accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Josh Elser (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ACCUMULO-3683) Scanner retry time does not back off or include jitter
Date Thu, 19 Mar 2015 03:22:38 GMT

    [ https://issues.apache.org/jira/browse/ACCUMULO-3683?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14368437#comment-14368437
] 

Josh Elser commented on ACCUMULO-3683:
--------------------------------------

This would obviously have impact outside of just scanner, but it would be good to also look
at classes like MasterClient and ServerClient and include the same jitter/backoff logic there
(I know MasterClient intentionally spins against the master *very* fast, but I still think
it might be excessive in common failure cases).

> Scanner retry time does not back off or include jitter
> ------------------------------------------------------
>
>                 Key: ACCUMULO-3683
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-3683
>             Project: Accumulo
>          Issue Type: Improvement
>          Components: client
>            Reporter: Eric Newton
>            Assignee: Eric Newton
>             Fix For: 1.7.0
>
>
> When a tablet server goes down, scans will begin failing. The {{ThriftScanner}} has a
hardcoded 100ms pause between retries.
> There are two things that would improve this: an exponential back-off, and some jitter,
so the tservers aren't hammering the metadata table with requests for the new location at
the same time the master is trying to re-assign the tablets.



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

Mime
View raw message