X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/afa1386ed7caa6075c8e145035cccd6186f404ce..2492231f7381ce33f2164322f799c84b2d7cceef:/lib/config.py diff --git a/lib/config.py b/lib/config.py index 52bfe0f..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 @@ -31,6 +31,9 @@ much memory. """ +# pylint: disable-msg=R0904 +# R0904: Too many public methods + import os import random import logging @@ -43,12 +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() +_config_lock = locking.SharedLock("ConfigWriter") # job id used for resource management at config upgrade time -_UPGRADE_CONFIG_JID="jid-cfg-upgrade" +_UPGRADE_CONFIG_JID = "jid-cfg-upgrade" def _ValidateConfig(data): @@ -61,10 +67,7 @@ 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: @@ -78,15 +81,15 @@ class TemporaryReservationManager: self._ec_reserved = {} def Reserved(self, resource): - for holder_reserved in self._ec_reserved.items(): + 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." % - (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: @@ -122,11 +125,22 @@ class TemporaryReservationManager: 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 @@ -135,17 +149,22 @@ class ConfigWriter: self._cfg_file = constants.CLUSTER_CONF_FILE else: self._cfg_file = cfg_file + 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._my_hostname = netutils.Hostname.GetSysName() self._last_cluster_serial = -1 - self._OpenConfig() + self._cfg_id = None + self._OpenConfig(accept_foreign) # this method needs to be static, so that we can call it on the class @staticmethod @@ -167,6 +186,18 @@ class ConfigWriter: 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, ec_id): """Generate a MAC for an instance. @@ -188,7 +219,21 @@ class ConfigWriter: if mac in all_macs: raise errors.ReservationError("mac already in use") else: - self._temporary_macs.Reserve(mac, ec_id) + self._temporary_macs.Reserve(ec_id, mac) + + @locking.ssynchronized(_config_lock, shared=1) + 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): @@ -200,6 +245,7 @@ class ConfigWriter: return self._temporary_secrets.Generate(self._AllDRBDSecrets(), utils.GenerateSecret, ec_id) + def _AllLVs(self): """Compute the list of all LVs. @@ -328,28 +374,59 @@ class ConfigWriter: 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 data.cluster.enabled_hypervisors: + if not cluster.enabled_hypervisors: result.append("enabled hypervisors list doesn't have any entries") - invalid_hvs = set(data.cluster.enabled_hypervisors) - constants.HYPER_TYPES + 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 data.cluster.master_node not in data.nodes: + if cluster.master_node not in data.nodes: result.append("cluster has invalid primary node '%s'" % - data.cluster.master_node) + 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)) @@ -363,6 +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: @@ -385,7 +473,7 @@ class ConfigWriter: 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")) @@ -396,16 +484,16 @@ 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") # master candidate checks @@ -415,36 +503,87 @@ class ConfigWriter: (mc_now, mc_max)) # node checks - for node in data.nodes.values(): + 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.drain, + (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 - d_map, duplicates = self._UnlockedComputeDRBDMap() + _, 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 - ips = { data.cluster.master_ip: ["cluster_ip"] } - def _helper(ip, name): - if ip in ips: - ips[ip].append(name) - else: - ips[ip] = [name] + 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(): - _helper(node.primary_ip, "node:%s/primary" % node.name) + _AddIpAddress(node.primary_ip, "node:%s/primary" % node.name) if node.secondary_ip != node.primary_ip: - _helper(node.secondary_ip, "node:%s/secondary" % node.name) + _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, ", ".join(owners))) + (ip, utils.CommaJoin(owners))) + return result @locking.ssynchronized(_config_lock, shared=1) @@ -746,6 +885,13 @@ 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. @@ -762,6 +908,180 @@ 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 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. @@ -804,9 +1124,9 @@ class ConfigWriter: """ if not item.uuid: item.uuid = self._GenerateUniqueID(ec_id) - elif item.uuid in self._AllIDs(temporary=True): - raise errors.ConfigurationError("Cannot add '%s': UUID already in use" % - (item.name, item.uuid)) + 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. @@ -862,10 +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() @@ -895,14 +1219,12 @@ 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(), - case_sensitive=False) + # Locking