Fix gnt-cluster getmaster on non-master nodes

[ganeti-local] / lib / cmdlib.py

diff --git a/lib/cmdlib.py b/lib/cmdlib.py
index a3150b1..0e2e184 100644 (file)
--- a/lib/cmdlib.py
+++ b/lib/cmdlib.py
@@ -1011,8 +1011,16 @@ class LUVerifyCluster(LogicalUnit):
 
       node_drbd = {}
       for minor, instance in all_drbd_map[node].items():
-        instance = instanceinfo[instance]
-        node_drbd[minor] = (instance.name, instance.admin_up)
+        if instance not in instanceinfo:
+          feedback_fn("  - ERROR: ghost instance '%s' in temporary DRBD map" %
+                      instance)
+          # ghost instance should not be running, but otherwise we
+          # don't give double warnings (both ghost instance and
+          # unallocated minor in use)
+          node_drbd[minor] = (instance, False)
+        else:
+          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,
                                 node_drbd, vg_name)
@@ -1065,9 +1073,17 @@ class LUVerifyCluster(LogicalUnit):
         }
         # FIXME: devise a free space model for file based instances as well
         if vg_name is not None:
+          if (constants.NV_VGLIST not in nresult or
+              vg_name not in nresult[constants.NV_VGLIST]):
+            feedback_fn("  - ERROR: node %s didn't return data for the"
+                        " volume group '%s' - it is either missing or broken" %
+                        (node, vg_name))
+            bad = True
+            continue
           node_info[node]["dfree"] = int(nresult[constants.NV_VGLIST][vg_name])
-      except ValueError:
-        feedback_fn("  - ERROR: invalid value returned from node %s" % (node,))
+      except (ValueError, KeyError):
+        feedback_fn("  - ERROR: invalid nodeinfo value returned"
+                    " from node %s" % (node,))
         bad = True
         continue
 
@@ -1274,6 +1290,7 @@ class LUVerifyDisks(NoHooksLU):
       if isinstance(lvs, basestring):
         logging.warning("Error enumerating LVs on node %s: %s", node, lvs)
         res_nlvm[node] = lvs
+        continue
       elif not isinstance(lvs, dict):
         logging.warning("Connection to node %s failed or invalid data"
                         " returned", node)
@@ -1665,9 +1682,11 @@ class LUDiagnoseOS(NoHooksLU):
                        selected=self.op.output_fields)
 
     # Lock all nodes, in shared mode
+    # Temporary removal of locks, should be reverted later
+    # TODO: reintroduce locks when they are lighter-weight
     self.needed_locks = {}
-    self.share_locks[locking.LEVEL_NODE] = 1
-    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+    #self.share_locks[locking.LEVEL_NODE] = 1
+    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -1691,6 +1710,11 @@ class LUDiagnoseOS(NoHooksLU):
 
     """
     all_os = {}
+    # we build here the list of nodes that didn't fail the RPC (at RPC
+    # level), so that nodes with a non-responding node daemon don't
+    # make all OSes invalid
+    good_nodes = [node_name for node_name in rlist
+                  if not rlist[node_name].failed]
     for node_name, nr in rlist.iteritems():
       if nr.failed or not nr.data:
         continue
@@ -1699,7 +1723,7 @@ class LUDiagnoseOS(NoHooksLU):
           # build a list of nodes for this os containing empty lists
           # for each node in node_list
           all_os[os_obj.name] = {}
-          for nname in node_list:
+          for nname in good_nodes:
             all_os[os_obj.name][nname] = []
         all_os[os_obj.name][node_name].append(os_obj)
     return all_os
@@ -1708,9 +1732,7 @@ class LUDiagnoseOS(NoHooksLU):
     """Compute the list of OSes.
 
     """
-    node_list = self.acquired_locks[locking.LEVEL_NODE]
-    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()
-                   if node in node_list]
+    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
     node_data = self.rpc.call_os_diagnose(valid_nodes)
     if node_data == False:
       raise errors.OpExecError("Can't gather the list of OSes")
@@ -2327,7 +2349,7 @@ class LUSetNodeParams(LogicalUnit):
         ((node.offline and not self.op.offline == False) or
          (node.drained and not self.op.drained == False))):
       raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
-                                 " to master_candidate")
+                                 " to master_candidate" % node.name)
 
     return