Statistics
| Branch: | Tag: | Revision:

root / lib / bootstrap.py @ 7845b8c8

History | View | Annotate | Download (24.7 kB)

1
#
2
#
3

    
4
# Copyright (C) 2006, 2007, 2008, 2010 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
"""Functions to bootstrap a new cluster.
23

24
"""
25

    
26
import os
27
import os.path
28
import re
29
import logging
30
import time
31

    
32
from ganeti import rpc
33
from ganeti import ssh
34
from ganeti import utils
35
from ganeti import errors
36
from ganeti import config
37
from ganeti import constants
38
from ganeti import objects
39
from ganeti import ssconf
40
from ganeti import serializer
41
from ganeti import hypervisor
42
from ganeti import bdev
43
from ganeti import netutils
44
from ganeti import backend
45

    
46

    
47
def _InitSSHSetup():
48
  """Setup the SSH configuration for the cluster.
49

50
  This generates a dsa keypair for root, adds the pub key to the
51
  permitted hosts and adds the hostkey to its own known hosts.
52

53
  """
54
  priv_key, pub_key, auth_keys = ssh.GetUserFiles(constants.GANETI_RUNAS)
55

    
56
  for name in priv_key, pub_key:
57
    if os.path.exists(name):
58
      utils.CreateBackup(name)
59
    utils.RemoveFile(name)
60

    
61
  result = utils.RunCmd(["ssh-keygen", "-t", "dsa",
62
                         "-f", priv_key,
63
                         "-q", "-N", ""])
64
  if result.failed:
65
    raise errors.OpExecError("Could not generate ssh keypair, error %s" %
66
                             result.output)
67

    
68
  utils.AddAuthorizedKey(auth_keys, utils.ReadFile(pub_key))
69

    
70

    
71
def GenerateHmacKey(file_name):
72
  """Writes a new HMAC key.
73

74
  @type file_name: str
75
  @param file_name: Path to output file
76

77
  """
78
  utils.WriteFile(file_name, data="%s\n" % utils.GenerateSecret(), mode=0400,
79
                  backup=True)
80

    
81

    
82
def GenerateClusterCrypto(new_cluster_cert, new_rapi_cert, new_confd_hmac_key,
83
                          new_cds, rapi_cert_pem=None, cds=None,
84
                          nodecert_file=constants.NODED_CERT_FILE,
85
                          rapicert_file=constants.RAPI_CERT_FILE,
86
                          hmackey_file=constants.CONFD_HMAC_KEY,
87
                          cds_file=constants.CLUSTER_DOMAIN_SECRET_FILE):
88
  """Updates the cluster certificates, keys and secrets.
89

90
  @type new_cluster_cert: bool
91
  @param new_cluster_cert: Whether to generate a new cluster certificate
92
  @type new_rapi_cert: bool
93
  @param new_rapi_cert: Whether to generate a new RAPI certificate
94
  @type new_confd_hmac_key: bool
95
  @param new_confd_hmac_key: Whether to generate a new HMAC key
96
  @type new_cds: bool
97
  @param new_cds: Whether to generate a new cluster domain secret
98
  @type rapi_cert_pem: string
99
  @param rapi_cert_pem: New RAPI certificate in PEM format
100
  @type cds: string
101
  @param cds: New cluster domain secret
102
  @type nodecert_file: string
103
  @param nodecert_file: optional override of the node cert file path
104
  @type rapicert_file: string
105
  @param rapicert_file: optional override of the rapi cert file path
106
  @type hmackey_file: string
107
  @param hmackey_file: optional override of the hmac key file path
108

109
  """
110
  # noded SSL certificate
111
  cluster_cert_exists = os.path.exists(nodecert_file)
112
  if new_cluster_cert or not cluster_cert_exists:
113
    if cluster_cert_exists:
114
      utils.CreateBackup(nodecert_file)
115

    
116
    logging.debug("Generating new cluster certificate at %s", nodecert_file)
117
    utils.GenerateSelfSignedSslCert(nodecert_file)
118

    
119
  # confd HMAC key
120
  if new_confd_hmac_key or not os.path.exists(hmackey_file):
121
    logging.debug("Writing new confd HMAC key to %s", hmackey_file)
122
    GenerateHmacKey(hmackey_file)
123

    
124
  # RAPI
125
  rapi_cert_exists = os.path.exists(rapicert_file)
126

    
127
  if rapi_cert_pem:
128
    # Assume rapi_pem contains a valid PEM-formatted certificate and key
129
    logging.debug("Writing RAPI certificate at %s", rapicert_file)
130
    utils.WriteFile(rapicert_file, data=rapi_cert_pem, backup=True)
131

    
132
  elif new_rapi_cert or not rapi_cert_exists:
133
    if rapi_cert_exists:
134
      utils.CreateBackup(rapicert_file)
135

    
136
    logging.debug("Generating new RAPI certificate at %s", rapicert_file)
137
    utils.GenerateSelfSignedSslCert(rapicert_file)
138

    
139
  # Cluster domain secret
140
  if cds:
141
    logging.debug("Writing cluster domain secret to %s", cds_file)
142
    utils.WriteFile(cds_file, data=cds, backup=True)
143

    
144
  elif new_cds or not os.path.exists(cds_file):
145
    logging.debug("Generating new cluster domain secret at %s", cds_file)
146
    GenerateHmacKey(cds_file)
147

    
148

    
149
def _InitGanetiServerSetup(master_name):
150
  """Setup the necessary configuration for the initial node daemon.
151

152
  This creates the nodepass file containing the shared password for
153
  the cluster, generates the SSL certificate and starts the node daemon.
154

155
  @type master_name: str
156
  @param master_name: Name of the master node
157

158
  """
159
  # Generate cluster secrets
160
  GenerateClusterCrypto(True, False, False, False)
161

    
162
  result = utils.RunCmd([constants.DAEMON_UTIL, "start", constants.NODED])
163
  if result.failed:
164
    raise errors.OpExecError("Could not start the node daemon, command %s"
165
                             " had exitcode %s and error %s" %
166
                             (result.cmd, result.exit_code, result.output))
167

    
168
  _WaitForNodeDaemon(master_name)
169

    
170

    
171
def _WaitForNodeDaemon(node_name):
172
  """Wait for node daemon to become responsive.
173

174
  """
175
  def _CheckNodeDaemon():
176
    result = rpc.RpcRunner.call_version([node_name])[node_name]
177
    if result.fail_msg:
178
      raise utils.RetryAgain()
179

    
180
  try:
181
    utils.Retry(_CheckNodeDaemon, 1.0, 10.0)
182
  except utils.RetryTimeout:
183
    raise errors.OpExecError("Node daemon on %s didn't answer queries within"
184
                             " 10 seconds" % node_name)
185

    
186

    
187
def _InitFileStorage(file_storage_dir):
188
  """Initialize if needed the file storage.
189

190
  @param file_storage_dir: the user-supplied value
191
  @return: either empty string (if file storage was disabled at build
192
      time) or the normalized path to the storage directory
193

194
  """
195
  if not constants.ENABLE_FILE_STORAGE:
196
    return ""
197

    
198
  file_storage_dir = os.path.normpath(file_storage_dir)
199

    
200
  if not os.path.isabs(file_storage_dir):
201
    raise errors.OpPrereqError("The file storage directory you passed is"
202
                               " not an absolute path.", errors.ECODE_INVAL)
203

    
204
  if not os.path.exists(file_storage_dir):
205
    try:
206
      os.makedirs(file_storage_dir, 0750)
207
    except OSError, err:
208
      raise errors.OpPrereqError("Cannot create file storage directory"
209
                                 " '%s': %s" % (file_storage_dir, err),
210
                                 errors.ECODE_ENVIRON)
211

    
212
  if not os.path.isdir(file_storage_dir):
213
    raise errors.OpPrereqError("The file storage directory '%s' is not"
214
                               " a directory." % file_storage_dir,
215
                               errors.ECODE_ENVIRON)
216
  return file_storage_dir
217

    
218

    
219
#pylint: disable-msg=R0913
220
def InitCluster(cluster_name, mac_prefix,
221
                master_netdev, file_storage_dir, candidate_pool_size,
222
                secondary_ip=None, vg_name=None, beparams=None,
223
                nicparams=None, hvparams=None, enabled_hypervisors=None,
224
                modify_etc_hosts=True, modify_ssh_setup=True,
225
                maintain_node_health=False, drbd_helper=None,
226
                uid_pool=None, default_iallocator=None):
227
  """Initialise the cluster.
228

229
  @type candidate_pool_size: int
230
  @param candidate_pool_size: master candidate pool size
231

232
  """
233
  # TODO: complete the docstring
234
  if config.ConfigWriter.IsCluster():
235
    raise errors.OpPrereqError("Cluster is already initialised",
236
                               errors.ECODE_STATE)
237

    
238
  if not enabled_hypervisors:
239
    raise errors.OpPrereqError("Enabled hypervisors list must contain at"
240
                               " least one member", errors.ECODE_INVAL)
241
  invalid_hvs = set(enabled_hypervisors) - constants.HYPER_TYPES
242
  if invalid_hvs:
243
    raise errors.OpPrereqError("Enabled hypervisors contains invalid"
244
                               " entries: %s" % invalid_hvs,
245
                               errors.ECODE_INVAL)
246

    
247
  hostname = netutils.GetHostname()
248

    
249
  if netutils.IP4Address.IsLoopback(hostname.ip):
250
    raise errors.OpPrereqError("This host's IP (%s) resolves to a loopback"
251
                               " address. Please fix DNS or %s." %
252
                               (hostname.ip, constants.ETC_HOSTS),
253
                               errors.ECODE_ENVIRON)
254

    
255
  if not netutils.IPAddress.Own(hostname.ip):
256
    raise errors.OpPrereqError("Inconsistency: this host's name resolves"
257
                               " to %s,\nbut this ip address does not"
258
                               " belong to this host. Aborting." %
259
                               hostname.ip, errors.ECODE_ENVIRON)
260

    
261
  clustername = netutils.GetHostname(name=cluster_name)
262

    
263
  if netutils.TcpPing(clustername.ip, constants.DEFAULT_NODED_PORT,
264
                   timeout=5):
265
    raise errors.OpPrereqError("Cluster IP already active. Aborting.",
266
                               errors.ECODE_NOTUNIQUE)
267

    
268
  if secondary_ip:
269
    if not netutils.IP4Address.IsValid(secondary_ip):
270
      raise errors.OpPrereqError("Invalid secondary ip given",
271
                                 errors.ECODE_INVAL)
272
    if (secondary_ip != hostname.ip and
273
        not netutils.IPAddress.Own(secondary_ip)):
274
      raise errors.OpPrereqError("You gave %s as secondary IP,"
275
                                 " but it does not belong to this host." %
276
                                 secondary_ip, errors.ECODE_ENVIRON)
277
  else:
278
    secondary_ip = hostname.ip
279

    
280
  if vg_name is not None:
281
    # Check if volume group is valid
282
    vgstatus = utils.CheckVolumeGroupSize(utils.ListVolumeGroups(), vg_name,
283
                                          constants.MIN_VG_SIZE)
284
    if vgstatus:
285
      raise errors.OpPrereqError("Error: %s\nspecify --no-lvm-storage if"
286
                                 " you are not using lvm" % vgstatus,
287
                                 errors.ECODE_INVAL)
288

    
289
  if drbd_helper is not None:
290
    try:
291
      curr_helper = bdev.BaseDRBD.GetUsermodeHelper()
292
    except errors.BlockDeviceError, err:
293
      raise errors.OpPrereqError("Error while checking drbd helper"
294
                                 " (specify --no-drbd-storage if you are not"
295
                                 " using drbd): %s" % str(err),
296
                                 errors.ECODE_ENVIRON)
297
    if drbd_helper != curr_helper:
298
      raise errors.OpPrereqError("Error: requiring %s as drbd helper but %s"
299
                                 " is the current helper" % (drbd_helper,
300
                                                             curr_helper),
301
                                 errors.ECODE_INVAL)
302

