Thx Joris. Would wait for Ben M's comments. The reasoning is described in the
blog because we witnessed it first hand running hundreds of frameworks (jenkins
in this case). Ben H. was in favor of normalizing the counter and make it a
level playing field as new frameworks joined and just resetting it to 0 for all
was a quick/easy fix.
Thx
From: Joris Van Remoortere <[email protected]>
To: "[email protected]" <[email protected]>; meghdoot bhattacharya
<[email protected]>
Cc: Benjamin Mahler <[email protected]>
Sent: Thursday, June 16, 2016 4:43 AM
Subject: Re: Multiple framework DRF tuning
Hey Meghdoot,
Mesos-4687 does indeed fix the precision math problem. From the JIRA you
can see it is fixed in 0.28.0, 0.27.2, 0.26.1, 0.25.1, 0.24.2.
For #1 it doesn't look like an allocation count reset was ever merged.
I've CCd BenM so we can discuss whether there are any bad consequences from
doing this.
—
*Joris Van Remoortere*
Mesosphere
On Wed, Jun 15, 2016 at 8:33 PM, meghdoot bhattacharya <
[email protected]> wrote:
> Resending with fixed links.
>
> I wanted to follow up on 2 issues that we discussed few years back on the
> below blog described in this section Mesos delayed offers to frameworks
>
>
> http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/
>
>
>
> We had sample commits back then for private build. and I am sure code has
> changed drastically since.
>
>
> 1. Reset allocation counter across all frameworks when new frameworks join
> (some sort of normalization)
> https://github.com/asathye/mesos/commit/b7ad40dc254390751069f55a6756efcf7d1bb9aa
>
>
>
>
> 2. Precision math not resulting in 0 share value.
>
> https://github.com/asathye/mesos/commit/56c144f8cf48bb14c3f0fb639e9791170c1b6267
>
>
>
>
>
> Has these been addressed in some form? Does Mesos-4687 address point 2
> above?
>
>
> Thx
>
> From: Joris Van Remoortere <[email protected]>
> To: "[email protected]" <[email protected]>; meghdoot bhattacharya
> <[email protected]>
> Sent: Wednesday, June 15, 2016 12:47 AM
> Subject: Re: Multiple framework DRF tuning
>
> Your e-mail seems truncated or mangled in some way (at least for me). Could
> you please resend it?
>
> —
> *Joris Van Remoortere*
> Mesosphere
>
> On Tue, Jun 14, 2016 at 10:00 PM, meghdoot bhattacharya <
> [email protected]> wrote:
>
> >
> >
> > I wanted to follow up on 2 issues that we discussed few years back on the
> > below blog described in this section Mesos delayed offers to frameworks
> >
> >
> >
> http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/
> >
> >
> >
> > We had sample commits back then for private build. and I am sure code has
> > changed drastically since.
> >
> >
> > 1. Reset allocation counter across all frameworks when new frameworks
> join
> > (some sort of normalization)* ben's fix for allocations
> > https://issues.apache.org/jira/browse/MES… · asathye/mesos@b7ad40d
> >
> >
> > 2. Precision math not resulting in 0 share value.
> > * Set double precision at 0.000001 · asathye/mesos@56c144f
> >
> >
> >
> > Has these been addressed in some form? Does Mesos-4687 address point 2
> > above?
> >
> >
> >
> > Thx
> >
> >
> > |
> > |
> > |
> > | | |
> >
> > |
> >
> > |
> > |
> > | |
> > * Set double precision at 0.000001 · asathye/mesos@56c144f
> > mesos - Mirror of Apache Mesos | |
> >
> > |
> >
> > |
> >
> >
> >
> >
> >
> >
> > |
> > |
> > |
> > | | |
> >
> > |
> >
> > |
> > |
> > | |
> > * ben's fix for allocations https://issues.apache.org/jira/browse/MES…
> ...
> > …OS-1086 * reset allocations on new framework addition | |
> >
> > |
> >
> > |
> >
> >
> >
>
>
>
