4:26 p.m.
Hey everyone,
Mike Turek and I are currently getting back to working on getting
tripleo container build jobs functional for ppc64le. The patch we'd
uploaded [0] got hit with a -2 a while back, which resulted in an email
[1] showing us tooling to properly request and destroy duffy nodes.
The current issue is that this tooling is currently only used on
ci.centos.org, not review.rdoproject.org. When asked in a meeting a
while back, amoralej suggested that the job would best be handled over
on the centos ci side of things, rather than the RDO review page.[2]
However, that would mean that we'd need to spin up a container build job
for centos ci, which is something they currently don't do/haven't done -
it'd be entirely new, and Power specific. On the RDO end of things, we
already have the correct jobs, we theoretically just need to pull in the
builders that allow for us to pull power nodes from Duffy [4] for the job.
There have been test day experiment patches that leverage the CI cloud's
resources and report results review.rdo [3], so it seems that doing
something like this would be possible, the question is whether or not
this would be allowed/if it would ruffle any feathers.
We think it'd be a lot easier to pull a couple builders in on the RDO
end of things rather than stand up a container job over on centos ci,
but for the most part we're just looking for direction, because we've
heard seemingly conflicting information from either side. We don't know
what the 'right' answer is.
Any input would be appreciated!
[0] -https://review.rdoproject.org/r/#/c/13606/
[1] -https://lists.rdoproject.org/pipermail/dev/2018-May/008705.html
[2]
-http://eavesdrop.openstack.org/irclogs/%23rdo/%23rdo.2018-05-30.log.html#t2018-05-30T15:21:21
[3] -https://review.rdoproject.org/r/#/c/10984/
[4] -https://wiki.centos.org/QaWiki/CI/Multiarch
Thanks,
Adam Kimball [baha on freenode]
6:27 p.m.
We think it'd be a lot easier to pull a couple builders in on the
RDO end of
If by "RDO" you mean in RDO Cloud, there are two answers:
1) adding a multiarch computes was not included into design, there's
separate ops team managing RDO Cloud which we would need to consult
and get estimates, I'd expect this to be
2) we actually do not need to add physical nodes, you just need to
provide publicly available openstack cloud account which we could add
as a separate nodepool provider. Is that an option?
things rather than stand up a container job over on centos ci, but
for the
most part we're just looking for direction, because we've heard seemingly
conflicting information from either side.
What is the conflicting info and why is standing up a container job on
ci.centos.org a problem?
As it was pointed out, job is already defined there, you just need to
assign it to ppc64e duffy nodes.
Cheers,
Alan
7:24 p.m.
Hey Alan!
Responses in line, but just to reiterate the main question:
Considering that we need to use duffy, which is preferred?:
1) create new jobs in ci.centos.org (rdo-infra/ci-config) to build
ppc64le containers
- con is that this is not where the other container build jobs live
2) create new jobs in review.rdoproject.org
(rdo-infra/review.rdoproject.org-config)
- con is that this would use duffy vms, which we believe is
unique to the jobs defined here
On 6/21/18 7:27 PM, Alan Pevec wrote:
> We think it'd be a lot easier to pull a couple builders in on
the RDO end of
If by "RDO" you mean in RDO Cloud, there are two answers:ai
We're
more referring to the jobs defined for
review.rdoproject.org/jenkins (where we believe all the container build
jobs are). The couple of builders we're referring to are
'weirdo-get-node' and 'weirdo-release-node' which manage ppc64le nodes
from duffy.
1) adding a multiarch computes was not included into design,
there's
separate ops team managing RDO Cloud which we would need to consult
and get estimates, I'd expect this to be
If we could somehow get the hardware
would this be possible? Either way,
this would more than likely take some time. Our best bet still seems to
be leveraging the existing ppc64le vms available from Duffy
2) we actually do not need to add physical nodes, you just need to
provide publicly available openstack cloud account which we could add
as a separate nodepool provider. Is that an option?
We unfortunately do not have a
publicly available openstack cloud. This
may be possible at some point in the future, but right now we just don't
have the cycles.
> things rather than stand up a container job over on centos ci, but for the
> most part we're just looking for direction, because we've heard seemingly
> conflicting information from either side.
What is the conflicting info and why is standing up a container job on
ci.centos.org a problem?
As far as we know, the container build jobs for x86 exist
exclusively in
review.rdoproject.org/jenkins so spinning up the job would be a little
more work. We'd need to define a containers build job similar to the x86
one, but in a different place, so we'd probably wind up duplicating some
configs from rdo-infra/review.rdoproject.org-config into
rdo-infra/ci-config. Not necessarily a problem, we just want to make
sure we're doing something the community would approve of. Conflicting
info is my fault, as I was originally interpreting the conversations as
jobs belonging in review.rdoproject.org/jenkins but later on hearing
that we they may belong in ci.centos.org instead.
As it was pointed out, job is already defined there, you just need
to
assign it to ppc64e duffy nodes.
Maybe I'm misunderstanding, but the container
build jobs are defined on
review.rdoproject.org and not ci.centos.org.
Cheers,
Alan
_______________________________________________
dev mailing list
dev(a)lists.rdoproject.org
http://lists.rdoproject.org/mailman/listinfo/dev
To unsubscribe: dev-unsubscribe(a)lists.rdoproject.org
Sorry if we're misunderstanding anything!
Thanks!
Mike Turek <mjturek on freenode>
2344
days inactive
2345
days old
2 comments
3 participants
participants (3)
-
Adam Kimball -
Alan Pevec -
Michael Turek