Deprecate kubeflow/training related ECR repos

[PatrickXYS]

What steps did you take and what happened:
As a post item of deprecating optional-test-infra, we'll deprecate kubeflow/training ECR repos provided by optional-test-infra.
Serves as sub-issue of kubeflow/testing#1008

What did you expect to happen:
Deprecate kubeflow/training ECR repos provided by optional-test-infra

Private ECR Repo Lists:

kubeflow/training related: 809251082950.dkr.ecr.us-west-2.amazonaws.com/

• pytorch-operator
• tf-operator
• training-operator

Public ECR Repo Lists:
kubeflow/training related: public.ecr.aws/j1r0q0g6/training

• tf-operator
• training-operator

@kubeflow/wg-training-leads I believe those repos are not used in deployment, please confirm those repos are not in-use, if not, let's find alternatives by July.

[johnugeorge]

kubeflow/testing#1008 (comment)

[thesuperzapper]

@johnugeorge @kubeflow/wg-training-leads I have made a script to migrate all the Notebook WG images from ECR to GitHub Container Registry.

I will use it to migrate the PUBLIC training ones ASAP (so we don't lose access to the old images):

• public.ecr.aws/j1r0q0g6/training/training-operator --> ghcr.io/kubeflow/training/training-operator
• public.ecr.aws/j1r0q0g6/training/tf-operator --> ghcr.io/kubeflow/training/tf-operator

NOTE: I will only be migrating the PUBLIC existing images, the @kubeflow/wg-training-leads will still need to:

1. migrate their docker build CI/CD process
2. migrate any private images (if they care about them)
3. update any manifests to use the new image location
4. communicate with users that old images will FAIL TO PULL after next month

[thesuperzapper]

@johnugeorge the mirroring is now finished, you can see them at:

• ghcr.io/kubeflow/training/training-operator
• ghcr.io/kubeflow/training/tf-operator

For now they are just a one-time mirror from ECR (in the interest of not losing the old images if ECR dies).
If you want to start using them going forwards, I am happy to run you through how it might work.

NOTE: We still need a kubeflow org admin to associate them to the kubeflow/training-operator repo, so they show up under "packages":

[PatrickXYS]

Looks like currently only training and notebooks WGs are actively using ECR repos for production deployment.

I can hold for training and notebooks repos until the end of July.

[PatrickXYS]

There are public ECR repos and Private ECR repos, I don't think training WG are using private ECR repos:

809251082950.dkr.ecr.us-west-2.amazonaws.com/pytorch-operator
809251082950.dkr.ecr.us-west-2.amazonaws.com/tf-operator
809251082950.dkr.ecr.us-west-2.amazonaws.com/training-operator

Can you help confirm? @johnugeorge Thanks!

[github-actions]

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.

[github-actions]

This issue has been automatically closed because it has not had recent activity. Please comment "/reopen" to reopen it.