4 # Copyright (C) 2006, 2007, 2008, 2010 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
42 from ganeti import bdev
43 from ganeti import netutils
47 """Setup the SSH configuration for the cluster.
49 This generates a dsa keypair for root, adds the pub key to the
50 permitted hosts and adds the hostkey to its own known hosts.
53 priv_key, pub_key, auth_keys = ssh.GetUserFiles(constants.GANETI_RUNAS)
55 for name in priv_key, pub_key:
56 if os.path.exists(name):
57 utils.CreateBackup(name)
58 utils.RemoveFile(name)
60 result = utils.RunCmd(["ssh-keygen", "-t", "dsa",
64 raise errors.OpExecError("Could not generate ssh keypair, error %s" %
67 utils.AddAuthorizedKey(auth_keys, utils.ReadFile(pub_key))
70 def GenerateHmacKey(file_name):
71 """Writes a new HMAC key.
74 @param file_name: Path to output file
77 utils.WriteFile(file_name, data="%s\n" % utils.GenerateSecret(), mode=0400,
81 def GenerateClusterCrypto(new_cluster_cert, new_rapi_cert, new_confd_hmac_key,
82 new_cds, rapi_cert_pem=None, cds=None,
83 nodecert_file=constants.NODED_CERT_FILE,
84 rapicert_file=constants.RAPI_CERT_FILE,
85 hmackey_file=constants.CONFD_HMAC_KEY,
86 cds_file=constants.CLUSTER_DOMAIN_SECRET_FILE):
87 """Updates the cluster certificates, keys and secrets.
89 @type new_cluster_cert: bool
90 @param new_cluster_cert: Whether to generate a new cluster certificate
91 @type new_rapi_cert: bool
92 @param new_rapi_cert: Whether to generate a new RAPI certificate
93 @type new_confd_hmac_key: bool
94 @param new_confd_hmac_key: Whether to generate a new HMAC key
96 @param new_cds: Whether to generate a new cluster domain secret
97 @type rapi_cert_pem: string
98 @param rapi_cert_pem: New RAPI certificate in PEM format
100 @param cds: New cluster domain secret
101 @type nodecert_file: string
102 @param nodecert_file: optional override of the node cert file path
103 @type rapicert_file: string
104 @param rapicert_file: optional override of the rapi cert file path
105 @type hmackey_file: string
106 @param hmackey_file: optional override of the hmac key file path
109 # noded SSL certificate
110 cluster_cert_exists = os.path.exists(nodecert_file)
111 if new_cluster_cert or not cluster_cert_exists:
112 if cluster_cert_exists:
113 utils.CreateBackup(nodecert_file)
115 logging.debug("Generating new cluster certificate at %s", nodecert_file)
116 utils.GenerateSelfSignedSslCert(nodecert_file)
119 if new_confd_hmac_key or not os.path.exists(hmackey_file):
120 logging.debug("Writing new confd HMAC key to %s", hmackey_file)
121 GenerateHmacKey(hmackey_file)
124 rapi_cert_exists = os.path.exists(rapicert_file)
127 # Assume rapi_pem contains a valid PEM-formatted certificate and key
128 logging.debug("Writing RAPI certificate at %s", rapicert_file)
129 utils.WriteFile(rapicert_file, data=rapi_cert_pem, backup=True)
131 elif new_rapi_cert or not rapi_cert_exists:
133 utils.CreateBackup(rapicert_file)
135 logging.debug("Generating new RAPI certificate at %s", rapicert_file)
136 utils.GenerateSelfSignedSslCert(rapicert_file)
138 # Cluster domain secret
140 logging.debug("Writing cluster domain secret to %s", cds_file)
141 utils.WriteFile(cds_file, data=cds, backup=True)
143 elif new_cds or not os.path.exists(cds_file):
144 logging.debug("Generating new cluster domain secret at %s", cds_file)
145 GenerateHmacKey(cds_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, generates the SSL certificate and starts the node daemon.
154 @type master_name: str
155 @param master_name: Name of the master node
158 # Generate cluster secrets
159 GenerateClusterCrypto(True, False, False, False)
161 result = utils.RunCmd([constants.DAEMON_UTIL, "start", constants.NODED])
163 raise errors.OpExecError("Could not start the node daemon, command %s"
164 " had exitcode %s and error %s" %
165 (result.cmd, result.exit_code, result.output))
167 _WaitForNodeDaemon(master_name)
170 def _WaitForNodeDaemon(node_name):
171 """Wait for node daemon to become responsive.
174 def _CheckNodeDaemon():
175 result = rpc.RpcRunner.call_version([node_name])[node_name]
177 raise utils.RetryAgain()
180 utils.Retry(_CheckNodeDaemon, 1.0, 10.0)
181 except utils.RetryTimeout:
182 raise errors.OpExecError("Node daemon on %s didn't answer queries within"
183 " 10 seconds" % node_name)
186 def _InitFileStorage(file_storage_dir):
187 """Initialize if needed the file storage.
189 @param file_storage_dir: the user-supplied value
190 @return: either empty string (if file storage was disabled at build
191 time) or the normalized path to the storage directory
194 if not constants.ENABLE_FILE_STORAGE:
197 file_storage_dir = os.path.normpath(file_storage_dir)
199 if not os.path.isabs(file_storage_dir):
200 raise errors.OpPrereqError("The file storage directory you passed is"
201 " not an absolute path.", errors.ECODE_INVAL)
203 if not os.path.exists(file_storage_dir):
205 os.makedirs(file_storage_dir, 0750)
207 raise errors.OpPrereqError("Cannot create file storage directory"
208 " '%s': %s" % (file_storage_dir, err),
209 errors.ECODE_ENVIRON)
211 if not os.path.isdir(file_storage_dir):
212 raise errors.OpPrereqError("The file storage directory '%s' is not"
213 " a directory." % file_storage_dir,
214 errors.ECODE_ENVIRON)
215 return file_storage_dir
218 #pylint: disable-msg=R0913
219 def InitCluster(cluster_name, mac_prefix,
220 master_netdev, file_storage_dir, candidate_pool_size,
221 secondary_ip=None, vg_name=None, beparams=None,
222 nicparams=None, hvparams=None, enabled_hypervisors=None,
223 modify_etc_hosts=True, modify_ssh_setup=True,
224 maintain_node_health=False, drbd_helper=None,
225 uid_pool=None, default_iallocator=None):
226 """Initialise the cluster.
228 @type candidate_pool_size: int
229 @param candidate_pool_size: master candidate pool size
232 # TODO: complete the docstring
233 if config.ConfigWriter.IsCluster():
234 raise errors.OpPrereqError("Cluster is already initialised",
237 if not enabled_hypervisors:
238 raise errors.OpPrereqError("Enabled hypervisors list must contain at"
239 " least one member", errors.ECODE_INVAL)
240 invalid_hvs = set(enabled_hypervisors) - constants.HYPER_TYPES
242 raise errors.OpPrereqError("Enabled hypervisors contains invalid"
243 " entries: %s" % invalid_hvs,
246 hostname = netutils.GetHostInfo()
248 if hostname.ip.startswith("127."):
249 raise errors.OpPrereqError("This host's IP resolves to the private"
250 " range (%s). Please fix DNS or %s." %
251 (hostname.ip, constants.ETC_HOSTS),
252 errors.ECODE_ENVIRON)
254 if not netutils.OwnIpAddress(hostname.ip):
255 raise errors.OpPrereqError("Inconsistency: this host's name resolves"
256 " to %s,\nbut this ip address does not"
257 " belong to this host. Aborting." %
258 hostname.ip, errors.ECODE_ENVIRON)
261 netutils.GetHostInfo(netutils.HostInfo.NormalizeName(cluster_name))
263 if netutils.TcpPing(clustername.ip, constants.DEFAULT_NODED_PORT,
265 raise errors.OpPrereqError("Cluster IP already active. Aborting.",
266 errors.ECODE_NOTUNIQUE)
269 if not netutils.IsValidIP4(secondary_ip):
270 raise errors.OpPrereqError("Invalid secondary ip given",
272 if (secondary_ip != hostname.ip and
273 not netutils.OwnIpAddress(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)
278 secondary_ip = hostname.ip
