☂️ Release 0.17 plan

[mimowo]

The release is planned tentatively for 16th March, let's discuss the scope.

Nice to haves:

• TAS: support for ElasticWorkloads #8160
• Concurrent Admission (Alpha) #8691
• MultiKueue: prevent starting preemptions in multiple worker clusters #8303
• TAS: support ResourceTransformations #8860
• Fix throughput issues at scale for BestEffortFIFO CQs with CPU and GPU flavors #8095
• Workload Scheduling Constraints (Workload-Level) and Preference-Aware MultiKueue Dispatching #8729
• Feature request: Support configuring Kueue to suspend a Job till told otherwise on a per job basis #6915
• Graduate LocalQueue metrics to Beta #8830
• KEP-2941: Add extended resources design for DRA integration #8734
• Add a mechanism to consider preemption cost when finding preemption candidates #7990
• Provide a way to specify in the configuration what resources kueue will manage #7513
• Kubeflow SparkApplication integration #7268
• MultiKueue: Support Scheduling of StatefulSet and LWS #4915
• ☂️ Review the state of feature gates and drop those no longer needed for 0.17 #8855
• Finalizer patch on the pod may overwrite other changes to the finalizer #3899
• KEP: add kueueviz auth #8971
• Support for running hero workloads #8826
• [Time Sharing] Support Time-Sharing in ClusterQueue #8522
• MultiKueue: invesigate alternative API for dispatcher API which does not suffer the SSA issues #8607

[mimowo]

cc @tenzen-y @gabesaba @mwysokin @kannon92 @sohankunkerkar

[kannon92]

Could we include #6915?

This seems to have slipped radar and missed reviews the last two cycles.

[sohankunkerkar]

We would like to make progress on the DRA side as well. Can you add these items in the release plan as well?

• Add MultiKueue + DRA integration and e2e tests #8775
• KEP-2941: Add extended resources design for DRA integration #8734

I will also add a design document for the partitionable devices.

[mimowo]

We would like to make progress on the DRA side as well. Can you add these items in the release plan as well?

Sure, progress with DRA is quite important, but I will add the second only, because here we focus on functional changes.

[mimowo]

Cool, ok we have a long list. Please remind us if something else should be watched :)

As usual - these are nice to haves - we will not block the release because of one of the features. We may still consider delaying the release for up to 2 weeks if some important features are "almost done".

[gabesaba]

As the release is just a month away, @mimowo and I discussed the priorities for our KEP reviews, and decided on the following:

Priority

• Concurrent Admission (Alpha) #8691
• MultiKueue: prevent starting preemptions in multiple worker clusters #8303
• Workload Scheduling Constraints (Workload-Level) and Preference-Aware MultiKueue Dispatching #8729
• Add a mechanism to consider preemption cost when finding preemption candidates #7990
• Graduate LocalQueue metrics to Beta #8830

Nice to Have

• Provide a way to specify in the configuration what resources kueue will manage #7513
• KEP-2941: Add extended resources design for DRA integration #8734
• MultiKueue: prevent starting preemptions in multiple worker clusters #8303
• Feature request: Support configuring Kueue to suspend a Job till told otherwise on a per job basis #6915
• [Time Sharing] Support Time-Sharing in ClusterQueue #8522
• Support for running hero workloads #8826

Please let us know if you have any feedback/concerns

[mimowo]

+1, but be aware that the categorization is just a "direction", which means:

• no features in the Priority are blocking for the release anyway if not done on time
• features if the Nice to Have section are best effort, but can be included if reviewed actively, and maintainers assess they are good to go

[tenzen-y]

IMO, #8734 would be supportive since extended resources could mitigate migration costs from device plugin to DRA ecosystem, which means it can accelerate DRA introduction in production. I believe that DRA is a big point for native WAS / TAS enhancements.