Improve error message for replace-disks

[ganeti-github.git] / lib / cmdlib / instance_storage.py

diff --git a/lib/cmdlib/instance_storage.py b/lib/cmdlib/instance_storage.py
index 871007c..b1fea8b 100644 (file)
--- a/lib/cmdlib/instance_storage.py
+++ b/lib/cmdlib/instance_storage.py
@@ -1049,7 +1049,8 @@ def ExpandCheckDisks(instance, disks):
   else:
     if not set(disks).issubset(instance.disks):
       raise errors.ProgrammerError("Can only act on disks belonging to the"
-                                   " target instance")
+                                   " target instance: expected a subset of %r,"
+                                   " got %r" % (instance.disks, disks))
     return disks
 
 
@@ -1138,6 +1139,7 @@ def ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
   ignored.
 
   """
+  lu.cfg.MarkInstanceDisksInactive(instance.name)
   all_result = True
   disks = ExpandCheckDisks(instance, disks)
 
@@ -1204,6 +1206,10 @@ def AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
   # into any other network-connected state (Connected, SyncTarget,
   # SyncSource, etc.)
 
+  # mark instance disks as active before doing actual work, so watcher does
+  # not try to shut them down erroneously
+  lu.cfg.MarkInstanceDisksActive(iname)
+
   # 1st pass, assemble on all nodes in secondary mode
   for idx, inst_disk in enumerate(disks):
     for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
@@ -1255,6 +1261,9 @@ def AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
   for disk in disks:
     lu.cfg.SetDiskID(disk, instance.primary_node)
 
+  if not disks_ok:
+    lu.cfg.MarkInstanceDisksInactive(iname)
+
   return disks_ok, device_info
 
 
@@ -1403,6 +1412,7 @@ class LUInstanceGrowDisk(LogicalUnit):
 
     if wipe_disks:
       # Get disk size from primary node for wiping
+      self.cfg.SetDiskID(disk, instance.primary_node)
       result = self.rpc.call_blockdev_getsize(instance.primary_node, [disk])
       result.Raise("Failed to retrieve disk size from node '%s'" %
                    instance.primary_node)
@@ -1459,9 +1469,9 @@ class LUInstanceGrowDisk(LogicalUnit):
       if disk_abort:
         self.LogWarning("Disk syncing has not returned a good status; check"
                         " the instance")
-      if instance.admin_state != constants.ADMINST_UP:
+      if not instance.disks_active:
         _SafeShutdownInstanceDisks(self, instance, disks=[disk])
-    elif instance.admin_state != constants.ADMINST_UP:
+    elif not instance.disks_active:
       self.LogWarning("Not shutting down the disk even if the instance is"
                       " not supposed to be running because no wait for"
                       " sync mode was requested")
@@ -1649,6 +1659,7 @@ class LUInstanceActivateDisks(NoHooksLU):
 
     if self.op.wait_for_sync:
       if not WaitForSync(self, self.instance):
+        self.cfg.MarkInstanceDisksInactive(self.instance.name)
         raise errors.OpExecError("Some disks of the instance are degraded!")
 
     return disks_info
@@ -2034,7 +2045,7 @@ class TLReplaceDisks(Tasklet):
     feedback_fn("Current seconary node: %s" %
                 utils.CommaJoin(self.instance.secondary_nodes))
 
-    activate_disks = (self.instance.admin_state != constants.ADMINST_UP)
+    activate_disks = not self.instance.disks_active
 
     # Activate the instance disks if we're replacing them on a down instance
     if activate_disks:
@@ -2100,8 +2111,14 @@ class TLReplaceDisks(Tasklet):
         if msg or not result.payload:
           if not msg:
             msg = "disk not found"
-          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
-                                   (idx, node, msg))
+          if not self._CheckDisksActivated(self.instance):
+            extra_hint = ("\nDisks seem to be not properly activated. Try"
+                          " running activate-disks on the instance before"
+                          " using replace-disks.")
+          else:
+            extra_hint = ""
+          raise errors.OpExecError("Can't find disk/%d on node %s: %s%s" %
+                                   (idx, node, msg, extra_hint))
 
   def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
     for idx, dev in enumerate(self.instance.disks):
@@ -2156,9 +2173,12 @@ class TLReplaceDisks(Tasklet):
 
       # we pass force_create=True to force the LVM creation
       for new_lv in new_lvs:
-        _CreateBlockDevInner(self.lu, node_name, self.instance, new_lv, True,
-                             GetInstanceInfoText(self.instance), False,
-                             excl_stor)
+        try:
+          _CreateBlockDevInner(self.lu, node_name, self.instance, new_lv, True,
+                               GetInstanceInfoText(self.instance), False,
+                               excl_stor)
+        except errors.DeviceCreationError, e:
+          raise errors.OpExecError("Can't create block device: %s" % e.message)
 
     return iv_names
 
@@ -2373,9 +2393,12 @@ class TLReplaceDisks(Tasklet):
                       (self.new_node, idx))
       # we pass force_create=True to force LVM creation
       for new_lv in dev.children:
-        _CreateBlockDevInner(self.lu, self.new_node, self.instance, new_lv,
-                             True, GetInstanceInfoText(self.instance), False,
-                             excl_stor)
+        try:
+          _CreateBlockDevInner(self.lu, self.new_node, self.instance, new_lv,
+                               True, GetInstanceInfoText(self.instance), False,
+                               excl_stor)
+        except errors.DeviceCreationError, e:
+          raise errors.OpExecError("Can't create block device: %s" % e.message)
 
     # Step 4: dbrd minors and drbd setups changes
     # after this, we must manually remove the drbd minors on both the