Re: [javagroups-users] NAKACK - range is null, sender not found in xmit_table

[Bela B. <be...@ya...>]

I've never tested JGroups on NLB.

How do you stop the boxes ? If you simply kill the processes, JGroups 
should exclude them from the cluster gracefully, the time taken is 
determined by FD_SOCK/FD/FD_ALL.

Did you set UDP.loopback to true and do you have MERGE2 in your stack ?


Kristof Jozsa wrote:
> We have our application using JGroups set up on 2 win2k3 test-server
> boxes running in VMware and pulled into a Windows Network Load
> Balancer setup. Our daily backup procedure stops both boxes, does the
> backup and puts them back online. Sometime in this process, JGroups
> starts spamming the logs with the following two lines of error:
>
> 2008-09-19 17:11:46,362 ERROR 60152285
> [org.jgroups.protocols.pbcast.NAKACK] - range is null
> 2008-09-19 17:11:46,362 ERROR 60152285
> [org.jgroups.protocols.pbcast.NAKACK] - sender OTHER_HOST_IP:4931 not
> found in xmit_table
>
> where OTHER_HOST_IP always refers to the ip address of the other
> cluster node (while the problem happens on both nodes equally). We get
> thousands of these errors per second and it doesn't appear to stop
> ever (we actually realized the problem first when the servers had ran
> out of disk space. We are using JGroups 2.6.3 (the previous stable
> version), didn't yet have a chance to upgrade to the latest. Any idea
> what can cause this problem?
>
> Slightly related, should JGroups with the stock udp.xml setup work
> with windows nodes being in an NLB setup?
>
>
>

-- 
Bela Ban
Lead JGroups / Clustering Team
JBoss - a division of Red Hat