[
https://issues.apache.org/jira/browse/LUCENE-5024?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13670416#comment-13670416
]
Robert Muir commented on LUCENE-5024:
-------------------------------------
The best solution here i think, is removal of create-or-append.
really index creation can be a one time thing you must do separately before you
can use the directory. this is typically how its done: lucene is wierd and has
this broken mechanism today instead.
> Can we reliably detect an incomplete first commit vs index corruption?
> ----------------------------------------------------------------------
>
> Key: LUCENE-5024
> URL: https://issues.apache.org/jira/browse/LUCENE-5024
> Project: Lucene - Core
> Issue Type: Bug
> Components: core/index
> Reporter: Michael McCandless
> Fix For: 5.0, 4.4
>
>
> Normally, if something bad happens (OS, JVM, hardware crashes) while
> IndexWriter is committing, we will just fallback to the prior commit
> and no intervention necessary from the app.
> But if that commit is the first commit, then on restart IndexWriter
> will now throw CorruptIndexException, as of LUCENE-4738.
> Prior to LUCENE-4738, in LUCENE-2812, we used to try to detect the
> corrupt first commit, but that logic was dangerous and could result in
> falsely believing no index is present when one is, e.g. when transient
> IOExceptions are thrown due to file descriptor exhaustion.
> But now two users have hit this change ... see "CorruptIndexException
> when opening Index during first commit" and "Calling
> IndexWriter.commit() immediately after creating the writer", both on
> java-user.
> It would be nice to get back to not marking an incomplete first commit
> as corruption ... but we have to proceed carefully.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
