Re: pcp updates: pmdadocker

Hi Marko,

----- Original Message -----
> On 2016-12-09 22:39, Lukas Berk wrote:
> > 
> > If I could get a review on my pmdadocker branch it'd be most
> > appreciated.  More qa needs to be added with a fake http request/reply
> > but the jist of the pmda is there.
> 
> This is certainly an interesting area of work! Few higher level questions:
> 
> - Is this PMDA intended to be purely for Docker-based containers also in
> the long run or could it be expanded to be a generic pmdacontainer to
> support alternatives like rkt and OCI as well? Since there is a growing
> momentum to avoid fragmentation and standardize in this area (see
> https://www.opencontainers.org/) perhaps it would be wise for PCP to be
> format/vendor neutral in this area, too?

pmdaroot fills that niche FWIW, and supports both LXC and Docker (and can
support Rocket easily too).  The domain of the docker PMDA should be just
docker though (IMO), since its talking to the docker daemon directly.

> - How do you envision this PMDA to work with container management
> systems like OpenShift / Kubernetes or Mesos and how it would
> intermingle with concepts like pods where there is 1-N containers per
> pod? Are there perhaps plans to provide higher level PCP components

Sounds like a different domain, possibly for other PMDAs or for core PCP
extension...?  (ala --containers)  Not sure, noone has really done this
level of analysis yet and come up with a viable plan here.

> which would better match to container management systems' abstraction
> levels (see also the next question)?
> 
> - Since containers / pods are most often considered ephemeral and there
> may be N replicas of a pod running at any time, is this kind of dynamics
> something in scope for PCP to be aware of?

Absolutely.  At the very least, we would want metrics similar to the
hinv.map metrics which would allow the high level subsystems (Kubernetes,
et al) to to be mapped to the lower-level containers they manage.

> I mean, for example, if there
> are N replicas of a pod which each happen to contain M containers, and
> I'd want to get an overview of them, how would this work with PCP?

Someone would need to think that through and come up with a feasible
design, and some code.  Possibly using labels to sensibly distinguish
the values from each other?  ;)  Maybe, probably ... dunno, I haven't
done that level of analysis myself.

> would you say that this is the intent or my imagination only or merely
> the current state of affairs which is aimed to be improved during the
> next steps on this front?

The latter, I think - PCP should continue to evolve to meet demand.

cheers.

--
Nathan

-=-=-=-=-=-=-=-=-=-=-=-
pcp mailing list
[email protected]
https://groups.io/g/pcp/messages
-=-=-
Groups.io Links:

You receive all messages sent to this group.

View/Reply Online (#14887): https://groups.io/g/pcp/message/14887
View All Messages In Topic (3): https://groups.io/g/pcp/topic/3780835
Mute This Topic: https://groups.io/mt/3780835?uid=174580
New Topic: https://groups.io/g/pcp/post

Change Your Subscription: https://groups.io/g/pcp/editsub?uid=174580

Group Home: https://groups.io/g/pcp
Contact Group Owner: [email protected]

Terms of Service: https://groups.io/static/tos

Unsubscribe: https://groups.io/g/pcp/leave/354243/563757577/xyzzy
-=-=-=-=-=-=-=-=-=-=-=-