Skip to content

CLI --load interrupt e2e: SIGTERM case exits by signal (code null) under load #2973

Closed

Description

@cliffhall

Problem

clients/cli/__tests__/load-interrupt-e2e.test.ts › "prints the partial summary and exits 130 on one SIGTERM" fails intermittently under machine load: the child's exit code is null (it died by the signal) instead of 130.

AssertionError: expected null to be 130
 ❯ __tests__/load-interrupt-e2e.test.ts:141:20
    139|       run.child.kill(signal);
    140|       const { code, stdout, stderr } = await run.done;
    141|       expect(code).toBe(EXIT_CODES.INTERRUPTED);

Observed while working #2922 (which touches only clients/mcpdo tests): it failed the local:gate coverage stage once, and in isolated reruns of just this file it failed 3 of 6 times with the load average around 17–22 (several concurrent agent gates). Passes on a quiet run.

Expected

The case is deterministic: either the CLI's signal handler is guaranteed to be installed before the test sends the signal (and the test waits on that condition rather than calls >= 10), or the CLI's handling is fixed if a real window exists where a SIGTERM kills it outright mid---load. Per AGENTS.md, a race is fixed with an awaited condition, never headroom.

Activity

  1. added this to the v2.16.0 milestone on Nov 12, 2026
  2. added
    bugSomething isn't working
    v2Issues and PRs for v2
    on Nov 12, 2026
  3. self-assigned this
    on Nov 12, 2026
  4. added a commit that references this issue on Nov 12, 2026
  5. cliffhall commented on Nov 12, 2026

    @cliffhall
    MemberAuthor

    Fixed by #2978 (merged as d330df9).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

bugSomething isn't workingv2Issues and PRs for v2

Type

No type

Projects

No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions