[jira] [Comment Edited] (MESOS-7197) Requesting tiny amount of CPU crashes master

["Benjamin Bannier (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MESOS-7197?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15891917#comment-15891917 ] 

Benjamin Bannier edited comment on MESOS-7197 at 3/2/17 9:33 AM:
-----------------------------------------------------------------

[~bmerry]: I was able to reproduce this; in an optimized build created with clang-trunk I see a critical check failure even for {{cpus:0.001;mem:1}}.

[~neilc]: These permille numbers appear suspiciously close to the edge of what we ignore for fixed point resource math. Is this a bug in the fixed point math or are we just missing validation?


was (Author: bbannier):
[~bmerry]: I was able to reproduce this, and with and an optimized build created with clang-trunk shows a critical check failure even for {{cpus:0.001;mem:1}}.

[~neilc]: These permille numbers appear suspiciously close to the edge of what we ignore for fixed point resource math. Is this a bug in the fixed point math or are we just missing validation?

> Requesting tiny amount of CPU crashes master
> --------------------------------------------
>
>                 Key: MESOS-7197
>                 URL: https://issues.apache.org/jira/browse/MESOS-7197
>             Project: Mesos
>          Issue Type: Bug
>          Components: allocation
>    Affects Versions: 1.1.0, 1.2.0
>         Environment: Ubuntu 14.04, using Mesosphere PPA to install Mesos
>            Reporter: Bruce Merry
>            Priority: Critical
>
> If a task is submitted with a tiny CPU request e.g. 0.0004, then when it completes the master crashes due to a CHECK failure:
> {noformat}
> F0302 10:48:26.654909 15391 sorter.cpp:291] Check failed: allocations[name].resources[slaveId].contains(resources) 
> {noformat}
> I can reproduce this with the following command:
> {noformat}
> mesos-execute --command='sleep 5' --master=$MASTER --name=crashtest --resources='cpus:0.0004;mem:128'
> {noformat}
> If I replace 0.0004 with 0.001 the issue no longer occurs.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)