cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Brandon Williams (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-2662) Nodes get ignored by dynamic snitch when read repair chance is zero
Date Wed, 18 May 2011 15:20:47 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-2662?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13035431#comment-13035431
] 

Brandon Williams commented on CASSANDRA-2662:
---------------------------------------------

To clarify what this patch does: on the first round, behavior is unchanged, it will fall back
to the subsnitch as usual.  On the second round however, the nodes will have scores so the
dynamic snitch will determine the best host.  This seems like a good idea, but I think we
should proceed cautiously with regard to CASSANDRA-1314.  This is certainly going to blow
the cache on the second round, since the unpinned hosts at zero are guaranteed to be ranked
higher than the pinned one.  However, in the current implementation, the badness threshold
in CASSANDRA-1519 can never take effect at RR = 0 since the scores of the other replicas will
always be null, and thus diversion to the subsnitch will always occur, returning the pinned
replica.  With this patch, the second replica is likely to end up being the pinned one, which
I think will be ok, as long as all coordinators agree, though this might not always be the
case.

> Nodes get ignored by dynamic snitch when read repair chance is zero
> -------------------------------------------------------------------
>
>                 Key: CASSANDRA-2662
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-2662
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 0.7.6
>            Reporter: Daniel Doubleday
>            Priority: Trivial
>         Attachments: dynsnitch.patch
>
>
> DynamicEndpointSnitch falls back to subsnitch when one of the scores of the endpoints
being compared is missing.
> This leads to a stable order of hosts until reads will lead to recorded scores. 
> If setting read repair chance to 0 and reads are performed with quorum then (rf - # quorum
nodes) will never get reads.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message