[
https://issues.apache.org/jira/browse/LUCENE-5958?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Michael McCandless updated LUCENE-5958:
---------------------------------------
Attachment: LUCENE-5958.patch
The changes caused a risk of deadlock in IW, which we hit in recent nightly
test failure.
Here's a patch to fix that; the issue is we cannot be sync'd on IW when dealing
with a tragedy ... there were two places in IW where we were doing that. I
added an assert that we don't hold the lock, and fixed those two places.
I tried to simplify finishCommit as much as I could, and get the infoStream
logging out of the tragedy handling as much as I could ...
> OOM or exceptions during checkpoint make IndexWriter have a bad day
> -------------------------------------------------------------------
>
> Key: LUCENE-5958
> URL: https://issues.apache.org/jira/browse/LUCENE-5958
> Project: Lucene - Core
> Issue Type: Bug
> Reporter: Robert Muir
> Fix For: 4.10.1, 5.0, Trunk
>
> Attachments: LUCENE-5958.patch, LUCENE-5958.patch
>
>
> During finishCommit(), we run checkpoint after we wrote the commit to disk,
> but if things go wrong here (e.g. IOError when IFD deletes a pending file,
> OOM), then everything will go wrong (we won't even properly incref things,
> and may end out deleting wrong files if the user calls rollback, leaving a
> corrupt index).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
