Hi, Karanbir and CentOS community.
I am starting this thread to understand what is needed to enable ppc64le architecture in ci.centos.org, be it hardware, software, or any roadblock. As soon as we understand all the requirements, I will be pleased to work with you to sort them out.
Ideally speaking, we would like to have all relevant checks and validations performed against amd64 to be executed on ppc64le as well, for all the relevant SIGs.
We also would like to see docker and cloud images, in general, being created and validated for ppc64le as part of CentOS pipelines.
As of 7.3, CentOS on ppc64le became a lot easier, meaning, for example, that nested qemu/KVM virtualization works out of the box; and creating and running docker images in CentOS guest works just fine.
To get the ball rolling, below is a short list of needed changes to create, run, and test docker images on CentOS 7.3:
- https://github.com/rhinstaller/lorax/pull/149 - https://github.com/virt-manager/virt-manager/commit/cd35470e3c55aa64976cd0a9... - https://github.com/CentOS/sig-cloud-instance-build/pull/65 - https://github.com/CentOS/sig-core-t_docker/pull/2
I am looking forward to work with you and define our next steps together.
Thank you!
Hi Murilo,
This might be better to host on an office hours session perhaps, and post followup notes to this thread - the ask from your side is quite vast, and email might take a while to get traction. But a few pieces of info here :
On 23/01/17 20:14, Murilo Opsfelder Araújo wrote:
Hi, Karanbir and CentOS community.
I am starting this thread to understand what is needed to enable ppc64le architecture in ci.centos.org, be it hardware, software, or any roadblock. As soon as we understand all the requirements, I will be pleased to work with you to sort them out.
Welcome!
Ideally speaking, we would like to have all relevant checks and validations performed against amd64 to be executed on ppc64le as well, for all the relevant SIGs.
This would be a fantastic goal to aim for. The CentOS CI infrastructure is evolving at the moment, and this is certainly a good time to get involved in that process.
There are a few pieces we likely want to address alongside and make sure we cover the entire work flow, from code control to build and test through to release.
There are a few SIGs already working against ppc targets, and we could perhaps use those as a model to work end to end with.
We also would like to see docker and cloud images, in general, being created and validated for ppc64le as part of CentOS pipelines.
As of 7.3, CentOS on ppc64le became a lot easier, meaning, for example, that nested qemu/KVM virtualization works out of the box; and creating and running docker images in CentOS guest works just fine.
The docker images etc might be a good place to start from, but there is no infra or process in place to handle these in a sane manner from our side. So I think the first steps we should likely kick off with is making sure we have the ppc64 and ppc64le targets available and setup in cbs.centos.org while we talk about how and where and when we can work on the other bits.
eg. even if we were building the cloud images, there is no place to test them.
w.r.t the docker images, I think it might be good to engage into the CentOS Container Pipeline and handle it that way.
regards,
On 01/23/2017 09:59 PM, Karanbir Singh wrote:
Hi Murilo,
Hi, Karanbir.
Thanks for replying me. And pardon me if I am flooding you with too many questions, I just want to have topics organized in my mind.
This might be better to host on an office hours session perhaps, and post followup notes to this thread - the ask from your side is quite vast, and email might take a while to get traction. But a few pieces of info here :
Definitely. I will get in touch with you in one of the office hours sessions ahead. Meanwhile, I think we can keep discussions here as well -- web archives are easier to search than IRC logs.
[...]
This would be a fantastic goal to aim for. The CentOS CI infrastructure is evolving at the moment, and this is certainly a good time to get involved in that process.
That is great. Can you share more details about what is changing, e.g. moving to new facility, hardware upgrade, etc?
[...]
The docker images etc might be a good place to start from, but there is no infra or process in place to handle these in a sane manner from our side.
I would like to understand the details behind infra and process.
As to infra, is it a collocation issue, i.e. no room for more servers? Or is it an issue of lack of POWER hardware? The latter seems to be easier to get addressed.
As to process, is it something that is already being worked by any SIG? Perhaps SIG Core or SIG Cloud? Who is the owner of this process?
So I think the first steps we should likely kick off with is making sure we have the ppc64 and ppc64le targets available and setup in cbs.centos.org while we talk about how and where and when we can work on the other bits.
It is not clear to me what is the relationship between hosts in cbs.centos.org and ci.centos.org.
My understanding is that cbs hosts are only used by koji to build packages for the cbs system; and that ci hosts are used to run the Jenkins jobs, which, eventually, allocate and release machines from the Duffy pool via admin.ci.centos.org. Did I get it correctly?
I can see that 2 ppc64 nodes are setup for cbs (http://cbs.centos.org/koji/hosts). Is the plan to setup more ppc64 nodes on cbs.centos.org?
eg. even if we were building the cloud images, there is no place to test them.
My vision is that we can mimic some of the amd64 jobs, e.g. https://ci.centos.org/view/CentOS-Core-QA/job/CentOS-Core-QA-t_docker-c6-64/, to build and test docker images on ppc64le as well.
With CentOS 7.3, it is now possible to build, run, and test docker images in CentOS guest.
w.r.t the docker images, I think it might be good to engage into the CentOS Container Pipeline and handle it that way.
That would also require ppc64 hardware to run the CentOS Container Pipeline jobs.
Is there any room to creating more ppc64le guests and adding them to ci.centos.org?
Thanks!