4 # Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 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 """Configuration management for Ganeti
24 This module provides the interface to the Ganeti cluster configuration.
26 The configuration data is stored on every node but is updated on the master
27 only. After each update, the master distributes the data to the other nodes.
29 Currently, the data storage format is JSON. YAML was slow and consuming too
34 # pylint: disable-msg=R0904
35 # R0904: Too many public methods
42 from ganeti import errors
43 from ganeti import locking
44 from ganeti import utils
45 from ganeti import constants
46 from ganeti import rpc
47 from ganeti import objects
48 from ganeti import serializer
49 from ganeti import uidpool
50 from ganeti import netutils
51 from ganeti import runtime
54 _config_lock = locking.SharedLock("ConfigWriter")
56 # job id used for resource management at config upgrade time
57 _UPGRADE_CONFIG_JID = "jid-cfg-upgrade"
60 def _ValidateConfig(data):
61 """Verifies that a configuration objects looks valid.
63 This only verifies the version of the configuration.
65 @raise errors.ConfigurationError: if the version differs from what
69 if data.version != constants.CONFIG_VERSION:
70 raise errors.ConfigVersionMismatch(constants.CONFIG_VERSION, data.version)
73 class TemporaryReservationManager:
74 """A temporary resource reservation manager.
76 This is used to reserve resources in a job, before using them, making sure
77 other jobs cannot get them in the meantime.
81 self._ec_reserved = {}
83 def Reserved(self, resource):
84 for holder_reserved in self._ec_reserved.values():
85 if resource in holder_reserved:
89 def Reserve(self, ec_id, resource):
90 if self.Reserved(resource):
91 raise errors.ReservationError("Duplicate reservation for resource '%s'"
93 if ec_id not in self._ec_reserved:
94 self._ec_reserved[ec_id] = set([resource])
96 self._ec_reserved[ec_id].add(resource)
98 def DropECReservations(self, ec_id):
99 if ec_id in self._ec_reserved:
100 del self._ec_reserved[ec_id]
102 def GetReserved(self):
104 for holder_reserved in self._ec_reserved.values():
105 all_reserved.update(holder_reserved)
108 def Generate(self, existing, generate_one_fn, ec_id):
109 """Generate a new resource of this type
112 assert callable(generate_one_fn)
114 all_elems = self.GetReserved()
115 all_elems.update(existing)
118 new_resource = generate_one_fn()
119 if new_resource is not None and new_resource not in all_elems:
122 raise errors.ConfigurationError("Not able generate new resource"
123 " (last tried: %s)" % new_resource)
124 self.Reserve(ec_id, new_resource)
128 def _MatchNameComponentIgnoreCase(short_name, names):
129 """Wrapper around L{utils.text.MatchNameComponent}.
132 return utils.MatchNameComponent(short_name, names, case_sensitive=False)
136 """The interface to the cluster configuration.
138 @ivar _temporary_lvs: reservation manager for temporary LVs
139 @ivar _all_rms: a list of all temporary reservation managers
142 def __init__(self, cfg_file=None, offline=False, _getents=runtime.GetEnts,
143 accept_foreign=False):
145 self._lock = _config_lock
146 self._config_data = None
147 self._offline = offline
149 self._cfg_file = constants.CLUSTER_CONF_FILE
151 self._cfg_file = cfg_file
152 self._getents = _getents
153 self._temporary_ids = TemporaryReservationManager()
154 self._temporary_drbds = {}
155 self._temporary_macs = TemporaryReservationManager()
156 self._temporary_secrets = TemporaryReservationManager()
157 self._temporary_lvs = TemporaryReservationManager()
158 self._all_rms = [self._temporary_ids, self._temporary_macs,
159 self._temporary_secrets, self._temporary_lvs]
160 # Note: in order to prevent errors when resolving our name in
161 # _DistributeConfig, we compute it here once and reuse it; it's
162 # better to raise an error before starting to modify the config
163 # file than after it was modified
164 self._my_hostname = netutils.Hostname.GetSysName()
165 self._last_cluster_serial = -1
167 self._OpenConfig(accept_foreign)
169 # this method needs to be static, so that we can call it on the class
172 """Check if the cluster is configured.
175 return os.path.exists(constants.CLUSTER_CONF_FILE)
177 def _GenerateOneMAC(self):
178 """Generate one mac address
181 prefix = self._config_data.cluster.mac_prefix
182 byte1 = random.randrange(0, 256)
183 byte2 = random.randrange(0, 256)
184 byte3 = random.randrange(0, 256)
185 mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3)
188 @locking.ssynchronized(_config_lock, shared=1)
189 def GetNdParams(self, node):
190 """Get the node params populated with cluster defaults.
192 @type node: L{objects.Node}
193 @param node: The node we want to know the params for
194 @return: A dict with the filled in node params
197 nodegroup = self._UnlockedGetNodeGroup(node.group)
198 return self._config_data.cluster.FillND(node, nodegroup)
200 @locking.ssynchronized(_config_lock, shared=1)
201 def GenerateMAC(self, ec_id):
202 """Generate a MAC for an instance.
204 This should check the current instances for duplicates.
207 existing = self._AllMACs()
208 return self._temporary_ids.Generate(existing, self._GenerateOneMAC, ec_id)
210 @locking.ssynchronized(_config_lock, shared=1)
211 def ReserveMAC(self, mac, ec_id):
212 """Reserve a MAC for an instance.
214 This only checks instances managed by this cluster, it does not
215 check for potential collisions elsewhere.
218 all_macs = self._AllMACs()
220 raise errors.ReservationError("mac already in use")
222 self._temporary_macs.Reserve(ec_id, mac)
224 @locking.ssynchronized(_config_lock, shared=1)
225 def ReserveLV(self, lv_name, ec_id):
226 """Reserve an VG/LV pair for an instance.
228 @type lv_name: string
229 @param lv_name: the logical volume name to reserve
232 all_lvs = self._AllLVs()
233 if lv_name in all_lvs:
234 raise errors.ReservationError("LV already in use")
236 self._temporary_lvs.Reserve(ec_id, lv_name)
238 @locking.ssynchronized(_config_lock, shared=1)
239 def GenerateDRBDSecret(self, ec_id):
240 """Generate a DRBD secret.
242 This checks the current disks for duplicates.
245 return self._temporary_secrets.Generate(self._AllDRBDSecrets(),
246 utils.GenerateSecret,
250 """Compute the list of all LVs.
254 for instance in self._config_data.instances.values():
255 node_data = instance.MapLVsByNode()
256 for lv_list in node_data.values():
257 lvnames.update(lv_list)
260 def _AllIDs(self, include_temporary):
261 """Compute the list of all UUIDs and names we have.
263 @type include_temporary: boolean
264 @param include_temporary: whether to include the _temporary_ids set
266 @return: a set of IDs
270 if include_temporary:
271 existing.update(self._temporary_ids.GetReserved())
272 existing.update(self._AllLVs())
273 existing.update(self._config_data.instances.keys())
274 existing.update(self._config_data.nodes.keys())
275 existing.update([i.uuid for i in self._AllUUIDObjects() if i.uuid])
278 def _GenerateUniqueID(self, ec_id):
279 """Generate an unique UUID.
281 This checks the current node, instances and disk names for
285 @return: the unique id
288 existing = self._AllIDs(include_temporary=False)
289 return self._temporary_ids.Generate(existing, utils.NewUUID, ec_id)
291 @locking.ssynchronized(_config_lock, shared=1)
292 def GenerateUniqueID(self, ec_id):
293 """Generate an unique ID.
295 This is just a wrapper over the unlocked version.
298 @param ec_id: unique id for the job to reserve the id to
301 return self._GenerateUniqueID(ec_id)
304 """Return all MACs present in the config.
307 @return: the list of all MACs
311 for instance in self._config_data.instances.values():
312 for nic in instance.nics:
313 result.append(nic.mac)
317 def _AllDRBDSecrets(self):
318 """Return all DRBD secrets present in the config.
321 @return: the list of all DRBD secrets
324 def helper(disk, result):
325 """Recursively gather secrets from this disk."""
326 if disk.dev_type == constants.DT_DRBD8:
327 result.append(disk.logical_id[5])
329 for child in disk.children:
330 helper(child, result)
333 for instance in self._config_data.instances.values():
334 for disk in instance.disks:
339 def _CheckDiskIDs(self, disk, l_ids, p_ids):
340 """Compute duplicate disk IDs
342 @type disk: L{objects.Disk}
343 @param disk: the disk at which to start searching
345 @param l_ids: list of current logical ids
347 @param p_ids: list of current physical ids
349 @return: a list of error messages
353 if disk.logical_id is not None:
354 if disk.logical_id in l_ids:
355 result.append("duplicate logical id %s" % str(disk.logical_id))
357 l_ids.append(disk.logical_id)
358 if disk.physical_id is not None:
359 if disk.physical_id in p_ids:
360 result.append("duplicate physical id %s" % str(disk.physical_id))
362 p_ids.append(disk.physical_id)
365 for child in disk.children:
366 result.extend(self._CheckDiskIDs(child, l_ids, p_ids))
369 def _UnlockedVerifyConfig(self):
373 @return: a list of error messages; a non-empty list signifies
377 # pylint: disable-msg=R0914
381 data = self._config_data
382 cluster = data.cluster
386 # global cluster checks
387 if not cluster.enabled_hypervisors:
388 result.append("enabled hypervisors list doesn't have any entries")
389 invalid_hvs = set(cluster.enabled_hypervisors) - constants.HYPER_TYPES
391 result.append("enabled hypervisors contains invalid entries: %s" %
393 missing_hvp = (set(cluster.enabled_hypervisors) -
394 set(cluster.hvparams.keys()))
396 result.append("hypervisor parameters missing for the enabled"
397 " hypervisor(s) %s" % utils.CommaJoin(missing_hvp))
399 if cluster.master_node not in data.nodes:
400 result.append("cluster has invalid primary node '%s'" %
403 def _helper(owner, attr, value, template):
405 utils.ForceDictType(value, template)
406 except errors.GenericError, err:
407 result.append("%s has invalid %s: %s" % (owner, attr, err))
409 def _helper_nic(owner, params):
411 objects.NIC.CheckParameterSyntax(params)
412 except errors.ConfigurationError, err:
413 result.append("%s has invalid nicparams: %s" % (owner, err))
415 # check cluster parameters
416 _helper("cluster", "beparams", cluster.SimpleFillBE({}),
417 constants.BES_PARAMETER_TYPES)
418 _helper("cluster", "nicparams", cluster.SimpleFillNIC({}),
419 constants.NICS_PARAMETER_TYPES)
420 _helper_nic("cluster", cluster.SimpleFillNIC({}))
421 _helper("cluster", "ndparams", cluster.SimpleFillND({}),
422 constants.NDS_PARAMETER_TYPES)
424 # per-instance checks
425 for instance_name in data.instances:
426 instance = data.instances[instance_name]
427 if instance.name != instance_name:
428 result.append("instance '%s' is indexed by wrong name '%s'" %
429 (instance.name, instance_name))
430 if instance.primary_node not in data.nodes:
431 result.append("instance '%s' has invalid primary node '%s'" %
432 (instance_name, instance.primary_node))
433 for snode in instance.secondary_nodes:
434 if snode not in data.nodes:
435 result.append("instance '%s' has invalid secondary node '%s'" %
436 (instance_name, snode))
437 for idx, nic in enumerate(instance.nics):
438 if nic.mac in seen_macs:
439 result.append("instance '%s' has NIC %d mac %s duplicate" %
440 (instance_name, idx, nic.mac))
442 seen_macs.append(nic.mac)
444 filled = cluster.SimpleFillNIC(nic.nicparams)
445 owner = "instance %s nic %d" % (instance.name, idx)
446 _helper(owner, "nicparams",
447 filled, constants.NICS_PARAMETER_TYPES)
448 _helper_nic(owner, filled)
451 if instance.beparams:
452 _helper("instance %s" % instance.name, "beparams",
453 cluster.FillBE(instance), constants.BES_PARAMETER_TYPES)
455 # gather the drbd ports for duplicate checks
456 for dsk in instance.disks:
457 if dsk.dev_type in constants.LDS_DRBD:
458 tcp_port = dsk.logical_id[2]
459 if tcp_port not in ports:
461 ports[tcp_port].append((instance.name, "drbd disk %s" % dsk.iv_name))
462 # gather network port reservation
463 net_port = getattr(instance, "network_port", None)
464 if net_port is not None:
465 if net_port not in ports:
467 ports[net_port].append((instance.name, "network port"))
469 # instance disk verify
470 for idx, disk in enumerate(instance.disks):
471 result.extend(["instance '%s' disk %d error: %s" %
472 (instance.name, idx, msg) for msg in disk.Verify()])
473 result.extend(self._CheckDiskIDs(disk, seen_lids, seen_pids))
475 # cluster-wide pool of free ports
476 for free_port in cluster.tcpudp_port_pool:
477 if free_port not in ports:
478 ports[free_port] = []
479 ports[free_port].append(("cluster", "port marked as free"))
481 # compute tcp/udp duplicate ports
487 txt = utils.CommaJoin(["%s/%s" % val for val in pdata])
488 result.append("tcp/udp port %s has duplicates: %s" % (pnum, txt))
490 # highest used tcp port check
492 if keys[-1] > cluster.highest_used_port:
493 result.append("Highest used port mismatch, saved %s, computed %s" %
494 (cluster.highest_used_port, keys[-1]))
496 if not data.nodes[cluster.master_node].master_candidate:
497 result.append("Master node is not a master candidate")
499 # master candidate checks
500 mc_now, mc_max, _ = self._UnlockedGetMasterCandidateStats()
502 result.append("Not enough master candidates: actual %d, target %d" %
506 for node_name, node in data.nodes.items():
507 if node.name != node_name:
508 result.append("Node '%s' is indexed by wrong name '%s'" %
509 (node.name, node_name))
510 if [node.master_candidate, node.drained, node.offline].count(True) > 1:
511 result.append("Node %s state is invalid: master_candidate=%s,"
512 " drain=%s, offline=%s" %
513 (node.name, node.master_candidate, node.drained,
515 if node.group not in data.nodegroups:
516 result.append("Node '%s' has invalid group '%s'" %
517 (node.name, node.group))
519 _helper("node %s" % node.name, "ndparams",
520 cluster.FillND(node, data.nodegroups[node.group]),
521 constants.NDS_PARAMETER_TYPES)
524 nodegroups_names = set()
525 for nodegroup_uuid in data.nodegroups:
526 nodegroup = data.nodegroups[nodegroup_uuid]
527 if nodegroup.uuid != nodegroup_uuid:
528 result.append("node group '%s' (uuid: '%s') indexed by wrong uuid '%s'"
529 % (nodegroup.name, nodegroup.uuid, nodegroup_uuid))
530 if utils.UUID_RE.match(nodegroup.name.lower()):
531 result.append("node group '%s' (uuid: '%s') has uuid-like name" %
532 (nodegroup.name, nodegroup.uuid))
533 if nodegroup.name in nodegroups_names:
534 result.append("duplicate node group name '%s'" % nodegroup.name)
536 nodegroups_names.add(nodegroup.name)
537 if nodegroup.ndparams:
538 _helper("group %s" % nodegroup.name, "ndparams",
539 cluster.SimpleFillND(nodegroup.ndparams),
540 constants.NDS_PARAMETER_TYPES)
544 _, duplicates = self._UnlockedComputeDRBDMap()
545 for node, minor, instance_a, instance_b in duplicates:
546 result.append("DRBD minor %d on node %s is assigned twice to instances"
547 " %s and %s" % (minor, node, instance_a, instance_b))
550 default_nicparams = cluster.nicparams[constants.PP_DEFAULT]
553 def _AddIpAddress(ip, name):
554 ips.setdefault(ip, []).append(name)
556 _AddIpAddress(cluster.master_ip, "cluster_ip")
558 for node in data.nodes.values():
559 _AddIpAddress(node.primary_ip, "node:%s/primary" % node.name)
560 if node.secondary_ip != node.primary_ip:
561 _AddIpAddress(node.secondary_ip, "node:%s/secondary" % node.name)
563 for instance in data.instances.values():
564 for idx, nic in enumerate(instance.nics):
568 nicparams = objects.FillDict(default_nicparams, nic.nicparams)
569 nic_mode = nicparams[constants.NIC_MODE]
570 nic_link = nicparams[constants.NIC_LINK]
572 if nic_mode == constants.NIC_MODE_BRIDGED:
573 link = "bridge:%s" % nic_link
574 elif nic_mode == constants.NIC_MODE_ROUTED:
575 link = "route:%s" % nic_link
577 raise errors.ProgrammerError("NIC mode '%s' not handled" % nic_mode)
579 _AddIpAddress("%s/%s" % (link, nic.ip),
580 "instance:%s/nic:%d" % (instance.name, idx))
582 for ip, owners in ips.items():
584 result.append("IP address %s is used by multiple owners: %s" %
585 (ip, utils.CommaJoin(owners)))
589 @locking.ssynchronized(_config_lock, shared=1)
590 def VerifyConfig(self):
