Skip to content

fix: count events in the window, share the object budget, trust the Lease's own times, keep CronJob runs in one incident - #589

Merged
abahmed merged 5 commits into
mainfrom
fix/event-budget-lease-noise
Nov 1, 2026
Merged

abahmed merged 5 commits into
mainfrom
fix/event-budget-lease-noise

Conversation

@abahmed

@abahmed abahmed commented Nov 1, 2026

Copy link
Copy Markdown
Owner

Unusual events: a note now keeps its occurrences with their times, and a Warning reason reports on the occurrences inside EventWindow. Three distinct events on a custom resource with no health status report; three spread over hours do not. Healthy custom resources stay silent.

Object budget: removes the per-group report exception. The first 100 objects of every dynamic kind no longer draw on the shared budget, so a high-volume kind cannot starve a small one. Admission rather than ranking, because object counts are unknown until a list has run and watches start together. Worst case grows by 100 objects per watched kind.

Lease: a renewal is now read from the Lease itself, renewTime past acquireTime. This survives a restart and does not count a stable identity taking the Lease again. A Lease without acquireTime keeps the previous in-memory check.

CronJob: blame on a Job or ReplicaSet maps to its top owner, so a run no longer opens an incident beside its CronJob. Blame moving between a CronJob and its runs is not a material change; a new reason still is. New labelled scenario cron-runs-keep-failing; SCORECARD refreshed.

@abahmed abahmed added the e2e Run the real-cluster E2E suite on this pull request label Nov 1, 2026
@abahmed
abahmed merged commit 6417a20 into main Nov 1, 2026
53 of 55 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

e2e Run the real-cluster E2E suite on this pull request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant