Repository navigation
0.70.2 T36: a push to main re-runs CI on the sha the merge queue just passed #4743
Description
Measured on 316dee2: the merge_group run 37092945880 took 9 jobs and 33.3 runner-min; the push run 37094246639 on the same sha took 9 jobs and 105.4 runner-min. Across the last 5 merges the push runs cost 513.6 runner-min more than the queue runs, about 103 per merge. Of that, 389.2 is the three test shards re-running what the queue had passed, and 109.5 is x86-main re-running a passed job.
Fix: a push to main takes the reuse tier only when a merge_group run on the same sha passed workspace-test. Its case table fails on the old file and passes on the new one. x86-main still runs in full, because its origin/main ratchets are not pure.
It changes what runs on main, so it needed the operator's sign-off: signed, effective on a passing review.
Status: branch pushed; the second review round is running. In the CI lane after T42.
K^ 2. Owner aprender-w4650. Refs #4678.
Activity
fix(t36): the aggregator rows run the ci.yml extract as a script, not…
noahgift commentedon Oct 8, 2026
Close criterion from #4763 is met: the first push runs after the merge show workspace-test reused on the merge queue's sha.
Measured on main ad8fafc:
tier=reuse. Shard 1 logged "workspace-test REUSED: push: merge_group run 37757303856 tested this very commit (ad8fafc) and its workspace-test succeeded". The shards took 1-3 min (before ci(tier): a push to main reuses the merge queue's workspace-test result on the same sha (T36) #4763: 389.2 runner-min across 5 merges went to shards re-running the queue's tests).Closing as done. #4678 stays open for its other rows.
Agent: aprender-clerk