4 # Copyright (C) 2006, 2007, 2008 Google Inc.
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.
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.
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
22 """Functions to bootstrap a new cluster.
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
45 """Setup the SSH configuration for the cluster.
47 This generates a dsa keypair for root, adds the pub key to the
48 permitted hosts and adds the hostkey to its own known hosts.
51 priv_key, pub_key, auth_keys = ssh.GetUserFiles(constants.GANETI_RUNAS)
53 for name in priv_key, pub_key:
54 if os.path.exists(name):
55 utils.CreateBackup(name)
56 utils.RemoveFile(name)
58 result = utils.RunCmd(["ssh-keygen", "-t", "dsa",
62 raise errors.OpExecError("Could not generate ssh keypair, error %s" %
65 utils.AddAuthorizedKey(auth_keys, utils.ReadFile(pub_key))
68 def GenerateHmacKey(file_name):
69 """Writes a new HMAC key.
72 @param file_name: Path to output file
75 utils.WriteFile(file_name, data="%s\n" % utils.GenerateSecret(), mode=0400,
79 def GenerateClusterCrypto(new_cluster_cert, new_rapi_cert, new_confd_hmac_key,
80 new_cds, rapi_cert_pem=None, cds=None,
81 nodecert_file=constants.NODED_CERT_FILE,
82 rapicert_file=constants.RAPI_CERT_FILE,
83 hmackey_file=constants.CONFD_HMAC_KEY):
84 """Updates the cluster certificates, keys and secrets.
86 @type new_cluster_cert: bool
87 @param new_cluster_cert: Whether to generate a new cluster certificate
88 @type new_rapi_cert: bool
89 @param new_rapi_cert: Whether to generate a new RAPI certificate
90 @type new_confd_hmac_key: bool
91 @param new_confd_hmac_key: Whether to generate a new HMAC key
93 @param new_cds: Whether to generate a new cluster domain secret
94 @type rapi_cert_pem: string
95 @param rapi_cert_pem: New RAPI certificate in PEM format
97 @param cds: New cluster domain secret
98 @type nodecert_file: string
99 @param nodecert_file: optional override of the node cert file path
100 @type rapicert_file: string
101 @param rapicert_file: optional override of the rapi cert file path
102 @type hmackey_file: string
103 @param hmackey_file: optional override of the hmac key file path
106 # noded SSL certificate
107 cluster_cert_exists = os.path.exists(nodecert_file)
108 if new_cluster_cert or not cluster_cert_exists:
109 if cluster_cert_exists:
110 utils.CreateBackup(nodecert_file)
112 logging.debug("Generating new cluster certificate at %s", nodecert_file)
113 utils.GenerateSelfSignedSslCert(nodecert_file)
116 if new_confd_hmac_key or not os.path.exists(hmackey_file):
117 logging.debug("Writing new confd HMAC key to %s", hmackey_file)
118 GenerateHmacKey(hmackey_file)
121 rapi_cert_exists = os.path.exists(rapicert_file)
124 # Assume rapi_pem contains a valid PEM-formatted certificate and key
125 logging.debug("Writing RAPI certificate at %s", rapicert_file)
126 utils.WriteFile(rapicert_file, data=rapi_cert_pem, backup=True)
128 elif new_rapi_cert or not rapi_cert_exists:
130 utils.CreateBackup(rapicert_file)
132 logging.debug("Generating new RAPI certificate at %s", rapicert_file)
133 utils.GenerateSelfSignedSslCert(rapicert_file)
135 # Cluster domain secret
137 logging.debug("Writing cluster domain secret to %s",
138 constants.CLUSTER_DOMAIN_SECRET_FILE)
139 utils.WriteFile(constants.CLUSTER_DOMAIN_SECRET_FILE,
140 data=cds, backup=True)
142 elif new_cds or not os.path.exists(constants.CLUSTER_DOMAIN_SECRET_FILE):
143 logging.debug("Generating new cluster domain secret at %s",
144 constants.CLUSTER_DOMAIN_SECRET_FILE)
145 GenerateHmacKey(constants.CLUSTER_DOMAIN_SECRET_FILE)
148 def _InitGanetiServerSetup(master_name):
149 """Setup the necessary configuration for the initial node daemon.
