test(docker): poll for settled dashboard 'down' slot instead of racing boot

[jethac]

Problem

tests/docker/test_dashboard.py::test_dashboard_slot_reports_down_when_disabled is flaky in the Docker build/test job:

AssertionError: Dashboard slot should be 'down' without HERMES_DASHBOARD;
svstat reports: 'up (pid 155 pgid 155) 0 seconds'
assert 'down' in 'up (pid 155 pgid 155) 0 seconds'
tests/docker/test_dashboard.py:51

Note the 0 seconds uptime — svstat is catching the service in its first moments.

Root cause

The dashboard is an always-declared supervised s6 longrun (docker/s6-rc.d/dashboard/). When HERMES_DASHBOARD is unset:

• run exits 0 immediately — no dashboard process is ever exec'd;
• finish exits 125, s6's "permanent failure, do not restart" marker (≈ s6-svc -O), so s6-supervise leaves the slot down and never restarts it.

So the slot's steady state is down, with no process running — corroborated by the sibling test_dashboard_not_running_by_default (a pgrep check) which passes.

But s6-rc exec's the run wrapper once when it brings the user-services bundle up, so s6-svstat can briefly report up (...) 0 seconds before run exits and finish settles the slot to down. The failing test did a single immediate s6-svstat check with no retry — unlike its sibling test_dashboard_slot_reports_up_when_enabled, which polls — so it intermittently reads that one-time boot flap.

Fix

Poll for the settled down state (mirroring the "up" test), anchored on ^down so a transitioning up (...) want down line can't false-match a bare down. Also corrects the docstring, which still described the deleted cont-init.d/03-dashboard-toggle down-marker mechanism (the real mechanism is the run-exit-0 / finish-exit-125 pair).

Test-only change — no s6 service or product code touched; it only relaxes an immediate check into a bounded poll for the same condition the service already guarantees in steady state.

[jethac]

Depends on #63560 (adds jethachan@gmail.com to AUTHOR_MAP). The first commit on this branch is that mapping — the check-attribution job needs it present to pass. Once #63560 merges to main, this rebases and that commit drops out, leaving just the test fix.

[teknium1]

Thanks for isolating this Docker-test race. Current main still performs the single immediate s6-svstat assertion at tests/docker/test_dashboard.py:44-53, while the disabled service takes the run exit-0 path (docker/s6-rc.d/dashboard/run:9-18) and the permanent-down finish path (docker/s6-rc.d/dashboard/finish:19-28).

The proposed poll_container(... | grep -q '^down') use is bounded by the existing helper in tests/docker/conftest.py:224-246, checks the returned success value, and mirrors the enabled-slot polling pattern at tests/docker/test_dashboard.py:74-75. The change is test-only apart from the explained attribution mapping dependency; GitHub reports PR head 879c4832548d as mergeable.

Automated hermes-sweeper review.

[jethac]

Closing — the prune in #74383 removed test_dashboard_slot_reports_down_when_disabled (and its siblings) entirely, so there's nothing left for this fix to stabilize. Thanks for landing #63522 via #74517!