Skip to content

Flake sighting: sighup_reload e2e restart_pending assertion in full-gate-build #708

Description

@forkwright

Finding

One gate-run flake sighting: archon::config_reload_e2e sighup_reload_applies_live_rotates_jwt_and_holds_back_restart_class failed in the full-gate-build job on PR #700 (run 31566727588, 2026-08-12) with restart_pending must list database.db_path, got: [] — while the regular CI Test job on the same SHA passed, and the rerun passed. Main's gate before and after has been green.

Evidence

Why this matters

The full-gate-build job runs nextest with different parallelism/timing than the standard Test job; a SIGHUP-driven e2e that asserts on reload timing is exactly the class that goes flaky under load. Per the fleet flake policy this is one sighting, filed now so a recurrence has a home and count — not yet quarantined.

Desired correction

If it recurs: make the assertion poll for the restart_pending state with a bounded deadline instead of asserting at a fixed point, or quarantine per TESTING.md. A second occurrence on this test should move it to quarantine + redesign.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions