accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Eric Newton (JIRA)" <j...@apache.org>
Subject [jira] [Created] (ACCUMULO-3684) balance going too slow
Date Wed, 18 Mar 2015 23:32:38 GMT
Eric Newton created ACCUMULO-3684:
-------------------------------------

             Summary: balance going too slow
                 Key: ACCUMULO-3684
                 URL: https://issues.apache.org/jira/browse/ACCUMULO-3684
             Project: Accumulo
          Issue Type: Improvement
          Components: master
         Environment: large test cluster cluster, hadoop 2, accumulo 1.6.2
            Reporter: Eric Newton
            Assignee: Eric Newton
             Fix For: 1.7.0, 1.6.3


While testing a new balancer on a large cluster, performing a large number of migrations appears
to have been going slowly (20 minutes to migrate 10K tablets). Since this balancer requires
several rounds, the cluster was in transition for couple of hours.

The master queues requests for tablet servers, and then pushes them out in a "flush". Presently,
it will not queue more than 5000 requests *for the entire cluster* so as to not overwhelm
the tservers. But if the cluster has hundreds of servers, they would not be overwhelmed anyhow.

Perhaps we should remove the pause after any intermediary flush.




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

Mime
View raw message