Uploaded image for project: 'mod_cluster'
  1. mod_cluster
  2. MODCLUSTER-408

Requests are not properly drained during undeploy

XMLWordPrintable

    • Hide

      1) Setup httpd/mod_cluster and JBoss with a <distributable/> application (undistributable apps spend the stop-context time in session draining by default so use distributable)
      2) Establish a session with a request so that subsequent requests will go through mod_cluster's sticky session logic
      3) Send a long running request through to JBoss and shutdown JBoss before it completes
      4) Note mod_cluster logging indicates request draining finishes immediately

      Show
      1) Setup httpd/mod_cluster and JBoss with a <distributable/> application (undistributable apps spend the stop-context time in session draining by default so use distributable) 2) Establish a session with a request so that subsequent requests will go through mod_cluster's sticky session logic 3) Send a long running request through to JBoss and shutdown JBoss before it completes 4) Note mod_cluster logging indicates request draining finishes immediately

      mod_cluster doesn't properly drain requests during undeploy and/or shutdown before it sends REMOVE-APP. org.jboss.modcluster TRACE logging shows JBoss receives a STOP-APP response indicating 0 requests and so the JBoss side thinks requests are drained.

      It looks like mod_cluster only sets "BALANCER_CONTEXT_ID" when a worker is selected by "internal_find_best_byrequests". So it doesn't count stickied requests and allows shutdown to proceed even if stickied requests are in progress.

              rhn-engineering-jclere Jean-Frederic Clere
              rhn-support-aogburn Aaron Ogburn
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

                Created:
                Updated:
                Resolved: