[jira] [Updated] (LUCENE-6579) Unexpected merge exceptions should be tragic to IndexWriter

["Michael McCandless (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/LUCENE-6579?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Michael McCandless updated LUCENE-6579:
---------------------------------------
    Attachment: LUCENE-6579.patch

Here's a tentative patch ... it was tricky because calling IW.tragicEvent from a merge thread quickly lead to deadlock.  I had to relax a few locking/blocking places to work around that ...

I also had to fix some tests to expected that IW is closed after hitting a merge exc, but I'm sure there's a long tail still there ... this will be destabilizing at first.

I still need to add a dedicated test that confirms IW is closed after a merge exc.

> Unexpected merge exceptions should be tragic to IndexWriter
> -----------------------------------------------------------
>
>                 Key: LUCENE-6579
>                 URL: https://issues.apache.org/jira/browse/LUCENE-6579
>             Project: Lucene - Core
>          Issue Type: Bug
>            Reporter: Michael McCandless
>            Assignee: Michael McCandless
>             Fix For: 5.3, Trunk
>
>         Attachments: LUCENE-6579.patch
>
>
> Today our behavior is weird: we will fail the merge (which is running in a background thread if you are using the default CMS), pause for 1.0 seconds, and then the next chance we get, kick off the merge again.
> I think this is a poor default, e.g. on disk full we will just keep "trying" and filling up disk again, wasting IO/CPU.
> I think IW should declare this a tragedy instead?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

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