1 Reply Latest reply on Dec 20, 2002 9:48 AM by Sacha Labourey

    Need help

    Rajatish Newbie

      We are trying to get clustering and load-balancing going for a group of two servers. We are using an Alteon load balancer to load balance.
      We got the clustering, failover and load balancing to work. thanks a lot to slaboure.
      We ran tests to handle more than 50 clients per second (suffices for our application) and it worked. We use apache Jmeter for our tests.
      However, here is a scenario where things fail. We start apache Jmeter running 10 threads. Our load balancer (alteon) worked correctly by delegating each request to an available server (5 requests went to server 1 and 5 to server 2). We kill one of the servers in the middle of load test (when it is in the process of serving requests). It does not work, Apache Jmeter shows several failed requests. What happens if you kill JBoss in the middle of a request?
      Here is a scenario where it works if you are not in the middle of serving a request. We hit a server, and issue requests. Then, we stop at a page, and kill the server. Now we send requests from this page and we have to wait about 25-30 seconds before the request is served. But session replication works and all requests go to the active server henceforth.
      So, the real problem is killing a server in the middle of a request? Why?
      Any pointers will help.
      Thanks,
      Raj

        • 1. Re: Need help
          Sacha Labourey Master

          When you say "it fails", what do mean? accuratly? Do you have a stack trace somewhere or a description.

          Otherwise it is hard to say. Which kind of http session replication do you use? (which package? i.e. the jboss one or the jetty one?)

          When you are waiting 25-30 seconds (in the "working" scenario, could you do a thread stack dump (press CTRL+BREAK in the console) and send it in a file? We have to try to remove this long wait.

          Cheers,


          sacha