[
http://jira.jboss.com/jira/browse/JGRP-455?page=comments#action_12363687 ]
Bela Ban commented on JGRP-455:
-------------------------------
The issue was in STABLE: the heard_from list could be incorrectly updated, so we'd
sometimes send a STABILITY message even though we had not yet received votes from all
members.
I changed the design slightly, so that now instead of heard_from we have votes, which is
empty initially and only when all members have sent their votes, we send a STABILITY
message and clear votes again.
The clearing of votes and resetting of the digest is now done with the same lock
("lock"), as there was a race condition when synchronizing separately on digest
and votes.
Tests with 1M 1K messages per sender on 8 nodes passed multiple times (before the error
occurred after 100K messages).
Problem with retransmission: message not found
----------------------------------------------
Key: JGRP-455
URL:
http://jira.jboss.com/jira/browse/JGRP-455
Project: JGroups
Issue Type: Bug
Affects Versions: 2.4.1 SP1
Reporter: Bela Ban
Assigned To: Bela Ban
Priority: Blocker
Fix For: 2.5
Attachments: tmp.txt, tmp2.txt, udp-atl.xml
Sometimes a sender cannot find a message with a seqno as requested by a receiver. The
seqno requested (47176) was already purged, sent_msgs: [48059 - 48090] (32)
This could be either an issue in STABLE, which prematurely purges messages although not
all members agreed on the highest seqnos seen, or NAKACK, which might incorrectly
'change' seqnos. Very hard to reproduce !
Trace:
42032 [ERROR] [OOB Thread,perf,10.68.32.33:32968] NAKACK.handleXmitReq(): -
(requester=10.68.28.33:33763, local_addr=10.68.32.33:32968) message
10.68.32.33:32968::47176 not found in sent msgs [48059 - 48090] (32)
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://jira.jboss.com/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira