[
https://issues.apache.org/jira/browse/MESOS-312?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13499021#comment-13499021
]
Benjamin Mahler commented on MESOS-312:
---------------------------------------
Looks like we only offer when there's more than 1 cpu and more than 32MB of
memory available.
https://github.com/apache/mesos/blob/trunk/src/master/hierarchical_allocator_process.hpp#L656
https://github.com/apache/mesos/blob/trunk/src/master/constants.cpp#L29
Seems like both can be problematic, we should revisit this logic and these
constants.
> Mesos never makes offers with only cpus
> ---------------------------------------
>
> Key: MESOS-312
> URL: https://issues.apache.org/jira/browse/MESOS-312
> Project: Mesos
> Issue Type: Bug
> Components: master
> Reporter: Charles Reiss
> Priority: Minor
>
> If a framework already has an executor placed on a machine, it may be able to
> use an offer that contains no or almost no memory but includes cpus. The
> current allocator will never make such offers.
> This can lead to non-intuitive results: If one runs Spark against Mesos
> configuring Spark to use almost exactly the available-to-Mesos memory, Spark
> will start running but eventually stall.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
