cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benedict (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-8124) Stopping a node during compaction can make already written files stay around
Date Mon, 09 Feb 2015 18:23:34 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-8124?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14312565#comment-14312565
] 

Benedict commented on CASSANDRA-8124:
-------------------------------------

I'm confused. Isn't this what CFS.removeUnfinishedCompactionLeftovers is for? Doesn't this
still leave a window for it to fail, otherwise, since the creatoin of the non-temp files and
swapping in is not atomic...

> Stopping a node during compaction can make already written files stay around
> ----------------------------------------------------------------------------
>
>                 Key: CASSANDRA-8124
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-8124
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Marcus Eriksson
>            Assignee: Marcus Eriksson
>              Labels: triaged
>             Fix For: 2.1.2
>
>         Attachments: 0001-8124-wip.patch
>
>
> In leveled compaction we generally create many files during compaction, in 2.0 we left
the ones we had written as -tmp- files, in 2.1 we close and open the readers, removing the
-tmp- markers.
> This means that any ongoing compactions will leave the resulting files around if we restart.
Note that stop:ing the compaction will cause an exception and that makes us call abort() on
the SSTableRewriter which removes the files.
> Guess a fix could be to keep the -tmp- marker and make -tmplink- files until we are actually
done with the compaction.



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

Mime
View raw message