X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/c41eea6ebec742e9579153666e51fcc5769c2752..2492231f7381ce33f2164322f799c84b2d7cceef:/lib/config.py diff --git a/lib/config.py b/lib/config.py index b1661fd..7d81b37 100644 --- a/lib/config.py +++ b/lib/config.py @@ -1,7 +1,7 @@ # # -# Copyright (C) 2006, 2007 Google Inc. +# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 Google Inc. # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by @@ -31,10 +31,13 @@ much memory. """ +# pylint: disable-msg=R0904 +# R0904: Too many public methods + import os -import tempfile import random import logging +import time from ganeti import errors from ganeti import locking @@ -43,9 +46,15 @@ from ganeti import constants from ganeti import rpc from ganeti import objects from ganeti import serializer +from ganeti import uidpool +from ganeti import netutils +from ganeti import runtime + +_config_lock = locking.SharedLock("ConfigWriter") -_config_lock = locking.SharedLock() +# job id used for resource management at config upgrade time +_UPGRADE_CONFIG_JID = "jid-cfg-upgrade" def _ValidateConfig(data): @@ -58,17 +67,80 @@ def _ValidateConfig(data): """ if data.version != constants.CONFIG_VERSION: - raise errors.ConfigurationError("Cluster configuration version" - " mismatch, got %s instead of %s" % - (data.version, - constants.CONFIG_VERSION)) + raise errors.ConfigVersionMismatch(constants.CONFIG_VERSION, data.version) + + +class TemporaryReservationManager: + """A temporary resource reservation manager. + + This is used to reserve resources in a job, before using them, making sure + other jobs cannot get them in the meantime. + + """ + def __init__(self): + self._ec_reserved = {} + + def Reserved(self, resource): + for holder_reserved in self._ec_reserved.values(): + if resource in holder_reserved: + return True + return False + + def Reserve(self, ec_id, resource): + if self.Reserved(resource): + raise errors.ReservationError("Duplicate reservation for resource '%s'" + % str(resource)) + if ec_id not in self._ec_reserved: + self._ec_reserved[ec_id] = set([resource]) + else: + self._ec_reserved[ec_id].add(resource) + + def DropECReservations(self, ec_id): + if ec_id in self._ec_reserved: + del self._ec_reserved[ec_id] + + def GetReserved(self): + all_reserved = set() + for holder_reserved in self._ec_reserved.values(): + all_reserved.update(holder_reserved) + return all_reserved + + def Generate(self, existing, generate_one_fn, ec_id): + """Generate a new resource of this type + + """ + assert callable(generate_one_fn) + + all_elems = self.GetReserved() + all_elems.update(existing) + retries = 64 + while retries > 0: + new_resource = generate_one_fn() + if new_resource is not None and new_resource not in all_elems: + break + else: + raise errors.ConfigurationError("Not able generate new resource" + " (last tried: %s)" % new_resource) + self.Reserve(ec_id, new_resource) + return new_resource + + +def _MatchNameComponentIgnoreCase(short_name, names): + """Wrapper around L{utils.text.MatchNameComponent}. + + """ + return utils.MatchNameComponent(short_name, names, case_sensitive=False) class ConfigWriter: """The interface to the cluster configuration. + @ivar _temporary_lvs: reservation manager for temporary LVs + @ivar _all_rms: a list of all temporary reservation managers + """ - def __init__(self, cfg_file=None, offline=False): + def __init__(self, cfg_file=None, offline=False, _getents=runtime.GetEnts, + accept_foreign=False): self.write_count = 0 self._lock = _config_lock self._config_data = None @@ -77,14 +149,22 @@ class ConfigWriter: self._cfg_file = constants.CLUSTER_CONF_FILE else: self._cfg_file = cfg_file - self._temporary_ids = set() + self._getents = _getents + self._temporary_ids = TemporaryReservationManager() self._temporary_drbds = {} + self._temporary_macs = TemporaryReservationManager() + self._temporary_secrets = TemporaryReservationManager() + self._temporary_lvs = TemporaryReservationManager() + self._all_rms = [self._temporary_ids, self._temporary_macs, + self._temporary_secrets, self._temporary_lvs] # Note: in order to prevent errors when resolving our name in # _DistributeConfig, we compute it here once and reuse it; it's # better to raise an error before starting to modify the config # file than after it was modified - self._my_hostname = utils.HostInfo().name - self._OpenConfig() + self._my_hostname = netutils.Hostname.GetSysName() + self._last_cluster_serial = -1 + self._cfg_id = None + self._OpenConfig(accept_foreign) # this method needs to be static, so that we can call it on the class @staticmethod @@ -94,58 +174,79 @@ class ConfigWriter: """ return os.path.exists(constants.CLUSTER_CONF_FILE) + def _GenerateOneMAC(self): + """Generate one mac address + + """ + prefix = self._config_data.cluster.mac_prefix + byte1 = random.randrange(0, 256) + byte2 = random.randrange(0, 256) + byte3 = random.randrange(0, 256) + mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3) + return mac + + @locking.ssynchronized(_config_lock, shared=1) + def GetNdParams(self, node): + """Get the node params populated with cluster defaults. + + @type node: L{object.Node} + @param node: The node we want to know the params for + @return: A dict with the filled in node params + + """ + nodegroup = self._UnlockedGetNodeGroup(node.group) + return self._config_data.cluster.FillND(node, nodegroup) + @locking.ssynchronized(_config_lock, shared=1) - def GenerateMAC(self): + def GenerateMAC(self, ec_id): """Generate a MAC for an instance. This should check the current instances for duplicates. """ - prefix = self._config_data.cluster.mac_prefix - all_macs = self._AllMACs() - retries = 64 - while retries > 0: - byte1 = random.randrange(0, 256) - byte2 = random.randrange(0, 256) - byte3 = random.randrange(0, 256) - mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3) - if mac not in all_macs: - break - retries -= 1 - else: - raise errors.ConfigurationError("Can't generate unique MAC") - return mac + existing = self._AllMACs() + return self._temporary_ids.Generate(existing, self._GenerateOneMAC, ec_id) @locking.ssynchronized(_config_lock, shared=1) - def IsMacInUse(self, mac): - """Predicate: check if the specified MAC is in use in the Ganeti cluster. + def ReserveMAC(self, mac, ec_id): + """Reserve a MAC for an instance. This only checks instances managed by this cluster, it does not check for potential collisions elsewhere. """ all_macs = self._AllMACs() - return mac in all_macs + if mac in all_macs: + raise errors.ReservationError("mac already in use") + else: + self._temporary_macs.Reserve(ec_id, mac) @locking.ssynchronized(_config_lock, shared=1) - def GenerateDRBDSecret(self): + def ReserveLV(self, lv_name, ec_id): + """Reserve an VG/LV pair for an instance. + + @type lv_name: string + @param lv_name: the logical volume name to reserve + + """ + all_lvs = self._AllLVs() + if lv_name in all_lvs: + raise errors.ReservationError("LV already in use") + else: + self._temporary_lvs.Reserve(ec_id, lv_name) + + @locking.ssynchronized(_config_lock, shared=1) + def GenerateDRBDSecret(self, ec_id): """Generate a DRBD secret. This checks the current disks for duplicates. """ - all_secrets = self._AllDRBDSecrets() - retries = 64 - while retries > 0: - secret = utils.GenerateSecret() - if secret not in all_secrets: - break - retries -= 1 - else: - raise errors.ConfigurationError("Can't generate unique DRBD secret") - return secret + return self._temporary_secrets.Generate(self._AllDRBDSecrets(), + utils.GenerateSecret, + ec_id) - def _ComputeAllLVs(self): + def _AllLVs(self): """Compute the list of all LVs. """ @@ -156,39 +257,48 @@ class ConfigWriter: lvnames.update(lv_list) return lvnames - @locking.ssynchronized(_config_lock, shared=1) - def GenerateUniqueID(self, exceptions=None): - """Generate an unique disk name. + def _AllIDs(self, include_temporary): + """Compute the list of all UUIDs and names we have. + + @type include_temporary: boolean + @param include_temporary: whether to include the _temporary_ids set + @rtype: set + @return: a set of IDs + + """ + existing = set() + if include_temporary: + existing.update(self._temporary_ids.GetReserved()) + existing.update(self._AllLVs()) + existing.update(self._config_data.instances.keys()) + existing.update(self._config_data.nodes.keys()) + existing.update([i.uuid for i in self._AllUUIDObjects() if i.uuid]) + return existing + + def _GenerateUniqueID(self, ec_id): + """Generate an unique UUID. This checks the current node, instances and disk names for duplicates. - @param exceptions: a list with some other names which should be checked - for uniqueness (used for example when you want to get - more than one id at one time without adding each one in - turn to the config file) - @rtype: string @return: the unique id """ - existing = set() - existing.update(self._temporary_ids) - existing.update(self._ComputeAllLVs()) - existing.update(self._config_data.instances.keys()) - existing.update(self._config_data.nodes.keys()) - if exceptions is not None: - existing.update(exceptions) - retries = 64 - while retries > 0: - unique_id = utils.NewUUID() - if unique_id not in existing and unique_id is not None: - break - else: - raise errors.ConfigurationError("Not able generate an unique ID" - " (last tried ID: %s" % unique_id) - self._temporary_ids.add(unique_id) - return unique_id + existing = self._AllIDs(include_temporary=False) + return self._temporary_ids.Generate(existing, utils.NewUUID, ec_id) + + @locking.ssynchronized(_config_lock, shared=1) + def GenerateUniqueID(self, ec_id): + """Generate an unique ID. + + This is just a wrapper over the unlocked version. + + @type ec_id: string + @param ec_id: unique id for the job to reserve the id to + + """ + return self._GenerateUniqueID(ec_id) def _AllMACs(self): """Return all MACs present in the config. @@ -226,17 +336,97 @@ class ConfigWriter: return result - @locking.ssynchronized(_config_lock, shared=1) - def VerifyConfig(self): + def _CheckDiskIDs(self, disk, l_ids, p_ids): + """Compute duplicate disk IDs + + @type disk: L{objects.Disk} + @param disk: the disk at which to start searching + @type l_ids: list + @param l_ids: list of current logical ids + @type p_ids: list + @param p_ids: list of current physical ids + @rtype: list + @return: a list of error messages + + """ + result = [] + if disk.logical_id is not None: + if disk.logical_id in l_ids: + result.append("duplicate logical id %s" % str(disk.logical_id)) + else: + l_ids.append(disk.logical_id) + if disk.physical_id is not None: + if disk.physical_id in p_ids: + result.append("duplicate physical id %s" % str(disk.physical_id)) + else: + p_ids.append(disk.physical_id) + + if disk.children: + for child in disk.children: + result.extend(self._CheckDiskIDs(child, l_ids, p_ids)) + return result + + def _UnlockedVerifyConfig(self): """Verify function. + @rtype: list + @return: a list of error messages; a non-empty list signifies + configuration errors + """ + # pylint: disable-msg=R0914 result = [] seen_macs = [] ports = {} data = self._config_data + cluster = data.cluster + seen_lids = [] + seen_pids = [] + + # global cluster checks + if not cluster.enabled_hypervisors: + result.append("enabled hypervisors list doesn't have any entries") + invalid_hvs = set(cluster.enabled_hypervisors) - constants.HYPER_TYPES + if invalid_hvs: + result.append("enabled hypervisors contains invalid entries: %s" % + invalid_hvs) + missing_hvp = (set(cluster.enabled_hypervisors) - + set(cluster.hvparams.keys())) + if missing_hvp: + result.append("hypervisor parameters missing for the enabled" + " hypervisor(s) %s" % utils.CommaJoin(missing_hvp)) + + if cluster.master_node not in data.nodes: + result.append("cluster has invalid primary node '%s'" % + cluster.master_node) + + def _helper(owner, attr, value, template): + try: + utils.ForceDictType(value, template) + except errors.GenericError, err: + result.append("%s has invalid %s: %s" % (owner, attr, err)) + + def _helper_nic(owner, params): + try: + objects.NIC.CheckParameterSyntax(params) + except errors.ConfigurationError, err: + result.append("%s has invalid nicparams: %s" % (owner, err)) + + # check cluster parameters + _helper("cluster", "beparams", cluster.SimpleFillBE({}), + constants.BES_PARAMETER_TYPES) + _helper("cluster", "nicparams", cluster.SimpleFillNIC({}), + constants.NICS_PARAMETER_TYPES) + _helper_nic("cluster", cluster.SimpleFillNIC({})) + _helper("cluster", "ndparams", cluster.SimpleFillND({}), + constants.NDS_PARAMETER_TYPES) + + # per-instance checks for instance_name in data.instances: instance = data.instances[instance_name] + if instance.name != instance_name: + result.append("instance '%s' is indexed by wrong name '%s'" % + (instance.name, instance_name)) if instance.primary_node not in data.nodes: result.append("instance '%s' has invalid primary node '%s'" % (instance_name, instance.primary_node)) @@ -250,6 +440,17 @@ class ConfigWriter: (instance_name, idx, nic.mac)) else: seen_macs.append(nic.mac) + if nic.nicparams: + filled = cluster.SimpleFillNIC(nic.nicparams) + owner = "instance %s nic %d" % (instance.name, idx) + _helper(owner, "nicparams", + filled, constants.NICS_PARAMETER_TYPES) + _helper_nic(owner, filled) + + # parameter checks + if instance.beparams: + _helper("instance %s" % instance.name, "beparams", + cluster.FillBE(instance), constants.BES_PARAMETER_TYPES) # gather the drbd ports for duplicate checks for dsk in instance.disks: @@ -265,8 +466,14 @@ class ConfigWriter: ports[net_port] = [] ports[net_port].append((instance.name, "network port")) + # instance disk verify + for idx, disk in enumerate(instance.disks): + result.extend(["instance '%s' disk %d error: %s" % + (instance.name, idx, msg) for msg in disk.Verify()]) + result.extend(self._CheckDiskIDs(disk, seen_lids, seen_pids)) + # cluster-wide pool of free ports - for free_port in data.cluster.tcpudp_port_pool: + for free_port in cluster.tcpudp_port_pool: if free_port not in ports: ports[free_port] = [] ports[free_port].append(("cluster", "port marked as free")) @@ -277,25 +484,121 @@ class ConfigWriter: for pnum in keys: pdata = ports[pnum] if len(pdata) > 1: - txt = ", ".join(["%s/%s" % val for val in pdata]) + txt = utils.CommaJoin(["%s/%s" % val for val in pdata]) result.append("tcp/udp port %s has duplicates: %s" % (pnum, txt)) # highest used tcp port check if keys: - if keys[-1] > data.cluster.highest_used_port: + if keys[-1] > cluster.highest_used_port: result.append("Highest used port mismatch, saved %s, computed %s" % - (data.cluster.highest_used_port, keys[-1])) + (cluster.highest_used_port, keys[-1])) - if not data.nodes[data.cluster.master_node].master_candidate: + if not data.nodes[cluster.master_node].master_candidate: result.append("Master node is not a master candidate") - mc_now, mc_max = self._UnlockedGetMasterCandidateStats() + # master candidate checks + mc_now, mc_max, _ = self._UnlockedGetMasterCandidateStats() if mc_now < mc_max: result.append("Not enough master candidates: actual %d, target %d" % (mc_now, mc_max)) + # node checks + for node_name, node in data.nodes.items(): + if node.name != node_name: + result.append("Node '%s' is indexed by wrong name '%s'" % + (node.name, node_name)) + if [node.master_candidate, node.drained, node.offline].count(True) > 1: + result.append("Node %s state is invalid: master_candidate=%s," + " drain=%s, offline=%s" % + (node.name, node.master_candidate, node.drained, + node.offline)) + if node.group not in data.nodegroups: + result.append("Node '%s' has invalid group '%s'" % + (node.name, node.group)) + else: + _helper("node %s" % node.name, "ndparams", + cluster.FillND(node, data.nodegroups[node.group]), + constants.NDS_PARAMETER_TYPES) + + # nodegroups checks + nodegroups_names = set() + for nodegroup_uuid in data.nodegroups: + nodegroup = data.nodegroups[nodegroup_uuid] + if nodegroup.uuid != nodegroup_uuid: + result.append("node group '%s' (uuid: '%s') indexed by wrong uuid '%s'" + % (nodegroup.name, nodegroup.uuid, nodegroup_uuid)) + if utils.UUID_RE.match(nodegroup.name.lower()): + result.append("node group '%s' (uuid: '%s') has uuid-like name" % + (nodegroup.name, nodegroup.uuid)) + if nodegroup.name in nodegroups_names: + result.append("duplicate node group name '%s'" % nodegroup.name) + else: + nodegroups_names.add(nodegroup.name) + if nodegroup.ndparams: + _helper("group %s" % nodegroup.name, "ndparams", + cluster.SimpleFillND(nodegroup.ndparams), + constants.NDS_PARAMETER_TYPES) + + + # drbd minors check + _, duplicates = self._UnlockedComputeDRBDMap() + for node, minor, instance_a, instance_b in duplicates: + result.append("DRBD minor %d on node %s is assigned twice to instances" + " %s and %s" % (minor, node, instance_a, instance_b)) + + # IP checks + default_nicparams = cluster.nicparams[constants.PP_DEFAULT] + ips = {} + + def _AddIpAddress(ip, name): + ips.setdefault(ip, []).append(name) + + _AddIpAddress(cluster.master_ip, "cluster_ip") + + for node in data.nodes.values(): + _AddIpAddress(node.primary_ip, "node:%s/primary" % node.name) + if node.secondary_ip != node.primary_ip: + _AddIpAddress(node.secondary_ip, "node:%s/secondary" % node.name) + + for instance in data.instances.values(): + for idx, nic in enumerate(instance.nics): + if nic.ip is None: + continue + + nicparams = objects.FillDict(default_nicparams, nic.nicparams) + nic_mode = nicparams[constants.NIC_MODE] + nic_link = nicparams[constants.NIC_LINK] + + if nic_mode == constants.NIC_MODE_BRIDGED: + link = "bridge:%s" % nic_link + elif nic_mode == constants.NIC_MODE_ROUTED: + link = "route:%s" % nic_link + else: + raise errors.ProgrammerError("NIC mode '%s' not handled" % nic_mode) + + _AddIpAddress("%s/%s" % (link, nic.ip), + "instance:%s/nic:%d" % (instance.name, idx)) + + for ip, owners in ips.items(): + if len(owners) > 1: + result.append("IP address %s is used by multiple owners: %s" % + (ip, utils.CommaJoin(owners))) + return result + @locking.ssynchronized(_config_lock, shared=1) + def VerifyConfig(self): + """Verify function. + + This is just a wrapper over L{_UnlockedVerifyConfig}. + + @rtype: list + @return: a list of error messages; a non-empty list signifies + configuration errors + + """ + return self._UnlockedVerifyConfig() + def _UnlockedSetDiskID(self, disk, node_name): """Convert the unique ID to the ID needed on the target nodes. @@ -388,37 +691,61 @@ class ConfigWriter: self._WriteConfig() return port - def _ComputeDRBDMap(self, instance): + def _UnlockedComputeDRBDMap(self): """Compute the used DRBD minor/nodes. + @rtype: (dict, list) @return: dictionary of node_name: dict of minor: instance_name; the returned dict will have all the nodes in it (even if with - an empty list). + an empty list), and a list of duplicates; if the duplicates + list is not empty, the configuration is corrupted and its caller + should raise an exception """ def _AppendUsedPorts(instance_name, disk, used): + duplicates = [] if disk.dev_type == constants.LD_DRBD8 and len(disk.logical_id) >= 5: - nodeA, nodeB, dummy, minorA, minorB = disk.logical_id[:5] - for node, port in ((nodeA, minorA), (nodeB, minorB)): - assert node in used, "Instance node not found in node list" + node_a, node_b, _, minor_a, minor_b = disk.logical_id[:5] + for node, port in ((node_a, minor_a), (node_b, minor_b)): + assert node in used, ("Node '%s' of instance '%s' not found" + " in node list" % (node, instance_name)) if port in used[node]: - raise errors.ProgrammerError("DRBD minor already used:" - " %s/%s, %s/%s" % - (node, port, instance_name, - used[node][port])) - - used[node][port] = instance_name + duplicates.append((node, port, instance_name, used[node][port])) + else: + used[node][port] = instance_name if disk.children: for child in disk.children: - _AppendUsedPorts(instance_name, child, used) + duplicates.extend(_AppendUsedPorts(instance_name, child, used)) + return duplicates + duplicates = [] my_dict = dict((node, {}) for node in self._config_data.nodes) - for (node, minor), instance in self._temporary_drbds.iteritems(): - my_dict[node][minor] = instance for instance in self._config_data.instances.itervalues(): for disk in instance.disks: - _AppendUsedPorts(instance.name, disk, my_dict) - return my_dict + duplicates.extend(_AppendUsedPorts(instance.name, disk, my_dict)) + for (node, minor), instance in self._temporary_drbds.iteritems(): + if minor in my_dict[node] and my_dict[node][minor] != instance: + duplicates.append((node, minor, instance, my_dict[node][minor])) + else: + my_dict[node][minor] = instance + return my_dict, duplicates + + @locking.ssynchronized(_config_lock) + def ComputeDRBDMap(self): + """Compute the used DRBD minor/nodes. + + This is just a wrapper over L{_UnlockedComputeDRBDMap}. + + @return: dictionary of node_name: dict of minor: instance_name; + the returned dict will have all the nodes in it (even if with + an empty list). + + """ + d_map, duplicates = self._UnlockedComputeDRBDMap() + if duplicates: + raise errors.ConfigurationError("Duplicate DRBD ports detected: %s" % + str(duplicates)) + return d_map @locking.ssynchronized(_config_lock) def AllocateDRBDMinor(self, nodes, instance): @@ -429,8 +756,17 @@ class ConfigWriter: multiple minors. The result is the list of minors, in the same order as the passed nodes. + @type instance: string + @param instance: the instance for which we allocate minors + """ - d_map = self._ComputeDRBDMap(instance) + assert isinstance(instance, basestring), \ + "Invalid argument '%s' passed to AllocateDRBDMinor" % instance + + d_map, duplicates = self._UnlockedComputeDRBDMap() + if duplicates: + raise errors.ConfigurationError("Duplicate DRBD ports detected: %s" % + str(duplicates)) result = [] for nname in nodes: ndata = d_map[nname] @@ -449,31 +785,55 @@ class ConfigWriter: minor = keys[-1] + 1 else: minor = ffree - result.append(minor) + # double-check minor against current instances + assert minor not in d_map[nname], \ + ("Attempt to reuse allocated DRBD minor %d on node %s," + " already allocated to instance %s" % + (minor, nname, d_map[nname][minor])) ndata[minor] = instance - assert (nname, minor) not in self._temporary_drbds, \ - "Attempt to reuse reserved DRBD minor" - self._temporary_drbds[(nname, minor)] = instance + # double-check minor against reservation + r_key = (nname, minor) + assert r_key not in self._temporary_drbds, \ + ("Attempt to reuse reserved DRBD minor %d on node %s," + " reserved for instance %s" % + (minor, nname, self._temporary_drbds[r_key])) + self._temporary_drbds[r_key] = instance + result.append(minor) logging.debug("Request to allocate drbd minors, input: %s, returning %s", nodes, result) return result - @locking.ssynchronized(_config_lock) - def ReleaseDRBDMinors(self, instance): + def _UnlockedReleaseDRBDMinors(self, instance): """Release temporary drbd minors allocated for a given instance. - This should be called on both the error paths and on the success - paths (after the instance has been added or updated). - @type instance: string @param instance: the instance for which temporary minors should be released """ + assert isinstance(instance, basestring), \ + "Invalid argument passed to ReleaseDRBDMinors" for key, name in self._temporary_drbds.items(): if name == instance: del self._temporary_drbds[key] + @locking.ssynchronized(_config_lock) + def ReleaseDRBDMinors(self, instance): + """Release temporary drbd minors allocated for a given instance. + + This should be called on the error paths, on the success paths + it's automatically called by the ConfigWriter add and update + functions. + + This function is just a wrapper over L{_UnlockedReleaseDRBDMinors}. + + @type instance: string + @param instance: the instance for which temporary minors should be + released + + """ + self._UnlockedReleaseDRBDMinors(instance) + @locking.ssynchronized(_config_lock, shared=1) def GetConfigVersion(self): """Get the configuration version. @@ -525,11 +885,18 @@ class ConfigWriter: return self._config_data.cluster.file_storage_dir @locking.ssynchronized(_config_lock, shared=1) + def GetSharedFileStorageDir(self): + """Get the shared file storage dir for this cluster. + + """ + return self._config_data.cluster.shared_file_storage_dir + + @locking.ssynchronized(_config_lock, shared=1) def GetHypervisorType(self): """Get the hypervisor type for this cluster. """ - return self._config_data.cluster.default_hypervisor + return self._config_data.cluster.enabled_hypervisors[0] @locking.ssynchronized(_config_lock, shared=1) def GetHostKey(self): @@ -541,8 +908,182 @@ class ConfigWriter: """ return self._config_data.cluster.rsahostkeypub + @locking.ssynchronized(_config_lock, shared=1) + def GetDefaultIAllocator(self): + """Get the default instance allocator for this cluster. + + """ + return self._config_data.cluster.default_iallocator + + @locking.ssynchronized(_config_lock, shared=1) + def GetPrimaryIPFamily(self): + """Get cluster primary ip family. + + @return: primary ip family + + """ + return self._config_data.cluster.primary_ip_family + + @locking.ssynchronized(_config_lock) + def AddNodeGroup(self, group, ec_id, check_uuid=True): + """Add a node group to the configuration. + + This method calls group.UpgradeConfig() to fill any missing attributes + according to their default values. + + @type group: L{objects.NodeGroup} + @param group: the NodeGroup object to add + @type ec_id: string + @param ec_id: unique id for the job to use when creating a missing UUID + @type check_uuid: bool + @param check_uuid: add an UUID to the group if it doesn't have one or, if + it does, ensure that it does not exist in the + configuration already + + """ + self._UnlockedAddNodeGroup(group, ec_id, check_uuid) + self._WriteConfig() + + def _UnlockedAddNodeGroup(self, group, ec_id, check_uuid): + """Add a node group to the configuration. + + """ + logging.info("Adding node group %s to configuration", group.name) + + # Some code might need to add a node group with a pre-populated UUID + # generated with ConfigWriter.GenerateUniqueID(). We allow them to bypass + # the "does this UUID" exist already check. + if check_uuid: + self._EnsureUUID(group, ec_id) + + try: + existing_uuid = self._UnlockedLookupNodeGroup(group.name) + except errors.OpPrereqError: + pass + else: + raise errors.OpPrereqError("Desired group name '%s' already exists as a" + " node group (UUID: %s)" % + (group.name, existing_uuid), + errors.ECODE_EXISTS) + + group.serial_no = 1 + group.ctime = group.mtime = time.time() + group.UpgradeConfig() + + self._config_data.nodegroups[group.uuid] = group + self._config_data.cluster.serial_no += 1 + @locking.ssynchronized(_config_lock) - def AddInstance(self, instance): + def RemoveNodeGroup(self, group_uuid): + """Remove a node group from the configuration. + + @type group_uuid: string + @param group_uuid: the UUID of the node group to remove + + """ + logging.info("Removing node group %s from configuration", group_uuid) + + if group_uuid not in self._config_data.nodegroups: + raise errors.ConfigurationError("Unknown node group '%s'" % group_uuid) + + assert len(self._config_data.nodegroups) != 1, \ + "Group '%s' is the only group, cannot be removed" % group_uuid + + del self._config_data.nodegroups[group_uuid] + self._config_data.cluster.serial_no += 1 + self._WriteConfig() + + def _UnlockedLookupNodeGroup(self, target): + """Lookup a node group's UUID. + + @type target: string or None + @param target: group name or UUID or None to look for the default + @rtype: string + @return: nodegroup UUID + @raises errors.OpPrereqError: when the target group cannot be found + + """ + if target is None: + if len(self._config_data.nodegroups) != 1: + raise errors.OpPrereqError("More than one node group exists. Target" + " group must be specified explicitely.") + else: + return self._config_data.nodegroups.keys()[0] + if target in self._config_data.nodegroups: + return target + for nodegroup in self._config_data.nodegroups.values(): + if nodegroup.name == target: + return nodegroup.uuid + raise errors.OpPrereqError("Node group '%s' not found" % target, + errors.ECODE_NOENT) + + @locking.ssynchronized(_config_lock, shared=1) + def LookupNodeGroup(self, target): + """Lookup a node group's UUID. + + This function is just a wrapper over L{_UnlockedLookupNodeGroup}. + + @type target: string or None + @param target: group name or UUID or None to look for the default + @rtype: string + @return: nodegroup UUID + + """ + return self._UnlockedLookupNodeGroup(target) + + def _UnlockedGetNodeGroup(self, uuid): + """Lookup a node group. + + @type uuid: string + @param uuid: group UUID + @rtype: L{objects.NodeGroup} or None + @return: nodegroup object, or None if not found + + """ + if uuid not in self._config_data.nodegroups: + return None + + return self._config_data.nodegroups[uuid] + + @locking.ssynchronized(_config_lock, shared=1) + def GetNodeGroup(self, uuid): + """Lookup a node group. + + @type uuid: string + @param uuid: group UUID + @rtype: L{objects.NodeGroup} or None + @return: nodegroup object, or None if not found + + """ + return self._UnlockedGetNodeGroup(uuid) + + @locking.ssynchronized(_config_lock, shared=1) + def GetAllNodeGroupsInfo(self): + """Get the configuration of all node groups. + + """ + return dict(self._config_data.nodegroups) + + @locking.ssynchronized(_config_lock, shared=1) + def GetNodeGroupList(self): + """Get a list of node groups. + + """ + return self._config_data.nodegroups.keys() + + @locking.ssynchronized(_config_lock, shared=1) + def GetNodeGroupMembersByNodes(self, nodes): + """Get nodes which are member in the same nodegroups as the given nodes. + + """ + ngfn = lambda node_name: self._UnlockedGetNodeInfo(node_name).group + return frozenset(member_name + for node_name in nodes + for member_name in + self._UnlockedGetNodeGroup(ngfn(node_name)).members) + + @locking.ssynchronized(_config_lock) + def AddInstance(self, instance, ec_id): """Add an instance to the config. This should be used after creating a new instance. @@ -558,26 +1099,50 @@ class ConfigWriter: all_lvs = instance.MapLVsByNode() logging.info("Instance '%s' DISK_LAYOUT: %s", instance.name, all_lvs) + all_macs = self._AllMACs() + for nic in instance.nics: + if nic.mac in all_macs: + raise errors.ConfigurationError("Cannot add instance %s:" + " MAC address '%s' already in use." % + (instance.name, nic.mac)) + + self._EnsureUUID(instance, ec_id) + instance.serial_no = 1 + instance.ctime = instance.mtime = time.time() self._config_data.instances[instance.name] = instance + self._config_data.cluster.serial_no += 1 + self._UnlockedReleaseDRBDMinors(instance.name) self._WriteConfig() + def _EnsureUUID(self, item, ec_id): + """Ensures a given object has a valid UUID. + + @param item: the instance or node to be checked + @param ec_id: the execution context id for the uuid reservation + + """ + if not item.uuid: + item.uuid = self._GenerateUniqueID(ec_id) + elif item.uuid in self._AllIDs(include_temporary=True): + raise errors.ConfigurationError("Cannot add '%s': UUID %s already" + " in use" % (item.name, item.uuid)) + def _SetInstanceStatus(self, instance_name, status): """Set the instance's status to a given value. """ - if status not in ("up", "down"): - raise errors.ProgrammerError("Invalid status '%s' passed to" - " ConfigWriter._SetInstanceStatus()" % - status) + assert isinstance(status, bool), \ + "Invalid status '%s' passed to SetInstanceStatus" % (status,) if instance_name not in self._config_data.instances: raise errors.ConfigurationError("Unknown instance '%s'" % instance_name) instance = self._config_data.instances[instance_name] - if instance.status != status: - instance.status = status + if instance.admin_up != status: + instance.admin_up = status instance.serial_no += 1 + instance.mtime = time.time() self._WriteConfig() @locking.ssynchronized(_config_lock) @@ -585,7 +1150,7 @@ class ConfigWriter: """Mark the instance status to up in the config. """ - self._SetInstanceStatus(instance_name, "up") + self._SetInstanceStatus(instance_name, True) @locking.ssynchronized(_config_lock) def RemoveInstance(self, instance_name): @@ -595,6 +1160,7 @@ class ConfigWriter: if instance_name not in self._config_data.instances: raise errors.ConfigurationError("Unknown instance '%s'" % instance_name) del self._config_data.instances[instance_name] + self._config_data.cluster.serial_no += 1 self._WriteConfig() @locking.ssynchronized(_config_lock) @@ -616,10 +1182,14 @@ class ConfigWriter: if disk.dev_type == constants.LD_FILE: # rename the file paths in logical and physical id file_storage_dir = os.path.dirname(os.path.dirname(disk.logical_id[1])) + disk_fname = "disk%s" % disk.iv_name.split("/")[1] disk.physical_id = disk.logical_id = (disk.logical_id[0], - os.path.join(file_storage_dir, - inst.name, - disk.iv_name)) + utils.PathJoin(file_storage_dir, + inst.name, + disk_fname)) + + # Force update of ssconf files + self._config_data.cluster.serial_no += 1 self._config_data.instances[inst.name] = inst self._WriteConfig() @@ -629,7 +1199,7 @@ class ConfigWriter: """Mark the status of an instance to down in the configuration. """ - self._SetInstanceStatus(instance_name, "down") + self._SetInstanceStatus(instance_name, False) def _UnlockedGetInstanceList(self): """Get the list of instances. @@ -649,16 +1219,15 @@ class ConfigWriter: """ return self._UnlockedGetInstanceList() - @locking.ssynchronized(_config_lock, shared=1) def ExpandInstanceName(self, short_name): """Attempt to expand an incomplete instance name. """ - return utils.MatchNameComponent(short_name, - self._config_data.instances.keys()) + # Locking is done in L{ConfigWriter.GetInstanceList} + return _MatchNameComponentIgnoreCase(short_name, self.GetInstanceList()) def _UnlockedGetInstanceInfo(self, instance_name): - """Returns informations about an instance. + """Returns information about an instance. This function is for internal use, when the config lock is already held. @@ -670,9 +1239,9 @@ class ConfigWriter: @locking.ssynchronized(_config_lock, shared=1) def GetInstanceInfo(self, instance_name): - """Returns informations about an instance. + """Returns information about an instance. - It takes the information from the configuration file. Other informations of + It takes the information from the configuration file. Other information of an instance are taken from the live systems. @param instance_name: name of the instance, e.g. @@ -685,11 +1254,30 @@ class ConfigWriter: return self._UnlockedGetInstanceInfo(instance_name) @locking.ssynchronized(_config_lock, shared=1) + def GetInstanceNodeGroups(self, instance_name, primary_only=False): + """Returns set of node group UUIDs for instance's nodes. + + @rtype: frozenset + + """ + instance = self._UnlockedGetInstanceInfo(instance_name) + if not instance: + raise errors.ConfigurationError("Unknown instance '%s'" % instance_name) + + if primary_only: + nodes = [instance.primary_node] + else: + nodes = instance.all_nodes + + return frozenset(self._UnlockedGetNodeInfo(node_name).group + for node_name in nodes) + + @locking.ssynchronized(_config_lock, shared=1) def GetAllInstancesInfo(self): """Get the configuration of all instances. @rtype: dict - @returns: dict of (instance, instance_info), where instance_info is what + @return: dict of (instance, instance_info), where instance_info is what would GetInstanceInfo return for the node """ @@ -698,16 +1286,20 @@ class ConfigWriter: return my_dict @locking.ssynchronized(_config_lock) - def AddNode(self, node): + def AddNode(self, node, ec_id): """Add a node to the configuration. @type