KEP-2238keps/sig-node/2238-liveness-probe-grace-period
Liveness Probe Grace Period
Liveness probes currently use the terminationGracePeriodSeconds on both normal shutdown and when probes fail. Hence, if a long termination period is set, and a liveness probe fails, a workload will not be promptly restarted because it will wait for the full termination period. I propose adding a new field to probes, probe.terminationGracePeriodSeconds. When… Full summary
Liveness probes currently use the terminationGracePeriodSeconds on both normal shutdown and when probes fail. Hence, if a long termination period is set, and a liveness probe fails, a workload will not be promptly restarted because it will wait for the full termination period. I propose adding a new field to probes, probe.terminationGracePeriodSeconds. When set, it will override the terminationGracePeriodSeconds for liveness or startup termination, and will be ignored for readiness probes. This maintains the current behaviour if desired while providing configuration to address this unintended behaviour.
Facts
- Authors
- @ehashman
- Approvers
- @derekwaynecarr
- PRR approvers
- @johnbelamaricalpha@johnbelamaricbeta@johnbelamaricstable
- Tracking issue
- #2238(opens GitHub in a new tab) closednot_plannedclosed 2024-06-15 01:33 UTCsig/nodestage/stablelifecycle/rotten
- Milestones
- alpha v1.21beta v1.22stable v1.28
- Participating SIGs
- none
- Feature gates
- ProbeTerminationGracePeriod
- Created / last updated
- 2021-01-07 / none