accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Keith Turner (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ACCUMULO-3480) GroupBalancer improvements
Date Thu, 22 Jan 2015 16:00:36 GMT

    [ https://issues.apache.org/jira/browse/ACCUMULO-3480?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14287641#comment-14287641
] 

Keith Turner commented on ACCUMULO-3480:
----------------------------------------

Interesting.  I suspect we could reduce time on those two places calling {{TreeMap.containsKey()}}
by using a {{HashSet}} instead of a {{TreeMap}}.

bq. the test failed (more than 1000 migrations were generated)

I looked into this, the assertion in the unit test is too tight.  In the code it adds multiple
moves sometimes and then checks.  So if there were 997 moves and 6 were added, it would see
that 1003 is >= 1000 and stop.  I will look into fixing that.

bq. the test takes a very long time to run (around 45 seconds)

In the test I ran yesterday that took 5.1 secs, I set max migrations to 5000.  When max migrations
is 1000, it takes 14.9 secs.  I changed the {{TreeMap.containsKey()}} to use an existing {{HashMap()}}
and it took 12.9 secs.

> GroupBalancer improvements
> --------------------------
>
>                 Key: ACCUMULO-3480
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-3480
>             Project: Accumulo
>          Issue Type: Bug
>            Reporter: Keith Turner
>            Assignee: Keith Turner
>             Fix For: 1.7.0
>
>         Attachments: ACCUMULO-3480-1.patch
>
>
> [~bfloss] made the following comment on ACCUMULO-3439
> bq. The balance method in GroupBalancer does nothing if there are any incoming migrations.
You probably want to check the incoming migrations and only bail out if any of those migrations
are for the specific table that the GroupBalancer is managing.
> I chatted w/ Brian and he also mentioned that the group balancer should record time at
the end, and not the beginning.  



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

Mime
View raw message