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

Sentinel: shared:01m23vygyffx5hg2f5vm8r74sr:2hwyx3x8nx is past its budget: trevato/forge: "/map on the forge-sized fixture, warm > (unnamed)" failed on 3 pulls in the last hour — the lane's red,… #3064

Merged opened this

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

Merged

The forge-sized `/map` warm test no longer decides pass or fail on wall-clock time, so the check machine's load can't turn it red. I couldn't reproduce the failure: the file passes locally in about 6 seconds. The type check wouldn't run here, because `bun x tsc --noEmit` errors with "bun is unable to write files to tempdir: ReadOnlyFileSystem". The change is pushed and the forge's check will be th…

shipped production is serving it deploy ee374c7

opened this

The forge-sized /map warm test no longer decides pass or fail on wall-clock time, so the check machine's load can't turn it red. I couldn't reproduce the failure: the file passes locally in about 6 seconds. The type check wouldn't run here, because bun x tsc --noEmit errors with "bun is unable to write files to tempdir: ReadOnlyFileSystem". The change is pushed and the forge's check will be the first type check on it.

What changed: In test/map-cache.test.ts, the assertion was median * 3 < cold, a ratio of two timings taken on the same machine. A noisy cold or warm sample could flip it. I think that is the cause of the failure, but I haven't confirmed it. The test now counts SQL statements for the first, cold request and requires that the warm steady-state count times 3 is below it. Cold runs about 350–390 statements and warm about 27–35, so a missing cache still fails it by a wide margin. The existing statement-count bound (fixed chrome cost plus 2 per lane) is unchanged.

What I left alone: The 5 s p95 check stays as a guard against a hang. The 50 ms budget is still measured by bench/map-home.ts, not asserted in bun test.

To try it: bun test test/map-cache.test.ts. The output now prints the cold statement count next to the warm one.

🤖 Generated with Claude Code

Closes #3060

commented

The forge-sized /map warm test no longer decides pass or fail on wall-clock time, so the check machine's load can't turn it red. I couldn't reproduce the failure: the file passes locally in about 6 seconds. The type check wouldn't run here, because bun x tsc --noEmit errors with "bun is unable to write files to tempdir: ReadOnlyFileSystem". The change is pushed and the forge's check will be the first type check on it.

What changed: In test/map-cache.test.ts, the assertion was median * 3 < cold, a ratio of two timings taken on the same machine. A noisy cold or warm sample could flip it. I think that is the cause of the failure, but I haven't confirmed it. The test now counts SQL statements for the first, cold request and requires that the warm steady-state count times 3 is below it. Cold runs about 350–390 statements and warm about 27–35, so a missing cache still fails it by a wide margin. The existing statement-count bound (fixed chrome cost plus 2 per lane) is unchanged.

What I left alone: The 5 s p95 check stays as a guard against a hang. The 50 ms budget is still measured by bench/map-home.ts, not asserted in bun test.

To try it: bun test test/map-cache.test.ts. The output now prints the cold statement count next to the warm one.

🤖 Generated with Claude Code

Approved

Test-only change, and a sound one. The cold-versus-warm check now compares SQL statement counts instead of two wall-clock timings, so check-machine load cannot flip it. A missing cache still fails it: the cold count is about 350-390 and the warm count about 30. The 5 s p95 hang guard and the CHROME + 2*lanes bound stay. The counter is stopped in a finally, so a failed request does not leak it. No user-facing surface changed, so there is nothing to see in the preview. Two things to do next: (1) The comment below the assertions still says cold was ~2,700 statements, but the PR measures about 350-390; update it so the number is true. (2) Rename cold to coldStatements, and warm to page (it holds the page body, not a timing), so the assertion reads clearly.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs