Just my 2 cents here ;)
Several months ago we discussed about this improvement, as I also
noticed this limitation.
While some times later, I have another thought, if framework owner is
concerned about this issue,
and if framework resourceOffers logic notice such condition, by all
easily means, the new design
could apply a thresh hold for resource usage per Mesos agent when
launching task to fix this issue.
One more thing, although no practical per agent resources statistics, I
have a hunch this behavior
will more like system memory usage pattern, i.e. got fragmented over
time, then no need to worry
about the issue in question anymore.
On 2016/4/28 0:41, Guangya Liu wrote:
Hi,
Can anyone help to shepherd MESOS-3765
<https://issues.apache.org/jira/browse/MESOS-3765>? It would be a very
important feature if customer running mesos on some powerful machines.
Thanks,
Guangya
