Skip to content

tests/chaos: sigkill_test.go mid-stream keeps the free-runner gate #2840 just removed #2841

Description

@devarismeroxa

Found during the adversarial review of #2840 (property2 fix).

sigkill_test.go:110-118 + assertSigkillIsGapFree:195 runs killAfterReads=95, paceMS=15, total=400 — no cap, no watermark gate. Identical parameters to property2's pre-fix mid-position-write, which #2836's stall injection made fail.

It has not flaked only because its verdict is stable (no stale assertion, so a blown-through kill lands in a different window rather than failing). Under the same stall the case silently stops testing the window it claims. #2840's own rationale — an uncapped child that finishes its run would fail the kill — applies verbatim.

Fix shape: the #2840 pattern — production cap (holdAt) + kill gated on the durable committed watermark, lower bound observed, upper bound structural.

Filed from review; not blocking #2840.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions