]
Tim Fox commented on JBMESSAGING-1691:
--------------------------------------
A couple of questions:
1) Why do you have the live and backup on the same node? If that node crashes the backup
will crash too and will be useless
2) Does the problem occur if you run the tests *outside* eclipse? I don't recommend
running test clients in eclipse - if you set breakpoints then that will cause pings not to
be sent and the system will fail over.
JBM2 cluster fails under heavy load
-----------------------------------
Key: JBMESSAGING-1691
URL:
https://jira.jboss.org/jira/browse/JBMESSAGING-1691
Project: JBoss Messaging
Issue Type: Bug
Components: JMS Clustering
Affects Versions: 2.0.0.beta4
Environment: Linux-debian 64bits 8G RAM - Java1.6 update14
Reporter: Abdel Dridi
Assignee: Clebert Suconic
Attachments: build.xml, build.xml, cluster-nodes-config-files.zip,
Jbm2ClusterTest-NodeA.zip, Jbm2ClusterTest-NodeB.zip, jbmTest.JPG, lib.zip
I have a cluster of 2 nodes; A and B:
In nodeA, I have JMB2-NodeA-live and JBM2-NodeB-backup.
In nodeB, I have JMB2-NodeB-live and JBM2-NodeA-backup.
I have a client running in NodeA (eclipseProject "Jbm2ClusterTest-NodeA.zip")
and an other client running in NodeB (eclipseProject
"Jbm2ClusterTest-NodeB.zip") - Both client starts on the same time.
In each project there's a Junit testcase (TestLauncher.testTenantPool) to start the
test.
I also attached the configuration (config/stand-alone/clustered) for each node
(Live/Backup).
BTW; I tried with the trunk build (07-23-2009) and JBM2-Beta3.
I faced the following issues:
Connection failure after producing 600k and consuming 400K messages of the InboundQueue
in each node.
If the failover kick off then after few min both BackUp and Live servers will be
listening to the cluster which leads to "more than one broadcasters on the network
broadcasting the same node id".
I also see "failed to accept connection: too many open files".
Let me know if you have any question,
Abdel
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: