JBoss 4.2.2 nodes start to cluster then suspect each other

Question

I have a website running with JBoss 4.2.2 on an existing Red Hat server. I'm setting up a second server so as to have a clustered pair (which will then be load-balanced). However, I can't get them to cluster successfully.

The existing server starts up JBoss with:

run.sh -c default -b 0.0.0.0

(I know the 'default' configuration doesn't support clustering out of the box - I'm using a modified version of it which includes clustering support.) When I start the second JBoss instance with the same command, it forms its own cluster without noticing the first. Both use the same partition name and multicast address and port.

I tried the McastReceiverTest and McastSenderTest programs to check that the machines could communicate over multicast; they could.

I then noticed the info at http://docs.jboss.org/jbossas/docs/Clustering_Guide/beta422/html/ch07s07s07.html, saying that JGroups cannot bind to all interfaces, and instead binds to the default interface; so presumably it was binding to 127.0.0.1, and thereby not getting the messages through. So instead I set the instances to tell JGroups to use the internal IPs:

run.sh -c default -b 0.0.0.0 -Djgroups.bind_addr=10.51.1.131
run.sh -c default -b 0.0.0.0 -Djgroups.bind_addr=10.51.1.141

(.131 is the existing server, .141 is the new server).

The nodes now notice each other and form a cluster - at first. However, while trying to deploy the .ear, the server log says this:

2010-08-07 22:26:39,321 DEBUG [org.jgroups.protocols.FD] sending are-you-alive msg to 10.51.1.131:46294 (own address=10.51.1.141:47629)
2010-08-07 22:26:45,412 WARN  [org.jgroups.protocols.FD] I was suspected by 10.51.1.131:48733; ignoring the SUSPECT message and sending back a HEARTBEAT_ACK
2010-08-07 22:26:49,324 DEBUG [org.jgroups.protocols.FD] sending are-you-alive msg to 10.51.1.131:46294 (own address=10.51.1.141:47629)
2010-08-07 22:26:49,324 DEBUG [org.jgroups.protocols.FD] heartbeat missing from 10.51.1.131:46294 (number=0)
2010-08-07 22:26:49,529 DEBUG [org.jgroups.protocols.MERGE2] initial_mbrs=[[own_addr=10.51.1.141:60365, coord_addr=10.51.1.141:60365, is_server=true]]
2010-08-07 22:26:52,092 WARN  [org.jboss.cache.TreeCache] replication failure with method_call optimisticPrepare; id:18; Args: ( arg[0] = GlobalTransaction:<10.51.1.131:46294>:5421085 ...) exception org.jboss.cache.lock.TimeoutException: failure acquiring lock: fqn=/Yudu_ear,Yudu-ejb_jar,Yudu-ejbPU/com/yudu/ejb/entity, caller=GlobalTransaction:<10.51.1.131:46294>:5421085, lock=read owners=[GlobalTransaction:<10.51.1.131:46294>:5421081] (activeReaders=1, activeWriter=null, waitingReaders=0, waitingWriters=1, waitingUpgrader=0)

...and the .ear fails to deploy.

If I change CacheMode in ejb3-entity-cache-service.xml from REPL_SYNC to LOCAL, the .ear deploys correctly, although of course the entity cache replication then doesn't happen. However, the log still shows interesting signs of the same problem.

It looks like:

• first the new node finds the existing one and forms a cluster
• then the FD checks fail, and after a set number of failures the new node splits off from the cluster and forms its own cluster of one
• then it finds it again, re-clusters and this time the FD checks work.

Relevant bits of the log file:

2010-08-07 23:47:07,423 INFO  [org.jgroups.protocols.UDP] socket information: local_addr=10.51.1.141:35666, mcast_addr=228.1.2.3:45566, bind_addr=/10.51.1.141, ttl=2 sock: bound to 10.51.1.141:35666, receive buffer size=131071, send buffer size=131071 mcast_recv_sock: bound to 0.0.0.0:45566, send buffer size=131071, receive buffer size=131071 mcast_send_sock: bound to 10.51.1.141:59196, send buffer size=131071, receive buffer size=131071
2010-08-07 23:47:07,431 DEBUG [org.jgroups.protocols.UDP] created unicast receiver thread
2010-08-07 23:47:09,445 DEBUG [org.jgroups.protocols.pbcast.GMS] initial_mbrs are [[own_addr=10.51.1.131:48888, coord_addr=10.51.1.131:48888, is_server=true]]
2010-08-07 23:47:09,446 DEBUG [org.jgroups.protocols.pbcast.GMS] election results: {10.51.1.131:48888=1}
2010-08-07 23:47:09,446 DEBUG [org.jgroups.protocols.pbcast.GMS] sending handleJoin(10.51.1.141:35666) to 10.51.1.131:48888
2010-08-07 23:47:09,751 DEBUG [org.jgroups.protocols.pbcast.GMS] [10.51.1.141:35666]: JoinRsp=[10.51.1.131:48888|61] [10.51.1.131:48888, 10.51.1.141:35666] [size=2]
2010-08-07 23:47:09,752 DEBUG [org.jgroups.protocols.pbcast.GMS] new_view=[10.51.1.131:48888|61] [10.51.1.131:48888, 10.51.1.141:35666]
...
2010-08-07 23:47:10,047 INFO  [org.jboss.ha.framework.interfaces.HAPartition.DefaultPartition] Number of cluster members: 2
2010-08-07 23:47:10,047 INFO  [org.jboss.ha.framework.interfaces.HAPartition.DefaultPartition] Other members: 1
...
2010-08-07 23:47:20,034 DEBUG [org.jgroups.protocols.FD] sending are-you-alive msg to 10.51.1.131:48888 (own address=10.51.1.141:35666)
2010-08-07 23:47:30,037 DEBUG [org.jgroups.protocols.FD] sending are-you-alive msg to 10.51.1.131:48888 (own address=10.51.1.141:35666)
2010-08-07 23:47:30,038 DEBUG [org.jgroups.protocols.FD] heartbeat missing from 10.51.1.131:48888 (number=0)
2010-08-07 23:47:40,040 DEBUG [org.jgroups.protocols.FD] sending are-you-alive msg to 10.51.1.131:48888 (own address=10.51.1.141:35666)
2010-08-07 23:47:40,040 DEBUG [org.jgroups.protocols.FD] heartbeat missing from 10.51.1.131:48888 (number=1)
...
2010-08-07 23:48:19,758 WARN  [org.jgroups.protocols.FD] I was suspected by 10.51.1.131:48888; ignoring the SUSPECT message and sending back a HEARTBEAT_ACK
2010-08-07 23:48:20,054 DEBUG [org.jgroups.protocols.FD] sending are-you-alive msg to 10.51.1.131:48888 (own address=10.51.1.141:35666)
2010-08-07 23:48:20,055 DEBUG [org.jgroups.protocols.FD] [10.51.1.141:35666]: received no heartbeat ack from 10.51.1.131:48888 for 6 times (60000 milliseconds), suspecting it
2010-08-07 23:48:20,058 DEBUG [org.jgroups.protocols.FD] broadcasting SUSPECT message [suspected_mbrs=[10.51.1.131:48888]] to group
...
2010-08-07 23:48:21,691 DEBUG [org.jgroups.protocols.pbcast.NAKACK] removing 10.51.1.131:48888 from received_msgs (not member anymore)
2010-08-07 23:48:21,691 INFO  [org.jboss.ha.framework.server.DistributedReplicantManagerImpl.DefaultPartition] I am (127.0.0.1:1099) received membershipChanged event:
2010-08-07 23:48:21,691 INFO  [org.jboss.ha.framework.server.DistributedReplicantManagerImpl.DefaultPartition] Dead members: 0 ([])
2010-08-07 23:48:21,691 INFO  [org.jboss.ha.framework.server.DistributedReplicantManagerImpl.DefaultPartition] New Members : 0 ([])
2010-08-07 23:48:21,691 INFO  [org.jboss.ha.framework.server.DistributedReplicantManagerImpl.DefaultPartition] All Members : 1 ([127.0.0.1:1099])
...
2010-08-07 23:49:59,793 WARN  [org.jgroups.protocols.FD] I was suspected by 10.51.1.131:48888; ignoring the SUSPECT message and sending back a HEARTBEAT_ACK
2010-08-07 23:50:09,796 WARN  [org.jgroups.protocols.FD] I was suspected by 10.51.1.131:48888; ignoring the SUSPECT message and sending back a HEARTBEAT_ACK
2010-08-07 23:50:19,144 DEBUG [org.jgroups.protocols.FD] Recevied Ack. is invalid (was from: 10.51.1.131:48888),
2010-08-07 23:50:19,144 DEBUG [org.jgroups.protocols.FD] Recevied Ack. is invalid (was from: 10.51.1.131:48888),
...
2010-08-07 23:50:21,791 DEBUG [org.jgroups.protocols.pbcast.GMS] new=[10.51.1.131:48902], suspected=[], leaving=[], new view: [10.51.1.141:35666|63] [10.51.1.141:35666, 10.51.1.131:48902]
...
2010-08-07 23:50:21,792 DEBUG [org.jgroups.protocols.pbcast.GMS] view=[10.51.1.141:35666|63] [10.51.1.141:35666, 10.51.1.131:48902]
2010-08-07 23:50:21,792 DEBUG [org.jgroups.protocols.pbcast.GMS] [local_addr=10.51.1.141:35666] view is [10.51.1.141:35666|63] [10.51.1.141:35666, 10.51.1.131:48902]
2010-08-07 23:50:21,822 INFO  [org.jboss.ha.framework.interfaces.HAPartition.lifecycle.DefaultPartition] New cluster view for partition DefaultPartition (id: 63, delta: 1) : [127.0.0.1:1099, 127.0.0.1:1099]
2010-08-07 23:50:21,822 DEBUG [org.jboss.ha.framework.interfaces.HAPartition.DefaultPartition] membership changed from 1 to 2
...
2010-08-07 23:50:31,825 DEBUG [org.jgroups.protocols.FD] sending are-you-alive msg to 10.51.1.131:48902 (own address=10.51.1.141:35666)
2010-08-07 23:50:31,832 DEBUG [org.jgroups.protocols.FD] received ack from 10.51.1.131:48902

But I'm at a loss to understand why the FD checks fail the first time round; and although it eventually seems to cluster with the other node, the initial failure seems to be enough to mess up the deployment when it tries to share entity state, and thereby prevent it from actually working in a useful way.

If anyone can shed light on this I'll be hugely grateful!

Bylaw

That's a perplexing issue, to be sure. I assume there's a reason you're using JBoss 4.2.2 & a customized server configuration, but can you recreate it with JBoss 4.2.3 (which included some jgroups changes) and/or the "all" configuration? – Aras

The customized configuration is intended to be a compromise between 'default' and 'all' - i.e. 'all' without the bits we're not using. It'll take some work to be able to try alternative configurations (because I can't freely mess with the existing node, so will need to add a third one), but I'll see if I can try those - thanks for the suggestion. – Bylaw

I strongly suggest cutting down the all configuration, rather than adding stuff to the default configuration. It's too easy to miss some critical component that doesn't look useful. – Tuchun

Answer 1

I think that before you move on to JBoss 4.2.3 (which is probably a good place to be eventually) or building a new configuration (I agree with @skaffman about pruning being easier than adding), you might want to try the following:

On 10.51.1.131:

run.sh -c default -b 10.51.1.131 -Djgroups.bind_addr=10.51.1.131

On 10.51.1.141:

run.sh -c default -b 10.51.1.141 -Djgroups.bind_addr=10.51.1.141

According to all the documentation I can find on this, the -b parameter is the server instance bind address, and having them be different might be creating some significant schizophrenia for JGroups. I had a four-server clustered environment working successfully for over three years, and that was part of the recommended configuration from RH/JBoss (we had a support contract, and got help from Bela Ban).

Isom