5 Replies Latest reply on Jun 29, 2004 10:05 AM by Patrick Jaromin

    Failover Server has Old Replicated Session

    Patrick Jaromin Newbie

      I've just updated from 3.2.3 to 3.2.4 final. Everything seemed to be working just fine until my final test - failover. I'm also using Apache and MOD_JK2 with sticky sessions (and an appropriate jvmRoute).

      I've got a very simple test webapp with a session-persisted counter that increments with each page refresh.

      Here's what happens:
      1 - start both server instances (A and B) - I've used the "all" configuration for testing. As-is out of the box.

      2 - Visit the webapp - Apache appropriately choses one of the JBoss instances. I use a different browser to visit the other and ensure it's initialized on both systems.

      3 - Refresh the test app on the server - through Apache connected to instance "A". I refresh multiple times - with the count around 20-50. I've tried this with multiple browser instances on multiple systems, various counts.

      4 - shutdown instance "A".

      5 - Hit "Refresh" in the browsers. They properly now failover to instance "B" - however, the counts are significantly lower (say 13 instead of 33). They are NOT 1, however - it is apparent that the sessions WERE replicated at some point, but that they were not updated. I've even waited for a few minutes of inactivity before killing server "A" just in case there was some refresh delay. Obviously, failover to an outdated session in many cases will be as bad or worse as not having a session at all.

      Please note that I've used this same webapp and same procedure on the exact same hardware/network to test JBoss 3.2.3 prior to deploying on our production servers - and all worked as expected. I also performed these intial tests with a fresh install of JBoss 3.2.4 without any modifications.

      I've reviewed the new tomcat setup and have played with the UseLocalCache attribute (set to false) but this didn't help.

      Am I missing some new attribute? Did something else change that I need to modify in my tests?

      Please help - I've already delayed my migration from 3.2.1 for two weeks awaiting 3.2.4 final (for a different fix) and really hope not to delay any further.

      Thanks in advance!

      - Patrick