☂️ Requirements for v0.3.0 release

[danielvegamyhre]

Deadline November 1st. We are still gathering requirements for v0.3.0, please feel free to comment suggestions.

Must have:

• Improve scheduling throughput for exclusive job placement per topology #310
• Scalability tests for exclusive Job to topology placement feature #174
• Handle race conditions in exclusive placement #341
• Bug: unable to setup indexes   #263
• Handle service name exceeds max length error gracefully #277
• Add validation to JobSet name to ensure child job names derived from it have no invalid chars #283

Nice to have:

• Feature: Configurable Failure Policy API #262
• Support for startup sequence #104
• Add a task explaining how to run TF training on JobSet #78
• Support for MPI jobs (JobSet extensions) #146
• Add a task explaining how to run MPI on JobSet  #77
• Create Jobs in batches #130

[kannon92]

Considering the light "Must-Haves", what do you and @ahg-g think about a strech-goal of including #104?

I know there is some questions around it but it will keep us honest if we put it on our release. IMO I also think it will be a really useful feature for JobSet users. /cc @vsoch @Gekko0114

[ahg-g]

Sounds good to me if you have a proposal.

[kannon92]

Sounds good to me if you have a proposal.

I drafted a KEP.

I am working on a draft of an implementation now.

[kannon92]

Not sure if you all have promoted this somewhere but it would be nice to target a blog post or an announcement with maybe this release.

I was thinking something around the level of https://kubernetes.io/blog/2022/10/04/introducing-kueue/

[vsoch]

I'd be in support! I'm definitely enjoying using Jobset - it's the basis of my new operator: https://converged-computing.github.io/metrics-operator/getting_started/metrics.html 🥳

[danielvegamyhre]

Added #104, #262, #263 to must-haves. We can revisit laster

[danielvegamyhre]

Release v0.3.0 has been published. I'll move the outstanding issues to v0.4.0