JobSet v0.2.1

What's Changed

• Automated cherry pick of #223: use pyproject.toml by @kannon92 in #226
• Automated cherry pick of #225: use get_namespaced_custom_object for getting jobset by @danielvegamyhre in #227
• Automated cherry pick of #248: add patch rbac for events by @kannon92 in #259

Full Changelog: v0.2.0...v0.2.1

JobSet v0.2.0

What's Changed

• Added tracking for the status of ReplicatedJobs (number of active/succeeded/failed jobs for each replicated job) to support Kueue integration.
• NodeAffinity is now mutable for suspended JobSets to support Kueue integration. When the JobSet is unsuspended, any ReplicatedJob with a modified NodeAffinity will result in the child jobs having their NodeAffinities updated to match.
• Added Go and Python client libraries for interacting with JobSet API
• JobSets now have a configurable SuccessPolicy. The policy targets a subset of ReplicatedJobs, and can wait for any or all jobs within each ReplicatedJob to complete before marking the JobSet complete.
• 1 headless service per JobSet is created, instead of 1 headless service per ReplicatedJob.
• User can specify the network subdomain, otherwise it will be defaulted to the JobSet name.
• Bug fix ensuring we only count jobs defined in the JobSet spec when determining if the JobSet is completed or not.
• Testgrid improvements (each test name shows up with success/failure now).

Acknowledgements

Thanks to all our contributors in this release, in no particular order:
@ahg-g @danielvegamyhre @kannon92 @vsoch @tenzen-y @mcariatm @clearnicki @a-hilaly @stuton @charles-chenzz

Full Changelog: v0.1.3...v0.2.0

JobSet v0.1.3

What's Changed

• Automated cherry pick of #118: Propagation errors to trigger reconciliation when they happen by @ahg-g in #121
• Automated cherry pick of #119: fix default success policy by @danielvegamyhre in #122

Full Changelog: v0.1.1...v0.1.3

JobSet v0.1.2

What's Changed

• Add JobSet name and ReplicatedJob name as annotations to jobs and pods.

JobSet v0.1.0

Features

• Support creating Jobs of different templates or many jobs from the same template
• Enable pod-to-pod communication via pod hostnames
• Support for 1:1 job to topology domain assignment
• Default success policy: JobSet succeeds when all jobs are completed
• Default failure policy: Jobset fails if any job fails
• Support for specifying a failure policy with a maximum number of times a jobset can be restarted before failing
• Support for suspending and resuming a JobSet
• Defaulting and validation webhook
• Webhook uses internal cert by default; user can use cert-manager as second choice

Acknowledgments

Thanks to our contributors in this release, in no particular order:
@danielvegamyhre @kannon92 @charles-chenzz @ArangoGutierrez @ahg-g @tenzen-y