[jira] [Commented] (CASSANDRA-11817) Change the way LCS major compaction works

["Jonathan Ellis (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-11817?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15286941#comment-15286941 ] 

Jonathan Ellis commented on CASSANDRA-11817:
--------------------------------------------

From IRC, some more context:

[marcus]
bq.  I guess we could avoid the CASSANDRA-7414 problem if we pick the level to write to in a good way, maybe pick the max level that is "full"? if so we should have most data there and quickly start involving that level in compactions.  Or maybe the biggest level, if max level is a single sstable it would probably be a bad pick, if it is 80% full, it is probably the best choice.

> Change the way LCS major compaction works
> -----------------------------------------
>
>                 Key: CASSANDRA-11817
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-11817
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Marcus Eriksson
>             Fix For: 3.x
>
>
> Currently we start writing in L1 and then continue in L2 once L1 is full, it is probably better to pick a high level and write all sstables to that level.
> I suspect that the end result will be basically the same without a bunch of pointless single-sstable compactions into an empty space in level+1



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