[
https://issues.apache.org/jira/browse/CASSANDRA-6296?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13819153#comment-13819153
]
T Jake Luciani commented on CASSANDRA-6296:
-------------------------------------------
Oh streaming fails over now? Great.
It would still seem like for first attempt it should try to use the same levels
as the replica. If it sends random sstables from random levels 6323 would
cause a lot of unnecessary compaction still.
Perhaps we could put logic in 6323 that will prefer the hinted level when it
picks and if that level is not available it will fall back to any open level?
> Bootstrapping a new node should not require much compaction when leveled
> compaction is used
> -------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-6296
> URL: https://issues.apache.org/jira/browse/CASSANDRA-6296
> Project: Cassandra
> Issue Type: Improvement
> Reporter: T Jake Luciani
> Priority: Minor
>
> boostrapping a new node currently requires recompacting all data. In the
> case of leveled compaction we already have the data in non-overlapping
> levels. Since we stream from a single host for a single range this shouldn't
> require any compaction if the sstable is > level 0
--
This message was sent by Atlassian JIRA
(v6.1#6144)
