[
https://issues.apache.org/jira/browse/LUCENE-3365?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13080942#comment-13080942
]
Geoff Cooney commented on LUCENE-3365:
--------------------------------------
I'm uploading a patch that contains a test that demonstrates this issue as well
as a proposed fix.
The proposed fix is simply to move the create or append decision a step lower
so it occurs after the lock is obtained.
> Create or Append mode determined before obtaining write lock
> ------------------------------------------------------------
>
> Key: LUCENE-3365
> URL: https://issues.apache.org/jira/browse/LUCENE-3365
> Project: Lucene - Java
> Issue Type: Bug
> Components: core/index
> Affects Versions: 3.1, 3.2, 3.3
> Reporter: Geoff Cooney
> Priority: Minor
> Labels: IndexWriter
>
> If an IndexWriter("writer1") is opened in CREATE_OR_APPEND mode, it
> determines whether to CREATE or APPEND before obtaining the write lock. When
> another IndexWriter("writer2") is in the process of creating the index, this
> can result in writer1 entering create mode and then waiting to obtain the
> lock. When writer2 commits and releases the lock, writer1 is already in
> create mode and overwrites the index created by write2.
> This bug was probably effected by LUCENE-2386 as prior to that Lucene
> generated an empty commit when a new index was created. I think the issue
> could still have occurred prior to that but the two IndexWriters would have
> needed to be opened nearly simultaneously and the first IndexWriter would
> need to release the lock before the second timed out.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
