Skip to content

tests: root-cause the plan-kernel execute hang the blame collector named #81

Closed

Description

@AresConnor

First named lead from the new CI hang forensics (PR #80's --blame-hang wiring): during the PR #80 CI experiment, one run was a genuine 8-minute hang, and the blame collector's sequence artifact names ProvisioningPlanKernelTests.Execute_AcceptedTerminalOutcomeConsumesPlanAndRejectsReplay as the test running at collection time (run 30701399331, the hung attempt; ~1252 tests had completed; a hang dump was produced alongside the sequence XML — download the run artifacts before they expire).

Context: this is the previously-unattributable hang class (master had an 11.5-minute-silent cancelled run on 2026-08-01, and a 6-hour cancelled run on 0cb89c65 before that). It survived the scheduler-unpinning work merged in PR #80 — the catalogued rotating flakes went to zero across 14 CI runs, but this hang fired once, so its mechanism is something else (the test coordinates a plan-kernel execute with blocking fixture waits; note the related finding that OperationCoordinator-family test fixtures hold _persistGate across blocking waits, tracked separately).

Acceptance criteria:

  • Root-cause the hang from the dump/sequence artifacts (dotnet-stack is installed globally on the dev box), not from timeout tuning.
  • A deterministic regression test or fixture restructuring that removes the hang mechanism.
  • No guard widening without a measurement that justifies the number.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions