cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Stu Hood (JIRA)" <>
Subject [jira] Commented: (CASSANDRA-1817) Dynamic Read Repair
Date Thu, 23 Dec 2010 08:06:00 GMT


Stu Hood commented on CASSANDRA-1817:

> This could have a large impact on performance,
The level of dynamic read repair would actually make a really good metric for inconsistency,
so exposing it properly would be a requirement for this issue to be a success. We wouldn't
want to be varying the read repair chance inside of a black box: the fact that it was happening
more frequently would be an important thing for the operator to know, since it might indicate
problems with the cluster and affect performance.

> I kind of think that for workloads where RR < 1.0 makes sense, it makes sense even
when actual repairs are happening.
Then I would propose that 1) our current RR configurable become the cap for the dynamic algorithm,
2) we add a hardcoded minimum chance of 1-5%, 3) we enable dynamic read repair by default.

> Dynamic Read Repair
> -------------------
>                 Key: CASSANDRA-1817
>                 URL:
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Core
>            Reporter: Stu Hood
>             Fix For: 0.7.1
> Read repair could (temporarily) adjust its own frequency (from the baseline) based on
the necessity of the repair for particular nodes. For example, a successful read repair (caused
data to be repaired) should bump the frequency for the node that needed repair, with the goal
that a node that has been offline for a while should trend toward 100% read repair while it
is successful.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message