Skip to content

Make provider soak timing robust to early event-loop wakeups #58

Description

@stefan-jansen

The Windows Python 3.12 ecosystem job for Dependabot PR #7 reproduced a repository-owned timing failure in test_provider_soak_runs_continuously_and_reconnects_once.

asyncio.sleep() returned just before the configured deadline. run_provider_soak() recorded duration_seconds below SOAK_DURATION_SECONDS and rejected the report, even though all scheduled stages completed. The production six-hour provider qualification uses the same deadline logic.

Acceptance criteria:

  • wait-until-deadline logic tolerates early event-loop wakeups
  • the public provider-soak report still requires the full configured duration
  • deterministic tests reproduce an early wakeup and pass with the fix
  • full repository gates and required cross-platform checks pass

Observed in https://github.com/ml4t/live/actions/runs/31518293363/job/93868796869

Metadata

Metadata

Assignees

No one assigned

    Labels

    compatibility: noneNo known compatibility impactpriority: highMaterial user impact requiring prompt remediationstatus: acceptedAccepted for implementationtype: bugConfirmed or reported incorrect behavior

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions