X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/21faa3ad42ab8dcf4e9405df5a40f2babdffa531..d4724b140175e9672becb46a8a92692e33524345:/lib/cmdlib.py diff --git a/lib/cmdlib.py b/lib/cmdlib.py index f105d74..c91c2d1 100644 --- a/lib/cmdlib.py +++ b/lib/cmdlib.py @@ -138,13 +138,18 @@ class LogicalUnit(object): self.owned_locks = context.glm.list_owned self.context = context self.rpc = rpc_runner - # Dicts used to declare locking needs to mcpu + + # Dictionaries used to declare locking needs to mcpu self.needed_locks = None self.share_locks = dict.fromkeys(locking.LEVELS, 0) + self.opportunistic_locks = dict.fromkeys(locking.LEVELS, False) + self.add_locks = {} self.remove_locks = {} + # Used to force good behavior when calling helper functions self.recalculate_locks = {} + # logging self.Log = processor.Log # pylint: disable=C0103 self.LogWarning = processor.LogWarning # pylint: disable=C0103 @@ -692,6 +697,39 @@ def _SupportsOob(cfg, node): return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM] +def _IsExclusiveStorageEnabledNode(cfg, node): + """Whether exclusive_storage is in effect for the given node. + + @type cfg: L{config.ConfigWriter} + @param cfg: The cluster configuration + @type node: L{objects.Node} + @param node: The node + @rtype: bool + @return: The effective value of exclusive_storage + + """ + return cfg.GetNdParams(node)[constants.ND_EXCLUSIVE_STORAGE] + + +def _IsExclusiveStorageEnabledNodeName(cfg, nodename): + """Whether exclusive_storage is in effect for the given node. + + @type cfg: L{config.ConfigWriter} + @param cfg: The cluster configuration + @type nodename: string + @param nodename: The node + @rtype: bool + @return: The effective value of exclusive_storage + @raise errors.OpPrereqError: if no node exists with the given name + + """ + ni = cfg.GetNodeInfo(nodename) + if ni is None: + raise errors.OpPrereqError("Invalid node name %s" % nodename, + errors.ECODE_NOENT) + return _IsExclusiveStorageEnabledNode(cfg, ni) + + def _CopyLockList(names): """Makes a copy of a list of lock names. @@ -959,7 +997,8 @@ def _RunPostHook(lu, node_name): try: hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name]) except Exception, err: # pylint: disable=W0703 - lu.LogWarning("Errors occurred running hooks on %s: %s" % (node_name, err)) + lu.LogWarning("Errors occurred running hooks on %s: %s", + node_name, err) def _CheckOutputFields(static, dynamic, selected): @@ -1100,7 +1139,8 @@ def _CheckInstanceState(lu, instance, req_states, msg=None): """ if msg is None: - msg = "can't use instance from outside %s states" % ", ".join(req_states) + msg = ("can't use instance from outside %s states" % + utils.CommaJoin(req_states)) if instance.admin_state not in req_states: raise errors.OpPrereqError("Instance '%s' is marked to be %s, %s" % (instance.name, instance.admin_state, msg), @@ -1360,27 +1400,6 @@ def _BuildNetworkHookEnv(name, subnet, gateway, network6, gateway6, return env -def _BuildNetworkHookEnvByObject(net): - """Builds network related env varliables for hooks - - @type net: L{objects.Network} - @param net: the network object - - """ - args = { - "name": net.name, - "subnet": net.network, - "gateway": net.gateway, - "network6": net.network6, - "gateway6": net.gateway6, - "network_type": net.network_type, - "mac_prefix": net.mac_prefix, - "tags": net.tags, - } - - return _BuildNetworkHookEnv(**args) # pylint: disable=W0142 - - def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status, minmem, maxmem, vcpus, nics, disk_template, disks, bep, hvp, hypervisor_name, tags): @@ -2226,6 +2245,11 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors): locking.LEVEL_INSTANCE: inst_names, locking.LEVEL_NODEGROUP: [self.group_uuid], locking.LEVEL_NODE: [], + + # This opcode is run by watcher every five minutes and acquires all nodes + # for a group. It doesn't run for a long time, so it's better to acquire + # the node allocation lock as well. + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, } self.share_locks = _ShareAll() @@ -2433,19 +2457,20 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors): constants.MIN_VG_SIZE) _ErrorIf(vgstatus, constants.CV_ENODELVM, node, vgstatus) - # check pv names - pvlist = nresult.get(constants.NV_PVLIST, None) - test = pvlist is None + # check pv names (and possibly sizes) + pvlist_dict = nresult.get(constants.NV_PVLIST, None) + test = pvlist_dict is None _ErrorIf(test, constants.CV_ENODELVM, node, "Can't get PV list from node") if not test: + pvlist = map(objects.LvmPvInfo.FromDict, pvlist_dict) # check that ':' is not present in PV names, since it's a # special character for lvcreate (denotes the range of PEs to # use on the PV) - for _, pvname, owner_vg in pvlist: - test = ":" in pvname + for pv in pvlist: + test = ":" in pv.name _ErrorIf(test, constants.CV_ENODELVM, node, "Invalid character ':' in PV '%s' of VG '%s'", - pvname, owner_vg) + pv.name, pv.vg_name) def _VerifyNodeBridges(self, ninfo, nresult, bridges): """Check the node bridges. @@ -3673,6 +3698,12 @@ class LUGroupVerifyDisks(NoHooksLU): locking.LEVEL_INSTANCE: [], locking.LEVEL_NODEGROUP: [], locking.LEVEL_NODE: [], + + # This opcode is acquires all node locks in a group. LUClusterVerifyDisks + # starts one instance of this opcode for every group, which means all + # nodes will be locked for a short amount of time, so it's better to + # acquire the node allocation lock as well. + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, } def DeclareLocks(self, level): @@ -3779,6 +3810,8 @@ class LUClusterRepairDiskSizes(NoHooksLU): def ExpandNames(self): if self.op.instances: self.wanted_names = _GetWantedInstances(self, self.op.instances) + # Not getting the node allocation lock as only a specific set of + # instances (and their nodes) is going to be acquired self.needed_locks = { locking.LEVEL_NODE_RES: [], locking.LEVEL_INSTANCE: self.wanted_names, @@ -3789,10 +3822,15 @@ class LUClusterRepairDiskSizes(NoHooksLU): self.needed_locks = { locking.LEVEL_NODE_RES: locking.ALL_SET, locking.LEVEL_INSTANCE: locking.ALL_SET, + + # This opcode is acquires the node locks for all instances + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, } + self.share_locks = { locking.LEVEL_NODE_RES: 1, locking.LEVEL_INSTANCE: 0, + locking.LEVEL_NODE_ALLOC: 1, } def DeclareLocks(self, level): @@ -4034,16 +4072,15 @@ class LUClusterSetParams(LogicalUnit): def ExpandNames(self): # FIXME: in the future maybe other cluster params won't require checking on # all nodes to be modified. + # FIXME: This opcode changes cluster-wide settings. Is acquiring all + # resource locks the right thing, shouldn't it be the BGL instead? self.needed_locks = { locking.LEVEL_NODE: locking.ALL_SET, locking.LEVEL_INSTANCE: locking.ALL_SET, locking.LEVEL_NODEGROUP: locking.ALL_SET, + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, } - self.share_locks = { - locking.LEVEL_NODE: 1, - locking.LEVEL_INSTANCE: 1, - locking.LEVEL_NODEGROUP: 1, - } + self.share_locks = _ShareAll() def BuildHooksEnv(self): """Build hooks env. @@ -4448,7 +4485,7 @@ def _UploadHelper(lu, nodes, fname): if msg: msg = ("Copy of file %s to node %s failed: %s" % (fname, to_node, msg)) - lu.proc.LogWarning(msg) + lu.LogWarning(msg) def _ComputeAncillaryFiles(cluster, redist): @@ -4590,8 +4627,9 @@ class LUClusterRedistConf(NoHooksLU): def ExpandNames(self): self.needed_locks = { locking.LEVEL_NODE: locking.ALL_SET, + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, } - self.share_locks[locking.LEVEL_NODE] = 1 + self.share_locks = _ShareAll() def Exec(self, feedback_fn): """Redistribute the configuration. @@ -4641,7 +4679,7 @@ def _WaitForSync(lu, instance, disks=None, oneshot=False): disks = _ExpandCheckDisks(instance, disks) if not oneshot: - lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name) + lu.LogInfo("Waiting for instance %s to sync disks", instance.name) node = instance.primary_node @@ -4684,8 +4722,8 @@ def _WaitForSync(lu, instance, disks=None, oneshot=False): max_time = mstat.estimated_time else: rem_time = "no time estimate" - lu.proc.LogInfo("- device %s: %5.2f%% done, %s" % - (disks[i].iv_name, mstat.sync_percent, rem_time)) + lu.LogInfo("- device %s: %5.2f%% done, %s", + disks[i].iv_name, mstat.sync_percent, rem_time) # if we're done but degraded, let's do a few small retries, to # make sure we see a stable and not transient situation; therefore @@ -4702,7 +4740,8 @@ def _WaitForSync(lu, instance, disks=None, oneshot=False): time.sleep(min(60, max_time)) if done: - lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name) + lu.LogInfo("Instance %s's disks are in sync", instance.name) + return not cumul_degraded @@ -4788,6 +4827,12 @@ class LUOobCommand(NoHooksLU): locking.LEVEL_NODE: lock_names, } + self.share_locks[locking.LEVEL_NODE_ALLOC] = 1 + + if not self.op.node_names: + # Acquire node allocation lock only if all nodes are affected + self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET + def CheckPrereq(self): """Check prerequisites. @@ -5220,6 +5265,7 @@ class _NodeQuery(_QueryBase): if self.do_locking: # If any non-static field is requested we need to lock the nodes lu.needed_locks[locking.LEVEL_NODE] = self.wanted + lu.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET def DeclareLocks(self, lu, level): pass @@ -5314,13 +5360,16 @@ class LUNodeQueryvols(NoHooksLU): def ExpandNames(self): self.share_locks = _ShareAll() - self.needed_locks = {} - if not self.op.nodes: - self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET + if self.op.nodes: + self.needed_locks = { + locking.LEVEL_NODE: _GetWantedNodes(self, self.op.nodes), + } else: - self.needed_locks[locking.LEVEL_NODE] = \ - _GetWantedNodes(self, self.op.nodes) + self.needed_locks = { + locking.LEVEL_NODE: locking.ALL_SET, + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, + } def Exec(self, feedback_fn): """Computes the list of nodes and their attributes. @@ -5383,13 +5432,16 @@ class LUNodeQueryStorage(NoHooksLU): def ExpandNames(self): self.share_locks = _ShareAll() - self.needed_locks = {} if self.op.nodes: - self.needed_locks[locking.LEVEL_NODE] = \ - _GetWantedNodes(self, self.op.nodes) + self.needed_locks = { + locking.LEVEL_NODE: _GetWantedNodes(self, self.op.nodes), + } else: - self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET + self.needed_locks = { + locking.LEVEL_NODE: locking.ALL_SET, + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, + } def Exec(self, feedback_fn): """Computes the list of nodes and their attributes. @@ -6015,19 +6067,28 @@ class LUNodeSetParams(LogicalUnit): def ExpandNames(self): if self.lock_all: - self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET} + self.needed_locks = { + locking.LEVEL_NODE: locking.ALL_SET, + + # Block allocations when all nodes are locked + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, + } else: - self.needed_locks = {locking.LEVEL_NODE: self.op.node_name} + self.needed_locks = { + locking.LEVEL_NODE: self.op.node_name, + } # Since modifying a node can have severe effects on currently running # operations the resource lock is at least acquired in shared mode self.needed_locks[locking.LEVEL_NODE_RES] = \ self.needed_locks[locking.LEVEL_NODE] - # Get node resource and instance locks in shared mode; they are not used - # for anything but read-only access - self.share_locks[locking.LEVEL_NODE_RES] = 1 - self.share_locks[locking.LEVEL_INSTANCE] = 1 + # Get all locks except nodes in shared mode; they are not used for anything + # but read-only access + self.share_locks = _ShareAll() + self.share_locks[locking.LEVEL_NODE] = 0 + self.share_locks[locking.LEVEL_NODE_RES] = 0 + self.share_locks[locking.LEVEL_NODE_ALLOC] = 0 if self.lock_instances: self.needed_locks[locking.LEVEL_INSTANCE] = \ @@ -6194,7 +6255,8 @@ class LUNodeSetParams(LogicalUnit): if master_singlehomed and self.op.secondary_ip != node.primary_ip: if self.op.force and node.name == master.name: self.LogWarning("Transitioning from single-homed to multi-homed" - " cluster. All nodes will require a secondary ip.") + " cluster; all nodes will require a secondary IP" + " address") else: raise errors.OpPrereqError("Changing the secondary ip on a" " single-homed cluster requires the" @@ -6204,7 +6266,8 @@ class LUNodeSetParams(LogicalUnit): elif not master_singlehomed and self.op.secondary_ip == node.primary_ip: if self.op.force and node.name == master.name: self.LogWarning("Transitioning from multi-homed to single-homed" - " cluster. Secondary IPs will have to be removed.") + " cluster; secondary IP addresses will have to be" + " removed") else: raise errors.OpPrereqError("Cannot set the secondary IP to be the" " same as the primary IP on a multi-homed" @@ -6481,7 +6544,13 @@ class _ClusterQuery(_QueryBase): drain_flag = NotImplemented if query.CQ_WATCHER_PAUSE in self.requested_data: - watcher_pause = utils.ReadWatcherPauseFile(pathutils.WATCHER_PAUSEFILE) + master_name = lu.cfg.GetMasterNode() + + result = lu.rpc.call_get_watcher_pause(master_name) + result.Raise("Can't retrieve watcher pause from master node '%s'" % + master_name) + + watcher_pause = result.payload else: watcher_pause = NotImplemented @@ -6582,9 +6651,9 @@ def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False, if msg: is_offline_secondary = (node in instance.secondary_nodes and result.offline) - lu.proc.LogWarning("Could not prepare block device %s on node %s" - " (is_primary=False, pass=1): %s", - inst_disk.iv_name, node, msg) + lu.LogWarning("Could not prepare block device %s on node %s" + " (is_primary=False, pass=1): %s", + inst_disk.iv_name, node, msg) if not (ignore_secondaries or is_offline_secondary): disks_ok = False @@ -6605,9 +6674,9 @@ def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False, True, idx) msg = result.fail_msg if msg: - lu.proc.LogWarning("Could not prepare block device %s on node %s" - " (is_primary=True, pass=2): %s", - inst_disk.iv_name, node, msg) + lu.LogWarning("Could not prepare block device %s on node %s" + " (is_primary=True, pass=2): %s", + inst_disk.iv_name, node, msg) disks_ok = False else: dev_path = result.payload @@ -6632,9 +6701,9 @@ def _StartInstanceDisks(lu, instance, force): if not disks_ok: _ShutdownInstanceDisks(lu, instance) if force is not None and not force: - lu.proc.LogWarning("", hint="If the message above refers to a" - " secondary node," - " you can retry the operation using '--force'.") + lu.LogWarning("", + hint=("If the message above refers to a secondary node," + " you can retry the operation using '--force'")) raise errors.OpExecError("Disk consistency error") @@ -6732,9 +6801,9 @@ def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False): def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name): """Checks if a node has enough free memory. - This function check if a given node has the needed amount of free + This function checks if a given node has the needed amount of free memory. In case the node has less memory or we cannot get the - information from the node, this function raise an OpPrereqError + information from the node, this function raises an OpPrereqError exception. @type lu: C{LogicalUnit} @@ -6772,11 +6841,11 @@ def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name): def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes): - """Checks if nodes have enough free disk space in the all VGs. + """Checks if nodes have enough free disk space in all the VGs. - This function check if all given nodes have the needed amount of + This function checks if all given nodes have the needed amount of free disk. In case any node has less disk or we cannot get the - information from the node, this function raise an OpPrereqError + information from the node, this function raises an OpPrereqError exception. @type lu: C{LogicalUnit} @@ -6797,9 +6866,9 @@ def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes): def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested): """Checks if nodes have enough free disk space in the specified VG. - This function check if all given nodes have the needed amount of + This function checks if all given nodes have the needed amount of free disk. In case any node has less disk or we cannot get the - information from the node, this function raise an OpPrereqError + information from the node, this function raises an OpPrereqError exception. @type lu: C{LogicalUnit} @@ -6937,10 +7006,10 @@ class LUInstanceStartup(LogicalUnit): self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline if self.primary_offline and self.op.ignore_offline_nodes: - self.proc.LogWarning("Ignoring offline primary node") + self.LogWarning("Ignoring offline primary node") if self.op.hvparams or self.op.beparams: - self.proc.LogWarning("Overridden parameters are ignored") + self.LogWarning("Overridden parameters are ignored") else: _CheckNodeOnline(self, instance.primary_node) @@ -6972,7 +7041,7 @@ class LUInstanceStartup(LogicalUnit): if self.primary_offline: assert self.op.ignore_offline_nodes - self.proc.LogInfo("Primary node offline, marked instance as started") + self.LogInfo("Primary node offline, marked instance as started") else: node_current = instance.primary_node @@ -7121,13 +7190,16 @@ class LUInstanceShutdown(LogicalUnit): assert self.instance is not None, \ "Cannot retrieve locked instance %s" % self.op.instance_name - _CheckInstanceState(self, self.instance, INSTANCE_ONLINE) + if not self.op.force: + _CheckInstanceState(self, self.instance, INSTANCE_ONLINE) + else: + self.LogWarning("Ignoring offline instance check") self.primary_offline = \ self.cfg.GetNodeInfo(self.instance.primary_node).offline if self.primary_offline and self.op.ignore_offline_nodes: - self.proc.LogWarning("Ignoring offline primary node") + self.LogWarning("Ignoring offline primary node") else: _CheckNodeOnline(self, self.instance.primary_node) @@ -7139,17 +7211,19 @@ class LUInstanceShutdown(LogicalUnit): node_current = instance.primary_node timeout = self.op.timeout - if not self.op.no_remember: + # If the instance is offline we shouldn't mark it as down, as that + # resets the offline flag. + if not self.op.no_remember and instance.admin_state in INSTANCE_ONLINE: self.cfg.MarkInstanceDown(instance.name) if self.primary_offline: assert self.op.ignore_offline_nodes - self.proc.LogInfo("Primary node offline, marked instance as stopped") + self.LogInfo("Primary node offline, marked instance as stopped") else: result = self.rpc.call_instance_shutdown(node_current, instance, timeout) msg = result.fail_msg if msg: - self.proc.LogWarning("Could not shutdown instance: %s" % msg) + self.LogWarning("Could not shutdown instance: %s", msg) _ShutdownInstanceDisks(self, instance) @@ -7250,7 +7324,7 @@ class LUInstanceRecreateDisks(LogicalUnit): HTYPE = constants.HTYPE_INSTANCE REQ_BGL = False - _MODIFYABLE = frozenset([ + _MODIFYABLE = compat.UniqueFrozenset([ constants.IDISK_SIZE, constants.IDISK_MODE, ]) @@ -7262,6 +7336,7 @@ class LUInstanceRecreateDisks(LogicalUnit): # TODO: Implement support changing VG while recreating constants.IDISK_VG, constants.IDISK_METAVG, + constants.IDISK_PROVIDER, ])) def _RunAllocator(self): @@ -7345,6 +7420,7 @@ class LUInstanceRecreateDisks(LogicalUnit): def ExpandNames(self): self._ExpandAndLockInstance() self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND + if self.op.nodes: self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes] self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes) @@ -7353,6 +7429,8 @@ class LUInstanceRecreateDisks(LogicalUnit): if self.op.iallocator: # iallocator will select a new node in the same group self.needed_locks[locking.LEVEL_NODEGROUP] = [] + self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET + self.needed_locks[locking.LEVEL_NODE_RES] = [] def DeclareLocks(self, level): @@ -7382,6 +7460,8 @@ class LUInstanceRecreateDisks(LogicalUnit): for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP): self.needed_locks[locking.LEVEL_NODE].extend( self.cfg.GetNodeGroup(group_uuid).members) + + assert locking.NAL in self.owned_locks(locking.LEVEL_NODE_ALLOC) elif not self.op.nodes: self._LockInstancesNodes(primary_only=False) elif level == locking.LEVEL_NODE_RES: @@ -7472,6 +7552,9 @@ class LUInstanceRecreateDisks(LogicalUnit): # Release unneeded node and node resource locks _ReleaseLocks(self, locking.LEVEL_NODE, keep=self.op.nodes) _ReleaseLocks(self, locking.LEVEL_NODE_RES, keep=self.op.nodes) + _ReleaseLocks(self, locking.LEVEL_NODE_ALLOC) + + assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC) def Exec(self, feedback_fn): """Recreate the disks. @@ -7615,6 +7698,7 @@ class LUInstanceRename(LogicalUnit): # Change the instance lock. This is definitely safe while we hold the BGL. # Otherwise the new lock would have to be added in acquired mode. assert self.REQ_BGL + assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER) self.glm.remove(locking.LEVEL_INSTANCE, old_name) self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name) @@ -7649,7 +7733,7 @@ class LUInstanceRename(LogicalUnit): msg = ("Could not run OS rename script for instance %s on node %s" " (but the instance has been renamed in Ganeti): %s" % (inst.name, inst.primary_node, msg)) - self.proc.LogWarning(msg) + self.LogWarning(msg) finally: _ShutdownInstanceDisks(self, inst) @@ -7792,6 +7876,10 @@ def _ExpandNamesForMigration(lu): lu.needed_locks[locking.LEVEL_NODE_RES] = [] lu.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE + # The node allocation lock is actually only needed for replicated instances + # (e.g. DRBD8) and if an iallocator is used. + lu.needed_locks[locking.LEVEL_NODE_ALLOC] = [] + def _DeclareLocksForMigration(lu, level): """Declares locks for L{TLMigrateInstance}. @@ -7800,17 +7888,29 @@ def _DeclareLocksForMigration(lu, level): @param level: Lock level """ - if level == locking.LEVEL_NODE: + if level == locking.LEVEL_NODE_ALLOC: + assert lu.op.instance_name in lu.owned_locks(locking.LEVEL_INSTANCE) + instance = lu.cfg.GetInstanceInfo(lu.op.instance_name) + + # Node locks are already declared here rather than at LEVEL_NODE as we need + # the instance object anyway to declare the node allocation lock. if instance.disk_template in constants.DTS_EXT_MIRROR: if lu.op.target_node is None: lu.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET + lu.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET else: lu.needed_locks[locking.LEVEL_NODE] = [instance.primary_node, lu.op.target_node] del lu.recalculate_locks[locking.LEVEL_NODE] else: lu._LockInstancesNodes() # pylint: disable=W0212 + + elif level == locking.LEVEL_NODE: + # Node locks are declared together with the node allocation lock + assert (lu.needed_locks[locking.LEVEL_NODE] or + lu.needed_locks[locking.LEVEL_NODE] is locking.ALL_SET) + elif level == locking.LEVEL_NODE_RES: # Copy node locks lu.needed_locks[locking.LEVEL_NODE_RES] = \ @@ -8063,10 +8163,10 @@ class LUInstanceMove(LogicalUnit): msg = result.fail_msg if msg: if self.op.ignore_consistency: - self.proc.LogWarning("Could not shutdown instance %s on node %s." - " Proceeding anyway. Please make sure node" - " %s is down. Error details: %s", - instance.name, source_node, source_node, msg) + self.LogWarning("Could not shutdown instance %s on node %s." + " Proceeding anyway. Please make sure node" + " %s is down. Error details: %s", + instance.name, source_node, source_node, msg) else: raise errors.OpExecError("Could not shutdown instance %s on" " node %s: %s" % @@ -8284,6 +8384,8 @@ class TLMigrateInstance(Tasklet): errors.ECODE_STATE) if instance.disk_template in constants.DTS_EXT_MIRROR: + assert locking.NAL in self.lu.owned_locks(locking.LEVEL_NODE_ALLOC) + _CheckIAllocatorOrNode(self.lu, "iallocator", "target_node") if self.lu.op.iallocator: @@ -8315,8 +8417,11 @@ class TLMigrateInstance(Tasklet): # in the LU _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=[instance.primary_node, self.target_node]) + _ReleaseLocks(self.lu, locking.LEVEL_NODE_ALLOC) else: + assert not self.lu.glm.is_owned(locking.LEVEL_NODE_ALLOC) + secondary_nodes = instance.secondary_nodes if not secondary_nodes: raise errors.ConfigurationError("No secondary node but using" @@ -8418,6 +8523,8 @@ class TLMigrateInstance(Tasklet): """Run the allocator based on input opcode. """ + assert locking.NAL in self.lu.owned_locks(locking.LEVEL_NODE_ALLOC) + # FIXME: add a self.ignore_ipolicy option req = iallocator.IAReqRelocate(name=self.instance_name, relocate_from=[self.instance.primary_node]) @@ -8777,9 +8884,9 @@ 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: + # If the instance's disk template is `rbd' or `ext' and there was a + # successful migration, unmap the device from the source node. + if self.instance.disk_template in (constants.DT_RBD, constants.DT_EXT): disks = _ExpandCheckDisks(instance, instance.disks) self.feedback_fn("* unmapping instance's disks from %s" % source_node) for disk in disks: @@ -8910,12 +9017,13 @@ def _CreateBlockDev(lu, node, instance, device, force_create, info, """ (disk,) = _AnnotateDiskParams(instance, [device], lu.cfg) + excl_stor = _IsExclusiveStorageEnabledNodeName(lu.cfg, node) return _CreateBlockDevInner(lu, node, instance, disk, force_create, info, - force_open) + force_open, excl_stor) def _CreateBlockDevInner(lu, node, instance, device, force_create, - info, force_open): + info, force_open, excl_stor): """Create a tree of block devices on a given node. If this device type has to be created on secondaries, create it and @@ -8942,6 +9050,8 @@ def _CreateBlockDevInner(lu, node, instance, device, force_create, L{backend.BlockdevCreate} function where it specifies whether we run on primary or not, and it affects both the child assembly and the device own Open() execution + @type excl_stor: boolean + @param excl_stor: Whether exclusive_storage is active for the node """ if device.CreateOnSecondary(): @@ -8950,15 +9060,17 @@ def _CreateBlockDevInner(lu, node, instance, device, force_create, if device.children: for child in device.children: _CreateBlockDevInner(lu, node, instance, child, force_create, - info, force_open) + info, force_open, excl_stor) if not force_create: return - _CreateSingleBlockDev(lu, node, instance, device, info, force_open) + _CreateSingleBlockDev(lu, node, instance, device, info, force_open, + excl_stor) -def _CreateSingleBlockDev(lu, node, instance, device, info, force_open): +def _CreateSingleBlockDev(lu, node, instance, device, info, force_open, + excl_stor): """Create a single block device on a given node. This will not recurse over children of the device, so they must be @@ -8977,11 +9089,14 @@ def _CreateSingleBlockDev(lu, node, instance, device, info, force_open): L{backend.BlockdevCreate} function where it specifies whether we run on primary or not, and it affects both the child assembly and the device own Open() execution + @type excl_stor: boolean + @param excl_stor: Whether exclusive_storage is active for the node """ lu.cfg.SetDiskID(device, node) result = lu.rpc.call_blockdev_create(node, device, device.size, - instance.name, force_open, info) + instance.name, force_open, info, + excl_stor) result.Raise("Can't create block device %s on" " node %s for instance %s" % (device, node, instance.name)) if device.physical_id is None: @@ -9029,6 +9144,7 @@ def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names, _DISK_TEMPLATE_NAME_PREFIX = { constants.DT_PLAIN: "", constants.DT_RBD: ".rbd", + constants.DT_EXT: ".ext", } @@ -9038,6 +9154,7 @@ _DISK_TEMPLATE_DEVICE_TYPE = { constants.DT_SHARED_FILE: constants.LD_FILE, constants.DT_BLOCK: constants.LD_BLOCKDEV, constants.DT_RBD: constants.LD_RBD, + constants.DT_EXT: constants.LD_EXT, } @@ -9049,8 +9166,6 @@ def _GenerateDiskTemplate( """Generate the entire disk layout for a given template type. """ - #TODO: compute space requirements - vgname = lu.cfg.GetVGName() disk_count = len(disk_info) disks = [] @@ -9119,12 +9234,27 @@ def _GenerateDiskTemplate( disk[constants.IDISK_ADOPT]) elif template_name == constants.DT_RBD: logical_id_fn = lambda idx, _, disk: ("rbd", names[idx]) + elif template_name == constants.DT_EXT: + def logical_id_fn(idx, _, disk): + provider = disk.get(constants.IDISK_PROVIDER, None) + if provider is None: + raise errors.ProgrammerError("Disk template is %s, but '%s' is" + " not found", constants.DT_EXT, + constants.IDISK_PROVIDER) + return (provider, names[idx]) else: raise errors.ProgrammerError("Unknown disk template '%s'" % template_name) dev_type = _DISK_TEMPLATE_DEVICE_TYPE[template_name] for idx, disk in enumerate(disk_info): + params = {} + # Only for the Ext template add disk_info to params + if template_name == constants.DT_EXT: + params[constants.IDISK_PROVIDER] = disk[constants.IDISK_PROVIDER] + for key in disk: + if key not in constants.IDISK_PARAMS: + params[key] = disk[key] disk_index = idx + base_index size = disk[constants.IDISK_SIZE] feedback_fn("* disk %s, size %s" % @@ -9133,7 +9263,7 @@ def _GenerateDiskTemplate( logical_id=logical_id_fn(idx, disk_index, disk), iv_name="disk/%d" % disk_index, mode=disk[constants.IDISK_MODE], - params={})) + params=params)) return