Small improvements for cluster verify
[ganeti-local] / lib / cmdlib.py
index 782cc2f..1e0ac3e 100644 (file)
@@ -583,6 +583,30 @@ def _CheckInstanceNodeGroups(cfg, instance_name, owned_groups):
   return inst_groups
 
 
+def _CheckNodeGroupInstances(cfg, group_uuid, owned_instances):
+  """Checks if the instances in a node group are still correct.
+
+  @type cfg: L{config.ConfigWriter}
+  @param cfg: The cluster configuration
+  @type group_uuid: string
+  @param group_uuid: Node group UUID
+  @type owned_instances: set or frozenset
+  @param owned_instances: List of currently owned instances
+
+  """
+  wanted_instances = cfg.GetNodeGroupInstances(group_uuid)
+  if owned_instances != wanted_instances:
+    raise errors.OpPrereqError("Instances in node group '%s' changed since"
+                               " locks were acquired, wanted '%s', have '%s';"
+                               " retry the operation" %
+                               (group_uuid,
+                                utils.CommaJoin(wanted_instances),
+                                utils.CommaJoin(owned_instances)),
+                               errors.ECODE_STATE)
+
+  return wanted_instances
+
+
 def _SupportsOob(cfg, node):
   """Tells if node supports OOB.
 
@@ -1486,6 +1510,47 @@ class _VerifyErrors(object):
       self.bad = self.bad or cond
 
 
+class LUClusterVerify(NoHooksLU):
+  """Submits all jobs necessary to verify the cluster.
+
+  """
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self.needed_locks = {}
+
+  def Exec(self, feedback_fn):
+    jobs = []
+
+    if self.op.group_name:
+      groups = [self.op.group_name]
+      depends_fn = lambda: None
+    else:
+      groups = self.cfg.GetNodeGroupList()
+
+      # Verify global configuration
+      jobs.append([opcodes.OpClusterVerifyConfig()])
+
+      # Always depend on global verification
+      depends_fn = lambda: [(-len(jobs), [])]
+
+    jobs.extend([opcodes.OpClusterVerifyGroup(group_name=group,
+                                              depends=depends_fn())]
+                for group in groups)
+
+    # Fix up all parameters
+    for op in itertools.chain(*jobs): # pylint: disable-msg=W0142
+      op.debug_simulate_errors = self.op.debug_simulate_errors
+      op.verbose = self.op.verbose
+      op.error_codes = self.op.error_codes
+      try:
+        op.skip_checks = self.op.skip_checks
+      except AttributeError:
+        assert not isinstance(op, opcodes.OpClusterVerifyGroup)
+
+    return ResultWithJobs(jobs)
+
+
 class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
   """Verifies the cluster config.
 
@@ -1509,6 +1574,7 @@ class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
   def ExpandNames(self):
     # Information can be safely retrieved as the BGL is acquired in exclusive
     # mode
+    assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER)
     self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
     self.all_node_info = self.cfg.GetAllNodesInfo()
     self.all_inst_info = self.cfg.GetAllInstancesInfo()
@@ -1570,7 +1636,7 @@ class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
                   "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()])
+    return not self.bad
 
 
 class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
@@ -1668,7 +1734,10 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       self.needed_locks[locking.LEVEL_NODE] = nodes
 
   def CheckPrereq(self):
-    group_nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
+    assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
+    self.group_info = self.cfg.GetNodeGroup(self.group_uuid)
+
+    group_nodes = set(self.group_info.members)
     group_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
 
     unlocked_nodes = \
@@ -2500,6 +2569,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     """
     # This method has too many local variables. pylint: disable-msg=R0914
+    feedback_fn("* Verifying group '%s'" % self.group_info.name)
 
     if not self.my_node_names:
       # empty node group
@@ -2997,27 +3067,21 @@ class LUGroupVerifyDisks(NoHooksLU):
     assert self.group_uuid in owned_groups
 
     # Check if locked instances are still correct
-    wanted_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
-    if owned_instances != wanted_instances:
-      raise errors.OpPrereqError("Instances in node group %s changed since"
-                                 " locks were acquired, wanted %s, have %s;"
-                                 " retry the operation" %
-                                 (self.op.group_name,
-                                  utils.CommaJoin(wanted_instances),
-                                  utils.CommaJoin(owned_instances)),
-                                 errors.ECODE_STATE)
+    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
 
     # Get instance information
     self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
 
     # Check if node groups for locked instances are still correct
     for (instance_name, inst) in self.instances.items():
-      assert self.group_uuid in self.cfg.GetInstanceNodeGroups(instance_name), \
-        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
       assert owned_nodes.issuperset(inst.all_nodes), \
         "Instance %s's nodes changed while we kept the lock" % instance_name
 
-      _CheckInstanceNodeGroups(self.cfg, instance_name, owned_groups)
+      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
+                                             owned_groups)
+
+      assert self.group_uuid in inst_groups, \
+        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
 
   def Exec(self, feedback_fn):
     """Verify integrity of cluster disks.
@@ -4279,7 +4343,7 @@ class _NodeQuery(_QueryBase):
 
   def ExpandNames(self, lu):
     lu.needed_locks = {}
-    lu.share_locks[locking.LEVEL_NODE] = 1
+    lu.share_locks = _ShareAll()
 
     if self.names:
       self.wanted = _GetWantedNodes(lu, self.names)
@@ -4290,7 +4354,7 @@ class _NodeQuery(_QueryBase):
                        query.NQ_LIVE in self.requested_data)
 
     if self.do_locking:
-      # if we don't request only static fields, we need to lock the nodes
+      # If any non-static field is requested we need to lock the nodes
       lu.needed_locks[locking.LEVEL_NODE] = self.wanted
 
   def DeclareLocks(self, lu, level):
@@ -4662,7 +4726,7 @@ class LUQuery(NoHooksLU):
   def CheckArguments(self):
     qcls = _GetQueryImplementation(self.op.what)
 
-    self.impl = qcls(self.op.filter, self.op.fields, False)
+    self.impl = qcls(self.op.filter, self.op.fields, self.op.use_locking)
 
   def ExpandNames(self):
     self.impl.ExpandNames(self)
@@ -8873,7 +8937,6 @@ class LUInstanceCreate(LogicalUnit):
       disk_abort = not _WaitForSync(self, iobj)
     elif iobj.disk_template in constants.DTS_INT_MIRROR:
       # make sure the disks are not degraded (still sync-ing is ok)
-      time.sleep(15)
       feedback_fn("* checking mirrors status")
       disk_abort = not _WaitForSync(self, iobj, oneshot=True)
     else:
@@ -9251,6 +9314,9 @@ class TLReplaceDisks(Tasklet):
     return remote_node_name
 
   def _FindFaultyDisks(self, node_name):
+    """Wrapper for L{_FindFaultyInstanceDisks}.
+
+    """
     return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
                                     node_name, True)
 
@@ -11996,6 +12062,9 @@ class LUGroupQuery(NoHooksLU):
   def ExpandNames(self):
     self.gq.ExpandNames(self)
 
+  def DeclareLocks(self, level):
+    self.gq.DeclareLocks(self, level)
+
   def Exec(self, feedback_fn):
     return self.gq.OldStyleQuery(self)
 
@@ -12293,15 +12362,7 @@ class LUGroupEvacuate(LogicalUnit):
     assert self.group_uuid in owned_groups
 
     # Check if locked instances are still correct
-    wanted_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
-    if owned_instances != wanted_instances:
-      raise errors.OpPrereqError("Instances in node group to be evacuated (%s)"
-                                 " changed since locks were acquired, wanted"
-                                 " %s, have %s; retry the operation" %
-                                 (self.group_uuid,
-                                  utils.CommaJoin(wanted_instances),
-                                  utils.CommaJoin(owned_instances)),
-                                 errors.ECODE_STATE)
+    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
 
     # Get instance information
     self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
@@ -12747,7 +12808,6 @@ class IAllocator(object):
     self.hypervisor = None
     self.relocate_from = None
     self.name = None
-    self.evac_nodes = None
     self.instances = None
     self.evac_mode = None
     self.target_groups = []
@@ -13029,15 +13089,6 @@ class IAllocator(object):
       }
     return request
 
-  def _AddEvacuateNodes(self):
-    """Add evacuate nodes data to allocator structure.
-
-    """
-    request = {
-      "evac_nodes": self.evac_nodes
-      }
-    return request
-
   def _AddNodeEvacuate(self):
     """Get data for node-evacuate requests.
 
@@ -13118,9 +13169,6 @@ class IAllocator(object):
       (_AddRelocateInstance,
        [("name", ht.TString), ("relocate_from", _STRING_LIST)],
        ht.TList),
-    constants.IALLOCATOR_MODE_MEVAC:
-      (_AddEvacuateNodes, [("evac_nodes", _STRING_LIST)],
-       ht.TListOf(ht.TAnd(ht.TIsLength(2), _STRING_LIST))),
      constants.IALLOCATOR_MODE_NODE_EVAC:
       (_AddNodeEvacuate, [
         ("instances", _STRING_LIST),
@@ -13179,39 +13227,25 @@ class IAllocator(object):
                                (self._result_check, self.result),
                                errors.ECODE_INVAL)
 
-    if self.mode in (constants.IALLOCATOR_MODE_RELOC,
-                     constants.IALLOCATOR_MODE_MEVAC):
+    if self.mode == constants.IALLOCATOR_MODE_RELOC:
+      assert self.relocate_from is not None
+      assert self.required_nodes == 1
+
       node2group = dict((name, ndata["group"])
                         for (name, ndata) in self.in_data["nodes"].items())
 
       fn = compat.partial(self._NodesToGroups, node2group,
                           self.in_data["nodegroups"])
 
-      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)
+      instance = self.cfg.GetInstanceInfo(self.name)
+      request_groups = fn(self.relocate_from + [instance.primary_node])
+      result_groups = fn(rdict["result"] + [instance.primary_node])
+
+      if self.success and not set(result_groups).issubset(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_NODE_EVAC:
       assert self.evac_mode in constants.IALLOCATOR_NEVAC_MODES
@@ -13295,10 +13329,6 @@ class LUTestAllocator(NoHooksLU):
       self.op.name = fname
       self.relocate_from = \
           list(self.cfg.GetInstanceInfo(fname).secondary_nodes)
-    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
-      if not hasattr(self.op, "evac_nodes"):
-        raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
-                                   " opcode input", errors.ECODE_INVAL)
     elif self.op.mode in (constants.IALLOCATOR_MODE_CHG_GROUP,
                           constants.IALLOCATOR_MODE_NODE_EVAC):
       if not self.op.instances:
@@ -13339,10 +13369,6 @@ class LUTestAllocator(NoHooksLU):
                        name=self.op.name,
                        relocate_from=list(self.relocate_from),
                        )
-    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
-      ial = IAllocator(self.cfg, self.rpc,
-                       mode=self.op.mode,
-                       evac_nodes=self.op.evac_nodes)
     elif self.op.mode == constants.IALLOCATOR_MODE_CHG_GROUP:
       ial = IAllocator(self.cfg, self.rpc,
                        mode=self.op.mode,