You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Benedict (JIRA)" <ji...@apache.org> on 2018/09/13 07:21:00 UTC
[jira] [Commented] (CASSANDRA-14743) Better health based routing
for transient reads
[ https://issues.apache.org/jira/browse/CASSANDRA-14743?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16613122#comment-16613122 ]
Benedict commented on CASSANDRA-14743:
--------------------------------------
{quote}As a heuristic a transient replica will be a little faster for a read compared to a full replica given two endpoints with similar scores so maybe we don't need additional health information.
{quote}
This is only true while the transient status is healthy. When performing a QUORUM in a multi-DC scenario, for instance, you might prefer a remote digest over a local transient response if there's a down local node. Or in a scenario where repair has been unable to run for a period, or inter-DC bandwidth is saturated, digest queries might entail less time in flight.
If nothing else, separate statistics for transient reads is a useful health barometer for the system. Ideally, though, we would be tracking separate statistics for each endpoint; this might be impractical for systems where all nodes are coordinators (though perhaps we could begin tracking stats for peers separately, since these are the expected and optimal case.
> Better health based routing for transient reads
> ------------------------------------------------
>
> Key: CASSANDRA-14743
> URL: https://issues.apache.org/jira/browse/CASSANDRA-14743
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Ariel Weisberg
> Priority: Major
> Fix For: 4.x
>
>
> Right now replica selection for reads with transient replication is purely health based, but it is health based in the sense that it health information is shared across both transient and full ranges. This means you can't use the health information to tell if an endpoint will be faster for a specific query.
> As a heuristic a transient replica will be a little faster for a read compared to a full replica given two endpoints with similar scores so maybe we don't need additional health information.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cassandra.apache.org
For additional commands, e-mail: commits-help@cassandra.apache.org