aurora-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Brian Wickman <wick...@apache.org>
Subject Re: thermos observer resource monitoring
Date Tue, 10 Feb 2015 19:10:15 GMT
Interesting -- I hadn't considered per-process stats.  Would a cli command
like 'thermos top' be sufficient in this case?  I'm just wondering if this
is an "always on" need or if a stat-a-la-carte solution would work.

On Mon, Feb 9, 2015 at 4:47 PM, Steve Niemitz <steve@tellapart.com> wrote:

> While we don't rely on the stats, I do think that they are nice to have.
> Mesos doesn't export stats down to the per-process level, so we lose some
> resolution there.
>
> That being said, currently we run about 1/2 our jobs in docker, which
> doesn't have per-process stats in the observer right now, so we're already
> running without them now anyways.
>
> What about the runners themselves exporting stats for the processes they're
> running, and the observer reading those?
>
> On Mon, Feb 9, 2015 at 7:34 PM, Zameer Manji <zmanji@apache.org> wrote:
>
> > Would removing this also remove the psutil dependency from the observer?
> >
> > On Mon, Feb 9, 2015 at 3:59 PM, Brian Wickman <wickman@apache.org>
> wrote:
> >
> > > Does anybody rely upon the resource monitoring feature in the Thermos
> > > observer UI (the UI on port :1338)?  Specifically the cpu/ram/disk
> > columns.
> > >
> > > As far as I know, nobody is using Thermos in the absence of Mesos, and
> > > because the Mesos slave provides the same information more accurately,
> > I'm
> > > inclined to remove this feature from the UI.  This will simplify the
> code
> > > quite a bit and also reduce the resources consumed by the observer as
> > well
> > > (we find it burns 2-4% CPU per box in practice.)
> > >
> > > ~brian
> > >
> > > --
> > > Zameer Manji
> > >
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message