LUInstanceRename: Amend comment about lock

[ganeti-local] / lib / cmdlib.py

diff --git a/lib/cmdlib.py b/lib/cmdlib.py
index 4b62876..08b6bf0 100644 (file)
--- a/lib/cmdlib.py
+++ b/lib/cmdlib.py
@@ -1265,6 +1265,7 @@ class LUClusterVerify(LogicalUnit):
 
   ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
   ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")
 
   ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
   ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")

+  ECLUSTERFILECHECK = (TCLUSTER, "ECLUSTERFILECHECK")

   EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
   EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
   EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
   EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
   EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
   EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")

@@ -1694,51 +1695,94 @@ class LUClusterVerify(LogicalUnit):
         test = n_img.mfree < needed_mem
         self._ErrorIf(test, self.ENODEN1, node,
                       "not enough memory to accomodate instance failovers"
         test = n_img.mfree < needed_mem
         self._ErrorIf(test, self.ENODEN1, node,
                       "not enough memory to accomodate instance failovers"

-                      " should node %s fail", prinode)
+                      " should node %s fail (%dMiB needed, %dMiB available)",
+                      prinode, needed_mem, n_img.mfree)

 
 

-  def _VerifyNodeFiles(self, ninfo, nresult, file_list, local_cksum,
-                       master_files):
-    """Verifies and computes the node required file checksums.
+  @classmethod
+  def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
+                   (files_all, files_all_opt, files_mc, files_vm)):
+    """Verifies file checksums collected from all nodes.

 
 

-    @type ninfo: L{objects.Node}
-    @param ninfo: the node to check
-    @param nresult: the remote results for the node
-    @param file_list: required list of files
-    @param local_cksum: dictionary of local files and their checksums
-    @param master_files: list of files that only masters should have
+    @param errorif: Callback for reporting errors
+    @param nodeinfo: List of L{objects.Node} objects
+    @param master_node: Name of master node
+    @param all_nvinfo: RPC results

 
     """
 
     """

-    node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    node_names = frozenset(node.name for node in nodeinfo)

 
 

-    remote_cksum = nresult.get(constants.NV_FILELIST, None)
-    test = not isinstance(remote_cksum, dict)
-    _ErrorIf(test, self.ENODEFILECHECK, node,
-             "node hasn't returned file checksum data")
-    if test:
-      return
+    assert master_node in node_names
+    assert (len(files_all | files_all_opt | files_mc | files_vm) ==
+            sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
+           "Found file listed in more than one file list"
+
+    # Define functions determining which nodes to consider for a file
+    file2nodefn = dict([(filename, fn)
+      for (files, fn) in [(files_all, None),
+                          (files_all_opt, None),
+                          (files_mc, lambda node: (node.master_candidate or
+                                                   node.name == master_node)),
+                          (files_vm, lambda node: node.vm_capable)]
+      for filename in files])
+
+    fileinfo = dict((filename, {}) for filename in file2nodefn.keys())
+
+    for node in nodeinfo:
+      nresult = all_nvinfo[node.name]
+
+      if nresult.fail_msg or not nresult.payload:
+        node_files = None
+      else:
+        node_files = nresult.payload.get(constants.NV_FILELIST, None)
+
+      test = not (node_files and isinstance(node_files, dict))
+      errorif(test, cls.ENODEFILECHECK, node.name,
+              "Node did not return file checksum data")
+      if test:
+        continue
+
+      for (filename, checksum) in node_files.items():
+        # Check if the file should be considered for a node
+        fn = file2nodefn[filename]
+        if fn is None or fn(node):
+          fileinfo[filename].setdefault(checksum, set()).add(node.name)
+
+    for (filename, checksums) in fileinfo.items():
+      assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
+
+      # Nodes having the file
+      with_file = frozenset(node_name
+                            for nodes in fileinfo[filename].values()
+                            for node_name in nodes)
+
+      # Nodes missing file
+      missing_file = node_names - with_file
+
+      if filename in files_all_opt:
+        # All or no nodes
+        errorif(missing_file and missing_file != node_names,
+                cls.ECLUSTERFILECHECK, None,
+                "File %s is optional, but it must exist on all or no nodes (not"
+                " found on %s)",
+                filename, utils.CommaJoin(utils.NiceSort(missing_file)))
+      else:
+        errorif(missing_file, cls.ECLUSTERFILECHECK, None,
+                "File %s is missing from node(s) %s", filename,
+                utils.CommaJoin(utils.NiceSort(missing_file)))
+
+      # See if there are multiple versions of the file
+      test = len(checksums) > 1
+      if test:
+        variants = ["variant %s on %s" %
+                    (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
+                    for (idx, (checksum, nodes)) in
+                      enumerate(sorted(checksums.items()))]
+      else:
+        variants = []

 
 

-    for file_name in file_list:
-      node_is_mc = ninfo.master_candidate
-      must_have = (file_name not in master_files) or node_is_mc
-      # missing
-      test1 = file_name not in remote_cksum
-      # invalid checksum
-      test2 = not test1 and remote_cksum[file_name] != local_cksum[file_name]
-      # existing and good
-      test3 = not test1 and remote_cksum[file_name] == local_cksum[file_name]
-      _ErrorIf(test1 and must_have, self.ENODEFILECHECK, node,
-               "file '%s' missing", file_name)
-      _ErrorIf(test2 and must_have, self.ENODEFILECHECK, node,
-               "file '%s' has wrong checksum", file_name)
-      # not candidate and this is not a must-have file
-      _ErrorIf(test2 and not must_have, self.ENODEFILECHECK, node,
-               "file '%s' should not exist on non master"
-               " candidates (and the file is outdated)", file_name)
-      # all good, except non-master/non-must have combination
-      _ErrorIf(test3 and not must_have, self.ENODEFILECHECK, node,
-               "file '%s' should not exist"
-               " on non master candidates", file_name)
+      errorif(test, cls.ECLUSTERFILECHECK, None,
+              "File %s found with %s different checksums (%s)",
+              filename, len(checksums), "; ".join(variants))

 
   def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
                       drbd_map):
 
   def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
                       drbd_map):

