☂️ Requirements for v0.4

[alculquicondor]

Targeting end of June

Committed

Beta Give feedback

1. Account for terminating pods when doing preemption #510
   kind/feature +1
   
2. WaitForPodsReady: Requeue at the back of the queue after timeout #599
   kind/feature +1
   
3. WaitForPodsReady: a mode where jobs don't block the queue head #610
   kind/feature +1
   
4. Apply LimitRange validation on workloads. #612
   kind/feature +1
   
5. Prevent starvation of large jobs #534
   kind/feature +1
   
6. Support JobSet in Kueue #738
   kind/feature +1
   

Nice to have (will not block the release)

Beta Give feedback

1. Dynamically reclaiming resources #78
   kind/feature priority/important-longterm +2
   
2. Support kubeflow operator #297
   kind/feature lifecycle/frozen +2
   
3. Support for RayJob #666
   kind/feature +1
   
4. Allow partial admission of jobs #420
   kind/feature +1
   

[alculquicondor]

cc @denkensk @ahg-g @mwielgus @kerthcet @tenzen-y

[ahg-g]

are the candidates listed in some rough priority?

[kerthcet]

Thanks for bringing this in! @alculquicondor 🎉

One concern is about the release cycle. This is a long term before we release v0.4.0, hope we can release minor versions regularly, users can experiment with the newest features randomly rather than building with master branch. We haven't released any versions since August last year, I think it's not a good signal. I recalled we talked about this before but couldn't find the records.

[kerthcet]

I will vote for them as they're the shortcomings Kueue has

• WaitForPodsReady: Requeue at the back of the queue after timeout #599
• Prevent starvation of large jobs #534
• Account for terminating pods when doing preemption #510
• Dynamically reclaiming resources #78

Also good to have:

• WaitForPodsReady: a mode where jobs don't block the queue head #610
• Preemption based on resource flavor order #582
• Support dynamically sized (elastic) jobs #77
• Allow partial admission of jobs #420

[alculquicondor]

are the candidates listed in some rough priority?

Not yet. I just grabbed everything from the backlog that might be feasible in 3-4 months (individually).

[mwielgus]

After some chat with @alculquicondor we agreed to aim at 2-3 month release cycle. The project is still young and more frequent, narrow-scoped releases will probably both work better and look better. Thus the list can be roughly prioritised into three categories:

• Will do - we will postpone release if the task is not finished.
• Maybe - we will try/strongly consider but will not postpone the release if the task is not finished.
• For next releases - was considered for the release but most likely the task will not even be started.

Will do:

• Prevent starvation of large jobs #534
• Account for terminating pods when doing preemption #510
• WaitForPodsReady: Requeue at the back of the queue after timeout #599
• WaitForPodsReady: a mode where jobs don't block the queue head #610
• Apply LimitRange validation on workloads. #612

Maybe:

• Dynamically reclaiming resources #78
• Support for RayJob #666 - the work has already been started - [Feature] Support suspend in RayJob ray-project/kuberay#926 but due to some refactoring the important dependency is on-hold
• Support kubeflow operator #297 - depends on Add job suspend semantics kubeflow/common#196

For next releases:

• Support dynamically sized (elastic) jobs #77 - the needed infrastructure will land in 1.27 (Mutable pod scheduling directives kubernetes/kubernetes#116161) and it will be a while before 1.27 is actually in use.
• Preemption based on resource flavor order #582 - nice optimisation but no explicit need at this moment.
• Allow partial admission of jobs #420 - not sure about proper UX, most likely won't fit anyway.
• Add support for budgets #28 - won't fit, too big, no clear API/UX yet.
• Support Argo/Tekton workflows #74 - 2 frameworks already in the pipeline, will wait for the next release(s).

[denkensk]

The project is still young and more frequent, narrow-scoped releases will probably both work better and look better.

By the way, can we collect the cases used by some companies in the production environment, which will encourage more people to use and participate in this project?

[alculquicondor]

For sure! If you know of any particular case, please share. Some of the requests above come from our customers.

[alculquicondor]

I have updated the issue description. @kerthcet @denkensk anything else you would like to add?
I will also share this list in the next WG meeting and in kubecon.

[kerthcet]

LGTM.

[alculquicondor]

I meant to add #534 as well

[alculquicondor]

I added #738 to the committed list

[alculquicondor]

/close
with https://github.com/kubernetes-sigs/kueue/releases/tag/v0.4.0

[k8s-ci-robot]

@alculquicondor: Closing this issue.

In response to this:

/close
with https://github.com/kubernetes-sigs/kueue/releases/tag/v0.4.0

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.