Unify the kubeflow project image location

[chenzhiwei]

Kubeflow project includes many docker images, but they are in different locations.

Can we put them into a unified location?

And we can also use same logic to build and publish images, such as Makefile.

This will make the images are easy to maintain.

There are following works and I am working on first one:

1. Unify the image dockerfiles

   Makes the dockerfiles can build both amd64 and ppc64le docker images.

   • refactor the dockerfile training-operator#893
   • update dockerfile and build steps pipelines#562
   • update dockerfile and add build step of frontend pipelines#567
   • build backend in travis ci training-operator#892

2. Unify the image build logics

   • make build-image to build image
   • make build-image-ppc64le to build ppc64le image
   • make push-image to push image to registry
   • make mirror-image to mirror image

3. Unify the image locations

[jlewi]

Thanks for helping out.

Can you elaborate on what you are working on? Which Dockerfiles are you planning on refactoring?

[chenzhiwei]

@jlewi I just updated the issue to include the related PRs, and will also include new PRs when available. My plan is first to make all docker images are buildable by makefile, then unify the location. There are two alternatives: docker.io/kubeflow and gcr.io/kubeflow-images-public.

[jlewi]

I'd like to support both docker and gcr.io and possibly others. It would be great to setup the ability to automatically mirror our docker images across registries.

There's been a long standing issue to fix this. see #211. Would love to find someone willing to drive that.

[gyliu513]

@dreamryx

[dreamryx]

I will continue the work. The dockerfiles of related image are list as below:

image/dockerfile	repo
centraldashboard	kubeflow
ambassador	ambassador-envoy
jupyterhub-k8s	kubeflow
vizier-core	katib
vizier-core-rest	katib
studyjobcontroller	katib
metricscollector	katib
suggestion-random	katib
suggestion-grid	katib
suggestion-hyperband	katib
suggestion-bayesianoptimization	katib
katib-ui	katib
metacontroller	kubeflow
pytorch-operator	pytorch-operator

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[gyliu513]

we are still working for this.

[jlewi]

@gyliu513 What are the next steps?

[gyliu513]

@jlewi we are now working for multi-arch support at #2633 , and will resume this after #2633 finished, hope it is OK.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.