[
https://issues.apache.org/jira/browse/CASSANDRA-4419?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13407993#comment-13407993
]
Sylvain Lebresne commented on CASSANDRA-4419:
---------------------------------------------
The problem is that compaction is considering uncompressed size for position
(and thus cut segments at 96MB of uncompressed data) while the manifest
consider compressed sizes for level sizes. This was fixed by CASSANDRA-4341 for
cassandra-1.1, but I guess we might want to backport that part of the patch to
1.0.
> leveled compaction generates too small sstables
> -----------------------------------------------
>
> Key: CASSANDRA-4419
> URL: https://issues.apache.org/jira/browse/CASSANDRA-4419
> Project: Cassandra
> Issue Type: Bug
> Affects Versions: 1.0.10
> Reporter: Noa Resare
> Priority: Minor
>
> When I set sstable_size_in_mb to 96 I end up with sstable data files no
> larger than 60M.
> This in turn messes up the LeveledManifest calculation since it finds
> compaction candidates by summing up the size of sstables at a particular
> level and comparing it to the configured size multiplied by the desired
> number of sstables at a level, resulting in ~20 sstables in level 1 instead
> of the 10 that one would expect from looking at LeveledManifest.
> Some additional logging here reveals that the position parameter passed to
> LeveledCompactionTask.newSSTableSegmentThresholdReached() is significantly
> higher than the size of the output file.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