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)
285 raise errors.OpPrereqError("Error: %s\nspecify --no-lvm-storage if"
286 " you are not using lvm" % vgstatus,
289 if drbd_helper is not None:
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,
303 file_storage_dir = _InitFileStorage(file_storage_dir)
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,
309 result = utils.RunCmd(["ip", "link", "show", "dev", master_netdev])
311 raise errors.OpPrereqError("Invalid master netdev given (%s): '%s'" %
313 result.output.strip()), errors.ECODE_INVAL)
315 dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE)]
316 utils.EnsureDirs(dirs)
318 utils.ForceDictType(beparams, constants.BES_PARAMETER_TYPES)
319 utils.ForceDictType(nicparams, constants.NICS_PARAMETER_TYPES)
320 objects.NIC.CheckParameterSyntax(nicparams)
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)
328 # set up the inter-node password and certificate, start noded
329 _InitGanetiServerSetup(hostname.name)
331 # set up ssh config and /etc/hosts
332 sshline = utils.ReadFile(constants.SSH_HOST_RSA_PUB)
333 sshkey = sshline.split(" ")[1]
336 utils.AddHostToEtcHosts(hostname.name)
341 if default_iallocator is not None:
342 alloc_script = utils.FindFile(default_iallocator,
343 constants.IALLOCATOR_SEARCH_PATH,
345 if alloc_script is None:
346 raise errors.OpPrereqError("Invalid default iallocator script '%s'"
347 " specified" % default_iallocator,
352 # init of cluster config file
353 cluster_config = objects.Cluster(
355 rsahostkeypub=sshkey,
356 highest_used_port=(constants.FIRST_DRBD_PORT - 1),
357 mac_prefix=mac_prefix,
358 volume_group_name=vg_name,
359 tcpudp_port_pool=set(),
360 master_node=hostname.name,
361 master_ip=clustername.ip,
362 master_netdev=master_netdev,
363 cluster_name=clustername.name,
364 file_storage_dir=file_storage_dir,
365 enabled_hypervisors=enabled_hypervisors,
366 beparams={constants.PP_DEFAULT: beparams},
367 nicparams={constants.PP_DEFAULT: nicparams},
369 candidate_pool_size=candidate_pool_size,
370 modify_etc_hosts=modify_etc_hosts,
371 modify_ssh_setup=modify_ssh_setup,
375 uuid=utils.NewUUID(),
376 maintain_node_health=maintain_node_health,
377 drbd_usermode_helper=drbd_helper,
378 default_iallocator=default_iallocator,
380 master_node_config = objects.Node(name=hostname.name,
381 primary_ip=hostname.ip,
382 secondary_ip=secondary_ip,
384 master_candidate=True,
385 offline=False, drained=False,
387 InitConfig(constants.CONFIG_VERSION, cluster_config, master_node_config)
388 cfg = config.ConfigWriter()
389 ssh.WriteKnownHostsFile(cfg, constants.SSH_KNOWN_HOSTS_FILE)
390 cfg.Update(cfg.GetClusterInfo(), logging.error)
392 # start the master ip
393 # TODO: Review rpc call from bootstrap
394 # TODO: Warn on failed start master
395 rpc.RpcRunner.call_node_start_master(hostname.name, True, False)
398 def InitConfig(version, cluster_config, master_node_config,
399 cfg_file=constants.CLUSTER_CONF_FILE):
400 """Create the initial cluster configuration.
402 It will contain the current node, which will also be the master
403 node, and no instances.
406 @param version: configuration version
407 @type cluster_config: L{objects.Cluster}
408 @param cluster_config: cluster configuration
409 @type master_node_config: L{objects.Node}
410 @param master_node_config: master node configuration
411 @type cfg_file: string
412 @param cfg_file: configuration file path
416 master_node_config.name: master_node_config,
420 config_data = objects.ConfigData(version=version,
421 cluster=cluster_config,
425 ctime=now, mtime=now)
426 utils.WriteFile(cfg_file,
427 data=serializer.Dump(config_data.ToDict()),
431 def FinalizeClusterDestroy(master):
432 """Execute the last steps of cluster destroy
434 This function shuts down all the daemons, completing the destroy
435 begun in cmdlib.LUDestroyOpcode.
438 cfg = config.ConfigWriter()
439 modify_ssh_setup = cfg.GetClusterInfo().modify_ssh_setup
440 result = rpc.RpcRunner.call_node_stop_master(master, True)
441 msg = result.fail_msg
443 logging.warning("Could not disable the master role: %s", msg)
444 result = rpc.RpcRunner.call_node_leave_cluster(master, modify_ssh_setup)
445 msg = result.fail_msg
447 logging.warning("Could not shutdown the node daemon and cleanup"
448 " the node: %s", msg)
451 def SetupNodeDaemon(cluster_name, node, ssh_key_check):
452 """Add a node to the cluster.
454 This function must be called before the actual opcode, and will ssh
455 to the remote node, copy the needed files, and start ganeti-noded,
456 allowing the master to do the rest via normal rpc calls.
458 @param cluster_name: the cluster name
459 @param node: the name of the new node
460 @param ssh_key_check: whether to do a strict key check
463 sshrunner = ssh.SshRunner(cluster_name)
465 noded_cert = utils.ReadFile(constants.NODED_CERT_FILE)
466 rapi_cert = utils.ReadFile(constants.RAPI_CERT_FILE)
467 confd_hmac_key = utils.ReadFile(constants.CONFD_HMAC_KEY)
469 # in the base64 pem encoding, neither '!' nor '.' are valid chars,
470 # so we use this to detect an invalid certificate; as long as the
471 # cert doesn't contain this, the here-document will be correctly
472 # parsed by the shell sequence below. HMAC keys are hexadecimal strings,
473 # so the same restrictions apply.
474 for content in (noded_cert, rapi_cert, confd_hmac_key):
475 if re.search('^!EOF\.', content, re.MULTILINE):
476 raise errors.OpExecError("invalid SSL certificate or HMAC key")
478 if not noded_cert.endswith("\n"):
480 if not rapi_cert.endswith("\n"):
482 if not confd_hmac_key.endswith("\n"):
483 confd_hmac_key += "\n"
485 # set up inter-node password and certificate and restarts the node daemon
486 # and then connect with ssh to set password and start ganeti-noded
487 # note that all the below variables are sanitized at this point,
488 # either by being constants or by the checks above
489 # TODO: Could this command exceed a shell's maximum command length?
490 mycommand = ("umask 077 && "
491 "cat > '%s' << '!EOF.' && \n"
493 "cat > '%s' << '!EOF.' && \n"
495 "cat > '%s' << '!EOF.' && \n"
497 "chmod 0400 %s %s %s && "
499 (constants.NODED_CERT_FILE, noded_cert,
500 constants.RAPI_CERT_FILE, rapi_cert,
501 constants.CONFD_HMAC_KEY, confd_hmac_key,
502 constants.NODED_CERT_FILE, constants.RAPI_CERT_FILE,
503 constants.CONFD_HMAC_KEY,
504 constants.DAEMON_UTIL, constants.NODED))
506 result = sshrunner.Run(node, 'root', mycommand, batch=False,
507 ask_key=ssh_key_check,
508 use_cluster_key=False,
509 strict_host_check=ssh_key_check)
511 raise errors.OpExecError("Remote command on node %s, error: %s,"
513 (node, result.fail_reason, result.output))
515 _WaitForNodeDaemon(node)
518 def MasterFailover(no_voting=False):
519 """Failover the master node.
521 This checks that we are not already the master, and will cause the
522 current master to cease being master, and the non-master to become
525 @type no_voting: boolean
526 @param no_voting: force the operation without remote nodes agreement
530 sstore = ssconf.SimpleStore()
532 old_master, new_master = ssconf.GetMasterAndMyself(sstore)
533 node_list = sstore.GetNodeList()
534 mc_list = sstore.GetMasterCandidates()
536 if old_master == new_master:
537 raise errors.OpPrereqError("This commands must be run on the node"
538 " where you want the new master to be."
539 " %s is already the master" %
540 old_master, errors.ECODE_INVAL)
542 if new_master not in mc_list:
543 mc_no_master = [name for name in mc_list if name != old_master]
544 raise errors.OpPrereqError("This node is not among the nodes marked"
545 " as master candidates. Only these nodes"
546 " can become masters. Current list of"
547 " master candidates is:\n"
548 "%s" % ('\n'.join(mc_no_master)),
552 vote_list = GatherMasterVotes(node_list)
555 voted_master = vote_list[0][0]
556 if voted_master is None:
557 raise errors.OpPrereqError("Cluster is inconsistent, most nodes did"
558 " not respond.", errors.ECODE_ENVIRON)
559 elif voted_master != old_master:
560 raise errors.OpPrereqError("I have a wrong configuration, I believe"
561 " the master is %s but the other nodes"
562 " voted %s. Please resync the configuration"
564 (old_master, voted_master),
570 logging.info("Setting master to %s, old master: %s", new_master, old_master)
572 result = rpc.RpcRunner.call_node_stop_master(old_master, True)
573 msg = result.fail_msg
575 logging.error("Could not disable the master role on the old master"
576 " %s, please disable manually: %s", old_master, msg)
578 master_ip = sstore.GetMasterIP()
580 # Here we have a phase where no master should be running
582 if netutils.TcpPing(master_ip, constants.DEFAULT_NODED_PORT):
583 raise utils.RetryAgain()
586 utils.Retry(_check_ip, (1, 1.5, 5), total_timeout)
587 except utils.RetryTimeout:
588 logging.warning("The master IP is still reachable after %s seconds,"
589 " continuing but activating the master on the current"
590 " node will probably fail", total_timeout)
592 # instantiate a real config writer, as we now know we have the
594 cfg = config.ConfigWriter()
596 cluster_info = cfg.GetClusterInfo()
597 cluster_info.master_node = new_master
598 # this will also regenerate the ssconf files, since we updated the
600 cfg.Update(cluster_info, logging.error)
602 result = rpc.RpcRunner.call_node_start_master(new_master, True, no_voting)
603 msg = result.fail_msg
605 logging.error("Could not start the master role on the new master"
606 " %s, please check: %s", new_master, msg)
613 """Returns the current master node.
615 This is a separate function in bootstrap since it's needed by
616 gnt-cluster, and instead of importing directly ssconf, it's better
617 to abstract it in bootstrap, where we do use ssconf in other
621 sstore = ssconf.SimpleStore()
623 old_master, _ = ssconf.GetMasterAndMyself(sstore)
628 def GatherMasterVotes(node_list):
629 """Check the agreement on who is the master.
631 This function will return a list of (node, number of votes), ordered
632 by the number of votes. Errors will be denoted by the key 'None'.
634 Note that the sum of votes is the number of nodes this machine
635 knows, whereas the number of entries in the list could be different
636 (if some nodes vote for another master).
638 We remove ourselves from the list since we know that (bugs aside)
639 since we use the same source for configuration information for both
640 backend and boostrap, we'll always vote for ourselves.
642 @type node_list: list
643 @param node_list: the list of nodes to query for master info; the current
644 node will be removed if it is in the list
646 @return: list of (node, votes)
649 myself = netutils.HostInfo().name
651 node_list.remove(myself)
655 # no nodes left (eventually after removing myself)
657 results = rpc.RpcRunner.call_master_info(node_list)
658 if not isinstance(results, dict):
659 # this should not happen (unless internal error in rpc)
660 logging.critical("Can't complete rpc call, aborting master startup")
661 return [(None, len(node_list))]
669 logging.warning("Error contacting node %s: %s", node, msg)
671 elif not isinstance(data, (tuple, list)) or len(data) < 3:
672 logging.warning("Invalid data received from node %s: %s", node, data)
675 if None not in votes:
679 master_node = data[2]
680 if master_node not in votes:
681 votes[master_node] = 0
682 votes[master_node] += 1
684 vote_list = [v for v in votes.items()]
685 # sort first on number of votes then on name, since we want None
686 # sorted later if we have the half of the nodes not responding, and
687 # half voting all for the same master
688 vote_list.sort(key=lambda x: (x[1], x[0]), reverse=True)