0

I am probing my server for clusters using jgroups probe.sh script, 2.4.1 and after shutting down Weblogic servers (not AdminServer) I see some zombie - like clusters and I dont know how to get rid of them. They are blocking deployment. They might have stayed in memory because of kill -9 command that I did several times on server.

How to get rid of them?

Slava Vedenin
  • 58,326
  • 13
  • 40
  • 59
michealAtmi
  • 1,012
  • 2
  • 14
  • 36

1 Answers1

0

I assume by shutting down Weblogic server, you mean that they're disabled, but the JVM is still running? Otherwise, the JGroups channels would not exist anymore.

When shutting down gracefully, you need to make sure to do a JChannel.close(), to de-allocate resources used by JGroups.

If you kill -9 your servers, are you sure the nodes you're still seeing via probe.sh are not from different clusters? The cluster names should indicate to which cluster they belong...

Bela Ban
  • 2,186
  • 13
  • 12
  • Hi, I am sure that they are from the same cluster, I see it by name. I know, kill -9 was wrong thing but I think I need a way to kill these zombies somehow - how can I do that in 2.4.1?. Weblogic admin console is working only, all servers are down and probe still detects zombies (I dont know why) In newer Jgroups I see chapter "Removal of zombie files" under FILE_PING topic, is this related with my problem ? Unfortunatelly I have 2.4.1 version which doesnt have FILE_PING. During graceful shutdown everything will be ok, Treecache.stopService and Jchannel.close seems to work – michealAtmi Mar 05 '16 at 21:13
  • When I set new multicast ip address, everything works again - probe detects new clusters with new multicast address and after graceful shutdown clusters dissapear. There are still old zombies with old multicast ip address and I want to remove them ... I must learn how to do that otherwise in future I will have to change multicast address before deployment as a workaround. I use jgroups 2.4.1 so if it possible I ask for solution for this version. – michealAtmi Mar 05 '16 at 21:18
  • I don't understand how kill -9 would render you with a zombie process. FILE_PING has nothing to do with this. Note that JBossCache has been deprecated long ago... – Bela Ban Mar 07 '16 at 07:58
  • Lets forget about kill -9 for a second. Stragest thing is that after restarting computer (localhost) with my application deployed on Jetty, I observe the same thing. Only after restart I do "./probe.sh -timeout 5000 -query jmx -query props:" and this shows me 4 results with old settings of UDP. How is it possible?? – michealAtmi Mar 07 '16 at 10:29
  • I changed probe class so that it logs ip address of the node that responses to probe packet sent by datagram socket (one should see the Probe class source to understand that). Then it turned out that totally different servers in the same network from other project responds. They are forked from the same product and has the same cluster configuration and that was the reason... I thought I am lonely in my network but that's not true. Thanks for support! – michealAtmi Mar 07 '16 at 18:01
  • ah, ok, good to know. Seems you're using quite an old version of JGroups as well! – Bela Ban Mar 08 '16 at 06:35