Release 0.4.0

[tenzen-y]

Maybe we want to cut a new mpi-operator release once we have completed the following tasks:

• Elastic resize broken in v2 operator: Horovod requires the slots parameter #445
• Mismatch kube-open api refs #483
• Remove mpi-operator v1 from this repository #486
• adding support for linux-ppc64le in CI  #487
• Support coscheduling plugin #500
• Add suspend semantics #504
• Adapt mpi-operator v2 to Kueue #505 (We will work on the kueue side: Add MPIJob controller kubernetes-sigs/kueue#578)
• Need to add "coordination.k8s.io" and "-leases" to clusterrole #508
• Respect SchedulingPolicy #518
• MPI-Operator: Add docs for the schedulingPolicy website#3453

[tenzen-y]

/cc @terrytangyuan @alculquicondor @gaocegege @zw0610

[terrytangyuan]

Sounds good to me

[ByronHsu]

@tenzen-y @terrytangyuan Wondering what is the estimated release date for this task? Our company depends on mpi-operator v2. I can also help on a few if needed :)

[terrytangyuan]

@tenzen-y @alculquicondor Any estimates on those pending issues? Perhaps @ByronHsu could help some of those.

[tenzen-y]

@ByronHsu We have yet to set a release date for 0.4.0. However, progress has been good.

I can also help on a few if needed

Thanks.
#500 and #518 have almost been completed (https://github.com/tenzen-y/mpi-operator/tree/support-scheduler-plugins).
Also, we can not work on #505 yet since this issue depends on kubernetes-sigs/kueue#360.

However, I'm open to other tasks not mentioned above!

[alculquicondor]

we can leave #505 to the kueue repo as well

[tenzen-y]

As another option, we might be able to include kueue related enhancements after the 0.4.0 release (0.5.0?)

[ByronHsu]

Sounds good! Thanks for the amazing effort!

[tenzen-y]

It would be better to include #521 in MPI Operator v0.4.0.

[mimowo]

Releasing 0.4.0 will help for the Kueue-MPI integration: kubernetes-sigs/kueue#65. With the decision that the integration is happening inside Kueue we need to have a dependency on the mpi-operator. For now, I draft (kubernetes-sigs/kueue#578) the integration using the master of the mpi-operator, so it is not blocking progress, but at some point we need to switch.

cc @alculquicondor @mwielgus

[alculquicondor]

We are pretty much ready for a release.

@terrytangyuan how can we do a release?
I remember we had to upload images, but now I think that's not necessary. Although tags might still be necessary.
What else do we need?

[alculquicondor]

Ah, this also needs to be updated https://github.com/kubeflow/mpi-operator/blob/master/RELEASE.md

@tenzen-y could you take it?

[tenzen-y]

We are pretty much ready for a release.

@terrytangyuan how can we do a release? I remember we had to upload images, but now I think that's not necessary. Although tags might still be necessary. What else do we need?

@alculquicondor We also need to add e2e for the coscheduling plugins (#500) before releasing v0.4.0. So I will update the change log once implementing e2e is done.

[terrytangyuan]

We should release through GitHub Release (in the UI). Yes please update the release notes.

[tenzen-y]

Note that: Probably, we need to create CI pipelines to build example images or manually build those images on our local machine and push the registry before we cut a new release.

[terrytangyuan]

Yep those should be automated. Here's a reference GitHub Action that we can borrow, e.g. docker image push and GitHub release. https://github.com/argoproj/argo-workflows/blob/master/.github/workflows/release.yaml

[tenzen-y]

created a issue: #541

[alculquicondor]

Can we manually create the images for this release?

Are we missing anything else for the release?

[tenzen-y]

Can we manually create the images for this release?

I don't have permission to publish images to Dockerhub, although building images on my locally is possible.

Are we missing anything else for the release?

I'm working on fixing the below bug:

Oh, this is a bug...
I will create a separate PR to fix that.

W0403 20:47:56.968863   15661 podgroup.go:314] Ignore replica "Launcher" priority class "non-existence": priorityclass.scheduling.k8s.io "non-existence" not found
    podgroup_test.go:624: Unexpected calculatePGMinResources for the scheduler-plugins (-want,+got):
          &v1.ResourceList{
        - 	s"cpu":    {i: resource.int64Amount{value: 7}, s: "7", Format: "DecimalSI"},
        + 	s"cpu":    {i: resource.int64Amount{value: 12}, Format: "DecimalSI"},
        - 	s"memory": {i: resource.int64Amount{value: 19327352832}, s: "18Gi", Format: "BinarySI"},
        + 	s"memory": {i: resource.int64Amount{value: 36507222016}, Format: "BinarySI"},
          }
https://github.com/kubeflow/mpi-operator/actions/runs/4601155665/jobs/8128664833?pr=540#step:8:208

#540 (comment)

And also, we might need to create CHANGELOG, as you mentioned.

[alculquicondor]

I do have permissions. Once you give me the green light, I could build and upload.

[tenzen-y]

I do have permissions. Once you give me the green light, I could build and upload.

Great!

Note that to support the multi-architectures, we must specify the platforms when we build the operator image:

$ make images PLATFORMS=linux/amd64,linux/arm64,linux/ppc64le

[alculquicondor]

Also need to run with IMG_BUILDER="docker buildx". However, the base images need some versioning. I'll work on this tomorrow.

[tenzen-y]

Released v0.4.0 🎉

https://github.com/kubeflow/mpi-operator/releases/tag/v0.4.0

Only kubeflow/website#3453 remains.

[tenzen-y]

All tasks are completed!
Thanks to everyone!

/close

[google-oss-prow]

@tenzen-y: Closing this issue.

In response to this:

All tasks are completed!
Thanks to everyone!

/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.