[hellices]

Closes #511

Summary

test_absolute_deadline_reaps_reader_that_suppresses_cancellation failed twice in a row on Windows in #509's CI, in code #509 does not touch.

• Cause: the test used a 10 ms deadline and the real monotonic clock. DeploymentOutcomeController._observe subtracts the time since registration from the deadline before its first read. A runner that started the worker more than 10 ms late therefore marked the deadline elapsed without ever calling the reader, and the test waited 30 s for started.
• Fix (test only): the test freezes the controller's clock (clock=lambda: 0.0), so the run gets its whole deadline however late its worker starts. It uses deadline_seconds=0.5, which the run's real asyncio.timeout still enforces, leaving room for the read task to start on a coarse Windows timer (about 15 ms).
• What it proves is unchanged: the absolute deadline reaps a reader that suppresses cancellation, and the outcome is INCOMPLETE with a deadline summary.

Production is unaffected; the default deadline is 300 s.

Verification

• Reproduced the race deterministically with a temporary test, not committed. With the old settings (deadline_seconds=0.01) and a clock that moves 20 ms on between registration and the worker running, the reader is never called and started.wait() times out. That is the CI failure.
• uv run pytest -p no:tach tests/ui/test_deployment_outcome_controller.py -k absolute_deadline_reaps -q: passed 5 runs in a row, about 0.8 s each.
• uv run pytest -p no:tach -n 8 tests/ui/test_deployment_outcome_controller.py -q: 35 passed.
• The fixed test still catches both behaviours it is about. Dropping the second cancellation of a read that suppresses the first fails it, and so does not marking the deadline elapsed on TimeoutError.
• ruff check, ruff format --check and mypy on the test file: clean.

Checklist

• This closes an accepted issue in this repository that is assigned to me.
• Tests cover the change, written to fail first where behavior changed. (Test-only change; the race was reproduced first.)
• make check passes locally. (Test-only change. The affected file and lint/type checks were run; the full suite runs in CI.)
• User-facing docs and the README match the change, or nothing user-facing changed.
• No security invariant is weakened: writes still need a keystroke approval
  and a fail-closed audit entry, and sensitive reads are still masked.

[Copilot]

🟢 Approved

No active review issues were identified for the test-only Windows timing adjustment.

No open issues 🎉

What changed in this PR

This test-only change addresses the Windows timing race in #511 by giving the deployment outcome observer its full deadline regardless of worker startup delay. Production behavior is unchanged.

Changes:

• Freeze the controller’s clock and increase the test deadline from 10 ms to 500 ms.
• Retain the real asyncio timeout to verify that the deadline reaps a reader that suppresses cancellation.
• Explain the timing adjustment in an inline comment.

File	Description
tests/​ui/​test_deployment_outcome_controller.py	Adjust the clock and deadline to stabilize the cancellation test on Windows.

_{Review effort: Balanced}

[my-reviewer-agent]

검토 범위에서 보고할 이슈가 없습니다.