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

"the asks page > when nothing asks, it says All fine, and so does the header" passes every time on a loaded check machine #5518

Merged opened this

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

Merged

I couldn't reproduce the failure here, so the cause below is a diagnosis from the code, not something I watched fail. `--rerun-each 30` passes 30 of 30 on this machine, which isn't loaded.

shipped production is serving it deploy ad44386

opened this

I couldn't reproduce the failure here, so the cause below is a diagnosis from the code, not something I watched fail. --rerun-each 30 passes 30 of 30 on this machine, which isn't loaded.

Cause. The asks page and the header count both include the forge's own needs. Those come from vitals(), which reads the real disk under the test's data directory. When free space there is under 10%, vitals() adds a "Disk is nearly full" ask. On a loaded check machine that makes the test see 1 ask where it expects "All fine" and 0. The cause is in the test, so no product code changed.

Change. The test in test/asks.test.ts now builds its forge with the existing disk option, which is how test/domains.test.ts pins it. The disk reads 95% free regardless of the host. Nothing the test promises is weakened. It has no longer timeouts, no retries, no sleeps and no skipped assertions. The other asks tests still pass and tsc --noEmit is clean.

Left out. The other host-dependent inputs to vitals() are the error and refused-call ledgers, which beforeEach already clears, and the run queue. This forge has no runs, so I didn't touch them. I didn't run the full check.

Try it. bun test --rerun-each 30 test/asks.test.ts -t "when nothing asks, it says All fine"

Closes #5327

commented

I couldn't reproduce the failure here, so the cause below is a diagnosis from the code, not something I watched fail. --rerun-each 30 passes 30 of 30 on this machine, which isn't loaded.

Cause. The asks page and the header count both include the forge's own needs. Those come from vitals(), which reads the real disk under the test's data directory. When free space there is under 10%, vitals() adds a "Disk is nearly full" ask. On a loaded check machine that makes the test see 1 ask where it expects "All fine" and 0. The cause is in the test, so no product code changed.

Change. The test in test/asks.test.ts now builds its forge with the existing disk option, which is how test/domains.test.ts pins it. The disk reads 95% free regardless of the host. Nothing the test promises is weakened. It has no longer timeouts, no retries, no sleeps and no skipped assertions. The other asks tests still pass and tsc --noEmit is clean.

Left out. The other host-dependent inputs to vitals() are the error and refused-call ledgers, which beforeEach already clears, and the run queue. This forge has no runs, so I didn't touch them. I didn't run the full check.

Try it. bun test --rerun-each 30 test/asks.test.ts -t "when nothing asks, it says All fine"

Approved

The cause is right: vitals() reads the real disk, and the test now pins it with the existing disk option, the same way test/domains.test.ts does. The option reaches the health path (src/actions/health.ts:54). It is test-only, with no product change, and no assertion is weakened. Two things I would do next: (1) pin the run-queue input the same way if a loaded machine can ever reach it; (2) share the roomy helper between domains.test.ts and asks.test.ts instead of a second copy.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs