[
https://issues.apache.org/jira/browse/CASSANDRA-10574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15004122#comment-15004122
]
Aleksey Yeschenko commented on CASSANDRA-10574:
-----------------------------------------------
The only reason we compact them at all is to clear out the tombstones. I'm not
sure that this is a workaround we want to commit, as it doesn't fix the issue
fundamentally (and with the flag you are risking failures due to tombstone
overwhelmed exceptions when reading hints).
The true fix is the new hints implementation in 3.0 (which, I know, doesn't
help you much right now, sorry).
I'm afraid you'll have to patch your C* locally if you want to run this patch
(again, sorry).
> Add the option to skip compaction for hints cf when doing the handoff.
> ----------------------------------------------------------------------
>
> Key: CASSANDRA-10574
> URL: https://issues.apache.org/jira/browse/CASSANDRA-10574
> Project: Cassandra
> Issue Type: Improvement
> Reporter: Dikang Gu
> Assignee: Dikang Gu
> Fix For: 2.1.x
>
> Attachments: hints.patch
>
>
> In our production env, we stop the gossip for about 1 hour on around 1/3 of
> nodes, and then we enable the gossip and it took very long time for the other
> 2/3 nodes to do the hints handoff.
> I find that most of the time is spent to do the compaction before each hints
> handoff, while in most case, there is only one sstable, and it's really not
> necessary to do the compaction.
> So I add the option to skip the compaction to allow the nodes to finish hints
> handoff very quickly.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
