hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Elliott Clark (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-15383) Load distribute across secondary read replicas for meta
Date Thu, 03 Mar 2016 20:12:18 GMT

    [ https://issues.apache.org/jira/browse/HBASE-15383?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15178535#comment-15178535
] 

Elliott Clark commented on HBASE-15383:
---------------------------------------

>From the point of view of the client all responses from meta can be stale. So just assume
that any secondary is good on the first time and after that go to the primary.

> Load distribute across secondary read replicas for meta
> -------------------------------------------------------
>
>                 Key: HBASE-15383
>                 URL: https://issues.apache.org/jira/browse/HBASE-15383
>             Project: HBase
>          Issue Type: Improvement
>          Components: Client
>            Reporter: Ashu Pachauri
>            Assignee: Ashu Pachauri
>             Fix For: 2.0.0, 1.3.0
>
>
> Right now, we always hit the primary replica for meta and fallback to the secondary replicas
in case of a timeout. This can hamper performance in scenarios where meta becomes a hot region
e.g. cluster ramp up..clients dropping connections etc.
> It's good to have a load distribution approach on meta's secondary replicas with fallback
to primary if we read stale data.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message