ticketsLog in with ident

antcolony #28

A worker killed for parking is recorded as failed

review

opened by Architect · assigned to Caramboleyo

opened · updated

When the scheduler stops (restart), it kills a running worker to park it and logs "→ parked", but the worker's normal end handler then reads the empty result and records WORK FAILED ("not JSON, exit 143"), writes no park marker, and the next start begins a fresh session instead of resuming. Test:

  1. In the e2e: a session running, stop the run (both SIGTERMs): the log says parked, no WORK FAILED, the park marker is on the ticket, and the next run RESUMES the same session. Details: seen live 2026-09-26 13:03 on tickets.worldapi.org#20 (s-20260926T1044-278e20) at a docker compose restart: 'RUN: killing worker pid … → parked' followed by 'WORK FAILED — the worker output is not JSON (exit 143)'. The kill for parking must win over the end handler.

History

  1. Architect opened the ticket
  2. AntColonyScheduler changed the state progress

    A worker started on this ticket.

    colony-lease: s-20260926T1538-a97668 · host Loreana · started · ms 1790444319025

  3. AntColonyScheduler changed the state open

    Back to open: the worker stopped without a usable result (it ended without a result) — the next worker tries again.

    colony-run: s-20260926T1538-a97668 · host Loreana · no report

  4. AntColonyScheduler changed the state progress

    A worker started on this ticket.

    colony-lease: s-20260926T1606-b6d5d7 · host Loreana · started · ms 1790446014036

  5. AntColonyScheduler commented

    Half done: A worker stopped to park is now recorded as parked, not failed, and resumes the same session; the next worker continues.

    • Still open: The park/resume behaviour is not proven by a run: e2e suites break on tickets' new projec… (and 1 more)
    • Decided: Same live map fix in the agent as in the scheduler.

    colony-report: s-20260926T1606-b6d5d7 · sha256 6d840fe48c75 · controller pass · half done

  6. AntColonyScheduler changed the state open

    Back to open: half done — the next worker continues.

  7. AntColonyScheduler changed the state progress

    A worker started on this ticket.

    colony-lease: s-20260926T1608-854d39 · host Loreana · started · ms 1790446121637

  8. Architect changed the state pending

    Held until antcolony#31 (e2e setup) is done: the e2e is broken by tickets' projects, so this fix cannot be proven and it restarts half done (the #30 loop).

  9. Architect assigned the ticket to Caramboleyo
  10. AntColonyScheduler commented

    Half done: A worker stopped to park is recorded as parked, not failed, and resumes; a run proving it is still missing; the next worker continues.

    • Still open: Park/resume behaviour not proven by a run: after antcolony#31 is fixed, run `node tests/e… (and 1 more)
    • Decided: Did not fix antcolony#31 here; it is a separate ticket.

    colony-report: s-20260926T1608-854d39 · sha256 729cfb4f4b52 · controller pass · half done

  11. Architect changed the state open

    antcolony#31 is done: the e2e passes again (348 + 46) — back in the queue.

  12. AntColonyScheduler changed the state progress

    A worker started on this ticket.

    colony-lease: s-20260926T1704-b64caf · host Loreana · started · ms 1790449477026

  13. AntColonyScheduler commented

    Built — goes live with the next deploy.

    colony-report: s-20260926T1704-b64caf · sha256 a428b7658cb5 · controller pass · ready to deploy

  14. AntColonyScheduler commented

    Now live on container antcolony-scheduler (restart by the architect) — A worker stopped to park is recorded as parked, not failed, and the next run resumes its session.

    Test:

    1. After the restart, stop the scheduler while a worker runs.
    2. The log says parked, no 'WORK FAILED'; the ticket stays in progress.
    3. Start it again: the same session is resumed, not a fresh one.
    • Decided: README unchanged: no user-facing change; STATUS.md already describes the fix.

    colony-deployed: s-20260926T1704-b64caf · sha256 a428b7658cb5

  15. AntColonyScheduler changed the state review

    Ready for you to test — see the last comment.

Reading is open to everyone. To comment or change the state, log in with ident (top right) and choose a display name.