    
303
  file_storage_dir = _InitFileStorage(file_storage_dir)
304

    
305
  if not re.match("^[0-9a-z]{2}:[0-9a-z]{2}:[0-9a-z]{2}$", mac_prefix):
306
    raise errors.OpPrereqError("Invalid mac prefix given '%s'" % mac_prefix,
307
                               errors.ECODE_INVAL)
308

    
309
  result = utils.RunCmd(["ip", "link", "show", "dev", master_netdev])
310
  if result.failed:
311
    raise errors.OpPrereqError("Invalid master netdev given (%s): '%s'" %
312
                               (master_netdev,
313
                                result.output.strip()), errors.ECODE_INVAL)
314

    
315
  dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE)]
316
  utils.EnsureDirs(dirs)
317

    
318
  utils.ForceDictType(beparams, constants.BES_PARAMETER_TYPES)
319
  utils.ForceDictType(nicparams, constants.NICS_PARAMETER_TYPES)
320
  objects.NIC.CheckParameterSyntax(nicparams)
321

    
322
  # hvparams is a mapping of hypervisor->hvparams dict
323
  for hv_name, hv_params in hvparams.iteritems():
324
    utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
325
    hv_class = hypervisor.GetHypervisor(hv_name)
326
    hv_class.CheckParameterSyntax(hv_params)
327

    
328
  # set up ssh config and /etc/hosts
329
  sshline = utils.ReadFile(constants.SSH_HOST_RSA_PUB)
330
  sshkey = sshline.split(" ")[1]
331

    
332
  if modify_etc_hosts:
333
    utils.AddHostToEtcHosts(hostname.name)
334

    
335
  if modify_ssh_setup:
336
    _InitSSHSetup()
337

    
338
  if default_iallocator is not None:
339
    alloc_script = utils.FindFile(default_iallocator,
340
                                  constants.IALLOCATOR_SEARCH_PATH,
341
                                  os.path.isfile)
342
    if alloc_script is None:
343
      raise errors.OpPrereqError("Invalid default iallocator script '%s'"
344
                                 " specified" % default_iallocator,
345
                                 errors.ECODE_INVAL)
346

    
347
  now = time.time()
348

    
349
  # init of cluster config file
350
  cluster_config = objects.Cluster(
351
    serial_no=1,
352
    rsahostkeypub=sshkey,
353
    highest_used_port=(constants.FIRST_DRBD_PORT - 1),
354
    mac_prefix=mac_prefix,
355
    volume_group_name=vg_name,
356
    tcpudp_port_pool=set(),
357
    master_node=hostname.name,
358
    master_ip=clustername.ip,
359
    master_netdev=master_netdev,
360
    cluster_name=clustername.name,
361
    file_storage_dir=file_storage_dir,
362
    enabled_hypervisors=enabled_hypervisors,
363
    beparams={constants.PP_DEFAULT: beparams},
364
    nicparams={constants.PP_DEFAULT: nicparams},
365
    hvparams=hvparams,
366
    candidate_pool_size=candidate_pool_size,
367
    modify_etc_hosts=modify_etc_hosts,
368
    modify_ssh_setup=modify_ssh_setup,
369
    uid_pool=uid_pool,
370
    ctime=now,
371
    mtime=now,
372
    uuid=utils.NewUUID(),
373
    maintain_node_health=maintain_node_health,
374
    drbd_usermode_helper=drbd_helper,
375
    default_iallocator=default_iallocator,
376
    )
377
  master_node_config = objects.Node(name=hostname.name,
378
                                    primary_ip=hostname.ip,
379
                                    secondary_ip=secondary_ip,
380
                                    serial_no=1,
381
                                    master_candidate=True,
382
                                    offline=False, drained=False,
383
                                    )
384
  InitConfig(constants.CONFIG_VERSION, cluster_config, master_node_config)
385
  cfg = config.ConfigWriter(offline=True)
386
  ssh.WriteKnownHostsFile(cfg, constants.SSH_KNOWN_HOSTS_FILE)
387
  cfg.Update(cfg.GetClusterInfo(), logging.error)
388
  backend.WriteSsconfFiles(cfg.GetSsconfValues())
389