151 This creates the nodepass file containing the shared password for
152 the cluster and also generates the SSL certificate.
155 # Generate cluster secrets
156 GenerateClusterCrypto(True, False, False, False)
158 result = utils.RunCmd([constants.DAEMON_UTIL, "start", constants.NODED])
160 raise errors.OpExecError("Could not start the node daemon, command %s"
161 " had exitcode %s and error %s" %
162 (result.cmd, result.exit_code, result.output))
164 _WaitForNodeDaemon(master_name)
167 def _WaitForNodeDaemon(node_name):
168 """Wait for node daemon to become responsive.
171 def _CheckNodeDaemon():
172 result = rpc.RpcRunner.call_version([node_name])[node_name]
174 raise utils.RetryAgain()
177 utils.Retry(_CheckNodeDaemon, 1.0, 10.0)
178 except utils.RetryTimeout:
179 raise errors.OpExecError("Node daemon on %s didn't answer queries within"
180 " 10 seconds" % node_name)
183 def _InitFileStorage(file_storage_dir):
184 """Initialize if needed the file storage.
186 @param file_storage_dir: the user-supplied value
187 @return: either empty string (if file storage was disabled at build
188 time) or the normalized path to the storage directory
191 if not constants.ENABLE_FILE_STORAGE:
194 file_storage_dir = os.path.normpath(file_storage_dir)
196 if not os.path.isabs(file_storage_dir):
197 raise errors.OpPrereqError("The file storage directory you passed is"
198 " not an absolute path.", errors.ECODE_INVAL)
200 if not os.path.exists(file_storage_dir):
202 os.makedirs(file_storage_dir, 0750)
204 raise errors.OpPrereqError("Cannot create file storage directory"
205 " '%s': %s" % (file_storage_dir, err),
206 errors.ECODE_ENVIRON)
208 if not os.path.isdir(file_storage_dir):
209 raise errors.OpPrereqError("The file storage directory '%s' is not"
210 " a directory." % file_storage_dir,
211 errors.ECODE_ENVIRON)
212 return file_storage_dir
215 def InitCluster(cluster_name, mac_prefix,
216 master_netdev, file_storage_dir, candidate_pool_size,
217 secondary_ip=None, vg_name=None, beparams=None,
218 nicparams=None, hvparams=None, enabled_hypervisors=None,
219 modify_etc_hosts=True, modify_ssh_setup=True,
220 maintain_node_health=False):
221 """Initialise the cluster.
223 @type candidate_pool_size: int
224 @param candidate_pool_size: master candidate pool size
227 # TODO: complete the docstring
228 if config.ConfigWriter.IsCluster():
229 raise errors.OpPrereqError("Cluster is already initialised",
232 if not enabled_hypervisors:
233 raise errors.OpPrereqError("Enabled hypervisors list must contain at"
234 " least one member", errors.ECODE_INVAL)
235 invalid_hvs = set(enabled_hypervisors) - constants.HYPER_TYPES
237 raise errors.OpPrereqError("Enabled hypervisors contains invalid"
238 " entries: %s" % invalid_hvs,
241 hostname = utils.GetHostInfo()
243 if hostname.ip.startswith("127."):
244 raise errors.OpPrereqError("This host's IP resolves to the private"
245 " range (%s). Please fix DNS or %s." %
246 (hostname.ip, constants.ETC_HOSTS),
247 errors.ECODE_ENVIRON)
249 if not utils.OwnIpAddress(hostname.ip):
250 raise errors.OpPrereqError("Inconsistency: this host's name resolves"
251 " to %s,\nbut this ip address does not"
252 " belong to this host. Aborting." %
253 hostname.ip, errors.ECODE_ENVIRON)
255 clustername = utils.GetHostInfo(utils.HostInfo.NormalizeName(cluster_name))
257 if utils.TcpPing(clustername.ip, constants.DEFAULT_NODED_PORT,
259 raise errors.OpPrereqError("Cluster IP already active. Aborting.",
