Failback continues to work for me -- But I am currently running on one machine with two instances of JBoss so I suspect that the EJB calls take the 'short cut' to the same server rather than marshalling out. I wonder if this makes it look like it's working when it isn't.
When I bring one server down, the other one picks up those sessions, and when I bring the server back up, the sessions get directed to that server and it knows about them -- whether it's an instance of the browser that I did refresh while the server was down or whether its an instance that I did not refresh.
Again, I need to explicitly install on two machines to better test. Also, I'm not explicitly changing anything in the session after the first request. I need to test more dynamic sessions.
It's nice to know the session does not get replicated unless something is changed on it -- for efficiency reasons.
Found out the circumstances when round robin stops working and it's a redirector.dll error. After one of the machines comes down it is disregarded by the redirector (unless the jvmRoute is already stamped in it's session cookie). For all new sessions, the redirector "sticks" to the machine that never went down. If I stop and restart the IIS server (hence the isapi_redirector.dll), then round robin resumes !
Maybe the mod_jk has this fixed. I just found the mod_jk stuff out on Jakarta and believe it to be a more recent incarnation of isapi_redirect. Thanks for pointing it out to me.
1) I do have log files - attached. They are extracts from the console windows of both serverOne and serverTwo and the log file from serverOne when it comes back online.
It shows that when the server comes back online, it knows how many cluster members there are, and does a 'setState for HASessionStateTransfer' and then 'State was retrieved successfully'. So I think it reads the state even if it's not changed.
I think this stuff is going to work!
2) I see that mod_jk is simply the apache version of isapi_redirector afterall. I've re-build with the latest isapi_redirector code and still it doesn't like a server once it's gone down -- even after it comes back up again. Probably both mod_jk and isapi_redirector will behave this way as I see they use the same exact code -- the module is probably common/jk_lb_worker.c.
Sorry for my inexcusable delay answering to this thread
Thank you very much for your help (patches, logs, messages, recommendations, etc.)
Any updates on this one? I found that using JBoss3-0.3_tomcat4.0.5 , when i specify jvmRoute i get session affinity and replication, however when i switch off server1 i get a NullPointerException on server2.. at ClusterManager.findSession().server1 & server2 are on different machines.Iam using mod_jk for loadbalancing.
so i have session affinity + replication , no session failback.Anyone who has achieved all three please do post a reply.(servers on different machines)
I am trying to build JBoss and seem to have the same problem as you did: AppConfigurationEntry class variable cannot resolve '.module'. What was the resolution of this problem for you? Thank you.