Mailing-List: contact commits-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@cassandra.apache.org
Date: Thu, 24 Jul 2014 20:37:38 +0000 (UTC)
From: "Jonathan Ellis (JIRA)" <jira@apache.org>
To: commits@cassandra.apache.org
Message-ID: <JIRA.12703613.1395785329607.39971.1406234258689@arcas>
In-Reply-To: <JIRA.12703613.1395785329607@arcas>
References: <JIRA.12703613.1395785329607@arcas>
Subject: [jira] [Updated] (CASSANDRA-6930) Dynamic Snitch
 isWorthMergingForRangeQuery Doesn't Handle Some Cases Optimally
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


     [ https://issues.apache.org/jira/browse/CASSANDRA-6930?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jonathan Ellis updated CASSANDRA-6930:
--------------------------------------

    Reviewer: Tyler Hobbs

[~thobbs] to review

> Dynamic Snitch isWorthMergingForRangeQuery Doesn't Handle Some Cases Optimally
> ------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-6930
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-6930
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>            Reporter: Tyler Hobbs
>            Assignee: Ala' Alkhaldi
>            Priority: Minor
>              Labels: lhf
>             Fix For: 2.0.10
>
>         Attachments: 9630-v1.txt
>
>
> I was investigating slow responses for queries like {{select * from system.peers}} and noticed that the dynamic endpoint snitch was reporting that the query was _not_ worth merging.  In this case, the local host had a score of 0, so {{return maxMerged < maxL1 + maxL2}} was returning false.  I believe using a {{<=}} condition is the proper fix there.
> Additionally, because scores are looked up three separate times, this method is a prone to race conditions.  I don't think it's worth fixing the race condition for a multi-node scenario, but at least in the single-node case, we can immediately return true and avoid any race conditions that would cause it to erroneously return false.


--
This message was sent by Atlassian JIRA
(v6.2#6252)