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

test/landing.test.ts drives the crew directly and names the step that did not happen, so it cannot go red on a slow tick #3493

shipped production is serving it opened this has it
opened this

Replaces #3355. Main went red for 16 minutes at 742eef2 on "a pull closed under a running agent > the run is cancelled, and its clone goes after it, not before". Two runs could not reproduce it; it passes five of five in a sandbox (5.9 s each). The test waits on the crew's own five-second tick and generous deadlines, and its failure says only "did not happen in time", so a loaded check gives no evidence.

Done:

  • [ ] The test starts the run and the cancellation by calling the crew's dispatch step directly (or a tick interval the test sets), not by waiting for the default five-second tick.
  • [ ] Every wait's failure message names the step and the state it saw: the run's state and summary, whether the clone exists, the last 20 lines of the run log.
  • [ ] It runs under 3 s locally, and 30 runs in a row in the sandbox pass (say so in the pull, with the command).
  • [ ] What it proves is unchanged: cancelled with "the pull closed", the clone removed only after the run ended, no ENOENT in the log, nothing queued after.

Files: test/landing.test.ts; at most a test-only option on the crew's start (named in the pull). Do not change how the crew cancels.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs