where can I config the timeout of jboss clustering, very urgent, waiting online

[bondchan921]

we have 4 nodes clustering, while one of them have long GC pause, which cause this node(node52) drop of clustering, and having the following log

2012-06-14 03:27:48,277 INFO [org.jboss.messaging.core.impl.postoffice.GroupMember] org.jboss.messaging.core.impl.postoffice.GroupMember$ControlMembershipListener@6225352b got new view [10.164.218.18:7910|10] [10.164.218.18:7910, 10.164.107.69:7910, 10.164.107.65:7910], old view is [10.164.218.14:7910|9] [10.164.218.14:7910, 10.164.218.18:7910, 10.164.107.69:7910, 10.164.107.65:7910] 2012-06-14 03:27:48,277 INFO [org.jboss.messaging.core.impl.postoffice.GroupMember] I am (10.164.218.18:7910) 2012-06-14 03:27:48,998 INFO [org.jboss.messaging.core.impl.postoffice.MessagingPostOffice] JBoss Messaging is failing over for failed node 52. If there are many messages to reload this may take some time...

can anyone tell me where can I config the timeout of the clustering node drop? It seem 2 minutes in my case, but I can't find in each configuration files...

4.3.0.GA_CP07

[jaikiran]

Moved to JBoss Messaging forum.

[bondchan921]

Anyone can do me a favor?

[gaohoward]

guess you need to change jgroups config for JBM. What's your current config look like?

[bondchan921]

Hi Yong Hao,

Thanks for your reply,

1), The Jgroup config is reside in the cluster-service.xml ? I have zipped in the attachment, and I don't figure why the timeout is 2 minutes.

2), There are several places (postoffice control channel/ postoffice data channel, Jgroup, others? ), which one will be the checkpoint of timeout to drop node from clustering ?

Thanks again!

[gaohoward]

looks like it in oracle-persistence-service.xml. You need to adjust the configurations in ControlChannelConfig section. I think it is the 'timeout' of 'FD' protocol.

Howard

[bondchan921]

Thanks,

found it, and it's 3 minutes in our cases.

2 another questions

Q1: As I found this FD timeout have been configure to 'Jgroup', 'Postoffice's DataChannel, Postoffice's ControlChannel',each of the timeout will cause node drop? and should be the shortest timeout will be the trigger in the majority of the cases?

--Cache timeout

Q2.1: Jboss TreeCache have been configure FD timeout too, is this the real timeout or is the value of parameter SyncReplTimeout ? I think is the former one according to your last reply

Q2.2: in the following cache log clip, it seems node lost in the JGroup, cause the node lost in the CacheView, my question is the CacheView  maintainance combined by the answer of Q2.1 and the JGroupView ?

(1099 is JNDI port, 7930 is replication cache port, 7920 is invalidation cache port, 7910 is the ControlChannel port, 7900 is the DataChannel port 7800 is JGroup port)

2012-06-14 03:28:01,049 New cluster view for partition PRODPartition_HA: 12 ([10.164.107.69:1099, 10.164.218.18:1099, 10.164.107.65:1099] delta: -1)

2012-06-14 03:28:01,091 viewAccepted(): [10.164.107.69:7930|12] [10.164.107.69:7930, 10.164.218.18:7930, 10.164.107.65:7930]

2012-06-14 03:28:01,057 I am (10.164.107.65:1099) received membershipChanged event:

2012-06-14 03:28:02,047 Dead members: 1 ([10.164.218.14:1099])

2012-06-14 03:28:02,161 New Members : 0 ([])

2012-06-14 03:28:02,161 All Members : 3 ([10.164.107.69:1099, 10.164.218.18:1099, 10.164.107.65:1099])

2012-06-14 03:28:02,161 Suspected member: 10.164.218.14:7800

2012-06-14 03:28:02,293 viewAccepted(): [10.164.107.69:7920|12] [10.164.107.69:7920, 10.164.218.18:7920, 10.164.107.65:7920

Bond

[gaohoward]

Q1: only "ControlChannel" matters. The data channel is used for some internal cluster message exchange, it is not used for membership management. However you may also need to adjust this channel because if this channel timeout the communication between members of cluster will be broken.

Q2 1&2: I don't know TreeCache configuration. Anyway it has nothing to do with JBM, so its config won't affect JBM.

[bondchan921]

Thanks Howard, really helps me !