node: L{objects.Node} @param node: a Node instance """ - logging.info("Adding node %s to configuration" % node.name) + logging.info("Adding node %s to configuration", node.name) + + self._EnsureUUID(node, ec_id) node.serial_no = 1 + node.ctime = node.mtime = time.time() + self._UnlockedAddNodeToGroup(node.name, node.group) self._config_data.nodes[node.name] = node self._config_data.cluster.serial_no += 1 self._WriteConfig() @@ -717,22 +1309,22 @@ class ConfigWriter: """Remove a node from the configuration. """ - logging.info("Removing node %s from configuration" % node_name) + logging.info("Removing node %s from configuration", node_name) if node_name not in self._config_data.nodes: raise errors.ConfigurationError("Unknown node '%s'" % node_name) + self._UnlockedRemoveNodeFromGroup(self._config_data.nodes[node_name]) del self._config_data.nodes[node_name] self._config_data.cluster.serial_no += 1 self._WriteConfig() - @locking.ssynchronized(_config_lock, shared=1) def ExpandNodeName(self, short_name): - """Attempt to expand an incomplete instance name. + """Attempt to expand an incomplete node name. """ - return utils.MatchNameComponent(short_name, - self._config_data.nodes.keys()) + # Locking is done in L{ConfigWriter.GetNodeList} + return _MatchNameComponentIgnoreCase(short_name, self.GetNodeList()) def _UnlockedGetNodeInfo(self, node_name): """Get the configuration of a node, as stored in the config. @@ -751,7 +1343,6 @@ class ConfigWriter: return self._config_data.nodes[node_name] - @locking.ssynchronized(_config_lock, shared=1) def GetNodeInfo(self, node_name): """Get the configuration of a node, as stored in the config. @@ -766,6 +1357,45 @@ class ConfigWriter: """ return self._UnlockedGetNodeInfo(node_name) + @locking.ssynchronized(_config_lock, shared=1) + def GetNodeInstances(self, node_name): + """Get the instances of a node, as stored in the config. + + @param node_name: the node name, e.g. I{node1.example.com} + + @rtype: (list, list) + @return: a tuple with two lists: the primary and the secondary instances + + """ + pri = [] + sec = [] + for inst in self._config_data.instances.values(): + if inst.primary_node == node_name: + pri.append(inst.name) + if node_name in inst.secondary_nodes: + sec.append(inst.name) + return (pri, sec) + + @locking.ssynchronized(_config_lock, shared=1) + def GetNodeGroupInstances(self, uuid, primary_only=False): + """Get the instances of a node group. + + @param uuid: Node group UUID + @param primary_only: Whether to only consider primary nodes + @rtype: frozenset + @return: List of instance names in node group + + """ + if primary_only: + nodes_fn = lambda inst: [inst.primary_node] + else: + nodes_fn = lambda inst: inst.all_nodes + + return frozenset(inst.name + for inst in self._config_data.instances.values() + for node_name in nodes_fn(inst) + if self._UnlockedGetNodeInfo(node_name).group == uuid) + def _UnlockedGetNodeList(self): """Return the list of nodes which are in the configuration. @@ -777,7 +1407,6 @@ class ConfigWriter: """ return self._config_data.nodes.keys() - @locking.ssynchronized(_config_lock, shared=1) def GetNodeList(self): """Return the list of nodes which are in the configuration. @@ -785,6 +1414,39 @@ class ConfigWriter: """ return self._UnlockedGetNodeList() + def _UnlockedGetOnlineNodeList(self): + """Return the list of nodes which are online. + + """ + all_nodes = [self._UnlockedGetNodeInfo(node) + for node in self._UnlockedGetNodeList()] + return [node.name for node in all_nodes if not node.offline] + + @locking.ssynchronized(_config_lock, shared=1) + def GetOnlineNodeList(self): + """Return the list of nodes which are online. + + """ + return self._UnlockedGetOnlineNodeList() + + @locking.ssynchronized(_config_lock, shared=1) + def GetVmCapableNodeList(self): + """Return the list of nodes which are not vm capable. + + """ + all_nodes = [self._UnlockedGetNodeInfo(node) + for node in self._UnlockedGetNodeList()] + return [node.name for node in all_nodes if node.vm_capable] + + @locking.ssynchronized(_config_lock, shared=1) + def GetNonVmCapableNodeList(self): + """Return the list of nodes which are not vm capable. + + """ + all_nodes = [self._UnlockedGetNodeInfo(node) + for node in self._UnlockedGetNodeList()] + return [node.name for node in all_nodes if not node.vm_capable] + @locking.ssynchronized(_config_lock, shared=1) def GetAllNodesInfo(self): """Get the configuration of all nodes. @@ -798,43 +1460,62 @@ class ConfigWriter: for node in self._UnlockedGetNodeList()]) return my_dict - def _UnlockedGetMasterCandidateStats(self): + @locking.ssynchronized(_config_lock, shared=1) + def GetNodeGroupsFromNodes(self, nodes): + """Returns groups for a list of nodes. + + @type nodes: list of string + @param nodes: List of node names + @rtype: frozenset + + """ + return frozenset(self._UnlockedGetNodeInfo(name).group for name in nodes) + + def _UnlockedGetMasterCandidateStats(self, exceptions=None): """Get the number of current and maximum desired and possible candidates. + @type exceptions: list + @param exceptions: if passed, list of nodes that should be ignored @rtype: tuple - @return: tuple of (current, desired and possible) + @return: tuple of (current, desired and possible, possible) """ - mc_now = mc_max = 0 - for node in self._config_data.nodes.itervalues(): - if not node.offline: + mc_now = mc_should = mc_max = 0 + for node in self._config_data.nodes.values(): + if exceptions and node.name in exceptions: + continue + if not (node.offline or node.drained) and node.master_capable: mc_max += 1 if node.master_candidate: mc_now += 1 - mc_max = min(mc_max, self._config_data.cluster.candidate_pool_size) - return (mc_now, mc_max) + mc_should = min(mc_max, self._config_data.cluster.candidate_pool_size) + return (mc_now, mc_should, mc_max) @locking.ssynchronized(_config_lock, shared=1) - def GetMasterCandidateStats(self): + def GetMasterCandidateStats(self, exceptions=None): """Get the number of current and maximum possible candidates. This is just a wrapper over L{_UnlockedGetMasterCandidateStats}. + @type exceptions: list + @param exceptions: if passed, list of nodes that should be ignored @rtype: tuple @return: tuple of (current, max) """ - return self._UnlockedGetMasterCandidateStats() + return self._UnlockedGetMasterCandidateStats(exceptions) @locking.ssynchronized(_config_lock) - def MaintainCandidatePool(self): + def MaintainCandidatePool(self, exceptions): """Try to grow the candidate pool to the desired size. + @type exceptions: list + @param exceptions: if passed, list of nodes that should be ignored @rtype: list @return: list with the adjusted nodes (L{objects.Node} instances) """ - mc_now, mc_max = self._UnlockedGetMasterCandidateStats() + mc_now, mc_max, _ = self._UnlockedGetMasterCandidateStats(exceptions) mod_list = [] if mc_now < mc_max: node_list = self._config_data.nodes.keys() @@ -843,7 +1524,8 @@ class ConfigWriter: if mc_now >= mc_max: break node = self._config_data.nodes[name] - if node.master_candidate or node.offline: + if (node.master_candidate or node.offline or node.drained or + node.name in exceptions or not node.master_capable): continue mod_list.append(node) node.master_candidate = True @@ -859,24 +1541,60 @@ class ConfigWriter: return mod_list + def _UnlockedAddNodeToGroup(self, node_name, nodegroup_uuid): + """Add a given node to the specified group. + + """ + if nodegroup_uuid not in self._config_data.nodegroups: + # This can happen if a node group gets deleted between its lookup and + # when we're adding the first node to it, since we don't keep a lock in + # the meantime. It's ok though, as we'll fail cleanly if the node group + # is not found anymore. + raise errors.OpExecError("Unknown node group: %s" % nodegroup_uuid) + if node_name not in self._config_data.nodegroups[nodegroup_uuid].members: + self._config_data.nodegroups[nodegroup_uuid].members.append(node_name) + + def _UnlockedRemoveNodeFromGroup(self, node): + """Remove a given node from its group. + + """ + nodegroup = node.group + if nodegroup not in self._config_data.nodegroups: + logging.warning("Warning: node '%s' has unknown node group '%s'" + " (while being removed from it)", node.name, nodegroup) + nodegroup_obj = self._config_data.nodegroups[nodegroup] + if node.name not in nodegroup_obj.members: + logging.warning("Warning: node '%s' not a member of its node group '%s'" + " (while being removed from it)", node.name, nodegroup) + else: + nodegroup_obj.members.remove(node.name) + def _BumpSerialNo(self): """Bump up the serial number of the config. """ self._config_data.serial_no += 1 + self._config_data.mtime = time.time() - def _OpenConfig(self): + def _AllUUIDObjects(self): + """Returns all objects with uuid attributes. + + """ + return (self._config_data.instances.values() + + self._config_data.nodes.values() + + self._config_data.nodegroups.values() + + [self._config_data.cluster]) + + def _OpenConfig(self, accept_foreign): """Read the config data from disk. """ - f = open(self._cfg_file, 'r') + raw_data = utils.ReadFile(self._cfg_file) + try: - try: - data = objects.ConfigData.FromDict(serializer.Load(f.read())) - except Exception, err: - raise errors.ConfigurationError(err) - finally: - f.close() + data = objects.ConfigData.FromDict(serializer.Load(raw_data)) + except Exception, err: + raise errors.ConfigurationError(err) # Make sure the configuration has the right version _ValidateConfig(data) @@ -885,12 +1603,67 @@ class ConfigWriter: not hasattr(data.cluster, 'rsahostkeypub')): raise errors.ConfigurationError("Incomplete configuration" " (missing cluster.rsahostkeypub)") + + if data.cluster.master_node != self._my_hostname and not accept_foreign: + msg = ("The configuration denotes node %s as master, while my" + " hostname is %s; opening a foreign configuration is only" + " possible in accept_foreign mode" % + (data.cluster.master_node, self._my_hostname)) + raise errors.ConfigurationError(msg) + + # Upgrade configuration if needed + data.UpgradeConfig() + self._config_data = data - # init the last serial as -1 so that the next write will cause + # reset the last serial as -1 so that the next write will cause # ssconf update self._last_cluster_serial = -1 - def _DistributeConfig(self): + # And finally run our (custom) config upgrade sequence + self._UpgradeConfig() + + self._cfg_id = utils.GetFileID(path=self._cfg_file) + + def _UpgradeConfig(self): + """Run upgrade steps that cannot be done purely in the objects. + + This is because some data elements need uniqueness across the + whole configuration, etc. + + @warning: this function will call L{_WriteConfig()}, but also + L{DropECReservations} so it needs to be called only from a + "safe" place (the constructor). If one wanted to call it with + the lock held, a DropECReservationUnlocked would need to be + created first, to avoid causing deadlock. + + """ + modified = False + for item in self._AllUUIDObjects(): + if item.uuid is None: + item.uuid = self._GenerateUniqueID(_UPGRADE_CONFIG_JID) + modified = True + if not self._config_data.nodegroups: + default_nodegroup_name = constants.INITIAL_NODE_GROUP_NAME + default_nodegroup = objects.NodeGroup(name=default_nodegroup_name, + members=[]) + self._UnlockedAddNodeGroup(default_nodegroup, _UPGRADE_CONFIG_JID, True) + modified = True + for node in self._config_data.nodes.values(): + if not node.group: + node.group = self.LookupNodeGroup(None) + modified = True + # This is technically *not* an upgrade, but needs to be done both when + # nodegroups are being added, and upon normally loading the config, + # because the members list of a node group is discarded upon + # serializing/deserializing the object. + self._UnlockedAddNodeToGroup(node.name, node.group) + if modified: + self._WriteConfig() + # This is ok even if it acquires the internal lock, as _UpgradeConfig is + # only called at config init time, without the lock held + self.DropECReservations(_UPGRADE_CONFIG_JID) + + def _DistributeConfig(self, feedback_fn): """Distribute the configuration to the other nodes. Currently, this only copies the configuration file. In the future, @@ -899,6 +1672,7 @@ class ConfigWriter: """ if self._offline: return True + bad = False node_list = [] @@ -919,41 +1693,78 @@ class ConfigWriter: result = rpc.RpcRunner.call_upload_file(node_list, self._cfg_file, address_list=addr_list) - for node in node_list: - if not result[node]: - logging.error("copy of file %s to node %s failed", - self._cfg_file, node) + for to_node, to_result in result.items(): + msg = to_result.fail_msg + if msg: + msg = ("Copy of file %s to node %s failed: %s" % + (self._cfg_file, to_node, msg)) + logging.error(msg) + + if feedback_fn: + feedback_fn(msg) + bad = True + return not bad - def _WriteConfig(self, destination=None): + def _WriteConfig(self, destination=None, feedback_fn=None): """Write the configuration data to persistent storage. """ + assert feedback_fn is None or callable(feedback_fn) + + # Warn on config errors, but don't abort the save - the + # configuration has already been modified, and we can't revert; + # the best we can do is to warn the user and save as is, leaving + # recovery to the user + config_errors = self._UnlockedVerifyConfig() + if config_errors: + errmsg = ("Configuration data is not consistent: %s" % + (utils.CommaJoin(config_errors))) + logging.critical(errmsg) + if feedback_fn: + feedback_fn(errmsg) + if destination is None: destination = self._cfg_file self._BumpSerialNo() txt = serializer.Dump(self._config_data.ToDict()) - dir_name, file_name = os.path.split(destination) - fd, name = tempfile.mkstemp('.newconfig', file_name, dir_name) - f = os.fdopen(fd, 'w') + + getents = self._getents() + try: + fd = utils.SafeWriteFile(destination, self._cfg_id, data=txt, + close=False, gid=getents.confd_gid, mode=0640) + except errors.LockError: + raise errors.ConfigurationError("The configuration file has been" + " modified since the last write, cannot" + " update") try: - f.write(txt) - os.fsync(f.fileno()) + self._cfg_id = utils.GetFileID(fd=fd) finally: - f.close() - # we don't need to do os.close(fd) as f.close() did it - os.rename(name, destination) + os.close(fd) + self.write_count += 1 # and redistribute the config file to