disks @@ -9462,13 +9592,15 @@ def _CheckOSParams(lu, required, nodenames, osname, osparams): osname, node) -def _CreateInstanceAllocRequest(op, disks, nics, beparams): +def _CreateInstanceAllocRequest(op, disks, nics, beparams, node_whitelist): """Wrapper around IAReqInstanceAlloc. @param op: The instance opcode @param disks: The computed disks @param nics: The computed nics @param beparams: The full filled beparams + @param node_whitelist: List of nodes which should appear as online to the + allocator (unless the node is already marked offline) @returns: A filled L{iallocator.IAReqInstanceAlloc} @@ -9483,17 +9615,18 @@ def _CreateInstanceAllocRequest(op, disks, nics, beparams): spindle_use=spindle_use, disks=disks, nics=[n.ToDict() for n in nics], - hypervisor=op.hypervisor) + hypervisor=op.hypervisor, + node_whitelist=node_whitelist) -def _ComputeNics(op, cluster, default_ip, cfg, proc): +def _ComputeNics(op, cluster, default_ip, cfg, ec_id): """Computes the nics. @param op: The instance opcode @param cluster: Cluster configuration object @param default_ip: The default ip to assign @param cfg: An instance of the configuration object - @param proc: The executer instance + @param ec_id: Execution context ID @returns: The build up nics @@ -9553,7 +9686,7 @@ def _ComputeNics(op, cluster, default_ip, cfg, proc): try: # TODO: We need to factor this out - cfg.ReserveMAC(mac, proc.GetECId()) + cfg.ReserveMAC(mac, ec_id) except errors.ReservationError: raise errors.OpPrereqError("MAC address %s already in use" " in cluster" % mac, @@ -9580,7 +9713,7 @@ def _ComputeDisks(op, default_vg): @param op: The instance opcode @param default_vg: The default_vg to assume - @return: The computer disks + @return: The computed disks """ disks = [] @@ -9598,16 +9731,37 @@ def _ComputeDisks(op, default_vg): raise errors.OpPrereqError("Invalid disk size '%s'" % size, errors.ECODE_INVAL) + ext_provider = disk.get(constants.IDISK_PROVIDER, None) + if ext_provider and op.disk_template != constants.DT_EXT: + raise errors.OpPrereqError("The '%s' option is only valid for the %s" + " disk template, not %s" % + (constants.IDISK_PROVIDER, constants.DT_EXT, + op.disk_template), errors.ECODE_INVAL) + data_vg = disk.get(constants.IDISK_VG, default_vg) new_disk = { constants.IDISK_SIZE: size, constants.IDISK_MODE: mode, constants.IDISK_VG: 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] + + # For extstorage, demand the `provider' option and add any + # additional parameters (ext-params) to the dict + if op.disk_template == constants.DT_EXT: + if ext_provider: + new_disk[constants.IDISK_PROVIDER] = ext_provider + for key in disk: + if key not in constants.IDISK_PARAMS: + new_disk[key] = disk[key] + else: + raise errors.OpPrereqError("Missing provider for template '%s'" % + constants.DT_EXT, errors.ECODE_INVAL) + disks.append(new_disk) return disks @@ -9631,6 +9785,16 @@ def _ComputeFullBeParams(op, cluster): return cluster.SimpleFillBE(op.beparams) +def _CheckOpportunisticLocking(op): + """Generate error if opportunistic locking is not possible. + + """ + if op.opportunistic_locking and not op.iallocator: + raise errors.OpPrereqError("Opportunistic locking is only available in" + " combination with an instance allocator", + errors.ECODE_INVAL) + + class LUInstanceCreate(LogicalUnit): """Create an instance. @@ -9664,7 +9828,8 @@ class LUInstanceCreate(LogicalUnit): # check disks. parameter names and consistent adopt/no-adopt strategy has_adopt = has_no_adopt = False for disk in self.op.disks: - utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES) + if self.op.disk_template != constants.DT_EXT: + utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES) if constants.IDISK_ADOPT in disk: has_adopt = True else: @@ -9726,6 +9891,8 @@ class LUInstanceCreate(LogicalUnit): " template") self.op.snode = None + _CheckOpportunisticLocking(self.op) + self._cds = _GetClusterDomainSecret() if self.op.mode == constants.INSTANCE_IMPORT: @@ -9816,7 +9983,11 @@ class LUInstanceCreate(LogicalUnit): # specifying a group on instance creation and then selecting nodes from # that group self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET - self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET + self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET + + if self.op.opportunistic_locking: + self.opportunistic_locks[locking.LEVEL_NODE] = True + self.opportunistic_locks[locking.LEVEL_NODE_RES] = True else: self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode) nodelist = [self.op.pnode] @@ -9824,9 +9995,6 @@ class LUInstanceCreate(LogicalUnit): self.op.snode = _ExpandNodeName(self.cfg, self.op.snode) nodelist.append(self.op.snode) self.needed_locks[locking.LEVEL_NODE] = nodelist - # Lock resources of instance's primary and secondary nodes (copy to - # prevent accidential modification) - self.needed_locks[locking.LEVEL_NODE_RES] = list(nodelist) # in case of import lock the source node too if self.op.mode == constants.INSTANCE_IMPORT: @@ -9838,6 +10006,7 @@ class LUInstanceCreate(LogicalUnit): if src_node is None: self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET + self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET self.op.src_node = None if os.path.isabs(src_path): raise errors.OpPrereqError("Importing an instance from a path" @@ -9851,23 +10020,40 @@ class LUInstanceCreate(LogicalUnit): self.op.src_path = src_path = \ utils.PathJoin(pathutils.EXPORT_DIR, src_path) + self.needed_locks[locking.LEVEL_NODE_RES] = \ + _CopyLockList(self.needed_locks[locking.LEVEL_NODE]) + def _RunAllocator(self): """Run the allocator based on input opcode. """ + if self.op.opportunistic_locking: + # Only consider nodes for which a lock is held + node_whitelist = self.owned_locks(locking.LEVEL_NODE) + else: + node_whitelist = None + #TODO Export network to iallocator so that it chooses a pnode # in a nodegroup that has the desired network connected to req = _CreateInstanceAllocRequest(self.op, self.disks, - self.nics, self.be_full) + self.nics, self.be_full, + node_whitelist) ial = iallocator.IAllocator(self.cfg, self.rpc, req) ial.Run(self.op.iallocator) if not ial.success: + # When opportunistic locks are used only a temporary failure is generated + if self.op.opportunistic_locking: + ecode = errors.ECODE_TEMP_NORES + else: + ecode = errors.ECODE_NORES + raise errors.OpPrereqError("Can't compute nodes using" " iallocator '%s': %s" % (self.op.iallocator, ial.info), - errors.ECODE_NORES) + ecode) + self.op.pnode = ial.result[0] self.LogInfo("Selected nodes for instance %s via iallocator %s: %s", self.op.instance_name, self.op.iallocator, @@ -10167,8 +10353,8 @@ class LUInstanceCreate(LogicalUnit): self._RevertToDefaults(cluster) # NIC buildup - self.nics = _ComputeNics(self.op, cluster, self.hostname1.ip, self.cfg, - self.proc) + self.nics = _ComputeNics(self.op, cluster, self.check_ip, self.cfg, + self.proc.GetECId()) # disk checks/pre-build default_vg = self.cfg.GetVGName() @@ -10221,12 +10407,14 @@ class LUInstanceCreate(LogicalUnit): self._RunAllocator() # Release all unneeded node locks - _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])) + keep_locks = filter(None, [self.op.pnode, self.op.snode, self.op.src_node]) + _ReleaseLocks(self, locking.LEVEL_NODE, keep=keep_locks) + _ReleaseLocks(self, locking.LEVEL_NODE_RES, keep=keep_locks) + _ReleaseLocks(self, locking.LEVEL_NODE_ALLOC) + + assert (self.owned_locks(locking.LEVEL_NODE) == + self.owned_locks(locking.LEVEL_NODE_RES)), \ + "Node locks differ from node resource locks" #### node related checks @@ -10278,10 +10466,10 @@ class LUInstanceCreate(LogicalUnit): " or does not belong to network %s" % (nic.ip, net), errors.ECODE_NOTUNIQUE) - else: - # net is None, ip None or given - if self.op.conflicts_check: - _CheckForConflictingIp(self, nic.ip, self.pnode.name) + + # net is None, ip None or given + elif self.op.conflicts_check: + _CheckForConflictingIp(self, nic.ip, self.pnode.name) # mirror node verification if self.op.disk_template in constants.DTS_INT_MIRROR: @@ -10327,6 +10515,9 @@ class LUInstanceCreate(LogicalUnit): # Any function that checks prerequisites can be placed here. # Check if there is enough space on the RADOS cluster. _CheckRADOSFreeSpace() + elif self.op.disk_template == constants.DT_EXT: + # FIXME: Function that checks prereqs if needed + pass else: # Check lv size requirements, if not adopting req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks) @@ -10384,7 +10575,7 @@ class LUInstanceCreate(LogicalUnit): if baddisks: raise errors.OpPrereqError("Device node(s) %s lie outside %s and" " cannot be adopted" % - (", ".join(baddisks), + (utils.CommaJoin(baddisks), constants.ADOPTABLE_BLOCKDEV_ROOT), errors.ECODE_INVAL) @@ -10431,6 +10622,9 @@ class LUInstanceCreate(LogicalUnit): _CheckNicsBridgesExist(self, self.nics, self.pnode.name) + #TODO: _CheckExtParams (remotely) + # Check parameters for extstorage + # memory check on primary node #TODO(dynmem): use MINMEM for checking if self.op.start: @@ -10451,6 +10645,7 @@ class LUInstanceCreate(LogicalUnit): assert not (self.owned_locks(locking.LEVEL_NODE_RES) - self.owned_locks(locking.LEVEL_NODE)), \ "Node locks differ from node resource locks" + assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC) ht_kind = self.op.hypervisor if ht_kind in constants.HTS_REQ_PORT: @@ -10719,6 +10914,8 @@ class LUInstanceMultiAlloc(NoHooksLU): " or set a cluster-wide default iallocator", errors.ECODE_INVAL) + _CheckOpportunisticLocking(self.op) + dups = utils.FindDuplicates([op.instance_name for op in self.op.instances]) if dups: raise errors.OpPrereqError("There are duplicate instance names: %s" % @@ -10729,11 +10926,19 @@ class LUInstanceMultiAlloc(NoHooksLU): """ self.share_locks = _ShareAll() - self.needed_locks = {} + self.needed_locks = { + # iallocator will select nodes and even if no iallocator is used, + # collisions with LUInstanceCreate should be avoided + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, + } if self.op.iallocator: self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET + + if self.op.opportunistic_locking: + self.opportunistic_locks[locking.LEVEL_NODE] = True + self.opportunistic_locks[locking.LEVEL_NODE_RES] = True else: nodeslist = [] for inst in self.op.instances: @@ -10754,11 +10959,21 @@ class LUInstanceMultiAlloc(NoHooksLU): """ cluster = self.cfg.GetClusterInfo() default_vg = self.cfg.GetVGName() + ec_id = self.proc.GetECId() + + if self.op.opportunistic_locking: + # Only consider nodes for which a lock is held + node_whitelist = self.owned_locks(locking.LEVEL_NODE) + else: + node_whitelist = None + insts = [_CreateInstanceAllocRequest(op, _ComputeDisks(op, default_vg), _ComputeNics(op, cluster, None, - self.cfg, self.proc), - _ComputeFullBeParams(op, cluster)) + self.cfg, ec_id), + _ComputeFullBeParams(op, cluster), + node_whitelist) for op in self.op.instances] + req = iallocator.IAReqMultiInstanceAlloc(instances=insts) ial = iallocator.IAllocator(self.cfg, self.rpc, req) @@ -10773,7 +10988,7 @@ class LUInstanceMultiAlloc(NoHooksLU): self.ia_result = ial.result if self.op.dry_run: - self.dry_run_rsult = objects.FillDict(self._ConstructPartialResult(), { + self.dry_run_result = objects.FillDict(self._ConstructPartialResult(), { constants.JOB_IDS_KEY: [], }) @@ -10947,6 +11162,7 @@ class LUInstanceReplaceDisks(LogicalUnit): if self.op.iallocator is not None: # iallocator will select a new node in the same group self.needed_locks[locking.LEVEL_NODEGROUP] = [] + self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET self.needed_locks[locking.LEVEL_NODE_RES] = [] @@ -10973,6 +11189,7 @@ class LUInstanceReplaceDisks(LogicalUnit): if self.op.iallocator is not None: assert self.op.remote_node is None assert not self.needed_locks[locking.LEVEL_NODE] + assert locking.NAL in self.owned_locks(locking.LEVEL_NODE_ALLOC) # Lock member nodes of all locked groups self.needed_locks[locking.LEVEL_NODE] = \ @@ -10980,7 +11197,10 @@ class LUInstanceReplaceDisks(LogicalUnit): for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP) for node_name in self.cfg.GetNodeGroup(group_uuid).members] else: + assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC) + self._LockInstancesNodes() + elif level == locking.LEVEL_NODE_RES: # Reuse node locks self.needed_locks[locking.LEVEL_NODE_RES] = \ @@ -11252,10 +11472,10 @@ class TLReplaceDisks(Tasklet): # Release unneeded node and node resource locks _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes) _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES, keep=touched_nodes) + _ReleaseLocks(self.lu, locking.LEVEL_NODE_ALLOC) # Release any owned node group - if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP): - _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP) + _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP) # Check whether disks are valid for disk_idx in self.disks: @@ -11279,6 +11499,7 @@ class TLReplaceDisks(Tasklet): (owned_nodes, self.node_secondary_ip.keys())) assert (self.lu.owned_locks(locking.LEVEL_NODE) == self.lu.owned_locks(locking.LEVEL_NODE_RES)) + assert not self.lu.glm.is_owned(locking.LEVEL_NODE_ALLOC) owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE) assert list(owned_instances) == [self.instance_name], \ @@ -11294,8 +11515,8 @@ class TLReplaceDisks(Tasklet): feedback_fn("Replacing disk(s) %s for instance '%s'" % (utils.CommaJoin(self.disks), self.instance.name)) - feedback_fn("Current primary node: %s", self.instance.primary_node) - feedback_fn("Current seconary node: %s", + feedback_fn("Current primary node: %s" % self.instance.primary_node) + feedback_fn("Current seconary node: %s" % utils.CommaJoin(self.instance.secondary_nodes)) activate_disks = (self.instance.admin_state != constants.ADMINST_UP) @@ -11355,7 +11576,7 @@ class TLReplaceDisks(Tasklet): continue for node in nodes: - self.lu.LogInfo("Checking disk/%d on %s" % (idx, node)) + self.lu.LogInfo("Checking disk/%d on %s", idx, node) self.cfg.SetDiskID(dev, node) result = _BlockdevFind(self, node, dev, self.instance) @@ -11395,7 +11616,7 @@ class TLReplaceDisks(Tasklet): if idx not in self.disks: continue - self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx)) + self.lu.LogInfo("Adding storage on %s for disk/%d", node_name, idx) self.cfg.SetDiskID(dev, node_name) @@ -11416,11 +11637,13 @@ class TLReplaceDisks(Tasklet): new_lvs = [lv_data, lv_meta] old_lvs = [child.Copy() for child in dev.children] iv_names[dev.iv_name] = (dev, old_lvs, new_lvs) + excl_stor = _IsExclusiveStorageEnabledNodeName(self.lu.cfg, node_name) # we pass force_create=True to force the LVM creation for new_lv in new_lvs: _CreateBlockDevInner(self.lu, node_name, self.instance, new_lv, True, - _GetInstanceInfoText(self.instance), False) + _GetInstanceInfoText(self.instance), False, + excl_stor) return iv_names @@ -11442,14 +11665,14 @@ class TLReplaceDisks(Tasklet): def _RemoveOldStorage(self, node_name, iv_names): for name, (_, old_lvs, _) in iv_names.iteritems(): - self.lu.LogInfo("Remove logical volumes for %s" % name) + self.lu.LogInfo("Remove logical volumes for %s", name) for lv in old_lvs: self.cfg.SetDiskID(lv, node_name) msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg if msg: - self.lu.LogWarning("Can't remove old LV: %s" % msg, + self.lu.LogWarning("Can't remove old LV: %s", msg, hint="remove unused LVs manually") def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613 @@ -11494,7 +11717,7 @@ class TLReplaceDisks(Tasklet): # Step: for each lv, detach+rename*2+attach self.lu.LogStep(4, steps_total, "Changing drbd configuration") for dev, old_lvs, new_lvs in iv_names.itervalues(): - self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name) + self.lu.LogInfo("Detaching %s drbd from local storage", dev.iv_name) result = self.rpc.call_blockdev_removechildren(self.target_node, dev, old_lvs) @@ -11548,7 +11771,7 @@ class TLReplaceDisks(Tasklet): self.cfg.SetDiskID(disk, self.target_node) # Now that the new lvs have the old name, we can add them to the device - self.lu.LogInfo("Adding new mirror component on %s" % self.target_node) + self.lu.LogInfo("Adding new mirror component on %s", self.target_node) result = self.rpc.call_blockdev_addchildren(self.target_node, (dev, self.instance), new_lvs) msg = result.fail_msg @@ -11629,13 +11852,15 @@ class TLReplaceDisks(Tasklet): # Step: create new storage self.lu.LogStep(3, steps_total, "Allocate new storage") disks = _AnnotateDiskParams(self.instance, self.instance.disks, self.cfg) + excl_stor = _IsExclusiveStorageEnabledNodeName(self.lu.cfg, self.new_node) for idx, dev in enumerate(disks): self.lu.LogInfo("Adding new local storage on %s for disk/%d" % (self.new_node, idx)) # we pass force_create=True to force LVM creation for new_lv in dev.children: _CreateBlockDevInner(self.lu, self.new_node, self.instance, new_lv, - True, _GetInstanceInfoText(self.instance), False) + True, _GetInstanceInfoText(self.instance), False, + excl_stor) # Step 4: dbrd minors and drbd setups changes # after this, we must manually remove the drbd minors on both the @@ -11679,14 +11904,15 @@ class TLReplaceDisks(Tasklet): try: _CreateSingleBlockDev(self.lu, self.new_node, self.instance, anno_new_drbd, - _GetInstanceInfoText(self.instance), False) + _GetInstanceInfoText(self.instance), False, + excl_stor) except errors.GenericError: self.cfg.ReleaseDRBDMinors(self.instance.name) raise # We have new devices, shutdown the drbd on the old secondary for idx, dev in enumerate(self.instance.disks): - self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx) + self.lu.LogInfo("Shutting down drbd for disk/%d on old node", idx) self.cfg.SetDiskID(dev, self.target_node) msg = self.rpc.call_blockdev_shutdown(self.target_node, (dev, self.instance)).fail_msg @@ -11798,7 +12024,7 @@ class LURepairNodeStorage(NoHooksLU): errors.ECODE_STATE) except errors.OpPrereqError, err: if self.op.ignore_consistency: - self.proc.LogWarning(str(err.args[0])) + self.LogWarning(str(err.args[0])) else: raise @@ -12185,7 +12411,8 @@ class LUInstanceGrowDisk(LogicalUnit): if instance.disk_template not in (constants.DT_FILE, constants.DT_SHARED_FILE, - constants.DT_RBD): + constants.DT_RBD, + constants.DT_EXT): # TODO: check the free disk space for file, when that feature will be # supported _CheckNodesFreeDiskPerVG(self, nodenames, @@ -12276,14 +12503,14 @@ class LUInstanceGrowDisk(LogicalUnit): if self.op.wait_for_sync: disk_abort = not _WaitForSync(self, instance, disks=[disk]) if disk_abort: - self.proc.LogWarning("Disk sync-ing has not returned a good" - " status; please check the instance") + self.LogWarning("Disk syncing has not returned a good status; check" + " the instance") if instance.admin_state != constants.ADMINST_UP: _SafeShutdownInstanceDisks(self, instance, disks=[disk]) elif instance.admin_state != constants.ADMINST_UP: - self.proc.LogWarning("Not shutting down the disk even if the instance is" - " not supposed to be running because no wait for" - " sync mode was requested") + self.LogWarning("Not shutting down the disk even if the instance is" + " not supposed to be running because no wait for" + " sync mode was requested") assert self.owned_locks(locking.LEVEL_NODE_RES) assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE) @@ -12685,7 +12912,10 @@ class LUInstanceSetParams(LogicalUnit): for (op, _, params) in mods: assert ht.TDict(params) - utils.ForceDictType(params, key_types) + # If 'key_types' is an empty dict, we assume we have an + # 'ext' template and thus do not ForceDictType + if key_types: + utils.ForceDictType(params, key_types) if op == constants.DDM_REMOVE: if params: @@ -12721,9 +12951,18 @@ class LUInstanceSetParams(LogicalUnit): params[constants.IDISK_SIZE] = size - elif op == constants.DDM_MODIFY and constants.IDISK_SIZE in params: - raise errors.OpPrereqError("Disk size change not possible, use" - " grow-disk", errors.ECODE_INVAL) + elif op == constants.DDM_MODIFY: + if constants.IDISK_SIZE in params: + raise errors.OpPrereqError("Disk size change not possible, use" + " grow-disk", errors.ECODE_INVAL) + if constants.IDISK_MODE not in params: + raise errors.OpPrereqError("Disk 'mode' is the only kind of" + " modification supported, but missing", + errors.ECODE_NOENT) + if len(params) > 1: + raise errors.OpPrereqError("Disk modification doesn't support" + " additional arbitrary parameters", + errors.ECODE_INVAL) @staticmethod def _VerifyNicModification(op, params): @@ -12787,10 +13026,6 @@ class LUInstanceSetParams(LogicalUnit): self.op.nics = self._UpgradeDiskNicMods( "NIC", self.op.nics, opcodes.OpInstanceSetParams.TestNicModifications) - # Check disk modifications - self._CheckMods("disk", self.op.disks, constants.IDISK_PARAMS_TYPES, - self._VerifyDiskModification) - if self.op.disks and self.op.disk_template is not None: raise errors.OpPrereqError("Disk template conversion and other disk" " changes not supported at the same time", @@ -12892,7 +13127,7 @@ class LUInstanceSetParams(LogicalUnit): netparams = self.cfg.GetGroupNetParams(new_net, pnode) if netparams is None: raise errors.OpPrereqError("No netparams found for the network" - " %s, propably not connected." % new_net, + " %s, probably not connected" % new_net, errors.ECODE_INVAL) new_params = dict(netparams) else: @@ -12920,6 +13155,10 @@ class LUInstanceSetParams(LogicalUnit): raise errors.OpPrereqError("Cannot set the NIC IP address to None" " on a routed NIC", errors.ECODE_INVAL) + elif new_mode == constants.NIC_MODE_OVS: + # TODO: check OVS link + self.LogInfo("OVS links are currently not checked for correctness") + if constants.INIC_MAC in params: mac = params[constants.INIC_MAC] if mac is None: @@ -12978,10 +13217,10 @@ class LUInstanceSetParams(LogicalUnit): elif new_ip.lower() == constants.NIC_IP_POOL: raise errors.OpPrereqError("ip=pool, but no network found", errors.ECODE_INVAL) - else: - # new net is None - if self.op.conflicts_check: - _CheckForConflictingIp(self, new_ip, pnode) + + # new net is None + elif self.op.conflicts_check: + _CheckForConflictingIp(self, new_ip, pnode) if old_ip: if old_net: @@ -12995,13 +13234,13 @@ class LUInstanceSetParams(LogicalUnit): elif (old_net is not None and (req_link is not None or req_mode is not None)): raise errors.OpPrereqError("Not allowed to change link or mode of" - " a NIC that is connected to a network.", + " a NIC that is connected to a network", errors.ECODE_INVAL) private.params = new_params private.filled = new_filled_params - def CheckPrereq(self): + def CheckPrereq(self): # pylint: disable=R0914 """Check prerequisites. This only checks the instance list against the existing names. @@ -13027,10 +13266,46 @@ class LUInstanceSetParams(LogicalUnit): # dictionary with instance information after the modification ispec = {} + # Check disk modifications. This is done here and not in CheckArguments + # (as with NICs), because we need to know the instance's disk template + if instance.disk_template == constants.DT_EXT: + self._CheckMods("disk", self.op.disks, {}, + self._VerifyDiskModification) + else: + self._CheckMods("disk", self.op.disks, constants.IDISK_PARAMS_TYPES, + self._VerifyDiskModification) + # Prepare disk/NIC modifications self.diskmod = PrepareContainerMods(self.op.disks, None) self.nicmod = PrepareContainerMods(self.op.nics, _InstNicModPrivate) + # Check the validity of the `provider' parameter + if instance.disk_template in constants.DT_EXT: + for mod in self.diskmod: + ext_provider = mod[2].get(constants.IDISK_PROVIDER, None) + if mod[0] == constants.DDM_ADD: + if ext_provider is None: + raise errors.OpPrereqError("Instance template is '%s' and parameter" + " '%s' missing, during disk add" % + (constants.DT_EXT, + constants.IDISK_PROVIDER), + errors.ECODE_NOENT) + elif mod[0] == constants.DDM_MODIFY: + if ext_provider: + raise errors.OpPrereqError("Parameter '%s' is invalid during disk" + " modification" % + constants.IDISK_PROVIDER, + errors.ECODE_INVAL) + else: + for mod in self.diskmod: + ext_provider = mod[2].get(constants.IDISK_PROVIDER, None) + if ext_provider is not None: + raise errors.OpPrereqError("Parameter '%s' is only valid for" + " instances of type '%s'" % + (constants.IDISK_PROVIDER, + constants.DT_EXT), + errors.ECODE_INVAL) + # OS change if self.op.os_name and not self.op.force: _CheckNodeHasOS(self, instance.primary_node, self.op.os_name, @@ -13276,16 +13551,13 @@ class LUInstanceSetParams(LogicalUnit): errors.ECODE_STATE) disk_sizes = [disk.size for disk in instance.disks] disk_sizes.extend(params["size"] for (op, idx, params, private) in - self.diskmod) + self.diskmod if op == constants.DDM_ADD) ispec[constants.ISPEC_DISK_COUNT] = len(disk_sizes) ispec[constants.ISPEC_DISK_SIZE] = disk_sizes - if self.op.offline is not None: - if self.op.offline: - msg = "can't change to offline" - else: - msg = "can't change to online" - _CheckInstanceState(self, instance, CAN_CHANGE_INSTANCE_OFFLINE, msg=msg) + if self.op.offline is not None and self.op.offline: + _CheckInstanceState(self, instance, CAN_CHANGE_INSTANCE_OFFLINE, + msg="can't change to offline") # Pre-compute NIC changes (necessary to use result in hooks) self._nic_chgdesc = [] @@ -13349,15 +13621,18 @@ class LUInstanceSetParams(LogicalUnit): self.diskparams) anno_disks = rpc.AnnotateDiskParams(constants.DT_DRBD8, new_disks, self.diskparams) + p_excl_stor = _IsExclusiveStorageEnabledNodeName(self.cfg, pnode) + s_excl_stor = _IsExclusiveStorageEnabledNodeName(self.cfg, snode) info = _GetInstanceInfoText(instance) feedback_fn("Creating additional volumes...") # first, create the missing data and meta devices for disk in anno_disks: # unfortunately this is... not too nice _CreateSingleBlockDev(self, pnode, instance, disk.children[1], - info, True) + info, True, p_excl_stor) for child in disk.children: - _CreateSingleBlockDev(self, snode, instance, child, info, True) + _CreateSingleBlockDev(self, snode, instance, child, info, True, + s_excl_stor) # at this stage, all new LVs have been created, we can rename the # old ones feedback_fn("Renaming original volumes...") @@ -13369,9 +13644,10 @@ class LUInstanceSetParams(LogicalUnit): feedback_fn("Initializing DRBD devices...") # all child devices are in place, we can now create the DRBD devices for disk in anno_disks: - for node in [pnode, snode]: + for (node, excl_stor) in [(pnode, p_excl_stor), (snode, s_excl_stor)]: f_create = node == pnode - _CreateSingleBlockDev(self, node, instance, disk, info, f_create) + _CreateSingleBlockDev(self, node, instance, disk, info, f_create, + excl_stor) # at this point, the instance has been modified instance.disk_template = constants.DT_DRBD8 @@ -13669,9 +13945,11 @@ class LUInstanceChangeGroup(LogicalUnit): def ExpandNames(self): self.share_locks = _ShareAll() + self.needed_locks = { locking.LEVEL_NODEGROUP: [], locking.LEVEL_NODE: [], + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, } self._ExpandAndLockInstance() @@ -13854,6 +14132,9 @@ class _ExportQuery(_QueryBase): locking.LEVEL_NODE: self.wanted, } + if not self.names: + lu.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET + def DeclareLocks(self, lu, level): pass @@ -13971,6 +14252,11 @@ class LUBackupExport(LogicalUnit): # - removing the removal operation altogether self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET + # Allocations should be stopped while this LU runs with node locks, but + # it doesn't have to be exclusive + self.share_locks[locking.LEVEL_NODE_ALLOC] = 1 + self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET + def DeclareLocks(self, level): """Last minute lock declaration.""" # All nodes are locked anyway, so nothing to do here. @@ -14237,11 +14523,19 @@ class LUBackupRemove(NoHooksLU): REQ_BGL = False def ExpandNames(self): - self.needed_locks = {} - # We need all nodes to be locked in order for RemoveExport to work, but we - # don't need to lock the instance itself, as nothing will happen to it (and - # we can remove exports also for a removed instance) - self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET + self.needed_locks = { + # We need all nodes to be locked in order for RemoveExport to work, but + # we don't need to lock the instance itself, as nothing will happen to it + # (and we can remove exports also for a removed instance) + locking.LEVEL_NODE: locking.ALL_SET, + + # Removing backups is quick, so blocking allocations is justified + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, + } + + # Allocations should be stopped while this LU runs with node locks, but it + # doesn't have to be exclusive + self.share_locks[locking.LEVEL_NODE_ALLOC] = 1 def Exec(self, feedback_fn): """Remove any export. @@ -15280,7 +15574,7 @@ class LUTestDelay(NoHooksLU): else: top_value = self.op.repeat - 1 for i in range(self.op.repeat): - self.LogInfo("Test delay iteration %d/%d" % (i, top_value)) + self.LogInfo("Test delay iteration %d/%d", i, top_value) self._TestDelay() @@ -15521,7 +15815,7 @@ class LUTestAllocator(NoHooksLU): self.op.mode, errors.ECODE_INVAL) if self.op.direction == constants.IALLOCATOR_DIR_OUT: - if self.op.allocator is None: + if self.op.iallocator is None: raise errors.OpPrereqError("Missing allocator name", errors.ECODE_INVAL) elif self.op.direction != constants.IALLOCATOR_DIR_IN: @@ -15574,12 +15868,11 @@ class LUTestAllocator(NoHooksLU): if self.op.direction == constants.IALLOCATOR_DIR_IN: result = ial.in_text else: - ial.Run(self.op.allocator, validate=False) + ial.Run(self.op.iallocator, validate=False) result = ial.out_text return result -# Network LUs class LUNetworkAdd(LogicalUnit): """Logical unit for creating networks. @@ -15595,18 +15888,27 @@ class LUNetworkAdd(LogicalUnit): mn = self.cfg.GetMasterNode() return ([mn], [mn]) + def CheckArguments(self): + if self.op.mac_prefix: + self.op.mac_prefix = \ + utils.NormalizeAndValidateThreeOctetMacPrefix(self.op.mac_prefix) + def ExpandNames(self): self.network_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId()) - self.needed_locks = {} - self.add_locks[locking.LEVEL_NETWORK] = self.network_uuid - def CheckPrereq(self): - """Check prerequisites. + if self.op.conflicts_check: + self.share_locks[locking.LEVEL_NODE] = 1 + self.share_locks[locking.LEVEL_NODE_ALLOC] = 1 + self.needed_locks = { + locking.LEVEL_NODE: locking.ALL_SET, + locking.LEVEL_NODE_ALLOC: locking.ALL_SET, + } + else: + self.needed_locks = {} - This checks that the given group name is not an existing node group - already. + self.add_locks[locking.LEVEL_NETWORK] = self.network_uuid - """ + def CheckPrereq(self): if self.op.network is None: raise errors.OpPrereqError("Network must be given", errors.ECODE_INVAL) @@ -15614,11 +15916,8 @@ class LUNetworkAdd(LogicalUnit): uuid = self.cfg.LookupNetwork(self.op.network_name) if uuid: - raise errors.OpPrereqError("Network '%s' already defined" % - self.op.network, errors.ECODE_EXISTS) - - if self.op.mac_prefix: - utils.NormalizeAndValidateMac(self.op.mac_prefix + ":00:00:00") + raise errors.OpPrereqError(("Network with name '%s' already exists" % + self.op.network_name), errors.ECODE_EXISTS) # Check tag validity for tag in self.op.tags: @@ -15652,31 +15951,36 @@ class LUNetworkAdd(LogicalUnit): mac_prefix=self.op.mac_prefix, network_type=self.op.network_type, uuid=self.network_uuid, - family=4) + family=constants.IP4_VERSION) # Initialize the associated address pool try: pool = network.AddressPool.InitializeNetwork(nobj) except errors.AddressPoolError, e: - raise errors.OpExecError("Cannot create IP pool for this network. %s" % e) + raise errors.OpExecError("Cannot create IP pool for this network: %s" % e) # Check if we need to reserve the nodes and the cluster master IP # These may not be allocated to any instances in routed mode, as # they wouldn't function anyway. - for node in self.cfg.GetAllNodesInfo().values(): - for ip in [node.primary_ip, node.secondary_ip]: - try: - pool.Reserve(ip) - self.LogInfo("Reserved node %s's IP (%s)", node.name, ip) - - except errors.AddressPoolError: - pass + if self.op.conflicts_check: + for node in self.cfg.GetAllNodesInfo().values(): + for ip in [node.primary_ip, node.secondary_ip]: + try: + if pool.Contains(ip): + pool.Reserve(ip) + self.LogInfo("Reserved IP address of node '%s' (%s)", + node.name, ip) + except errors.AddressPoolError: + self.LogWarning("Cannot reserve IP address of node '%s' (%s)", + node.name, ip) - master_ip = self.cfg.GetClusterInfo().master_ip - try: - pool.Reserve(master_ip) - self.LogInfo("Reserved cluster master IP (%s)", master_ip) - except errors.AddressPoolError: - pass + master_ip = self.cfg.GetClusterInfo().master_ip + try: + if pool.Contains(master_ip): + pool.Reserve(master_ip) + self.LogInfo("Reserved cluster master IP address (%s)", master_ip) + except errors.AddressPoolError: + self.LogWarning("Cannot reserve cluster master IP address (%s)", + master_ip) if self.op.add_reserved_ips: for ip in self.op.add_reserved_ips: @@ -15702,10 +16006,13 @@ class LUNetworkRemove(LogicalUnit): self.network_uuid = self.cfg.LookupNetwork(self.op.network_name) if not self.network_uuid: - raise errors.OpPrereqError("Network %s not found" % self.op.network_name, - errors.ECODE_INVAL) + raise errors.OpPrereqError(("Network '%s' not found" % + self.op.network_name), errors.ECODE_NOENT) + + self.share_locks[locking.LEVEL_NODEGROUP] = 1 self.needed_locks = { locking.LEVEL_NETWORK: [self.network_uuid], + locking.LEVEL_NODEGROUP: locking.ALL_SET, } def CheckPrereq(self): @@ -15716,19 +16023,17 @@ class LUNetworkRemove(LogicalUnit): cluster. """ - # Verify that the network is not conncted. node_groups = [group.name for group in self.cfg.GetAllNodeGroupsInfo().values() - for net in group.networks.keys() - if net == self.network_uuid] + if self.network_uuid in group.networks] if node_groups: - self.LogWarning("Nework '%s' is connected to the following" - " node groups: %s" % (self.op.network_name, - utils.CommaJoin(utils.NiceSort(node_groups)))) - raise errors.OpPrereqError("Network still connected", - errors.ECODE_STATE) + self.LogWarning("Network '%s' is connected to the following" + " node groups: %s" % + (self.op.network_name, + utils.CommaJoin(utils.NiceSort(node_groups)))) + raise errors.OpPrereqError("Network still connected", errors.ECODE_STATE) def BuildHooksEnv(self): """Build hooks env. @@ -15772,11 +16077,10 @@ class LUNetworkSetParams(LogicalUnit): def ExpandNames(self): self.network_uuid = self.cfg.LookupNetwork(self.op.network_name) - self.network = self.cfg.GetNetwork(self.network_uuid) - if self.network is None: - raise errors.OpPrereqError("Could not retrieve network '%s' (UUID: %s)" % - (self.op.network_name, self.network_uuid), - errors.ECODE_INVAL) + if self.network_uuid is None: + raise errors.OpPrereqError(("Network '%s' not found" % + self.op.network_name), errors.ECODE_NOENT) + self.needed_locks = { locking.LEVEL_NETWORK: [self.network_uuid], } @@ -15785,6 +16089,7 @@ class LUNetworkSetParams(LogicalUnit): """Check prerequisites. """ + self.network = self.cfg.GetNetwork(self.network_uuid) self.gateway = self.network.gateway self.network_type = self.network.network_type self.mac_prefix = self.network.mac_prefix @@ -15800,8 +16105,9 @@ class LUNetworkSetParams(LogicalUnit): else: self.gateway = self.op.gateway if self.pool.IsReserved(self.gateway): - raise errors.OpPrereqError("%s is already reserved" % - self.gateway, errors.ECODE_INVAL) + raise errors.OpPrereqError("Gateway IP address '%s' is already" + " reserved" % self.gateway, + errors.ECODE_STATE) if self.op.network_type: if self.op.network_type == constants.VALUE_NONE: @@ -15813,8 +16119,8 @@ class LUNetworkSetParams(LogicalUnit): if self.op.mac_prefix == constants.VALUE_NONE: self.mac_prefix = None else: - utils.NormalizeAndValidateMac(self.op.mac_prefix + ":00:00:00") - self.mac_prefix = self.op.mac_prefix + self.mac_prefix = \ + utils.NormalizeAndValidateThreeOctetMacPrefix(self.op.mac_prefix) if self.op.gateway6: if self.op.gateway6 == constants.VALUE_NONE: @@ -15859,7 +16165,7 @@ class LUNetworkSetParams(LogicalUnit): # extend cfg.ReserveIp/ReleaseIp with the external flag if self.op.gateway: if self.gateway == self.network.gateway: - self.LogWarning("Gateway is already %s" % self.gateway) + self.LogWarning("Gateway is already %s", self.gateway) else: if self.gateway: self.pool.Reserve(self.gateway, external=True) @@ -15871,11 +16177,11 @@ class LUNetworkSetParams(LogicalUnit): for ip in self.op.add_reserved_ips: try: if self.pool.IsReserved(ip): - self.LogWarning("IP %s is already reserved" % ip) + self.LogWarning("IP address %s is already reserved", ip) else: self.pool.Reserve(ip, external=True) - except errors.AddressPoolError, e: - self.LogWarning("Cannot reserve ip %s. %s" % (ip, e)) + except errors.AddressPoolError, err: + self.LogWarning("Cannot reserve IP address %s: %s", ip, err) if self.op.remove_reserved_ips: for ip in self.op.remove_reserved_ips: @@ -15884,11 +16190,11 @@ class LUNetworkSetParams(LogicalUnit): continue try: if not self.pool.IsReserved(ip): - self.LogWarning("IP %s is already unreserved" % ip) + self.LogWarning("IP address %s is already unreserved", ip) else: self.pool.Release(ip, external=True) - except errors.AddressPoolError, e: - self.LogWarning("Cannot release ip %s. %s" % (ip, e)) + except errors.AddressPoolError, err: + self.LogWarning("Cannot release IP address %s: %s", ip, err) if self.op.mac_prefix: self.network.mac_prefix = self.mac_prefix @@ -15946,11 +16252,8 @@ class _NetworkQuery(_QueryBase): """ do_instances = query.NETQ_INST in self.requested_data do_groups = do_instances or (query.NETQ_GROUP in self.requested_data) - do_stats = query.NETQ_STATS in self.requested_data - network_to_groups = None network_to_instances = None - stats = None # For NETQ_GROUP, we need to map network->[groups] if do_groups: @@ -15969,32 +16272,30 @@ class _NetworkQuery(_QueryBase): group_instances = [instance for instance in all_instances.values() if instance.primary_node in group_nodes] - for net_uuid in group.networks.keys(): - if net_uuid in network_to_groups: - netparams = group.networks[net_uuid] - mode = netparams[constants.NIC_MODE] - link = netparams[constants.NIC_LINK] - info = group.name + "(" + mode + ", " + link + ")" + for net_uuid in self.wanted: + netparams = group.networks.get(net_uuid, None) + if netparams: + info = (group.name, netparams[constants.NIC_MODE], + netparams[constants.NIC_LINK]) + network_to_groups[net_uuid].append(info) - if do_instances: - for instance in group_instances: - for nic in instance.nics: - if nic.network == self._all_networks[net_uuid].name: - network_to_instances[net_uuid].append(instance.name) - break - - if do_stats: - stats = {} - for uuid, net in self._all_networks.items(): - if uuid in self.wanted: - pool = network.AddressPool(net) - stats[uuid] = { - "free_count": pool.GetFreeCount(), - "reserved_count": pool.GetReservedCount(), - "map": pool.GetMap(), - "external_reservations": ", ".join(pool.GetExternalReservations()), - } + if do_instances: + for instance in group_instances: + for nic in instance.nics: + if nic.network == self._all_networks[net_uuid].name: + network_to_instances[net_uuid].append(instance.name) + break + else: + network_to_groups = None + + if query.NETQ_STATS in self.requested_data: + stats = \ + dict((uuid, + self._GetStats(network.AddressPool(self._all_networks[uuid]))) + for uuid in self.wanted) + else: + stats = None return query.NetworkQueryData([self._all_networks[uuid] for uuid in self.wanted], @@ -16002,6 +16303,19 @@ class _NetworkQuery(_QueryBase): network_to_instances, stats) + @staticmethod + def _GetStats(pool): + """Returns statistics for a network address pool. + + """ + return { + "free_count": pool.GetFreeCount(), + "reserved_count": pool.GetReservedCount(), + "map": pool.GetMap(), + "external_reservations": + utils.CommaJoin(pool.GetExternalReservations()), + } + class LUNetworkQuery(NoHooksLU): """Logical unit for querying networks. @@ -16035,16 +16349,14 @@ class LUNetworkConnect(LogicalUnit): self.network_link = self.op.network_link self.network_uuid = self.cfg.LookupNetwork(self.network_name) - self.network = self.cfg.GetNetwork(self.network_uuid) - if self.network is None: - raise errors.OpPrereqError("Network %s does not exist" % - self.network_name, errors.ECODE_INVAL) + if self.network_uuid is None: + raise errors.OpPrereqError("Network '%s' does not exist" % + self.network_name, errors.ECODE_NOENT) self.group_uuid = self.cfg.LookupNodeGroup(self.group_name) - self.group = self.cfg.GetNodeGroup(self.group_uuid) - if self.group is None: - raise errors.OpPrereqError("Group %s does not exist" % - self.group_name, errors.ECODE_INVAL) + if self.group_uuid is None: + raise errors.OpPrereqError("Group '%s' does not exist" % + self.group_name, errors.ECODE_NOENT) self.needed_locks = { locking.LEVEL_INSTANCE: [], @@ -16052,14 +16364,19 @@ class LUNetworkConnect(LogicalUnit): } self.share_locks[locking.LEVEL_INSTANCE] = 1 + if self.op.conflicts_check: + self.needed_locks[locking.LEVEL_NETWORK] = [self.network_uuid] + self.share_locks[locking.LEVEL_NETWORK] = 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) + if self.op.conflicts_check: + self.needed_locks[locking.LEVEL_INSTANCE] = \ + self.cfg.GetNodeGroupInstances(self.group_uuid) def BuildHooksEnv(self): ret = { @@ -16067,7 +16384,6 @@ class LUNetworkConnect(LogicalUnit): "GROUP_NETWORK_MODE": self.network_mode, "GROUP_NETWORK_LINK": self.network_link, } - ret.update(_BuildNetworkHookEnvByObject(self.network)) return ret def BuildHooksNodes(self): @@ -16075,8 +16391,9 @@ class LUNetworkConnect(LogicalUnit): return (nodes, nodes) def CheckPrereq(self): - l = lambda value: ", ".join("%s: %s/%s" % (i[0], i[1], i[2]) - for i in value) + owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) + + assert self.group_uuid in owned_groups self.netparams = { constants.NIC_MODE: self.network_mode, @@ -16084,6 +16401,7 @@ class LUNetworkConnect(LogicalUnit): } objects.NIC.CheckParameterSyntax(self.netparams) + self.group = self.cfg.GetNodeGroup(self.group_uuid) #if self.network_mode == constants.NIC_MODE_BRIDGED: # _CheckNodeGroupBridgesExist(self, self.network_link, self.group_uuid) self.connected = False @@ -16093,24 +16411,11 @@ class LUNetworkConnect(LogicalUnit): self.connected = True return - pool = network.AddressPool(self.network) if self.op.conflicts_check: - groupinstances = [] - for n in self.cfg.GetNodeGroupInstances(self.group_uuid): - groupinstances.append(self.cfg.GetInstanceInfo(n)) - instances = [(instance.name, idx, nic.ip) - for instance in groupinstances - for idx, nic in enumerate(instance.nics) - if (not nic.network and pool.Contains(nic.ip))] - if instances: - self.LogWarning("Following occurences use IPs from network %s" - " that is about to connect to nodegroup %s: %s" % - (self.network_name, self.group.name, - l(instances))) - raise errors.OpPrereqError("Conflicting IPs found." - " Please remove/modify" - " corresponding NICs", - errors.ECODE_INVAL) + pool = network.AddressPool(self.cfg.GetNetwork(self.network_uuid)) + + _NetworkConflictCheck(self, lambda nic: pool.Contains(nic.ip), + "connect to") def Exec(self, feedback_fn): if self.connected: @@ -16120,6 +16425,53 @@ class LUNetworkConnect(LogicalUnit): self.cfg.Update(self.group, feedback_fn) +def _NetworkConflictCheck(lu, check_fn, action): + """Checks for network interface conflicts with a network. + + @type lu: L{LogicalUnit} + @type check_fn: callable receiving one parameter (L{objects.NIC}) and + returning boolean + @param check_fn: Function checking for conflict + @type action: string + @param action: Part of error message (see code) + @raise errors.OpPrereqError: If conflicting IP addresses are found. + + """ + # Check if locked instances are still correct + owned_instances = frozenset(lu.owned_locks(locking.LEVEL_INSTANCE)) + _CheckNodeGroupInstances(lu.cfg, lu.group_uuid, owned_instances) + + conflicts = [] + + for (_, instance) in lu.cfg.GetMultiInstanceInfo(owned_instances): + instconflicts = [(idx, nic.ip) + for (idx, nic) in enumerate(instance.nics) + if check_fn(nic)] + + if instconflicts: + conflicts.append((instance.name, instconflicts)) + + if conflicts: + lu.LogWarning("IP addresses from network '%s', which is about to %s" + " node group '%s', are in use: %s" % + (lu.network_name, action, lu.group.name, + utils.CommaJoin(("%s: %s" % + (name, _FmtNetworkConflict(details))) + for (name, details) in conflicts))) + + raise errors.OpPrereqError("Conflicting IP addresses found; " + " remove/modify the corresponding network" + " interfaces", errors.ECODE_STATE) + + +def _FmtNetworkConflict(details): + """Utility for L{_NetworkConflictCheck}. + + """ + return utils.CommaJoin("nic%s/%s" % (idx, ipaddr) + for (idx, ipaddr) in details) + + class LUNetworkDisconnect(LogicalUnit): """Disconnect a network to a nodegroup @@ -16133,16 +16485,14 @@ class LUNetworkDisconnect(LogicalUnit): self.group_name = self.op.group_name self.network_uuid = self.cfg.LookupNetwork(self.network_name) - self.network = self.cfg.GetNetwork(self.network_uuid) - if self.network is None: - raise errors.OpPrereqError("Network %s does not exist" % - self.network_name, errors.ECODE_INVAL) + if self.network_uuid is None: + raise errors.OpPrereqError("Network '%s' does not exist" % + self.network_name, errors.ECODE_NOENT) self.group_uuid = self.cfg.LookupNodeGroup(self.group_name) - self.group = self.cfg.GetNodeGroup(self.group_uuid) - if self.group is None: - raise errors.OpPrereqError("Group %s does not exist" % - self.group_name, errors.ECODE_INVAL) + if self.group_uuid is None: + raise errors.OpPrereqError("Group '%s' does not exist" % + self.group_name, errors.ECODE_NOENT) self.needed_locks = { locking.LEVEL_INSTANCE: [], @@ -16156,14 +16506,14 @@ class LUNetworkDisconnect(LogicalUnit): # Lock instances optimistically, needs verification once group lock has # been acquired - self.needed_locks[locking.LEVEL_INSTANCE] = \ + if self.op.conflicts_check: + self.needed_locks[locking.LEVEL_INSTANCE] = \ self.cfg.GetNodeGroupInstances(self.group_uuid) def BuildHooksEnv(self): ret = { "GROUP_NAME": self.group_name, } - ret.update(_BuildNetworkHookEnvByObject(self.network)) return ret def BuildHooksNodes(self): @@ -16171,35 +16521,21 @@ class LUNetworkDisconnect(LogicalUnit): return (nodes, nodes) def CheckPrereq(self): - l = lambda value: ", ".join("%s: %s/%s" % (i[0], i[1], i[2]) - for i in value) + owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) + + assert self.group_uuid in owned_groups + self.group = self.cfg.GetNodeGroup(self.group_uuid) self.connected = True if self.network_uuid not in self.group.networks: - self.LogWarning("Network '%s' is" - " not mapped to group '%s'" % - (self.network_name, self.group.name)) + self.LogWarning("Network '%s' is not mapped to group '%s'", + self.network_name, self.group.name) self.connected = False return if self.op.conflicts_check: - groupinstances = [] - for n in self.cfg.GetNodeGroupInstances(self.group_uuid): - groupinstances.append(self.cfg.GetInstanceInfo(n)) - instances = [(instance.name, idx, nic.ip) - for instance in groupinstances - for idx, nic in enumerate(instance.nics) - if nic.network == self.network_name] - if instances: - self.LogWarning("Following occurences use IPs from network %s" - " that is about to disconnected from the nodegroup" - " %s: %s" % - (self.network_name, self.group.name, - l(instances))) - raise errors.OpPrereqError("Conflicting IPs." - " Please remove/modify" - " corresponding NICS", - errors.ECODE_INVAL) + _NetworkConflictCheck(self, lambda nic: nic.network == self.network_name, + "disconnect from") def Exec(self, feedback_fn): if not self.connected: @@ -16237,18 +16573,18 @@ def _GetQueryImplementation(name): def _CheckForConflictingIp(lu, ip, node): - """In case of conflicting ip raise error. + """In case of conflicting IP address raise error. @type ip: string - @param ip: ip address + @param ip: IP address @type node: string @param node: node name """ (conf_net, _) = lu.cfg.CheckIPInNodeGroup(ip, node) if conf_net is not None: - raise errors.OpPrereqError("Conflicting IP found:" - " %s <> %s." % (ip, conf_net), - errors.ECODE_INVAL) + raise errors.OpPrereqError(("Conflicting IP address found: '%s' != '%s'" % + (ip, conf_net)), + errors.ECODE_STATE) return (None, None)