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
44 from ganeti import backend
48 """Setup the SSH configuration for the cluster.
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.
54 priv_key, pub_key, auth_keys = ssh.GetUserFiles(constants.GANETI_RUNAS)
56 for name in priv_key, pub_key:
57 if os.path.exists(name):
58 utils.CreateBackup(name)
59 utils.RemoveFile(name)
61 result = utils.RunCmd(["ssh-keygen", "-t", "dsa",
65 raise errors.OpExecError("Could not generate ssh keypair, error %s" %
68 utils.AddAuthorizedKey(auth_keys, utils.ReadFile(pub_key))
71 def GenerateHmacKey(file_name):
72 """Writes a new HMAC key.
75 @param file_name: Path to output file
78 utils.WriteFile(file_name, data="%s\n" % utils.GenerateSecret(), mode=0400,
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.
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
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
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
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)
116 logging.debug("Generating new cluster certificate at %s", nodecert_file)
117 utils.GenerateSelfSignedSslCert(nodecert_file)
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)
125 rapi_cert_exists = os.path.exists(rapicert_file)
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)
132 elif new_rapi_cert or not rapi_cert_exists:
134 utils.CreateBackup(rapicert_file)
136 logging.debug("Generating new RAPI certificate at %s", rapicert_file)
137 utils.GenerateSelfSignedSslCert(rapicert_file)
139 # Cluster domain secret
141 logging.debug("Writing cluster domain secret to %s", cds_file)
142 utils.WriteFile(cds_file, data=cds, backup=True)
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)
149 def _InitGanetiServerSetup(master_name):
150 """Setup the necessary configuration for the initial node daemon.
152 This creates the nodepass file containing the shared password for
153 the cluster, generates the SSL certificate and starts the node daemon.
155 @type master_name: str
156 @param master_name: Name of the master node
159 # Generate cluster secrets
160 GenerateClusterCrypto(True, False, False, False)
162 result = utils.RunCmd([constants.DAEMON_UTIL, "start", constants.NODED])
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))
168 _WaitForNodeDaemon(master_name)
171 def _WaitForNodeDaemon(node_name):
172 """Wait for node daemon to become responsive.
175 def _CheckNodeDaemon():
176 result = rpc.RpcRunner.call_version([node_name])[node_name]
178 raise utils.RetryAgain()
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)
187 def _InitFileStorage(file_storage_dir):
188 """Initialize if needed the file storage.
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
195 if not constants.ENABLE_FILE_STORAGE:
198 file_storage_dir = os.path.normpath(file_storage_dir)
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)
204 if not os.path.exists(file_storage_dir):
206 os.makedirs(file_storage_dir, 0750)
208 raise errors.OpPrereqError("Cannot create file storage directory"
209 " '%s': %s" % (file_storage_dir, err),
210 errors.ECODE_ENVIRON)
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
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 primary_ip_version=None):
228 """Initialise the cluster.
230 @type candidate_pool_size: int
231 @param candidate_pool_size: master candidate pool size
234 # TODO: complete the docstring
235 if config.ConfigWriter.IsCluster():
236 raise errors.OpPrereqError("Cluster is already initialised",
239 if not enabled_hypervisors:
240 raise errors.OpPrereqError("Enabled hypervisors list must contain at"
241 " least one member", errors.ECODE_INVAL)
242 invalid_hvs = set(enabled_hypervisors) - constants.HYPER_TYPES
244 raise errors.OpPrereqError("Enabled hypervisors contains invalid"
245 " entries: %s" % invalid_hvs,
250 if primary_ip_version == constants.IP4_VERSION:
251 ipcls = netutils.IP4Address
252 elif primary_ip_version == constants.IP6_VERSION:
253 ipcls = netutils.IP6Address
255 raise errors.OpPrereqError("Invalid primary ip version: %d." %
258 hostname = netutils.GetHostname(family=ipcls.family)
259 if not ipcls.IsValid(hostname.ip):
260 raise errors.OpPrereqError("This host's IP (%s) is not a valid IPv%d"
261 " address." % (hostname.ip, primary_ip_version))
263 if ipcls.IsLoopback(hostname.ip):
264 raise errors.OpPrereqError("This host's IP (%s) resolves to a loopback"
265 " address. Please fix DNS or %s." %
266 (hostname.ip, constants.ETC_HOSTS),
267 errors.ECODE_ENVIRON)
269 if not ipcls.Own(hostname.ip):
270 raise errors.OpPrereqError("Inconsistency: this host's name resolves"
271 " to %s,\nbut this ip address does not"
272 " belong to this host" %
273 hostname.ip, errors.ECODE_ENVIRON)
275 clustername = netutils.GetHostname(name=cluster_name, family=ipcls.family)
277 if netutils.TcpPing(clustername.ip, constants.DEFAULT_NODED_PORT, timeout=5):
278 raise errors.OpPrereqError("Cluster IP already active",
279 errors.ECODE_NOTUNIQUE)
282 if primary_ip_version == constants.IP6_VERSION:
283 raise errors.OpPrereqError("When using a IPv6 primary address, a valid"
284 " IPv4 address must be given as secondary",
286 secondary_ip = hostname.ip
288 if not netutils.IP4Address.IsValid(secondary_ip):
289 raise errors.OpPrereqError("Secondary IP address (%s) has to be a valid"
290 " IPv4 address." % secondary_ip,
293 if not netutils.IP4Address.Own(secondary_ip):
294 raise errors.OpPrereqError("You gave %s as secondary IP,"
295 " but it does not belong to this host." %
296 secondary_ip, errors.ECODE_ENVIRON)
298 if vg_name is not None:
299 # Check if volume group is valid
300 vgstatus = utils.CheckVolumeGroupSize(utils.ListVolumeGroups(), vg_name,
301 constants.MIN_VG_SIZE)
303 raise errors.OpPrereqError("Error: %s\nspecify --no-lvm-storage if"
304 " you are not using lvm" % vgstatus,
307 if drbd_helper is not None:
309 curr_helper = bdev.BaseDRBD.GetUsermodeHelper()
310 except errors.BlockDeviceError, err:
311 raise errors.OpPrereqError("Error while checking drbd helper"
312 " (specify --no-drbd-storage if you are not"
313 " using drbd): %s" % str(err),
314 errors.ECODE_ENVIRON)
315 if drbd_helper != curr_helper:
316 raise errors.OpPrereqError("Error: requiring %s as drbd helper but %s"
317 " is the current helper" % (drbd_helper,
321 file_storage_dir = _InitFileStorage(file_storage_dir)
323 if not re.match("^[0-9a-z]{2}:[0-9a-z]{2}:[0-9a-z]{2}$", mac_prefix):
324 raise errors.OpPrereqError("Invalid mac prefix given '%s'" % mac_prefix,
327 result = utils.RunCmd(["ip", "link", "show", "dev", master_netdev])
329 raise errors.OpPrereqError("Invalid master netdev given (%s): '%s'" %
331 result.output.strip()), errors.ECODE_INVAL)
333 dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE)]
334 utils.EnsureDirs(dirs)
336 utils.ForceDictType(beparams, constants.BES_PARAMETER_TYPES)
337 utils.ForceDictType(nicparams, constants.NICS_PARAMETER_TYPES)
338 objects.NIC.CheckParameterSyntax(nicparams)
340 # hvparams is a mapping of hypervisor->hvparams dict
