[backend] When will Kubeflow Pipelines (KFP) images be supported for linux/arm64 platform?

[VeniaGao]

Environment

• How did you deploy Kubeflow Pipelines (KFP)?
  Excuted while ! kustomize build example | kubectl apply -f -; do echo "Retrying to apply resources"; sleep 10; done
• KFP version:
  Kubeflow AI reference platform 1.11.0: https://github.com/kubeflow/manifests/releases/tag/v1.11.0
• KFP SDK version:

Steps to reproduce

1、Executed while ! kustomize build example | kubectl apply -f -; do echo "Retrying to apply resources"; sleep 10; done
2、The below Kubeflow Pipelines (KFP) pods failed to be in running status, and the pod container logs show exec format error.

• cache-server-xxx
• metadata-envoy-deployment-xxx
• metadata-grpc-deployment-xxx
• metadata-writer-xxx
• ml-pipeline-xxx
• ml-pipeline-persistenceagent-xxx
• ml-pipeline-scheduledworkflow-xxx
• ml-pipeline-ui-xxx
• ml-pipeline-viewer-crd-xxx
• ml-pipeline-visualizationserver-xxx

3. Tried to pull the images using command docker pull --platform=linux/arm64 ghcr.io/kubeflow/kfp-xxx-xxx:xxx, but got the warning message as following.

WARNING: image with reference ghcr.io/kubeflow/kfp-viewer-crd-controller:2.15.0 was found but its platform (linux/amd64) does not match the specified platform (linux/arm64)

Expected result

The kfp-xxx-xxx images can also support the platform linux/arm64.

Materials and Reference

---

Impacted by this bug? Give it a 👍.

[github-actions]

🎉 Welcome to the Kubeflow Pipelines repo! 🎉

Thanks for opening your first issue! We're happy to have you as part of our community 🚀

Next steps:

• Check out the Contributing Guide for repo-specific guidelines and the Kubeflow Contributor Guide for general community standards
• Join our #kubeflow-pipelines Slack channel
• Attend the Kubeflow Pipelines WG meetings

Feel free to ask questions in the comments.

[ShreyasN707]

Hi @VeniaGao
I might be able to help with this.

From what I understand, the current KFP images on ghcr.io are only published as linux/amd64, so when they’re pulled on an ARM64 Kubernetes cluster the containers fail at runtime with exec format error because the binaries don’t match the CPU architecture.

This looks like it would need enabling multi-arch (amd64 + arm64) builds for all kfp-* images in the release pipelines and publishing them as proper manifest lists.

If that’s the right direction, I’d be happy to take this up and work on a fix — could I get assigned to this issue

[VeniaGao]

Hi @VeniaGao I might be able to help with this.

From what I understand, the current KFP images on ghcr.io are only published as linux/amd64, so when they’re pulled on an ARM64 Kubernetes cluster the containers fail at runtime with exec format error because the binaries don’t match the CPU architecture.

This looks like it would need enabling multi-arch (amd64 + arm64) builds for all kfp-* images in the release pipelines and publishing them as proper manifest lists.

If that’s the right direction, I’d be happy to take this up and work on a fix — could I get assigned to this issue

Hi @ShreyasN707 Thanks for your reply. Yes, that should be the right direction, could you please assign this issue to yourself and help enable multi-arch (amd64 + arm64) builds for all kfp-* images in the release pipelines and publish them as proper manifest lists? Thanks a lot!

[ShreyasN707]

Hi @VeniaGao I might be able to help with this.
From what I understand, the current KFP images on ghcr.io are only published as linux/amd64, so when they’re pulled on an ARM64 Kubernetes cluster the containers fail at runtime with exec format error because the binaries don’t match the CPU architecture.
This looks like it would need enabling multi-arch (amd64 + arm64) builds for all kfp-* images in the release pipelines and publishing them as proper manifest lists.
If that’s the right direction, I’d be happy to take this up and work on a fix — could I get assigned to this issue

Hi @ShreyasN707 Thanks for your reply. Yes, that should be the right direction, could you please assign this issue to yourself and help enable multi-arch (amd64 + arm64) builds for all kfp-* images in the release pipelines and publish them as proper manifest lists? Thanks a lot!

Thankyou for acknowledgement @VeniaGao . I am happy to help with this! Can I please know how do I assign it to myself I am new here.

[VeniaGao]

Hi @VeniaGao I might be able to help with this.
From what I understand, the current KFP images on ghcr.io are only published as linux/amd64, so when they’re pulled on an ARM64 Kubernetes cluster the containers fail at runtime with exec format error because the binaries don’t match the CPU architecture.
This looks like it would need enabling multi-arch (amd64 + arm64) builds for all kfp-* images in the release pipelines and publishing them as proper manifest lists.
If that’s the right direction, I’d be happy to take this up and work on a fix — could I get assigned to this issue

Hi @ShreyasN707 Thanks for your reply. Yes, that should be the right direction, could you please assign this issue to yourself and help enable multi-arch (amd64 + arm64) builds for all kfp-* images in the release pipelines and publish them as proper manifest lists? Thanks a lot!

Thankyou for acknowledgement @VeniaGao . I am happy to help with this! Can I please know how do I assign it to myself I am new here.

@ShreyasN707 Sorry that I don't know either, maybe you can ask help from other contributors? Many thanks.

[ShreyasN707]

Sure @VeniaGao I will try to figure it out once.

[ShreyasN707]

/assign

[VeniaGao]

Thanks a lot! @ShreyasN707 👍🏻