_info[node]["dfree"] = int(nresult[constants.NV_VGLIST][vg_name])
1121
      except (ValueError, KeyError):
1122
        feedback_fn("  - ERROR: invalid nodeinfo value returned"
1123
                    " from node %s" % (node,))
1124
        bad = True
1125
        continue
1126

    
1127
    node_vol_should = {}
1128

    
1129
    for instance in instancelist:
1130
      feedback_fn("* Verifying instance %s" % instance)
1131
      inst_config = instanceinfo[instance]
1132
      result =  self._VerifyInstance(instance, inst_config, node_volume,
1133
                                     node_instance, feedback_fn, n_offline)
1134
      bad = bad or result
1135
      inst_nodes_offline = []
1136

    
1137
      inst_config.MapLVsByNode(node_vol_should)
1138

    
1139
      instance_cfg[instance] = inst_config
1140

    
1141
      pnode = inst_config.primary_node
1142
      if pnode in node_info:
1143
        node_info[pnode]['pinst'].append(instance)
1144
      elif pnode not in n_offline:
1145
        feedback_fn("  - ERROR: instance %s, connection to primary node"
1146
                    " %s failed" % (instance, pnode))
1147
        bad = True
1148

    
1149
      if pnode in n_offline:
1150
        inst_nodes_offline.append(pnode)
1151

    
1152
      # If the instance is non-redundant we cannot survive losing its primary
1153
      # node, so we are not N+1 compliant. On the other hand we have no disk
1154
      # templates with more than one secondary so that situation is not well
1155
      # supported either.
1156
      # FIXME: does not support file-backed instances
1157
      if len(inst_config.secondary_nodes) == 0:
1158
        i_non_redundant.append(instance)
1159
      elif len(inst_config.secondary_nodes) > 1:
1160
        feedback_fn("  - WARNING: multiple secondaries for instance %s"
1161
                    % instance)
1162

    
1163
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
1164
        i_non_a_balanced.append(instance)
1165

    
1166
      for snode in inst_config.secondary_nodes:
1167
        if snode in node_info:
1168
          node_info[snode]['sinst'].append(instance)
1169
          if pnode not in node_info[snode]['sinst-by-pnode']:
1170
            node_info[snode]['sinst-by-pnode'][pnode] = []
1171
          node_info[snode]['sinst-by-pnode'][pnode].append(instance)
1172
        elif snode not in n_offline:
1173
          feedback_fn("  - ERROR: instance %s, connection to secondary node"
1174
                      " %s failed" % (instance, snode))
1175
          bad = True
1176
        if snode in n_offline:
1177
          inst_nodes_offline.append(snode)
1178

    
1179
      if inst_nodes_offline:
1180
        # warn that the instance lives on offline nodes, and set bad=True
1181
        feedback_fn("  - ERROR: instance lives on offline node(s) %s" %
1182
                    ", ".join(inst_nodes_offline))
1183
        bad = True
1184

    
1185
    feedback_fn("* Verifying orphan volumes")
1186
    result = self._VerifyOrphanVolumes(node_vol_should, node_volume,
1187
                                       feedback_fn)
1188
    bad = bad or result
1189

    
1190
    feedback_fn("* Verifying remaining instances")
1191
    result = self._VerifyOrphanInstances(instancelist, node_instance,
1192
                                         feedback_fn)
1193
    bad = bad or result
1194

    
1195
    if constants.VERIFY_NPLUSONE_MEM not in self.skip_set:
1196
      feedback_fn("* Verifying N+1 Memory redundancy")
1197
      result = self._VerifyNPlusOneMemory(node_info, instance_cfg, feedback_fn)
1198
      bad = bad or result
1199

    
1200
    feedback_fn("* Other Notes")
1201
    if i_non_redundant:
1202
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
1203
                  % len(i_non_redundant))
1204

    
1205
    if i_non_a_balanced:
1206
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
1207
                  % len(i_non_a_balanced))
1208

    
1209
    if n_offline:
1210
      feedback_fn("  - NOTICE: %d offline node(s) found." % len(n_offline))
1211

    
1212
    if n_drained:
1213
      feedback_fn("  - NOTICE: %d drained node(s) found." % len(n_drained))
1214

    
1215
    return not bad
1216

    
1217
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
1218
    """Analize the post-hooks' result
1219

1220
    This method analyses the hook result, handles it, and sends some
1221
    nicely-formatted feedback back to the user.
1222

1223
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
1224
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
1225
    @param hooks_results: the results of the multi-node hooks rpc call
1226
    @param feedback_fn: function used send feedback back to the caller
1227
    @param lu_result: previous Exec result
1228
    @return: the new Exec result, based on the previous result
1229
        and hook results
1230

1231
    """
1232
    # We only really run POST phase hooks, and are only interested in
1233
    # their results
1234
    if phase == constants.HOOKS_PHASE_POST:
1235
      # Used to change hooks' output to proper indentation
1236
      indent_re = re.compile('^', re.M)
1237
      feedback_fn("* Hooks Results")
1238
      if not hooks_results:
1239
        feedback_fn("  - ERROR: general communication failure")
1240
        lu_result = 1
1241
      else:
1242
        for node_name in hooks_results:
1243
          show_node_header = True
1244
          res = hooks_results[node_name]
1245
          if res.failed or res.data is False or not isinstance(res.data, list):
1246
            if res.offline:
1247
              # no need to warn or set fail return value
1248
              continue
1249
            feedback_fn("    Communication failure in hooks execution")
1250
            lu_result = 1
1251
            continue
1252
          for script, hkr, output in res.data:
1253
            if hkr == constants.HKR_FAIL:
1254
              # The node header is only shown once, if there are
1255
              # failing hooks on that node
1256
              if show_node_header:
1257
                feedback_fn("  Node %s:" % node_name)
1258
                show_node_header = False
1259
              feedback_fn("    ERROR: Script %s failed, output:" % script)
1260
              output = indent_re.sub('      ', output)
1261
              feedback_fn("%s" % output)
1262
              lu_result = 1
1263

    
1264
      return lu_result
1265

    
1266

    
1267
class LUVerifyDisks(NoHooksLU):
1268
  """Verifies the cluster disks status.
1269

1270
  """
1271
  _OP_REQP = []
1272
  REQ_BGL = False
1273

    
1274
  def ExpandNames(self):
1275
    self.needed_locks = {
1276
      locking.LEVEL_NODE: locking.ALL_SET,
1277
      locking.LEVEL_INSTANCE: locking.ALL_SET,
1278
    }
1279
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
1280

    
1281
  def CheckPrereq(self):
1282
    """Check prerequisites.
1283

1284
    This has no prerequisites.
1285

1286
    """
1287
    pass
1288

    
1289
  def Exec(self, feedback_fn):
1290
    """Verify integrity of cluster disks.
1291

1292
    @rtype: tuple of three items
1293
    @return: a tuple of (dict of node-to-node_error, list of instances
1294
        which need activate-disks, dict of instance: (node, volume) for
1295
        missing volumes
1296

1297
    """
1298
    result = res_nodes, res_instances, res_missing = {}, [], {}
1299

    
1300
    vg_name = self.cfg.GetVGName()
1301
    nodes = utils.NiceSort(self.cfg.GetNodeList())
1302
    instances = [self.cfg.GetInstanceInfo(name)
1303
                 for name in self.cfg.GetInstanceList()]
1304

    
1305
    nv_dict = {}
1306
    for inst in instances:
1307
      inst_lvs = {}
1308
      if (not inst.admin_up or
1309
          inst.disk_template not in constants.DTS_NET_MIRROR):
1310
        continue
1311
      inst.MapLVsByNode(inst_lvs)
1312
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
1313
      for node, vol_list in inst_lvs.iteritems():
1314
        for vol in vol_list:
1315
          nv_dict[(node, vol)] = inst
1316

    
1317
    if not nv_dict:
1318
      return result
1319

    
1320
    node_lvs = self.rpc.call_volume_list(nodes, vg_name)
1321

    
1322
    to_act = set()
1323
    for node in nodes:
1324
      # node_volume
1325
      node_res = node_lvs[node]
1326
      if node_res.offline:
1327
        continue
1328
      msg = node_res.RemoteFailMsg()
1329
      if msg:
1330
        logging.warning("Error enumerating LVs on node %s: %s", node, msg)
1331
        res_nodes[node] = msg
1332
        continue
1333

    
1334
      lvs = node_res.payload
1335
      for lv_name, (_, lv_inactive, lv_online) in lvs.items():
1336
        inst = nv_dict.pop((node, lv_name), None)
1337
        if (not lv_online and inst is not None
1338
            and inst.name not in res_instances):
1339
          res_instances.append(inst.name)
1340

    
1341
    # any leftover items in nv_dict are missing LVs, let's arrange the
1342
    # data better
1343
    for key, inst in nv_dict.iteritems():
1344
      if inst.name not in res_missing:
1345
        res_missing[inst.name] = []
1346
      res_missing[inst.name].append(key)
1347

    
1348
    return result
1349

    
1350

    
1351
class LURenameCluster(LogicalUnit):
1352
  """Rename the cluster.
1353

1354
  """
1355
  HPATH = "cluster-rename"
1356
  HTYPE = constants.HTYPE_CLUSTER
1357
  _OP_REQP = ["name"]
1358

    
1359
  def BuildHooksEnv(self):
1360
    """Build hooks env.
1361

1362
    """
1363
    env = {
1364
      "OP_TARGET": self.cfg.GetClusterName(),
1365
      "NEW_NAME": self.op.name,
1366
      }
1367
    mn = self.cfg.GetMasterNode()
1368
    return env, [mn], [mn]
1369

    
1370
  def CheckPrereq(self):
1371
    """Verify that the passed name is a valid one.
1372

1373
    """
1374
    hostname = utils.HostInfo(self.op.name)
1375

    
1376
    new_name = hostname.name
1377
    self.ip = new_ip = hostname.ip
1378
    old_name = self.cfg.GetClusterName()
1379
    old_ip = self.cfg.GetMasterIP()
1380
    if new_name == old_name and new_ip == old_ip:
1381
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
1382
                                 " cluster has changed")
1383
    if new_ip != old_ip:
1384
      if utils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
1385
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
1386
                                   " reachable on the network. Aborting." %
1387
                                   new_ip)
1388

    
1389
    self.op.name = new_name
1390

    
1391
  def Exec(self, feedback_fn):
1392
    """Rename the cluster.
1393

1394
    """
1395
    clustername = self.op.name
1396
    ip = self.ip
1397

    
1398
    # shutdown the master IP
1399
    master = self.cfg.GetMasterNode()
1400
    result = self.rpc.call_node_stop_master(master, False)
1401
    msg = result.RemoteFailMsg()
1402
    if msg:
1403
      raise errors.OpExecError("Could not disable the master role: %s" % msg)
1404

    
1405
    try:
1406
      cluster = self.cfg.GetClusterInfo()
1407
      cluster.cluster_name = clustername
1408
      cluster.master_ip = ip
1409
      self.cfg.Update(cluster)
1410

    
1411
      # update the known hosts file
1412
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
1413
      node_list = self.cfg.GetNodeList()
1414
      try:
1415
        node_list.remove(master)
1416
      except ValueError:
1417
        pass
1418
      result = self.rpc.call_upload_file(node_list,
1419
                                         constants.SSH_KNOWN_HOSTS_FILE)
1420
      for to_node, to_result in result.iteritems():
1421
         msg = to_result.RemoteFailMsg()
1422
         if msg:
1423
           msg = ("Copy of file %s to node %s failed: %s" %
1424
                   (constants.SSH_KNOWN_HOSTS_FILE, to_node, msg))
1425
           self.proc.LogWarning(msg)
1426

    
1427
    finally:
1428
      result = self.rpc.call_node_start_master(master, False)
1429
      msg = result.RemoteFailMsg()
1430
      if msg:
1431
        self.LogWarning("Could not re-enable the master role on"
1432
                        " the master, please restart manually: %s", msg)
1433

    
1434

    
1435
def _RecursiveCheckIfLVMBased(disk):
1436
  """Check if the given disk or its children are lvm-based.
1437

1438
  @type disk: L{objects.Disk}
1439
  @param disk: the disk to check
1440
  @rtype: booleean
1441
  @return: boolean indicating whether a LD_LV dev_type was found or not
1442

1443
  """
1444
  if disk.children:
1445
    for chdisk in disk.children:
1446
      if _RecursiveCheckIfLVMBased(chdisk):
1447
        return True
1448
  return disk.dev_type == constants.LD_LV
1449

    
1450

    
1451
class LUSetClusterParams(LogicalUnit):
1452
  """Change the parameters of the cluster.
1453

1454
  """
1455
  HPATH = "cluster-modify"
1456
  HTYPE = constants.HTYPE_CLUSTER
1457
  _OP_REQP = []
1458
  REQ_BGL = False
1459

    
1460
  def CheckArguments(self):
1461
    """Check parameters
1462

1463
    """
1464
    if not hasattr(self.op, "candidate_pool_size"):
1465
      self.op.candidate_pool_size = None
1466
    if self.op.candidate_pool_size is not None:
1467
      try:
1468
        self.op.candidate_pool_size = int(self.op.candidate_pool_size)
1469
      except (ValueError, TypeError), err:
1470
        raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
1471
                                   str(err))
1472
      if self.op.candidate_pool_size < 1:
1473
        raise errors.OpPrereqError("At least one master candidate needed")
1474

    
1475
  def ExpandNames(self):
1476
    # FIXME: in the future maybe other cluster params won't require checking on
1477
    # all nodes to be modified.
1478
    self.needed_locks = {
1479
      locking.LEVEL_NODE: locking.ALL_SET,
1480
    }
1481
    self.share_locks[locking.LEVEL_NODE] = 1
1482

    
1483
  def BuildHooksEnv(self):
1484
    """Build hooks env.
1485

1486
    """
1487
    env = {
1488
      "OP_TARGET": self.cfg.GetClusterName(),
1489
      "NEW_VG_NAME": self.op.vg_name,
1490
      }
1491
    mn = self.cfg.GetMasterNode()
1492
    return env, [mn], [mn]
1493

    
1494
  def CheckPrereq(self):
1495
    """Check prerequisites.
1496

1497
    This checks whether the given params don't conflict and
1498
    if the given volume group is valid.
1499

1500
    """
1501
    if self.op.vg_name is not None and not self.op.vg_name:
1502
      instances = self.cfg.GetAllInstancesInfo().values()
1503
      for inst in instances:
1504
        for disk in inst.disks:
1505
          if _RecursiveCheckIfLVMBased(disk):
1506
            raise errors.OpPrereqError("Cannot disable lvm storage while"
1507
                                       " lvm-based instances exist")
1508

    
1509
    node_list = self.acquired_locks[locking.LEVEL_NODE]
1510

    
1511
    # if vg_name not None, checks given volume group on all nodes
1512
    if self.op.vg_name:
1513
      vglist = self.rpc.call_vg_list(node_list)
1514
      for node in node_list:
1515
        msg = vglist[node].RemoteFailMsg()
1516
        if msg:
1517
          # ignoring down node
1518
          self.LogWarning("Error while gathering data on node %s"
1519
                          " (ignoring node): %s", node, msg)
1520
          continue
1521
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
1522
                                              self.op.vg_name,
1523
                                              constants.MIN_VG_SIZE)
1524
        if vgstatus:
1525
          raise errors.OpPrereqError("Error on node '%s': %s" %
1526
                                     (node, vgstatus))
1527

    
1528
    self.cluster = cluster = self.cfg.GetClusterInfo()
1529
    # validate params changes
1530
    if self.op.beparams:
1531
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
1532
      self.new_beparams = objects.FillDict(
1533
        cluster.beparams[constants.PP_DEFAULT], self.op.beparams)
1534

    
1535
    if self.op.nicparams:
1536
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
1537
      self.new_nicparams = objects.FillDict(
1538
        cluster.nicparams[constants.PP_DEFAULT], self.op.nicparams)
1539
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
1540

    
1541
    # hypervisor list/parameters
1542
    self.new_hvparams = objects.FillDict(cluster.hvparams, {})
1543
    if self.op.hvparams:
1544
      if not isinstance(self.op.hvparams, dict):
1545
        raise errors.OpPrereqError("Invalid 'hvparams' parameter on input")
1546
      for hv_name, hv_dict in self.op.hvparams.items():
1547
        if hv_name not in self.new_hvparams:
1548
          self.new_hvparams[hv_name] = hv_dict
1549
        else:
1550
          self.new_hvparams[hv_name].update(hv_dict)
1551

    
1552
    if self.op.enabled_hypervisors is not None:
1553
      self.hv_list = self.op.enabled_hypervisors
1554
    else:
1555
      self.hv_list = cluster.enabled_hypervisors
1556

    
1557
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
1558
      # either the enabled list has changed, or the parameters have, validate
1559
      for hv_name, hv_params in self.new_hvparams.items():
1560
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
1561
            (self.op.enabled_hypervisors and
1562
             hv_name in self.op.enabled_hypervisors)):
1563
          # either this is a new hypervisor, or its parameters have changed
1564
          hv_class = hypervisor.GetHypervisor(hv_name)
1565
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1566
          hv_class.CheckParameterSyntax(hv_params)
1567
          _CheckHVParams(self, node_list, hv_name, hv_params)
1568

    
1569
  def Exec(self, feedback_fn):
1570
    """Change the parameters of the cluster.
1571

1572
    """
1573
    if self.op.vg_name is not None:
1574
      new_volume = self.op.vg_name
1575
      if not new_volume:
1576
        new_volume = None
1577
      if new_volume != self.cfg.GetVGName():
1578
        self.cfg.SetVGName(new_volume)
1579
      else:
1580
        feedback_fn("Cluster LVM configuration already in desired"
1581
                    " state, not changing")
1582
    if self.op.hvparams:
1583
      self.cluster.hvparams = self.new_hvparams
1584
    if self.op.enabled_hypervisors is not None:
1585
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
1586
    if self.op.beparams:
1587
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
1588
    if self.op.nicparams:
1589
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
1590

    
1591
    if self.op.candidate_pool_size is not None:
1592
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
1593

    
1594
    self.cfg.Update(self.cluster)
1595

    
1596
    # we want to update nodes after the cluster so that if any errors
1597
    # happen, we have recorded and saved the cluster info
1598
    if self.op.candidate_pool_size is not None:
1599
      _AdjustCandidatePool(self)
1600

    
1601

    
1602
def _RedistributeAncillaryFiles(lu, additional_nodes=None):
1603
  """Distribute additional files which are part of the cluster configuration.
1604

1605
  ConfigWriter takes care of distributing the config and ssconf files, but
1606
  there are more files which should be distributed to all nodes. This function
1607
  makes sure those are copied.
1608

1609
  @param lu: calling logical unit
1610
  @param additional_nodes: list of nodes not in the config to distribute to
1611

1612
  """
1613
  # 1. Gather target nodes
1614
  myself = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
1615
  dist_nodes = lu.cfg.GetNodeList()
1616
  if additional_nodes is not None:
1617
    dist_nodes.extend(additional_nodes)
1618
  if myself.name in dist_nodes:
1619
    dist_nodes.remove(myself.name)
1620
  # 2. Gather files to distribute
1621
  dist_files = set([constants.ETC_HOSTS,
1622
                    constants.SSH_KNOWN_HOSTS_FILE,
1623
                    constants.RAPI_CERT_FILE,
1624
                    constants.RAPI_USERS_FILE,
1625
                   ])
1626

    
1627
  enabled_hypervisors = lu.cfg.GetClusterInfo().enabled_hypervisors
1628
  for hv_name in enabled_hypervisors:
1629
    hv_class = hypervisor.GetHypervisor(hv_name)
1630
    dist_files.update(hv_class.GetAncillaryFiles())
1631

    
1632
  # 3. Perform the files upload
1633
  for fname in dist_files:
1634
    if os.path.exists(fname):
1635
      result = lu.rpc.call_upload_file(dist_nodes, fname)
1636
      for to_node, to_result in result.items():
1637
         msg = to_result.RemoteFailMsg()
1638
         if msg:
1639
           msg = ("Copy of file %s to node %s failed: %s" %
1640
                   (fname, to_node, msg))
1641
           lu.proc.LogWarning(msg)
1642

    
1643

    
1644
class LURedistributeConfig(NoHooksLU):
1645
  """Force the redistribution of cluster configuration.
1646

1647
  This is a very simple LU.
1648

1649
  """
1650
  _OP_REQP = []
1651
  REQ_BGL = False
1652

    
1653
  def ExpandNames(self):
1654
    self.needed_locks = {
1655
      locking.LEVEL_NODE: locking.ALL_SET,
1656
    }
1657
    self.share_locks[locking.LEVEL_NODE] = 1
1658

    
1659
  def CheckPrereq(self):
1660
    """Check prerequisites.
1661

1662
    """
1663

    
1664
  def Exec(self, feedback_fn):
1665
    """Redistribute the configuration.
1666

1667
    """
1668
    self.cfg.Update(self.cfg.GetClusterInfo())
1669
    _RedistributeAncillaryFiles(self)
1670

    
1671

    
1672
def _WaitForSync(lu, instance, oneshot=False, unlock=False):
1673
  """Sleep and poll for an instance's disk to sync.
1674

1675
  """
1676
  if not instance.disks:
1677
    return True
1678

    
1679
  if not oneshot:
1680
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
1681

    
1682
  node = instance.primary_node
1683

    
1684
  for dev in instance.disks:
1685
    lu.cfg.SetDiskID(dev, node)
1686

    
1687
  retries = 0
1688
  while True:
1689
    max_time = 0
1690
    done = True
1691
    cumul_degraded = False
1692
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
1693
    msg = rstats.RemoteFailMsg()
1694
    if msg:
1695
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
1696
      retries += 1
1697
      if retries >= 10:
1698
        raise errors.RemoteError("Can't contact node %s for mirror data,"
1699
                                 " aborting." % node)
1700
      time.sleep(6)
1701
      continue
1702
    rstats = rstats.payload
1703
    retries = 0
1704
    for i, mstat in enumerate(rstats):
1705
      if mstat is None:
1706
        lu.LogWarning("Can't compute data for node %s/%s",
1707
                           node, instance.disks[i].iv_name)
1708
        continue
1709
      # we ignore the ldisk parameter
1710
      perc_done, est_time, is_degraded, _ = mstat
1711
      cumul_degraded = cumul_degraded or (is_degraded and perc_done is None)
1712
      if perc_done is not None:
1713
        done = False
1714
        if est_time is not None:
1715
          rem_time = "%d estimated seconds remaining" % est_time
1716
          max_time = est_time
1717
        else:
1718
          rem_time = "no time estimate"
1719
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
1720
                        (instance.disks[i].iv_name, perc_done, rem_time))
1721
    if done or oneshot:
1722
      break
1723

    
1724
    time.sleep(min(60, max_time))
1725

    
1726
  if done:
1727
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
1728
  return not cumul_degraded
1729

    
1730

    
1731
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
1732
  """Check that mirrors are not degraded.
1733

1734
  The ldisk parameter, if True, will change the test from the
1735
  is_degraded attribute (which represents overall non-ok status for
1736
  the device(s)) to the ldisk (representing the local storage status).
1737

1738
  """
1739
  lu.cfg.SetDiskID(dev, node)
1740
  if ldisk:
1741
    idx = 6
1742
  else:
1743
    idx = 5
1744

    
1745
  result = True
1746
  if on_primary or dev.AssembleOnSecondary():
1747
    rstats = lu.rpc.call_blockdev_find(node, dev)
1748
    msg = rstats.RemoteFailMsg()
1749
    if msg:
1750
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
1751
      result = False
1752
    elif not rstats.payload:
1753
      lu.LogWarning("Can't find disk on node %s", node)
1754
      result = False
1755
    else:
1756
      result = result and (not rstats.payload[idx])
1757
  if dev.children:
1758
    for child in dev.children:
1759
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
1760

    
1761
  return result
1762

    
1763

    
1764
class LUDiagnoseOS(NoHooksLU):
1765
  """Logical unit for OS diagnose/query.
1766

1767
  """
1768
  _OP_REQP = ["output_fields", "names"]
1769
  REQ_BGL = False
1770
  _FIELDS_STATIC = utils.FieldSet()
1771
  _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status")
1772

    
1773
  def ExpandNames(self):
1774
    if self.op.names:
1775
      raise errors.OpPrereqError("Selective OS query not supported")
1776

    
1777
    _CheckOutputFields(static=self._FIELDS_STATIC,
1778
                       dynamic=self._FIELDS_DYNAMIC,
1779
                       selected=self.op.output_fields)
1780

    
1781
    # Lock all nodes, in shared mode
1782
    # Temporary removal of locks, should be reverted later
1783
    # TODO: reintroduce locks when they are lighter-weight
1784
    self.needed_locks = {}
1785
    #self.share_locks[locking.LEVEL_NODE] = 1
1786
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
1787

    
1788
  def CheckPrereq(self):
1789
    """Check prerequisites.
1790

1791
    """
1792

    
1793
  @staticmethod
1794
  def _DiagnoseByOS(node_list, rlist):
1795
    """Remaps a per-node return list into an a per-os per-node dictionary
1796

1797
    @param node_list: a list with the names of all nodes
1798
    @param rlist: a map with node names as keys and OS objects as values
1799

1800
    @rtype: dict
1801
    @return: a dictionary with osnames as keys and as value another map, with
1802
        nodes as keys and list of OS objects as values, eg::
1803

1804
          {"debian-etch": {"node1": [<object>,...],
1805
                           "node2": [<object>,]}
1806
          }
1807

1808
    """
1809
    all_os = {}
1810
    # we build here the list of nodes that didn't fail the RPC (at RPC
1811
    # level), so that nodes with a non-responding node daemon don't
1812
    # make all OSes invalid
1813
    good_nodes = [node_name for node_name in rlist
1814
                  if not rlist[node_name].failed]
1815
    for node_name, nr in rlist.iteritems():
1816
      if nr.failed or not nr.data:
1817
        continue
1818
      for os_obj in nr.data:
1819
        if os_obj.name not in all_os:
1820
          # build a list of nodes for this os containing empty lists
1821
          # for each node in node_list
1822
          all_os[os_obj.name] = {}
1823
          for nname in good_nodes:
1824
            all_os[os_obj.name][nname] = []
1825
        all_os[os_obj.name][node_name].append(os_obj)
1826
    return all_os
1827

    
1828
  def Exec(self, feedback_fn):
1829
    """Compute the list of OSes.
1830

1831
    """
1832
    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
1833
    node_data = self.rpc.call_os_diagnose(valid_nodes)
1834
    if node_data == False:
1835
      raise errors.OpExecError("Can't gather the list of OSes")
1836
    pol = self._DiagnoseByOS(valid_nodes, node_data)
1837
    output = []
1838
    for os_name, os_data in pol.iteritems():
1839
      row = []
1840
      for field in self.op.output_fields:
1841
        if field == "name":
1842
          val = os_name
1843
        elif field == "valid":
1844
          val = utils.all([osl and osl[0] for osl in os_data.values()])
1845
        elif field == "node_status":
1846
          val = {}
1847
          for node_name, nos_list in os_data.iteritems():
1848
            val[node_name] = [(v.status, v.path) for v in nos_list]
1849
        else:
1850
          raise errors.ParameterError(field)
1851
        row.append(val)
1852
      output.append(row)
1853

    
1854
    return output
1855

    
1856

    
1857
class LURemoveNode(LogicalUnit):
1858
  """Logical unit for removing a node.
1859

1860
  """
1861
  HPATH = "node-remove"
1862
  HTYPE = constants.HTYPE_NODE
1863
  _OP_REQP = ["node_name"]
1864

    
1865
  def BuildHooksEnv(self):
1866
    """Build hooks env.
1867

1868
    This doesn't run on the target node in the pre phase as a failed
1869
    node would then be impossible to remove.
1870

1871
    """
1872
    env = {
1873
      "OP_TARGET": self.op.node_name,
1874
      "NODE_NAME": self.op.node_name,
1875
      }
1876
    all_nodes = self.cfg.GetNodeList()
1877
    all_nodes.remove(self.op.node_name)
1878
    return env, all_nodes, all_nodes
1879

    
1880
  def CheckPrereq(self):
1881
    """Check prerequisites.
1882

1883
    This checks:
1884
     - the node exists in the configuration
1885
     - it does not have primary or secondary instances
1886
     - it's not the master
1887

1888
    Any errors are signalled by raising errors.OpPrereqError.
1889

1890
    """
1891
    node = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.node_name))
1892
    if node is None:
1893
      raise errors.OpPrereqError, ("Node '%s' is unknown." % self.op.node_name)
1894

    
1895
    instance_list = self.cfg.GetInstanceList()
1896

    
1897
    masternode = self.cfg.GetMasterNode()
1898
    if node.name == masternode:
1899
      raise errors.OpPrereqError("Node is the master node,"
1900
                                 " you need to failover first.")
1901

    
1902
    for instance_name in instance_list:
1903
      instance = self.cfg.GetInstanceInfo(instance_name)
1904
      if node.name in instance.all_nodes:
1905
        raise errors.OpPrereqError("Instance %s is still running on the node,"
1906
                                   " please remove first." % instance_name)
1907
    self.op.node_name = node.name
1908
    self.node = node
1909

    
1910
  def Exec(self, feedback_fn):
1911
    """Removes the node from the cluster.
1912

1913
    """
1914
    node = self.node
1915
    logging.info("Stopping the node daemon and removing configs from node %s",
1916
                 node.name)
1917

    
1918
    self.context.RemoveNode(node.name)
1919

    
1920
    result = self.rpc.call_node_leave_cluster(node.name)
1921
    msg = result.RemoteFailMsg()
1922
    if msg:
1923
      self.LogWarning("Errors encountered on the remote node while leaving"
1924
                      " the cluster: %s", msg)
1925

    
1926
    # Promote nodes to master candidate as needed
1927
    _AdjustCandidatePool(self)
1928

    
1929

    
1930
class LUQueryNodes(NoHooksLU):
1931
  """Logical unit for querying nodes.
1932

1933
  """
1934
  _OP_REQP = ["output_fields", "names", "use_locking"]
1935
  REQ_BGL = False
1936
  _FIELDS_DYNAMIC = utils.FieldSet(
1937
    "dtotal", "dfree",
1938
    "mtotal", "mnode", "mfree",
1939
    "bootid",
1940
    "ctotal", "cnodes", "csockets",
1941
    )
1942

    
1943
  _FIELDS_STATIC = utils.FieldSet(
1944
    "name", "pinst_cnt", "sinst_cnt",
1945
    "pinst_list", "sinst_list",
1946
    "pip", "sip", "tags",
1947
    "serial_no",
1948
    "master_candidate",
1949
    "master",
1950
    "offline",
1951
    "drained",
1952
    )
1953

    
1954
  def ExpandNames(self):
1955
    _CheckOutputFields(static=self._FIELDS_STATIC,
1956
                       dynamic=self._FIELDS_DYNAMIC,
1957
                       selected=self.op.output_fields)
1958

    
1959
    self.needed_locks = {}
1960
    self.share_locks[locking.LEVEL_NODE] = 1
1961

    
1962
    if self.op.names:
1963
      self.wanted = _GetWantedNodes(self, self.op.names)
1964
    else:
1965
      self.wanted = locking.ALL_SET
1966

    
1967
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
1968
    self.do_locking = self.do_node_query and self.op.use_locking
1969
    if self.do_locking:
1970
      # if we don't request only static fields, we need to lock the nodes
1971
      self.needed_locks[locking.LEVEL_NODE] = self.wanted
1972

    
1973

    
1974
  def CheckPrereq(self):
1975
    """Check prerequisites.
1976

1977
    """
1978
    # The validation of the node list is done in the _GetWantedNodes,
1979
    # if non empty, and if empty, there's no validation to do
1980
    pass
1981

    
1982
  def Exec(self, feedback_fn):
1983
    """Computes the list of nodes and their attributes.
1984

1985
    """
1986
    all_info = self.cfg.GetAllNodesInfo()
1987
    if self.do_locking:
1988
      nodenames = self.acquired_locks[locking.LEVEL_NODE]
1989
    elif self.wanted != locking.ALL_SET:
1990
      nodenames = self.wanted
1991
      missing = set(nodenames).difference(all_info.keys())
1992
      if missing:
1993
        raise errors.OpExecError(
1994
          "Some nodes were removed before retrieving their data: %s" % missing)
1995
    else:
1996
      nodenames = all_info.keys()
1997

    
1998
    nodenames = utils.NiceSort(nodenames)
1999
    nodelist = [all_info[name] for name in nodenames]
2000

    
2001
    # begin data gathering
2002

    
2003
    if self.do_node_query:
2004
      live_data = {}
2005
      node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
2006
                                          self.cfg.GetHypervisorType())
2007
      for name in nodenames:
2008
        nodeinfo = node_data[name]
2009
        if not nodeinfo.RemoteFailMsg() and nodeinfo.payload:
2010
          nodeinfo = nodeinfo.payload
2011
          fn = utils.TryConvert
2012
          live_data[name] = {
2013
            "mtotal": fn(int, nodeinfo.get('memory_total', None)),
2014
            "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
2015
            "mfree": fn(int, nodeinfo.get('memory_free', None)),
2016
            "dtotal": fn(int, nodeinfo.get('vg_size', None)),
2017
            "dfree": fn(int, nodeinfo.get('vg_free', None)),
2018
            "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
2019
            "bootid": nodeinfo.get('bootid', None),
2020
            "cnodes": fn(int, nodeinfo.get('cpu_nodes', None)),
2021
            "csockets": fn(int, nodeinfo.get('cpu_sockets', None)),
2022
            }
2023
        else:
2024
          live_data[name] = {}
2025
    else:
2026
      live_data = dict.fromkeys(nodenames, {})
2027

    
2028
    node_to_primary = dict([(name, set()) for name in nodenames])
2029
    node_to_secondary = dict([(name, set()) for name in nodenames])
2030

    
2031
    inst_fields = frozenset(("pinst_cnt", "pinst_list",
2032
                             "sinst_cnt", "sinst_list"))
2033
    if inst_fields & frozenset(self.op.output_fields):
2034
      instancelist = self.cfg.GetInstanceList()
2035

    
2036
      for instance_name in instancelist:
2037
        inst = self.cfg.GetInstanceInfo(instance_name)
2038
        if inst.primary_node in node_to_primary:
2039
          node_to_primary[inst.primary_node].add(inst.name)
2040
        for secnode in inst.secondary_nodes:
2041
          if secnode in node_to_secondary:
2042
            node_to_secondary[secnode].add(inst.name)
2043

    
2044
    master_node = self.cfg.GetMasterNode()
2045

    
2046
    # end data gathering
2047

    
2048
    output = []
2049
    for node in nodelist:
2050
      node_output = []
2051
      for field in self.op.output_fields:
2052
        if field == "name":
2053
          val = node.name
2054
        elif field == "pinst_list":
2055
          val = list(node_to_primary[node.name])
2056
        elif field == "sinst_list":
2057
          val = list(node_to_secondary[node.name])
2058
        elif field == "pinst_cnt":
2059
          val = len(node_to_primary[node.name])
2060
        elif field == "sinst_cnt":
2061
          val = len(node_to_secondary[node.name])
2062
        elif field == "pip":
2063
          val = node.primary_ip
2064
        elif field == "sip":
2065
          val = node.secondary_ip
2066
        elif field == "tags":
2067
          val = list(node.GetTags())
2068
        elif field == "serial_no":
2069
          val = node.serial_no
2070
        elif field == "master_candidate":
2071
          val = node.master_candidate
2072
        elif field == "master":
2073
          val = node.name == master_node
2074
        elif field == "offline":
2075
          val = node.offline
2076
        elif field == "drained":
2077
          val = node.drained
2078
        elif self._FIELDS_DYNAMIC.Matches(field):
2079
          val = live_data[node.name].get(field, None)
2080
        else:
2081
          raise errors.ParameterError(field)
2082
        node_output.append(val)
2083
      output.append(node_output)
2084

    
2085
    return output
2086

    
2087

    
2088
class LUQueryNodeVolumes(NoHooksLU):
2089
  """Logical unit for getting volumes on node(s).
2090

2091
  """
2092
  _OP_REQP = ["nodes", "output_fields"]
2093
  REQ_BGL = False
2094
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
2095
  _FIELDS_STATIC = utils.FieldSet("node")
2096

    
2097
  def ExpandNames(self):
2098
    _CheckOutputFields(static=self._FIELDS_STATIC,
2099
                       dynamic=self._FIELDS_DYNAMIC,
2100
                       selected=self.op.output_fields)
2101

    
2102
    self.needed_locks = {}
2103
    self.share_locks[locking.LEVEL_NODE] = 1
2104
    if not self.op.nodes:
2105
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
2106
    else:
2107
      self.needed_locks[locking.LEVEL_NODE] = \
2108
        _GetWantedNodes(self, self.op.nodes)
2109

    
2110
  def CheckPrereq(self):
2111
    """Check prerequisites.
2112

2113
    This checks that the fields required are valid output fields.
2114

2115
    """
2116
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
2117

    
2118
  def Exec(self, feedback_fn):
2119
    """Computes the list of nodes and their attributes.
2120

2121
    """
2122
    nodenames = self.nodes
2123
    volumes = self.rpc.call_node_volumes(nodenames)
2124

    
2125
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
2126
             in self.cfg.GetInstanceList()]
2127

    
2128
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
2129

    
2130
    output = []
2131
    for node in nodenames:
2132
      if node not in volumes or volumes[node].failed or not volumes[node].data:
2133
        continue
2134

    
2135
      node_vols = volumes[node].data[:]
2136
      node_vols.sort(key=lambda vol: vol['dev'])
2137

    
2138
      for vol in node_vols:
2139
        node_output = []
2140
        for field in self.op.output_fields:
2141
          if field == "node":
2142
            val = node
2143
          elif field == "phys":
2144
            val = vol['dev']
2145
          elif field == "vg":
2146
            val = vol['vg']
2147
          elif field == "name":
2148
            val = vol['name']
2149
          elif field == "size":
2150
            val = int(float(vol['size']))
2151
          elif field == "instance":
2152
            for inst in ilist:
2153
              if node not in lv_by_node[inst]:
2154
                continue
2155
              if vol['name'] in lv_by_node[inst][node]:
2156
                val = inst.name
2157
                break
2158
            else:
2159
              val = '-'
2160
          else:
2161
            raise errors.ParameterError(field)
2162
          node_output.append(str(val))
2163

    
2164
        output.append(node_output)
2165

    
2166
    return output
2167

    
2168

    
2169
class LUAddNode(LogicalUnit):
2170
  """Logical unit for adding node to the cluster.
2171

2172
  """
2173
  HPATH = "node-add"
2174
  HTYPE = constants.HTYPE_NODE
2175
  _OP_REQP = ["node_name"]
2176

    
2177
  def BuildHooksEnv(self):
2178
    """Build hooks env.
2179

2180
    This will run on all nodes before, and on all nodes + the new node after.
2181

2182
    """
2183
    env = {
2184
      "OP_TARGET": self.op.node_name,
2185
      "NODE_NAME": self.op.node_name,
2186
      "NODE_PIP": self.op.primary_ip,
2187
      "NODE_SIP": self.op.secondary_ip,
2188
      }
2189
    nodes_0 = self.cfg.GetNodeList()
2190
    nodes_1 = nodes_0 + [self.op.node_name, ]
2191
    return env, nodes_0, nodes_1
2192

    
2193
  def CheckPrereq(self):
2194
    """Check prerequisites.
2195

2196
    This checks:
2197
     - the new node is not already in the config
2198
     - it is resolvable
2199
     - its parameters (single/dual homed) matches the cluster
2200

2201
    Any errors are signalled by raising errors.OpPrereqError.
2202

2203
    """
2204
    node_name = self.op.node_name
2205
    cfg = self.cfg
2206

    
2207
    dns_data = utils.HostInfo(node_name)
2208

    
2209
    node = dns_data.name
2210
    primary_ip = self.op.primary_ip = dns_data.ip
2211
    secondary_ip = getattr(self.op, "secondary_ip", None)
2212
    if secondary_ip is None:
2213
      secondary_ip = primary_ip
2214
    if not utils.IsValidIP(secondary_ip):
2215
      raise errors.OpPrereqError("Invalid secondary IP given")
2216
    self.op.secondary_ip = secondary_ip
2217

    
2218
    node_list = cfg.GetNodeList()
2219
    if not self.op.readd and node in node_list:
2220
      raise errors.OpPrereqError("Node %s is already in the configuration" %
2221
                                 node)
2222
    elif self.op.readd and node not in node_list:
2223
      raise errors.OpPrereqError("Node %s is not in the configuration" % node)
2224

    
2225
    for existing_node_name in node_list:
2226
      existing_node = cfg.GetNodeInfo(existing_node_name)
2227

    
2228
      if self.op.readd and node == existing_node_name:
2229
        if (existing_node.primary_ip != primary_ip or
2230
            existing_node.secondary_ip != secondary_ip):
2231
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
2232
                                     " address configuration as before")
2233
        continue
2234

    
2235
      if (existing_node.primary_ip == primary_ip or
2236
          existing_node.secondary_ip == primary_ip or
2237
          existing_node.primary_ip == secondary_ip or
2238
          existing_node.secondary_ip == secondary_ip):
2239
        raise errors.OpPrereqError("New node ip address(es) conflict with"
2240
                                   " existing node %s" % existing_node.name)
2241

    
2242
    # check that the type of the node (single versus dual homed) is the
2243
    # same as for the master
2244
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
2245
    master_singlehomed = myself.secondary_ip == myself.primary_ip
2246
    newbie_singlehomed = secondary_ip == primary_ip
2247
    if master_singlehomed != newbie_singlehomed:
2248
      if master_singlehomed:
2249
        raise errors.OpPrereqError("The master has no private ip but the"
2250
                                   " new node has one")
2251
      else:
2252
        raise errors.OpPrereqError("The master has a private ip but the"
2253
                                   " new node doesn't have one")
2254

    
2255
    # checks reachablity
2256
    if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
2257
      raise errors.OpPrereqError("Node not reachable by ping")
2258

    
2259
    if not newbie_singlehomed:
2260
      # check reachability from my secondary ip to newbie's secondary ip
2261
      if not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
2262
                           source=myself.secondary_ip):
2263
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
2264
                                   " based ping to noded port")
2265

    
2266
    cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2267
    mc_now, _ = self.cfg.GetMasterCandidateStats()
2268
    master_candidate = mc_now < cp_size
2269

    
2270
    self.new_node = objects.Node(name=node,
2271
                                 primary_ip=primary_ip,
2272
                                 secondary_ip=secondary_ip,
2273
                                 master_candidate=master_candidate,
2274
                                 offline=False, drained=False)
2275

    
2276
  def Exec(self, feedback_fn):
2277
    """Adds the new node to the cluster.
2278

2279
    """
2280
    new_node = self.new_node
2281
    node = new_node.name
2282

    
2283
    # check connectivity
2284
    result = self.rpc.call_version([node])[node]
2285
    result.Raise()
2286
    if result.data:
2287
      if constants.PROTOCOL_VERSION == result.data:
2288
        logging.info("Communication to node %s fine, sw version %s match",
2289
                     node, result.data)
2290
      else:
2291
        raise errors.OpExecError("Version mismatch master version %s,"
2292
                                 " node version %s" %
2293
                                 (constants.PROTOCOL_VERSION, result.data))
2294
    else:
2295
      raise errors.OpExecError("Cannot get version from the new node")
2296

    
2297
    # setup ssh on node
2298
    logging.info("Copy ssh key to node %s", node)
2299
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
2300
    keyarray = []
2301
    keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
2302
                constants.SSH_HOST_RSA_PRIV, constants.SSH_HOST_RSA_PUB,
2303
                priv_key, pub_key]
2304

    
2305
    for i in keyfiles:
2306
      f = open(i, 'r')
2307
      try:
2308
        keyarray.append(f.read())
2309
      finally:
2310
        f.close()
2311

    
2312
    result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
2313
                                    keyarray[2],
2314
                                    keyarray[3], keyarray[4], keyarray[5])
2315

    
2316
    msg = result.RemoteFailMsg()
2317
    if msg:
2318
      raise errors.OpExecError("Cannot transfer ssh keys to the"
2319
                               " new node: %s" % msg)
2320

    
2321
    # Add node to our /etc/hosts, and add key to known_hosts
2322
    if self.cfg.GetClusterInfo().modify_etc_hosts:
2323
      utils.AddHostToEtcHosts(new_node.name)
2324

    
2325
    if new_node.secondary_ip != new_node.primary_ip:
2326
      result = self.rpc.call_node_has_ip_address(new_node.name,
2327
                                                 new_node.secondary_ip)
2328
      msg = result.RemoteFailMsg()
2329
      if msg:
2330
        raise errors.OpPrereqError("Failure checking secondary ip"
2331
                                   " on node %s: %s" % (new_node.name, msg))
2332
      if not result.payload:
2333
        raise errors.OpExecError("Node claims it doesn't have the secondary ip"
2334
                                 " you gave (%s). Please fix and re-run this"
2335
                                 " command." % new_node.secondary_ip)
2336

    
2337
    node_verify_list = [self.cfg.GetMasterNode()]
2338
    node_verify_param = {
2339
      'nodelist': [node],
2340
      # TODO: do a node-net-test as well?
2341
    }
2342

    
2343
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
2344
                                       self.cfg.GetClusterName())
2345
    for verifier in node_verify_list:
2346
      msg = result[verifier].RemoteFailMsg()
2347
      if msg:
2348
        raise errors.OpExecError("Cannot communicate with node %s: %s" %
2349
                                 (verifier, msg))
2350
      nl_payload = result[verifier].payload['nodelist']
2351
      if nl_payload:
2352
        for failed in nl_payload:
2353
          feedback_fn("ssh/hostname verification failed %s -> %s" %
2354
                      (verifier, nl_payload[failed]))
2355
        raise errors.OpExecError("ssh/hostname verification failed.")
2356

    
2357
    if self.op.readd:
2358
      _RedistributeAncillaryFiles(self)
2359
      self.context.ReaddNode(new_node)
2360
    else:
2361
      _RedistributeAncillaryFiles(self, additional_nodes=[node])
2362
      self.context.AddNode(new_node)
2363

    
2364

    
2365
class LUSetNodeParams(LogicalUnit):
2366
  """Modifies the parameters of a node.
2367

2368
  """
2369
  HPATH = "node-modify"
2370
  HTYPE = constants.HTYPE_NODE
2371
  _OP_REQP = ["node_name"]
2372
  REQ_BGL = False
2373

    
2374
  def CheckArguments(self):
2375
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2376
    if node_name is None:
2377
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2378
    self.op.node_name = node_name
2379
    _CheckBooleanOpField(self.op, 'master_candidate')
2380
    _CheckBooleanOpField(self.op, 'offline')
2381
    _CheckBooleanOpField(self.op, 'drained')
2382
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained]
2383
    if all_mods.count(None) == 3:
2384
      raise errors.OpPrereqError("Please pass at least one modification")
2385
    if all_mods.count(True) > 1:
2386
      raise errors.OpPrereqError("Can't set the node into more than one"
2387
                                 " state at the same time")
2388

    
2389
  def ExpandNames(self):
2390
    self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
2391

    
2392
  def BuildHooksEnv(self):
2393
    """Build hooks env.
2394

2395
    This runs on the master node.
2396

2397
    """
2398
    env = {
2399
      "OP_TARGET": self.op.node_name,
2400
      "MASTER_CANDIDATE": str(self.op.master_candidate),
2401
      "OFFLINE": str(self.op.offline),
2402
      "DRAINED": str(self.op.drained),
2403
      }
2404
    nl = [self.cfg.GetMasterNode(),
2405
          self.op.node_name]
2406
    return env, nl, nl
2407

    
2408
  def CheckPrereq(self):
2409
    """Check prerequisites.
2410

2411
    This only checks the instance list against the existing names.
2412

2413
    """
2414
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
2415

    
2416
    if ((self.op.master_candidate == False or self.op.offline == True or
2417
         self.op.drained == True) and node.master_candidate):
2418
      # we will demote the node from master_candidate
2419
      if self.op.node_name == self.cfg.GetMasterNode():
2420
        raise errors.OpPrereqError("The master node has to be a"
2421
                                   " master candidate, online and not drained")
2422
      cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2423
      num_candidates, _ = self.cfg.GetMasterCandidateStats()
2424
      if num_candidates <= cp_size:
2425
        msg = ("Not enough master candidates (desired"
2426
               " %d, new value will be %d)" % (cp_size, num_candidates-1))
2427
        if self.op.force:
2428
          self.LogWarning(msg)
2429
        else:
2430
          raise errors.OpPrereqError(msg)
2431

    
2432
    if (self.op.master_candidate == True and
2433
        ((node.offline and not self.op.offline == False) or
2434
         (node.drained and not self.op.drained == False))):
2435
      raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
2436
                                 " to master_candidate" % node.name)
2437

    
2438
    return
2439

    
2440
  def Exec(self, feedback_fn):
2441
    """Modifies a node.
2442

2443
    """
2444
    node = self.node
2445

    
2446
    result = []
2447
    changed_mc = False
2448

    
2449
    if self.op.offline is not None:
2450
      node.offline = self.op.offline
2451
      result.append(("offline", str(self.op.offline)))
2452
      if self.op.offline == True:
2453
        if node.master_candidate:
2454
          node.master_candidate = False
2455
          changed_mc = True
2456
          result.append(("master_candidate", "auto-demotion due to offline"))
2457
        if node.drained:
2458
          node.drained = False
2459
          result.append(("drained", "clear drained status due to offline"))
2460

    
2461
    if self.op.master_candidate is not None:
2462
      node.master_candidate = self.op.master_candidate
2463
      changed_mc = True
2464
      result.append(("master_candidate", str(self.op.master_candidate)))
2465
      if self.op.master_candidate == False:
2466
        rrc = self.rpc.call_node_demote_from_mc(node.name)
2467
        msg = rrc.RemoteFailMsg()
2468
        if msg:
2469
          self.LogWarning("Node failed to demote itself: %s" % msg)
2470

    
2471
    if self.op.drained is not None:
2472
      node.drained = self.op.drained
2473
      result.append(("drained", str(self.op.drained)))
2474
      if self.op.drained == True:
2475
        if node.master_candidate:
2476
          node.master_candidate = False
2477
          changed_mc = True
2478
          result.append(("master_candidate", "auto-demotion due to drain"))
2479
        if node.offline:
2480
          node.offline = False
2481
          result.append(("offline", "clear offline status due to drain"))
2482

    
2483
    # this will trigger configuration file update, if needed
2484
    self.cfg.Update(node)
2485
    # this will trigger job queue propagation or cleanup
2486
    if changed_mc:
2487
      self.context.ReaddNode(node)
2488

    
2489
    return result
2490

    
2491

    
2492
class LUPowercycleNode(NoHooksLU):
2493
  """Powercycles a node.
2494

2495
  """
2496
  _OP_REQP = ["node_name", "force"]
2497
  REQ_BGL = False
2498

    
2499
  def CheckArguments(self):
2500
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2501
    if node_name is None:
2502
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2503
    self.op.node_name = node_name
2504
    if node_name == self.cfg.GetMasterNode() and not self.op.force:
2505
      raise errors.OpPrereqError("The node is the master and the force"
2506
                                 " parameter was not set")
2507

    
2508
  def ExpandNames(self):
2509
    """Locking for PowercycleNode.
2510

2511
    This is a last-resource option and shouldn't block on other
2512
    jobs. Therefore, we grab no locks.
2513

2514
    """
2515
    self.needed_locks = {}
2516

    
2517
  def CheckPrereq(self):
2518
    """Check prerequisites.
2519

2520
    This LU has no prereqs.
2521

2522
    """
2523
    pass
2524

    
2525
  def Exec(self, feedback_fn):
2526
    """Reboots a node.
2527

2528
    """
2529
    result = self.rpc.call_node_powercycle(self.op.node_name,
2530
                                           self.cfg.GetHypervisorType())
2531
    msg = result.RemoteFailMsg()
2532
    if msg:
2533
      raise errors.OpExecError("Failed to schedule the reboot: %s" % msg)
2534
    return result.payload
2535

    
2536

    
2537
class LUQueryClusterInfo(NoHooksLU):
2538
  """Query cluster configuration.
2539

2540
  """
2541
  _OP_REQP = []
2542
  REQ_BGL = False
2543

    
2544
  def ExpandNames(self):
2545
    self.needed_locks = {}
2546

    
2547
  def CheckPrereq(self):
2548
    """No prerequsites needed for this LU.
2549

2550
    """
2551
    pass
2552

    
2553
  def Exec(self, feedback_fn):
2554
    """Return cluster config.
2555

2556
    """
2557
    cluster = self.cfg.GetClusterInfo()
2558
    result = {
2559
      "software_version": constants.RELEASE_VERSION,
2560
      "protocol_version": constants.PROTOCOL_VERSION,
2561
      "config_version": constants.CONFIG_VERSION,
2562
      "os_api_version": constants.OS_API_VERSION,
2563
      "export_version": constants.EXPORT_VERSION,
2564
      "architecture": (platform.architecture()[0], platform.machine()),
2565
      "name": cluster.cluster_name,
2566
      "master": cluster.master_node,
2567
      "default_hypervisor": cluster.default_hypervisor,
2568
      "enabled_hypervisors": cluster.enabled_hypervisors,
2569
      "hvparams": dict([(hypervisor, cluster.hvparams[hypervisor])
2570
                        for hypervisor in cluster.enabled_hypervisors]),
2571
      "beparams": cluster.beparams,
2572
      "nicparams": cluster.nicparams,
2573
      "candidate_pool_size": cluster.candidate_pool_size,
2574
      "master_netdev": cluster.master_netdev,
2575
      "volume_group_name": cluster.volume_group_name,
2576
      "file_storage_dir": cluster.file_storage_dir,
2577
      }
2578

    
2579
    return result
2580

    
2581

    
2582
class LUQueryConfigValues(NoHooksLU):
2583
  """Return configuration values.
2584

2585
  """
2586
  _OP_REQP = []
2587
  REQ_BGL = False
2588
  _FIELDS_DYNAMIC = utils.FieldSet()
2589
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag")
2590

    
2591
  def ExpandNames(self):
2592
    self.needed_locks = {}
2593

    
2594
    _CheckOutputFields(static=self._FIELDS_STATIC,
2595
                       dynamic=self._FIELDS_DYNAMIC,
2596
                       selected=self.op.output_fields)
2597

    
2598
  def CheckPrereq(self):
2599
    """No prerequisites.
2600

2601
    """
2602
    pass
2603

    
2604
  def Exec(self, feedback_fn):
2605
    """Dump a representation of the cluster config to the standard output.
2606

2607
    """
2608
    values = []
2609
    for field in self.op.output_fields:
2610
      if field == "cluster_name":
2611
        entry = self.cfg.GetClusterName()
2612
      elif field == "master_node":
2613
        entry = self.cfg.GetMasterNode()
2614
      elif field == "drain_flag":
2615
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
2616
      else:
2617
        raise errors.ParameterError(field)
2618
      values.append(entry)
2619
    return values
2620

    
2621

    
2622
class LUActivateInstanceDisks(NoHooksLU):
2623
  """Bring up an instance's disks.
2624

2625
  """
2626
  _OP_REQP = ["instance_name"]
2627
  REQ_BGL = False
2628

    
2629
  def ExpandNames(self):
2630
    self._ExpandAndLockInstance()
2631
    self.needed_locks[locking.LEVEL_NODE] = []
2632
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2633

    
2634
  def DeclareLocks(self, level):
2635
    if level == locking.LEVEL_NODE:
2636
      self._LockInstancesNodes()
2637

    
2638
  def CheckPrereq(self):
2639
    """Check prerequisites.
2640

2641
    This checks that the instance is in the cluster.
2642

2643
    """
2644
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2645
    assert self.instance is not None, \
2646
      "Cannot retrieve locked instance %s" % self.op.instance_name
2647
    _CheckNodeOnline(self, self.instance.primary_node)
2648

    
2649
  def Exec(self, feedback_fn):
2650
    """Activate the disks.
2651

2652
    """
2653
    disks_ok, disks_info = _AssembleInstanceDisks(self, self.instance)
2654
    if not disks_ok:
2655
      raise errors.OpExecError("Cannot activate block devices")
2656

    
2657
    return disks_info
2658

    
2659

    
2660
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False):
2661
  """Prepare the block devices for an instance.
2662

2663
  This sets up the block devices on all nodes.
2664

2665
  @type lu: L{LogicalUnit}
2666
  @param lu: the logical unit on whose behalf we execute
2667
  @type instance: L{objects.Instance}
2668
  @param instance: the instance for whose disks we assemble
2669
  @type ignore_secondaries: boolean
2670
  @param ignore_secondaries: if true, errors on secondary nodes
2671
      won't result in an error return from the function
2672
  @return: False if the operation failed, otherwise a list of
2673
      (host, instance_visible_name, node_visible_name)
2674
      with the mapping from node devices to instance devices
2675

2676
  """
2677
  device_info = []
2678
  disks_ok = True
2679
  iname = instance.name
2680
  # With the two passes mechanism we try to reduce the window of
2681
  # opportunity for the race condition of switching DRBD to primary
2682
  # before handshaking occured, but we do not eliminate it
2683

    
2684
  # The proper fix would be to wait (with some limits) until the
2685
  # connection has been made and drbd transitions from WFConnection
2686
  # into any other network-connected state (Connected, SyncTarget,
2687
  # SyncSource, etc.)
2688

    
2689
  # 1st pass, assemble on all nodes in secondary mode
2690
  for inst_disk in instance.disks:
2691
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2692
      lu.cfg.SetDiskID(node_disk, node)
2693
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
2694
      msg = result.RemoteFailMsg()
2695
      if msg:
2696
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2697
                           " (is_primary=False, pass=1): %s",
2698
                           inst_disk.iv_name, node, msg)
2699
        if not ignore_secondaries:
2700
          disks_ok = False
2701

    
2702
  # FIXME: race condition on drbd migration to primary
2703

    
2704
  # 2nd pass, do only the primary node
2705
  for inst_disk in instance.disks:
2706
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2707
      if node != instance.primary_node:
2708
        continue
2709
      lu.cfg.SetDiskID(node_disk, node)
2710
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
2711
      msg = result.RemoteFailMsg()
2712
      if msg:
2713
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2714
                           " (is_primary=True, pass=2): %s",
2715
                           inst_disk.iv_name, node, msg)
2716
        disks_ok = False
2717
    device_info.append((instance.primary_node, inst_disk.iv_name,
2718
                        result.payload))
2719

    
2720
  # leave the disks configured for the primary node
2721
  # this is a workaround that would be fixed better by
2722
  # improving the logical/physical id handling
2723
  for disk in instance.disks:
2724
    lu.cfg.SetDiskID(disk, instance.primary_node)
2725

    
2726
  return disks_ok, device_info
2727

    
2728

    
2729
def _StartInstanceDisks(lu, instance, force):
2730
  """Start the disks of an instance.
2731

2732
  """
2733
  disks_ok, dummy = _AssembleInstanceDisks(lu, instance,
2734
                                           ignore_secondaries=force)
2735
  if not disks_ok:
2736
    _ShutdownInstanceDisks(lu, instance)
2737
    if force is not None and not force:
2738
      lu.proc.LogWarning("", hint="If the message above refers to a"
2739
                         " secondary node,"
2740
                         " you can retry the operation using '--force'.")
2741
    raise errors.OpExecError("Disk consistency error")
2742

    
2743

    
2744
class LUDeactivateInstanceDisks(NoHooksLU):
2745
  """Shutdown an instance's disks.
2746

2747
  """
2748
  _OP_REQP = ["instance_name"]
2749
  REQ_BGL = False
2750

    
2751
  def ExpandNames(self):
2752
    self._ExpandAndLockInstance()
2753
    self.needed_locks[locking.LEVEL_NODE] = []
2754
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2755

    
2756
  def DeclareLocks(self, level):
2757
    if level == locking.LEVEL_NODE:
2758
      self._LockInstancesNodes()
2759

    
2760
  def CheckPrereq(self):
2761
    """Check prerequisites.
2762

2763
    This checks that the instance is in the cluster.
2764

2765
    """
2766
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2767
    assert self.instance is not None, \
2768
      "Cannot retrieve locked instance %s" % self.op.instance_name
2769

    
2770
  def Exec(self, feedback_fn):
2771
    """Deactivate the disks
2772

2773
    """
2774
    instance = self.instance
2775
    _SafeShutdownInstanceDisks(self, instance)
2776

    
2777

    
2778
def _SafeShutdownInstanceDisks(lu, instance):
2779
  """Shutdown block devices of an instance.
2780

2781
  This function checks if an instance is running, before calling
2782
  _ShutdownInstanceDisks.
2783

2784
  """
2785
  pnode = instance.primary_node
2786
  ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])
2787
  ins_l = ins_l[pnode]
2788
  msg = ins_l.RemoteFailMsg()
2789
  if msg:
2790
    raise errors.OpExecError("Can't contact node %s: %s" % (pnode, msg))
2791

    
2792
  if instance.name in ins_l.payload:
2793
    raise errors.OpExecError("Instance is running, can't shutdown"
2794
                             " block devices.")
2795

    
2796
  _ShutdownInstanceDisks(lu, instance)
2797

    
2798

    
2799
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
2800
  """Shutdown block devices of an instance.
2801

2802
  This does the shutdown on all nodes of the instance.
2803

2804
  If the ignore_primary is false, errors on the primary node are
2805
  ignored.
2806

2807
  """
2808
  all_result = True
2809
  for disk in instance.disks:
2810
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
2811
      lu.cfg.SetDiskID(top_disk, node)
2812
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
2813
      msg = result.RemoteFailMsg()
2814
      if msg:
2815
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
2816
                      disk.iv_name, node, msg)
2817
        if not ignore_primary or node != instance.primary_node:
2818
          all_result = False
2819
  return all_result
2820

    
2821

    
2822
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
2823
  """Checks if a node has enough free memory.
2824

2825
  This function check if a given node has the needed amount of free
2826
  memory. In case the node has less memory or we cannot get the
2827
  information from the node, this function raise an OpPrereqError
2828
  exception.
2829

2830
  @type lu: C{LogicalUnit}
2831
  @param lu: a logical unit from which we get configuration data
2832
  @type node: C{str}
2833
  @param node: the node to check
2834
  @type reason: C{str}
2835
  @param reason: string to use in the error message
2836
  @type requested: C{int}
2837
  @param requested: the amount of memory in MiB to check for
2838
  @type hypervisor_name: C{str}
2839
  @param hypervisor_name: the hypervisor to ask for memory stats
2840
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
2841
      we cannot check the node
2842

2843
  """
2844
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
2845
  msg = nodeinfo[node].RemoteFailMsg()
2846
  if msg:
2847
    raise errors.OpPrereqError("Can't get data from node %s: %s" % (node, msg))
2848
  free_mem = nodeinfo[node].payload.get('memory_free', None)
2849
  if not isinstance(free_mem, int):
2850
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
2851
                               " was '%s'" % (node, free_mem))
2852
  if requested > free_mem:
2853
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
2854
                               " needed %s MiB, available %s MiB" %
2855
                               (node, reason, requested, free_mem))
2856

    
2857

    
2858
class LUStartupInstance(LogicalUnit):
2859
  """Starts an instance.
2860

2861
  """
2862
  HPATH = "instance-start"
2863
  HTYPE = constants.HTYPE_INSTANCE
2864
  _OP_REQP = ["instance_name", "force"]
2865
  REQ_BGL = False
2866

    
2867
  def ExpandNames(self):
2868
    self._ExpandAndLockInstance()
2869

    
2870
  def BuildHooksEnv(self):
2871
    """Build hooks env.
2872

2873
    This runs on master, primary and secondary nodes of the instance.
2874

2875
    """
2876
    env = {
2877
      "FORCE": self.op.force,
2878
      }
2879
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2880
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2881
    return env, nl, nl
2882

    
2883
  def CheckPrereq(self):
2884
    """Check prerequisites.
2885

2886
    This checks that the instance is in the cluster.
2887

2888
    """
2889
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2890
    assert self.instance is not None, \
2891
      "Cannot retrieve locked instance %s" % self.op.instance_name
2892

    
2893
    # extra beparams
2894
    self.beparams = getattr(self.op, "beparams", {})
2895
    if self.beparams:
2896
      if not isinstance(self.beparams, dict):
2897
        raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
2898
                                   " dict" % (type(self.beparams), ))
2899
      # fill the beparams dict
2900
      utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
2901
      self.op.beparams = self.beparams
2902

    
2903
    # extra hvparams
2904
    self.hvparams = getattr(self.op, "hvparams", {})
2905
    if self.hvparams:
2906
      if not isinstance(self.hvparams, dict):
2907
        raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
2908
                                   " dict" % (type(self.hvparams), ))
2909

    
2910
      # check hypervisor parameter syntax (locally)
2911
      cluster = self.cfg.GetClusterInfo()
2912
      utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
2913
      filled_hvp = objects.FillDict(cluster.hvparams[instance.hypervisor],
2914
                                    instance.hvparams)
2915
      filled_hvp.update(self.hvparams)
2916
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
2917
      hv_type.CheckParameterSyntax(filled_hvp)
2918
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
2919
      self.op.hvparams = self.hvparams
2920

    
2921
    _CheckNodeOnline(self, instance.primary_node)
2922

    
2923
    bep = self.cfg.GetClusterInfo().FillBE(instance)
2924
    # check bridges existance
2925
    _CheckInstanceBridgesExist(self, instance)
2926

    
2927
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2928
                                              instance.name,
2929
                                              instance.hypervisor)
2930
    msg = remote_info.RemoteFailMsg()
2931
    if msg:
2932
      raise errors.OpPrereqError("Error checking node %s: %s" %
2933
                                 (instance.primary_node, msg))
2934
    if not remote_info.payload: # not running already
2935
      _CheckNodeFreeMemory(self, instance.primary_node,
2936
                           "starting instance %s" % instance.name,
2937
                           bep[constants.BE_MEMORY], instance.hypervisor)
2938

    
2939
  def Exec(self, feedback_fn):
2940
    """Start the instance.
2941

2942
    """
2943
    instance = self.instance
2944
    force = self.op.force
2945

    
2946
    self.cfg.MarkInstanceUp(instance.name)
2947

    
2948
    node_current = instance.primary_node
2949

    
2950
    _StartInstanceDisks(self, instance, force)
2951

    
2952
    result = self.rpc.call_instance_start(node_current, instance,
2953
                                          self.hvparams, self.beparams)
2954
    msg = result.RemoteFailMsg()
2955
    if msg:
2956
      _ShutdownInstanceDisks(self, instance)
2957
      raise errors.OpExecError("Could not start instance: %s" % msg)
2958

    
2959

    
2960
class LURebootInstance(LogicalUnit):
2961
  """Reboot an instance.
2962

2963
  """
2964
  HPATH = "instance-reboot"
2965
  HTYPE = constants.HTYPE_INSTANCE
2966
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
2967
  REQ_BGL = False
2968

    
2969
  def ExpandNames(self):
2970
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
2971
                                   constants.INSTANCE_REBOOT_HARD,
2972
                                   constants.INSTANCE_REBOOT_FULL]:
2973
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
2974
                                  (constants.INSTANCE_REBOOT_SOFT,
2975
                                   constants.INSTANCE_REBOOT_HARD,
2976
                                   constants.INSTANCE_REBOOT_FULL))
2977
    self._ExpandAndLockInstance()
2978

    
2979
  def BuildHooksEnv(self):
2980
    """Build hooks env.
2981

2982
    This runs on master, primary and secondary nodes of the instance.
2983

2984
    """
2985
    env = {
2986
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
2987
      "REBOOT_TYPE": self.op.reboot_type,
2988
      }
2989
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2990
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2991
    return env, nl, nl
2992

    
2993
  def CheckPrereq(self):
2994
    """Check prerequisites.
2995

2996
    This checks that the instance is in the cluster.
2997

2998
    """
2999
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3000
    assert self.instance is not None, \
3001
      "Cannot retrieve locked instance %s" % self.op.instance_name
3002

    
3003
    _CheckNodeOnline(self, instance.primary_node)
3004

    
3005
    # check bridges existance
3006
    _CheckInstanceBridgesExist(self, instance)
3007

    
3008
  def Exec(self, feedback_fn):
3009
    """Reboot the instance.
3010

3011
    """
3012
    instance = self.instance
3013
    ignore_secondaries = self.op.ignore_secondaries
3014
    reboot_type = self.op.reboot_type
3015

    
3016
    node_current = instance.primary_node
3017

    
3018
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
3019
                       constants.INSTANCE_REBOOT_HARD]:
3020
      for disk in instance.disks:
3021
        self.cfg.SetDiskID(disk, node_current)
3022
      result = self.rpc.call_instance_reboot(node_current, instance,
3023
                                             reboot_type)
3024
      msg = result.RemoteFailMsg()
3025
      if msg:
3026
        raise errors.OpExecError("Could not reboot instance: %s" % msg)
3027
    else:
3028
      result = self.rpc.call_instance_shutdown(node_current, instance)
3029
      msg = result.RemoteFailMsg()
3030
      if msg:
3031
        raise errors.OpExecError("Could not shutdown instance for"
3032
                                 " full reboot: %s" % msg)
3033
      _ShutdownInstanceDisks(self, instance)
3034
      _StartInstanceDisks(self, instance, ignore_secondaries)
3035
      result = self.rpc.call_instance_start(node_current, instance, None, None)
3036
      msg = result.RemoteFailMsg()
3037
      if msg:
3038
        _ShutdownInstanceDisks(self, instance)
3039
        raise errors.OpExecError("Could not start instance for"
3040
                                 " full reboot: %s" % msg)
3041

    
3042
    self.cfg.MarkInstanceUp(instance.name)
3043

    
3044

    
3045
class LUShutdownInstance(LogicalUnit):
3046
  """Shutdown an instance.
3047

3048
  """
3049
  HPATH = "instance-stop"
3050
  HTYPE = constants.HTYPE_INSTANCE
3051
  _OP_REQP = ["instance_name"]
3052
  REQ_BGL = False
3053

    
3054
  def ExpandNames(self):
3055
    self._ExpandAndLockInstance()
3056

    
3057
  def BuildHooksEnv(self):
3058
    """Build hooks env.
3059

3060
    This runs on master, primary and secondary nodes of the instance.
3061

3062
    """
3063
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3064
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3065
    return env, nl, nl
3066

    
3067
  def CheckPrereq(self):
3068
    """Check prerequisites.
3069

3070
    This checks that the instance is in the cluster.
3071

3072
    """
3073
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3074
    assert self.instance is not None, \
3075
      "Cannot retrieve locked instance %s" % self.op.instance_name
3076
    _CheckNodeOnline(self, self.instance.primary_node)
3077

    
3078
  def Exec(self, feedback_fn):
3079
    """Shutdown the instance.
3080

3081
    """
3082
    instance = self.instance
3083
    node_current = instance.primary_node
3084
    self.cfg.MarkInstanceDown(instance.name)
3085
    result = self.rpc.call_instance_shutdown(node_current, instance)
3086
    msg = result.RemoteFailMsg()
3087
    if msg:
3088
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
3089

    
3090
    _ShutdownInstanceDisks(self, instance)
3091

    
3092

    
3093
class LUReinstallInstance(LogicalUnit):
3094
  """Reinstall an instance.
3095

3096
  """
3097
  HPATH = "instance-reinstall"
3098
  HTYPE = constants.HTYPE_INSTANCE
3099
  _OP_REQP = ["instance_name"]
3100
  REQ_BGL = False
3101

    
3102
  def ExpandNames(self):
3103
    self._ExpandAndLockInstance()
3104

    
3105
  def BuildHooksEnv(self):
3106
    """Build hooks env.
3107

3108
    This runs on master, primary and secondary nodes of the instance.
3109

3110
    """
3111
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3112
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3113
    return env, nl, nl
3114

    
3115
  def CheckPrereq(self):
3116
    """Check prerequisites.
3117

3118
    This checks that the instance is in the cluster and is not running.
3119

3120
    """
3121
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3122
    assert instance is not None, \
3123
      "Cannot retrieve locked instance %s" % self.op.instance_name
3124
    _CheckNodeOnline(self, instance.primary_node)
3125

    
3126
    if instance.disk_template == constants.DT_DISKLESS:
3127
      raise errors.OpPrereqError("Instance '%s' has no disks" %
3128
                                 self.op.instance_name)
3129
    if instance.admin_up:
3130
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3131
                                 self.op.instance_name)
3132
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3133
                                              instance.name,
3134
                                              instance.hypervisor)
3135
    msg = remote_info.RemoteFailMsg()
3136
    if msg:
3137
      raise errors.OpPrereqError("Error checking node %s: %s" %
3138
                                 (instance.primary_node, msg))
3139
    if remote_info.payload:
3140
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3141
                                 (self.op.instance_name,
3142
                                  instance.primary_node))
3143

    
3144
    self.op.os_type = getattr(self.op, "os_type", None)
3145
    if self.op.os_type is not None:
3146
      # OS verification
3147
      pnode = self.cfg.GetNodeInfo(
3148
        self.cfg.ExpandNodeName(instance.primary_node))
3149
      if pnode is None:
3150
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
3151
                                   self.op.pnode)
3152
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
3153
      result.Raise()
3154
      if not isinstance(result.data, objects.OS):
3155
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
3156
                                   " primary node"  % self.op.os_type)
3157

    
3158
    self.instance = instance
3159

    
3160
  def Exec(self, feedback_fn):
3161
    """Reinstall the instance.
3162

3163
    """
3164
    inst = self.instance
3165

    
3166
    if self.op.os_type is not None:
3167
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
3168
      inst.os = self.op.os_type
3169
      self.cfg.Update(inst)
3170

    
3171
    _StartInstanceDisks(self, inst, None)
3172
    try:
3173
      feedback_fn("Running the instance OS create scripts...")
3174
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True)
3175
      msg = result.RemoteFailMsg()
3176
      if msg:
3177
        raise errors.OpExecError("Could not install OS for instance %s"
3178
                                 " on node %s: %s" %
3179
                                 (inst.name, inst.primary_node, msg))
3180
    finally:
3181
      _ShutdownInstanceDisks(self, inst)
3182

    
3183

    
3184
class LURenameInstance(LogicalUnit):
3185
  """Rename an instance.
3186

3187
  """
3188
  HPATH = "instance-rename"
3189
  HTYPE = constants.HTYPE_INSTANCE
3190
  _OP_REQP = ["instance_name", "new_name"]
3191

    
3192
  def BuildHooksEnv(self):
3193
    """Build hooks env.
3194

3195
    This runs on master, primary and secondary nodes of the instance.
3196

3197
    """
3198
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3199
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3200
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3201
    return env, nl, nl
3202

    
3203
  def CheckPrereq(self):
3204
    """Check prerequisites.
3205

3206
    This checks that the instance is in the cluster and is not running.
3207

3208
    """
3209
    instance = self.cfg.GetInstanceInfo(
3210
      self.cfg.ExpandInstanceName(self.op.instance_name))
3211
    if instance is None:
3212
      raise errors.OpPrereqError("Instance '%s' not known" %
3213
                                 self.op.instance_name)
3214
    _CheckNodeOnline(self, instance.primary_node)
3215

    
3216
    if instance.admin_up:
3217
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3218
                                 self.op.instance_name)
3219
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3220
                                              instance.name,
3221
                                              instance.hypervisor)
3222
    msg = remote_info.RemoteFailMsg()
3223
    if msg:
3224
      raise errors.OpPrereqError("Error checking node %s: %s" %
3225
                                 (instance.primary_node, msg))
3226
    if remote_info.payload:
3227
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3228
                                 (self.op.instance_name,
3229
                                  instance.primary_node))
3230
    self.instance = instance
3231

    
3232
    # new name verification
3233
    name_info = utils.HostInfo(self.op.new_name)
3234

    
3235
    self.op.new_name = new_name = name_info.name
3236
    instance_list = self.cfg.GetInstanceList()
3237
    if new_name in instance_list:
3238
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3239
                                 new_name)
3240

    
3241
    if not getattr(self.op, "ignore_ip", False):
3242
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3243
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3244
                                   (name_info.ip, new_name))
3245

    
3246

    
3247
  def Exec(self, feedback_fn):
3248
    """Reinstall the instance.
3249

3250
    """
3251
    inst = self.instance
3252
    old_name = inst.name
3253

    
3254
    if inst.disk_template == constants.DT_FILE:
3255
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3256

    
3257
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3258
    # Change the instance lock. This is definitely safe while we hold the BGL
3259
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3260
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3261

    
3262
    # re-read the instance from the configuration after rename
3263
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3264

    
3265
    if inst.disk_template == constants.DT_FILE:
3266
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3267
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3268
                                                     old_file_storage_dir,
3269
                                                     new_file_storage_dir)
3270
      result.Raise()
3271
      if not result.data:
3272
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3273
                                 " directory '%s' to '%s' (but the instance"
3274
                                 " has been renamed in Ganeti)" % (
3275
                                 inst.primary_node, old_file_storage_dir,
3276
                                 new_file_storage_dir))
3277

    
3278
      if not result.data[0]:
3279
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3280
                                 " (but the instance has been renamed in"
3281
                                 " Ganeti)" % (old_file_storage_dir,
3282
                                               new_file_storage_dir))
3283

    
3284
    _StartInstanceDisks(self, inst, None)
3285
    try:
3286
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3287
                                                 old_name)
3288
      msg = result.RemoteFailMsg()
3289
      if msg:
3290
        msg = ("Could not run OS rename script for instance %s on node %s"
3291
               " (but the instance has been renamed in Ganeti): %s" %
3292
               (inst.name, inst.primary_node, msg))
3293
        self.proc.LogWarning(msg)
3294
    finally:
3295
      _ShutdownInstanceDisks(self, inst)
3296

    
3297

    
3298
class LURemoveInstance(LogicalUnit):
3299
  """Remove an instance.
3300

3301
  """
3302
  HPATH = "instance-remove"
3303
  HTYPE = constants.HTYPE_INSTANCE
3304
  _OP_REQP = ["instance_name", "ignore_failures"]
3305
  REQ_BGL = False
3306

    
3307
  def ExpandNames(self):
3308
    self._ExpandAndLockInstance()
3309
    self.needed_locks[locking.LEVEL_NODE] = []
3310
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3311

    
3312
  def DeclareLocks(self, level):
3313
    if level == locking.LEVEL_NODE:
3314
      self._LockInstancesNodes()
3315

    
3316
  def BuildHooksEnv(self):
3317
    """Build hooks env.
3318

3319
    This runs on master, primary and secondary nodes of the instance.
3320

3321
    """
3322
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3323
    nl = [self.cfg.GetMasterNode()]
3324
    return env, nl, nl
3325

    
3326
  def CheckPrereq(self):
3327
    """Check prerequisites.
3328

3329
    This checks that the instance is in the cluster.
3330

3331
    """
3332
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3333
    assert self.instance is not None, \
3334
      "Cannot retrieve locked instance %s" % self.op.instance_name
3335

    
3336
  def Exec(self, feedback_fn):
3337
    """Remove the instance.
3338

3339
    """
3340
    instance = self.instance
3341
    logging.info("Shutting down instance %s on node %s",
3342
                 instance.name, instance.primary_node)
3343

    
3344
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3345
    msg = result.RemoteFailMsg()
3346
    if msg:
3347
      if self.op.ignore_failures:
3348
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3349
      else:
3350
        raise errors.OpExecError("Could not shutdown instance %s on"
3351
                                 " node %s: %s" %
3352
                                 (instance.name, instance.primary_node, msg))
3353

    
3354
    logging.info("Removing block devices for instance %s", instance.name)
3355

    
3356
    if not _RemoveDisks(self, instance):
3357
      if self.op.ignore_failures:
3358
        feedback_fn("Warning: can't remove instance's disks")
3359
      else:
3360
        raise errors.OpExecError("Can't remove instance's disks")
3361

    
3362
    logging.info("Removing instance %s out of cluster config", instance.name)
3363

    
3364
    self.cfg.RemoveInstance(instance.name)
3365
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3366

    
3367

    
3368
class LUQueryInstances(NoHooksLU):
3369
  """Logical unit for querying instances.
3370

3371
  """
3372
  _OP_REQP = ["output_fields", "names", "use_locking"]
3373
  REQ_BGL = False
3374
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3375
                                    "admin_state",
3376
                                    "disk_template", "ip", "mac", "bridge",
3377
                                    "sda_size", "sdb_size", "vcpus", "tags",
3378
                                    "network_port", "beparams",
3379
                                    r"(disk)\.(size)/([0-9]+)",
3380
                                    r"(disk)\.(sizes)", "disk_usage",
3381
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3382
                                    r"(nic)\.(macs|ips|bridges)",
3383
                                    r"(disk|nic)\.(count)",
3384
                                    "serial_no", "hypervisor", "hvparams",] +
3385
                                  ["hv/%s" % name
3386
                                   for name in constants.HVS_PARAMETERS] +
3387
                                  ["be/%s" % name
3388
                                   for name in constants.BES_PARAMETERS])
3389
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3390

    
3391

    
3392
  def ExpandNames(self):
3393
    _CheckOutputFields(static=self._FIELDS_STATIC,
3394
                       dynamic=self._FIELDS_DYNAMIC,
3395
                       selected=self.op.output_fields)
3396

    
3397
    self.needed_locks = {}
3398
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3399
    self.share_locks[locking.LEVEL_NODE] = 1
3400

    
3401
    if self.op.names:
3402
      self.wanted = _GetWantedInstances(self, self.op.names)
3403
    else:
3404
      self.wanted = locking.ALL_SET
3405

    
3406
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3407
    self.do_locking = self.do_node_query and self.op.use_locking
3408
    if self.do_locking:
3409
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3410
      self.needed_locks[locking.LEVEL_NODE] = []
3411
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3412

    
3413
  def DeclareLocks(self, level):
3414
    if level == locking.LEVEL_NODE and self.do_locking:
3415
      self._LockInstancesNodes()
3416

    
3417
  def CheckPrereq(self):
3418
    """Check prerequisites.
3419

3420
    """
3421
    pass
3422

    
3423
  def Exec(self, feedback_fn):
3424
    """Computes the list of nodes and their attributes.
3425

3426
    """
3427
    all_info = self.cfg.GetAllInstancesInfo()
3428
    if self.wanted == locking.ALL_SET:
3429
      # caller didn't specify instance names, so ordering is not important
3430
      if self.do_locking:
3431
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3432
      else:
3433
        instance_names = all_info.keys()
3434
      instance_names = utils.NiceSort(instance_names)
3435
    else:
3436
      # caller did specify names, so we must keep the ordering
3437
      if self.do_locking:
3438
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3439
      else:
3440
        tgt_set = all_info.keys()
3441
      missing = set(self.wanted).difference(tgt_set)
3442
      if missing:
3443
        raise errors.OpExecError("Some instances were removed before"
3444
                                 " retrieving their data: %s" % missing)
3445
      instance_names = self.wanted
3446

    
3447
    instance_list = [all_info[iname] for iname in instance_names]
3448

    
3449
    # begin data gathering
3450

    
3451
    nodes = frozenset([inst.primary_node for inst in instance_list])
3452
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3453

    
3454
    bad_nodes = []
3455
    off_nodes = []
3456
    if self.do_node_query:
3457
      live_data = {}
3458
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3459
      for name in nodes:
3460
        result = node_data[name]
3461
        if result.offline:
3462
          # offline nodes will be in both lists
3463
          off_nodes.append(name)
3464
        if result.failed or result.RemoteFailMsg():
3465
          bad_nodes.append(name)
3466
        else:
3467
          if result.payload:
3468
            live_data.update(result.payload)
3469
          # else no instance is alive
3470
    else:
3471
      live_data = dict([(name, {}) for name in instance_names])
3472

    
3473
    # end data gathering
3474

    
3475
    HVPREFIX = "hv/"
3476
    BEPREFIX = "be/"
3477
    output = []
3478
    for instance in instance_list:
3479
      iout = []
3480
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3481
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3482
      for field in self.op.output_fields:
3483
        st_match = self._FIELDS_STATIC.Matches(field)
3484
        if field == "name":
3485
          val = instance.name
3486
        elif field == "os":
3487
          val = instance.os
3488
        elif field == "pnode":
3489
          val = instance.primary_node
3490
        elif field == "snodes":
3491
          val = list(instance.secondary_nodes)
3492
        elif field == "admin_state":
3493
          val = instance.admin_up
3494
        elif field == "oper_state":
3495
          if instance.primary_node in bad_nodes:
3496
            val = None
3497
          else:
3498
            val = bool(live_data.get(instance.name))
3499
        elif field == "status":
3500
          if instance.primary_node in off_nodes:
3501
            val = "ERROR_nodeoffline"
3502
          elif instance.primary_node in bad_nodes:
3503
            val = "ERROR_nodedown"
3504
          else:
3505
            running = bool(live_data.get(instance.name))
3506
            if running:
3507
              if instance.admin_up:
3508
                val = "running"
3509
              else:
3510
                val = "ERROR_up"
3511
            else:
3512
              if instance.admin_up:
3513
                val = "ERROR_down"
3514
              else:
3515
                val = "ADMIN_down"
3516
        elif field == "oper_ram":
3517
          if instance.primary_node in bad_nodes:
3518
            val = None
3519
          elif instance.name in live_data:
3520
            val = live_data[instance.name].get("memory", "?")
3521
          else:
3522
            val = "-"
3523
        elif field == "disk_template":
3524
          val = instance.disk_template
3525
        elif field == "ip":
3526
          val = instance.nics[0].ip
3527
        elif field == "bridge":
3528
          val = instance.nics[0].bridge
3529
        elif field == "mac":
3530
          val = instance.nics[0].mac
3531
        elif field == "sda_size" or field == "sdb_size":
3532
          idx = ord(field[2]) - ord('a')
3533
          try:
3534
            val = instance.FindDisk(idx).size
3535
          except errors.OpPrereqError:
3536
            val = None
3537
        elif field == "disk_usage": # total disk usage per node
3538
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3539
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3540
        elif field == "tags":
3541
          val = list(instance.GetTags())
3542
        elif field == "serial_no":
3543
          val = instance.serial_no
3544
        elif field == "network_port":
3545
          val = instance.network_port
3546
        elif field == "hypervisor":
3547
          val = instance.hypervisor
3548
        elif field == "hvparams":
3549
          val = i_hv
3550
        elif (field.startswith(HVPREFIX) and
3551
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3552
          val = i_hv.get(field[len(HVPREFIX):], None)
3553
        elif field == "beparams":
3554
          val = i_be
3555
        elif (field.startswith(BEPREFIX) and
3556
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3557
          val = i_be.get(field[len(BEPREFIX):], None)
3558
        elif st_match and st_match.groups():
3559
          # matches a variable list
3560
          st_groups = st_match.groups()
3561
          if st_groups and st_groups[0] == "disk":
3562
            if st_groups[1] == "count":
3563
              val = len(instance.disks)
3564
            elif st_groups[1] == "sizes":
3565
              val = [disk.size for disk in instance.disks]
3566
            elif st_groups[1] == "size":
3567
              try:
3568
                val = instance.FindDisk(st_groups[2]).size
3569
              except errors.OpPrereqError:
3570
                val = None
3571
            else:
3572
              assert False, "Unhandled disk parameter"
3573
          elif st_groups[0] == "nic":
3574
            if st_groups[1] == "count":
3575
              val = len(instance.nics)
3576
            elif st_groups[1] == "macs":
3577
              val = [nic.mac for nic in instance.nics]
3578
            elif st_groups[1] == "ips":
3579
              val = [nic.ip for nic in instance.nics]
3580
            elif st_groups[1] == "bridges":
3581
              val = [nic.bridge for nic in instance.nics]
3582
            else:
3583
              # index-based item
3584
              nic_idx = int(st_groups[2])
3585
              if nic_idx >= len(instance.nics):
3586
                val = None
3587
              else:
3588
                if st_groups[1] == "mac":
3589
                  val = instance.nics[nic_idx].mac
3590
                elif st_groups[1] == "ip":
3591
                  val = instance.nics[nic_idx].ip
3592
                elif st_groups[1] == "bridge":
3593
                  val = instance.nics[nic_idx].bridge
3594
                else:
3595
                  assert False, "Unhandled NIC parameter"
3596
          else:
3597
            assert False, "Unhandled variable parameter"
3598
        else:
3599
          raise errors.ParameterError(field)
3600
        iout.append(val)
3601
      output.append(iout)
3602

    
3603
    return output
3604

    
3605

    
3606
class LUFailoverInstance(LogicalUnit):
3607
  """Failover an instance.
3608

3609
  """
3610
  HPATH = "instance-failover"
3611
  HTYPE = constants.HTYPE_INSTANCE
3612
  _OP_REQP = ["instance_name", "ignore_consistency"]
3613
  REQ_BGL = False
3614

    
3615
  def ExpandNames(self):
3616
    self._ExpandAndLockInstance()
3617
    self.needed_locks[locking.LEVEL_NODE] = []
3618
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3619

    
3620
  def DeclareLocks(self, level):
3621
    if level == locking.LEVEL_NODE:
3622
      self._LockInstancesNodes()
3623

    
3624
  def BuildHooksEnv(self):
3625
    """Build hooks env.
3626

3627
    This runs on master, primary and secondary nodes of the instance.
3628

3629
    """
3630
    env = {
3631
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3632
      }
3633
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3634
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3635
    return env, nl, nl
3636

    
3637
  def CheckPrereq(self):
3638
    """Check prerequisites.
3639

3640
    This checks that the instance is in the cluster.
3641

3642
    """
3643
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3644
    assert self.instance is not None, \
3645
      "Cannot retrieve locked instance %s" % self.op.instance_name
3646

    
3647
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3648
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3649
      raise errors.OpPrereqError("Instance's disk layout is not"
3650
                                 " network mirrored, cannot failover.")
3651

    
3652
    secondary_nodes = instance.secondary_nodes
3653
    if not secondary_nodes:
3654
      raise errors.ProgrammerError("no secondary node but using "
3655
                                   "a mirrored disk template")
3656

    
3657
    target_node = secondary_nodes[0]
3658
    _CheckNodeOnline(self, target_node)
3659
    _CheckNodeNotDrained(self, target_node)
3660
    # check memory requirements on the secondary node
3661
    _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3662
                         instance.name, bep[constants.BE_MEMORY],
3663
                         instance.hypervisor)
3664
    # check bridge existance
3665
    _CheckInstanceBridgesExist(self, instance, node=target_node)
3666

    
3667
  def Exec(self, feedback_fn):
3668
    """Failover an instance.
3669

3670
    The failover is done by shutting it down on its present node and
3671
    starting it on the secondary.
3672

3673
    """
3674
    instance = self.instance
3675

    
3676
    source_node = instance.primary_node
3677
    target_node = instance.secondary_nodes[0]
3678

    
3679
    feedback_fn("* checking disk consistency between source and target")
3680
    for dev in instance.disks:
3681
      # for drbd, these are drbd over lvm
3682
      if not _CheckDiskConsistency(self, dev, target_node, False):
3683
        if instance.admin_up and not self.op.ignore_consistency:
3684
          raise errors.OpExecError("Disk %s is degraded on target node,"
3685
                                   " aborting failover." % dev.iv_name)
3686

    
3687
    feedback_fn("* shutting down instance on source node")
3688
    logging.info("Shutting down instance %s on node %s",
3689
                 instance.name, source_node)
3690

    
3691
    result = self.rpc.call_instance_shutdown(source_node, instance)
3692
    msg = result.RemoteFailMsg()
3693
    if msg:
3694
      if self.op.ignore_consistency:
3695
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3696
                             " Proceeding anyway. Please make sure node"
3697
                             " %s is down. Error details: %s",
3698
                             instance.name, source_node, source_node, msg)
3699
      else:
3700
        raise errors.OpExecError("Could not shutdown instance %s on"
3701
                                 " node %s: %s" %
3702
                                 (instance.name, source_node, msg))
3703

    
3704
    feedback_fn("* deactivating the instance's disks on source node")
3705
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3706
      raise errors.OpExecError("Can't shut down the instance's disks.")
3707

    
3708
    instance.primary_node = target_node
3709
    # distribute new instance config to the other nodes
3710
    self.cfg.Update(instance)
3711

    
3712
    # Only start the instance if it's marked as up
3713
    if instance.admin_up:
3714
      feedback_fn("* activating the instance's disks on target node")
3715
      logging.info("Starting instance %s on node %s",
3716
                   instance.name, target_node)
3717

    
3718
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3719
                                               ignore_secondaries=True)
3720
      if not disks_ok:
3721
        _ShutdownInstanceDisks(self, instance)
3722
        raise errors.OpExecError("Can't activate the instance's disks")
3723

    
3724
      feedback_fn("* starting the instance on the target node")
3725
      result = self.rpc.call_instance_start(target_node, instance, None, None)
3726
      msg = result.RemoteFailMsg()
3727
      if msg:
3728
        _ShutdownInstanceDisks(self, instance)
3729
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3730
                                 (instance.name, target_node, msg))
3731

    
3732

    
3733
class LUMigrateInstance(LogicalUnit):
3734
  """Migrate an instance.
3735

3736
  This is migration without shutting down, compared to the failover,
3737
  which is done with shutdown.
3738

3739
  """
3740
  HPATH = "instance-migrate"
3741
  HTYPE = constants.HTYPE_INSTANCE
3742
  _OP_REQP = ["instance_name", "live", "cleanup"]
3743

    
3744
  REQ_BGL = False
3745

    
3746
  def ExpandNames(self):
3747
    self._ExpandAndLockInstance()
3748
    self.needed_locks[locking.LEVEL_NODE] = []
3749
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3750

    
3751
  def DeclareLocks(self, level):
3752
    if level == locking.LEVEL_NODE:
3753
      self._LockInstancesNodes()
3754

    
3755
  def BuildHooksEnv(self):
3756
    """Build hooks env.
3757

3758
    This runs on master, primary and secondary nodes of the instance.
3759

3760
    """
3761
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3762
    env["MIGRATE_LIVE"] = self.op.live
3763
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3764
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3765
    return env, nl, nl
3766

    
3767
  def CheckPrereq(self):
3768
    """Check prerequisites.
3769

3770
    This checks that the instance is in the cluster.
3771

3772
    """
3773
    instance = self.cfg.GetInstanceInfo(
3774
      self.cfg.ExpandInstanceName(self.op.instance_name))
3775
    if instance is None:
3776
      raise errors.OpPrereqError("Instance '%s' not known" %
3777
                                 self.op.instance_name)
3778

    
3779
    if instance.disk_template != constants.DT_DRBD8:
3780
      raise errors.OpPrereqError("Instance's disk layout is not"
3781
                                 " drbd8, cannot migrate.")
3782

    
3783
    secondary_nodes = instance.secondary_nodes
3784
    if not secondary_nodes:
3785
      raise errors.ConfigurationError("No secondary node but using"
3786
                                      " drbd8 disk template")
3787

    
3788
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3789

    
3790
    target_node = secondary_nodes[0]
3791
    # check memory requirements on the secondary node
3792
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3793
                         instance.name, i_be[constants.BE_MEMORY],
3794
                         instance.hypervisor)
3795

    
3796
    # check bridge existance
3797
    _CheckInstanceBridgesExist(self, instance, node=target_node)
3798

    
3799
    if not self.op.cleanup:
3800
      _CheckNodeNotDrained(self, target_node)
3801
      result = self.rpc.call_instance_migratable(instance.primary_node,
3802
                                                 instance)
3803
      msg = result.RemoteFailMsg()
3804
      if msg:
3805
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3806
                                   msg)
3807

    
3808
    self.instance = instance
3809

    
3810
  def _WaitUntilSync(self):
3811
    """Poll with custom rpc for disk sync.
3812

3813
    This uses our own step-based rpc call.
3814

3815
    """
3816
    self.feedback_fn("* wait until resync is done")
3817
    all_done = False
3818
    while not all_done:
3819
      all_done = True
3820
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3821
                                            self.nodes_ip,
3822
                                            self.instance.disks)
3823
      min_percent = 100
3824
      for node, nres in result.items():
3825
        msg = nres.RemoteFailMsg()
3826
        if msg:
3827
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3828
                                   (node, msg))
3829
        node_done, node_percent = nres.payload
3830
        all_done = all_done and node_done
3831
        if node_percent is not None:
3832
          min_percent = min(min_percent, node_percent)
3833
      if not all_done:
3834
        if min_percent < 100:
3835
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3836
        time.sleep(2)
3837

    
3838
  def _EnsureSecondary(self, node):
3839
    """Demote a node to secondary.
3840

3841
    """
3842
    self.feedback_fn("* switching node %s to secondary mode" % node)
3843

    
3844
    for dev in self.instance.disks:
3845
      self.cfg.SetDiskID(dev, node)
3846

    
3847
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3848
                                          self.instance.disks)
3849
    msg = result.RemoteFailMsg()
3850
    if msg:
3851
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3852
                               " error %s" % (node, msg))
3853

    
3854
  def _GoStandalone(self):
3855
    """Disconnect from the network.
3856

3857
    """
3858
    self.feedback_fn("* changing into standalone mode")
3859
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3860
                                               self.instance.disks)
3861
    for node, nres in result.items():
3862
      msg = nres.RemoteFailMsg()
3863
      if msg:
3864
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3865
                                 " error %s" % (node, msg))
3866

    
3867
  def _GoReconnect(self, multimaster):
3868
    """Reconnect to the network.
3869

3870
    """
3871
    if multimaster:
3872
      msg = "dual-master"
3873
    else:
3874
      msg = "single-master"
3875
    self.feedback_fn("* changing disks into %s mode" % msg)
3876
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3877
                                           self.instance.disks,
3878
                                           self.instance.name, multimaster)
3879
    for node, nres in result.items():
3880
      msg = nres.RemoteFailMsg()
3881
      if msg:
3882
        raise errors.OpExecError("Cannot change disks config on node %s,"
3883
                                 " error: %s" % (node, msg))
3884

    
3885
  def _ExecCleanup(self):
3886
    """Try to cleanup after a failed migration.
3887

3888
    The cleanup is done by:
3889
      - check that the instance is running only on one node
3890
        (and update the config if needed)
3891
      - change disks on its secondary node to secondary
3892
      - wait until disks are fully synchronized
3893
      - disconnect from the network
3894
      - change disks into single-master mode
3895
      - wait again until disks are fully synchronized
3896

3897
    """
3898
    instance = self.instance
3899
    target_node = self.target_node
3900
    source_node = self.source_node
3901

    
3902
    # check running on only one node
3903
    self.feedback_fn("* checking where the instance actually runs"
3904
                     " (if this hangs, the hypervisor might be in"
3905
                     " a bad state)")
3906
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3907
    for node, result in ins_l.items():
3908
      msg = result.RemoteFailMsg()
3909
      if msg:
3910
        raise errors.OpExecError("Can't contact node %s: %s" % (node, msg))
3911

    
3912
    runningon_source = instance.name in ins_l[source_node].payload
3913
    runningon_target = instance.name in ins_l[target_node].payload
3914

    
3915
    if runningon_source and runningon_target:
3916
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3917
                               " or the hypervisor is confused. You will have"
3918
                               " to ensure manually that it runs only on one"
3919
                               " and restart this operation.")
3920

    
3921
    if not (runningon_source or runningon_target):
3922
      raise errors.OpExecError("Instance does not seem to be running at all."
3923
                               " In this case, it's safer to repair by"
3924
                               " running 'gnt-instance stop' to ensure disk"
3925
                               " shutdown, and then restarting it.")
3926

    
3927
    if runningon_target:
3928
      # the migration has actually succeeded, we need to update the config
3929
      self.feedback_fn("* instance running on secondary node (%s),"
3930
                       " updating config" % target_node)
3931
      instance.primary_node = target_node
3932
      self.cfg.Update(instance)
3933
      demoted_node = source_node
3934
    else:
3935
      self.feedback_fn("* instance confirmed to be running on its"
3936
                       " primary node (%s)" % source_node)
3937
      demoted_node = target_node
3938

    
3939
    self._EnsureSecondary(demoted_node)
3940
    try:
3941
      self._WaitUntilSync()
3942
    except errors.OpExecError:
3943
      # we ignore here errors, since if the device is standalone, it
3944
      # won't be able to sync
3945
      pass
3946
    self._GoStandalone()
3947
    self._GoReconnect(False)
3948
    self._WaitUntilSync()
3949

    
3950
    self.feedback_fn("* done")
3951

    
3952
  def _RevertDiskStatus(self):
3953
    """Try to revert the disk status after a failed migration.
3954

3955
    """
3956
    target_node = self.target_node
3957
    try:
3958
      self._EnsureSecondary(target_node)
3959
      self._GoStandalone()
3960
      self._GoReconnect(False)
3961
      self._WaitUntilSync()
3962
    except errors.OpExecError, err:
3963
      self.LogWarning("Migration failed and I can't reconnect the"
3964
                      " drives: error '%s'\n"
3965
                      "Please look and recover the instance status" %
3966
                      str(err))
3967

    
3968
  def _AbortMigration(self):
3969
    """Call the hypervisor code to abort a started migration.
3970

3971
    """
3972
    instance = self.instance
3973
    target_node = self.target_node
3974
    migration_info = self.migration_info
3975

    
3976
    abort_result = self.rpc.call_finalize_migration(target_node,
3977
                                                    instance,
3978
                                                    migration_info,
3979
                                                    False)
3980
    abort_msg = abort_result.RemoteFailMsg()
3981
    if abort_msg:
3982
      logging.error("Aborting migration failed on target node %s: %s" %
3983
                    (target_node, abort_msg))
3984
      # Don't raise an exception here, as we stil have to try to revert the
3985
      # disk status, even if this step failed.
3986

    
3987
  def _ExecMigration(self):
3988
    """Migrate an instance.
3989

3990
    The migrate is done by:
3991
      - change the disks into dual-master mode
3992
      - wait until disks are fully synchronized again<