X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/8a69b3a8193cab2c46daceccf8686028a43cecc3..cf572b1378a61db132d84a5e03b0e5a157848caa:/lib/cmdlib.py diff --git a/lib/cmdlib.py b/lib/cmdlib.py index 561a9bf..f867a4d 100644 --- a/lib/cmdlib.py +++ b/lib/cmdlib.py @@ -1,7 +1,7 @@ # # -# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 Google Inc. +# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012 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 @@ -721,6 +721,109 @@ def _GetUpdatedParams(old_params, update_dict, return params_copy +def _GetUpdatedIPolicy(old_ipolicy, new_ipolicy, group_policy=False): + """Return the new version of a instance policy. + + @param group_policy: whether this policy applies to a group and thus + we should support removal of policy entries + + """ + use_none = use_default = group_policy + ipolicy = copy.deepcopy(old_ipolicy) + for key, value in new_ipolicy.items(): + if key not in constants.IPOLICY_ALL_KEYS: + raise errors.OpPrereqError("Invalid key in new ipolicy: %s" % key, + errors.ECODE_INVAL) + if key in constants.IPOLICY_PARAMETERS: + utils.ForceDictType(value, constants.ISPECS_PARAMETER_TYPES) + ipolicy[key] = _GetUpdatedParams(old_ipolicy.get(key, {}), value, + use_none=use_none, + use_default=use_default) + else: + # FIXME: we assume all others are lists; this should be redone + # in a nicer way + if not value or value == [constants.VALUE_DEFAULT]: + if group_policy: + del ipolicy[key] + else: + raise errors.OpPrereqError("Can't unset ipolicy attribute '%s'" + " on the cluster'" % key, + errors.ECODE_INVAL) + else: + ipolicy[key] = list(value) + try: + objects.InstancePolicy.CheckParameterSyntax(ipolicy) + except errors.ConfigurationError, err: + raise errors.OpPrereqError("Invalid instance policy: %s" % err, + errors.ECODE_INVAL) + return ipolicy + + +def _UpdateAndVerifySubDict(base, updates, type_check): + """Updates and verifies a dict with sub dicts of the same type. + + @param base: The dict with the old data + @param updates: The dict with the new data + @param type_check: Dict suitable to ForceDictType to verify correct types + @returns: A new dict with updated and verified values + + """ + def fn(old, value): + new = _GetUpdatedParams(old, value) + utils.ForceDictType(new, type_check) + return new + + ret = copy.deepcopy(base) + ret.update(dict((key, fn(base.get(key, {}), value)) + for key, value in updates.items())) + return ret + + +def _MergeAndVerifyHvState(op_input, obj_input): + """Combines the hv state from an opcode with the one of the object + + @param op_input: The input dict from the opcode + @param obj_input: The input dict from the objects + @return: The verified and updated dict + + """ + if op_input: + invalid_hvs = set(op_input) - constants.HYPER_TYPES + if invalid_hvs: + raise errors.OpPrereqError("Invalid hypervisor(s) in hypervisor state:" + " %s" % utils.CommaJoin(invalid_hvs), + errors.ECODE_INVAL) + if obj_input is None: + obj_input = {} + type_check = constants.HVSTS_PARAMETER_TYPES + return _UpdateAndVerifySubDict(obj_input, op_input, type_check) + + return None + + +def _MergeAndVerifyDiskState(op_input, obj_input): + """Combines the disk state from an opcode with the one of the object + + @param op_input: The input dict from the opcode + @param obj_input: The input dict from the objects + @return: The verified and updated dict + """ + if op_input: + invalid_dst = set(op_input) - constants.DS_VALID_TYPES + if invalid_dst: + raise errors.OpPrereqError("Invalid storage type(s) in disk state: %s" % + utils.CommaJoin(invalid_dst), + errors.ECODE_INVAL) + type_check = constants.DSS_PARAMETER_TYPES + if obj_input is None: + obj_input = {} + return dict((key, _UpdateAndVerifySubDict(obj_input.get(key, {}), value, + type_check)) + for key, value in op_input.items()) + + return None + + def _ReleaseLocks(lu, level, names=None, keep=None): """Releases locks owned by an LU. @@ -936,8 +1039,8 @@ def _CheckInstanceState(lu, instance, req_states, msg=None): if msg is None: msg = "can't use instance from outside %s states" % ", ".join(req_states) if instance.admin_state not in req_states: - raise errors.OpPrereqError("Instance %s is marked to be %s, %s" % - (instance, instance.admin_state, msg), + raise errors.OpPrereqError("Instance '%s' is marked to be %s, %s" % + (instance.name, instance.admin_state, msg), errors.ECODE_STATE) if constants.ADMINST_UP not in req_states: @@ -951,6 +1054,161 @@ def _CheckInstanceState(lu, instance, req_states, msg=None): (instance.name, msg), errors.ECODE_STATE) +def _ComputeMinMaxSpec(name, ipolicy, value): + """Computes if value is in the desired range. + + @param name: name of the parameter for which we perform the check + @param ipolicy: dictionary containing min, max and std values + @param value: actual value that we want to use + @return: None or element not meeting the criteria + + + """ + if value in [None, constants.VALUE_AUTO]: + return None + max_v = ipolicy[constants.ISPECS_MAX].get(name, value) + min_v = ipolicy[constants.ISPECS_MIN].get(name, value) + if value > max_v or min_v > value: + return ("%s value %s is not in range [%s, %s]" % + (name, value, min_v, max_v)) + return None + + +def _ComputeIPolicySpecViolation(ipolicy, mem_size, cpu_count, disk_count, + nic_count, disk_sizes, + _compute_fn=_ComputeMinMaxSpec): + """Verifies ipolicy against provided specs. + + @type ipolicy: dict + @param ipolicy: The ipolicy + @type mem_size: int + @param mem_size: The memory size + @type cpu_count: int + @param cpu_count: Used cpu cores + @type disk_count: int + @param disk_count: Number of disks used + @type nic_count: int + @param nic_count: Number of nics used + @type disk_sizes: list of ints + @param disk_sizes: Disk sizes of used disk (len must match C{disk_count}) + @param _compute_fn: The compute function (unittest only) + @return: A list of violations, or an empty list of no violations are found + + """ + assert disk_count == len(disk_sizes) + + test_settings = [ + (constants.ISPEC_MEM_SIZE, mem_size), + (constants.ISPEC_CPU_COUNT, cpu_count), + (constants.ISPEC_DISK_COUNT, disk_count), + (constants.ISPEC_NIC_COUNT, nic_count), + ] + map((lambda d: (constants.ISPEC_DISK_SIZE, d)), disk_sizes) + + return filter(None, + (_compute_fn(name, ipolicy, value) + for (name, value) in test_settings)) + + +def _ComputeIPolicyInstanceViolation(ipolicy, instance, + _compute_fn=_ComputeIPolicySpecViolation): + """Compute if instance meets the specs of ipolicy. + + @type ipolicy: dict + @param ipolicy: The ipolicy to verify against + @type instance: L{objects.Instance} + @param instance: The instance to verify + @param _compute_fn: The function to verify ipolicy (unittest only) + @see: L{_ComputeIPolicySpecViolation} + + """ + mem_size = instance.beparams.get(constants.BE_MAXMEM, None) + cpu_count = instance.beparams.get(constants.BE_VCPUS, None) + disk_count = len(instance.disks) + disk_sizes = [disk.size for disk in instance.disks] + nic_count = len(instance.nics) + + return _compute_fn(ipolicy, mem_size, cpu_count, disk_count, nic_count, + disk_sizes) + + +def _ComputeIPolicyInstanceSpecViolation(ipolicy, instance_spec, + _compute_fn=_ComputeIPolicySpecViolation): + """Compute if instance specs meets the specs of ipolicy. + + @type ipolicy: dict + @param ipolicy: The ipolicy to verify against + @param instance_spec: dict + @param instance_spec: The instance spec to verify + @param _compute_fn: The function to verify ipolicy (unittest only) + @see: L{_ComputeIPolicySpecViolation} + + """ + mem_size = instance_spec.get(constants.ISPEC_MEM_SIZE, None) + cpu_count = instance_spec.get(constants.ISPEC_CPU_COUNT, None) + disk_count = instance_spec.get(constants.ISPEC_DISK_COUNT, 0) + disk_sizes = instance_spec.get(constants.ISPEC_DISK_SIZE, []) + nic_count = instance_spec.get(constants.ISPEC_NIC_COUNT, 0) + + return _compute_fn(ipolicy, mem_size, cpu_count, disk_count, nic_count, + disk_sizes) + + +def _ComputeIPolicyNodeViolation(ipolicy, instance, current_group, + target_group, + _compute_fn=_ComputeIPolicyInstanceViolation): + """Compute if instance meets the specs of the new target group. + + @param ipolicy: The ipolicy to verify + @param instance: The instance object to verify + @param current_group: The current group of the instance + @param target_group: The new group of the instance + @param _compute_fn: The function to verify ipolicy (unittest only) + @see: L{_ComputeIPolicySpecViolation} + + """ + if current_group == target_group: + return [] + else: + return _compute_fn(ipolicy, instance) + + +def _CheckTargetNodeIPolicy(lu, ipolicy, instance, node, ignore=False, + _compute_fn=_ComputeIPolicyNodeViolation): + """Checks that the target node is correct in terms of instance policy. + + @param ipolicy: The ipolicy to verify + @param instance: The instance object to verify + @param node: The new node to relocate + @param ignore: Ignore violations of the ipolicy + @param _compute_fn: The function to verify ipolicy (unittest only) + @see: L{_ComputeIPolicySpecViolation} + + """ + primary_node = lu.cfg.GetNodeInfo(instance.primary_node) + res = _compute_fn(ipolicy, instance, primary_node.group, node.group) + + if res: + msg = ("Instance does not meet target node group's (%s) instance" + " policy: %s") % (node.group, utils.CommaJoin(res)) + if ignore: + lu.LogWarning(msg) + else: + raise errors.OpPrereqError(msg, errors.ECODE_INVAL) + + +def _ComputeNewInstanceViolations(old_ipolicy, new_ipolicy, instances): + """Computes a set of any instances that would violate the new ipolicy. + + @param old_ipolicy: The current (still in-place) ipolicy + @param new_ipolicy: The new (to become) ipolicy + @param instances: List of instances to verify + @return: A list of instances which violates the new ipolicy but did not before + + """ + return (_ComputeViolatingInstances(old_ipolicy, instances) - + _ComputeViolatingInstances(new_ipolicy, instances)) + + def _ExpandItemName(fn, name, kind): """Expand an item name. @@ -1164,6 +1422,26 @@ def _DecideSelfPromotion(lu, exceptions=None): return mc_now < mc_should +def _CalculateGroupIPolicy(cluster, group): + """Calculate instance policy for group. + + """ + return cluster.SimpleFillIPolicy(group.ipolicy) + + +def _ComputeViolatingInstances(ipolicy, instances): + """Computes a set of instances who violates given ipolicy. + + @param ipolicy: The ipolicy to verify + @type instances: object.Instance + @param instances: List of instances to verify + @return: A frozenset of instance names violating the ipolicy + + """ + return frozenset([inst.name for inst in instances + if _ComputeIPolicyInstanceViolation(ipolicy, inst)]) + + def _CheckNicsBridgesExist(lu, target_nics, target_node): """Check that the brigdes needed by a list of nics exist. @@ -1405,7 +1683,9 @@ class LUClusterDestroy(LogicalUnit): ems = self.cfg.GetUseExternalMipScript() result = self.rpc.call_node_deactivate_master_ip(master_params.name, master_params, ems) - result.Raise("Could not disable the master role") + if result.fail_msg: + self.LogWarning("Error disabling the master IP address: %s", + result.fail_msg) return master_params.name @@ -2056,6 +2336,10 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors): node_vol_should = {} instanceconfig.MapLVsByNode(node_vol_should) + ipolicy = _CalculateGroupIPolicy(self.cfg.GetClusterInfo(), self.group_info) + err = _ComputeIPolicyInstanceViolation(ipolicy, instanceconfig) + _ErrorIf(err, constants.CV_EINSTANCEPOLICY, instance, err) + for node in node_vol_should: n_img = node_image[node] if n_img.offline or n_img.rpc_fail or n_img.lvm_fail: @@ -3505,8 +3789,14 @@ class LUClusterSetParams(LogicalUnit): # all nodes to be modified. self.needed_locks = { locking.LEVEL_NODE: locking.ALL_SET, + locking.LEVEL_INSTANCE: locking.ALL_SET, + locking.LEVEL_NODEGROUP: locking.ALL_SET, + } + self.share_locks = { + locking.LEVEL_NODE: 1, + locking.LEVEL_INSTANCE: 1, + locking.LEVEL_NODEGROUP: 1, } - self.share_locks[locking.LEVEL_NODE] = 1 def BuildHooksEnv(self): """Build hooks env. @@ -3595,6 +3885,42 @@ class LUClusterSetParams(LogicalUnit): self.new_ndparams["oob_program"] = \ constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM] + if self.op.hv_state: + new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, + self.cluster.hv_state_static) + self.new_hv_state = dict((hv, cluster.SimpleFillHvState(values)) + for hv, values in new_hv_state.items()) + + if self.op.disk_state: + new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state, + self.cluster.disk_state_static) + self.new_disk_state = \ + dict((storage, dict((name, cluster.SimpleFillDiskState(values)) + for name, values in svalues.items())) + for storage, svalues in new_disk_state.items()) + + if self.op.ipolicy: + self.new_ipolicy = _GetUpdatedIPolicy(cluster.ipolicy, self.op.ipolicy, + group_policy=False) + + all_instances = self.cfg.GetAllInstancesInfo().values() + violations = set() + for group in self.cfg.GetAllNodeGroupsInfo().values(): + instances = frozenset([inst for inst in all_instances + if compat.any(node in group.members + for node in inst.all_nodes)]) + new_ipolicy = objects.FillIPolicy(self.new_ipolicy, group.ipolicy) + new = _ComputeNewInstanceViolations(_CalculateGroupIPolicy(cluster, + group), + new_ipolicy, instances) + if new: + violations.update(new) + + if violations: + self.LogWarning("After the ipolicy change the following instances" + " violate them: %s", + utils.CommaJoin(violations)) + if self.op.nicparams: utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES) self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams) @@ -3755,12 +4081,18 @@ class LUClusterSetParams(LogicalUnit): self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams if self.op.nicparams: self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams + if self.op.ipolicy: + self.cluster.ipolicy = self.new_ipolicy if self.op.osparams: self.cluster.osparams = self.new_osp if self.op.ndparams: self.cluster.ndparams = self.new_ndparams if self.op.diskparams: self.cluster.diskparams = self.new_diskparams + if self.op.hv_state: + self.cluster.hv_state_static = self.new_hv_state + if self.op.disk_state: + self.cluster.disk_state_static = self.new_disk_state if self.op.candidate_pool_size is not None: self.cluster.candidate_pool_size = self.op.candidate_pool_size @@ -5225,6 +5557,12 @@ class LUNodeAdd(LogicalUnit): if self.op.ndparams: utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES) + if self.op.hv_state: + self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, None) + + if self.op.disk_state: + self.new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state, None) + def Exec(self, feedback_fn): """Adds the new node to the cluster. @@ -5263,6 +5601,12 @@ class LUNodeAdd(LogicalUnit): else: new_node.ndparams = {} + if self.op.hv_state: + new_node.hv_state_static = self.new_hv_state + + if self.op.disk_state: + new_node.disk_state_static = self.new_disk_state + # check connectivity result = self.rpc.call_version([node])[node] result.Raise("Can't get version information from node %s" % node) @@ -5349,7 +5693,8 @@ class LUNodeSetParams(LogicalUnit): self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name) all_mods = [self.op.offline, self.op.master_candidate, self.op.drained, self.op.master_capable, self.op.vm_capable, - self.op.secondary_ip, self.op.ndparams] + self.op.secondary_ip, self.op.ndparams, self.op.hv_state, + self.op.disk_state] if all_mods.count(None) == len(all_mods): raise errors.OpPrereqError("Please pass at least one modification", errors.ECODE_INVAL) @@ -5590,6 +5935,15 @@ class LUNodeSetParams(LogicalUnit): utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES) self.new_ndparams = new_ndparams + if self.op.hv_state: + self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, + self.node.hv_state_static) + + if self.op.disk_state: + self.new_disk_state = \ + _MergeAndVerifyDiskState(self.op.disk_state, + self.node.disk_state_static) + def Exec(self, feedback_fn): """Modifies a node. @@ -5606,6 +5960,12 @@ class LUNodeSetParams(LogicalUnit): if self.op.powered is not None: node.powered = self.op.powered + if self.op.hv_state: + node.hv_state_static = self.new_hv_state + + if self.op.disk_state: + node.disk_state_static = self.new_disk_state + for attr in ["master_capable", "vm_capable"]: val = getattr(self.op, attr) if val is not None: @@ -5720,6 +6080,7 @@ class LUClusterQuery(NoHooksLU): "os_hvp": os_hvp, "beparams": cluster.beparams, "osparams": cluster.osparams, + "ipolicy": cluster.ipolicy, "nicparams": cluster.nicparams, "ndparams": cluster.ndparams, "candidate_pool_size": cluster.candidate_pool_size, @@ -6171,6 +6532,11 @@ class LUInstanceStartup(LogicalUnit): def ExpandNames(self): self._ExpandAndLockInstance() + self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE + + def DeclareLocks(self, level): + if level == locking.LEVEL_NODE_RES: + self._LockInstancesNodes(primary_only=True, level=locking.LEVEL_NODE_RES) def BuildHooksEnv(self): """Build hooks env. @@ -6227,6 +6593,7 @@ class LUInstanceStartup(LogicalUnit): _CheckNodeOnline(self, instance.primary_node) bep = self.cfg.GetClusterInfo().FillBE(instance) + bep.update(self.op.beparams) # check bridges existence _CheckInstanceBridgesExist(self, instance) @@ -6239,7 +6606,7 @@ class LUInstanceStartup(LogicalUnit): if not remote_info.payload: # not running already _CheckNodeFreeMemory(self, instance.primary_node, "starting instance %s" % instance.name, - bep[constants.BE_MAXMEM], instance.hypervisor) + bep[constants.BE_MINMEM], instance.hypervisor) def Exec(self, feedback_fn): """Start the instance. @@ -6546,6 +6913,7 @@ class LUInstanceRecreateDisks(LogicalUnit): self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes) else: self.needed_locks[locking.LEVEL_NODE] = [] + self.needed_locks[locking.LEVEL_NODE_RES] = [] def DeclareLocks(self, level): if level == locking.LEVEL_NODE: @@ -6941,7 +7309,8 @@ class LUInstanceFailover(LogicalUnit): cleanup=False, failover=True, ignore_consistency=ignore_consistency, - shutdown_timeout=shutdown_timeout) + shutdown_timeout=shutdown_timeout, + ignore_ipolicy=self.op.ignore_ipolicy) self.tasklets = [self._migrater] def DeclareLocks(self, level): @@ -7015,7 +7384,8 @@ class LUInstanceMigrate(LogicalUnit): self._migrater = TLMigrateInstance(self, self.op.instance_name, cleanup=self.op.cleanup, failover=False, - fallback=self.op.allow_failover) + fallback=self.op.allow_failover, + ignore_ipolicy=self.op.ignore_ipolicy) self.tasklets = [self._migrater] def DeclareLocks(self, level): @@ -7144,6 +7514,10 @@ class LUInstanceMove(LogicalUnit): _CheckNodeOnline(self, target_node) _CheckNodeNotDrained(self, target_node) _CheckNodeVmCapable(self, target_node) + ipolicy = _CalculateGroupIPolicy(self.cfg.GetClusterInfo(), + self.cfg.GetNodeGroup(node.group)) + _CheckTargetNodeIPolicy(self, ipolicy, instance, node, + ignore=self.op.ignore_ipolicy) if instance.admin_state == constants.ADMINST_UP: # check memory requirements on the secondary node @@ -7304,7 +7678,8 @@ class LUNodeMigrate(LogicalUnit): mode=self.op.mode, live=self.op.live, iallocator=self.op.iallocator, - target_node=self.op.target_node)] + target_node=self.op.target_node, + ignore_ipolicy=self.op.ignore_ipolicy)] for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name) ] @@ -7341,6 +7716,8 @@ class TLMigrateInstance(Tasklet): and target node @type shutdown_timeout: int @ivar shutdown_timeout: In case of failover timeout of the shutdown + @type ignore_ipolicy: bool + @ivar ignore_ipolicy: If true, we can ignore instance policy when migrating """ @@ -7351,7 +7728,8 @@ class TLMigrateInstance(Tasklet): def __init__(self, lu, instance_name, cleanup=False, failover=False, fallback=False, ignore_consistency=False, - shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT): + shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT, + ignore_ipolicy=False): """Initializes this class. """ @@ -7365,6 +7743,7 @@ class TLMigrateInstance(Tasklet): self.fallback = fallback self.ignore_consistency = ignore_consistency self.shutdown_timeout = shutdown_timeout + self.ignore_ipolicy = ignore_ipolicy def CheckPrereq(self): """Check prerequisites. @@ -7376,6 +7755,7 @@ class TLMigrateInstance(Tasklet): instance = self.cfg.GetInstanceInfo(instance_name) assert instance is not None self.instance = instance + cluster = self.cfg.GetClusterInfo() if (not self.cleanup and not instance.admin_state == constants.ADMINST_UP and @@ -7403,6 +7783,13 @@ class TLMigrateInstance(Tasklet): # BuildHooksEnv self.target_node = self.lu.op.target_node + # Check that the target node is correct in terms of instance policy + nodeinfo = self.cfg.GetNodeInfo(self.target_node) + group_info = self.cfg.GetNodeGroup(nodeinfo.group) + ipolicy = _CalculateGroupIPolicy(cluster, group_info) + _CheckTargetNodeIPolicy(self.lu, ipolicy, instance, nodeinfo, + ignore=self.ignore_ipolicy) + # self.target_node is already populated, either directly or by the # iallocator run target_node = self.target_node @@ -7436,11 +7823,17 @@ class TLMigrateInstance(Tasklet): " node can be passed)" % (instance.disk_template, text), errors.ECODE_INVAL) + nodeinfo = self.cfg.GetNodeInfo(target_node) + group_info = self.cfg.GetNodeGroup(nodeinfo.group) + ipolicy = _CalculateGroupIPolicy(cluster, group_info) + _CheckTargetNodeIPolicy(self.lu, ipolicy, instance, nodeinfo, + ignore=self.ignore_ipolicy) - i_be = self.cfg.GetClusterInfo().FillBE(instance) + i_be = cluster.FillBE(instance) # check memory requirements on the secondary node - if not self.failover or instance.admin_state == constants.ADMINST_UP: + if (not self.cleanup and + (not self.failover or instance.admin_state == constants.ADMINST_UP)): _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" % instance.name, i_be[constants.BE_MAXMEM], instance.hypervisor) @@ -7448,6 +7841,18 @@ class TLMigrateInstance(Tasklet): self.lu.LogInfo("Not checking memory on the secondary node as" " instance will not be started") + # check if failover must be forced instead of migration + if (not self.cleanup and not self.failover and + i_be[constants.BE_ALWAYS_FAILOVER]): + if self.fallback: + self.lu.LogInfo("Instance configured to always failover; fallback" + " to failover") + self.failover = True + else: + raise errors.OpPrereqError("This instance has been configured to" + " always failover, please allow failover", + errors.ECODE_STATE) + # check bridge existance _CheckInstanceBridgesExist(self.lu, instance, node=target_node) @@ -7481,8 +7886,7 @@ class TLMigrateInstance(Tasklet): self.lu.op.live = None elif self.lu.op.mode is None: # read the default value from the hypervisor - i_hv = self.cfg.GetClusterInfo().FillHV(self.instance, - skip_globals=False) + i_hv = cluster.FillHV(self.instance, skip_globals=False) self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE] self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE @@ -7494,6 +7898,7 @@ class TLMigrateInstance(Tasklet): """Run the allocator based on input opcode. """ + # FIXME: add a self.ignore_ipolicy option ial = IAllocator(self.cfg, self.rpc, mode=constants.IALLOCATOR_MODE_RELOC, name=self.instance_name, @@ -7845,6 +8250,21 @@ class TLMigrateInstance(Tasklet): self._GoReconnect(False) self._WaitUntilSync() + # If the instance's disk template is `rbd' and there was a successful + # migration, unmap the device from the source node. + if self.instance.disk_template == constants.DT_RBD: + disks = _ExpandCheckDisks(instance, instance.disks) + self.feedback_fn("* unmapping instance's disks from %s" % source_node) + for disk in disks: + result = self.rpc.call_blockdev_shutdown(source_node, disk) + msg = result.fail_msg + if msg: + logging.error("Migration was successful, but couldn't unmap the" + " block device %s on source node %s: %s", + disk.iv_name, source_node, msg) + logging.error("You need to unmap the device %s manually on %s", + disk.iv_name, source_node) + self.feedback_fn("* done") def _ExecFailover(self): @@ -8058,9 +8478,18 @@ def _ComputeLDParams(disk_template, disk_params): dt_params = disk_params[disk_template] if disk_template == constants.DT_DRBD8: drbd_params = { - constants.RESYNC_RATE: dt_params[constants.DRBD_RESYNC_RATE], - constants.BARRIERS: dt_params[constants.DRBD_DISK_BARRIERS], - constants.NO_META_FLUSH: dt_params[constants.DRBD_META_BARRIERS], + constants.LDP_RESYNC_RATE: dt_params[constants.DRBD_RESYNC_RATE], + constants.LDP_BARRIERS: dt_params[constants.DRBD_DISK_BARRIERS], + constants.LDP_NO_META_FLUSH: dt_params[constants.DRBD_META_BARRIERS], + constants.LDP_DEFAULT_METAVG: dt_params[constants.DRBD_DEFAULT_METAVG], + constants.LDP_DISK_CUSTOM: dt_params[constants.DRBD_DISK_CUSTOM], + constants.LDP_NET_CUSTOM: dt_params[constants.DRBD_NET_CUSTOM], + constants.LDP_DYNAMIC_RESYNC: dt_params[constants.DRBD_DYNAMIC_RESYNC], + constants.LDP_PLAN_AHEAD: dt_params[constants.DRBD_PLAN_AHEAD], + constants.LDP_FILL_TARGET: dt_params[constants.DRBD_FILL_TARGET], + constants.LDP_DELAY_TARGET: dt_params[constants.DRBD_DELAY_TARGET], + constants.LDP_MAX_RATE: dt_params[constants.DRBD_MAX_RATE], + constants.LDP_MIN_RATE: dt_params[constants.DRBD_MIN_RATE], } drbd_params = \ @@ -8071,7 +8500,7 @@ def _ComputeLDParams(disk_template, disk_params): # data LV data_params = { - constants.STRIPES: dt_params[constants.DRBD_DATA_STRIPES], + constants.LDP_STRIPES: dt_params[constants.DRBD_DATA_STRIPES], } data_params = \ objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV], @@ -8080,7 +8509,7 @@ def _ComputeLDParams(disk_template, disk_params): # metadata LV meta_params = { - constants.STRIPES: dt_params[constants.DRBD_META_STRIPES], + constants.LDP_STRIPES: dt_params[constants.DRBD_META_STRIPES], } meta_params = \ objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV], @@ -8093,7 +8522,7 @@ def _ComputeLDParams(disk_template, disk_params): elif disk_template == constants.DT_PLAIN: params = { - constants.STRIPES: dt_params[constants.LV_STRIPES], + constants.LDP_STRIPES: dt_params[constants.LV_STRIPES], } params = \ objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV], @@ -8103,6 +8532,15 @@ def _ComputeLDParams(disk_template, disk_params): elif disk_template == constants.DT_BLOCK: result.append(constants.DISK_LD_DEFAULTS[constants.LD_BLOCKDEV]) + elif disk_template == constants.DT_RBD: + params = { + constants.LDP_POOL: dt_params[constants.RBD_POOL] + } + params = \ + objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_RBD], + params) + result.append(params) + return result @@ -8148,7 +8586,7 @@ def _GenerateDiskTemplate(lu, template_name, if template_name == constants.DT_DISKLESS: pass elif template_name == constants.DT_PLAIN: - if len(secondary_nodes) != 0: + if secondary_nodes: raise errors.ProgrammerError("Wrong template configuration") names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i) @@ -8179,8 +8617,9 @@ def _GenerateDiskTemplate(lu, template_name, names.append(lv_prefix + "_meta") for idx, disk in enumerate(disk_info): disk_index = idx + base_index + drbd_default_metavg = drbd_params[constants.LDP_DEFAULT_METAVG] data_vg = disk.get(constants.IDISK_VG, vgname) - meta_vg = disk.get(constants.IDISK_METAVG, data_vg) + meta_vg = disk.get(constants.IDISK_METAVG, drbd_default_metavg) disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node, disk[constants.IDISK_SIZE], [data_vg, meta_vg], @@ -8191,7 +8630,7 @@ def _GenerateDiskTemplate(lu, template_name, disk_dev.mode = disk[constants.IDISK_MODE] disks.append(disk_dev) elif template_name == constants.DT_FILE: - if len(secondary_nodes) != 0: + if secondary_nodes: raise errors.ProgrammerError("Wrong template configuration") opcodes.RequireFileStorage() @@ -8208,7 +8647,7 @@ def _GenerateDiskTemplate(lu, template_name, params=ld_params[0]) disks.append(disk_dev) elif template_name == constants.DT_SHARED_FILE: - if len(secondary_nodes) != 0: + if secondary_nodes: raise errors.ProgrammerError("Wrong template configuration") opcodes.RequireSharedFileStorage() @@ -8225,7 +8664,7 @@ def _GenerateDiskTemplate(lu, template_name, params=ld_params[0]) disks.append(disk_dev) elif template_name == constants.DT_BLOCK: - if len(secondary_nodes) != 0: + if secondary_nodes: raise errors.ProgrammerError("Wrong template configuration") for idx, disk in enumerate(disk_info): @@ -8238,6 +8677,22 @@ def _GenerateDiskTemplate(lu, template_name, mode=disk[constants.IDISK_MODE], params=ld_params[0]) disks.append(disk_dev) + elif template_name == constants.DT_RBD: + if secondary_nodes: + raise errors.ProgrammerError("Wrong template configuration") + + names = _GenerateUniqueNames(lu, [".rbd.disk%d" % (base_index + i) + for i in range(disk_count)]) + + for idx, disk in enumerate(disk_info): + disk_index = idx + base_index + disk_dev = objects.Disk(dev_type=constants.LD_RBD, + size=disk[constants.IDISK_SIZE], + logical_id=("rbd", names[idx]), + iv_name="disk/%d" % disk_index, + mode=disk[constants.IDISK_MODE], + params=ld_params[0]) + disks.append(disk_dev) else: raise errors.ProgrammerError("Invalid disk template '%s'" % template_name) @@ -8478,6 +8933,7 @@ def _ComputeDiskSize(disk_template, disks): constants.DT_FILE: None, constants.DT_SHARED_FILE: 0, constants.DT_BLOCK: 0, + constants.DT_RBD: 0, } if disk_template not in req_size_dict: @@ -9049,7 +9505,7 @@ class LUInstanceCreate(LogicalUnit): # pylint: disable=W0142 self.instance_file_storage_dir = utils.PathJoin(*joinargs) - def CheckPrereq(self): + def CheckPrereq(self): # pylint: disable=R0914 """Check prerequisites. """ @@ -9190,8 +9646,9 @@ class LUInstanceCreate(LogicalUnit): constants.IDISK_SIZE: size, constants.IDISK_MODE: mode, constants.IDISK_VG: data_vg, - constants.IDISK_METAVG: disk.get(constants.IDISK_METAVG, data_vg), } + if constants.IDISK_METAVG in disk: + new_disk[constants.IDISK_METAVG] = disk[constants.IDISK_METAVG] if constants.IDISK_ADOPT in disk: new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT] self.disks.append(new_disk) @@ -9247,6 +9704,9 @@ class LUInstanceCreate(LogicalUnit): _ReleaseLocks(self, locking.LEVEL_NODE, keep=filter(None, [self.op.pnode, self.op.snode, self.op.src_node])) + _ReleaseLocks(self, locking.LEVEL_NODE_RES, + keep=filter(None, [self.op.pnode, self.op.snode, + self.op.src_node])) #### node related checks @@ -9285,14 +9745,38 @@ class LUInstanceCreate(LogicalUnit): nodenames = [pnode.name] + self.secondaries + # Verify instance specs + ispec = { + constants.ISPEC_MEM_SIZE: self.be_full.get(constants.BE_MAXMEM, None), + constants.ISPEC_CPU_COUNT: self.be_full.get(constants.BE_VCPUS, None), + constants.ISPEC_DISK_COUNT: len(self.disks), + constants.ISPEC_DISK_SIZE: [disk["size"] for disk in self.disks], + constants.ISPEC_NIC_COUNT: len(self.nics), + } + + group_info = self.cfg.GetNodeGroup(pnode.group) + ipolicy = _CalculateGroupIPolicy(cluster, group_info) + res = _ComputeIPolicyInstanceSpecViolation(ipolicy, ispec) + if not self.op.ignore_ipolicy and res: + raise errors.OpPrereqError(("Instance allocation to group %s violates" + " policy: %s") % (pnode.group, + utils.CommaJoin(res)), + errors.ECODE_INVAL) + # disk parameters (not customizable at instance or node level) # just use the primary node parameters, ignoring the secondary. - self.diskparams = self.cfg.GetNodeGroup(pnode.group).diskparams + self.diskparams = group_info.diskparams if not self.adopt_disks: - # Check lv size requirements, if not adopting - req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks) - _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes) + if self.op.disk_template == constants.DT_RBD: + # _CheckRADOSFreeSpace() is just a placeholder. + # Any function that checks prerequisites can be placed here. + # Check if there is enough space on the RADOS cluster. + _CheckRADOSFreeSpace() + else: + # Check lv size requirements, if not adopting + req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks) + _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes) elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data all_lvs = set(["%s/%s" % (disk[constants.IDISK_VG], @@ -9605,6 +10089,14 @@ class LUInstanceCreate(LogicalUnit): return list(iobj.all_nodes) +def _CheckRADOSFreeSpace(): + """Compute disk size requirements inside the RADOS cluster. + + """ + # For the RADOS cluster we assume there is always enough space. + pass + + class LUInstanceConsole(NoHooksLU): """Connect to an instance's console. @@ -9720,7 +10212,8 @@ class LUInstanceReplaceDisks(LogicalUnit): self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode, self.op.iallocator, self.op.remote_node, - self.op.disks, False, self.op.early_release) + self.op.disks, False, self.op.early_release, + self.op.ignore_ipolicy) self.tasklets = [self.replacer] @@ -9802,7 +10295,7 @@ class TLReplaceDisks(Tasklet): """ def __init__(self, lu, instance_name, mode, iallocator_name, remote_node, - disks, delay_iallocator, early_release): + disks, delay_iallocator, early_release, ignore_ipolicy): """Initializes this class. """ @@ -9816,6 +10309,7 @@ class TLReplaceDisks(Tasklet): self.disks = disks self.delay_iallocator = delay_iallocator self.early_release = early_release + self.ignore_ipolicy = ignore_ipolicy # Runtime data self.instance = None @@ -10038,6 +10532,16 @@ class TLReplaceDisks(Tasklet): if not self.disks: self.disks = range(len(self.instance.disks)) + # TODO: This is ugly, but right now we can't distinguish between internal + # submitted opcode and external one. We should fix that. + if self.remote_node_info: + # We change the node, lets verify it still meets instance policy + new_group_info = self.cfg.GetNodeGroup(self.remote_node_info.group) + ipolicy = _CalculateGroupIPolicy(self.cfg.GetClusterInfo(), + new_group_info) + _CheckTargetNodeIPolicy(self, ipolicy, instance, self.remote_node_info, + ignore=self.ignore_ipolicy) + # TODO: compute disk parameters primary_node_info = self.cfg.GetNodeInfo(instance.primary_node) secondary_node_info = self.cfg.GetNodeInfo(secondary_node) @@ -10895,6 +11399,7 @@ class LUInstanceGrowDisk(LogicalUnit): self._ExpandAndLockInstance() self.needed_locks[locking.LEVEL_NODE] = [] self.needed_locks[locking.LEVEL_NODE_RES] = [] + self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE def DeclareLocks(self, level): @@ -10947,7 +11452,8 @@ class LUInstanceGrowDisk(LogicalUnit): self.disk = instance.FindDisk(self.op.disk) if instance.disk_template not in (constants.DT_FILE, - constants.DT_SHARED_FILE): + constants.DT_SHARED_FILE, + constants.DT_RBD): # TODO: check the free disk space for file, when that feature will be # supported _CheckNodesFreeDiskPerVG(self, nodenames, @@ -11194,7 +11700,8 @@ class LUInstanceSetParams(LogicalUnit): def CheckArguments(self): if not (self.op.nics or self.op.disks or self.op.disk_template or self.op.hvparams or self.op.beparams or self.op.os_name or - self.op.online_inst or self.op.offline_inst): + self.op.online_inst or self.op.offline_inst or + self.op.runtime_mem): raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL) if self.op.hvparams: @@ -11378,6 +11885,8 @@ class LUInstanceSetParams(LogicalUnit): env = _BuildInstanceHookEnvByObject(self, self.instance, override=args) if self.op.disk_template: env["NEW_DISK_TEMPLATE"] = self.op.disk_template + if self.op.runtime_mem: + env["RUNTIME_MEMORY"] = self.op.runtime_mem return env @@ -11442,6 +11951,10 @@ class LUInstanceSetParams(LogicalUnit): _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required) snode_info = self.cfg.GetNodeInfo(self.op.remote_node) + snode_group = self.cfg.GetNodeGroup(snode_info.group) + ipolicy = _CalculateGroupIPolicy(cluster, snode_group) + _CheckTargetNodeIPolicy(self, ipolicy, instance, snode_info, + ignore=self.op.ignore_ipolicy) if pnode_info.group != snode_info.group: self.LogWarning("The primary and secondary nodes are in two" " different node groups; the disk parameters" @@ -11578,6 +12091,33 @@ class LUInstanceSetParams(LogicalUnit): " %s, due to not enough memory" % node, errors.ECODE_STATE) + if self.op.runtime_mem: + remote_info = self.rpc.call_instance_info(instance.primary_node, + instance.name, + instance.hypervisor) + remote_info.Raise("Error checking node %s" % instance.primary_node) + if not remote_info.payload: # not running already + raise errors.OpPrereqError("Instance %s is not running" % instance.name, + errors.ECODE_STATE) + + current_memory = remote_info.payload["memory"] + if (not self.op.force and + (self.op.runtime_mem > self.be_proposed[constants.BE_MAXMEM] or + self.op.runtime_mem < self.be_proposed[constants.BE_MINMEM])): + raise errors.OpPrereqError("Instance %s must have memory between %d" + " and %d MB of memory unless --force is" + " given" % (instance.name, + self.be_proposed[constants.BE_MINMEM], + self.be_proposed[constants.BE_MAXMEM]), + errors.ECODE_INVAL) + + if self.op.runtime_mem > current_memory: + _CheckNodeFreeMemory(self, instance.primary_node, + "ballooning memory for instance %s" % + instance.name, + self.op.memory - current_memory, + instance.hypervisor) + # NIC processing self.nic_pnew = {} self.nic_pinst = {} @@ -11821,6 +12361,15 @@ class LUInstanceSetParams(LogicalUnit): result = [] instance = self.instance + + # runtime memory + if self.op.runtime_mem: + rpcres = self.rpc.call_instance_balloon_memory(instance.primary_node, + instance, + self.op.runtime_mem) + rpcres.Raise("Cannot modify instance runtime memory") + result.append(("runtime_memory", self.op.runtime_mem)) + # disk changes for disk_op, disk_dict in self.op.disks: if disk_op == constants.DDM_REMOVE: @@ -12594,6 +13143,16 @@ class LUGroupAdd(LogicalUnit): if self.op.ndparams: utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES) + if self.op.hv_state: + self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, None) + else: + self.new_hv_state = None + + if self.op.disk_state: + self.new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state, None) + else: + self.new_disk_state = None + if self.op.diskparams: for templ in constants.DISK_TEMPLATES: if templ not in self.op.diskparams: @@ -12602,6 +13161,15 @@ class LUGroupAdd(LogicalUnit): else: self.op.diskparams = self.cfg.GetClusterInfo().diskparams + if self.op.ipolicy: + cluster = self.cfg.GetClusterInfo() + full_ipolicy = cluster.SimpleFillIPolicy(self.op.ipolicy) + try: + objects.InstancePolicy.CheckParameterSyntax(full_ipolicy) + except errors.ConfigurationError, err: + raise errors.OpPrereqError("Invalid instance policy: %s" % err, + errors.ECODE_INVAL) + def BuildHooksEnv(self): """Build hooks env. @@ -12625,7 +13193,10 @@ class LUGroupAdd(LogicalUnit): uuid=self.group_uuid, alloc_policy=self.op.alloc_policy, ndparams=self.op.ndparams, - diskparams=self.op.diskparams) + diskparams=self.op.diskparams, + ipolicy=self.op.ipolicy, + hv_state_static=self.new_hv_state, + disk_state_static=self.new_disk_state) self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False) del self.remove_locks[locking.LEVEL_NODEGROUP] @@ -12771,6 +13342,7 @@ class _GroupQuery(_QueryBase): lu.needed_locks = {} self._all_groups = lu.cfg.GetAllNodeGroupsInfo() + self._cluster = lu.cfg.GetClusterInfo() name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values()) if not self.names: @@ -12836,7 +13408,8 @@ class _GroupQuery(_QueryBase): # Do not pass on node information if it was not requested. group_to_nodes = None - return query.GroupQueryData([self._all_groups[uuid] + return query.GroupQueryData(self._cluster, + [self._all_groups[uuid] for uuid in self.wanted], group_to_nodes, group_to_instances) @@ -12874,6 +13447,9 @@ class LUGroupSetParams(LogicalUnit): self.op.ndparams, self.op.diskparams, self.op.alloc_policy, + self.op.hv_state, + self.op.disk_state, + self.op.ipolicy, ] if all_changes.count(None) == len(all_changes): @@ -12885,14 +13461,32 @@ class LUGroupSetParams(LogicalUnit): self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name) self.needed_locks = { + locking.LEVEL_INSTANCE: [], locking.LEVEL_NODEGROUP: [self.group_uuid], } + self.share_locks[locking.LEVEL_INSTANCE] = 1 + + def DeclareLocks(self, level): + if level == locking.LEVEL_INSTANCE: + assert not self.needed_locks[locking.LEVEL_INSTANCE] + + # Lock instances optimistically, needs verification once group lock has + # been acquired + self.needed_locks[locking.LEVEL_INSTANCE] = \ + self.cfg.GetNodeGroupInstances(self.group_uuid) + def CheckPrereq(self): """Check prerequisites. """ + owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE)) + + # Check if locked instances are still correct + _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances) + self.group = self.cfg.GetNodeGroup(self.group_uuid) + cluster = self.cfg.GetClusterInfo() if self.group is None: raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" % @@ -12913,6 +13507,33 @@ class LUGroupSetParams(LogicalUnit): utils.ForceDictType(new_templ_params, constants.DISK_DT_TYPES) self.new_diskparams[templ] = new_templ_params + if self.op.hv_state: + self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state, + self.group.hv_state_static) + + if self.op.disk_state: + self.new_disk_state = \ + _MergeAndVerifyDiskState(self.op.disk_state, + self.group.disk_state_static) + + if self.op.ipolicy: + self.new_ipolicy = _GetUpdatedIPolicy(self.group.ipolicy, + self.op.ipolicy, + group_policy=True) + + new_ipolicy = cluster.SimpleFillIPolicy(self.new_ipolicy) + inst_filter = lambda inst: inst.name in owned_instances + instances = self.cfg.GetInstancesInfoByFilter(inst_filter).values() + violations = \ + _ComputeNewInstanceViolations(_CalculateGroupIPolicy(cluster, + self.group), + new_ipolicy, instances) + + if violations: + self.LogWarning("After the ipolicy change the following instances" + " violate them: %s", + utils.CommaJoin(violations)) + def BuildHooksEnv(self): """Build hooks env. @@ -12946,6 +13567,15 @@ class LUGroupSetParams(LogicalUnit): if self.op.alloc_policy: self.group.alloc_policy = self.op.alloc_policy + if self.op.hv_state: + self.group.hv_state_static = self.new_hv_state + + if self.op.disk_state: + self.group.disk_state_static = self.new_disk_state + + if self.op.ipolicy: + self.group.ipolicy = self.new_ipolicy + self.cfg.Update(self.group, feedback_fn) return result @@ -13695,9 +14325,11 @@ class IAllocator(object): """Compute node groups data. """ + cluster = cfg.GetClusterInfo() ng = dict((guuid, { "name": gdata.name, "alloc_policy": gdata.alloc_policy, + "ipolicy": _CalculateGroupIPolicy(cluster, gdata), }) for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())