-
30. Re: Changes on the JBM stack on JBoss5
Hi Guys,
Sorry to jump the thread ;) but also having getting the same issue. Tested it with 2.6.5 and the 2.6.6 version set up. I am able to reproduce it consistently.
Dont know if you have resolved the issue or not but will post details of what i have and if you want more let me know.
What i have is just a 2 node cluster, inside each node i have 2 channels, im not using the multiplexer but setting the "singleton_name" attribute to be the same so both channels share jgroup resources.
Basically whats happening is this, i havetry{ ChannelManager.getChannel().getState(dest, Path1, timeout); ChannelManager.getChannel().getState(dest, Path2, timeout); ChannelManager.getChannel().getState(dest, Path3, timeout); }catch(Exception e){ log.error("problem asking older node for its state",e); } ......... ChannelManager.broadcastMessage(msg);
Now whats happening is, the exception gets thrown and is caught and printed but as soon as i try to do the "broadcastMessage" it hangs and my whole thread goes into wait state.
Here is the thread dump."LiveFeed_ConnectionHandler 1" daemon prio=3 tid=0x0983cc00 nid=0x51 in Object.wait() [0x7817e000..0x7817eaf0] java.lang.Thread.State: WAITING (on object monitor) at java.lang.Object.wait(Native Method) - waiting on <0x7ed443e0> (a java.lang.Object) at java.lang.Object.wait(Object.java:485) at org.jgroups.protocols.pbcast.FLUSH.blockMessageDuringFlush(FLUSH.java:300) - locked <0x7ed443e0> (a java.lang.Object) at org.jgroups.protocols.pbcast.FLUSH.down(FLUSH.java:254) at org.jgroups.stack.ProtocolStack.down(ProtocolStack.java:475) at org.jgroups.JChannel.down(JChannel.java:1452) at org.jgroups.JChannel.send(JChannel.java:625)
What is getting transfered is this, i am using jboss cache, and ask another node to transfer a branch, this branch that i get this exception on has 260,000 records with each record containing about 10 fields. I havent seen the exception with smaller branches, just this one.
if you want anymore information let me know, if you think you have a fix let me know, i will check it out and see if it works for my scenario.
Cheers,
LL -
31. Re: Changes on the JBM stack on JBoss5
Ok ignore the hanging bit..........that happens when you set your timeout to be 0 ;)
-
32. Re: Changes on the JBM stack on JBoss5
LL,
You should probably open a thread on the JBoss Cache user's forum, posting a link to this thread.
It looks like Vladmir fixed this on JGroups 2.6.7 to be released by the end of this week. So maybe all you have to do is to give a try on that release.
Anyway you should open the thread on JBossCache User's forum -
33. Re: Changes on the JBM stack on JBoss5
BTW: This bug on JGroups is not just related to JBM. Brian just told me EJB3 had as similar failure yesterday on their testsuite.
-
34. Re: Changes on the JBM stack on JBoss5
The fix looks good.. I have already ran three successful runs with the clustered-testsuite, plus another with failures related to our QA environment (probably slow response from the DB).
When this will be released officially? -
35. Re: Changes on the JBM stack on JBoss5
Just for the record... I have executed a single clustered-test in a loop for 150 times.. so the issue is gone for sure.
Thanks for the help...
And it is good as this issue was not affecting JBM only -
36. Re: Changes on the JBM stack on JBoss5
We have another problem to solve now.
The latest JGroups (not sure if 2.6.7 or 2.6.6) changed one semantic over the channels.
We have a test that instantiate several channels inside the same VM. Before this was working fine, but now I can't connect more than one channel on the same singletonName on the stack. I have removed the singleton attribute but then the channels are not seeing each other.
I create multiple stacks (one for each instance) and they don't see each other either.
How can someone create multiple connected JChannels on the UDP inside the same VM for tests?
For the Record, the test I have problem now is ClusteredPostOfficeTest. -
37. Re: Changes on the JBM stack on JBoss5
I think this is more of a JGroups question than a messaging question.
-
38. Re: Changes on the JBM stack on JBoss5
"bstansberry@jboss.com" wrote:
Use loopback="true"; I'll post details as to why on Design of Clustering forum in a while and add a link to the post here.
Promised post as to why loopback="true" is a good default:
http://www.jboss.org/community/docs/DOC-10920 -
39. Re: Changes on the JBM stack on JBoss5
"timfox" wrote:
I think this is more of a JGroups question than a messaging question.
True, but the semantic changed and it is affecting JBM.
I tried all the possible current alternatives using the multiplexor, and I can' t get multiple channels inside the same VM to see each other. -
40. Re: Changes on the JBM stack on JBoss5
"clebert.suconic@jboss.com" wrote:
"timfox" wrote:
I think this is more of a JGroups question than a messaging question.
True, but the semantic changed and it is affecting JBM.
I tried all the possible current alternatives using the multiplexor, and I can' t get multiple channels inside the same VM to see each other.
My point is, if JGroups behaviour has changed between releases, one of the JGroups team can probably advise you better on what has changed and how to deal with it than one of the JBM team :) -
41. Re: Changes on the JBM stack on JBoss5
Is the version of this test that is expected to work currently checked in on Branch_Stable? If not, send it to me, including any required resource files.
-
42. Re: Changes on the JBM stack on JBoss5
And please provide instructions on how to run it as a single test.
-
43. Re: Changes on the JBM stack on JBoss5
I will commit some changes I made to the test, and send you the instructions.
-
44. Re: Changes on the JBM stack on JBoss5
"clebert.suconic@jboss.com" wrote:
We have another problem to solve now.
How can someone create multiple connected JChannels on the UDP inside the same VM for tests?
Heh, this problem with singleton testing has been bothering us as well. Brian, Bela and I tried to find a sensible solution but we have given up for now.