Weblogic Clustering - No server found

Question

          Hi folks,
          Occasionaly, when we weblogic ping our servers, we get "No server found" errors.
          In the server weblogic.log files, we don't see much in the way of information other
          than a heartbeat being the last entry usually in the log.
          These errors sometimes correspond with an inability to access the website from a
          browser as well, which obviously is a little difficult to explain to customers/partners.
          According to the weblogic console, garbage collection doesn't appear to be the cause
          of this problem. It's happening roughly every 30 minutes and behaves as it should
          (up to 99% down to 10-20%). The collection occurs instantaneously and doesn't appear
          to correspond to any problems in the servers.
          We have about 1200 java classes, 1100 jsps, ~200 ejb classes that are deployed,
          and 200-300 thousand page views per day.
          We have numerous jobs using the weblogic timers, loading up caches and performing
          other tasks. I'd like to blaim them, but there is no spike in CPU usage and, like
          I said, garbage collection is behaving.
          Session replication is not turned on, however clustering is enabled for failover.
          We're deployed on 2 sun servers, Solaris 2.7, Sun 420 / sparc architecture. We're
          using Jdk 1.2.1._04. We've allocated 1 Gigabyte of ram to each VM on each server.
          We've been battling these issues for over two years now, slowly upgrading with the
          patches. We've tweaked every configuration option about a half dozen times.
          The only two tweaks that I've found to be helpful were reducing from 4 VMs on 2 servers
          to 2 VMs on 2 servers. And changing the connection timeout from 5 minutes to 2
          minutes on the web layer. The first tweak makes sense, the second tweak is very
          mysterious.
          I'd like to upgrade further to later patches, however my developers are telling me
          that we need to use JDK 1.3 because our JSPs are too big for JDK 1.2 and there are
          memory leaks in JDK 1.3.
          The reason the JSPs are too big is because the latest service packs add more code
          to the JSPs, thus creating verify errors from too large code blocks all over the
          place.
          Questions:
          - is there a way to turn off multicasting, jndi, ebj clustering but still have failover?
          - anyone deploying JDK 1.3 on Solaris with weblogic? Are you running into memory
          problems?
          - shall we struggle on or just rewrite everything so it works in resin or some other
          servlet engine, dumping weblogic altogether?
          

Answer 1

Tim,
          Can you mention the WLS version & SP level?
          BTW, have you ever considered contacting weblogic support?
          This kind of issues are best solved by BEA support or calling
          BEA consultant to your site. ;-)
          Kumar
          Tim Shephard wrote:
          > Hi folks,
          >
          > Occasionaly, when we weblogic ping our servers, we get "No server found" errors.
          >
          > In the server weblogic.log files, we don't see much in the way of information other
          > than a heartbeat being the last entry usually in the log.
          >
          > These errors sometimes correspond with an inability to access the website from a
          > browser as well, which obviously is a little difficult to explain to customers/partners.
          >
          > According to the weblogic console, garbage collection doesn't appear to be the cause
          > of this problem. It's happening roughly every 30 minutes and behaves as it should
          > (up to 99% down to 10-20%). The collection occurs instantaneously and doesn't appear
          > to correspond to any problems in the servers.
          >
          > We have about 1200 java classes, 1100 jsps, ~200 ejb classes that are deployed,
          > and 200-300 thousand page views per day.
          >
          > We have numerous jobs using the weblogic timers, loading up caches and performing
          > other tasks. I'd like to blaim them, but there is no spike in CPU usage and, like
          > I said, garbage collection is behaving.
          >
          > Session replication is not turned on, however clustering is enabled for failover.
          >
          > We're deployed on 2 sun servers, Solaris 2.7, Sun 420 / sparc architecture. We're
          > using Jdk 1.2.1._04. We've allocated 1 Gigabyte of ram to each VM on each server.
          >
          > We've been battling these issues for over two years now, slowly upgrading with the
          > patches. We've tweaked every configuration option about a half dozen times.
          >
          > The only two tweaks that I've found to be helpful were reducing from 4 VMs on 2 servers
          > to 2 VMs on 2 servers. And changing the connection timeout from 5 minutes to 2
          > minutes on the web layer. The first tweak makes sense, the second tweak is very
          > mysterious.
          >
          > I'd like to upgrade further to later patches, however my developers are telling me
          > that we need to use JDK 1.3 because our JSPs are too big for JDK 1.2 and there are
          > memory leaks in JDK 1.3.
          >
          > The reason the JSPs are too big is because the latest service packs add more code
          > to the JSPs, thus creating verify errors from too large code blocks all over the
          > place.
          >
          > Questions:
          >
          > - is there a way to turn off multicasting, jndi, ebj clustering but still have failover?
          > - anyone deploying JDK 1.3 on Solaris with weblogic? Are you running into memory
          > problems?
          >
          > - shall we struggle on or just rewrite everything so it works in resin or some other
          > servlet engine, dumping weblogic altogether?
          >