kueue
kueue copied to clipboard
Support kubeflow operator
What would you like to be added: Support kubeflow training operator.
Why is this needed: It is to track the status of kueue to support kubeflow training operator.
- [ ] https://github.com/kubeflow/common/pull/196
- [ ] #65
Note that MPIJob latest version is not currently part of the training-operator https://github.com/kubeflow/training-operator/issues/1479
The Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs.
This bot triages issues and PRs according to the following rules:
- After 90d of inactivity,
lifecycle/stale
is applied - After 30d of inactivity since
lifecycle/stale
was applied,lifecycle/rotten
is applied - After 30d of inactivity since
lifecycle/rotten
was applied, the issue is closed
You can:
- Mark this issue or PR as fresh with
/remove-lifecycle stale
- Mark this issue or PR as rotten with
/lifecycle rotten
- Close this issue or PR with
/close
- Offer to help out with Issue Triage
Please send feedback to sig-contributor-experience at kubernetes/community.
/lifecycle stale
/lifecycle frozen
This is currently blocked on https://github.com/kubeflow/common/pull/196
/assign
As a first step, I opened a PR to add the PyTorchJob support, and then I will add the following framework support:
- TFJob
- MXJob
- XGboostJob
- PaddleJob
Also, I'm on the fence if we should support MPIJob v1 hosted only on kubeflow/training-operator (currently, MPIJob v2 hosted only on kubeflow/mpi-operator)
Regarding MPIJob v1 wdyt? @alculquicondor @mimowo @kerthcet @trasc
I'm ok leaving it out if it's not trivial to support 2 API versions. I think the CRD objects themselves are not compatible.
I think the CRD objects themselves are not compatible.
Right.
I'm ok leaving it out if it's not trivial to support 2 API versions.
We can not support v1 and v2 API by a single controller: https://github.com/kubernetes-sigs/kueue/tree/a103723023aa6c5a63cc8c1248fd38d8640d7003/pkg/controller/jobs/mpijob.
However, once we implement a separate controller for v1 like https://github.com/kubernetes-sigs/kueue/blob/3589969054023cb8b584a4639f4b9dec8c371a67/pkg/controller/jobs/kubeflow/jobs/pytorchjob/pytorchjob_controller.go, we can support v1.
Anyway, I think MPIJob v1 is a lower priority since we already support MPIJob v2.
+1 to defer the work unless we receive strong demands.
+1 to defer the work unless we receive strong demands.
I agree.
Tasks:
- [x] PyTorchJob: https://github.com/kubernetes-sigs/kueue/pull/995
- [x] TFJob: https://github.com/kubernetes-sigs/kueue/pull/1068
- [x] XGBoostJob: https://github.com/kubernetes-sigs/kueue/pull/1114
- [x] MXJob: https://github.com/kubernetes-sigs/kueue/pull/1183
- [x] PaddleJob: https://github.com/kubernetes-sigs/kueue/pull/1142