260 errors.ECODE_NOTUNIQUE)
263 if not utils.IsValidIP(secondary_ip):
264 raise errors.OpPrereqError("Invalid secondary ip given",
266 if (secondary_ip != hostname.ip and
267 not utils.OwnIpAddress(secondary_ip)):
268 raise errors.OpPrereqError("You gave %s as secondary IP,"
269 " but it does not belong to this host." %
270 secondary_ip, errors.ECODE_ENVIRON)
272 secondary_ip = hostname.ip
274 if vg_name is not None:
275 # Check if volume group is valid
276 vgstatus = utils.CheckVolumeGroupSize(utils.ListVolumeGroups(), vg_name,
277 constants.MIN_VG_SIZE)
279 raise errors.OpPrereqError("Error: %s\nspecify --no-lvm-storage if"
280 " you are not using lvm" % vgstatus,
283 file_storage_dir = _InitFileStorage(file_storage_dir)
285 if not re.match("^[0-9a-z]{2}:[0-9a-z]{2}:[0-9a-z]{2}$", mac_prefix):
286 raise errors.OpPrereqError("Invalid mac prefix given '%s'" % mac_prefix,
289 result = utils.RunCmd(["ip", "link", "show", "dev", master_netdev])
291 raise errors.OpPrereqError("Invalid master netdev given (%s): '%s'" %
293 result.output.strip()), errors.ECODE_INVAL)
295 dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE)]
296 utils.EnsureDirs(dirs)
298 utils.ForceDictType(beparams, constants.BES_PARAMETER_TYPES)
299 utils.ForceDictType(nicparams, constants.NICS_PARAMETER_TYPES)
300 objects.NIC.CheckParameterSyntax(nicparams)
302 # hvparams is a mapping of hypervisor->hvparams dict
303 for hv_name, hv_params in hvparams.iteritems():
304 utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
305 hv_class = hypervisor.GetHypervisor(hv_name)
306 hv_class.CheckParameterSyntax(hv_params)
308 # set up the inter-node password and certificate
309 _InitGanetiServerSetup(hostname.name)
311 # set up ssh config and /etc/hosts
312 sshline = utils.ReadFile(constants.SSH_HOST_RSA_PUB)
313 sshkey = sshline.split(" ")[1]
316 utils.AddHostToEtcHosts(hostname.name)
323 # init of cluster config file
324 cluster_config = objects.Cluster(
326 rsahostkeypub=sshkey,
327 highest_used_port=(constants.FIRST_DRBD_PORT - 1),
328 mac_prefix=mac_prefix,
329 volume_group_name=vg_name,
330 tcpudp_port_pool=set(),
331 master_node=hostname.name,
332 master_ip=clustername.ip,
333 master_netdev=master_netdev,
334 cluster_name=clustername.name,
335 file_storage_dir=file_storage_dir,
336 enabled_hypervisors=enabled_hypervisors,
337 beparams={constants.PP_DEFAULT: beparams},
338 nicparams={constants.PP_DEFAULT: nicparams},
340 candidate_pool_size=candidate_pool_size,
341 modify_etc_hosts=modify_etc_hosts,
342 modify_ssh_setup=modify_ssh_setup,
345 uuid=utils.NewUUID(),
346 maintain_node_health=maintain_node_health,
348 master_node_config = objects.Node(name=hostname.name,
349 primary_ip=hostname.ip,
350 secondary_ip=secondary_ip,
352 master_candidate=True,
353 offline=False, drained=False,
355 InitConfig(constants.CONFIG_VERSION, cluster_config, master_node_config)
356 cfg = config.ConfigWriter()
357 ssh.WriteKnownHostsFile(cfg, constants.SSH_KNOWN_HOSTS_FILE)
358 cfg.Update(cfg.GetClusterInfo(), logging.error)
360 # start the master ip
361 # TODO: Review rpc call from bootstrap
362 # TODO: Warn on failed start master
363 rpc.RpcRunner.call_node_start_master(hostname.name, True, False)
366 def InitConfig(version, cluster_config, master_node_config,
367 cfg_file=constants.CLUSTER_CONF_FILE):
368 """Create the initial cluster configuration.
370 It will contain the current node, which will also be the master
371 node, and no instances.
