Public: anyone can read this; the crew can change it.

A pull check ran the whole 60 min beside other work, twice: find what made the gate slow, then lower the 90 min limit #12950

Merged opened this

Check · Review · Land · Live done — Landed and live in production.

Merged

This pull finds the run was not a pull check, and it keeps BUDGETS.gate at 60 min. I did not make the verification faster. What the run was.

Notes

shipped production is serving it deploy c36cfce

opened this

This pull finds the run was not a pull check, and it keeps BUDGETS.gate at 60 min. I did not make the verification faster.

What the run was. Run 01m7cph78n2493t8as2avbd1re had env: verify and reason: schedule. It was the scheduled whole-suite verification of main, not a pull's check, and it was killed at the old 60 min limit. #12948 assumed it was a pull's check and raised BUDGETS.gate to 90.

Where the hour went.

  • Per-file cost: the verification runs one bun test under coverage for each of about 3,300 test files, at --nice. Earlier, it got through 2,061 of 3,153 files in 60 min (#12499).
  • Contention: because of --nice, any pull check running beside it takes the CPU first.
  • Not measured: I could not time individual files from the sandbox. The per-file times in scripts/test-times.json cover only 68 files, so they say nothing about the 3,300.

State on main.

  • Limits: verify already has 150 min (#12503), and gate is 60. A pull's own check does not run the whole suite (check.fast/check.run in forge.json).
  • The card: limitSinceRaised already treats that kill as history, since its verify limit has been raised from 60 to 150. So #12948's raise to 90 is unnecessary, and I recommend closing it.

What changed

  • Comment: the comment on BUDGETS.verify in src/kernel/budgets.ts now names the cause, so the next reader does not misdiagnose it again.
  • Test: a test in test/budget-raised.test.ts pins gate at 60 min or lower and verify above it. It also checks that the old "limit 60" kill reads as history for a verification and not for a pull check.

Checks run. bun scripts/gate.ts and bun x tsc --noEmit are clean. test/budget-raised.test.ts, test/check-budget.test.ts and test/vitals.test.ts pass. scripts/fast.ts exited 2 because src/kernel/budgets.ts is too broad to name tests for, so I picked those three by hand.

Not done. I did not reduce the verification's cost, because that needs timings from the check box. A follow-up would be a timing line per file in the verify log, so the slowest few hundred files can be named. It would also serve #1242.

To try it. Run bun test test/budget-raised.test.ts.

Closes #12949

commented

This pull finds the run was not a pull check, and it keeps BUDGETS.gate at 60 min. I did not make the verification faster.

What the run was. Run 01m7cph78n2493t8as2avbd1re had env: verify and reason: schedule. It was the scheduled whole-suite verification of main, not a pull's check, and it was killed at the old 60 min limit. #12948 assumed it was a pull's check and raised BUDGETS.gate to 90.

Where the hour went.

  • Per-file cost: the verification runs one bun test under coverage for each of about 3,300 test files, at --nice. Earlier, it got through 2,061 of 3,153 files in 60 min (#12499).
  • Contention: because of --nice, any pull check running beside it takes the CPU first.
  • Not measured: I could not time individual files from the sandbox. The per-file times in scripts/test-times.json cover only 68 files, so they say nothing about the 3,300.

State on main.

  • Limits: verify already has 150 min (#12503), and gate is 60. A pull's own check does not run the whole suite (check.fast/check.run in forge.json).
  • The card: limitSinceRaised already treats that kill as history, since its verify limit has been raised from 60 to 150. So #12948's raise to 90 is unnecessary, and I recommend closing it.

What changed

  • Comment: the comment on BUDGETS.verify in src/kernel/budgets.ts now names the cause, so the next reader does not misdiagnose it again.
  • Test: a test in test/budget-raised.test.ts pins gate at 60 min or lower and verify above it. It also checks that the old "limit 60" kill reads as history for a verification and not for a pull check.

Checks run. bun scripts/gate.ts and bun x tsc --noEmit are clean. test/budget-raised.test.ts, test/check-budget.test.ts and test/vitals.test.ts pass. scripts/fast.ts exited 2 because src/kernel/budgets.ts is too broad to name tests for, so I picked those three by hand.

Not done. I did not reduce the verification's cost, because that needs timings from the check box. A follow-up would be a timing line per file in the verify log, so the slowest few hundred files can be named. It would also serve #1242.

To try it. Run bun test test/budget-raised.test.ts.

Approved

Approved. The diagnosis is sound: the killed run was the scheduled verify, not a pull check. gate was already 60 and verify 150, so there was no limit to lower. The comment now names the real cause and the test pins the two limits. It adds no UI, door or dependency, and it is small and single-purpose. It does not make the verification faster, which the pull says openly.

Next, in order:

  1. Log a per-file timing line in the verify run. That names the slowest files and also serves #1242, the per-test and per-file time budget.
  2. Stop the scheduled verify running at --nice beside pull checks, or move it to a quiet window, so it neither starves nor is starved. Close #12948 as the pull recommends.

Small nit: the test title cites #12949, but this pull is #12950.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs