0 Replies Latest reply on Sep 20, 2004 10:24 AM by posselt

    JNDI lookup extremely slow when a clusternode is down

    posselt Newbie

      Hi,

      We have a cluster of 3 nodes to provide high availability to our clients.
      They use "jboss1:1100,jboss2:1100,jboss3:1100" as jndi url to
      connect to the server.

      Everything works fine as long as on all nodes at least the OS is running
      and at least one JBoss is up. However, if one node is completely down
      (turned off), the JNDI lookup takes several minutes.

      I checked org.jnp.interfaces.NamingContext and for the lookup a socket
      is opened to all nodes in the url. Now, if a node is down there is a long
      timeout which causes the whole lookup to be blocked. Hence, instead of
      high availability we have multiple points of failure (if any one of the three
      nodes crashes we have a downtime).

      Any workaround? We think about using a load balancer or patch the
      NamingContext class.

      PS: debian linux, JBoss-3.2.3RC1. From looking at the code it seems to
      be the same in JBoss-4.0.0RC1.

      Cheers,
      Dietmar