You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Branimir Lambov (Jira)" <ji...@apache.org> on 2021/08/03 14:59:00 UTC
[jira] [Commented] (CASSANDRA-16782) Improve the way we pick
sstables for STCS-in-L0 and in TWCS 'current' window
[ https://issues.apache.org/jira/browse/CASSANDRA-16782?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17392350#comment-17392350 ]
Branimir Lambov commented on CASSANDRA-16782:
---------------------------------------------
I have some general questions about the improvement. It looks like it is a combination of two things:
* A drastic increase of STCS's limit for max sstables in a compaction, and
* A preference of the STCS level to compact based on the number of sstables,
which only apply to uses of STCS in LCS and TWCS.
I can see a lot of value in the latter, but why not apply it always, including plain STCS? STCS's selection of bucket to compact does result in accumulation of sstables on the smallest-sstables bucket and at DataStax we were recently discussing selecting the most populous bucket as a possible solution to this problem, which might be good enough even without raising the max limit. I personally could not find any downsides to it, and I wonder why you would prefer to restrict its usage?
The former we can achieve by configuration, can't we? (Perhaps also adding a max size limit.) Speaking of which, is there a reason to place the new settings in {{cassandra.yaml}} instead of compaction parameters?
> Improve the way we pick sstables for STCS-in-L0 and in TWCS 'current' window
> ----------------------------------------------------------------------------
>
> Key: CASSANDRA-16782
> URL: https://issues.apache.org/jira/browse/CASSANDRA-16782
> Project: Cassandra
> Issue Type: Improvement
> Components: Local/Compaction/LCS, Local/Compaction/TWCS
> Reporter: Marcus Eriksson
> Assignee: Marcus Eriksson
> Priority: Normal
> Fix For: 4.x
>
>
> The goal when being behind in L0 should always be to get the number of sstables down to a reasonable level as soon as possible. Currently it is common that we run compactions on the large sstables but leave thousands of tiny sstables behind.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cassandra.apache.org
For additional commands, e-mail: commits-help@cassandra.apache.org