master candidates - self._DistributeConfig() + self._DistributeConfig(feedback_fn) # Write ssconf files on all nodes (including locally) if self._last_cluster_serial < self._config_data.cluster.serial_no: if not self._offline: - rpc.RpcRunner.call_write_ssconf_files(self._UnlockedGetNodeList(), - self._UnlockedGetSsconfValues()) + result = rpc.RpcRunner.call_write_ssconf_files( + self._UnlockedGetOnlineNodeList(), + self._UnlockedGetSsconfValues()) + + for nname, nresu in result.items(): + msg = nresu.fail_msg + if msg: + errmsg = ("Error while uploading ssconf files to" + " node %s: %s" % (nname, msg)) + logging.warning(errmsg) + + if feedback_fn: + feedback_fn(errmsg) + self._last_cluster_serial = self._config_data.cluster.serial_no def _UnlockedGetSsconfValues(self): @@ -965,50 +1776,72 @@ class ConfigWriter: """ fn = "\n".join + instance_names = utils.NiceSort(self._UnlockedGetInstanceList()) node_names = utils.NiceSort(self._UnlockedGetNodeList()) node_info = [self._UnlockedGetNodeInfo(name) for name in node_names] + node_pri_ips = ["%s %s" % (ninfo.name, ninfo.primary_ip) + for ninfo in node_info] + node_snd_ips = ["%s %s" % (ninfo.name, ninfo.secondary_ip) + for ninfo in node_info] + instance_data = fn(instance_names) off_data = fn(node.name for node in node_info if node.offline) + on_data = fn(node.name for node in node_info if not node.offline) mc_data = fn(node.name for node in node_info if node.master_candidate) + mc_ips_data = fn(node.primary_ip for node in node_info + if node.master_candidate) node_data = fn(node_names) + node_pri_ips_data = fn(node_pri_ips) + node_snd_ips_data = fn(node_snd_ips) cluster = self._config_data.cluster - return { + cluster_tags = fn(cluster.GetTags()) + + hypervisor_list = fn(cluster.enabled_hypervisors) + + uid_pool = uidpool.FormatUidPool(cluster.uid_pool, separator="\n") + + nodegroups = ["%s %s" % (nodegroup.uuid, nodegroup.name) for nodegroup in + self._config_data.nodegroups.values()] + nodegroups_data = fn(utils.NiceSort(nodegroups)) + + ssconf_values = { constants.SS_CLUSTER_NAME: cluster.cluster_name, + constants.SS_CLUSTER_TAGS: cluster_tags, constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir, + constants.SS_SHARED_FILE_STORAGE_DIR: cluster.shared_file_storage_dir, constants.SS_MASTER_CANDIDATES: mc_data, + constants.SS_MASTER_CANDIDATES_IPS: mc_ips_data, constants.SS_MASTER_IP: cluster.master_ip, constants.SS_MASTER_NETDEV: cluster.master_netdev, constants.SS_MASTER_NODE: cluster.master_node, constants.SS_NODE_LIST: node_data, + constants.SS_NODE_PRIMARY_IPS: node_pri_ips_data, + constants.SS_NODE_SECONDARY_IPS: node_snd_ips_data, constants.SS_OFFLINE_NODES: off_data, + constants.SS_ONLINE_NODES: on_data, + constants.SS_PRIMARY_IP_FAMILY: str(cluster.primary_ip_family), + constants.SS_INSTANCE_LIST: instance_data, + constants.SS_RELEASE_VERSION: constants.RELEASE_VERSION, + constants.SS_HYPERVISOR_LIST: hypervisor_list, + constants.SS_MAINTAIN_NODE_HEALTH: str(cluster.maintain_node_health), + constants.SS_UID_POOL: uid_pool, + constants.SS_NODEGROUPS: nodegroups_data, } + bad_values = [(k, v) for k, v in ssconf_values.items() + if not isinstance(v, (str, basestring))] + if bad_values: + err = utils.CommaJoin("%s=%s" % (k, v) for k, v in bad_values) + raise errors.ConfigurationError("Some ssconf key(s) have non-string" + " values: %s" % err) + return ssconf_values - @locking.ssynchronized(_config_lock) - def InitConfig(self, version, cluster_config, master_node_config): - """Create the initial cluster configuration. - - It will contain the current node, which will also be the master - node, and no instances. - - @type version: int - @param version: Configuration version - @type cluster_config: objects.Cluster - @param cluster_config: Cluster configuration - @type master_node_config: objects.Node - @param master_node_config: Master node configuration + @locking.ssynchronized(_config_lock, shared=1) + def GetSsconfValues(self): + """Wrapper using lock around _UnlockedGetSsconf(). """ - nodes = { - master_node_config.name: master_node_config, - } - - self._config_data = objects.ConfigData(version=version, - cluster=cluster_config, - nodes=nodes, - instances={}, - serial_no=1) - self._WriteConfig() + return self._UnlockedGetSsconfValues() @locking.ssynchronized(_config_lock, shared=1) def GetVGName(self): @@ -1027,11 +1860,20 @@ class ConfigWriter: self._WriteConfig() @locking.ssynchronized(_config_lock, shared=1) - def GetDefBridge(self): - """Return the default bridge. + def GetDRBDHelper(self): + """Return DRBD usermode helper. """ - return self._config_data.cluster.default_bridge + return self._config_data.cluster.drbd_usermode_helper + + @locking.ssynchronized(_config_lock) + def SetDRBDHelper(self, drbd_helper): + """Set DRBD usermode helper. + + """ + self._config_data.cluster.drbd_usermode_helper = drbd_helper + self._config_data.cluster.serial_no += 1 + self._WriteConfig() @locking.ssynchronized(_config_lock, shared=1) def GetMACPrefix(self): @@ -1042,7 +1884,7 @@ class ConfigWriter: @locking.ssynchronized(_config_lock, shared=1) def GetClusterInfo(self): - """Returns informations about the cluster + """Returns information about the cluster @rtype: L{objects.Cluster} @return: the cluster object @@ -1050,8 +1892,15 @@ class ConfigWriter: """ return self._config_data.cluster + @locking.ssynchronized(_config_lock, shared=1) + def HasAnyDiskOfType(self, dev_type): + """Check if in there is at disk of the given type in the configuration. + + """ + return self._config_data.HasAnyDiskOfType(dev_type) + @locking.ssynchronized(_config_lock) - def Update(self, target): + def Update(self, target, feedback_fn): """Notify function to be called after updates. This function must be called when an object (as returned by @@ -1063,6 +1912,7 @@ class ConfigWriter: @param target: an instance of either L{objects.Cluster}, L{objects.Node} or L{objects.Instance} which is existing in the cluster + @param feedback_fn: Callable feedback function """ if self._config_data is None: @@ -1076,6 +1926,8 @@ class ConfigWriter: update_serial = True elif isinstance(target, objects.Instance): test = target in self._config_data.instances.values() + elif isinstance(target, objects.NodeGroup): + test = target in self._config_data.nodegroups.values() else: raise errors.ProgrammerError("Invalid object type (%s) passed to" " ConfigWriter.Update" % type(target)) @@ -1083,9 +1935,22 @@ class ConfigWriter: raise errors.ConfigurationError("Configuration updated since object" " has been read or unknown object") target.serial_no += 1 + target.mtime = now = time.time() if update_serial: # for node updates, we need to increase the cluster serial too self._config_data.cluster.serial_no += 1 + self._config_data.cluster.mtime = now - self._WriteConfig() + if isinstance(target, objects.Instance): + self._UnlockedReleaseDRBDMinors(target.name) + + self._WriteConfig(feedback_fn=feedback_fn) + + @locking.ssynchronized(_config_lock) + def DropECReservations(self, ec_id): + """Drop per-execution-context reservations + + """ + for rm in self._all_rms: + rm.DropECReservations(ec_id)