LUAddNode: copy the vnc password file also for KVM
[ganeti-local] / lib / cmdlib.py
index c639dd3..d7f883b 100644 (file)
@@ -392,8 +392,8 @@ def _GetWantedInstances(lu, instances):
       wanted.append(instance)
 
   else:
-    wanted = lu.cfg.GetInstanceList()
-  return utils.NiceSort(wanted)
+    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
+  return wanted
 
 
 def _CheckOutputFields(static, dynamic, selected):
@@ -455,8 +455,8 @@ def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
   @param secondary_nodes: list of secondary nodes as strings
   @type os_type: string
   @param os_type: the name of the instance's OS
-  @type status: string
-  @param status: the desired status of the instances
+  @type status: boolean
+  @param status: the should_run status of the instance
   @type memory: string
   @param memory: the memory size of the instance
   @type vcpus: string
@@ -468,13 +468,17 @@ def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
   @return: the hook environment for this instance
 
   """
+  if status:
+    str_status = "up"
+  else:
+    str_status = "down"
   env = {
     "OP_TARGET": name,
     "INSTANCE_NAME": name,
     "INSTANCE_PRIMARY": primary_node,
     "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
     "INSTANCE_OS_TYPE": os_type,
-    "INSTANCE_STATUS": status,
+    "INSTANCE_STATUS": str_status,
     "INSTANCE_MEMORY": memory,
     "INSTANCE_VCPUS": vcpus,
   }
@@ -516,7 +520,7 @@ def _BuildInstanceHookEnvByObject(lu, instance, override=None):
     'primary_node': instance.primary_node,
     'secondary_nodes': instance.secondary_nodes,
     'os_type': instance.os,
-    'status': instance.os,
+    'status': instance.admin_up,
     'memory': bep[constants.BE_MEMORY],
     'vcpus': bep[constants.BE_VCPUS],
     'nics': [(nic.ip, nic.bridge, nic.mac) for nic in instance.nics],
@@ -613,7 +617,8 @@ class LUVerifyCluster(LogicalUnit):
     self.share_locks = dict(((i, 1) for i in locking.LEVELS))
 
   def _VerifyNode(self, nodeinfo, file_list, local_cksum,
-                  node_result, feedback_fn, master_files):
+                  node_result, feedback_fn, master_files,
+                  drbd_map):
     """Run multiple tests against a node.
 
     Test list:
@@ -630,6 +635,9 @@ class LUVerifyCluster(LogicalUnit):
     @param node_result: the results from the node
     @param feedback_fn: function used to accumulate results
     @param master_files: list of files that only masters should have
+    @param drbd_map: the useddrbd minors for this node, in
+        form of minor: (instance, must_exist) which correspond to instances
+        and their running status
 
     """
     node = nodeinfo.name
@@ -642,18 +650,28 @@ class LUVerifyCluster(LogicalUnit):
     # compares ganeti version
     local_version = constants.PROTOCOL_VERSION
     remote_version = node_result.get('version', None)
-    if not remote_version:
+    if not (remote_version and isinstance(remote_version, (list, tuple)) and
+            len(remote_version) == 2):
       feedback_fn("  - ERROR: connection to %s failed" % (node))
       return True
 
-    if local_version != remote_version:
-      feedback_fn("  - ERROR: sw version mismatch: master %s, node(%s) %s" %
-                      (local_version, node, remote_version))
+    if local_version != remote_version[0]:
+      feedback_fn("  - ERROR: incompatible protocol versions: master %s,"
+                  " node %s %s" % (local_version, node, remote_version[0]))
       return True
 
-    # checks vg existance and size > 20G
+    # node seems compatible, we can actually try to look into its results
 
     bad = False
+
+    # full package version
+    if constants.RELEASE_VERSION != remote_version[1]:
+      feedback_fn("  - WARNING: software version mismatch: master %s,"
+                  " node %s %s" %
+                  (constants.RELEASE_VERSION, node, remote_version[1]))
+
+    # checks vg existence and size > 20G
+
     vglist = node_result.get(constants.NV_VGLIST, None)
     if not vglist:
       feedback_fn("  - ERROR: unable to check volume groups on node %s." %
@@ -724,6 +742,19 @@ class LUVerifyCluster(LogicalUnit):
         if hv_result is not None:
           feedback_fn("  - ERROR: hypervisor %s verify failure: '%s'" %
                       (hv_name, hv_result))
+
+    # check used drbd list
+    used_minors = node_result.get(constants.NV_DRBDLIST, [])
+    for minor, (iname, must_exist) in drbd_map.items():
+      if minor not in used_minors and must_exist:
+        feedback_fn("  - ERROR: drbd minor %d of instance %s is not active" %
+                    (minor, iname))
+        bad = True
+    for minor in used_minors:
+      if minor not in drbd_map:
+        feedback_fn("  - ERROR: unallocated drbd minor %d is in use" % minor)
+        bad = True
+
     return bad
 
   def _VerifyInstance(self, instance, instanceconfig, node_vol_is,
@@ -751,7 +782,7 @@ class LUVerifyCluster(LogicalUnit):
                           (volume, node))
           bad = True
 
-    if not instanceconfig.status == 'down':
+    if instanceconfig.admin_up:
       if ((node_current not in node_instance or
           not instance in node_instance[node_current]) and
           node_current not in n_offline):
@@ -867,6 +898,8 @@ class LUVerifyCluster(LogicalUnit):
     nodelist = utils.NiceSort(self.cfg.GetNodeList())
     nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
     instancelist = utils.NiceSort(self.cfg.GetInstanceList())
+    instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
+                        for iname in instancelist)
     i_non_redundant = [] # Non redundant instances
     i_non_a_balanced = [] # Non auto-balanced instances
     n_offline = [] # List of offline nodes
@@ -900,12 +933,15 @@ class LUVerifyCluster(LogicalUnit):
       constants.NV_VGLIST: None,
       constants.NV_VERSION: None,
       constants.NV_HVINFO: self.cfg.GetHypervisorType(),
+      constants.NV_DRBDLIST: None,
       }
     all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
                                            self.cfg.GetClusterName())
 
     cluster = self.cfg.GetClusterInfo()
     master_node = self.cfg.GetMasterNode()
+    all_drbd_map = self.cfg.ComputeDRBDMap()
+
     for node_i in nodeinfo:
       node = node_i.name
       nresult = all_nvinfo[node].data
@@ -928,8 +964,13 @@ class LUVerifyCluster(LogicalUnit):
         bad = True
         continue
 
+      node_drbd = {}
+      for minor, instance in all_drbd_map[node].items():
+        instance = instanceinfo[instance]
+        node_drbd[minor] = (instance.name, instance.admin_up)
       result = self._VerifyNode(node_i, file_names, local_checksums,
-                                nresult, feedback_fn, master_files)
+                                nresult, feedback_fn, master_files,
+                                node_drbd)
       bad = bad or result
 
       lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
@@ -985,7 +1026,7 @@ class LUVerifyCluster(LogicalUnit):
 
     for instance in instancelist:
       feedback_fn("* Verifying instance %s" % instance)
-      inst_config = self.cfg.GetInstanceInfo(instance)
+      inst_config = instanceinfo[instance]
       result =  self._VerifyInstance(instance, inst_config, node_volume,
                                      node_instance, feedback_fn, n_offline)
       bad = bad or result
@@ -1154,7 +1195,7 @@ class LUVerifyDisks(NoHooksLU):
     nv_dict = {}
     for inst in instances:
       inst_lvs = {}
-      if (inst.status != "up" or
+      if (not inst.admin_up or
           inst.disk_template not in constants.DTS_NET_MIRROR):
         continue
       inst.MapLVsByNode(inst_lvs)
@@ -1495,8 +1536,7 @@ def _WaitForSync(lu, instance, oneshot=False, unlock=False):
       continue
     rstats = rstats.data
     retries = 0
-    for i in range(len(rstats)):
-      mstat = rstats[i]
+    for i, mstat in enumerate(rstats):
       if mstat is None:
         lu.LogWarning("Can't compute data for node %s/%s",
                            node, instance.disks[i].iv_name)
@@ -2138,8 +2178,10 @@ class LUAddNode(LogicalUnit):
           logging.error("Copy of file %s to node %s failed", fname, to_node)
 
     to_copy = []
-    if constants.HT_XEN_HVM in self.cfg.GetClusterInfo().enabled_hypervisors:
+    enabled_hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
+    if constants.HTS_USE_VNC.intersection(enabled_hypervisors):
       to_copy.append(constants.VNC_PASSWORD_FILE)
+
     for fname in to_copy:
       result = self.rpc.call_upload_file([node], fname)
       if result[node].failed or not result[node]:
@@ -2627,9 +2669,10 @@ class LUStartupInstance(LogicalUnit):
     _StartInstanceDisks(self, instance, force)
 
     result = self.rpc.call_instance_start(node_current, instance, extra_args)
-    if result.failed or not result.data:
+    msg = result.RemoteFailMsg()
+    if msg:
       _ShutdownInstanceDisks(self, instance)
-      raise errors.OpExecError("Could not start instance")
+      raise errors.OpExecError("Could not start instance: %s" % msg)
 
 
 class LURebootInstance(LogicalUnit):
@@ -2702,9 +2745,11 @@ class LURebootInstance(LogicalUnit):
       _ShutdownInstanceDisks(self, instance)
       _StartInstanceDisks(self, instance, ignore_secondaries)
       result = self.rpc.call_instance_start(node_current, instance, extra_args)
-      if result.failed or not result.data:
+      msg = result.RemoteFailMsg()
+      if msg:
         _ShutdownInstanceDisks(self, instance)
-        raise errors.OpExecError("Could not start instance for full reboot")
+        raise errors.OpExecError("Could not start instance for"
+                                 " full reboot: %s" % msg)
 
     self.cfg.MarkInstanceUp(instance.name)
 
@@ -2792,7 +2837,7 @@ class LUReinstallInstance(LogicalUnit):
     if instance.disk_template == constants.DT_DISKLESS:
       raise errors.OpPrereqError("Instance '%s' has no disks" %
                                  self.op.instance_name)
-    if instance.status != "down":
+    if instance.admin_up:
       raise errors.OpPrereqError("Instance '%s' is marked to be up" %
                                  self.op.instance_name)
     remote_info = self.rpc.call_instance_info(instance.primary_node,
@@ -2834,11 +2879,11 @@ class LUReinstallInstance(LogicalUnit):
     try:
       feedback_fn("Running the instance OS create scripts...")
       result = self.rpc.call_instance_os_add(inst.primary_node, inst)
-      result.Raise()
-      if not result.data:
+      msg = result.RemoteFailMsg()
+      if msg:
         raise errors.OpExecError("Could not install OS for instance %s"
-                                 " on node %s" %
-                                 (inst.name, inst.primary_node))
+                                 " on node %s: %s" %
+                                 (inst.name, inst.primary_node, msg))
     finally:
       _ShutdownInstanceDisks(self, inst)
 
@@ -2875,7 +2920,7 @@ class LURenameInstance(LogicalUnit):
                                  self.op.instance_name)
     _CheckNodeOnline(self, instance.primary_node)
 
-    if instance.status != "down":
+    if instance.admin_up:
       raise errors.OpPrereqError("Instance '%s' is marked to be up" %
                                  self.op.instance_name)
     remote_info = self.rpc.call_instance_info(instance.primary_node,
@@ -2944,10 +2989,11 @@ class LURenameInstance(LogicalUnit):
     try:
       result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
                                                  old_name)
-      if result.failed or not result.data:
+      msg = result.RemoteFailMsg()
+      if msg:
         msg = ("Could not run OS rename script for instance %s on node %s"
-               " (but the instance has been renamed in Ganeti)" %
-               (inst.name, inst.primary_node))
+               " (but the instance has been renamed in Ganeti): %s" %
+               (inst.name, inst.primary_node, msg))
         self.proc.LogWarning(msg)
     finally:
       _ShutdownInstanceDisks(self, inst)
@@ -3080,19 +3126,25 @@ class LUQueryInstances(NoHooksLU):
 
     """
     all_info = self.cfg.GetAllInstancesInfo()
-    if self.do_locking:
-      instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
-    elif self.wanted != locking.ALL_SET:
-      instance_names = self.wanted
-      missing = set(instance_names).difference(all_info.keys())
-      if missing:
-        raise errors.OpExecError(
-          "Some instances were removed before retrieving their data: %s"
-          % missing)
+    if self.wanted == locking.ALL_SET:
+      # caller didn't specify instance names, so ordering is not important
+      if self.do_locking:
+        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
+      else:
+        instance_names = all_info.keys()
+      instance_names = utils.NiceSort(instance_names)
     else:
-      instance_names = all_info.keys()
+      # caller did specify names, so we must keep the ordering
+      if self.do_locking:
+        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
+      else:
+        tgt_set = all_info.keys()
+      missing = set(self.wanted).difference(tgt_set)
+      if missing:
+        raise errors.OpExecError("Some instances were removed before"
+                                 " retrieving their data: %s" % missing)
+      instance_names = self.wanted
 
-    instance_names = utils.NiceSort(instance_names)
     instance_list = [all_info[iname] for iname in instance_names]
 
     # begin data gathering
@@ -3139,7 +3191,7 @@ class LUQueryInstances(NoHooksLU):
         elif field == "snodes":
           val = list(instance.secondary_nodes)
         elif field == "admin_state":
-          val = (instance.status != "down")
+          val = instance.admin_up
         elif field == "oper_state":
           if instance.primary_node in bad_nodes:
             val = None
@@ -3153,12 +3205,12 @@ class LUQueryInstances(NoHooksLU):
           else:
             running = bool(live_data.get(instance.name))
             if running:
-              if instance.status != "down":
+              if instance.admin_up:
                 val = "running"
               else:
                 val = "ERROR_up"
             else:
-              if instance.status != "down":
+              if instance.admin_up:
                 val = "ERROR_down"
               else:
                 val = "ADMIN_down"
@@ -3332,7 +3384,7 @@ class LUFailoverInstance(LogicalUnit):
     for dev in instance.disks:
       # for drbd, these are drbd over lvm
       if not _CheckDiskConsistency(self, dev, target_node, False):
-        if instance.status == "up" and not self.op.ignore_consistency:
+        if instance.admin_up and not self.op.ignore_consistency:
           raise errors.OpExecError("Disk %s is degraded on target node,"
                                    " aborting failover." % dev.iv_name)
 
@@ -3360,7 +3412,7 @@ class LUFailoverInstance(LogicalUnit):
     self.cfg.Update(instance)
 
     # Only start the instance if it's marked as up
-    if instance.status == "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)
@@ -3373,10 +3425,11 @@ class LUFailoverInstance(LogicalUnit):
 
       feedback_fn("* starting the instance on the target node")
       result = self.rpc.call_instance_start(target_node, instance, None)
-      if result.failed or not result.data:
+      msg = result.RemoteFailMsg()
+      if msg:
         _ShutdownInstanceDisks(self, instance)
-        raise errors.OpExecError("Could not start instance %s on node %s." %
-                                 (instance.name, target_node))
+        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
+                                 (instance.name, target_node, msg))
 
 
 class LUMigrateInstance(LogicalUnit):
@@ -3600,6 +3653,41 @@ class LUMigrateInstance(LogicalUnit):
 
     self.feedback_fn("* done")
 
+  def _RevertDiskStatus(self):
+    """Try to revert the disk status after a failed migration.
+
+    """
+    target_node = self.target_node
+    try:
+      self._EnsureSecondary(target_node)
+      self._GoStandalone()
+      self._GoReconnect(False)
+      self._WaitUntilSync()
+    except errors.OpExecError, err:
+      self.LogWarning("Migration failed and I can't reconnect the"
+                      " drives: error '%s'\n"
+                      "Please look and recover the instance status" %
+                      str(err))
+
+  def _AbortMigration(self):
+    """Call the hypervisor code to abort a started migration.
+
+    """
+    instance = self.instance
+    target_node = self.target_node
+    migration_info = self.migration_info
+
+    abort_result = self.rpc.call_finalize_migration(target_node,
+                                                    instance,
+                                                    migration_info,
+                                                    False)
+    abort_msg = abort_result.RemoteFailMsg()
+    if abort_msg:
+      logging.error("Aborting migration failed on target node %s: %s" %
+                    (target_node, abort_msg))
+      # Don't raise an exception here, as we stil have to try to revert the
+      # disk status, even if this step failed.
+
   def _ExecMigration(self):
     """Migrate an instance.
 
@@ -3623,11 +3711,38 @@ class LUMigrateInstance(LogicalUnit):
                                  " synchronized on target node,"
                                  " aborting migrate." % dev.iv_name)
 
+    # First get the migration information from the remote node
+    result = self.rpc.call_migration_info(source_node, instance)
+    msg = result.RemoteFailMsg()
+    if msg:
+      log_err = ("Failed fetching source migration information from %s: %s" %
+                  (source_node, msg))
+      logging.error(log_err)
+      raise errors.OpExecError(log_err)
+
+    self.migration_info = migration_info = result.data[1]
+
+    # Then switch the disks to master/master mode
     self._EnsureSecondary(target_node)
     self._GoStandalone()
     self._GoReconnect(True)
     self._WaitUntilSync()
 
+    self.feedback_fn("* preparing %s to accept the instance" % target_node)
+    result = self.rpc.call_accept_instance(target_node,
+                                           instance,
+                                           migration_info,
+                                           self.nodes_ip[target_node])
+
+    msg = result.RemoteFailMsg()
+    if msg:
+      logging.error("Instance pre-migration failed, trying to revert"
+                    " disk status: %s", msg)
+      self._AbortMigration()
+      self._RevertDiskStatus()
+      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
+                               (instance.name, msg))
+
     self.feedback_fn("* migrating instance to %s" % target_node)
     time.sleep(10)
     result = self.rpc.call_instance_migrate(source_node, instance,
@@ -3637,17 +3752,8 @@ class LUMigrateInstance(LogicalUnit):
     if msg:
       logging.error("Instance migration failed, trying to revert"
                     " disk status: %s", msg)
-      try:
-        self._EnsureSecondary(target_node)
-        self._GoStandalone()
-        self._GoReconnect(False)
-        self._WaitUntilSync()
-      except errors.OpExecError, err:
-        self.LogWarning("Migration failed and I can't reconnect the"
-                        " drives: error '%s'\n"
-                        "Please look and recover the instance status" %
-                        str(err))
-
+      self._AbortMigration()
+      self._RevertDiskStatus()
       raise errors.OpExecError("Could not migrate instance %s: %s" %
                                (instance.name, msg))
     time.sleep(10)
@@ -3656,6 +3762,17 @@ class LUMigrateInstance(LogicalUnit):
     # distribute new instance config to the other nodes
     self.cfg.Update(instance)
 
+    result = self.rpc.call_finalize_migration(target_node,
+                                              instance,
+                                              migration_info,
+                                              True)
+    msg = result.RemoteFailMsg()
+    if msg:
+      logging.error("Instance migration succeeded, but finalization failed:"
+                    " %s" % msg)
+      raise errors.OpExecError("Could not finalize instance migration: %s" %
+                               msg)
+
     self._EnsureSecondary(source_node)
     self._WaitUntilSync()
     self._GoStandalone()
@@ -3817,7 +3934,8 @@ def _GenerateDiskTemplate(lu, template_name,
       disk_index = idx + base_index
       disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
                               logical_id=(vgname, names[idx]),
-                              iv_name="disk/%d" % disk_index)
+                              iv_name="disk/%d" % disk_index,
+                              mode=disk["mode"])
       disks.append(disk_dev)
   elif template_name == constants.DT_DRBD8:
     if len(secondary_nodes) != 1:
@@ -3837,6 +3955,7 @@ def _GenerateDiskTemplate(lu, template_name,
                                       disk["size"], names[idx*2:idx*2+2],
                                       "disk/%d" % disk_index,
                                       minors[idx*2], minors[idx*2+1])
+      disk_dev.mode = disk["mode"]
       disks.append(disk_dev)
   elif template_name == constants.DT_FILE:
     if len(secondary_nodes) != 0:
@@ -3848,7 +3967,8 @@ def _GenerateDiskTemplate(lu, template_name,
                               iv_name="disk/%d" % disk_index,
                               logical_id=(file_driver,
                                           "%s/disk%d" % (file_storage_dir,
-                                                         idx)))
+                                                         idx)),
+                              mode=disk["mode"])
       disks.append(disk_dev)
   else:
     raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
@@ -3981,6 +4101,8 @@ def _CheckHVParams(lu, nodenames, hvname, hvparams):
                                                   hvparams)
   for node in nodenames:
     info = hvinfo[node]
+    if info.offline:
+      continue
     info.Raise()
     if not info.data or not isinstance(info.data, (tuple, list)):
       raise errors.OpPrereqError("Cannot get current information"
@@ -4094,7 +4216,9 @@ class LUCreateInstance(LogicalUnit):
           raise errors.OpPrereqError("Invalid MAC address specified: %s" %
                                      mac)
       # bridge verification
-      bridge = nic.get("bridge", self.cfg.GetDefBridge())
+      bridge = nic.get("bridge", None)
+      if bridge is None:
+        bridge = self.cfg.GetDefBridge()
       self.nics.append(objects.NIC(mac=mac, ip=nic_ip, bridge=bridge))
 
     # disk checks/pre-build
@@ -4396,10 +4520,7 @@ class LUCreateInstance(LogicalUnit):
                            self.be_full[constants.BE_MEMORY],
                            self.op.hypervisor)
 
-    if self.op.start:
-      self.instance_status = 'up'
-    else:
-      self.instance_status = 'down'
+    self.instance_status = self.op.start
 
   def Exec(self, feedback_fn):
     """Create and add the instance to the cluster.
@@ -4446,7 +4567,7 @@ class LUCreateInstance(LogicalUnit):
                             primary_node=pnode_name,
                             nics=self.nics, disks=disks,
                             disk_template=self.op.disk_template,
-                            status=self.instance_status,
+                            admin_up=self.instance_status,
                             network_port=network_port,
                             beparams=self.op.beparams,
                             hvparams=self.op.hvparams,
@@ -4470,8 +4591,6 @@ class LUCreateInstance(LogicalUnit):
     # Declare that we don't want to remove the instance lock anymore, as we've
     # added the instance to the config
     del self.remove_locks[locking.LEVEL_INSTANCE]
-    # Remove the temp. assignements for the instance's drbds
-    self.cfg.ReleaseDRBDMinors(instance)
     # Unlock all the nodes
     if self.op.mode == constants.INSTANCE_IMPORT:
       nodes_keep = [self.op.src_node]
@@ -4508,11 +4627,11 @@ class LUCreateInstance(LogicalUnit):
       if self.op.mode == constants.INSTANCE_CREATE:
         feedback_fn("* running the instance OS create scripts...")
         result = self.rpc.call_instance_os_add(pnode_name, iobj)
-        result.Raise()
-        if not result.data:
+        msg = result.RemoteFailMsg()
+        if msg:
           raise errors.OpExecError("Could not add os for instance %s"
-                                   " on node %s" %
-                                   (instance, pnode_name))
+                                   " on node %s: %s" %
+                                   (instance, pnode_name, msg))
 
       elif self.op.mode == constants.INSTANCE_IMPORT:
         feedback_fn("* running the instance OS import scripts...")
@@ -4537,9 +4656,9 @@ class LUCreateInstance(LogicalUnit):
       logging.info("Starting instance %s on node %s", instance, pnode_name)
       feedback_fn("* starting instance...")
       result = self.rpc.call_instance_start(pnode_name, iobj, None)
-      result.Raise()
-      if not result.data:
-        raise errors.OpExecError("Could not start instance")
+      msg = result.RemoteFailMsg()
+      if msg:
+        raise errors.OpExecError("Could not start instance: %s" % msg)
 
 
 class LUConnectConsole(NoHooksLU):
@@ -4584,7 +4703,12 @@ class LUConnectConsole(NoHooksLU):
     logging.debug("Connecting to console of %s on %s", instance.name, node)
 
     hyper = hypervisor.GetHypervisor(instance.hypervisor)
-    console_cmd = hyper.GetShellCommandForConsole(instance)
+    cluster = self.cfg.GetClusterInfo()
+    # beparams and hvparams are passed separately, to avoid editing the
+    # instance and then saving the defaults in the instance itself.
+    hvparams = cluster.FillHV(instance)
+    beparams = cluster.FillBE(instance)
+    console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
 
     # build ssh cmdline
     return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
@@ -4632,6 +4756,10 @@ class LUReplaceDisks(LogicalUnit):
         raise errors.OpPrereqError("Node '%s' not known" %
                                    self.op.remote_node)
       self.op.remote_node = remote_node
+      # Warning: do not remove the locking of the new secondary here
+      # unless DRBD8.AddChildren is changed to work in parallel;
+      # currently it doesn't since parallel invocations of
+      # FindUnusedMinor will conflict
       self.needed_locks[locking.LEVEL_NODE] = [remote_node]
       self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
     else:
@@ -5037,7 +5165,7 @@ class LUReplaceDisks(LogicalUnit):
       try:
         _CreateSingleBlockDev(self, new_node, instance, new_drbd,
                               _GetInstanceInfoText(instance), False)
-      except error.BlockDeviceError:
+      except errors.BlockDeviceError:
         self.cfg.ReleaseDRBDMinors(instance.name)
         raise
 
@@ -5068,9 +5196,6 @@ class LUReplaceDisks(LogicalUnit):
       dev.logical_id = new_logical_id
       cfg.SetDiskID(dev, pri_node)
     cfg.Update(instance)
-    # we can remove now the temp minors as now the new values are
-    # written to the config file (and therefore stable)
-    self.cfg.ReleaseDRBDMinors(instance.name)
 
     # and now perform the drbd attach
     info("attaching primary drbds to new secondary (standalone => connected)")
@@ -5117,7 +5242,7 @@ class LUReplaceDisks(LogicalUnit):
     instance = self.instance
 
     # Activate the instance disks if we're replacing them on a down instance
-    if instance.status == "down":
+    if not instance.admin_up:
       _StartInstanceDisks(self, instance, True)
 
     if self.op.mode == constants.REPLACE_DISK_CHG:
@@ -5128,7 +5253,7 @@ class LUReplaceDisks(LogicalUnit):
     ret = fn(feedback_fn)
 
