[jboss-cvs] JBossCache/src/org/jboss/cache/buddyreplication ...

[Manik Surtani]

  User: msurtani
  Date: 06/11/10 15:32:52

  Modified:    src/org/jboss/cache/buddyreplication   BuddyManager.java
  Added:       src/org/jboss/cache/buddyreplication  
                        BuddyNotInitException.java
  Log:
  Fixed BR deadlock issue on startup
  
  Revision  Changes    Path
  1.48      +80 -24    JBossCache/src/org/jboss/cache/buddyreplication/BuddyManager.java
  
  (In the diff below, changes in quantity of whitespace are not shown.)
  
  Index: BuddyManager.java
  ===================================================================
  RCS file: /cvsroot/jboss/JBossCache/src/org/jboss/cache/buddyreplication/BuddyManager.java,v
  retrieving revision 1.47
  retrieving revision 1.48
  diff -u -b -r1.47 -r1.48
  --- BuddyManager.java	6 Nov 2006 23:34:08 -0000	1.47
  +++ BuddyManager.java	10 Nov 2006 20:32:52 -0000	1.48
  @@ -97,6 +97,15 @@
      public static final Fqn BUDDY_BACKUP_SUBTREE_FQN = Fqn.fromString(BUDDY_BACKUP_SUBTREE);
   
      /**
  +    * number of times to retry communicating with a selected buddy if the buddy has not been initialised.
  +    */
  +   private static int UNINIT_BUDDIES_RETRIES = 3;
  +   /**
  +    * wait time between retries
  +    */
  +   private static final long UNINIT_BUDDIES_RETRY_NAPTIME = 500;
  +
  +   /**
       * Flag to prevent us receiving and processing remote calls before we've started
       */
      private boolean initialised = false;
  @@ -324,9 +333,10 @@
       * Called by TreeCache._remoteRemoveFromBuddyGroup(String groupName)
       * when a method call for this is received from a remote cache.
       */
  -   public void handleRemoveFromBuddyGroup(String groupName)
  +   public void handleRemoveFromBuddyGroup(String groupName) throws BuddyNotInitException
      {
  -      waitForInit();
  +      if (!initialised) throw new BuddyNotInitException("Not yet initialised");
  +
         if (log.isInfoEnabled()) log.info("Removing self from buddy group " + groupName);
         buddyGroupsIParticipateIn.remove(groupName);
   
  @@ -352,7 +362,8 @@
       */
      public void handleAssignToBuddyGroup(BuddyGroup newGroup, Map state) throws Exception
      {
  -      waitForInit();
  +      if (!initialised) throw new BuddyNotInitException("Not yet initialised");
  +
         if (log.isInfoEnabled()) log.info("Assigning self to buddy group " + newGroup);
         buddyGroupsIParticipateIn.put(newGroup.getGroupName(), newGroup);
   
  @@ -452,7 +463,7 @@
   
      // -------------- internal helpers methods --------------------
   
  -   private void removeFromGroup(List buddies)
  +   private void removeFromGroup(List buddies) throws InterruptedException
      {
         if (log.isDebugEnabled())
            log.debug("Removing obsolete buddies from buddy group [" + buddyGroup.getGroupName() + "].  Obsolete buddies are " + buddies);
  @@ -461,13 +472,34 @@
         MethodCall membershipCall = MethodCallFactory.create(MethodDeclarations.remoteRemoveFromBuddyGroupMethod, buddyGroup.getGroupName());
         MethodCall replicateCall = MethodCallFactory.create(MethodDeclarations.replicateMethod, membershipCall);
   
  +      int attemptsLeft = UNINIT_BUDDIES_RETRIES;
  +
  +      while (attemptsLeft-- > 0)
  +      {
         try
         {
            makeRemoteCall(buddies, replicateCall);
  +            break;
         }
         catch (Exception e)
         {
  -         log.error("Problems broadcasting removeFromGroup to specific buddies", e);
  +            if (e instanceof BuddyNotInitException || e.getCause() instanceof BuddyNotInitException)
  +            {
  +               if (attemptsLeft > 0)
  +               {
  +                  log.info("One of the buddies have not been initialised.  Will retry after a short nap.");
  +                  Thread.sleep(UNINIT_BUDDIES_RETRY_NAPTIME);
  +               }
  +               else
  +               {
  +                  throw new BuddyNotInitException("Unable to contact buddy after " + UNINIT_BUDDIES_RETRIES + " retries");
  +               }
  +            }
  +            else
  +            {
  +               log.error("Unable to communicate with Buddy for some reason", e);
  +            }
  +         }
         }
         log.trace("removeFromGroup notification complete");
      }
  @@ -528,14 +560,36 @@
         MethodCall membershipCall = MethodCallFactory.create(MethodDeclarations.remoteAssignToBuddyGroupMethod, buddyGroup, stateMap);
         MethodCall replicateCall = MethodCallFactory.create(MethodDeclarations.replicateMethod, membershipCall);
   
  +      int attemptsLeft = UNINIT_BUDDIES_RETRIES;
  +
  +      while (attemptsLeft-- > 0)
  +      {
         try
         {
            makeRemoteCall(buddies, replicateCall);
  +            break;
         }
         catch (Exception e)
         {
  -         log.error("Problems broadcasting assignToGroup to specific buddies", e);
  +            if (e instanceof BuddyNotInitException || e.getCause() instanceof BuddyNotInitException)
  +            {
  +               if (attemptsLeft > 0)
  +               {
  +                  log.info("One of the buddies have not been initialised.  Will retry after a short nap.");
  +                  Thread.sleep(UNINIT_BUDDIES_RETRY_NAPTIME);
         }
  +               else
  +               {
  +                  throw new BuddyNotInitException("Unable to contact buddy after " + UNINIT_BUDDIES_RETRIES + " retries");
  +               }
  +            }
  +            else
  +            {
  +               log.error("Unable to communicate with Buddy for some reason", e);
  +            }
  +         }
  +      }
  +
         log.trace("addToGroup notification complete");
      }
   
  @@ -740,6 +794,8 @@
   
         public void run()
         {
  +         // don't start this thread until the Buddy Manager has initialised as it cocks things up.
  +         waitForInit();
            while (!Thread.interrupted())
            {
               try
  
  
  
  1.2       +27 -0     JBossCache/src/org/jboss/cache/buddyreplication/BuddyNotInitException.java
  
  (In the diff below, changes in quantity of whitespace are not shown.)
  
  Index: BuddyNotInitException.java
  ===================================================================
  RCS file: BuddyNotInitException.java
  diff -N BuddyNotInitException.java
  --- /dev/null	1 Jan 1970 00:00:00 -0000
  +++ BuddyNotInitException.java	10 Nov 2006 20:32:52 -0000	1.2
  @@ -0,0 +1,27 @@
  +package org.jboss.cache.buddyreplication;
  +
  +import org.jboss.cache.CacheException;
  +
  +/**
  + * Exception to depict that a buddy has not been initialised to participate in any comms
  + *
  + * @author <a href="mailto:manik at jboss.org">Manik Surtani</a>
  + * @since 1.4.1
  + */
  +public class BuddyNotInitException extends CacheException
  +{
  +
  +   public BuddyNotInitException()
  +   {
  +   }
  +
  +   public BuddyNotInitException(String msg)
  +   {
  +      super(msg);
  +   }
  +
  +   public BuddyNotInitException(String msg, Throwable cause)
  +   {
  +      super(msg, cause);
  +   }
  +}