374 @param version: configuration version
375 @type cluster_config: L{objects.Cluster}
376 @param cluster_config: cluster configuration
377 @type master_node_config: L{objects.Node}
378 @param master_node_config: master node configuration
379 @type cfg_file: string
380 @param cfg_file: configuration file path
384 master_node_config.name: master_node_config,
388 config_data = objects.ConfigData(version=version,
389 cluster=cluster_config,
393 ctime=now, mtime=now)
394 utils.WriteFile(cfg_file,
395 data=serializer.Dump(config_data.ToDict()),
399 def FinalizeClusterDestroy(master):
400 """Execute the last steps of cluster destroy
402 This function shuts down all the daemons, completing the destroy
403 begun in cmdlib.LUDestroyOpcode.
406 cfg = config.ConfigWriter()
407 modify_ssh_setup = cfg.GetClusterInfo().modify_ssh_setup
408 result = rpc.RpcRunner.call_node_stop_master(master, True)
409 msg = result.fail_msg
411 logging.warning("Could not disable the master role: %s", msg)
412 result = rpc.RpcRunner.call_node_leave_cluster(master, modify_ssh_setup)
413 msg = result.fail_msg
415 logging.warning("Could not shutdown the node daemon and cleanup"
416 " the node: %s", msg)
419 def SetupNodeDaemon(cluster_name, node, ssh_key_check):
420 """Add a node to the cluster.
422 This function must be called before the actual opcode, and will ssh
423 to the remote node, copy the needed files, and start ganeti-noded,
424 allowing the master to do the rest via normal rpc calls.
426 @param cluster_name: the cluster name
427 @param node: the name of the new node
428 @param ssh_key_check: whether to do a strict key check
431 sshrunner = ssh.SshRunner(cluster_name)
433 noded_cert = utils.ReadFile(constants.NODED_CERT_FILE)
434 rapi_cert = utils.ReadFile(constants.RAPI_CERT_FILE)
435 confd_hmac_key = utils.ReadFile(constants.CONFD_HMAC_KEY)
437 # in the base64 pem encoding, neither '!' nor '.' are valid chars,
438 # so we use this to detect an invalid certificate; as long as the
439 # cert doesn't contain this, the here-document will be correctly
440 # parsed by the shell sequence below. HMAC keys are hexadecimal strings,
441 # so the same restrictions apply.
442 for content in (noded_cert, rapi_cert, confd_hmac_key):
443 if re.search('^!EOF\.', content, re.MULTILINE):
444 raise errors.OpExecError("invalid SSL certificate or HMAC key")
446 if not noded_cert.endswith("\n"):
448 if not rapi_cert.endswith("\n"):
450 if not confd_hmac_key.endswith("\n"):
451 confd_hmac_key += "\n"
453 # set up inter-node password and certificate and restarts the node daemon
454 # and then connect with ssh to set password and start ganeti-noded
455 # note that all the below variables are sanitized at this point,
456 # either by being constants or by the checks above
457 # TODO: Could this command exceed a shell's maximum command length?
458 mycommand = ("umask 077 && "
459 "cat > '%s' << '!EOF.' && \n"
461 "cat > '%s' << '!EOF.' && \n"
463 "cat > '%s' << '!EOF.' && \n"
465 "chmod 0400 %s %s %s && "
467 (constants.NODED_CERT_FILE, noded_cert,
468 constants.RAPI_CERT_FILE, rapi_cert,
469 constants.CONFD_HMAC_KEY, confd_hmac_key,
470 constants.NODED_CERT_FILE, constants.RAPI_CERT_FILE,
471 constants.CONFD_HMAC_KEY,
472 constants.DAEMON_UTIL, constants.NODED))
474 result = sshrunner.Run(node, 'root', mycommand, batch=False,
475 ask_key=ssh_key_check,
476 use_cluster_key=False,
477 strict_host_check=ssh_key_check)
479 raise errors.OpExecError("Remote command on node %s, error: %s,"
481 (node, result.fail_reason, result.output))
483 _WaitForNodeDaemon(node)
486 def MasterFailover(no_voting=False):
487 """Failover the master node.
