cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Yuki Morishita (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-4932) Agree on a gcbefore/expirebefore value for all replica during validation compaction
Date Mon, 08 Apr 2013 18:51:16 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-4932?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13625659#comment-13625659
] 

Yuki Morishita commented on CASSANDRA-4932:
-------------------------------------------

Overall, it looks good.
Is it better to calculate gc_before just before sending TreeRequest rather than doing it in
RepairJob constructor?
RepairJobs for all given CFs are created at once, but those are executed one by one.
                
> Agree on a gcbefore/expirebefore value for all replica during validation compaction 
> ------------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-4932
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-4932
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Sylvain Lebresne
>            Assignee: Marcus Eriksson
>            Priority: Minor
>             Fix For: 2.0
>
>         Attachments: 0001-CASSANDRA-4932.patch
>
>
> CASSANDRA-4905 made it so that we don't include gcable tombstones in merkle tree digests
because otherwise the fact that a gcable tombstone exists in a given node depends on the local
scheduling of compactions, which make the merkle-tree digets depends on that which is bad.
> However, we still depends on a value of gcbefore/expirebefore that is computed locally
on the node. This is not a huge deal because each node should compute those value roughly
at the same time so the discrepancies shouldn't exceed a few hundreds of milliseconds in general
and a few seconds if we're unlucky. Still, it's fairly easy to fix this problem by having
the repair coordinator pick the gcbefore/expirebefore and share that with all the replica.
It does require sending that info on the wire though, so will require a new messaging version
and thus can only be done in a major cycle. 

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message