[
https://issues.apache.org/jira/browse/CASSANDRA-10276?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14977863#comment-14977863
]
Jeff Jirsa edited comment on CASSANDRA-10276 at 10/28/15 6:51 AM:
------------------------------------------------------------------
I'm definitely not disagreeing with you, but like always, the devil's in the
details. If there were 3 40MB files, 3 120M files, 3 500M files, 3 2G files, 3
8G files, and 3 24G files all in one window, it might be nice to do one big
rollup compaction (and that's a pretty extreme example, but let's look at the
edge cases, because someone's bound to hit it - one single 40M flush away from
having a single 100G file in normal STCS operation, but in an old window, that
next 40M file may never show up, so we'll end up keeping 17 extra sstables
around for no reason).
was (Author: jjirsa):
I'm definitely not disagreeing with you, but like always, the devil's in the
details. If there were 3 40MB files, 3 120M files, 3 500M files, 3 2G files, 3
8G files, and 3 24G files all in one window, it might be nice to do one big
rollup compaction (and that's a pretty extreme example, but let's look at the
edge cases, because someone's bound to hit it - one single 40M flush away from
having a single 100G file in normal STCS operation).
> With DTCS, do STCS in windows if more than max_threshold sstables
> -----------------------------------------------------------------
>
> Key: CASSANDRA-10276
> URL: https://issues.apache.org/jira/browse/CASSANDRA-10276
> Project: Cassandra
> Issue Type: Sub-task
> Components: Core
> Reporter: Marcus Eriksson
> Assignee: Marcus Eriksson
> Fix For: 3.x, 2.1.x, 2.2.x
>
>
> To avoid constant recompaction of files in big ( > max threshold) DTCS
> windows, we should do STCS of those files.
> Patch here: https://github.com/krummas/cassandra/commits/marcuse/dtcs_stcs
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
