fix(test): test_kanban_core_functionality self-SIGTERM via heartbeat enforcement (broken-main, unblocks test(4) on #43/#44/#45)

[sahilm-ai]

Problem

tests/hermes_cli/test_kanban_core_functionality.py::test_dispatch_once_stale_disabled_when_timeout_zero deterministically SIGTERMs the pytest process itself (raw pytest RC=143, 5/5 runs). All tests pass, but pytest is killed before printing its summary line, so the parallel harness (scripts/run_tests_parallel.py) scrapes 0 passed/0 failed and buckets the file as "no tests ran". That is what makes the test (4) shard red on #43, #44, AND #45 — broken-main from the 170-commit upstream rebase, not introduced by any of those three PRs (diagnosed in t_8cd01b66).

Root cause

The test stores os.getpid() as the task's worker_pid with started_at 5h ago and no heartbeat, then calls dispatch_once(stale_timeout_seconds=0). dispatch_once runs enforce_missing_heartbeat (hermes_cli/kanban_db.py:6763) independently of stale_timeout_seconds, which os.kill(SIGTERM)s the stored PID = the pytest process. conftest's _guarded_kill explicitly allows self-kills, so the SIGTERM lands. enforce_missing_heartbeat-in-dispatch_once came from fork commit c699bb2 (#37); the test came from e286e68 — the rebase composed the two and exposed the latent self-kill.

Fix

Set a recent last_heartbeat_at on the run so enforce_missing_heartbeat skips the task (this test isolates STALE detection, not heartbeat enforcement). Also hardened test_enforce_max_runtime_integrates_with_dispatch against the same os.getpid()-as-worker_pid footgun on its real-os.kill dispatch_once call.

Test-only change. No production code touched.

Verification

• Raw venv/bin/python -m pytest test_kanban_core_functionality.py → RC=0, 166 passed, 1 skipped.
• Targeted: both modified tests pass (2 passed in 0.17s).
• N/A: no UI-emitting path in this change (test-only).

Sequencing

Merge this BEFORE #43/#44/#45 so their test (4) shard clears.

Summary by CodeRabbit

Tests

• Improved test reliability and isolation in core functionality tests by adjusting setup procedures to prevent background enforcement mechanisms from triggering during test execution.
• Refined two test scenarios to ensure accurate verification of timeout and stale detection behaviors without unintended external process interference.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: aa6a85e4-19f7-483d-b273-e90da17f1f6d

📥 Commits

Reviewing files that changed from the base of the PR and between 9b21fa5 and 4ab73a3.

📒 Files selected for processing (1)

• tests/hermes_cli/test_kanban_core_functionality.py

---

📝 Walkthrough

Walkthrough

Two test scenarios are adjusted to set last_heartbeat_at when backdating task_runs.started_at, preventing heartbeat enforcement from interfering with test isolation during dispatch_once execution.

Changes

Test Isolation via Heartbeat Control

Layer / File(s)	Summary
Heartbeat control in dispatch and stale-detection tests tests/hermes_cli/test_kanban_core_functionality.py	test_enforce_max_runtime_integrates_with_dispatch and test_dispatch_once_stale_disabled_when_timeout_zero now set last_heartbeat_at to a recent timestamp alongside the backdated started_at, preventing heartbeat enforcement from affecting test outcomes while each test isolates its specific behavior.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• sahilm-ti/hermes-agent#26: The test updates to explicitly set task_runs.last_heartbeat_at are directly tied to heartbeat-aware stuck-worker detection logic that depends on last_heartbeat_at.

Poem

🐇 Heartbeat checks, now tamed with grace,
Two tests set timestamps in their place,
No phantom signals shall distract,
Each test stays focused, sharp, intact! ❤️

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title directly addresses the main issue: fixing test failures caused by heartbeat enforcement triggering unintended SIGTERM in kanban core functionality tests.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch kanban/t_8cd01b66

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[github-actions]

🔎 Lint report: kanban/t_8cd01b66 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 9760 on HEAD, 9760 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 5053 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.

[sahilm-ti]

auto-review: approved → human_review (merge-sequencing call for Sahil).

• U1 in-scope: PASS — only tests/hermes_cli/test_kanban_core_functionality.py (+15/-4), test-only per AC.
• U4 AC coverage: PASS — test (4) is GREEN in CI (3m25s), the shard this PR fixes. The self-SIGTERM fix is confirmed beyond local.
• C5 worker identity: PASS — 1 commit 4ab73a3, Sahil (AI) <266772320+sahilm-ai>.
• C1 CI: one red — test (3). This is pre-existing broken-main, NOT caused by this PR: the 2 failures (test_unassigned_task_auto_assigned_with_default_assignee, test_dry_run_with_default_assignee_reports_without_mutating) live in test_kanban_default_assignee.py, which this PR does not touch. It's the default_assignee spawn regression diagnosed in t_8cd01b66 and fixed by fix(tests): hermetic kanban dispatch tests + default_assignee spawn regression #43.

Sequencing per AC: merge #46 first (clears test (4)), then #43 (clears test (3)). Auto-review will not merge — Sahil's call.