Exception request for KEP-3329 'Retriable and non-retriable Pod failures for Jobs'
Michał Woźniak
Hi, please review the exception request below.
There is one PR (#115331) which is important for k8s Jobs, but it also fixes an issue in Kubelet on its own (fixing the Kubelet issue #116410). The review is in-progress with one outstanding issue, which is understood.
Enhancement name: Retriable and non-retriable Pod failures for Jobs
Enhancement status (alpha/beta/stable): Beta
SIG: sig-apps (sig-node participating)
k/enhancements repo issue #: 3329
PR #’s: 115331
Additional time needed (in days): 3
Reason this enhancement is critical for this milestone: To fix the bug in Beta (see here). Also, to open the possibility of fixing #115844 in the next release.
Risks from adding code late: (to k8s stability, testing, etc.) The PR includes unit tests, and node e2e tests providing nearly 100% coverage for the new code. The e2e node tests for this PR were run in a loop with over 100 repeats to detect flakes.
Risks from cutting enhancement: (partial implementation, critical customer usecase, etc.):
Delayed bugfix for Beta, resulting in delayed adoption of the pod failure policy feature to support use-cases around avoiding unnecessary costs for running batch workloads. Also, a delayed bugfix for the Kubelet issue #116410. Finally, a delayed solution for the #115844 issue (see the comment) for TensorFlow users.Regards,
Michał
Xander Grzywinski
Following discussion in K8s Slack the release team is APPROVING this exception request. Your updated deadline to make any changes to your PR is 17:00 PDT Friday 17th March 2022.
If you have any questions, please reach out to us in the #sig-release Slack channel.
Thanks,
Xander Grzywinski,
1.27 Release Team Lead