@@ -2173,19 +2217,14 @@ class LUClusterVerify(LogicalUnit):
     node_vol_should = {}
 
     # FIXME: verify OS list
     node_vol_should = {}
 
     # FIXME: verify OS list

+
+    # File verification
+    filemap = _ComputeAncillaryFiles(cluster, False)
+

     # do local checksums
     # do local checksums

-    master_files = [constants.CLUSTER_CONF_FILE]

     master_node = self.master_node = self.cfg.GetMasterNode()
     master_ip = self.cfg.GetMasterIP()
 
     master_node = self.master_node = self.cfg.GetMasterNode()
     master_ip = self.cfg.GetMasterIP()
 

-    file_names = ssconf.SimpleStore().GetFileList()
-    file_names.extend(constants.ALL_CERT_FILES)
-    file_names.extend(master_files)
-    if cluster.modify_etc_hosts:
-      file_names.append(constants.ETC_HOSTS)
-
-    local_checksums = utils.FingerprintFiles(file_names)
-

     # Compute the set of hypervisor parameters
     hvp_data = []
     for hv_name in hypervisors:
     # Compute the set of hypervisor parameters
     hvp_data = []
     for hv_name in hypervisors:

@@ -2207,7 +2246,10 @@ class LUClusterVerify(LogicalUnit):
 
     feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
     node_verify_param = {
 
     feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
     node_verify_param = {

-      constants.NV_FILELIST: file_names,
+      constants.NV_FILELIST:
+        utils.UniqueSequence(filename
+                             for files in filemap
+                             for filename in files),

       constants.NV_NODELIST: [node.name for node in nodeinfo
                               if not node.offline],
       constants.NV_HYPERVISOR: hypervisors,
       constants.NV_NODELIST: [node.name for node in nodeinfo
                               if not node.offline],
       constants.NV_HYPERVISOR: hypervisors,

@@ -2289,6 +2331,9 @@ class LUClusterVerify(LogicalUnit):
     feedback_fn("* Gathering disk information (%s nodes)" % len(nodelist))
     instdisk = self._CollectDiskInfo(nodelist, node_image, instanceinfo)
 
     feedback_fn("* Gathering disk information (%s nodes)" % len(nodelist))
     instdisk = self._CollectDiskInfo(nodelist, node_image, instanceinfo)
 

+    feedback_fn("* Verifying configuration file consistency")
+    self._VerifyFiles(_ErrorIf, nodeinfo, master_node, all_nvinfo, filemap)
+

     feedback_fn("* Verifying node status")
 
     refos_img = None
     feedback_fn("* Verifying node status")
 
     refos_img = None

@@ -2326,9 +2371,6 @@ class LUClusterVerify(LogicalUnit):
       nimg.call_ok = self._VerifyNode(node_i, nresult)
       self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
       self._VerifyNodeNetwork(node_i, nresult)
       nimg.call_ok = self._VerifyNode(node_i, nresult)
       self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
       self._VerifyNodeNetwork(node_i, nresult)

-      self._VerifyNodeFiles(node_i, nresult, file_names, local_checksums,
-                            master_files)
-

       self._VerifyOob(node_i, nresult)
 
       if nimg.vm_capable:
       self._VerifyOob(node_i, nresult)
 
       if nimg.vm_capable:

@@ -2871,6 +2913,12 @@ class LUClusterSetParams(LogicalUnit):
       utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
       self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
 
       utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
       self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
 

+      # TODO: we need a more general way to handle resetting
+      # cluster-level parameters to default values
+      if self.new_ndparams["oob_program"] == "":
+        self.new_ndparams["oob_program"] = \
+            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
+

     if self.op.nicparams:
       utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
       self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
     if self.op.nicparams:
       utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
       self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)

@@ -3112,6 +3160,50 @@ def _UploadHelper(lu, nodes, fname):
         lu.proc.LogWarning(msg)
 
 
         lu.proc.LogWarning(msg)
 
 

+def _ComputeAncillaryFiles(cluster, redist):
+  """Compute files external to Ganeti which need to be consistent.
+
+  @type redist: boolean
+  @param redist: Whether to include files which need to be redistributed
+
+  """
+  # Compute files for all nodes
+  files_all = set([
+    constants.SSH_KNOWN_HOSTS_FILE,
+    constants.CONFD_HMAC_KEY,
+    constants.CLUSTER_DOMAIN_SECRET_FILE,
+    ])
+
+  if not redist:
+    files_all.update(constants.ALL_CERT_FILES)
+    files_all.update(ssconf.SimpleStore().GetFileList())
+
+  if cluster.modify_etc_hosts:
+    files_all.add(constants.ETC_HOSTS)
+
+  # Files which must either exist on all nodes or on none
+  files_all_opt = set([
+    constants.RAPI_USERS_FILE,
+    ])
+
+  # Files which should only be on master candidates
+  files_mc = set()
+  if not redist:
+    files_mc.add(constants.CLUSTER_CONF_FILE)
+
+  # Files which should only be on VM-capable nodes
+  files_vm = set(filename
+    for hv_name in cluster.enabled_hypervisors
+    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
+
+  # Filenames must be unique
+  assert (len(files_all | files_all_opt | files_mc | files_vm) ==
+          sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
+         "Found file listed in more than one file list"
+
+  return (files_all, files_all_opt, files_mc, files_vm)
+
+

 def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
   """Distribute additional files which are part of the cluster configuration.
 
 def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
   """Distribute additional files which are part of the cluster configuration.
 

@@ -3125,40 +3217,42 @@ def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
   @param additional_vm: whether the additional nodes are vm-capable or not
 
   """
   @param additional_vm: whether the additional nodes are vm-capable or not
 
   """

-  # 1. Gather target nodes
-  myself = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
-  dist_nodes = lu.cfg.GetOnlineNodeList()
-  nvm_nodes = lu.cfg.GetNonVmCapableNodeList()
-  vm_nodes = [name for name in dist_nodes if name not in nvm_nodes]
+  # Gather target nodes
+  cluster = lu.cfg.GetClusterInfo()
+  master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
+
+  online_nodes = lu.cfg.GetOnlineNodeList()
+  vm_nodes = lu.cfg.GetVmCapableNodeList()
+

   if additional_nodes is not None:
   if additional_nodes is not None:

-    dist_nodes.extend(additional_nodes)
+    online_nodes.extend(additional_nodes)

     if additional_vm:
       vm_nodes.extend(additional_nodes)
     if additional_vm:
       vm_nodes.extend(additional_nodes)

-  if myself.name in dist_nodes:
-    dist_nodes.remove(myself.name)
-  if myself.name in vm_nodes:
-    vm_nodes.remove(myself.name)
-
-  # 2. Gather files to distribute
-  dist_files = set([constants.ETC_HOSTS,
-                    constants.SSH_KNOWN_HOSTS_FILE,
-                    constants.RAPI_CERT_FILE,
-                    constants.RAPI_USERS_FILE,
-                    constants.CONFD_HMAC_KEY,
-                    constants.CLUSTER_DOMAIN_SECRET_FILE,
-                   ])
-
-  vm_files = set()
-  enabled_hypervisors = lu.cfg.GetClusterInfo().enabled_hypervisors
-  for hv_name in enabled_hypervisors:
-    hv_class = hypervisor.GetHypervisor(hv_name)
-    vm_files.update(hv_class.GetAncillaryFiles())
-
-  # 3. Perform the files upload
-  for fname in dist_files:
-    _UploadHelper(lu, dist_nodes, fname)
-  for fname in vm_files:
-    _UploadHelper(lu, vm_nodes, fname)
+
+  # Never distribute to master node
+  for nodelist in [online_nodes, vm_nodes]:
+    if master_info.name in nodelist:
+      nodelist.remove(master_info.name)
+
+  # Gather file lists
+  (files_all, files_all_opt, files_mc, files_vm) = \
+    _ComputeAncillaryFiles(cluster, True)
+
+  # Never re-distribute configuration file from here
+  assert not (constants.CLUSTER_CONF_FILE in files_all or
+              constants.CLUSTER_CONF_FILE in files_vm)
+  assert not files_mc, "Master candidates not handled in this function"
+
+  filemap = [
+    (online_nodes, files_all),
+    (online_nodes, files_all_opt),
+    (vm_nodes, files_vm),
+    ]
+
+  # Upload the files
+  for (node_list, files) in filemap:
+    for fname in files:
+      _UploadHelper(lu, node_list, fname)

 
 
 class LUClusterRedistConf(NoHooksLU):
 
 
 class LUClusterRedistConf(NoHooksLU):

@@ -4037,10 +4131,16 @@ class _InstanceQuery(_QueryBase):
           bad_nodes.append(name)
         elif result.payload:
           for inst in result.payload:
           bad_nodes.append(name)
         elif result.payload:
           for inst in result.payload:

-            if all_info[inst].primary_node == name:
-              live_data.update(result.payload)
+            if inst in all_info:
+              if all_info[inst].primary_node == name:
+                live_data.update(result.payload)
+              else:
+                wrongnode_inst.add(inst)

             else:
             else:

-              wrongnode_inst.add(inst)
+              # orphan instance; we don't list it here as we don't
+              # handle this case yet in the output of instance listing
+              logging.warning("Orphan instance '%s' found on node %s",
+                              inst, name)

         # else no instance is alive
     else:
       live_data = {}
         # else no instance is alive
     else:
       live_data = {}

@@ -5697,7 +5797,9 @@ class LUInstanceRename(LogicalUnit):
       rename_file_storage = True
 
     self.cfg.RenameInstance(inst.name, self.op.new_name)
       rename_file_storage = True
 
     self.cfg.RenameInstance(inst.name, self.op.new_name)

-    # Change the instance lock. This is definitely safe while we hold the BGL
+    # 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

     self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
     self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
 
     self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
     self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
 

@@ -5864,6 +5966,17 @@ class LUInstanceFailover(LogicalUnit):
     self.needed_locks[locking.LEVEL_NODE] = []
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
     self.needed_locks[locking.LEVEL_NODE] = []
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 

+    ignore_consistency = self.op.ignore_consistency
+    shutdown_timeout = self.op.shutdown_timeout
+    self._migrater = TLMigrateInstance(self, self.op.instance_name,
+                                       cleanup=False,
+                                       iallocator=self.op.iallocator,
+                                       target_node=self.op.target_node,
+                                       failover=True,
+                                       ignore_consistency=ignore_consistency,
+                                       shutdown_timeout=shutdown_timeout)
+    self.tasklets = [self._migrater]
+

   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
       instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
       instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)

@@ -5883,13 +5996,14 @@ class LUInstanceFailover(LogicalUnit):
     This runs on master, primary and secondary nodes of the instance.
 
     """
     This runs on master, primary and secondary nodes of the instance.
 
     """

-    instance = self.instance
+    instance = self._migrater.instance

     source_node = instance.primary_node
     source_node = instance.primary_node

+    target_node = self._migrater.target_node

     env = {
       "IGNORE_CONSISTENCY": self.op.ignore_consistency,
       "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
       "OLD_PRIMARY": source_node,
     env = {
       "IGNORE_CONSISTENCY": self.op.ignore_consistency,
       "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
       "OLD_PRIMARY": source_node,

-      "NEW_PRIMARY": self.op.target_node,
+      "NEW_PRIMARY": target_node,

       }
 
     if instance.disk_template in constants.DTS_INT_MIRROR:
       }
 
     if instance.disk_template in constants.DTS_INT_MIRROR:

@@ -5906,171 +6020,9 @@ class LUInstanceFailover(LogicalUnit):
     """Build hooks nodes.
 
     """
     """Build hooks nodes.
 
     """

-    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
-    return (nl, nl + [self.instance.primary_node])
-
-  def CheckPrereq(self):
-    """Check prerequisites.
-
-    This checks that the instance is in the cluster.
-
-    """
-    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
-    assert self.instance is not None, \
-      "Cannot retrieve locked instance %s" % self.op.instance_name
-
-    bep = self.cfg.GetClusterInfo().FillBE(instance)
-    if instance.disk_template not in constants.DTS_MIRRORED:
-      raise errors.OpPrereqError("Instance's disk layout is not"
-                                 " mirrored, cannot failover.",
-                                 errors.ECODE_STATE)
-
-    if instance.disk_template in constants.DTS_EXT_MIRROR:
-      _CheckIAllocatorOrNode(self, "iallocator", "target_node")
-      if self.op.iallocator:
-        self._RunAllocator()
-        # Release all unnecessary node locks
-        nodes_keep = [instance.primary_node, self.op.target_node]
-        nodes_rel = [node for node in self.acquired_locks[locking.LEVEL_NODE]
-                     if node not in nodes_keep]
-        self.context.glm.release(locking.LEVEL_NODE, nodes_rel)
-        self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
-
-      # self.op.target_node is already populated, either directly or by the
-      # iallocator run
-      target_node = self.op.target_node
-
-    else:
-      secondary_nodes = instance.secondary_nodes
-      if not secondary_nodes:
-        raise errors.ConfigurationError("No secondary node but using"
-                                        " %s disk template" %
-                                        instance.disk_template)
-      target_node = secondary_nodes[0]
-
-      if self.op.iallocator or (self.op.target_node and
-                                self.op.target_node != target_node):
-        raise errors.OpPrereqError("Instances with disk template %s cannot"
-                                   " be failed over to arbitrary nodes"
-                                   " (neither an iallocator nor a target"
-                                   " node can be passed)" %
-                                   instance.disk_template, errors.ECODE_INVAL)
-    _CheckNodeOnline(self, target_node)
-    _CheckNodeNotDrained(self, target_node)
-
-    # Save target_node so that we can use it in BuildHooksEnv
-    self.op.target_node = target_node
-
-    if instance.admin_up:
-      # check memory requirements on the secondary node
-      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
-                           instance.name, bep[constants.BE_MEMORY],
-                           instance.hypervisor)
-    else:
-      self.LogInfo("Not checking memory on the secondary node as"
-                   " instance will not be started")
-
-    # check bridge existance
-    _CheckInstanceBridgesExist(self, instance, node=target_node)
-
-  def Exec(self, feedback_fn):
-    """Failover an instance.
-
-    The failover is done by shutting it down on its present node and
-    starting it on the secondary.
-
-    """
-    instance = self.instance
-    primary_node = self.cfg.GetNodeInfo(instance.primary_node)
-
-    source_node = instance.primary_node
-    target_node = self.op.target_node
-
-    if instance.admin_up:
-      feedback_fn("* checking disk consistency between source and target")
-      for dev in instance.disks:
-        # for drbd, these are drbd over lvm
-        if not _CheckDiskConsistency(self, dev, target_node, False):
-          if not self.op.ignore_consistency:
-            raise errors.OpExecError("Disk %s is degraded on target node,"
-                                     " aborting failover." % dev.iv_name)
-    else:
-      feedback_fn("* not checking disk consistency as instance is not running")
-
-    feedback_fn("* shutting down instance on source node")
-    logging.info("Shutting down instance %s on node %s",
-                 instance.name, source_node)
-
-    result = self.rpc.call_instance_shutdown(source_node, instance,
-                                             self.op.shutdown_timeout)
-    msg = result.fail_msg
-    if msg:
-      if self.op.ignore_consistency or primary_node.offline:
-        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)
-      else:
-        raise errors.OpExecError("Could not shutdown instance %s on"
-                                 " node %s: %s" %
-                                 (instance.name, source_node, msg))
-
-    feedback_fn("* deactivating the instance's disks on source node")
-    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
-      raise errors.OpExecError("Can't shut down the instance's disks.")
-
-    instance.primary_node = target_node
-    # distribute new instance config to the other nodes
-    self.cfg.Update(instance, feedback_fn)
-
-    # Only start the instance if it's marked as up
-    if instance.admin_up:
-      feedback_fn("* activating the instance's disks on target node")
-      logging.info("Starting instance %s on node %s",
-                   instance.name, target_node)
-
-      disks_ok, _ = _AssembleInstanceDisks(self, instance,
-                                           ignore_secondaries=True)
-      if not disks_ok:
-        _ShutdownInstanceDisks(self, instance)
-        raise errors.OpExecError("Can't activate the instance's disks")
-
-      feedback_fn("* starting the instance on the target node")
-      result = self.rpc.call_instance_start(target_node, instance, None, None)
-      msg = result.fail_msg
-      if msg:
-        _ShutdownInstanceDisks(self, instance)
-        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
-                                 (instance.name, target_node, msg))
-
-  def _RunAllocator(self):
-    """Run the allocator based on input opcode.
-
-    """
-    ial = IAllocator(self.cfg, self.rpc,
-                     mode=constants.IALLOCATOR_MODE_RELOC,
-                     name=self.instance.name,
-                     # TODO See why hail breaks with a single node below
-                     relocate_from=[self.instance.primary_node,
-                                    self.instance.primary_node],
-                     )
-
-    ial.Run(self.op.iallocator)
-
-    if not ial.success:
-      raise errors.OpPrereqError("Can't compute nodes using"
-                                 " iallocator '%s': %s" %
-                                 (self.op.iallocator, ial.info),
-                                 errors.ECODE_NORES)
-    if len(ial.result) != ial.required_nodes:
-      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
-                                 " of nodes (%s), required %s" %
-                                 (self.op.iallocator, len(ial.result),
-                                  ial.required_nodes), errors.ECODE_FAULT)
-    self.op.target_node = ial.result[0]
-    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
-                 self.instance.name, self.op.iallocator,
-                 utils.CommaJoin(ial.result))
+    instance = self._migrater.instance
+    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
+    return (nl, nl + [instance.primary_node])

 
 
 class LUInstanceMigrate(LogicalUnit):
 
 
 class LUInstanceMigrate(LogicalUnit):

@@ -6094,8 +6046,11 @@ class LUInstanceMigrate(LogicalUnit):
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
     self._migrater = TLMigrateInstance(self, self.op.instance_name,
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
     self._migrater = TLMigrateInstance(self, self.op.instance_name,

-                                       self.op.cleanup, self.op.iallocator,
-                                       self.op.target_node)
+                                       cleanup=self.op.cleanup,
+                                       iallocator=self.op.iallocator,
+                                       target_node=self.op.target_node,
+                                       failover=False,
+                                       fallback=self.op.allow_failover)

     self.tasklets = [self._migrater]
 
   def DeclareLocks(self, level):
     self.tasklets = [self._migrater]
 
   def DeclareLocks(self, level):

@@ -6355,8 +6310,9 @@ class LUNodeMigrate(LogicalUnit):
       logging.debug("Migrating instance %s", inst.name)
       names.append(inst.name)
 
       logging.debug("Migrating instance %s", inst.name)
       names.append(inst.name)
 

-      tasklets.append(TLMigrateInstance(self, inst.name, False,
-                                        self.op.iallocator, None))
+      tasklets.append(TLMigrateInstance(self, inst.name, cleanup=False,
+                                        iallocator=self.op.iallocator,
+                                        taget_node=None))

 
       if inst.disk_template in constants.DTS_EXT_MIRROR:
         # We need to lock all nodes, as the iallocator will choose the
 
       if inst.disk_template in constants.DTS_EXT_MIRROR:
         # We need to lock all nodes, as the iallocator will choose the

@@ -6403,10 +6359,28 @@ class TLMigrateInstance(Tasklet):
   @type live: boolean
   @ivar live: whether the migration will be done live or non-live;
       this variable is initalized only after CheckPrereq has run
   @type live: boolean
   @ivar live: whether the migration will be done live or non-live;
       this variable is initalized only after CheckPrereq has run

+  @type cleanup: boolean
+  @ivar cleanup: Wheater we cleanup from a failed migration
+  @type iallocator: string
+  @ivar iallocator: The iallocator used to determine target_node
+  @type target_node: string
+  @ivar target_node: If given, the target_node to reallocate the instance to
+  @type failover: boolean
+  @ivar failover: Whether operation results in failover or migration
+  @type fallback: boolean
+  @ivar fallback: Whether fallback to failover is allowed if migration not
+                  possible
+  @type ignore_consistency: boolean
+  @ivar ignore_consistency: Wheter we should ignore consistency between source
+                            and target node
+  @type shutdown_timeout: int
+  @ivar shutdown_timeout: In case of failover timeout of the shutdown

 
   """
 
   """

-  def __init__(self, lu, instance_name, cleanup,
-               iallocator=None, target_node=None):
+  def __init__(self, lu, instance_name, cleanup=False, iallocator=None,
+               target_node=None, failover=False, fallback=False,
+               ignore_consistency=False,
+               shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT):

     """Initializes this class.
 
     """
     """Initializes this class.
 
     """

@@ -6418,6 +6392,10 @@ class TLMigrateInstance(Tasklet):
     self.live = False # will be overridden later
     self.iallocator = iallocator
     self.target_node = target_node
     self.live = False # will be overridden later
     self.iallocator = iallocator
     self.target_node = target_node

+    self.failover = failover
+    self.fallback = fallback
+    self.ignore_consistency = ignore_consistency
+    self.shutdown_timeout = shutdown_timeout

 
   def CheckPrereq(self):
     """Check prerequisites.
 
   def CheckPrereq(self):
     """Check prerequisites.

@@ -6430,9 +6408,19 @@ class TLMigrateInstance(Tasklet):
     assert instance is not None
     self.instance = instance
 
     assert instance is not None
     self.instance = instance
 

+    if (not self.cleanup and not instance.admin_up and not self.failover and
+        self.fallback):
+      self.lu.LogInfo("Instance is marked down, fallback allowed, switching"
+                      " to failover")
+      self.failover = True
+

     if instance.disk_template not in constants.DTS_MIRRORED:
     if instance.disk_template not in constants.DTS_MIRRORED:

+      if self.failover:
+        text = "failovers"
+      else:
+        text = "migrations"

       raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
       raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"

-                                 " migrations" % instance.disk_template,
+                                 " %s" % (instance.disk_template, text),

                                  errors.ECODE_STATE)
 
     if instance.disk_template in constants.DTS_EXT_MIRROR:
                                  errors.ECODE_STATE)
 
     if instance.disk_template in constants.DTS_EXT_MIRROR:

@@ -6460,31 +6448,47 @@ class TLMigrateInstance(Tasklet):
                                         " %s disk template" %
                                         instance.disk_template)
       target_node = secondary_nodes[0]
                                         " %s disk template" %
                                         instance.disk_template)
       target_node = secondary_nodes[0]

-      if self.lu.op.iallocator or (self.lu.op.target_node and
-                                   self.lu.op.target_node != target_node):
+      if self.iallocator or (self.target_node and
+                             self.target_node != target_node):
+        if self.failover:
+          text = "failed over"
+        else:
+          text = "migrated"

         raise errors.OpPrereqError("Instances with disk template %s cannot"
         raise errors.OpPrereqError("Instances with disk template %s cannot"

-                                   " be migrated over to arbitrary nodes"
+                                   " be %s over to arbitrary nodes"

                                    " (neither an iallocator nor a target"
                                    " node can be passed)" %
                                    " (neither an iallocator nor a target"
                                    " node can be passed)" %

-                                   instance.disk_template, errors.ECODE_INVAL)
+                                   (text, instance.disk_template),
+                                   errors.ECODE_INVAL)

 
     i_be = self.cfg.GetClusterInfo().FillBE(instance)
 
     # check memory requirements on the secondary node
 
     i_be = self.cfg.GetClusterInfo().FillBE(instance)
 
     # check memory requirements on the secondary node

-    _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
-                         instance.name, i_be[constants.BE_MEMORY],
-                         instance.hypervisor)
+    if not self.failover or instance.admin_up:
+      _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
+                           instance.name, i_be[constants.BE_MEMORY],
+                           instance.hypervisor)
+    else:
+      self.lu.LogInfo("Not checking memory on the secondary node as"
+                      " instance will not be started")

 
     # check bridge existance
     _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
 
     if not self.cleanup:
       _CheckNodeNotDrained(self.lu, target_node)
 
     # check bridge existance
     _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
 
     if not self.cleanup:
       _CheckNodeNotDrained(self.lu, target_node)

-      result = self.rpc.call_instance_migratable(instance.primary_node,
-                                                 instance)
-      result.Raise("Can't migrate, please use failover",
-                   prereq=True, ecode=errors.ECODE_STATE)
+      if not self.failover:
+        result = self.rpc.call_instance_migratable(instance.primary_node,
+                                                   instance)
+        if result.fail_msg and self.fallback:
+          self.lu.LogInfo("Can't migrate, instance offline, fallback to"
+                          " failover")
+          self.failover = True
+        else:
+          result.Raise("Can't migrate, please use failover",
+                       prereq=True, ecode=errors.ECODE_STATE)

 
 

+    assert not (self.failover and self.cleanup)

 
   def _RunAllocator(self):
     """Run the allocator based on input opcode.
 
   def _RunAllocator(self):
     """Run the allocator based on input opcode.

@@ -6515,24 +6519,29 @@ class TLMigrateInstance(Tasklet):
                  self.instance_name, self.iallocator,
                  utils.CommaJoin(ial.result))
 
                  self.instance_name, self.iallocator,
                  utils.CommaJoin(ial.result))
 

-    if self.lu.op.live is not None and self.lu.op.mode is not None:
-      raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
-                                 " parameters are accepted",
-                                 errors.ECODE_INVAL)
-    if self.lu.op.live is not None:
-      if self.lu.op.live:
-        self.lu.op.mode = constants.HT_MIGRATION_LIVE
-      else:
-        self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
-      # reset the 'live' parameter to None so that repeated
-      # invocations of CheckPrereq do not raise an exception
-      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)
-      self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
-
-    self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
+    if not self.failover:
+      if self.lu.op.live is not None and self.lu.op.mode is not None:
+        raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
+                                   " parameters are accepted",
+                                   errors.ECODE_INVAL)
+      if self.lu.op.live is not None:
+        if self.lu.op.live:
+          self.lu.op.mode = constants.HT_MIGRATION_LIVE
+        else:
+          self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
+        # reset the 'live' parameter to None so that repeated
+        # invocations of CheckPrereq do not raise an exception
+        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)
+        self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
+
+      self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
+    else:
+      # Failover is never live
+      self.live = False

 
   def _WaitUntilSync(self):
     """Poll with custom rpc for disk sync.
 
   def _WaitUntilSync(self):
     """Poll with custom rpc for disk sync.

@@ -6759,7 +6768,6 @@ class TLMigrateInstance(Tasklet):
                                (instance.name, msg))
 
     self.feedback_fn("* migrating instance to %s" % target_node)
                                (instance.name, msg))
 
     self.feedback_fn("* migrating instance to %s" % target_node)

-    time.sleep(10)

     result = self.rpc.call_instance_migrate(source_node, instance,
                                             self.nodes_ip[target_node],
                                             self.live)
     result = self.rpc.call_instance_migrate(source_node, instance,
                                             self.nodes_ip[target_node],
                                             self.live)

@@ -6772,7 +6780,6 @@ class TLMigrateInstance(Tasklet):
       self._RevertDiskStatus()
       raise errors.OpExecError("Could not migrate instance %s: %s" %
                                (instance.name, msg))
       self._RevertDiskStatus()
       raise errors.OpExecError("Could not migrate instance %s: %s" %
                                (instance.name, msg))

-    time.sleep(10)

 
     instance.primary_node = target_node
     # distribute new instance config to the other nodes
 
     instance.primary_node = target_node
     # distribute new instance config to the other nodes

@@ -6798,14 +6805,82 @@ class TLMigrateInstance(Tasklet):
 
     self.feedback_fn("* done")
 
 
     self.feedback_fn("* done")
 

+  def _ExecFailover(self):
+    """Failover an instance.
+
+    The failover is done by shutting it down on its present node and
+    starting it on the secondary.
+
+    """
+    instance = self.instance
+    primary_node = self.cfg.GetNodeInfo(instance.primary_node)
+
+    source_node = instance.primary_node
+    target_node = self.target_node
+
+    if instance.admin_up:
+      self.feedback_fn("* checking disk consistency between source and target")
+      for dev in instance.disks:
+        # for drbd, these are drbd over lvm
+        if not _CheckDiskConsistency(self, dev, target_node, False):
+          if not self.ignore_consistency:
+            raise errors.OpExecError("Disk %s is degraded on target node,"
+                                     " aborting failover." % dev.iv_name)
+    else:
+      self.feedback_fn("* not checking disk consistency as instance is not"
+                       " running")
+
+    self.feedback_fn("* shutting down instance on source node")
+    logging.info("Shutting down instance %s on node %s",
+                 instance.name, source_node)
+
+    result = self.rpc.call_instance_shutdown(source_node, instance,
+                                             self.shutdown_timeout)
+    msg = result.fail_msg
+    if msg:
+      if self.ignore_consistency or primary_node.offline:
+        self.lu.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" %
+                                 (instance.name, source_node, msg))
+
+    self.feedback_fn("* deactivating the instance's disks on source node")
+    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
+      raise errors.OpExecError("Can't shut down the instance's disks.")
+
+    instance.primary_node = target_node
+    # distribute new instance config to the other nodes
+    self.cfg.Update(instance, self.feedback_fn)
+
+    # Only start the instance if it's marked as up
+    if instance.admin_up:
+      self.feedback_fn("* activating the instance's disks on target node")
+      logging.info("Starting instance %s on node %s",
+                   instance.name, target_node)
+
+      disks_ok, _ = _AssembleInstanceDisks(self, instance,
+                                           ignore_secondaries=True)
+      if not disks_ok:
+        _ShutdownInstanceDisks(self, instance)
+        raise errors.OpExecError("Can't activate the instance's disks")
+
+      self.feedback_fn("* starting the instance on the target node")
+      result = self.rpc.call_instance_start(target_node, instance, None, None)
+      msg = result.fail_msg
+      if msg:
+        _ShutdownInstanceDisks(self, instance)
+        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
+                                 (instance.name, target_node, msg))
+

   def Exec(self, feedback_fn):
     """Perform the migration.
 
     """
   def Exec(self, feedback_fn):
     """Perform the migration.
 
     """

-    feedback_fn("Migrating instance %s" % self.instance.name)
-

     self.feedback_fn = feedback_fn
     self.feedback_fn = feedback_fn

-

     self.source_node = self.instance.primary_node
 
     # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
     self.source_node = self.instance.primary_node
 
     # FIXME: if we implement migrate-to-any in DRBD, this needs fixing

@@ -6820,10 +6895,16 @@ class TLMigrateInstance(Tasklet):
       self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
       }
 
       self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
       }
 

-    if self.cleanup:
-      return self._ExecCleanup()
+    if self.failover:
+      feedback_fn("Failover instance %s" % self.instance.name)
+      self._ExecFailover()

     else:
     else:

-      return self._ExecMigration()
+      feedback_fn("Migrating instance %s" % self.instance.name)
+
+      if self.cleanup:
+        return self._ExecCleanup()
+      else:
+        return self._ExecMigration()

 
 
 def _CreateBlockDev(lu, node, instance, device, force_create,
 
 
 def _CreateBlockDev(lu, node, instance, device, force_create,

@@ -7224,7 +7305,7 @@ def _ComputeDiskSizePerVG(disk_template, disks):
     vgs = {}
     for disk in disks:
       vgs[disk[constants.IDISK_VG]] = \
     vgs = {}
     for disk in disks:
       vgs[disk[constants.IDISK_VG]] = \

-        vgs.get(constants.IDISK_VG, 0) + disk[constants.IDISK_MODE] + payload
+        vgs.get(constants.IDISK_VG, 0) + disk[constants.IDISK_SIZE] + payload

 
     return vgs
 
 
     return vgs
 

@@ -9355,23 +9436,33 @@ class LUInstanceQueryData(NoHooksLU):
 
   def ExpandNames(self):
     self.needed_locks = {}
 
   def ExpandNames(self):
     self.needed_locks = {}

-    self.share_locks = dict.fromkeys(locking.LEVELS, 1)

 
 

-    if self.op.instances:
-      self.wanted_names = []
-      for name in self.op.instances:
-        full_name = _ExpandInstanceName(self.cfg, name)
-        self.wanted_names.append(full_name)
-      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
+    # Use locking if requested or when non-static information is wanted
+    if not (self.op.static or self.op.use_locking):
+      self.LogWarning("Non-static data requested, locks need to be acquired")
+      self.op.use_locking = True
+
+    if self.op.instances or not self.op.use_locking:
+      # Expand instance names right here
+      self.wanted_names = _GetWantedInstances(self, self.op.instances)

     else:
     else:

+      # Will use acquired locks

       self.wanted_names = None
       self.wanted_names = None

-      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET

 
 

-    self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+    if self.op.use_locking:
+      self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+
+      if self.wanted_names is None:
+        self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
+      else:
+        self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
+
+      self.needed_locks[locking.LEVEL_NODE] = []
+      self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE

 
   def DeclareLocks(self, level):
 
   def DeclareLocks(self, level):

-    if level == locking.LEVEL_NODE:
+    if self.op.use_locking and level == locking.LEVEL_NODE:

       self._LockInstancesNodes()
 
   def CheckPrereq(self):
       self._LockInstancesNodes()
 
   def CheckPrereq(self):

@@ -9381,10 +9472,11 @@ class LUInstanceQueryData(NoHooksLU):
 
     """
     if self.wanted_names is None:
 
     """
     if self.wanted_names is None:

+      assert self.op.use_locking, "Locking was not used"

       self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
 
       self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
 

-    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
-                             in self.wanted_names]
+    self.wanted_instances = [self.cfg.GetInstanceInfo(name)
+                             for name in self.wanted_names]

 
   def _ComputeBlockdevStatus(self, node, instance_name, dev):
     """Returns the status of a block device
 
   def _ComputeBlockdevStatus(self, node, instance_name, dev):
     """Returns the status of a block device

@@ -9430,7 +9522,7 @@ class LUInstanceQueryData(NoHooksLU):
     else:
       dev_children = []
 
     else:
       dev_children = []
 

-    data = {
+    return {

       "iv_name": dev.iv_name,
       "dev_type": dev.dev_type,
       "logical_id": dev.logical_id,
       "iv_name": dev.iv_name,
       "dev_type": dev.dev_type,
       "logical_id": dev.logical_id,

@@ -9442,8 +9534,6 @@ class LUInstanceQueryData(NoHooksLU):
       "size": dev.size,
       }
 
       "size": dev.size,
       }
 

-    return data
-

   def Exec(self, feedback_fn):
     """Gather and return data"""
     result = {}
   def Exec(self, feedback_fn):
     """Gather and return data"""
     result = {}

@@ -9471,7 +9561,7 @@ class LUInstanceQueryData(NoHooksLU):
       disks = [self._ComputeDiskStatus(instance, None, device)
                for device in instance.disks]
 
       disks = [self._ComputeDiskStatus(instance, None, device)
                for device in instance.disks]
 

-      idict = {
+      result[instance.name] = {

         "name": instance.name,
         "config_state": config_state,
         "run_state": remote_state,
         "name": instance.name,
         "config_state": config_state,
         "run_state": remote_state,

@@ -9496,8 +9586,6 @@ class LUInstanceQueryData(NoHooksLU):
         "uuid": instance.uuid,
         }
 
         "uuid": instance.uuid,
         }
 

-      result[instance.name] = idict
-

     return result
 
 
     return result
 
 

@@ -11084,8 +11172,8 @@ class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
   This is an abstract class which is the parent of all the other tags LUs.
 
   """
   This is an abstract class which is the parent of all the other tags LUs.
 
   """

-

   def ExpandNames(self):
   def ExpandNames(self):

+    self.group_uuid = None

     self.needed_locks = {}
     if self.op.kind == constants.TAG_NODE:
       self.op.name = _ExpandNodeName(self.cfg, self.op.name)
     self.needed_locks = {}
     if self.op.kind == constants.TAG_NODE:
       self.op.name = _ExpandNodeName(self.cfg, self.op.name)

@@ -11093,6 +11181,8 @@ class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
     elif self.op.kind == constants.TAG_INSTANCE:
       self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
       self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
     elif self.op.kind == constants.TAG_INSTANCE:
       self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
       self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name

+    elif self.op.kind == constants.TAG_NODEGROUP:
+      self.group_uuid = self.cfg.LookupNodeGroup(self.op.name)

 
     # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
     # not possible to acquire the BGL based on opcode parameters)
 
     # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
     # not possible to acquire the BGL based on opcode parameters)

@@ -11107,6 +11197,8 @@ class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
       self.target = self.cfg.GetNodeInfo(self.op.name)
     elif self.op.kind == constants.TAG_INSTANCE:
       self.target = self.cfg.GetInstanceInfo(self.op.name)
       self.target = self.cfg.GetNodeInfo(self.op.name)
     elif self.op.kind == constants.TAG_INSTANCE:
       self.target = self.cfg.GetInstanceInfo(self.op.name)

+    elif self.op.kind == constants.TAG_NODEGROUP:
+      self.target = self.cfg.GetNodeGroup(self.group_uuid)

     else:
       raise errors.OpPrereqError("Wrong tag type requested (%s)" %
                                  str(self.op.kind), errors.ECODE_INVAL)
     else:
       raise errors.OpPrereqError("Wrong tag type requested (%s)" %
                                  str(self.op.kind), errors.ECODE_INVAL)

@@ -11162,6 +11254,8 @@ class LUTagsSearch(NoHooksLU):
     tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
     nlist = cfg.GetAllNodesInfo().values()
     tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
     tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
     nlist = cfg.GetAllNodesInfo().values()
     tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])

+    tgts.extend(("/nodegroup/%s" % n.name, n)
+                for n in cfg.GetAllNodeGroupsInfo().values())

     results = []
     for path, target in tgts:
       for tag in target.GetTags():
     results = []
     for path, target in tgts:
       for tag in target.GetTags():

@@ -11799,8 +11893,61 @@ class IAllocator(object):
     if not isinstance(rdict["result"], list):
       raise errors.OpExecError("Can't parse iallocator results: 'result' key"
                                " is not a list")
     if not isinstance(rdict["result"], list):
       raise errors.OpExecError("Can't parse iallocator results: 'result' key"
                                " is not a list")

+
+    if self.mode == constants.IALLOCATOR_MODE_RELOC:
+      assert self.relocate_from is not None
+      assert self.required_nodes == 1
+
+      node2group = dict((name, ndata["group"])
+                        for (name, ndata) in self.in_data["nodes"].items())
+
+      fn = compat.partial(self._NodesToGroups, node2group,
+                          self.in_data["nodegroups"])
+
+      request_groups = fn(self.relocate_from)
+      result_groups = fn(rdict["result"])
+
+      if result_groups != request_groups:
+        raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
+                                 " differ from original groups (%s)" %
+                                 (utils.CommaJoin(result_groups),
+                                  utils.CommaJoin(request_groups)))
+

     self.out_data = rdict
 
     self.out_data = rdict
 

+  @staticmethod
+  def _NodesToGroups(node2group, groups, nodes):
+    """Returns a list of unique group names for a list of nodes.
+
+    @type node2group: dict
+    @param node2group: Map from node name to group UUID
+    @type groups: dict
+    @param groups: Group information
+    @type nodes: list
+    @param nodes: Node names
+
+    """
+    result = set()
+
+    for node in nodes:
+      try:
+        group_uuid = node2group[node]
+      except KeyError:
+        # Ignore unknown node
+        pass
+      else:
+        try:
+          group = groups[group_uuid]
+        except KeyError:
+          # Can't find group, let's use UUID
+          group_name = group_uuid
+        else:
+          group_name = group["name"]
+
+        result.add(group_name)
+
+    return sorted(result)
+

 
 class LUTestAllocator(NoHooksLU):
   """Run allocator tests.
 
 class LUTestAllocator(NoHooksLU):
   """Run allocator tests.