is done in L{ConfigWriter.GetInstanceList} + return _MatchNameComponentIgnoreCase(short_name, self.GetInstanceList()) def _UnlockedGetInstanceInfo(self, instance_name): """Returns information about an instance. @@ -932,6 +1254,25 @@ 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. @@ -958,6 +1299,7 @@ class ConfigWriter: 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() @@ -972,18 +1314,17 @@ class ConfigWriter: 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(), - case_sensitive=False) + # 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. @@ -1002,7 +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. @@ -1017,6 +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. @@ -1028,7 +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. @@ -1036,14 +1414,38 @@ 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 not node.offline] + 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): @@ -1058,6 +1460,17 @@ class ConfigWriter: for node in self._UnlockedGetNodeList()]) return my_dict + @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. @@ -1071,7 +1484,7 @@ class ConfigWriter: for node in self._config_data.nodes.values(): if exceptions and node.name in exceptions: continue - if not (node.offline or node.drained): + if not (node.offline or node.drained) and node.master_capable: mc_max += 1 if node.master_candidate: mc_now += 1 @@ -1112,7 +1525,7 @@ class ConfigWriter: break node = self._config_data.nodes[name] if (node.master_candidate or node.offline or node.drained or - node.name in exceptions): + node.name in exceptions or not node.master_capable): continue mod_list.append(node) node.master_candidate = True @@ -1128,6 +1541,34 @@ 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. @@ -1141,9 +1582,10 @@ class ConfigWriter: """ return (self._config_data.instances.values() + self._config_data.nodes.values() + + self._config_data.nodegroups.values() + [self._config_data.cluster]) - def _OpenConfig(self): + def _OpenConfig(self, accept_foreign): """Read the config data from disk. """ @@ -1162,6 +1604,13 @@ class ConfigWriter: 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() @@ -1173,15 +1622,19 @@ class ConfigWriter: # 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()}, so it needs - to either be called with the lock held or from a safe place - (the constructor) + @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 @@ -1189,13 +1642,27 @@ class ConfigWriter: 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. @@ -1253,7 +1720,7 @@ class ConfigWriter: config_errors = self._UnlockedVerifyConfig() if config_errors: errmsg = ("Configuration data is not consistent: %s" % - (", ".join(config_errors))) + (utils.CommaJoin(config_errors))) logging.critical(errmsg) if feedback_fn: feedback_fn(errmsg) @@ -1263,7 +1730,18 @@ class ConfigWriter: self._BumpSerialNo() txt = serializer.Dump(self._config_data.ToDict()) - utils.WriteFile(destination, data=txt) + 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: + self._cfg_id = utils.GetFileID(fd=fd) + finally: + os.close(fd) self.write_count += 1 @@ -1274,7 +1752,7 @@ class ConfigWriter: if self._last_cluster_serial < self._config_data.cluster.serial_no: if not self._offline: result = rpc.RpcRunner.call_write_ssconf_files( - self._UnlockedGetNodeList(), + self._UnlockedGetOnlineNodeList(), self._UnlockedGetSsconfValues()) for nname, nresu in result.items(): @@ -1318,10 +1796,20 @@ class ConfigWriter: cluster = self._config_data.cluster cluster_tags = fn(cluster.GetTags()) - return { + + 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, @@ -1332,9 +1820,28 @@ class ConfigWriter: 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, shared=1) + def GetSsconfValues(self): + """Wrapper using lock around _UnlockedGetSsconf(). + + """ + return self._UnlockedGetSsconfValues() @locking.ssynchronized(_config_lock, shared=1) def GetVGName(self): @@ -1353,6 +1860,22 @@ class ConfigWriter: self._WriteConfig() @locking.ssynchronized(_config_lock, shared=1) + def GetDRBDHelper(self): + """Return DRBD usermode helper. + + """ + 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): """Return the mac prefix. @@ -1369,6 +1892,13 @@ 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, feedback_fn): """Notify function to be called after updates. @@ -1396,6 +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)) @@ -1420,7 +1952,5 @@ class ConfigWriter: """Drop per-execution-context reservations """ - self._temporary_ids.DropECReservations(ec_id) - self._temporary_macs.DropECReservations(ec_id) - self._temporary_secrets.DropECReservations(ec_id) - + for rm in self._all_rms: + rm.DropECReservations(ec_id)