You are viewing a plain text version of this content. The canonical link for it is here.
Posted to notifications@accumulo.apache.org by "Bill Havanki (JIRA)" <ji...@apache.org> on 2014/06/02 15:16:03 UTC

[jira] [Commented] (ACCUMULO-2789) Add a system stress test based on key/cell/row size

    [ https://issues.apache.org/jira/browse/ACCUMULO-2789?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14015362#comment-14015362 ] 

Bill Havanki commented on ACCUMULO-2789:
----------------------------------------

I mentioned this to Wilhelm on Friday, but I'll record it here to avoid forgetting it. It would be beneficial if the write client could be configured to include a limited number of changes per mutation. The current implementation puts an entire row's worth of puts into a single mutation, and when using huge keys or values that could easily take up too much memory.

This update could either be part of this ticket or a follow-on, either is fine with me.

> Add a system stress test based on key/cell/row size
> ---------------------------------------------------
>
>                 Key: ACCUMULO-2789
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-2789
>             Project: Accumulo
>          Issue Type: Improvement
>          Components: test
>            Reporter: Sean Busbey
>            Assignee: William Slacum
>            Priority: Minor
>             Fix For: 1.5.2, 1.6.1, 1.7.0
>
>
> Add a system test that allows stress testing an Accumulo instance in the presence of large cells and rows.
> The test should run a configurable number of ingest and reading clients and allow setting:
> * configurable range of overall key size
> * configurable range of key component sizes
> * configurable value size
> * configurable number of cells per row
> * optional use of isolation on server and/or client
> This kind of test would help us preemptively find issues like ACCUMULO-2671, ACCUMULO-201, and the tserver memory crashes observed with some large mutation sizes while working on YCSB support for Accumulo.
> It will also help us provide test examples of supporting large cardinality on rowids, column families, and visibilities.



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