You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@drill.apache.org by "Daniel Barclay (Drill) (JIRA)" <ji...@apache.org> on 2015/05/15 01:19:01 UTC
[jira] [Comment Edited] (DRILL-2039) use forks-per-CPU units in
forkCount
[ https://issues.apache.org/jira/browse/DRILL-2039?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14544584#comment-14544584 ]
Daniel Barclay (Drill) edited comment on DRILL-2039 at 5/14/15 11:18 PM:
-------------------------------------------------------------------------
> The default value of 4 has been working decently for a while
No, actually it hasn't. It has chronically been causing problems for multiple developers here.
Why don't we have the known-reliable value of "1" in the checked-in version, and let people who want to optimize run time risk problems on their own machines? Since they consciously changed forkCount, if they run into problems because of that forkCount setting, they're likely to recognize the problem easily.
Right now, people who want to optimize run time risk causing problems on _other_ people's machines (by having the non-unity value of forkCount checked in). Those other people, not having consciously set forkCount, aren't going to easily recognize that timeouts on their machines are because of the forkCount setting.
was (Author: dsbos):
> The default value of 4 has been working decently for a while
No, actually it hasn't. It has chronically been causing problems for multiple developers here.
Why don't we have the known-reliable value of "1" in the checked-in version, and let people who want to optimize run time risk problems on their own machines? Since they consciously changed forkCount, if they run into problems because of that forkCount setting, they're likely to recognize the problem easily.
Right now, people who want to optimize run time risk causing problems on _other_ peoples machines (by having the non-unity value of forkCount checked in). Those other people, not having consciously set forkCount, aren't going to easily recognize that timeouts on their machines are because of the forkCount setting.
> use forks-per-CPU units in forkCount
> ------------------------------------
>
> Key: DRILL-2039
> URL: https://issues.apache.org/jira/browse/DRILL-2039
> Project: Apache Drill
> Issue Type: Improvement
> Components: Tools, Build & Test
> Reporter: Daniel Barclay (Drill)
> Assignee: Hanifi Gunes
> Priority: Minor
> Fix For: 0.9.0
>
> Attachments: DRILL-2039.1.patch.txt, DRILL-2039.2.patch
>
>
> Can Surefire's fork count be set using units of forks per CPU instead of the absolute number of forks (i.e., {{<forkCount>1C</forkCount>}} instead of {{<forkCount>4</forkcount>}})?
> That would let developers with more powerful machines still run tests in parallel while letting others avoid or reduce test timeout errors without the risk of forgetting to re-edit pom.xml after every switch to a different git branch.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)