Skip to content

pr-review: per-PR cancel-in-progress cancels racing runs (concurrency thrash) #534

Closed

Description

@don-petry

Parent: #495 · Unplanned — now the main friction draining the backlog. · Work type: Fix

pr-review's per-PR concurrency group uses cancel-in-progress: true, so racing runs on the same PR cancel each other (the long-standing ~34%% cancellation rate). Observed repeatedly during #497 validation: dispatched/real review runs sit queued then get cancelled, slowing autonomous backlog clearance.

Proposed: same-head-SHA review runs are idempotent, so they don't need cancelling — either drop cancel-in-progress for the review group, or key the group by head SHA and let same-SHA runs no-op via the existing marker rather than cancel. Re-evaluate after change against the cancel rate.

Not blocking the fan-out; high-leverage for agent effectiveness.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugBug reportsinitiativeEpic / initiative tracking issue

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions