0 Replies Latest reply on Mar 2, 2014 3:37 AM by yoavgur

    Inconsistency between ManagedConnectionPool and CachedConnectionManager

    yoavgur

      Hi all,

       

      We are working witn jboss-4.2.3.GA.

      Our application uses several connection pools.

      For all poos, MinSize and MaxSize are the same (60), and IdleTimeoutMinutes is set to "0".

       

      On the customer site, we encountered the following issue (All data was retrieved via jboss JMX console):

      1. Number of available connections in a specific pool reaches 0 very often.

      2. Accordingly, number of "In Use" connections for that pool is close to the MaxSize.

      3. There is a fairly high number of destroyed connections (>100), but the number of created connections is exactly (number of destroyed connections) + (pool MaxSize), which seems OK.

      4. Number of "In Use" connections according to CachedConnectionManager is low.

      5. listInUseConnections() from CachedConnectionManager lists low number of connection stack traces.

       

      As you can see, 1-2 is consistent, 4-5 is consistent, but 1-2 and 4-5 seems to contradict each other.

       

      We could just set the MaxSize of the pool to an higher number, but we suspect that the inconsistency described above
      might indicate a hidden issue that is causing the number of available connections to be low for reasons other than just real system stress.

       

      Have anyone encouneterd such inconsistency between a specific pool indicators and the CachedConnectionManager indicators?

      Any idea what could be the cause for such inconsistency?

      Can it be related to the high number of destroyed connections?

       

      Note that the issue happened only after several month of the system running OK, and did not reproduce (yet?) after a jboss restart.

      We tried to reproduce the issue in our own lab and failed to do so.

       

      Any help or hint will be appreciated.

       

      Thanks in advance,

      Yoav.