lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Michael McCandless (JIRA)" <j...@apache.org>
Subject [jira] Created: (LUCENE-1200) IndexWriter.addIndexes* can deadlock in rare cases
Date Tue, 04 Mar 2008 10:16:40 GMT
IndexWriter.addIndexes* can deadlock in rare cases
--------------------------------------------------

                 Key: LUCENE-1200
                 URL: https://issues.apache.org/jira/browse/LUCENE-1200
             Project: Lucene - Java
          Issue Type: Bug
          Components: Index
    Affects Versions: 2.4
            Reporter: Michael McCandless
            Assignee: Michael McCandless
            Priority: Minor
             Fix For: 2.4


In somewhat rare cases it's possible for addIndexes to deadlock
because it is a synchronized method.

Normally the merges that are necessary for addIndexes are done
serially (with the primary thread) because they involve segments from
an external directory.  However, if mergeFactor of these merges
complete then a merge becomes necessary for the merged segments, which
are not external, and so it can run in the background.  If too many BG
threads need to run (currently > 4) then the "pause primary thread"
approach adopted in LUCENE-1164 will deadlock, because the addIndexes
method is holding a lock on IndexWriter.

This was appearing as a intermittant deadlock in the
TestIndexWriterMerging test case.

This issue is not present in 2.3 (it was caused by LUCENE-1164).

The solution is to shrink the scope of synchronization: don't
synchronize on the whole method & wrap synchronized(this) in the right
places inside the methods.

-- 
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: java-dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-dev-help@lucene.apache.org


Mime
View raw message