341 for hv_name, hv_params in hvparams.iteritems():
342 utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
343 hv_class = hypervisor.GetHypervisor(hv_name)
344 hv_class.CheckParameterSyntax(hv_params)
346 # set up ssh config and /etc/hosts
347 sshline = utils.ReadFile(constants.SSH_HOST_RSA_PUB)
348 sshkey = sshline.split(" ")[1]
351 utils.AddHostToEtcHosts(hostname)
356 if default_iallocator is not None:
357 alloc_script = utils.FindFile(default_iallocator,
358 constants.IALLOCATOR_SEARCH_PATH,
360 if alloc_script is None:
361 raise errors.OpPrereqError("Invalid default iallocator script '%s'"
362 " specified" % default_iallocator,
367 # init of cluster config file
368 cluster_config = objects.Cluster(
370 rsahostkeypub=sshkey,
371 highest_used_port=(constants.FIRST_DRBD_PORT - 1),
372 mac_prefix=mac_prefix,
373 volume_group_name=vg_name,
374 tcpudp_port_pool=set(),
375 master_node=hostname.name,
376 master_ip=clustername.ip,
377 master_netdev=master_netdev,
378 cluster_name=clustername.name,
379 file_storage_dir=file_storage_dir,
380 enabled_hypervisors=enabled_hypervisors,
381 beparams={constants.PP_DEFAULT: beparams},
382 nicparams={constants.PP_DEFAULT: nicparams},
384 candidate_pool_size=candidate_pool_size,
385 modify_etc_hosts=modify_etc_hosts,
386 modify_ssh_setup=modify_ssh_setup,
390 uuid=utils.NewUUID(),
391 maintain_node_health=maintain_node_health,
392 drbd_usermode_helper=drbd_helper,
393 default_iallocator=default_iallocator,
394 primary_ip_family=ipcls.family,
396 master_node_config = objects.Node(name=hostname.name,
397 primary_ip=hostname.ip,
398 secondary_ip=secondary_ip,
400 master_candidate=True,
401 offline=False, drained=False,
403 InitConfig(constants.CONFIG_VERSION, cluster_config, master_node_config)
404 cfg = config.ConfigWriter(offline=True)
405 ssh.WriteKnownHostsFile(cfg, constants.SSH_KNOWN_HOSTS_FILE)
406 cfg.Update(cfg.GetClusterInfo(), logging.error)
407 backend.WriteSsconfFiles(cfg.GetSsconfValues())
409 # set up the inter-node password and certificate
410 _InitGanetiServerSetup(hostname.name)
412 # start the master ip
413 # TODO: Review rpc call from bootstrap
414 # TODO: Warn on failed start master
415 rpc.RpcRunner.call_node_start_master(hostname.name, True, False)
418 def InitConfig(version, cluster_config, master_node_config,
419 cfg_file=constants.CLUSTER_CONF_FILE):
420 """Create the initial cluster configuration.
422 It will contain the current node, which will also be the master
423 node, and no instances.
426 @param version: configuration version
427 @type cluster_config: L{objects.Cluster}
428 @param cluster_config: cluster configuration
429 @type master_node_config: L{objects.Node}
430 @param master_node_config: master node configuration
431 @type cfg_file: string
432 @param cfg_file: configuration file path
436 master_node_config.name: master_node_config,
440 config_data = objects.ConfigData(version=version,
441 cluster=cluster_config,
445 ctime=now, mtime=now)
446 utils.WriteFile(cfg_file,
447 data=serializer.Dump(config_data.ToDict()),
451 def FinalizeClusterDestroy(master):
452 """Execute the last steps of cluster destroy
454 This function shuts down all the daemons, completing the destroy
455 begun in cmdlib.LUDestroyOpcode.
458 cfg = config.ConfigWriter()
459 modify_ssh_setup = cfg.GetClusterInfo().modify_ssh_setup
460 result = rpc.RpcRunner.call_node_stop_master(master, True)
461 msg = result.fail_msg
463 logging.warning("Could not disable the master role: %s", msg)
464 result = rpc.RpcRunner.call_node_leave_cluster(master, modify_ssh_setup)
465 msg = result.fail_msg
467 logging.warning("Could not shutdown the node daemon and cleanup"
468 " the node: %s", msg)
471 def SetupNodeDaemon(cluster_name, node, ssh_key_check):
472 """Add a node to the cluster.
