Yuan Tang

> migrate the v2 implementations to the training operator Are you suggesting moving the entire codebase to training-operator? Or use mpi-operator as a library?

Sounds good

> there are many issues in the training operator (e.g. inconsistent job conditions, not using headless svc, and so on) Can you expand on this? This would be helpful for...

> We leave the individual mpi-operator, and the training-operator uses mpi-operatror as a library. It means that users can deploy MPIJob v2 as either part of the training operator or...

This would be a breaking change but I think it's relatively safe given how few people are using MXNet these days. cc @kubeflow/wg-training-leads

Any concerns of removing this? I don't think we want to continue supporting it.

Does anyone want to pick this up?

I don't think I'll have time to complete this. If you are interested in contributing, feel free to start a new PR.

Great. Thanks!

I'd like to get https://github.com/kubeflow/training-operator/pull/1953 merged as well. I think the risk is pretty low.