gnt-debug: rename allocator to iallocator
[ganeti-local] / lib / cmdlib.py
index 7ba8630..98a056c 100644 (file)
@@ -57,6 +57,7 @@ from ganeti import netutils
 from ganeti import query
 from ganeti import qlang
 from ganeti import opcodes
+from ganeti import ht
 
 import ganeti.masterd.instance # pylint: disable-msg=W0611
 
@@ -1153,7 +1154,7 @@ def _CheckIAllocatorOrNode(lu, iallocator_slot, node_slot):
   iallocator = getattr(lu.op, iallocator_slot, None)
 
   if node is not None and iallocator is not None:
-    raise errors.OpPrereqError("Do not specify both, iallocator and node.",
+    raise errors.OpPrereqError("Do not specify both, iallocator and node",
                                errors.ECODE_INVAL)
   elif node is None and iallocator is None:
     default_iallocator = lu.cfg.GetDefaultIAllocator()
@@ -1161,10 +1162,10 @@ def _CheckIAllocatorOrNode(lu, iallocator_slot, node_slot):
       setattr(lu.op, iallocator_slot, default_iallocator)
     else:
       raise errors.OpPrereqError("No iallocator or node given and no"
-                                 " cluster-wide default iallocator found."
-                                 " Please specify either an iallocator or a"
+                                 " cluster-wide default iallocator found;"
+                                 " please specify either an iallocator or a"
                                  " node, or set a cluster-wide default"
-                                 " iallocator.")
+                                 " iallocator")
 
 
 class LUClusterPostInit(LogicalUnit):
@@ -1253,7 +1254,7 @@ class LUClusterDestroy(LogicalUnit):
 
 
 def _VerifyCertificate(filename):
-  """Verifies a certificate for LUClusterVerify.
+  """Verifies a certificate for LUClusterVerifyConfig.
 
   @type filename: string
   @param filename: Path to PEM file
@@ -1263,7 +1264,7 @@ def _VerifyCertificate(filename):
     cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
                                            utils.ReadFile(filename))
   except Exception, err: # pylint: disable-msg=W0703
-    return (LUClusterVerify.ETYPE_ERROR,
+    return (LUClusterVerifyConfig.ETYPE_ERROR,
             "Failed to load X509 certificate %s: %s" % (filename, err))
 
   (errcode, msg) = \
@@ -1278,21 +1279,52 @@ def _VerifyCertificate(filename):
   if errcode is None:
     return (None, fnamemsg)
   elif errcode == utils.CERT_WARNING:
-    return (LUClusterVerify.ETYPE_WARNING, fnamemsg)
+    return (LUClusterVerifyConfig.ETYPE_WARNING, fnamemsg)
   elif errcode == utils.CERT_ERROR:
-    return (LUClusterVerify.ETYPE_ERROR, fnamemsg)
+    return (LUClusterVerifyConfig.ETYPE_ERROR, fnamemsg)
 
   raise errors.ProgrammerError("Unhandled certificate error code %r" % errcode)
 
 
-class LUClusterVerify(LogicalUnit):
-  """Verifies the cluster status.
+def _GetAllHypervisorParameters(cluster, instances):
+  """Compute the set of all hypervisor parameters.
+
+  @type cluster: L{objects.Cluster}
+  @param cluster: the cluster object
+  @param instances: list of L{objects.Instance}
+  @param instances: additional instances from which to obtain parameters
+  @rtype: list of (origin, hypervisor, parameters)
+  @return: a list with all parameters found, indicating the hypervisor they
+       apply to, and the origin (can be "cluster", "os X", or "instance Y")
 
   """
-  HPATH = "cluster-verify"
-  HTYPE = constants.HTYPE_CLUSTER
-  REQ_BGL = False
+  hvp_data = []
 
+  for hv_name in cluster.enabled_hypervisors:
+    hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
+
+  for os_name, os_hvp in cluster.os_hvp.items():
+    for hv_name, hv_params in os_hvp.items():
+      if hv_params:
+        full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
+        hvp_data.append(("os %s" % os_name, hv_name, full_params))
+
+  # TODO: collapse identical parameter values in a single one
+  for instance in instances:
+    if instance.hvparams:
+      hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
+                       cluster.FillHV(instance)))
+
+  return hvp_data
+
+
+class _VerifyErrors(object):
+  """Mix-in for cluster/group verify LUs.
+
+  It provides _Error and _ErrorIf, and updates the self.bad boolean. (Expects
+  self.op and self._feedback_fn to be available.)
+
+  """
   TCLUSTER = "cluster"
   TNODE = "node"
   TINSTANCE = "instance"
@@ -1300,6 +1332,8 @@ class LUClusterVerify(LogicalUnit):
   ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
   ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")
   ECLUSTERFILECHECK = (TCLUSTER, "ECLUSTERFILECHECK")
+  ECLUSTERDANGLINGNODES = (TNODE, "ECLUSTERDANGLINGNODES")
+  ECLUSTERDANGLINGINST = (TNODE, "ECLUSTERDANGLINGINST")
   EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
   EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
   EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
@@ -1329,6 +1363,138 @@ class LUClusterVerify(LogicalUnit):
   ETYPE_ERROR = "ERROR"
   ETYPE_WARNING = "WARNING"
 
+  def _Error(self, ecode, item, msg, *args, **kwargs):
+    """Format an error message.
+
+    Based on the opcode's error_codes parameter, either format a
+    parseable error code, or a simpler error string.
+
+    This must be called only from Exec and functions called from Exec.
+
+    """
+    ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
+    itype, etxt = ecode
+    # first complete the msg
+    if args:
+      msg = msg % args
+    # then format the whole message
+    if self.op.error_codes: # This is a mix-in. pylint: disable-msg=E1101
+      msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
+    else:
+      if item:
+        item = " " + item
+      else:
+        item = ""
+      msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
+    # and finally report it via the feedback_fn
+    self._feedback_fn("  - %s" % msg) # Mix-in. pylint: disable-msg=E1101
+
+  def _ErrorIf(self, cond, *args, **kwargs):
+    """Log an error message if the passed condition is True.
+
+    """
+    cond = (bool(cond)
+            or self.op.debug_simulate_errors) # pylint: disable-msg=E1101
+    if cond:
+      self._Error(*args, **kwargs)
+    # do not mark the operation as failed for WARN cases only
+    if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
+      self.bad = self.bad or cond
+
+
+class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
+  """Verifies the cluster config.
+
+  """
+  REQ_BGL = False
+
+  def _VerifyHVP(self, hvp_data):
+    """Verifies locally the syntax of the hypervisor parameters.
+
+    """
+    for item, hv_name, hv_params in hvp_data:
+      msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
+             (item, hv_name))
+      try:
+        hv_class = hypervisor.GetHypervisor(hv_name)
+        utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
+        hv_class.CheckParameterSyntax(hv_params)
+      except errors.GenericError, err:
+        self._ErrorIf(True, self.ECLUSTERCFG, None, msg % str(err))
+
+  def ExpandNames(self):
+    self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
+    self.all_node_info = self.cfg.GetAllNodesInfo()
+    self.all_inst_info = self.cfg.GetAllInstancesInfo()
+    self.needed_locks = {}
+
+  def Exec(self, feedback_fn):
+    """Verify integrity of cluster, performing various test on nodes.
+
+    """
+    self.bad = False
+    self._feedback_fn = feedback_fn
+
+    feedback_fn("* Verifying cluster config")
+
+    for msg in self.cfg.VerifyConfig():
+      self._ErrorIf(True, self.ECLUSTERCFG, None, msg)
+
+    feedback_fn("* Verifying cluster certificate files")
+
+    for cert_filename in constants.ALL_CERT_FILES:
+      (errcode, msg) = _VerifyCertificate(cert_filename)
+      self._ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
+
+    feedback_fn("* Verifying hypervisor parameters")
+
+    self._VerifyHVP(_GetAllHypervisorParameters(self.cfg.GetClusterInfo(),
+                                                self.all_inst_info.values()))
+
+    feedback_fn("* Verifying all nodes belong to an existing group")
+
+    # We do this verification here because, should this bogus circumstance
+    # occur, it would never be catched by VerifyGroup, which only acts on
+    # nodes/instances reachable from existing node groups.
+
+    dangling_nodes = set(node.name for node in self.all_node_info.values()
+                         if node.group not in self.all_group_info)
+
+    dangling_instances = {}
+    no_node_instances = []
+
+    for inst in self.all_inst_info.values():
+      if inst.primary_node in dangling_nodes:
+        dangling_instances.setdefault(inst.primary_node, []).append(inst.name)
+      elif inst.primary_node not in self.all_node_info:
+        no_node_instances.append(inst.name)
+
+    pretty_dangling = [
+        "%s (%s)" %
+        (node.name,
+         utils.CommaJoin(dangling_instances.get(node.name,
+                                                ["no instances"])))
+        for node in dangling_nodes]
+
+    self._ErrorIf(bool(dangling_nodes), self.ECLUSTERDANGLINGNODES, None,
+                  "the following nodes (and their instances) belong to a non"
+                  " existing group: %s", utils.CommaJoin(pretty_dangling))
+
+    self._ErrorIf(bool(no_node_instances), self.ECLUSTERDANGLINGINST, None,
+                  "the following instances have a non-existing primary-node:"
+                  " %s", utils.CommaJoin(no_node_instances))
+
+    return (not self.bad, [g.name for g in self.all_group_info.values()])
+
+
+class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
+  """Verifies the status of a node group.
+
+  """
+  HPATH = "cluster-verify"
+  HTYPE = constants.HTYPE_CLUSTER
+  REQ_BGL = False
+
   _HOOKS_INDENT_RE = re.compile("^", re.M)
 
   class NodeImage(object):
@@ -1382,48 +1548,90 @@ class LUClusterVerify(LogicalUnit):
       self.oslist = {}
 
   def ExpandNames(self):
+    # This raises errors.OpPrereqError on its own:
+    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
+
+    all_node_info = self.cfg.GetAllNodesInfo()
+    all_inst_info = self.cfg.GetAllInstancesInfo()
+
+    node_names = set(node.name
+                     for node in all_node_info.values()
+                     if node.group == self.group_uuid)
+
+    inst_names = [inst.name
+                  for inst in all_inst_info.values()
+                  if inst.primary_node in node_names]
+
+    # In Exec(), we warn about mirrored instances that have primary and
+    # secondary living in separate node groups. To fully verify that
+    # volumes for these instances are healthy, we will need to do an
+    # extra call to their secondaries. We ensure here those nodes will
+    # be locked.
+    for inst in inst_names:
+      if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
+        node_names.update(all_inst_info[inst].secondary_nodes)
+
     self.needed_locks = {
-      locking.LEVEL_NODE: locking.ALL_SET,
-      locking.LEVEL_INSTANCE: locking.ALL_SET,
+      locking.LEVEL_NODEGROUP: [self.group_uuid],
+      locking.LEVEL_NODE: list(node_names),
+      locking.LEVEL_INSTANCE: inst_names,
     }
+
     self.share_locks = dict.fromkeys(locking.LEVELS, 1)
 
-  def _Error(self, ecode, item, msg, *args, **kwargs):
-    """Format an error message.
+  def CheckPrereq(self):
+    self.all_node_info = self.cfg.GetAllNodesInfo()
+    self.all_inst_info = self.cfg.GetAllInstancesInfo()
 
-    Based on the opcode's error_codes parameter, either format a
-    parseable error code, or a simpler error string.
+    group_nodes = set(node.name
+                      for node in self.all_node_info.values()
+                      if node.group == self.group_uuid)
 
-    This must be called only from Exec and functions called from Exec.
+    group_instances = set(inst.name
+                          for inst in self.all_inst_info.values()
+                          if inst.primary_node in group_nodes)
 
-    """
-    ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
-    itype, etxt = ecode
-    # first complete the msg
-    if args:
-      msg = msg % args
-    # then format the whole message
-    if self.op.error_codes:
-      msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
-    else:
-      if item:
-        item = " " + item
-      else:
-        item = ""
-      msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
-    # and finally report it via the feedback_fn
-    self._feedback_fn("  - %s" % msg)
+    unlocked_nodes = \
+        group_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
 
-  def _ErrorIf(self, cond, *args, **kwargs):
-    """Log an error message if the passed condition is True.
+    unlocked_instances = \
+        group_instances.difference(self.glm.list_owned(locking.LEVEL_INSTANCE))
 
-    """
-    cond = bool(cond) or self.op.debug_simulate_errors
-    if cond:
-      self._Error(*args, **kwargs)
-    # do not mark the operation as failed for WARN cases only
-    if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
-      self.bad = self.bad or cond
+    if unlocked_nodes:
+      raise errors.OpPrereqError("missing lock for nodes: %s" %
+                                 utils.CommaJoin(unlocked_nodes))
+
+    if unlocked_instances:
+      raise errors.OpPrereqError("missing lock for instances: %s" %
+                                 utils.CommaJoin(unlocked_instances))
+
+    self.my_node_names = utils.NiceSort(group_nodes)
+    self.my_inst_names = utils.NiceSort(group_instances)
+
+    self.my_node_info = dict((name, self.all_node_info[name])
+                             for name in self.my_node_names)
+
+    self.my_inst_info = dict((name, self.all_inst_info[name])
+                             for name in self.my_inst_names)
+
+    # We detect here the nodes that will need the extra RPC calls for verifying
+    # split LV volumes; they should be locked.
+    extra_lv_nodes = set()
+
+    for inst in self.my_inst_info.values():
+      if inst.disk_template in constants.DTS_INT_MIRROR:
+        group = self.my_node_info[inst.primary_node].group
+        for nname in inst.secondary_nodes:
+          if self.all_node_info[nname].group != group:
+            extra_lv_nodes.add(nname)
+
+    unlocked_lv_nodes = \
+        extra_lv_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
+
+    if unlocked_lv_nodes:
+      raise errors.OpPrereqError("these nodes could be locked: %s" %
+                                 utils.CommaJoin(unlocked_lv_nodes))
+    self.extra_lv_nodes = list(extra_lv_nodes)
 
   def _VerifyNode(self, ninfo, nresult):
     """Perform some basic validation on data returned from a node.
@@ -1531,7 +1739,7 @@ class LUClusterVerify(LogicalUnit):
              ntime_diff)
 
   def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
-    """Check the node time.
+    """Check the node LVM results.
 
     @type ninfo: L{objects.Node}
     @param ninfo: the node to check
@@ -1567,8 +1775,31 @@ class LUClusterVerify(LogicalUnit):
         _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
                  " '%s' of VG '%s'", pvname, owner_vg)
 
+  def _VerifyNodeBridges(self, ninfo, nresult, bridges):
+    """Check the node bridges.
+
+    @type ninfo: L{objects.Node}
+    @param ninfo: the node to check
+    @param nresult: the remote results for the node
+    @param bridges: the expected list of bridges
+
+    """
+    if not bridges:
+      return
+
+    node = ninfo.name
+    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+
+    missing = nresult.get(constants.NV_BRIDGES, None)
+    test = not isinstance(missing, list)
+    _ErrorIf(test, self.ENODENET, node,
+             "did not return valid bridge information")
+    if not test:
+      _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
+               utils.CommaJoin(sorted(missing)))
+
   def _VerifyNodeNetwork(self, ninfo, nresult):
-    """Check the node time.
+    """Check the node network connectivity results.
 
     @type ninfo: L{objects.Node}
     @param ninfo: the node to check
@@ -1640,12 +1871,6 @@ class LUClusterVerify(LogicalUnit):
                "instance not running on its primary node %s",
                node_current)
 
-    for node, n_img in node_image.items():
-      if node != node_current:
-        test = instance in n_img.instances
-        _ErrorIf(test, self.EINSTANCEWRONGNODE, instance,
-                 "instance should not run on node %s", node)
-
     diskdata = [(nname, success, status, idx)
                 for (nname, disks) in diskstatus.items()
                 for idx, (success, status) in enumerate(disks)]
@@ -1685,18 +1910,6 @@ class LUClusterVerify(LogicalUnit):
         self._ErrorIf(test, self.ENODEORPHANLV, node,
                       "volume %s is unknown", volume)
 
-  def _VerifyOrphanInstances(self, instancelist, node_image):
-    """Verify the list of running instances.
-
-    This checks what instances are running but unknown to the cluster.
-
-    """
-    for node, n_img in node_image.items():
-      for o_inst in n_img.instances:
-        test = o_inst not in instancelist
-        self._ErrorIf(test, self.ENODEORPHANINSTANCE, node,
-                      "instance %s on node %s should not exist", o_inst, node)
-
   def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
     """Verify N+1 Memory Resilience.
 
@@ -2179,20 +2392,6 @@ class LUClusterVerify(LogicalUnit):
 
     return instdisk
 
-  def _VerifyHVP(self, hvp_data):
-    """Verifies locally the syntax of the hypervisor parameters.
-
-    """
-    for item, hv_name, hv_params in hvp_data:
-      msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
-             (item, hv_name))
-      try:
-        hv_class = hypervisor.GetHypervisor(hv_name)
-        utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
-        hv_class.CheckParameterSyntax(hv_params)
-      except errors.GenericError, err:
-        self._ErrorIf(True, self.ECLUSTERCFG, None, msg % str(err))
-
   def BuildHooksEnv(self):
     """Build hooks env.
 
@@ -2200,14 +2399,12 @@ class LUClusterVerify(LogicalUnit):
     the output be logged in the verify output and the verification to fail.
 
     """
-    cfg = self.cfg
-
     env = {
-      "CLUSTER_TAGS": " ".join(cfg.GetClusterInfo().GetTags())
+      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
       }
 
     env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
-               for node in cfg.GetAllNodesInfo().values())
+               for node in self.my_node_info.values())
 
     return env
 
@@ -2215,10 +2412,12 @@ class LUClusterVerify(LogicalUnit):
     """Build hooks nodes.
 
     """
-    return ([], self.cfg.GetNodeList())
+    assert self.my_node_names, ("Node list not gathered,"
+      " has CheckPrereq been executed?")
+    return ([], self.my_node_names)
 
   def Exec(self, feedback_fn):
-    """Verify integrity of cluster, performing various test on nodes.
+    """Verify integrity of the node group, performing various test on nodes.
 
     """
     # This method has too many local variables. pylint: disable-msg=R0914
@@ -2226,26 +2425,14 @@ class LUClusterVerify(LogicalUnit):
     _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
     verbose = self.op.verbose
     self._feedback_fn = feedback_fn
-    feedback_fn("* Verifying global settings")
-    for msg in self.cfg.VerifyConfig():
-      _ErrorIf(True, self.ECLUSTERCFG, None, msg)
-
-    # Check the cluster certificates
-    for cert_filename in constants.ALL_CERT_FILES:
-      (errcode, msg) = _VerifyCertificate(cert_filename)
-      _ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
 
     vg_name = self.cfg.GetVGName()
     drbd_helper = self.cfg.GetDRBDHelper()
-    hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
     cluster = self.cfg.GetClusterInfo()
-    nodelist = utils.NiceSort(self.cfg.GetNodeList())
-    nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
-    nodeinfo_byname = dict(zip(nodelist, nodeinfo))
-    instancelist = utils.NiceSort(self.cfg.GetInstanceList())
-    instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
-                        for iname in instancelist)
     groupinfo = self.cfg.GetAllNodeGroupsInfo()
+    hypervisors = cluster.enabled_hypervisors
+    node_data_list = [self.my_node_info[name] for name in self.my_node_names]
+
     i_non_redundant = [] # Non redundant instances
     i_non_a_balanced = [] # Non auto-balanced instances
     n_offline = 0 # Count of offline nodes
@@ -2261,37 +2448,32 @@ class LUClusterVerify(LogicalUnit):
     master_node = self.master_node = self.cfg.GetMasterNode()
     master_ip = self.cfg.GetMasterIP()
 
-    # Compute the set of hypervisor parameters
-    hvp_data = []
-    for hv_name in hypervisors:
-      hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
-    for os_name, os_hvp in cluster.os_hvp.items():
-      for hv_name, hv_params in os_hvp.items():
-        if not hv_params:
-          continue
-        full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
-        hvp_data.append(("os %s" % os_name, hv_name, full_params))
-    # TODO: collapse identical parameter values in a single one
-    for instance in instanceinfo.values():
-      if not instance.hvparams:
-        continue
-      hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
-                       cluster.FillHV(instance)))
-    # and verify them locally
-    self._VerifyHVP(hvp_data)
+    feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
+
+    # We will make nodes contact all nodes in their group, and one node from
+    # every other group.
+    # TODO: should it be a *random* node, different every time?
+    online_nodes = [node.name for node in node_data_list if not node.offline]
+    other_group_nodes = {}
+
+    for name in sorted(self.all_node_info):
+      node = self.all_node_info[name]
+      if (node.group not in other_group_nodes
+          and node.group != self.group_uuid
+          and not node.offline):
+        other_group_nodes[node.group] = node.name
 
-    feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
     node_verify_param = {
       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_NODELIST: online_nodes + other_group_nodes.values(),
       constants.NV_HYPERVISOR: hypervisors,
-      constants.NV_HVPARAMS: hvp_data,
-      constants.NV_NODENETTEST: [(node.name, node.primary_ip,
-                                  node.secondary_ip) for node in nodeinfo
+      constants.NV_HVPARAMS:
+        _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
+      constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
+                                 for node in node_data_list
                                  if not node.offline],
       constants.NV_INSTANCELIST: hypervisors,
       constants.NV_VERSION: None,
@@ -2312,15 +2494,30 @@ class LUClusterVerify(LogicalUnit):
     if drbd_helper:
       node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
 
+    # bridge checks
+    # FIXME: this needs to be changed per node-group, not cluster-wide
+    bridges = set()
+    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
+    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
+      bridges.add(default_nicpp[constants.NIC_LINK])
+    for instance in self.my_inst_info.values():
+      for nic in instance.nics:
+        full_nic = cluster.SimpleFillNIC(nic.nicparams)
+        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
+          bridges.add(full_nic[constants.NIC_LINK])
+
+    if bridges:
+      node_verify_param[constants.NV_BRIDGES] = list(bridges)
+
     # Build our expected cluster state
     node_image = dict((node.name, self.NodeImage(offline=node.offline,
                                                  name=node.name,
                                                  vm_capable=node.vm_capable))
-                      for node in nodeinfo)
+                      for node in node_data_list)
 
     # Gather OOB paths
     oob_paths = []
-    for node in nodeinfo:
+    for node in self.all_node_info.values():
       path = _SupportsOob(self.cfg, node)
       if path and path not in oob_paths:
         oob_paths.append(path)
@@ -2328,14 +2525,13 @@ class LUClusterVerify(LogicalUnit):
     if oob_paths:
       node_verify_param[constants.NV_OOB_PATHS] = oob_paths
 
-    for instance in instancelist:
-      inst_config = instanceinfo[instance]
+    for instance in self.my_inst_names:
+      inst_config = self.my_inst_info[instance]
 
       for nname in inst_config.all_nodes:
         if nname not in node_image:
-          # ghost node
           gnode = self.NodeImage(name=nname)
-          gnode.ghost = True
+          gnode.ghost = (nname not in self.all_node_info)
           node_image[nname] = gnode
 
       inst_config.MapLVsByNode(node_vol_should)
@@ -2358,23 +2554,59 @@ class LUClusterVerify(LogicalUnit):
     # time before and after executing the request, we can at least have a time
     # window.
     nvinfo_starttime = time.time()
-    all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
+    all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
+                                           node_verify_param,
                                            self.cfg.GetClusterName())
+    if self.extra_lv_nodes and vg_name is not None:
+      extra_lv_nvinfo = \
+          self.rpc.call_node_verify(self.extra_lv_nodes,
+                                    {constants.NV_LVLIST: vg_name},
+                                    self.cfg.GetClusterName())
+    else:
+      extra_lv_nvinfo = {}
     nvinfo_endtime = time.time()
 
     all_drbd_map = self.cfg.ComputeDRBDMap()
 
-    feedback_fn("* Gathering disk information (%s nodes)" % len(nodelist))
-    instdisk = self._CollectDiskInfo(nodelist, node_image, instanceinfo)
+    feedback_fn("* Gathering disk information (%s nodes)" %
+                len(self.my_node_names))
+    instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
+                                     self.my_inst_info)
 
     feedback_fn("* Verifying configuration file consistency")
-    self._VerifyFiles(_ErrorIf, nodeinfo, master_node, all_nvinfo, filemap)
+
+    # If not all nodes are being checked, we need to make sure the master node
+    # and a non-checked vm_capable node are in the list.
+    absent_nodes = set(self.all_node_info).difference(self.my_node_info)
+    if absent_nodes:
+      vf_nvinfo = all_nvinfo.copy()
+      vf_node_info = list(self.my_node_info.values())
+      additional_nodes = []
+      if master_node not in self.my_node_info:
+        additional_nodes.append(master_node)
+        vf_node_info.append(self.all_node_info[master_node])
+      # Add the first vm_capable node we find which is not included
+      for node in absent_nodes:
+        nodeinfo = self.all_node_info[node]
+        if nodeinfo.vm_capable and not nodeinfo.offline:
+          additional_nodes.append(node)
+          vf_node_info.append(self.all_node_info[node])
+          break
+      key = constants.NV_FILELIST
+      vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
+                                                 {key: node_verify_param[key]},
+                                                 self.cfg.GetClusterName()))
+    else:
+      vf_nvinfo = all_nvinfo
+      vf_node_info = self.my_node_info.values()
+
+    self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
 
     feedback_fn("* Verifying node status")
 
     refos_img = None
 
-    for node_i in nodeinfo:
+    for node_i in node_data_list:
       node = node_i.name
       nimg = node_image[node]
 
@@ -2411,23 +2643,41 @@ class LUClusterVerify(LogicalUnit):
 
       if nimg.vm_capable:
         self._VerifyNodeLVM(node_i, nresult, vg_name)
-        self._VerifyNodeDrbd(node_i, nresult, instanceinfo, drbd_helper,
+        self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
                              all_drbd_map)
 
         self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
         self._UpdateNodeInstances(node_i, nresult, nimg)
         self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
         self._UpdateNodeOS(node_i, nresult, nimg)
+
         if not nimg.os_fail:
           if refos_img is None:
             refos_img = nimg
           self._VerifyNodeOS(node_i, nimg, refos_img)
+        self._VerifyNodeBridges(node_i, nresult, bridges)
+
+        # Check whether all running instancies are primary for the node. (This
+        # can no longer be done from _VerifyInstance below, since some of the
+        # wrong instances could be from other node groups.)
+        non_primary_inst = set(nimg.instances).difference(nimg.pinst)
+
+        for inst in non_primary_inst:
+          test = inst in self.all_inst_info
+          _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
+                   "instance should not run on node %s", node_i.name)
+          _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
+                   "node is running unknown instance %s", inst)
+
+    for node, result in extra_lv_nvinfo.items():
+      self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
+                              node_image[node], vg_name)
 
     feedback_fn("* Verifying instance status")
-    for instance in instancelist:
+    for instance in self.my_inst_names:
       if verbose:
         feedback_fn("* Verifying instance %s" % instance)
-      inst_config = instanceinfo[instance]
+      inst_config = self.my_inst_info[instance]
       self._VerifyInstance(instance, inst_config, node_image,
                            instdisk[instance])
       inst_nodes_offline = []
@@ -2462,7 +2712,7 @@ class LUClusterVerify(LogicalUnit):
         instance_groups = {}
 
         for node in instance_nodes:
-          instance_groups.setdefault(nodeinfo_byname[node].group,
+          instance_groups.setdefault(self.all_node_info[node].group,
                                      []).append(node)
 
         pretty_list = [
@@ -2501,14 +2751,22 @@ class LUClusterVerify(LogicalUnit):
 
     feedback_fn("* Verifying orphan volumes")
     reserved = utils.FieldSet(*cluster.reserved_lvs)
-    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
 
-    feedback_fn("* Verifying orphan instances")
-    self._VerifyOrphanInstances(instancelist, node_image)
+    # We will get spurious "unknown volume" warnings if any node of this group
+    # is secondary for an instance whose primary is in another group. To avoid
+    # them, we find these instances and add their volumes to node_vol_should.
+    for inst in self.all_inst_info.values():
+      for secondary in inst.secondary_nodes:
+        if (secondary in self.my_node_info
+            and inst.name not in self.my_inst_info):
+          inst.MapLVsByNode(node_vol_should)
+          break
+
+    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
 
     if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
       feedback_fn("* Verifying N+1 Memory redundancy")
-      self._VerifyNPlusOneMemory(node_image, instanceinfo)
+      self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
 
     feedback_fn("* Other Notes")
     if i_non_redundant:
@@ -2974,8 +3232,8 @@ class LUClusterSetParams(LogicalUnit):
           # if we're moving instances to routed, check that they have an ip
           target_mode = params_filled[constants.NIC_MODE]
           if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
-            nic_errors.append("Instance %s, nic/%d: routed nick with no ip" %
-                              (instance.name, nic_idx))
+            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
+                              " address" % (instance.name, nic_idx))
       if nic_errors:
         raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
                                    "\n".join(nic_errors))
@@ -5420,7 +5678,8 @@ class LUInstanceStartup(LogicalUnit):
     instance = self.instance
     force = self.op.force
 
-    self.cfg.MarkInstanceUp(instance.name)
+    if not self.op.no_remember:
+      self.cfg.MarkInstanceUp(instance.name)
 
     if self.primary_offline:
       assert self.op.ignore_offline_nodes
@@ -5585,7 +5844,8 @@ class LUInstanceShutdown(LogicalUnit):
     node_current = instance.primary_node
     timeout = self.op.timeout
 
-    self.cfg.MarkInstanceDown(instance.name)
+    if not self.op.no_remember:
+      self.cfg.MarkInstanceDown(instance.name)
 
     if self.primary_offline:
       assert self.op.ignore_offline_nodes
@@ -5698,8 +5958,25 @@ class LUInstanceRecreateDisks(LogicalUnit):
   HTYPE = constants.HTYPE_INSTANCE
   REQ_BGL = False
 
+  def CheckArguments(self):
+    # normalise the disk list
+    self.op.disks = sorted(frozenset(self.op.disks))
+
   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)
+    else:
+      self.needed_locks[locking.LEVEL_NODE] = []
+
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_NODE:
+      # if we replace the nodes, we only need to lock the old primary,
+      # otherwise we need to lock all nodes for disk re-creation
+      primary_only = bool(self.op.nodes)
+      self._LockInstancesNodes(primary_only=primary_only)
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -5725,12 +6002,31 @@ class LUInstanceRecreateDisks(LogicalUnit):
     instance = self.cfg.GetInstanceInfo(self.op.instance_name)
     assert instance is not None, \
       "Cannot retrieve locked instance %s" % self.op.instance_name
-    _CheckNodeOnline(self, instance.primary_node)
+    if self.op.nodes:
+      if len(self.op.nodes) != len(instance.all_nodes):
+        raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
+                                   " %d replacement nodes were specified" %
+                                   (instance.name, len(instance.all_nodes),
+                                    len(self.op.nodes)),
+                                   errors.ECODE_INVAL)
+      assert instance.disk_template != constants.DT_DRBD8 or \
+          len(self.op.nodes) == 2
+      assert instance.disk_template != constants.DT_PLAIN or \
+          len(self.op.nodes) == 1
+      primary_node = self.op.nodes[0]
+    else:
+      primary_node = instance.primary_node
+    _CheckNodeOnline(self, primary_node)
 
     if instance.disk_template == constants.DT_DISKLESS:
       raise errors.OpPrereqError("Instance '%s' has no disks" %
                                  self.op.instance_name, errors.ECODE_INVAL)
-    _CheckInstanceDown(self, instance, "cannot recreate disks")
+    # if we replace nodes *and* the old primary is offline, we don't
+    # check
+    assert instance.primary_node in self.needed_locks[locking.LEVEL_NODE]
+    old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
+    if not (self.op.nodes and old_pnode.offline):
+      _CheckInstanceDown(self, instance, "cannot recreate disks")
 
     if not self.op.disks:
       self.op.disks = range(len(instance.disks))
@@ -5739,18 +6035,39 @@ class LUInstanceRecreateDisks(LogicalUnit):
         if idx >= len(instance.disks):
           raise errors.OpPrereqError("Invalid disk index '%s'" % idx,
                                      errors.ECODE_INVAL)
-
+    if self.op.disks != range(len(instance.disks)) and self.op.nodes:
+      raise errors.OpPrereqError("Can't recreate disks partially and"
+                                 " change the nodes at the same time",
+                                 errors.ECODE_INVAL)
     self.instance = instance
 
   def Exec(self, feedback_fn):
     """Recreate the disks.
 
     """
+    # change primary node, if needed
+    if self.op.nodes:
+      self.instance.primary_node = self.op.nodes[0]
+      self.LogWarning("Changing the instance's nodes, you will have to"
+                      " remove any disks left on the older nodes manually")
+
     to_skip = []
-    for idx, _ in enumerate(self.instance.disks):
+    for idx, disk in enumerate(self.instance.disks):
       if idx not in self.op.disks: # disk idx has not been passed in
         to_skip.append(idx)
         continue
+      # update secondaries for disks, if needed
+      if self.op.nodes:
+        if disk.dev_type == constants.LD_DRBD8:
+          # need to update the nodes
+          assert len(self.op.nodes) == 2
+          logical_id = list(disk.logical_id)
+          logical_id[0] = self.op.nodes[0]
+          logical_id[1] = self.op.nodes[1]
+          disk.logical_id = tuple(logical_id)
+
+    if self.op.nodes:
+      self.cfg.Update(self.instance, feedback_fn)
 
     _CreateDisks(self, self.instance, to_skip=to_skip)
 
@@ -5805,8 +6122,9 @@ class LUInstanceRename(LogicalUnit):
     new_name = self.op.new_name
     if self.op.name_check:
       hostname = netutils.GetHostname(name=new_name)
-      self.LogInfo("Resolved given name '%s' to '%s'", new_name,
-                   hostname.name)
+      if hostname != new_name:
+        self.LogInfo("Resolved given name '%s' to '%s'", new_name,
+                     hostname.name)
       if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
         raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"
                                     " same as given hostname '%s'") %
@@ -6469,6 +6787,10 @@ class TLMigrateInstance(Tasklet):
       # self.target_node is already populated, either directly or by the
       # iallocator run
       target_node = self.target_node
+      if self.target_node == instance.primary_node:
+        raise errors.OpPrereqError("Cannot migrate instance %s"
+                                   " to its primary (%s)" %
+                                   (instance.name, instance.primary_node))
 
       if len(self.lu.tasklets) == 1:
         # It is safe to release locks only when we're the only tasklet
@@ -9983,6 +10305,7 @@ class LUInstanceSetParams(LogicalUnit):
       self.be_inst = i_bedict # the new dict (without defaults)
     else:
       self.be_new = self.be_inst = {}
+    be_old = cluster.FillBE(instance)
 
     # osparams processing
     if self.op.osparams:
@@ -9994,7 +10317,8 @@ class LUInstanceSetParams(LogicalUnit):
 
     self.warn = []
 
-    if constants.BE_MEMORY in self.op.beparams and not self.op.force:
+    if (constants.BE_MEMORY in self.op.beparams and not self.op.force and
+        be_new[constants.BE_MEMORY] > be_old[constants.BE_MEMORY]):
       mem_check_list = [pnode]
       if be_new[constants.BE_AUTO_BALANCE]:
         # either we changed auto_balance to yes or it was from before
@@ -10035,16 +10359,17 @@ class LUInstanceSetParams(LogicalUnit):
         for node, nres in nodeinfo.items():
           if node not in instance.secondary_nodes:
             continue
-          msg = nres.fail_msg
-          if msg:
-            self.warn.append("Can't get info from secondary node %s: %s" %
-                             (node, msg))
-          elif not isinstance(nres.payload.get('memory_free', None), int):
-            self.warn.append("Secondary node %s didn't return free"
-                             " memory information" % node)
+          nres.Raise("Can't get info from secondary node %s" % node,
+                     prereq=True, ecode=errors.ECODE_STATE)
+          if not isinstance(nres.payload.get('memory_free', None), int):
+            raise errors.OpPrereqError("Secondary node %s didn't return free"
+                                       " memory information" % node,
+                                       errors.ECODE_STATE)
           elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
-            self.warn.append("Not enough memory to failover instance to"
-                             " secondary node %s" % node)
+            raise errors.OpPrereqError("This change will prevent the instance"
+                                       " from failover to its secondary node"
+                                       " %s, due to not enough memory" % node,
+                                       errors.ECODE_STATE)
 
     # NIC processing
     self.nic_pnew = {}
@@ -10196,7 +10521,8 @@ class LUInstanceSetParams(LogicalUnit):
     self.cfg.Update(instance, feedback_fn)
 
     # disks are created, waiting for sync
-    disk_abort = not _WaitForSync(self, instance)
+    disk_abort = not _WaitForSync(self, instance,
+                                  oneshot=not self.op.wait_for_sync)
     if disk_abort:
       raise errors.OpExecError("There are some degraded disks for"
                                " this instance, please cleanup manually")
@@ -10886,20 +11212,40 @@ class LUGroupAssignNodes(NoHooksLU):
 
     # We want to lock all the affected nodes and groups. We have readily
     # available the list of nodes, and the *destination* group. To gather the
-    # list of "source" groups, we need to fetch node information.
-    self.node_data = self.cfg.GetAllNodesInfo()
-    affected_groups = set(self.node_data[node].group for node in self.op.nodes)
-    affected_groups.add(self.group_uuid)
-
+    # list of "source" groups, we need to fetch node information later on.
     self.needed_locks = {
-      locking.LEVEL_NODEGROUP: list(affected_groups),
+      locking.LEVEL_NODEGROUP: set([self.group_uuid]),
       locking.LEVEL_NODE: self.op.nodes,
       }
 
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_NODEGROUP:
+      assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1
+
+      # Try to get all affected nodes' groups without having the group or node
+      # lock yet. Needs verification later in the code flow.
+      groups = self.cfg.GetNodeGroupsFromNodes(self.op.nodes)
+
+      self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
+
   def CheckPrereq(self):
     """Check prerequisites.
 
     """
+    assert self.needed_locks[locking.LEVEL_NODEGROUP]
+    assert (frozenset(self.glm.list_owned(locking.LEVEL_NODE)) ==
+            frozenset(self.op.nodes))
+
+    expected_locks = (set([self.group_uuid]) |
+                      self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
+    actual_locks = self.glm.list_owned(locking.LEVEL_NODEGROUP)
+    if actual_locks != expected_locks:
+      raise errors.OpExecError("Nodes changed groups since locks were acquired,"
+                               " current groups are '%s', used to be '%s'" %
+                               (utils.CommaJoin(expected_locks),
+                                utils.CommaJoin(actual_locks)))
+
+    self.node_data = self.cfg.GetAllNodesInfo()
     self.group = self.cfg.GetNodeGroup(self.group_uuid)
     instance_data = self.cfg.GetAllInstancesInfo()
 
@@ -10935,6 +11281,9 @@ class LUGroupAssignNodes(NoHooksLU):
     for node in self.op.nodes:
       self.node_data[node].group = self.group_uuid
 
+    # FIXME: Depends on side-effects of modifying the result of
+    # C{cfg.GetAllNodesInfo}
+
     self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
 
   @staticmethod
@@ -11651,16 +12000,6 @@ class IAllocator(object):
   """
   # pylint: disable-msg=R0902
   # lots of instance attributes
-  _ALLO_KEYS = [
-    "name", "mem_size", "disks", "disk_template",
-    "os", "tags", "nics", "vcpus", "hypervisor",
-    ]
-  _RELO_KEYS = [
-    "name", "relocate_from",
-    ]
-  _EVAC_KEYS = [
-    "evac_nodes",
-    ]
 
   def __init__(self, cfg, rpc, mode, **kwargs):
     self.cfg = cfg
@@ -11675,22 +12014,20 @@ class IAllocator(object):
     self.relocate_from = None
     self.name = None
     self.evac_nodes = None
+    self.instances = None
+    self.reloc_mode = None
+    self.target_groups = None
     # computed fields
     self.required_nodes = None
     # init result fields
     self.success = self.info = self.result = None
-    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
-      keyset = self._ALLO_KEYS
-      fn = self._AddNewInstance
-    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
-      keyset = self._RELO_KEYS
-      fn = self._AddRelocateInstance
-    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
-      keyset = self._EVAC_KEYS
-      fn = self._AddEvacuateNodes
-    else:
+
+    try:
+      (fn, keyset, self._result_check) = self._MODE_DATA[self.mode]
+    except KeyError:
       raise errors.ProgrammerError("Unknown mode '%s' passed to the"
                                    " IAllocator" % self.mode)
+
     for key in kwargs:
       if key not in keyset:
         raise errors.ProgrammerError("Invalid input parameter '%s' to"
@@ -11701,7 +12038,7 @@ class IAllocator(object):
       if key not in kwargs:
         raise errors.ProgrammerError("Missing input parameter '%s' to"
                                      " IAllocator" % key)
-    self._BuildInputData(fn)
+    self._BuildInputData(compat.partial(fn, self))
 
   def _ComputeClusterData(self):
     """Compute the generic allocator input data.
@@ -11730,7 +12067,8 @@ class IAllocator(object):
       hypervisor_name = self.hypervisor
     elif self.mode == constants.IALLOCATOR_MODE_RELOC:
       hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
-    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
+    elif self.mode in (constants.IALLOCATOR_MODE_MEVAC,
+                       constants.IALLOCATOR_MODE_MRELOC):
       hypervisor_name = cluster_info.enabled_hypervisors[0]
 
     node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
@@ -11756,12 +12094,12 @@ class IAllocator(object):
     """Compute node groups data.
 
     """
-    ng = {}
-    for guuid, gdata in cfg.GetAllNodeGroupsInfo().items():
-      ng[guuid] = {
-        "name": gdata.name,
-        "alloc_policy": gdata.alloc_policy,
-        }
+    ng = dict((guuid, {
+      "name": gdata.name,
+      "alloc_policy": gdata.alloc_policy,
+      })
+      for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())
+
     return ng
 
   @staticmethod
@@ -11772,22 +12110,19 @@ class IAllocator(object):
     @returns: a dict of name: (node dict, node config)
 
     """
-    node_results = {}
-    for ninfo in node_cfg.values():
-      # fill in static (config-based) values
-      pnr = {
-        "tags": list(ninfo.GetTags()),
-        "primary_ip": ninfo.primary_ip,
-        "secondary_ip": ninfo.secondary_ip,
-        "offline": ninfo.offline,
-        "drained": ninfo.drained,
-        "master_candidate": ninfo.master_candidate,
-        "group": ninfo.group,
-        "master_capable": ninfo.master_capable,
-        "vm_capable": ninfo.vm_capable,
-        }
-
-      node_results[ninfo.name] = pnr
+    # fill in static (config-based) values
+    node_results = dict((ninfo.name, {
+      "tags": list(ninfo.GetTags()),
+      "primary_ip": ninfo.primary_ip,
+      "secondary_ip": ninfo.secondary_ip,
+      "offline": ninfo.offline,
+      "drained": ninfo.drained,
+      "master_candidate": ninfo.master_candidate,
+      "group": ninfo.group,
+      "master_capable": ninfo.master_capable,
+      "vm_capable": ninfo.vm_capable,
+      })
+      for ninfo in node_cfg.values())
 
     return node_results
 
@@ -11861,11 +12196,12 @@ class IAllocator(object):
       nic_data = []
       for nic in iinfo.nics:
         filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
-        nic_dict = {"mac": nic.mac,
-                    "ip": nic.ip,
-                    "mode": filled_params[constants.NIC_MODE],
-                    "link": filled_params[constants.NIC_LINK],
-                   }
+        nic_dict = {
+          "mac": nic.mac,
+          "ip": nic.ip,
+          "mode": filled_params[constants.NIC_MODE],
+          "link": filled_params[constants.NIC_LINK],
+          }
         if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
           nic_dict["bridge"] = filled_params[constants.NIC_LINK]
         nic_data.append(nic_dict)
@@ -11905,6 +12241,7 @@ class IAllocator(object):
       self.required_nodes = 2
     else:
       self.required_nodes = 1
+
     request = {
       "name": self.name,
       "disk_template": self.disk_template,
@@ -11917,6 +12254,7 @@ class IAllocator(object):
       "nics": self.nics,
       "required_nodes": self.required_nodes,
       }
+
     return request
 
   def _AddRelocateInstance(self):
@@ -11964,6 +12302,16 @@ class IAllocator(object):
       }
     return request
 
+  def _AddMultiRelocate(self):
+    """Get data for multi-relocate requests.
+
+    """
+    return {
+      "instances": self.instances,
+      "reloc_mode": self.reloc_mode,
+      "target_groups": self.target_groups,
+      }
+
   def _BuildInputData(self, fn):
     """Build input data structures.
 
@@ -11976,6 +12324,28 @@ class IAllocator(object):
 
     self.in_text = serializer.Dump(self.in_data)
 
+  _MODE_DATA = {
+    constants.IALLOCATOR_MODE_ALLOC:
+      (_AddNewInstance,
+       ["name", "mem_size", "disks", "disk_template", "os", "tags", "nics",
+        "vcpus", "hypervisor"], ht.TList),
+    constants.IALLOCATOR_MODE_RELOC:
+      (_AddRelocateInstance, ["name", "relocate_from"], ht.TList),
+    constants.IALLOCATOR_MODE_MEVAC:
+      (_AddEvacuateNodes, ["evac_nodes"],
+       ht.TListOf(ht.TAnd(ht.TIsLength(2),
+                          ht.TListOf(ht.TString)))),
+    constants.IALLOCATOR_MODE_MRELOC:
+      (_AddMultiRelocate, ["instances", "reloc_mode", "target_groups"],
+       ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
+         # pylint: disable-msg=E1101
+         # Class '...' has no 'OP_ID' member
+         "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
+                              opcodes.OpInstanceMigrate.OP_ID,
+                              opcodes.OpInstanceReplaceDisks.OP_ID])
+         })))),
+    }
+
   def Run(self, name, validate=True, call_fn=None):
     """Run an instance allocator and return the results.
 
@@ -12016,28 +12386,45 @@ class IAllocator(object):
                                  " missing key '%s'" % key)
       setattr(self, key, rdict[key])
 
-    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
+    if not self._result_check(self.result):
+      raise errors.OpExecError("Iallocator returned invalid result,"
+                               " expected %s, got %s" %
+                               (self._result_check, self.result),
+                               errors.ECODE_INVAL)
 
+    if self.mode in (constants.IALLOCATOR_MODE_RELOC,
+                     constants.IALLOCATOR_MODE_MEVAC):
       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)))
+      if self.mode == constants.IALLOCATOR_MODE_RELOC:
+        assert self.relocate_from is not None
+        assert self.required_nodes == 1
+
+        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)))
+      elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
+        request_groups = fn(self.evac_nodes)
+        for (instance_name, secnode) in self.result:
+          result_groups = fn([secnode])
+          if result_groups != request_groups:
+            raise errors.OpExecError("Iallocator returned new secondary node"
+                                     " '%s' (group '%s') for instance '%s'"
+                                     " which is not in original group '%s'" %
+                                     (secnode, utils.CommaJoin(result_groups),
+                                      instance_name,
+                                      utils.CommaJoin(request_groups)))
+      else:
+        raise errors.ProgrammerError("Unhandled mode '%s'" % self.mode)
 
     self.out_data = rdict
 
@@ -12121,6 +12508,12 @@ class LUTestAllocator(NoHooksLU):
       if not hasattr(self.op, "evac_nodes"):
         raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
                                    " opcode input", errors.ECODE_INVAL)
+    elif self.op.mode == constants.IALLOCATOR_MODE_MRELOC:
+      if self.op.instances:
+        self.op.instances = _GetWantedInstances(self, self.op.instances)
+      else:
+        raise errors.OpPrereqError("Missing instances to relocate",
+                                   errors.ECODE_INVAL)
     else:
       raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
                                  self.op.mode, errors.ECODE_INVAL)
@@ -12160,6 +12553,12 @@ class LUTestAllocator(NoHooksLU):
       ial = IAllocator(self.cfg, self.rpc,
                        mode=self.op.mode,
                        evac_nodes=self.op.evac_nodes)
+    elif self.op.mode == constants.IALLOCATOR_MODE_MRELOC:
+      ial = IAllocator(self.cfg, self.rpc,
+                       mode=self.op.mode,
+                       instances=self.op.instances,
+                       reloc_mode=self.op.reloc_mode,
+                       target_groups=self.op.target_groups)
     else:
       raise errors.ProgrammerError("Uncatched mode %s in"
                                    " LUTestAllocator.Exec", self.op.mode)