474 This function must be called before the actual opcode, and will ssh
475 to the remote node, copy the needed files, and start ganeti-noded,
476 allowing the master to do the rest via normal rpc calls.
478 @param cluster_name: the cluster name
479 @param node: the name of the new node
480 @param ssh_key_check: whether to do a strict key check
483 family = ssconf.SimpleStore().GetPrimaryIPFamily()
484 sshrunner = ssh.SshRunner(cluster_name,
485 ipv6=family==netutils.IP6Address.family)
487 noded_cert = utils.ReadFile(constants.NODED_CERT_FILE)
488 rapi_cert = utils.ReadFile(constants.RAPI_CERT_FILE)
489 confd_hmac_key = utils.ReadFile(constants.CONFD_HMAC_KEY)
491 # in the base64 pem encoding, neither '!' nor '.' are valid chars,
492 # so we use this to detect an invalid certificate; as long as the
493 # cert doesn't contain this, the here-document will be correctly
494 # parsed by the shell sequence below. HMAC keys are hexadecimal strings,
495 # so the same restrictions apply.
496 for content in (noded_cert, rapi_cert, confd_hmac_key):
497 if re.search('^!EOF\.', content, re.MULTILINE):
498 raise errors.OpExecError("invalid SSL certificate or HMAC key")
500 if not noded_cert.endswith("\n"):
502 if not rapi_cert.endswith("\n"):
504 if not confd_hmac_key.endswith("\n"):
505 confd_hmac_key += "\n"
507 bind_address = constants.IP4_ADDRESS_ANY
508 if family == netutils.IP6Address.family:
509 bind_address = constants.IP6_ADDRESS_ANY
511 # set up inter-node password and certificate and restarts the node daemon
512 # and then connect with ssh to set password and start ganeti-noded
513 # note that all the below variables are sanitized at this point,
514 # either by being constants or by the checks above
515 # TODO: Could this command exceed a shell's maximum command length?
516 mycommand = ("umask 077 && "
517 "cat > '%s' << '!EOF.' && \n"
519 "cat > '%s' << '!EOF.' && \n"
521 "cat > '%s' << '!EOF.' && \n"
523 "chmod 0400 %s %s %s && "
524 "%s start %s -b '%s'" %
525 (constants.NODED_CERT_FILE, noded_cert,
526 constants.RAPI_CERT_FILE, rapi_cert,
527 constants.CONFD_HMAC_KEY, confd_hmac_key,
528 constants.NODED_CERT_FILE, constants.RAPI_CERT_FILE,
529 constants.CONFD_HMAC_KEY,
530 constants.DAEMON_UTIL, constants.NODED, bind_address))
532 result = sshrunner.Run(node, 'root', mycommand, batch=False,
533 ask_key=ssh_key_check,
534 use_cluster_key=False,
535 strict_host_check=ssh_key_check)
537 raise errors.OpExecError("Remote command on node %s, error: %s,"
539 (node, result.fail_reason, result.output))
541 _WaitForNodeDaemon(node)
544 def MasterFailover(no_voting=False):
545 """Failover the master node.
547 This checks that we are not already the master, and will cause the
548 current master to cease being master, and the non-master to become
551 @type no_voting: boolean
552 @param no_voting: force the operation without remote nodes agreement
556 sstore = ssconf.SimpleStore()
558 old_master, new_master = ssconf.GetMasterAndMyself(sstore)
559 node_list = sstore.GetNodeList()
560 mc_list = sstore.GetMasterCandidates()
562 if old_master == new_master:
563 raise errors.OpPrereqError("This commands must be run on the node"
564 " where you want the new master to be."
565 " %s is already the master" %
566 old_master, errors.ECODE_INVAL)
568 if new_master not in mc_list:
569 mc_no_master = [name for name in mc_list if name != old_master]
570 raise errors.OpPrereqError("This node is not among the nodes marked"
571 " as master candidates. Only these nodes"
572 " can become masters. Current list of"
573 " master candidates is:\n"
574 "%s" % ('\n'.join(mc_no_master)),
578 vote_list = GatherMasterVotes(node_list)
581 voted_master = vote_list[0][0]
582 if voted_master is None:
583 raise errors.OpPrereqError("Cluster is inconsistent, most nodes did"
584 " not respond.", errors.ECODE_ENVIRON)
585 elif voted_master != old_master:
586 raise errors.OpPrereqError("I have a wrong configuration, I believe"
587 " the master is %s but the other nodes"
588 " voted %s. Please resync the configuration"
590 (old_master, voted_master),
596 logging.info("Setting master to %s, old master: %s", new_master, old_master)
598 result = rpc.RpcRunner.call_node_stop_master(old_master, True)
599 msg = result.fail_msg
601 logging.error("Could not disable the master role on the old master"
602 " %s, please disable manually: %s", old_master, msg)
604 master_ip = sstore.GetMasterIP()
606 # Here we have a phase where no master should be running
608 if netutils.TcpPing(master_ip, constants.DEFAULT_NODED_PORT):
609 raise utils.RetryAgain()
612 utils.Retry(_check_ip, (1, 1.5, 5), total_timeout)
613 except utils.RetryTimeout:
614 logging.warning("The master IP is still reachable after %s seconds,"
615 " continuing but activating the master on the current"
616 " node will probably fail", total_timeout)
618 # instantiate a real config writer, as we now know we have the
620 cfg = config.ConfigWriter()
622 cluster_info = cfg.GetClusterInfo()
623 cluster_info.master_node = new_master
624 # this will also regenerate the ssconf files, since we updated the
626 cfg.Update(cluster_info, logging.error)
628 result = rpc.RpcRunner.call_node_start_master(new_master, True, no_voting)
629 msg = result.fail_msg
631 logging.error("Could not start the master role on the new master"
632 " %s, please check: %s", new_master, msg)
639 """Returns the current master node.
641 This is a separate function in bootstrap since it's needed by
642 gnt-cluster, and instead of importing directly ssconf, it's better
643 to abstract it in bootstrap, where we do use ssconf in other
647 sstore = ssconf.SimpleStore()
649 old_master, _ = ssconf.GetMasterAndMyself(sstore)
654 def GatherMasterVotes(node_list):
655 """Check the agreement on who is the master.
657 This function will return a list of (node, number of votes), ordered
658 by the number of votes. Errors will be denoted by the key 'None'.
660 Note that the sum of votes is the number of nodes this machine
661 knows, whereas the number of entries in the list could be different
662 (if some nodes vote for another master).
664 We remove ourselves from the list since we know that (bugs aside)
665 since we use the same source for configuration information for both
666 backend and boostrap, we'll always vote for ourselves.
668 @type node_list: list
669 @param node_list: the list of nodes to query for master info; the current
670 node will be removed if it is in the list
672 @return: list of (node, votes)
675 myself = netutils.Hostname.GetSysName()
677 node_list.remove(myself)
681 # no nodes left (eventually after removing myself)
683 results = rpc.RpcRunner.call_master_info(node_list)
684 if not isinstance(results, dict):
685 # this should not happen (unless internal error in rpc)
686 logging.critical("Can't complete rpc call, aborting master startup")
687 return [(None, len(node_list))]
695 logging.warning("Error contacting node %s: %s", node, msg)
697 # for now we accept both length 3 and 4 (data[3] is primary ip version)
698 elif not isinstance(data, (tuple, list)) or len(data) < 3:
699 logging.warning("Invalid data received from node %s: %s", node, data)
702 if None not in votes:
706 master_node = data[2]
707 if master_node not in votes:
708 votes[master_node] = 0
709 votes[master_node] += 1
711 vote_list = [v for v in votes.items()]
712 # sort first on number of votes then on name, since we want None
713 # sorted later if we have the half of the nodes not responding, and
714 # half voting all for the same master
715 vote_list.sort(key=lambda x: (x[1], x[0]), reverse=True)