593 This is just a wrapper over L{_UnlockedVerifyConfig}.
596 @return: a list of error messages; a non-empty list signifies
600 return self._UnlockedVerifyConfig()
602 def _UnlockedSetDiskID(self, disk, node_name):
603 """Convert the unique ID to the ID needed on the target nodes.
605 This is used only for drbd, which needs ip/port configuration.
607 The routine descends down and updates its children also, because
608 this helps when the only the top device is passed to the remote
611 This function is for internal use, when the config lock is already held.
615 for child in disk.children:
616 self._UnlockedSetDiskID(child, node_name)
618 if disk.logical_id is None and disk.physical_id is not None:
620 if disk.dev_type == constants.LD_DRBD8:
621 pnode, snode, port, pminor, sminor, secret = disk.logical_id
622 if node_name not in (pnode, snode):
623 raise errors.ConfigurationError("DRBD device not knowing node %s" %
625 pnode_info = self._UnlockedGetNodeInfo(pnode)
626 snode_info = self._UnlockedGetNodeInfo(snode)
627 if pnode_info is None or snode_info is None:
628 raise errors.ConfigurationError("Can't find primary or secondary node"
629 " for %s" % str(disk))
630 p_data = (pnode_info.secondary_ip, port)
631 s_data = (snode_info.secondary_ip, port)
632 if pnode == node_name:
633 disk.physical_id = p_data + s_data + (pminor, secret)
634 else: # it must be secondary, we tested above
635 disk.physical_id = s_data + p_data + (sminor, secret)
637 disk.physical_id = disk.logical_id
640 @locking.ssynchronized(_config_lock)
641 def SetDiskID(self, disk, node_name):
642 """Convert the unique ID to the ID needed on the target nodes.
644 This is used only for drbd, which needs ip/port configuration.
646 The routine descends down and updates its children also, because
647 this helps when the only the top device is passed to the remote
651 return self._UnlockedSetDiskID(disk, node_name)
653 @locking.ssynchronized(_config_lock)
654 def AddTcpUdpPort(self, port):
655 """Adds a new port to the available port pool.
658 if not isinstance(port, int):
659 raise errors.ProgrammerError("Invalid type passed for port")
661 self._config_data.cluster.tcpudp_port_pool.add(port)
664 @locking.ssynchronized(_config_lock, shared=1)
665 def GetPortList(self):
666 """Returns a copy of the current port list.
669 return self._config_data.cluster.tcpudp_port_pool.copy()
671 @locking.ssynchronized(_config_lock)
672 def AllocatePort(self):
675 The port will be taken from the available port pool or from the
676 default port range (and in this case we increase
680 # If there are TCP/IP ports configured, we use them first.
681 if self._config_data.cluster.tcpudp_port_pool:
682 port = self._config_data.cluster.tcpudp_port_pool.pop()
684 port = self._config_data.cluster.highest_used_port + 1
685 if port >= constants.LAST_DRBD_PORT:
686 raise errors.ConfigurationError("The highest used port is greater"
687 " than %s. Aborting." %
688 constants.LAST_DRBD_PORT)
689 self._config_data.cluster.highest_used_port = port
694 def _UnlockedComputeDRBDMap(self):
695 """Compute the used DRBD minor/nodes.
698 @return: dictionary of node_name: dict of minor: instance_name;
699 the returned dict will have all the nodes in it (even if with
700 an empty list), and a list of duplicates; if the duplicates
701 list is not empty, the configuration is corrupted and its caller
702 should raise an exception
705 def _AppendUsedPorts(instance_name, disk, used):
707 if disk.dev_type == constants.LD_DRBD8 and len(disk.logical_id) >= 5:
708 node_a, node_b, _, minor_a, minor_b = disk.logical_id[:5]
709 for node, port in ((node_a, minor_a), (node_b, minor_b)):
710 assert node in used, ("Node '%s' of instance '%s' not found"
711 " in node list" % (node, instance_name))
712 if port in used[node]:
713 duplicates.append((node, port, instance_name, used[node][port]))
715 used[node][port] = instance_name
717 for child in disk.children:
718 duplicates.extend(_AppendUsedPorts(instance_name, child, used))
722 my_dict = dict((node, {}) for node in self._config_data.nodes)
723 for instance in self._config_data.instances.itervalues():
724 for disk in instance.disks:
725 duplicates.extend(_AppendUsedPorts(instance.name, disk, my_dict))
726 for (node, minor), instance in self._temporary_drbds.iteritems():
727 if minor in my_dict[node] and my_dict[node][minor] != instance:
728 duplicates.append((node, minor, instance, my_dict[node][minor]))
730 my_dict[node][minor] = instance
731 return my_dict, duplicates
733 @locking.ssynchronized(_config_lock)
734 def ComputeDRBDMap(self):
735 """Compute the used DRBD minor/nodes.
737 This is just a wrapper over L{_UnlockedComputeDRBDMap}.
739 @return: dictionary of node_name: dict of minor: instance_name;
740 the returned dict will have all the nodes in it (even if with
744 d_map, duplicates = self._UnlockedComputeDRBDMap()
746 raise errors.ConfigurationError("Duplicate DRBD ports detected: %s" %
750 @locking.ssynchronized(_config_lock)
751 def AllocateDRBDMinor(self, nodes, instance):
752 """Allocate a drbd minor.
754 The free minor will be automatically computed from the existing
755 devices. A node can be given multiple times in order to allocate
756 multiple minors. The result is the list of minors, in the same
757 order as the passed nodes.
759 @type instance: string
760 @param instance: the instance for which we allocate minors
763 assert isinstance(instance, basestring), \
764 "Invalid argument '%s' passed to AllocateDRBDMinor" % instance
766 d_map, duplicates = self._UnlockedComputeDRBDMap()
768 raise errors.ConfigurationError("Duplicate DRBD ports detected: %s" %
774 # no minors used, we can start at 0
777 self._temporary_drbds[(nname, 0)] = instance
781 ffree = utils.FirstFree(keys)
783 # return the next minor
784 # TODO: implement high-limit check
788 # double-check minor against current instances
789 assert minor not in d_map[nname], \
790 ("Attempt to reuse allocated DRBD minor %d on node %s,"
791 " already allocated to instance %s" %
792 (minor, nname, d_map[nname][minor]))
793 ndata[minor] = instance
794 # double-check minor against reservation
795 r_key = (nname, minor)
796 assert r_key not in self._temporary_drbds, \
797 ("Attempt to reuse reserved DRBD minor %d on node %s,"
798 " reserved for instance %s" %
799 (minor, nname, self._temporary_drbds[r_key]))
800 self._temporary_drbds[r_key] = instance
802 logging.debug("Request to allocate drbd minors, input: %s, returning %s",
806 def _UnlockedReleaseDRBDMinors(self, instance):
807 """Release temporary drbd minors allocated for a given instance.
809 @type instance: string
810 @param instance: the instance for which temporary minors should be
814 assert isinstance(instance, basestring), \
815 "Invalid argument passed to ReleaseDRBDMinors"
816 for key, name in self._temporary_drbds.items():
818 del self._temporary_drbds[key]
820 @locking.ssynchronized(_config_lock)
821 def ReleaseDRBDMinors(self, instance):
822 """Release temporary drbd minors allocated for a given instance.
824 This should be called on the error paths, on the success paths
825 it's automatically called by the ConfigWriter add and update
828 This function is just a wrapper over L{_UnlockedReleaseDRBDMinors}.
830 @type instance: string
831 @param instance: the instance for which temporary minors should be
835 self._UnlockedReleaseDRBDMinors(instance)
837 @locking.ssynchronized(_config_lock, shared=1)
838 def GetConfigVersion(self):
839 """Get the configuration version.
841 @return: Config version
844 return self._config_data.version
846 @locking.ssynchronized(_config_lock, shared=1)
847 def GetClusterName(self):
850 @return: Cluster name
853 return self._config_data.cluster.cluster_name
855 @locking.ssynchronized(_config_lock, shared=1)
856 def GetMasterNode(self):
857 """Get the hostname of the master node for this cluster.
859 @return: Master hostname
862 return self._config_data.cluster.master_node
864 @locking.ssynchronized(_config_lock, shared=1)
865 def GetMasterIP(self):
866 """Get the IP of the master node for this cluster.
871 return self._config_data.cluster.master_ip
873 @locking.ssynchronized(_config_lock, shared=1)
874 def GetMasterNetdev(self):
875 """Get the master network device for this cluster.
878 return self._config_data.cluster.master_netdev
880 @locking.ssynchronized(_config_lock, shared=1)
881 def GetFileStorageDir(self):
882 """Get the file storage dir for this cluster.
885 return self._config_data.cluster.file_storage_dir
887 @locking.ssynchronized(_config_lock, shared=1)
888 def GetSharedFileStorageDir(self):
889 """Get the shared file storage dir for this cluster.
892 return self._config_data.cluster.shared_file_storage_dir
894 @locking.ssynchronized(_config_lock, shared=1)
895 def GetHypervisorType(self):
896 """Get the hypervisor type for this cluster.
899 return self._config_data.cluster.enabled_hypervisors[0]
901 @locking.ssynchronized(_config_lock, shared=1)
902 def GetHostKey(self):
903 """Return the rsa hostkey from the config.
906 @return: the rsa hostkey
909 return self._config_data.cluster.rsahostkeypub
911 @locking.ssynchronized(_config_lock, shared=1)
912 def GetDefaultIAllocator(self):
913 """Get the default instance allocator for this cluster.
916 return self._config_data.cluster.default_iallocator
918 @locking.ssynchronized(_config_lock, shared=1)
919 def GetPrimaryIPFamily(self):
920 """Get cluster primary ip family.
922 @return: primary ip family
925 return self._config_data.cluster.primary_ip_family
927 @locking.ssynchronized(_config_lock)
928 def AddNodeGroup(self, group, ec_id, check_uuid=True):
929 """Add a node group to the configuration.
931 This method calls group.UpgradeConfig() to fill any missing attributes
932 according to their default values.
934 @type group: L{objects.NodeGroup}
935 @param group: the NodeGroup object to add
937 @param ec_id: unique id for the job to use when creating a missing UUID
938 @type check_uuid: bool
939 @param check_uuid: add an UUID to the group if it doesn't have one or, if
940 it does, ensure that it does not exist in the
941 configuration already
944 self._UnlockedAddNodeGroup(group, ec_id, check_uuid)
947 def _UnlockedAddNodeGroup(self, group, ec_id, check_uuid):
948 """Add a node group to the configuration.
951 logging.info("Adding node group %s to configuration", group.name)
953 # Some code might need to add a node group with a pre-populated UUID
954 # generated with ConfigWriter.GenerateUniqueID(). We allow them to bypass
955 # the "does this UUID" exist already check.
957 self._EnsureUUID(group, ec_id)
960 existing_uuid = self._UnlockedLookupNodeGroup(group.name)
961 except errors.OpPrereqError:
964 raise errors.OpPrereqError("Desired group name '%s' already exists as a"
965 " node group (UUID: %s)" %
966 (group.name, existing_uuid),
970 group.ctime = group.mtime = time.time()
971 group.UpgradeConfig()
973 self._config_data.nodegroups[group.uuid] = group
974 self._config_data.cluster.serial_no += 1
976 @locking.ssynchronized(_config_lock)
977 def RemoveNodeGroup(self, group_uuid):
978 """Remove a node group from the configuration.
980 @type group_uuid: string
981 @param group_uuid: the UUID of the node group to remove
984 logging.info("Removing node group %s from configuration", group_uuid)
986 if group_uuid not in self._config_data.nodegroups:
987 raise errors.ConfigurationError("Unknown node group '%s'" % group_uuid)
989 assert len(self._config_data.nodegroups) != 1, \
990 "Group '%s' is the only group, cannot be removed" % group_uuid
992 del self._config_data.nodegroups[group_uuid]
993 self._config_data.cluster.serial_no += 1
996 def _UnlockedLookupNodeGroup(self, target):
997 """Lookup a node group's UUID.
999 @type target: string or None
1000 @param target: group name or UUID or None to look for the default
1002 @return: nodegroup UUID
1003 @raises errors.OpPrereqError: when the target group cannot be found
1007 if len(self._config_data.nodegroups) != 1:
1008 raise errors.OpPrereqError("More than one node group exists. Target"
1009 " group must be specified explicitely.")
1011 return self._config_data.nodegroups.keys()[0]
1012 if target in self._config_data.nodegroups:
1014 for nodegroup in self._config_data.nodegroups.values():
1015 if nodegroup.name == target:
1016 return nodegroup.uuid
1017 raise errors.OpPrereqError("Node group '%s' not found" % target,
1020 @locking.ssynchronized(_config_lock, shared=1)
1021 def LookupNodeGroup(self, target):
1022 """Lookup a node group's UUID.
1024 This function is just a wrapper over L{_UnlockedLookupNodeGroup}.
1026 @type target: string or None
1027 @param target: group name or UUID or None to look for the default
1029 @return: nodegroup UUID
1032 return self._UnlockedLookupNodeGroup(target)
1034 def _UnlockedGetNodeGroup(self, uuid):
1035 """Lookup a node group.
1038 @param uuid: group UUID
1039 @rtype: L{objects.NodeGroup} or None
1040 @return: nodegroup object, or None if not found
1043 if uuid not in self._config_data.nodegroups:
1046 return self._config_data.nodegroups[uuid]
1048 @locking.ssynchronized(_config_lock, shared=1)
1049 def GetNodeGroup(self, uuid):
1050 """Lookup a node group.
1053 @param uuid: group UUID
1054 @rtype: L{objects.NodeGroup} or None
1055 @return: nodegroup object, or None if not found
1058 return self._UnlockedGetNodeGroup(uuid)
1060 @locking.ssynchronized(_config_lock, shared=1)
1061 def GetAllNodeGroupsInfo(self):
1062 """Get the configuration of all node groups.
1065 return dict(self._config_data.nodegroups)
1067 @locking.ssynchronized(_config_lock, shared=1)
1068 def GetNodeGroupList(self):
1069 """Get a list of node groups.
1072 return self._config_data.nodegroups.keys()
1074 @locking.ssynchronized(_config_lock, shared=1)
1075 def GetNodeGroupMembersByNodes(self, nodes):
1076 """Get nodes which are member in the same nodegroups as the given nodes.
1079 ngfn = lambda node_name: self._UnlockedGetNodeInfo(node_name).group
1080 return frozenset(member_name
1081 for node_name in nodes
1083 self._UnlockedGetNodeGroup(ngfn(node_name)).members)
1085 @locking.ssynchronized(_config_lock)
1086 def AddInstance(self, instance, ec_id):
1087 """Add an instance to the config.
1089 This should be used after creating a new instance.
1091 @type instance: L{objects.Instance}
1092 @param instance: the instance object
1095 if not isinstance(instance, objects.Instance):
1096 raise errors.ProgrammerError("Invalid type passed to AddInstance")
1098 if instance.disk_template != constants.DT_DISKLESS:
1099 all_lvs = instance.MapLVsByNode()
1100 logging.info("Instance '%s' DISK_LAYOUT: %s", instance.name, all_lvs)
1102 all_macs = self._AllMACs()
1103 for nic in instance.nics:
1104 if nic.mac in all_macs:
1105 raise errors.ConfigurationError("Cannot add instance %s:"
1106 " MAC address '%s' already in use." %
1107 (instance.name, nic.mac))
1109 self._EnsureUUID(instance, ec_id)
1111 instance.serial_no = 1
1112 instance.ctime = instance.mtime = time.time()
1113 self._config_data.instances[instance.name] = instance
1114 self._config_data.cluster.serial_no += 1
1115 self._UnlockedReleaseDRBDMinors(instance.name)
1118 def _EnsureUUID(self, item, ec_id):
1119 """Ensures a given object has a valid UUID.
1121 @param item: the instance or node to be checked
1122 @param ec_id: the execution context id for the uuid reservation
1126 item.uuid = self._GenerateUniqueID(ec_id)
1127 elif item.uuid in self._AllIDs(include_temporary=True):
1128 raise errors.ConfigurationError("Cannot add '%s': UUID %s already"
1129 " in use" % (item.name, item.uuid))
1131 def _SetInstanceStatus(self, instance_name, status):
1132 """Set the instance's status to a given value.
1135 assert isinstance(status, bool), \
1136 "Invalid status '%s' passed to SetInstanceStatus" % (status,)
1138 if instance_name not in self._config_data.instances:
1139 raise errors.ConfigurationError("Unknown instance '%s'" %
1141 instance = self._config_data.instances[instance_name]
1142 if instance.admin_up != status:
1143 instance.admin_up = status
1144 instance.serial_no += 1
1145 instance.mtime = time.time()
1148 @locking.ssynchronized(_config_lock)
1149 def MarkInstanceUp(self, instance_name):
1150 """Mark the instance status to up in the config.
1153 self._SetInstanceStatus(instance_name, True)
1155 @locking.ssynchronized(_config_lock)
1156 def RemoveInstance(self, instance_name):
1157 """Remove the instance from the configuration.
1160 if instance_name not in self._config_data.instances:
1161 raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
1162 del self._config_data.instances[instance_name]
1163 self._config_data.cluster.serial_no += 1
1166 @locking.ssynchronized(_config_lock)
1167 def RenameInstance(self, old_name, new_name):
1168 """Rename an instance.
1170 This needs to be done in ConfigWriter and not by RemoveInstance
1171 combined with AddInstance as only we can guarantee an atomic
1175 if old_name not in self._config_data.instances:
1176 raise errors.ConfigurationError("Unknown instance '%s'" % old_name)
1177 inst = self._config_data.instances[old_name]
1178 del self._config_data.instances[old_name]
1179 inst.name = new_name
1181 for disk in inst.disks:
1182 if disk.dev_type == constants.LD_FILE:
1183 # rename the file paths in logical and physical id
1184 file_storage_dir = os.path.dirname(os.path.dirname(disk.logical_id[1]))
1185 disk_fname = "disk%s" % disk.iv_name.split("/")[1]
1186 disk.physical_id = disk.logical_id = (disk.logical_id[0],
1187 utils.PathJoin(file_storage_dir,
1191 # Force update of ssconf files
1192 self._config_data.cluster.serial_no += 1
1194 self._config_data.instances[inst.name] = inst
1197 @locking.ssynchronized(_config_lock)
1198 def MarkInstanceDown(self, instance_name):
1199 """Mark the status of an instance to down in the configuration.
1202 self._SetInstanceStatus(instance_name, False)
1204 def _UnlockedGetInstanceList(self):
1205 """Get the list of instances.
1207 This function is for internal use, when the config lock is already held.
1210 return self._config_data.instances.keys()
1212 @locking.ssynchronized(_config_lock, shared=1)
1213 def GetInstanceList(self):
1214 """Get the list of instances.
1216 @return: array of instances, ex. ['instance2.example.com',
1217 'instance1.example.com']
1220 return self._UnlockedGetInstanceList()
1222 def ExpandInstanceName(self, short_name):
1223 """Attempt to expand an incomplete instance name.
1226 # Locking is done in L{ConfigWriter.GetInstanceList}
1227 return _MatchNameComponentIgnoreCase(short_name, self.GetInstanceList())
1229 def _UnlockedGetInstanceInfo(self, instance_name):
1230 """Returns information about an instance.
1232 This function is for internal use, when the config lock is already held.
1235 if instance_name not in self._config_data.instances:
1238 return self._config_data.instances[instance_name]
1240 @locking.ssynchronized(_config_lock, shared=1)
1241 def GetInstanceInfo(self, instance_name):
1242 """Returns information about an instance.
1244 It takes the information from the configuration file. Other information of
1245 an instance are taken from the live systems.
1247 @param instance_name: name of the instance, e.g.
1248 I{instance1.example.com}
1250 @rtype: L{objects.Instance}
1251 @return: the instance object
1254 return self._UnlockedGetInstanceInfo(instance_name)
1256 @locking.ssynchronized(_config_lock, shared=1)
1257 def GetInstanceNodeGroups(self, instance_name, primary_only=False):
1258 """Returns set of node group UUIDs for instance's nodes.
1263 instance = self._UnlockedGetInstanceInfo(instance_name)
1265 raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
1268 nodes = [instance.primary_node]
1270 nodes = instance.all_nodes
1272 return frozenset(self._UnlockedGetNodeInfo(node_name).group
1273 for node_name in nodes)
1275 @locking.ssynchronized(_config_lock, shared=1)
1276 def GetMultiInstanceInfo(self, instances):
1277 """Get the configuration of multiple instances.
1279 @param instances: list of instance names
1281 @return: list of tuples (instance, instance_info), where
1282 instance_info is what would GetInstanceInfo return for the
1283 node, while keeping the original order
1286 return [(name, self._UnlockedGetInstanceInfo(name)) for name in instances]
1288 @locking.ssynchronized(_config_lock, shared=1)
1289 def GetAllInstancesInfo(self):
1290 """Get the configuration of all instances.
1293 @return: dict of (instance, instance_info), where instance_info is what
1294 would GetInstanceInfo return for the node
1297 my_dict = dict([(instance, self._UnlockedGetInstanceInfo(instance))
1298 for instance in self._UnlockedGetInstanceList()])
1301 @locking.ssynchronized(_config_lock)
1302 def AddNode(self, node, ec_id):
1303 """Add a node to the configuration.
1305 @type node: L{objects.Node}
1306 @param node: a Node instance
1309 logging.info("Adding node %s to configuration", node.name)
1311 self._EnsureUUID(node, ec_id)
1314 node.ctime = node.mtime = time.time()
1315 self._UnlockedAddNodeToGroup(node.name, node.group)
1316 self._config_data.nodes[node.name] = node
1317 self._config_data.cluster.serial_no += 1
1320 @locking.ssynchronized(_config_lock)
1321 def RemoveNode(self, node_name):
1322 """Remove a node from the configuration.
1325 logging.info("Removing node %s from configuration", node_name)
1327 if node_name not in self._config_data.nodes:
1328 raise errors.ConfigurationError("Unknown node '%s'" % node_name)
1330 self._UnlockedRemoveNodeFromGroup(self._config_data.nodes[node_name])
1331 del self._config_data.nodes[node_name]
1332 self._config_data.cluster.serial_no += 1
1335 def ExpandNodeName(self, short_name):
1336 """Attempt to expand an incomplete node name.
1339 # Locking is done in L{ConfigWriter.GetNodeList}
1340 return _MatchNameComponentIgnoreCase(short_name, self.GetNodeList())
1342 def _UnlockedGetNodeInfo(self, node_name):
1343 """Get the configuration of a node, as stored in the config.
1345 This function is for internal use, when the config lock is already
1348 @param node_name: the node name, e.g. I{node1.example.com}
1350 @rtype: L{objects.Node}
1351 @return: the node object
1354 if node_name not in self._config_data.nodes:
1357 return self._config_data.nodes[node_name]
1359 @locking.ssynchronized(_config_lock, shared=1)
1360 def GetNodeInfo(self, node_name):
1361 """Get the configuration of a node, as stored in the config.
1363 This is just a locked wrapper over L{_UnlockedGetNodeInfo}.
1365 @param node_name: the node name, e.g. I{node1.example.com}
1367 @rtype: L{objects.Node}
1368 @return: the node object
1371 return self._UnlockedGetNodeInfo(node_name)
1373 @locking.ssynchronized(_config_lock, shared=1)
1374 def GetNodeInstances(self, node_name):
1375 """Get the instances of a node, as stored in the config.
1377 @param node_name: the node name, e.g. I{node1.example.com}
1379 @rtype: (list, list)
1380 @return: a tuple with two lists: the primary and the secondary instances
1385 for inst in self._config_data.instances.values():
1386 if inst.primary_node == node_name:
1387 pri.append(inst.name)
1388 if node_name in inst.secondary_nodes:
1389 sec.append(inst.name)
1392 @locking.ssynchronized(_config_lock, shared=1)
1393 def GetNodeGroupInstances(self, uuid, primary_only=False):
1394 """Get the instances of a node group.
1396 @param uuid: Node group UUID
1397 @param primary_only: Whether to only consider primary nodes
1399 @return: List of instance names in node group
1403 nodes_fn = lambda inst: [inst.primary_node]
1405 nodes_fn = lambda inst: inst.all_nodes
1407 return frozenset(inst.name
1408 for inst in self._config_data.instances.values()
1409 for node_name in nodes_fn(inst)
1410 if self._UnlockedGetNodeInfo(node_name).group == uuid)
1412 def _UnlockedGetNodeList(self):
1413 """Return the list of nodes which are in the configuration.
1415 This function is for internal use, when the config lock is already
1421 return self._config_data.nodes.keys()
1423 @locking.ssynchronized(_config_lock, shared=1)
1424 def GetNodeList(self):
1425 """Return the list of nodes which are in the configuration.
1428 return self._UnlockedGetNodeList()
1430 def _UnlockedGetOnlineNodeList(self):
1431 """Return the list of nodes which are online.
1434 all_nodes = [self._UnlockedGetNodeInfo(node)
1435 for node in self._UnlockedGetNodeList()]
1436 return [node.name for node in all_nodes if not node.offline]
1438 @locking.ssynchronized(_config_lock, shared=1)
1439 def GetOnlineNodeList(self):
1440 """Return the list of nodes which are online.
1443 return self._UnlockedGetOnlineNodeList()
1445 @locking.ssynchronized(_config_lock, shared=1)
1446 def GetVmCapableNodeList(self):
1447 """Return the list of nodes which are not vm capable.
1450 all_nodes = [self._UnlockedGetNodeInfo(node)
1451 for node in self._UnlockedGetNodeList()]
1452 return [node.name for node in all_nodes if node.vm_capable]
1454 @locking.ssynchronized(_config_lock, shared=1)
1455 def GetNonVmCapableNodeList(self):
1456 """Return the list of nodes which are not vm capable.
1459 all_nodes = [self._UnlockedGetNodeInfo(node)
1460 for node in self._UnlockedGetNodeList()]
1461 return [node.name for node in all_nodes if not node.vm_capable]
1463 @locking.ssynchronized(_config_lock, shared=1)
1464 def GetMultiNodeInfo(self, nodes):
1465 """Get the configuration of multiple nodes.
1467 @param nodes: list of node names
1469 @return: list of tuples of (node, node_info), where node_info is
1470 what would GetNodeInfo return for the node, in the original
1474 return [(name, self._UnlockedGetNodeInfo(name)) for name in nodes]
1476 @locking.ssynchronized(_config_lock, shared=1)
1477 def GetAllNodesInfo(self):
1478 """Get the configuration of all nodes.
1481 @return: dict of (node, node_info), where node_info is what
1482 would GetNodeInfo return for the node
1485 my_dict = dict([(node, self._UnlockedGetNodeInfo(node))
1486 for node in self._UnlockedGetNodeList()])
1489 @locking.ssynchronized(_config_lock, shared=1)
1490 def GetNodeGroupsFromNodes(self, nodes):
1491 """Returns groups for a list of nodes.
1493 @type nodes: list of string
1494 @param nodes: List of node names
1498 return frozenset(self._UnlockedGetNodeInfo(name).group for name in nodes)
1500 def _UnlockedGetMasterCandidateStats(self, exceptions=None):
1501 """Get the number of current and maximum desired and possible candidates.
1503 @type exceptions: list
1504 @param exceptions: if passed, list of nodes that should be ignored
1506 @return: tuple of (current, desired and possible, possible)
1509 mc_now = mc_should = mc_max = 0
1510 for node in self._config_data.nodes.values():
1511 if exceptions and node.name in exceptions:
1513 if not (node.offline or node.drained) and node.master_capable:
1515 if node.master_candidate:
1517 mc_should = min(mc_max, self._config_data.cluster.candidate_pool_size)
1518 return (mc_now, mc_should, mc_max)
1520 @locking.ssynchronized(_config_lock, shared=1)
1521 def GetMasterCandidateStats(self, exceptions=None):
1522 """Get the number of current and maximum possible candidates.
1524 This is just a wrapper over L{_UnlockedGetMasterCandidateStats}.
1526 @type exceptions: list
1527 @param exceptions: if passed, list of nodes that should be ignored
1529 @return: tuple of (current, max)
1532 return self._UnlockedGetMasterCandidateStats(exceptions)
1534 @locking.ssynchronized(_config_lock)
1535 def MaintainCandidatePool(self, exceptions):
1536 """Try to grow the candidate pool to the desired size.
1538 @type exceptions: list
1539 @param exceptions: if passed, list of nodes that should be ignored
1541 @return: list with the adjusted nodes (L{objects.Node} instances)
1544 mc_now, mc_max, _ = self._UnlockedGetMasterCandidateStats(exceptions)
1547 node_list = self._config_data.nodes.keys()
1548 random.shuffle(node_list)
1549 for name in node_list:
1550 if mc_now >= mc_max:
1552 node = self._config_data.nodes[name]
1553 if (node.master_candidate or node.offline or node.drained or
1554 node.name in exceptions or not node.master_capable):
1556 mod_list.append(node)
1557 node.master_candidate = True
1560 if mc_now != mc_max:
1561 # this should not happen
1562 logging.warning("Warning: MaintainCandidatePool didn't manage to"
1563 " fill the candidate pool (%d/%d)", mc_now, mc_max)
1565 self._config_data.cluster.serial_no += 1
1570 def _UnlockedAddNodeToGroup(self, node_name, nodegroup_uuid):
1571 """Add a given node to the specified group.
1574 if nodegroup_uuid not in self._config_data.nodegroups:
1575 # This can happen if a node group gets deleted between its lookup and
1576 # when we're adding the first node to it, since we don't keep a lock in
1577 # the meantime. It's ok though, as we'll fail cleanly if the node group
1578 # is not found anymore.
1579 raise errors.OpExecError("Unknown node group: %s" % nodegroup_uuid)
1580 if node_name not in self._config_data.nodegroups[nodegroup_uuid].members:
1581 self._config_data.nodegroups[nodegroup_uuid].members.append(node_name)
1583 def _UnlockedRemoveNodeFromGroup(self, node):
1584 """Remove a given node from its group.
1587 nodegroup = node.group
1588 if nodegroup not in self._config_data.nodegroups:
1589 logging.warning("Warning: node '%s' has unknown node group '%s'"
1590 " (while being removed from it)", node.name, nodegroup)
1591 nodegroup_obj = self._config_data.nodegroups[nodegroup]
1592 if node.name not in nodegroup_obj.members:
1593 logging.warning("Warning: node '%s' not a member of its node group '%s'"
1594 " (while being removed from it)", node.name, nodegroup)
1596 nodegroup_obj.members.remove(node.name)
1598 def _BumpSerialNo(self):
1599 """Bump up the serial number of the config.
1602 self._config_data.serial_no += 1
1603 self._config_data.mtime = time.time()
1605 def _AllUUIDObjects(self):
1606 """Returns all objects with uuid attributes.
1609 return (self._config_data.instances.values() +
1610 self._config_data.nodes.values() +
1611 self._config_data.nodegroups.values() +
1612 [self._config_data.cluster])
1614 def _OpenConfig(self, accept_foreign):
1615 """Read the config data from disk.
1618 raw_data = utils.ReadFile(self._cfg_file)
1621 data = objects.ConfigData.FromDict(serializer.Load(raw_data))
1622 except Exception, err:
1623 raise errors.ConfigurationError(err)
1625 # Make sure the configuration has the right version
1626 _ValidateConfig(data)
1628 if (not hasattr(data, 'cluster') or
1629 not hasattr(data.cluster, 'rsahostkeypub')):
1630 raise errors.ConfigurationError("Incomplete configuration"
1631 " (missing cluster.rsahostkeypub)")
1633 if data.cluster.master_node != self._my_hostname and not accept_foreign:
1634 msg = ("The configuration denotes node %s as master, while my"
1635 " hostname is %s; opening a foreign configuration is only"
1636 " possible in accept_foreign mode" %
1637 (data.cluster.master_node, self._my_hostname))
1638 raise errors.ConfigurationError(msg)
1640 # Upgrade configuration if needed
1641 data.UpgradeConfig()
1643 self._config_data = data
1644 # reset the last serial as -1 so that the next write will cause
1646 self._last_cluster_serial = -1
1648 # And finally run our (custom) config upgrade sequence
1649 self._UpgradeConfig()
1651 self._cfg_id = utils.GetFileID(path=self._cfg_file)
1653 def _UpgradeConfig(self):
1654 """Run upgrade steps that cannot be done purely in the objects.
1656 This is because some data elements need uniqueness across the
1657 whole configuration, etc.
1659 @warning: this function will call L{_WriteConfig()}, but also
1660 L{DropECReservations} so it needs to be called only from a
1661 "safe" place (the constructor). If one wanted to call it with
1662 the lock held, a DropECReservationUnlocked would need to be
1663 created first, to avoid causing deadlock.
1667 for item in self._AllUUIDObjects():
1668 if item.uuid is None:
1669 item.uuid = self._GenerateUniqueID(_UPGRADE_CONFIG_JID)
1671 if not self._config_data.nodegroups:
1672 default_nodegroup_name = constants.INITIAL_NODE_GROUP_NAME
1673 default_nodegroup = objects.NodeGroup(name=default_nodegroup_name,
1675 self._UnlockedAddNodeGroup(default_nodegroup, _UPGRADE_CONFIG_JID, True)
1677 for node in self._config_data.nodes.values():
1679 node.group = self.LookupNodeGroup(None)
1681 # This is technically *not* an upgrade, but needs to be done both when
1682 # nodegroups are being added, and upon normally loading the config,
1683 # because the members list of a node group is discarded upon
1684 # serializing/deserializing the object.
1685 self._UnlockedAddNodeToGroup(node.name, node.group)
1688 # This is ok even if it acquires the internal lock, as _UpgradeConfig is
1689 # only called at config init time, without the lock held
1690 self.DropECReservations(_UPGRADE_CONFIG_JID)
1692 def _DistributeConfig(self, feedback_fn):
1693 """Distribute the configuration to the other nodes.
1695 Currently, this only copies the configuration file. In the future,
1696 it could be used to encapsulate the 2/3-phase update mechanism.
1706 myhostname = self._my_hostname
1707 # we can skip checking whether _UnlockedGetNodeInfo returns None
1708 # since the node list comes from _UnlocketGetNodeList, and we are
1709 # called with the lock held, so no modifications should take place
1711 for node_name in self._UnlockedGetNodeList():
1712 if node_name == myhostname:
1714 node_info = self._UnlockedGetNodeInfo(node_name)
1715 if not node_info.master_candidate:
1717 node_list.append(node_info.name)
1718 addr_list.append(node_info.primary_ip)
1720 result = rpc.RpcRunner.call_upload_file(node_list, self._cfg_file,
1721 address_list=addr_list)
1722 for to_node, to_result in result.items():
1723 msg = to_result.fail_msg
1725 msg = ("Copy of file %s to node %s failed: %s" %
1726 (self._cfg_file, to_node, msg))
1736 def _WriteConfig(self, destination=None, feedback_fn=None):
1737 """Write the configuration data to persistent storage.
1740 assert feedback_fn is None or callable(feedback_fn)
1742 # Warn on config errors, but don't abort the save - the
1743 # configuration has already been modified, and we can't revert;
1744 # the best we can do is to warn the user and save as is, leaving
1745 # recovery to the user
1746 config_errors = self._UnlockedVerifyConfig()
1748 errmsg = ("Configuration data is not consistent: %s" %
1749 (utils.CommaJoin(config_errors)))
1750 logging.critical(errmsg)
1754 if destination is None:
1755 destination = self._cfg_file
1756 self._BumpSerialNo()
1757 txt = serializer.Dump(self._config_data.ToDict())
1759 getents = self._getents()
1761 fd = utils.SafeWriteFile(destination, self._cfg_id, data=txt,
1762 close=False, gid=getents.confd_gid, mode=0640)
1763 except errors.LockError:
1764 raise errors.ConfigurationError("The configuration file has been"
1765 " modified since the last write, cannot"
1768 self._cfg_id = utils.GetFileID(fd=fd)
1772 self.write_count += 1
1774 # and redistribute the config file to master candidates
1775 self._DistributeConfig(feedback_fn)
1777 # Write ssconf files on all nodes (including locally)
1778 if self._last_cluster_serial < self._config_data.cluster.serial_no:
1779 if not self._offline:
1780 result = rpc.RpcRunner.call_write_ssconf_files(
1781 self._UnlockedGetOnlineNodeList(),
1782 self._UnlockedGetSsconfValues())
1784 for nname, nresu in result.items():
1785 msg = nresu.fail_msg
1787 errmsg = ("Error while uploading ssconf files to"
1788 " node %s: %s" % (nname, msg))
1789 logging.warning(errmsg)
1794 self._last_cluster_serial = self._config_data.cluster.serial_no
1796 def _UnlockedGetSsconfValues(self):
1797 """Return the values needed by ssconf.