     # Deactivate the instance disks if we're replacing them on a down instance
-    if instance.status == "down":
+    if not instance.admin_up:
       _SafeShutdownInstanceDisks(self, instance)
 
     return ret
@@ -5344,10 +5469,10 @@ class LUQueryInstanceData(NoHooksLU):
           remote_state = "down"
       else:
         remote_state = None
-      if instance.status == "down":
-        config_state = "down"
-      else:
+      if instance.admin_up:
         config_state = "up"
+      else:
+        config_state = "down"
 
       disks = [self._ComputeDiskStatus(instance, None, device)
                for device in instance.disks]
@@ -5412,7 +5537,7 @@ class LUSetInstanceParams(LogicalUnit):
           raise errors.OpPrereqError("Invalid disk index")
       if disk_op == constants.DDM_ADD:
         mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
-        if mode not in (constants.DISK_RDONLY, constants.DISK_RDWR):
+        if mode not in constants.DISK_ACCESS_SET:
           raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode)
         size = disk_dict.get('size', None)
         if size is None:
@@ -5682,13 +5807,12 @@ class LUSetInstanceParams(LogicalUnit):
         disk_idx_base = len(instance.disks)
         new_disk = _GenerateDiskTemplate(self,
                                          instance.disk_template,
-                                         instance, instance.primary_node,
+                                         instance.name, instance.primary_node,
                                          instance.secondary_nodes,
                                          [disk_dict],
                                          file_path,
                                          file_driver,
                                          disk_idx_base)[0]
-        new_disk.mode = disk_dict['mode']
         instance.disks.append(new_disk)
         info = _GetInstanceInfoText(instance)
 
@@ -5701,7 +5825,7 @@ class LUSetInstanceParams(LogicalUnit):
           try:
             _CreateBlockDev(self, node, instance, new_disk,
                             f_create, info, f_create)
-          except error.OpExecError, err:
+          except errors.OpExecError, err:
             self.LogWarning("Failed to create volume %s (%s) on"
                             " node %s: %s",
                             new_disk.iv_name, new_disk, node, err)
@@ -5903,11 +6027,12 @@ class LUExportInstance(LogicalUnit):
           snap_disks.append(new_dev)
 
     finally:
-      if self.op.shutdown and instance.status == "up":
+      if self.op.shutdown and instance.admin_up:
         result = self.rpc.call_instance_start(src_node, instance, None)
-        if result.failed or not result.data:
+        msg = result.RemoteFailMsg()
+        if msg:
           _ShutdownInstanceDisks(self, instance)
-          raise errors.OpExecError("Could not start instance")
+          raise errors.OpExecError("Could not start instance: %s" % msg)
 
     # TODO: check for size
 
@@ -6278,7 +6403,7 @@ class IAllocator(object):
       "version": 1,
       "cluster_name": cfg.GetClusterName(),
       "cluster_tags": list(cluster_info.GetTags()),
-      "enable_hypervisors": list(cluster_info.enabled_hypervisors),
+      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
       # we don't have job IDs
       }
     iinfo = cfg.GetAllInstancesInfo().values()
@@ -6297,52 +6422,60 @@ class IAllocator(object):
                                            hypervisor_name)
     node_iinfo = self.lu.rpc.call_all_instances_info(node_list,
                        cluster_info.enabled_hypervisors)
-    for nname in node_list:
+    for nname, nresult in node_data.items():
+      # first fill in static (config-based) values
       ninfo = cfg.GetNodeInfo(nname)
-      node_data[nname].Raise()
-      if not isinstance(node_data[nname].data, dict):
-        raise errors.OpExecError("Can't get data for node %s" % nname)
-      remote_info = node_data[nname].data
-      for attr in ['memory_total', 'memory_free', 'memory_dom0',
-                   'vg_size', 'vg_free', 'cpu_total']:
-        if attr not in remote_info:
-          raise errors.OpExecError("Node '%s' didn't return attribute '%s'" %
-                                   (nname, attr))
-        try:
-          remote_info[attr] = int(remote_info[attr])
-        except ValueError, err:
-          raise errors.OpExecError("Node '%s' returned invalid value for '%s':"
-                                   " %s" % (nname, attr, str(err)))
-      # compute memory used by primary instances
-      i_p_mem = i_p_up_mem = 0
-      for iinfo, beinfo in i_list:
-        if iinfo.primary_node == nname:
-          i_p_mem += beinfo[constants.BE_MEMORY]
-          if iinfo.name not in node_iinfo[nname]:
-            i_used_mem = 0
-          else:
-            i_used_mem = int(node_iinfo[nname][iinfo.name]['memory'])
-          i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
-          remote_info['memory_free'] -= max(0, i_mem_diff)
-
-          if iinfo.status == "up":
-            i_p_up_mem += beinfo[constants.BE_MEMORY]
-
-      # compute memory used by instances
       pnr = {
         "tags": list(ninfo.GetTags()),
-        "total_memory": remote_info['memory_total'],
-        "reserved_memory": remote_info['memory_dom0'],
-        "free_memory": remote_info['memory_free'],
-        "i_pri_memory": i_p_mem,
-        "i_pri_up_memory": i_p_up_mem,
-        "total_disk": remote_info['vg_size'],
-        "free_disk": remote_info['vg_free'],
         "primary_ip": ninfo.primary_ip,
         "secondary_ip": ninfo.secondary_ip,
-        "total_cpus": remote_info['cpu_total'],
         "offline": ninfo.offline,
+        "master_candidate": ninfo.master_candidate,
         }
+
+      if not ninfo.offline:
+        nresult.Raise()
+        if not isinstance(nresult.data, dict):
+          raise errors.OpExecError("Can't get data for node %s" % nname)
+        remote_info = nresult.data
+        for attr in ['memory_total', 'memory_free', 'memory_dom0',
+                     'vg_size', 'vg_free', 'cpu_total']:
+          if attr not in remote_info:
+            raise errors.OpExecError("Node '%s' didn't return attribute"
+                                     " '%s'" % (nname, attr))
+          try:
+            remote_info[attr] = int(remote_info[attr])
+          except ValueError, err:
+            raise errors.OpExecError("Node '%s' returned invalid value"
+                                     " for '%s': %s" % (nname, attr, err))
+        # compute memory used by primary instances
+        i_p_mem = i_p_up_mem = 0
+        for iinfo, beinfo in i_list:
+          if iinfo.primary_node == nname:
+            i_p_mem += beinfo[constants.BE_MEMORY]
+            if iinfo.name not in node_iinfo[nname].data:
+              i_used_mem = 0
+            else:
+              i_used_mem = int(node_iinfo[nname].data[iinfo.name]['memory'])
+            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
+            remote_info['memory_free'] -= max(0, i_mem_diff)
+
+            if iinfo.admin_up:
+              i_p_up_mem += beinfo[constants.BE_MEMORY]
+
+        # compute memory used by instances
+        pnr_dyn = {
+          "total_memory": remote_info['memory_total'],
+          "reserved_memory": remote_info['memory_dom0'],
+          "free_memory": remote_info['memory_free'],
+          "total_disk": remote_info['vg_size'],
+          "free_disk": remote_info['vg_free'],
+          "total_cpus": remote_info['cpu_total'],
+          "i_pri_memory": i_p_mem,
+          "i_pri_up_memory": i_p_up_mem,
+          }
+        pnr.update(pnr_dyn)
+
       node_results[nname] = pnr
     data["nodes"] = node_results
 
@@ -6353,13 +6486,13 @@ class IAllocator(object):
                   for n in iinfo.nics]
       pir = {
         "tags": list(iinfo.GetTags()),
-        "should_run": iinfo.status == "up",
+        "admin_up": iinfo.admin_up,
         "vcpus": beinfo[constants.BE_VCPUS],
         "memory": beinfo[constants.BE_MEMORY],
         "os": iinfo.os,
-        "nodes": list(iinfo.all_nodes),
+        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
         "nics": nic_data,
-        "disks": [{"size": dsk.size, "mode": "w"} for dsk in iinfo.disks],
+        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
         "disk_template": iinfo.disk_template,
         "hypervisor": iinfo.hypervisor,
         }
@@ -6380,8 +6513,6 @@ class IAllocator(object):
 
     """
     data = self.in_data
-    if len(self.disks) != 2:
-      raise errors.OpExecError("Only two-disk configurations supported")
 
     disk_space = _ComputeDiskSize(self.disk_template, self.disks)
 
@@ -6538,8 +6669,6 @@ class LUTestAllocator(NoHooksLU):
                                      " 'nics' parameter")
       if not isinstance(self.op.disks, list):
         raise errors.OpPrereqError("Invalid parameter 'disks'")
-      if len(self.op.disks) != 2:
-        raise errors.OpPrereqError("Only two-disk configurations supported")
       for row in self.op.disks:
         if (not isinstance(row, dict) or
             "size" not in row or
@@ -6548,7 +6677,7 @@ class LUTestAllocator(NoHooksLU):
             row["mode"] not in ['r', 'w']):
           raise errors.OpPrereqError("Invalid contents of the"
                                      " 'disks' parameter")
-      if self.op.hypervisor is None:
+      if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
         self.op.hypervisor = self.cfg.GetHypervisorType()
     elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
       if not hasattr(self.op, "name"):