TFJob 1.0

[richardsliu]

Tracking items needed for TFJob 1.0:

• Create API: Create TFJob v1 API and controller from v1beta2 #989
• Adding Prometheus support: Prometheus support in TF Job #988
• Documentation: Create TFJob v1 documentation #990
• Kustomize files: Update kustomize files for tf-operator v1 #991

/cc @gaocegege /cc @johnugeorge

[johnugeorge]

Sounds good

[gaocegege]

SGTM

[johnugeorge]

Performance issues: #965

[k82cn]

I can help on the "Scheduling policy" to integrate with related scheduler :)

BTW, should "Scheduling policy" also be part of common operator?

[richardsliu]

I've included SchedulingPolicy in the common API here: https://github.com/kubeflow/common/blob/master/operator/v1/types.go#L184

[richardsliu]

Discussed with community members - for TFJob 1.0, we will not be migrating to the new common API until the latter has been stabilized and tested with a new operator.

Instead the goal will be to release v1 from the v1beta2 implementation, with smaller improvements like enhancing metrics.

/cc @johnugeorge
/cc @gaocegege

[jlewi]

Update from TFJob 1.0
TFJob 1.0 API is done
But not for PyTorch

[johnugeorge]

@jlewi I will complete it in 2 days

[richardsliu]

We can close this after #965 is fixed.

[jlewi]

Regarding the UI bug: #1019. Should we get eventually rid of or keep the TFJobs dashboard? Would it be better to have a more general UI that isn't specific to TFJob? How far can we get with existing monitoring tools like the K8s dashboard?

Should we open up a separate issue to figure out the future of the TFJobs dashboard?

[johnugeorge]

Related slack conversation: https://kubeflow.slack.com/archives/C985VJN9F/p1560184440012500

[richardsliu]

/close

[k8s-ci-robot]

@richardsliu: Closing this issue.

In response to this:

/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.