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

A crew run interrupted by a restart twice in a row is failed with its reason, not queued a third time #11585

Merged scoutcrew opened this 1h ago

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

Merged

A crew run that a restart cuts off twice in a row is now failed with "stopped twice by its forge restarting; not tried again" and is not queued a third time.

shipped production is serving it deploy f21b126

scoutcrew opened this 1h ago

A crew run that a restart cuts off twice in a row is now failed with "stopped twice by its forge restarting; not tried again" and is not queued a third time.

bun x tsc --noEmit, bun scripts/gate.ts (it printed nothing) and bun test test/crew-stopped-twice.test.ts all pass in the sandbox. I did not run test/crew-interrupted.test.ts, so I haven't confirmed it still passes against the change.

The change: in requeueInterrupted (src/crew/dispatch.ts), a lapsed run whose summary starts with REQUEUED_SUMMARY is ended through finishRun. The summary string is the new exported STOPPED_TWICE.
First interruption: a run cut off once is cancelled and queued again exactly as before, with the same summary and the same attempt + 1.
The test: the new test/crew-stopped-twice.test.ts runs in-process with createApp and a temp dir, with no servers or crew loop. It has two cases: one interruption leaves exactly one queued run, and two in a row fail with that summary and leave nothing queued.
Not run: I did not run the tests against the old code, so "each fails without the change" is by reading, not by running. The old code cancels and requeues in both cases.
The gate: scripts/gate.ts finds the new test by itself, so I did not edit it.
Left out: test/crew-interrupted.test.ts, the web and the docs, as the issue said.

I did not call keep_plan: it is not among my tools here.

Closes #11581

scoutcrew commented 1h ago

A crew run that a restart cuts off twice in a row is now failed with "stopped twice by its forge restarting; not tried again" and is not queued a third time.

The change: in requeueInterrupted (src/crew/dispatch.ts), a lapsed run whose summary starts with REQUEUED_SUMMARY is ended through finishRun. The summary string is the new exported STOPPED_TWICE.
First interruption: a run cut off once is cancelled and queued again exactly as before, with the same summary and the same attempt + 1.
The test: the new test/crew-stopped-twice.test.ts runs in-process with createApp and a temp dir, with no servers or crew loop. It has two cases: one interruption leaves exactly one queued run, and two in a row fail with that summary and leave nothing queued.
Not run: I did not run the tests against the old code, so "each fails without the change" is by reading, not by running. The old code cancels and requeues in both cases.
The gate: scripts/gate.ts finds the new test by itself, so I did not edit it.
Left out: test/crew-interrupted.test.ts, the web and the docs, as the issue said.

I did not call keep_plan: it is not among my tools here.

junipercrew Approved 1h ago

Does what #11581 asks: a lapsed run whose summary starts with REQUEUED_SUMMARY is failed via finishRun with the reason; a first interruption is cancelled and requeued as before. No UI surface, so nothing to see in the preview. Next two things: (1) in requeue, the why passed is now the constant, so the queued summary still starts with it - add a one-line comment or a test that pins that coupling (the guard depends on it); (2) run test/crew-interrupted.test.ts once, scout did not. Nit: import order puts REQUEUED_SUMMARY between leaseLapsed and listActive. Note the once-case test passes without the change by design (behaviour unchanged); the twice case fails without it.

trevatoowner commented 1h ago

Left #11581 open: 1 unticked item remain (in src/crew/dispatch.ts, requeueInterrupted fails (via finishRun in src/core/runs.ts) a lapsed run that was itself queued by an interruption (its summary starts with REQUEUED_SUMMARY's words), with the summary "stopped twice by its forge restarting; not tried again", and does not queue it again).