[
https://issues.apache.org/jira/browse/CASSANDRA-7019?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15096196#comment-15096196
]
Marcus Eriksson commented on CASSANDRA-7019:
--------------------------------------------
Perhaps we could apply it to the higher (L3+?) levels in leveled compaction? Or
perhaps the highest and highest - 1 levels. I see two reasons for this;
* we would reduce the number of sstables we have to read when doing these
compactions - since most sstables in L0 overlap all other sstables we would
read everything on disk if we did this for all levels. An sstable in L3 will
only overlap 1 (or 2) sstables in each of L2 and L1.
* data in the higher levels should be older, meaning we most often can drop
more covered data/tombstones when doing the high level compactions
Question: (I only quickly skimmed the code) do we drop overwritten data as well?
This approach will not be 'perfect' like the 'compact all sstables
1:1'-approach outlined above, but I like it - it is quite simple to reason
about and should give good results (if the performance does not turn out to be
horrible).
> Improve tombstone compactions
> -----------------------------
>
> Key: CASSANDRA-7019
> URL: https://issues.apache.org/jira/browse/CASSANDRA-7019
> Project: Cassandra
> Issue Type: Improvement
> Components: Compaction
> Reporter: Marcus Eriksson
> Assignee: Branimir Lambov
> Labels: compaction
> Fix For: 3.x
>
>
> When there are no other compactions to do, we trigger a single-sstable
> compaction if there is more than X% droppable tombstones in the sstable.
> In this ticket we should try to include overlapping sstables in those
> compactions to be able to actually drop the tombstones. Might only be doable
> with LCS (with STCS we would probably end up including all sstables)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