489 This checks that we are not already the master, and will cause the
490 current master to cease being master, and the non-master to become
493 @type no_voting: boolean
494 @param no_voting: force the operation without remote nodes agreement
498 sstore = ssconf.SimpleStore()
500 old_master, new_master = ssconf.GetMasterAndMyself(sstore)
501 node_list = sstore.GetNodeList()
502 mc_list = sstore.GetMasterCandidates()
504 if old_master == new_master:
505 raise errors.OpPrereqError("This commands must be run on the node"
506 " where you want the new master to be."
507 " %s is already the master" %
508 old_master, errors.ECODE_INVAL)
510 if new_master not in mc_list:
511 mc_no_master = [name for name in mc_list if name != old_master]
512 raise errors.OpPrereqError("This node is not among the nodes marked"
513 " as master candidates. Only these nodes"
514 " can become masters. Current list of"
515 " master candidates is:\n"
516 "%s" % ('\n'.join(mc_no_master)),
520 vote_list = GatherMasterVotes(node_list)
523 voted_master = vote_list[0][0]
524 if voted_master is None:
525 raise errors.OpPrereqError("Cluster is inconsistent, most nodes did"
526 " not respond.", errors.ECODE_ENVIRON)
527 elif voted_master != old_master:
528 raise errors.OpPrereqError("I have a wrong configuration, I believe"
529 " the master is %s but the other nodes"
530 " voted %s. Please resync the configuration"
532 (old_master, voted_master),
538 logging.info("Setting master to %s, old master: %s", new_master, old_master)
540 result = rpc.RpcRunner.call_node_stop_master(old_master, True)
541 msg = result.fail_msg
543 logging.error("Could not disable the master role on the old master"
544 " %s, please disable manually: %s", old_master, msg)
546 # Here we have a phase where no master should be running
548 # instantiate a real config writer, as we now know we have the
550 cfg = config.ConfigWriter()
552 cluster_info = cfg.GetClusterInfo()
553 cluster_info.master_node = new_master
554 # this will also regenerate the ssconf files, since we updated the
556 cfg.Update(cluster_info, logging.error)
558 result = rpc.RpcRunner.call_node_start_master(new_master, True, no_voting)
559 msg = result.fail_msg
561 logging.error("Could not start the master role on the new master"
562 " %s, please check: %s", new_master, msg)
569 """Returns the current master node.
571 This is a separate function in bootstrap since it's needed by
572 gnt-cluster, and instead of importing directly ssconf, it's better
573 to abstract it in bootstrap, where we do use ssconf in other
577 sstore = ssconf.SimpleStore()
579 old_master, _ = ssconf.GetMasterAndMyself(sstore)
584 def GatherMasterVotes(node_list):
585 """Check the agreement on who is the master.
587 This function will return a list of (node, number of votes), ordered
588 by the number of votes. Errors will be denoted by the key 'None'.
590 Note that the sum of votes is the number of nodes this machine
591 knows, whereas the number of entries in the list could be different
592 (if some nodes vote for another master).
594 We remove ourselves from the list since we know that (bugs aside)
595 since we use the same source for configuration information for both
596 backend and boostrap, we'll always vote for ourselves.
598 @type node_list: list
599 @param node_list: the list of nodes to query for master info; the current
600 node will be removed if it is in the list
602 @return: list of (node, votes)
605 myself = utils.HostInfo().name
607 node_list.remove(myself)
611 # no nodes left (eventually after removing myself)
613 results = rpc.RpcRunner.call_master_info(node_list)
614 if not isinstance(results, dict):
615 # this should not happen (unless internal error in rpc)
616 logging.critical("Can't complete rpc call, aborting master startup")
617 return [(None, len(node_list))]
625 logging.warning("Error contacting node %s: %s", node, msg)
627 elif not isinstance(data, (tuple, list)) or len(data) < 3:
628 logging.warning("Invalid data received from node %s: %s", node, data)
631 if None not in votes:
635 master_node = data[2]
636 if master_node not in votes:
637 votes[master_node] = 0
638 votes[master_node] += 1
640 vote_list = [v for v in votes.items()]
641 # sort first on number of votes then on name, since we want None
642 # sorted later if we have the half of the nodes not responding, and
643 # half voting all for the same master
644 vote_list.sort(key=lambda x: (x[1], x[0]), reverse=True)