[
https://jira.jboss.org/jira/browse/JGRP-1103?page=com.atlassian.jira.plug...
]
Bela Ban commented on JGRP-1103:
--------------------------------
JGRP-1085 was the culprit: the optimzations where some xmit_table or retransmitter related
code was moved outside the lock scope caused duplicate message delivery in some edge
cases.
Reverting the changes, passes all NRW related tests all the time
NakReceiverWindow: concurrent insertion may insert the same seqno
multiple times
--------------------------------------------------------------------------------
Key: JGRP-1103
URL:
https://jira.jboss.org/jira/browse/JGRP-1103
Project: JGroups
Issue Type: Bug
Reporter: Bela Ban
Assignee: Bela Ban
Priority: Blocker
Fix For: 2.8
When concurrently inserting the same range of seqnos (e.g. 1-100) into the
NakReceiverWindow by N threads (e.g. 10), we sometimes end up with duplicate messages for
the *same* seqno ! This happens very rarely and only when many threads insert the same
message !
This can be observed in DuplicateTest (testOOBMulticastsToAll3Senders()): sometimes the
final list contains more than 10 elements.
I reproduced this by testing NakReceiverWindow in isolation, the test is
NakReceiverWindowTest2.
I assume this is a regression introduced in 2.8 by moving insertions to and removal from
xmit_table out of the write_lock scope in method NakReceiverWindow.add().
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
https://jira.jboss.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira