reference clearing for ConcurrentHashMap can deadlock with receive txn apply
----------------------------------------------------------------------------
Key: CDV-596
URL: https://jira.terracotta.org/jira//browse/CDV-596
Project: Community Development
Issue Type: Bug
Affects Versions: 2.5.0
Reporter: Tim Eck
Assignee: Issue Review Board
I'll attach some thread dumps of this in action, but the problem starts when
the memory manager in the client calls __tc_clearReferences() on
ConcurrentHashMap. The memory manager always takes the TCObject resolve lock
before calling __tc_clearReferences(). In ConcurrentHashMap,
__tc_clearReference() also takes distributed locks (on the segments), this
particular nesting is problematic. In order to obtain the segment locks, txns
from other nodes might need to be applied locally. Those incoming txns will
need to obtain the resolve lock to apply(), thus you end up with a deadlock.
The fix is to not take distributed locks on the segments (local only)
This has been making the ConcurrentHashMapSwapingTest fail for quite a while
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
https://jira.terracotta.org/jira//secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
_______________________________________________
tc-dev mailing list
[email protected]
http://lists.terracotta.org/mailman/listinfo/tc-dev
