cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Marcus Eriksson (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-9508) Avoid early opening of compaction results when doing anticompaction
Date Fri, 29 May 2015 07:13:17 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-9508?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14564327#comment-14564327
] 

Marcus Eriksson commented on CASSANDRA-9508:
--------------------------------------------

unit test pushed to the same repo (https://github.com/krummas/cassandra/commits/marcuse/9508)

> Avoid early opening of compaction results when doing anticompaction
> -------------------------------------------------------------------
>
>                 Key: CASSANDRA-9508
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-9508
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Marcus Eriksson
>            Assignee: Marcus Eriksson
>            Priority: Critical
>             Fix For: 3.x, 2.1.x, 2.2.0 rc1
>
>
> When doing anticompaction we have 2 sstable rewriters open and write the repaired range
to one writer and the unrepaired to one. This can cause some keys to become unreadable during
anticompaction, example:
> Say we are anticompacting one 150MB sstable into two new ones, 75MB each, first we start
writing to the first one, after 50MB, we open the result early and move the start point up
to the 50MB mark in the original file. Then, after 75MB written, we start writing to the next
file and after writing 50MB to that file, we open that file early and move the start point
up to 125MB in the original file. This makes 25MB unreadable that is only present in the first
file written, but we have moved the start point beyond that part of the file in the original
> Simple solution is probably to avoid doing early opening during anticompaction, and since
anticompaction is not as frequent as ordinary compactions, the performance implications would
not be too big



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

Mime
View raw message