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] Commented: (LUCENE-1566) Large Lucene index can hit false OOM due to Sun JRE issue
Date Wed, 15 Jul 2009 20:02:14 GMT

    [ https://issues.apache.org/jira/browse/LUCENE-1566?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12731651#action_12731651
] 

Michael McCandless commented on LUCENE-1566:
--------------------------------------------

Looks good Simon!

  * From your post above, I thought the "read in 100MB chunks into the
    big array" failed to work around the issue?  Ie that's why you
    allocated a tiny intermediate buffer?  Did that turn out not to be
    the case?

  * How about, instead of ignoring the chunk size on 64bit jre, we
    conditionally default it to Long.MAX_VALUE?

  * Can you link also to the Sun VM bug in the javadocs for these
    set/getChunkSize?


> Large Lucene index can hit false OOM due to Sun JRE issue
> ---------------------------------------------------------
>
>                 Key: LUCENE-1566
>                 URL: https://issues.apache.org/jira/browse/LUCENE-1566
>             Project: Lucene - Java
>          Issue Type: Bug
>          Components: Index
>    Affects Versions: 2.4.1
>            Reporter: Michael McCandless
>            Assignee: Simon Willnauer
>            Priority: Minor
>             Fix For: 2.9
>
>         Attachments: LUCENE-1566.patch, LUCENE-1566.patch, LUCENE_1566_IndexInput.patch
>
>
> This is not a Lucene issue, but I want to open this so future google
> diggers can more easily find it.
> There's this nasty bug in Sun's JRE:
>   http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6478546
> The gist seems to be, if you try to read a large (eg 200 MB) number of
> bytes during a single RandomAccessFile.read call, you can incorrectly
> hit OOM.  Lucene does this, with norms, since we read in one byte per
> doc per field with norms, as a contiguous array of length maxDoc().
> The workaround was a custom patch to do large file reads as several
> smaller reads.
> Background here:
>   http://www.nabble.com/problems-with-large-Lucene-index-td22347854.html

-- 
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