kubernetes-sigs / karpenter / 35323450290

83%

main: 83%

Build Type

Pull #3330

github

Committed by semx

Commit Message

fix: set DisruptionReason on forceful expiration and garbage collection

The DisruptionReason condition is set only by the disruption queue, so a
NodeClaim deleted by the expiration or garbage collection controller
carries no reason, and the pod drain metric reports it as a forceful
termination with no detail.

Set the condition before the delete in both controllers, with Expired or
GarbageCollected as the reason and a message that names the expireAfter
and timestamps, or the missing instance. The NodeClaim is re-read and
patched with an optimistic lock, retried on conflict, so a concurrent
status update is not overwritten; a patch failure returns before the
delete so the reconcile retries.

The reasons are condition reasons, not NodePool budget reasons, so they
are not added to the DisruptionReason enum, and they follow the
PascalCase of the voluntary reasons set on the same condition.

Fixes #2023

Signed-off-by: Sergey Sannikov <sergey@sannikov.dev>

Pull Request Pull Request #3330: fix: set DisruptionReason on forceful expiration and garbage collection

Coverage Stats

19 of 28 new or added lines in 2 files covered. (67.86%)

2 existing lines in 1 file now uncovered.

17523 of 21028 relevant lines covered (83.33%)

76697.36 hits per line