X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/5a3103e9ef4564d8e8cd9f9d60aac799ccf3c08f..0d0c1b4a16ae3edaebe231b67e657d148c8c3c8b:/daemons/ganeti-watcher

diff --git a/daemons/ganeti-watcher b/daemons/ganeti-watcher
index c912592..8b2b183 100755
--- a/daemons/ganeti-watcher
+++ b/daemons/ganeti-watcher
@@ -47,6 +47,9 @@ BAD_STATES = ['stopped']
 HELPLESS_STATES = ['(node down)']
 NOTICE = 'NOTICE'
 ERROR = 'ERROR'
+KEY_RESTART_COUNT = "restart_count"
+KEY_RESTART_WHEN = "restart_when"
+KEY_BOOT_ID = "bootid"
 
 
 class Error(Exception):
@@ -119,8 +122,8 @@ class WatcherState(object):
     except Exception, msg:
       # Ignore errors while loading the file and treat it as empty
       self.data = {}
-      sys.stderr.write("Empty or invalid state file. "
-          "Using defaults. Error message: %s\n" % msg)
+      sys.stderr.write("Empty or invalid state file."
+                       " Using defaults. Error message: %s\n" % msg)
 
     if "instance" not in self.data:
       self.data["instance"] = {}
@@ -151,8 +154,8 @@ class WatcherState(object):
     """
     ndata = self.data["node"]
 
-    if name in ndata and "bootid" in ndata[name]:
-      return ndata[name]["bootid"]
+    if name in ndata and KEY_BOOT_ID in ndata[name]:
+      return ndata[name][KEY_BOOT_ID]
     return None
 
   def SetNodeBootID(self, name, bootid):
@@ -166,7 +169,7 @@ class WatcherState(object):
     if name not in ndata:
       ndata[name] = {}
 
-    ndata[name]["bootid"] = bootid
+    ndata[name][KEY_BOOT_ID] = bootid
 
   def NumberOfRestartAttempts(self, instance):
     """Returns number of previous restart attempts.
@@ -178,7 +181,7 @@ class WatcherState(object):
     idata = self.data["instance"]
 
     if instance.name in idata:
-      return idata[instance.name]["restart_count"]
+      return idata[instance.name][KEY_RESTART_COUNT]
 
     return 0
 
@@ -196,8 +199,8 @@ class WatcherState(object):
     else:
       inst = idata[instance.name]
 
-    inst["restart_when"] = time.time()
-    inst["restart_count"] = idata.get("restart_count", 0) + 1
+    inst[KEY_RESTART_WHEN] = time.time()
+    inst[KEY_RESTART_COUNT] = inst.get(KEY_RESTART_COUNT, 0) + 1
 
   def RemoveInstance(self, instance):
     """Update state to reflect that a machine is running, i.e. remove record.
@@ -346,6 +349,7 @@ class Watcher(object):
     notepad = WatcherState()
     self.CheckInstances(notepad)
     self.CheckDisks(notepad)
+    self.VerifyDisks()
     notepad.Save()
 
   def CheckDisks(self, notepad):
@@ -364,9 +368,8 @@ class Watcher(object):
       # secondary node.
       for instance in GetInstanceList(with_secondaries=check_nodes):
         try:
-          self.messages.append(Message(NOTICE,
-                                       "Activating disks for %s." %
-                                       instance.name))
+          self.messages.append(Message(NOTICE, ("Activating disks for %s." %
+                                                instance.name)))
           instance.ActivateDisks()
         except Error, x:
           self.messages.append(Message(ERROR, str(x)))
@@ -399,9 +402,8 @@ class Watcher(object):
                                        (instance.name, MAXTRIES)))
           continue
         try:
-          self.messages.append(Message(NOTICE,
-                                       "Restarting %s%s." %
-                                       (instance.name, last)))
+          self.messages.append(Message(NOTICE, ("Restarting %s%s." %
+                                                (instance.name, last))))
           instance.Restart()
         except Error, x:
           self.messages.append(Message(ERROR, str(x)))
@@ -413,10 +415,17 @@ class Watcher(object):
       else:
         if notepad.NumberOfRestartAttempts(instance):
           notepad.RemoveInstance(instance)
-          msg = Message(NOTICE,
-                        "Restart of %s succeeded." % instance.name)
+          msg = Message(NOTICE, "Restart of %s succeeded." % instance.name)
           self.messages.append(msg)
 
+  def VerifyDisks(self):
+    """Run gnt-cluster verify-disks.
+
+    """
+    result = DoCmd(['gnt-cluster', 'verify-disks', '--lock-retries=15'])
+    if result.output:
+      self.messages.append(Message(NOTICE, result.output))
+
   def WriteReport(self, logfile):
     """Log all messages to file.
 
@@ -457,7 +466,11 @@ def main():
     sys.stderr = sys.stdout = open(constants.LOG_WATCHER, 'a')
 
   try:
-    watcher = Watcher()
+    try:
+      watcher = Watcher()
+    except errors.ConfigurationError:
+      # Just exit if there's no configuration
+      sys.exit(constants.EXIT_SUCCESS)
     watcher.Run()
     watcher.WriteReport(sys.stdout)
   except NotMasterError: