OKD in ARM platform

[SaravanaStorageNetwork]

Support for OKD to setup in ARM platform:

When can one expect the same? Any issues in enabling it in OKD?

This project ocp-on-arm tracks mainly the downstream bits.

[vrutkovs]

/assign @LorbusChris

[LorbusChris]

This is currently blocked by the fact that we do not yet have a functioning arm64 build farm in our OpenShift Prow instance. See https://issues.redhat.com/browse/DPTP-2718 for more details. The issue on the agenda for the Testing/CI Architecture call next tuesday, too, in case you want to join in.

[0xHexE]

@LorbusChris, is arm64 supported now in OKD?

[0xHexE]

I have a cluster with existing AMD64 machines, and I want to add ARM64 machines. The machine is getting provisioned, but not as a node.

[LorbusChris]

ARM64 is not yet supported in OKD unfortunately.

[0xHexE]

https://docs.okd.io/latest/installing/installing_aws/installing-aws-customizations.html#installation-configuration-parameters-optional_installing-aws-customizations

I think they support it, just not cross-architecture compute.architecture

[aleskandro]

https://docs.okd.io/latest/installing/installing_aws/installing-aws-customizations.html#installation-configuration-parameters-optional_installing-aws-customizations

I think they support it, just not cross-architecture compute.architecture

Hi @0xHexE yes, having an ARM-only cluster and a heterogeneous cluster (x86 + ARM nodes) are two different things.

But, unfortunately, OKD on ARM is not supported yet, as @LorbusChris stated above.

We need to fix the doc removing ARM from the supported architectures in OKD until we unblock DPTP-2718.

[anthr76]

Are there any update to this or any ways the community can help pitch in?

[AlexanderWurz]

would also be interessted in OKD on arm - mainly running on aws on graviton cpus, any update on that topic is appreciated

[ci-operator]

@aleskandro It appears that DPTP-2718 is now unblocked, closed complete, and the project discourse has moved to an epic shared between the teams collaborating on this - however, there are no links to this new collaboration to be found. What is the current status, and where can it now be followed?

[JaimeMagiera]

We will be discussing this at Tuesday's OKD Working Group meeting.

[kalik1]

TL;DR 1st quarter 2023

https://youtu.be/T42ri28tnYo?t=2297

@LorbusChris said: [From YT subs] so it is getting closer, let me put it that way: Alessandro who is our arm QE lead for openshift he's been a regular I think on this meeting too, he's been working on it and we essentially just need to get a compute on arm to be able to build on arm. he has a pipeline that actually creates a fully multi-arch manifest listed payloads and he has built it. Once it works we just essentially right now need to find resources to deploy that, and build it regularly so it is very close and yeah hopefully in the in the first quarter this year.

Thanks @JaimeMagiera @LorbusChris for sharing the updates :)

[gogeof]

We are also very interested in this feature. It is now the second quarter, and we are wondering if OKD already supports arm64 in the new version. Thank you.

[celevra]

+1 we've got a arm cluster with 44 nodes, need to switch to OKD

[dreaminghk]

+1 waiting for the ARM support

[japhar81]

@LorbusChris @JaimeMagiera @kalik1 since we're pretty well past Q1.. any chance we could get an update on ARM support?

[smuda]

After upgrading to 4.14.0-0.okd-2023-12-01-225814 to multi-arch (with oc adm upgrade --to-multi-arch) the only upgrade path is to 4.14.0-0.okd-2023-12-01-225814 which seems to be single-arch?

% oc adm upgrade
Cluster version is 4.14.0-0.okd-2023-12-01-225814

Upstream: https://amd64.origin.releases.ci.openshift.org/graph
Channel: stable-4

Recommended updates:

  VERSION                        IMAGE
  4.14.0-0.okd-2023-12-01-225814 registry.ci.openshift.org/origin/release@sha256:4510c50834c07e71521a68c021320b5e44978245c8031a858f74328e54e89e1c

Upgrading to the same version via OC does not work:

% oc adm upgrade --to=4.14.0-0.okd-2023-12-01-225814
info: Cluster is already at version 4.14.0-0.okd-2023-12-01-225814

However, upgrading via the console works, even though progress does not show in console or with oc adm upgrade.

Once that upgrade has completed, I'm able to upgrade to 4.14.0-0.okd-2024-01-06-084517 but it seems to not be multi-arch anymore.

% oc adm upgrade                                                    
Cluster version is 4.14.0-0.okd-2023-12-01-225814

Upstream: https://amd64.origin.releases.ci.openshift.org/graph
Channel: stable-4

Recommended updates:

  VERSION                        IMAGE
  4.14.0-0.okd-2024-01-06-084517 registry.ci.openshift.org/origin/release@sha256:c4a6b6850701202f629c0e451de784b02f0de079650a1b9ccbf610448ebc9227

Perhaps this has something to do with the channel? According to the docs the channel is supposed to be changed to stable-4.14 but that's probably a OpenShift thing in the docs?

Or perhaps I'm in uncharted waters and upgrading to multi-arch wasn't even supposed to work at all :-)

[japhar81]

@smuda its a red herring.. that multi-arch upgrade doesnt actually do anything.. I think we're all still waiting for a release that supports it..

[codespearhead]

https://docs.okd.io/latest/installing/installing_aws/installing-aws-customizations.html#installation-configuration-parameters-optional_installing-aws-customizations

I think they support it, just not cross-architecture compute.architecture

You have to have a cluster with multi-architecture compute machines if you want to run ARM worker nodes, because the control plane nodes (master nodes) cannot be ARM.

As per this issue, multi-arch compute machines became GA in Openshift v4.14, which was released 5 months ago.

However, it's worth noting that some things still don't work in this type of cluster, hence the disclosure:

In my opinion, this issue should be renamed to:

Support cluster with multi-architecture compute machines

Can you guys check whether this issue should be closed now that we have OKD v4.15?

[japhar81]

@codespearhead nope -- it works in OpenShift, but not in OKD. Even at 4.15, I can't do multi-arch..