You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Contributor nudge: CI-failure nudges churn on the same PR as the failing-check set and reason prefix flap #17328
On hive's own spoke (since this morning's roll) the stage-072 contributor nudge re-fires repeatedly on the same PR while CI keeps failing across pushes and reruns (#17303 / #17305 / #17308). Pod log excerpt for #17308, one head (60ed7e2), five minutes:
#17073 made the nudge one comment per PR + reason class, but a CI failure is three reason classes (check-failing, optional-check-failing, required-check-failing → slugs check-failing / optional-check-failing), so as the lane's reason prefix flaps between them every pass is a "new class": the comment is rewritten, the relay gets another fix-pr dispatch, and the per-class nudge count restarts (so the cap of 2 never holds). Each new red head is another relay dispatch too, with no window between them.
All CI-failure reasons are one check-failing class. Same failing set → the existing hive:contributor-nudge comment is edited in place with the new head and an attempt count; never a second comment.
On the same head the failing set only accumulates (union); another check going red folds into the comment without a relay re-dispatch.
A changed failing set on a new head within governor.labels.stage.contributor_nudge.ci_failure_cooldown (default 6h) folds into the existing comment as a rewrite (count unchanged); after the cooldown it counts as a fresh nudge.
State lives in the stage-label store (/data JSON) so a restart does not re-nudge.
Settings-UI parity for the new key (Labels tab, next to the re-nudge fields).
Activity
🐛 contributor nudge: per-PR CI-failure cooldown and failing-set dedupe (