Possible rare thread hazard in IW.commit
----------------------------------------
Key: LUCENE-2782
URL: https://issues.apache.org/jira/browse/LUCENE-2782
Project: Lucene - Java
Issue Type: Bug
Reporter: Michael McCandless
Assignee: Michael McCandless
Fix For: 2.9.4, 3.0.3, 3.1, 4.0
I was seeing a very rare intermittent failure in
TestIndexWriter.testCommitThreadSafety.
The issue happens if one thread calls commit while another is flushing, and is
exacerbated at high flush rates (eg maxBufferedDocs=2). The thread doing
commit will first flush, and then it syncs the files. However in between those
two, if other threads manage to add enough docs and trigger another flush, a
2nd new segment can sneak into the SegmentInfos before we sync.
This is normally harmless, in that it just means the commit includes a few more
docs that had been added by other threads, so it's fine. But, it can mean that
a committed segment references the still-open doc store files. Our tests now
catch this (I changed MockDirWrapper to throw an exception in this case), and
so testCommitThreadSafety can fail with this exception. If you hardwire the
maxBufferedDocs to 2 it happens quite often.
It's not clear this is really a problem in real apps vs just our anal
MockDirWrapper but I think we should fix it...
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