    
390
  # set up the inter-node password and certificate
391
  _InitGanetiServerSetup(hostname.name)
392

    
393
  # start the master ip
394
  # TODO: Review rpc call from bootstrap
395
  # TODO: Warn on failed start master
396
  rpc.RpcRunner.call_node_start_master(hostname.name, True, False)
397

    
398

    
399
def InitConfig(version, cluster_config, master_node_config,
400
               cfg_file=constants.CLUSTER_CONF_FILE):
401
  """Create the initial cluster configuration.
402

403
  It will contain the current node, which will also be the master
404
  node, and no instances.
405

406
  @type version: int
407
  @param version: configuration version
408
  @type cluster_config: L{objects.Cluster}
409
  @param cluster_config: cluster configuration
410
  @type master_node_config: L{objects.Node}
411
  @param master_node_config: master node configuration
412
  @type cfg_file: string
413
  @param cfg_file: configuration file path
414

415
  """
416
  nodes = {
417
    master_node_config.name: master_node_config,
418
    }
419

    
420
  now = time.time()
421
  config_data = objects.ConfigData(version=version,
422
                                   cluster=cluster_config,
423
                                   nodes=nodes,
424
                                   instances={},
425
                                   serial_no=1,
426
                                   ctime=now, mtime=now)
427
  utils.WriteFile(cfg_file,
428
                  data=serializer.Dump(config_data.ToDict()),
429
                  mode=0600)
430

    
431

    
432
def FinalizeClusterDestroy(master):
433
  """Execute the last steps of cluster destroy
434

435
  This function shuts down all the daemons, completing the destroy
436
  begun in cmdlib.LUDestroyOpcode.
437

438
  """
439
  cfg = config.ConfigWriter()
440
  modify_ssh_setup = cfg.GetClusterInfo().modify_ssh_setup
441
  result = rpc.RpcRunner.call_node_stop_master(master, True)
442
  msg = result.fail_msg
443
  if msg:
444
    logging.warning("Could not disable the master role: %s", msg)
445
  result = rpc.RpcRunner.call_node_leave_cluster(master, modify_ssh_setup)
446
  msg = result.fail_msg
447
  if msg:
448
    logging.warning("Could not shutdown the node daemon and cleanup"
449
                    " the node: %s", msg)
450

    
451

    
452
def SetupNodeDaemon(cluster_name, node, ssh_key_check):
453
  """Add a node to the cluster.
454

455
  This function must be called before the actual opcode, and will ssh
456
  to the remote node, copy the needed files, and start ganeti-noded,
457
  allowing the master to do the rest via normal rpc calls.
458

459
  @param cluster_name: the cluster name
460
  @param node: the name of the new node
461
  @param ssh_key_check: whether to do a strict key check
462

463
  """
464
  sshrunner = ssh.SshRunner(cluster_name)
465

    
466
  noded_cert = utils.ReadFile(constants.NODED_CERT_FILE)
467
  rapi_cert = utils.ReadFile(constants.RAPI_CERT_FILE)
468
  confd_hmac_key = utils.ReadFile(constants.CONFD_HMAC_KEY)
469

    
470
  # in the base64 pem encoding, neither '!' nor '.' are valid chars,
471
  # so we use this to detect an invalid certificate; as long as the
472
  # cert doesn't contain this, the here-document will be correctly
473
  # parsed by the shell sequence below. HMAC keys are hexadecimal strings,
474
  # so the same restrictions apply.
475
  for content in (noded_cert, rapi_cert, confd_hmac_key):
476
    if re.search('^!EOF\.', content, re.MULTILINE):
477
      raise errors.OpExecError("invalid SSL certificate or HMAC key")
478

    
479
  if not noded_cert.endswith("\n"):
480
    noded_cert += "\n"
481
  if not rapi_cert.endswith("\n"):
482
    rapi_cert += "\n"
483
  if not confd_hmac_key.endswith("\n"):
484
    confd_hmac_key += "\n"
485

    
486
  # set up inter-node password and certificate and restarts the node daemon
487
  # and then connect with ssh to set password and start ganeti-noded
488
  # note that all the below variables are sanitized at this point,
489
  # either by being constants or by the checks above
490
  # TODO: Could this command exceed a shell's maximum command length?
491
  mycommand = ("umask 077 && "
492
               "cat > '%s' << '!EOF.' && \n"
493
               "%s!EOF.\n"
494
               "cat > '%s' << '!EOF.' && \n"
495
               "%s!EOF.\n"
496
               "cat > '%s' << '!EOF.' && \n"
497
               "%s!EOF.\n"
498
               "chmod 0400 %s %s %s && "
499
               "%s start %s" %
500
               (constants.NODED_CERT_FILE, noded_cert,
501
                constants.RAPI_CERT_FILE, rapi_cert,
502
                constants.CONFD_HMAC_KEY, confd_hmac_key,
503
                constants.NODED_CERT_FILE, constants.RAPI_CERT_FILE,
504
                constants.CONFD_HMAC_KEY,
505
                constants.DAEMON_UTIL, constants.NODED))
506

    
507
  result = sshrunner.Run(node, 'root', mycommand, batch=False,
508
                         ask_key=ssh_key_check,
509
                         use_cluster_key=False,
510
                         strict_host_check=ssh_key_check)
511
  if result.failed:
512
    raise errors.OpExecError("Remote command on node %s, error: %s,"
513
                             " output: %s" %
514
                             (node, result.fail_reason, result.output))
515

    
516
  _WaitForNodeDaemon(node)
517

    
518

    
519
def MasterFailover(no_voting=False):
520
  """Failover the master node.
521

522
  This checks that we are not already the master, and will cause the
523
  current master to cease being master, and the non-master to become
524
  new master.
525

526
  @type no_voting: boolean
527
  @param no_voting: force the operation without remote nodes agreement
528
                      (dangerous)
529

530
  """
531
  sstore = ssconf.SimpleStore()
532

    
533
  old_master, new_master = ssconf.GetMasterAndMyself(sstore)
534
  node_list = sstore.GetNodeList()
535
  mc_list = sstore.GetMasterCandidates()
536

    
537
  if old_master == new_master:
538
    raise errors.OpPrereqError("This commands must be run on the node"
539
                               " where you want the new master to be."
540
                               " %s is already the master" %
541
                               old_master, errors.ECODE_INVAL)
542

    
543
  if new_master not in mc_list:
544
    mc_no_master = [name for name in mc_list if name != old_master]
545
    raise errors.OpPrereqError("This node is not among the nodes marked"
546
                               " as master candidates. Only these nodes"
547
                               " can become masters. Current list of"
548
                               " master candidates is:\n"
549
                               "%s" % ('\n'.join(mc_no_master)),
550
                               errors.ECODE_STATE)
551

    
552
  if not no_voting:
553
    vote_list = GatherMasterVotes(node_list)
554

    
555
    if vote_list:
556
      voted_master = vote_list[0][0]
557
      if voted_master is None:
558
        raise errors.OpPrereqError("Cluster is inconsistent, most nodes did"
559
                                   " not respond.", errors.ECODE_ENVIRON)
560
      elif voted_master != old_master:
561
        raise errors.OpPrereqError("I have a wrong configuration, I believe"
562
                                   " the master is %s but the other nodes"
563
                                   " voted %s. Please resync the configuration"
564
                                   " of this node." %
565
                                   (old_master, voted_master),
566
                                   errors.ECODE_STATE)
567
  # end checks
568

    
569
  rcode = 0
570

    
571
  logging.info("Setting master to %s, old master: %s", new_master, old_master)
572

    
573
  result = rpc.RpcRunner.call_node_stop_master(old_master, True)
574
  msg = result.fail_msg
575
  if msg:
576
    logging.error("Could not disable the master role on the old master"
577
                 " %s, please disable manually: %s", old_master, msg)
578

    
579
  master_ip = sstore.GetMasterIP()
580
  total_timeout = 30
581
  # Here we have a phase where no master should be running
582
  def _check_ip():
583
    if netutils.TcpPing(master_ip, constants.DEFAULT_NODED_PORT):
584
      raise utils.RetryAgain()
585

    
586
  try:
587
    utils.Retry(_check_ip, (1, 1.5, 5), total_timeout)
588
  except utils.RetryTimeout:
589
    logging.warning("The master IP is still reachable after %s seconds,"
590
                    " continuing but activating the master on the current"
591
                    " node will probably fail", total_timeout)
592

    
593
  # instantiate a real config writer, as we now know we have the
594
  # configuration data
595
  cfg = config.ConfigWriter()
596

    
597
  cluster_info = cfg.GetClusterInfo()
598
  cluster_info.master_node = new_master
599
  # this will also regenerate the ssconf files, since we updated the
600
  # cluster info
601
  cfg.Update(cluster_info, logging.error)
602

    
603
  result = rpc.RpcRunner.call_node_start_master(new_master, True, no_voting)
604
  msg = result.fail_msg
605
  if msg:
606
    logging.error("Could not start the master role on the new master"
607
                  " %s, please check: %s", new_master, msg)
608
    rcode = 1
609

    
610
  return rcode
611

    
612

    
613
def GetMaster():
614
  """Returns the current master node.
615

616
  This is a separate function in bootstrap since it's needed by
617
  gnt-cluster, and instead of importing directly ssconf, it's better
618
  to abstract it in bootstrap, where we do use ssconf in other
619
  functions too.
620

621
  """
622
  sstore = ssconf.SimpleStore()
623

    
624
  old_master, _ = ssconf.GetMasterAndMyself(sstore)
625

    
626
  return old_master
627

    
628

    
629
def GatherMasterVotes(node_list):
630
  """Check the agreement on who is the master.
631

632
  This function will return a list of (node, number of votes), ordered
633
  by the number of votes. Errors will be denoted by the key 'None'.
634

635
  Note that the sum of votes is the number of nodes this machine
636
  knows, whereas the number of entries in the list could be different
637
  (if some nodes vote for another master).
638

639
  We remove ourselves from the list since we know that (bugs aside)
640
  since we use the same source for configuration information for both
641
  backend and boostrap, we'll always vote for ourselves.
642

643
  @type node_list: list
644
  @param node_list: the list of nodes to query for master info; the current
645
      node will be removed if it is in the list
646
  @rtype: list
647
  @return: list of (node, votes)
648

649
  """
650
  myself = netutils.Hostname.GetSysName()
651
  try:
652
    node_list.remove(myself)
653
  except ValueError:
654
    pass
655
  if not node_list:
656
    # no nodes left (eventually after removing myself)
657
    return []
658
  results = rpc.RpcRunner.call_master_info(node_list)
659
  if not isinstance(results, dict):
660
    # this should not happen (unless internal error in rpc)
661
    logging.critical("Can't complete rpc call, aborting master startup")
662
    return [(None, len(node_list))]
663
  votes = {}
664
  for node in results:
665
    nres = results[node]
666
    data = nres.payload
667
    msg = nres.fail_msg
668
    fail = False
669
    if msg:
670
      logging.warning("Error contacting node %s: %s", node, msg)
671
      fail = True
672
    elif not isinstance(data, (tuple, list)) or len(data) < 3:
673
      logging.warning("Invalid data received from node %s: %s", node, data)
674
      fail = True
675
    if fail:
676
      if None not in votes:
677
        votes[None] = 0
678
      votes[None] += 1
679
      continue
680
    master_node = data[2]
681
    if master_node not in votes:
682
      votes[master_node] = 0
683
    votes[master_node] += 1
684

    
685
  vote_list = [v for v in votes.items()]
686
  # sort first on number of votes then on name, since we want None
687
  # sorted later if we have the half of the nodes not responding, and
688
  # half voting all for the same master
689
  vote_list.sort(key=lambda x: (x[1], x[0]), reverse=True)
690

    
691
  return vote_list