1800 @return: a dictionary with keys the ssconf names and values their
1805 instance_names = utils.NiceSort(self._UnlockedGetInstanceList())
1806 node_names = utils.NiceSort(self._UnlockedGetNodeList())
1807 node_info = [self._UnlockedGetNodeInfo(name) for name in node_names]
1808 node_pri_ips = ["%s %s" % (ninfo.name, ninfo.primary_ip)
1809 for ninfo in node_info]
1810 node_snd_ips = ["%s %s" % (ninfo.name, ninfo.secondary_ip)
1811 for ninfo in node_info]
1813 instance_data = fn(instance_names)
1814 off_data = fn(node.name for node in node_info if node.offline)
1815 on_data = fn(node.name for node in node_info if not node.offline)
1816 mc_data = fn(node.name for node in node_info if node.master_candidate)
1817 mc_ips_data = fn(node.primary_ip for node in node_info
1818 if node.master_candidate)
1819 node_data = fn(node_names)
1820 node_pri_ips_data = fn(node_pri_ips)
1821 node_snd_ips_data = fn(node_snd_ips)
1823 cluster = self._config_data.cluster
1824 cluster_tags = fn(cluster.GetTags())
1826 hypervisor_list = fn(cluster.enabled_hypervisors)
1828 uid_pool = uidpool.FormatUidPool(cluster.uid_pool, separator="\n")
1830 nodegroups = ["%s %s" % (nodegroup.uuid, nodegroup.name) for nodegroup in
1831 self._config_data.nodegroups.values()]
1832 nodegroups_data = fn(utils.NiceSort(nodegroups))
1835 constants.SS_CLUSTER_NAME: cluster.cluster_name,
1836 constants.SS_CLUSTER_TAGS: cluster_tags,
1837 constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir,
1838 constants.SS_SHARED_FILE_STORAGE_DIR: cluster.shared_file_storage_dir,
1839 constants.SS_MASTER_CANDIDATES: mc_data,
1840 constants.SS_MASTER_CANDIDATES_IPS: mc_ips_data,
1841 constants.SS_MASTER_IP: cluster.master_ip,
1842 constants.SS_MASTER_NETDEV: cluster.master_netdev,
1843 constants.SS_MASTER_NODE: cluster.master_node,
1844 constants.SS_NODE_LIST: node_data,
1845 constants.SS_NODE_PRIMARY_IPS: node_pri_ips_data,
1846 constants.SS_NODE_SECONDARY_IPS: node_snd_ips_data,
1847 constants.SS_OFFLINE_NODES: off_data,
1848 constants.SS_ONLINE_NODES: on_data,
1849 constants.SS_PRIMARY_IP_FAMILY: str(cluster.primary_ip_family),
1850 constants.SS_INSTANCE_LIST: instance_data,
1851 constants.SS_RELEASE_VERSION: constants.RELEASE_VERSION,
1852 constants.SS_HYPERVISOR_LIST: hypervisor_list,
1853 constants.SS_MAINTAIN_NODE_HEALTH: str(cluster.maintain_node_health),
1854 constants.SS_UID_POOL: uid_pool,
1855 constants.SS_NODEGROUPS: nodegroups_data,
1857 bad_values = [(k, v) for k, v in ssconf_values.items()
1858 if not isinstance(v, (str, basestring))]
1860 err = utils.CommaJoin("%s=%s" % (k, v) for k, v in bad_values)
1861 raise errors.ConfigurationError("Some ssconf key(s) have non-string"
1862 " values: %s" % err)
1863 return ssconf_values
1865 @locking.ssynchronized(_config_lock, shared=1)
1866 def GetSsconfValues(self):
1867 """Wrapper using lock around _UnlockedGetSsconf().
1870 return self._UnlockedGetSsconfValues()
1872 @locking.ssynchronized(_config_lock, shared=1)
1873 def GetVGName(self):
1874 """Return the volume group name.
1877 return self._config_data.cluster.volume_group_name
1879 @locking.ssynchronized(_config_lock)
1880 def SetVGName(self, vg_name):
1881 """Set the volume group name.
1884 self._config_data.cluster.volume_group_name = vg_name
1885 self._config_data.cluster.serial_no += 1
1888 @locking.ssynchronized(_config_lock, shared=1)
1889 def GetDRBDHelper(self):
1890 """Return DRBD usermode helper.
1893 return self._config_data.cluster.drbd_usermode_helper
1895 @locking.ssynchronized(_config_lock)
1896 def SetDRBDHelper(self, drbd_helper):
1897 """Set DRBD usermode helper.
1900 self._config_data.cluster.drbd_usermode_helper = drbd_helper
1901 self._config_data.cluster.serial_no += 1
1904 @locking.ssynchronized(_config_lock, shared=1)
1905 def GetMACPrefix(self):
1906 """Return the mac prefix.
1909 return self._config_data.cluster.mac_prefix
1911 @locking.ssynchronized(_config_lock, shared=1)
1912 def GetClusterInfo(self):
1913 """Returns information about the cluster
1915 @rtype: L{objects.Cluster}
1916 @return: the cluster object
1919 return self._config_data.cluster
1921 @locking.ssynchronized(_config_lock, shared=1)
1922 def HasAnyDiskOfType(self, dev_type):
1923 """Check if in there is at disk of the given type in the configuration.
1926 return self._config_data.HasAnyDiskOfType(dev_type)
1928 @locking.ssynchronized(_config_lock)
1929 def Update(self, target, feedback_fn):
1930 """Notify function to be called after updates.
1932 This function must be called when an object (as returned by
1933 GetInstanceInfo, GetNodeInfo, GetCluster) has been updated and the
1934 caller wants the modifications saved to the backing store. Note
1935 that all modified objects will be saved, but the target argument
1936 is the one the caller wants to ensure that it's saved.
1938 @param target: an instance of either L{objects.Cluster},
1939 L{objects.Node} or L{objects.Instance} which is existing in
1941 @param feedback_fn: Callable feedback function
1944 if self._config_data is None:
1945 raise errors.ProgrammerError("Configuration file not read,"
1947 update_serial = False
1948 if isinstance(target, objects.Cluster):
1949 test = target == self._config_data.cluster
1950 elif isinstance(target, objects.Node):
1951 test = target in self._config_data.nodes.values()
1952 update_serial = True
1953 elif isinstance(target, objects.Instance):
1954 test = target in self._config_data.instances.values()
1955 elif isinstance(target, objects.NodeGroup):
1956 test = target in self._config_data.nodegroups.values()
1958 raise errors.ProgrammerError("Invalid object type (%s) passed to"
1959 " ConfigWriter.Update" % type(target))
1961 raise errors.ConfigurationError("Configuration updated since object"
1962 " has been read or unknown object")
1963 target.serial_no += 1
1964 target.mtime = now = time.time()
1967 # for node updates, we need to increase the cluster serial too
1968 self._config_data.cluster.serial_no += 1
1969 self._config_data.cluster.mtime = now
1971 if isinstance(target, objects.Instance):
1972 self._UnlockedReleaseDRBDMinors(target.name)
1974 self._WriteConfig(feedback_fn=feedback_fn)
1976 @locking.ssynchronized(_config_lock)
1977 def DropECReservations(self, ec_id):
1978 """Drop per-execution-context reservations
1981 for rm in self._all_rms:
1982 rm.DropECReservations(ec_id)