Fix Windows flake in notifier dead-endpoint e2e test

[Mikola Lysenko (mikolalysenko)]

Problem

update_notifier_e2e::dead_endpoint_never_fails_the_command flakes on the Windows test leg:

panicked at crates\socket-patch-cli\tests\update\update_notifier_e2e.rs:515:5:
a failed check must still rate-limit itself to once a day; lastCheckAt=1791482144 now=1791572146

Seen on run 37972832939 (test (windows-latest, 2), --test update: 52 passed, 1 failed) on an unrelated docs PR. lastCheckAt is the STALE value the test seeds (~25h old), so the state file was never rewritten. It has hit a PR so far, not the merge queue. But this test runs on the Windows leg in merge_group too, and there each flake evicts an entry and costs a full rebuild.

Root cause

The lastCheckAt write happens in refresh_latest, which runs in a detached tokio::spawn. finish joins it under the production 500 ms grace, and then the child calls process::exit. This row used eligible_kit, which keeps the 500 ms grace. On a loaded Windows runner the task can miss the window and get killed before save_state lands. The comment on eligible_kit_await_fetch describes this same race. Every other row that asserts on the background task's effect already uses that helper; this one was missed.

Fix

Use eligible_kit_await_fetch("http://127.0.0.1:1") in this test, and add a doc note explaining why. No assertion changes:

• The fetch is still capped at 2 s, so the run waits at most about that long.
• The test still proves that a failed check advances lastCheckAt. Only the pre-fetch write exists on the failure path.
• grace_budget_bounds_command_latency still pins the real 500 ms ceiling.

Proof

To simulate a runner where the task misses the window, I added SOCKET_UPDATE_GRACE_MS=0 to eligible_kit locally (the change is not committed), then looped the test 20x:

	pass	fail
main	5	15
this PR	20	0

• With the real config, the fixed test passed 30/30 with every core saturated by busy loops.
• cargo test -p socket-patch-cli --test update update_notifier_e2e: 19 passed.
• cargo clippy -p socket-patch-cli --test update -- -D warnings: clean.
• rustfmt --check on the touched file: clean.

No tests were removed or moved.

🤖 Generated with Claude Code

https://claude.ai/code/session_01UWAaQAvuYc8RATeyXGm8rs

---

Generated by Claude Code

---

Note

Low Risk
Test-only change to e2e harness timing; no production notifier or CLI behavior is modified.

Overview
Fixes intermittent Windows CI failures in dead_endpoint_never_fails_the_command by aligning it with other notifier e2e tests that assert on background-check side effects.

The test now uses eligible_kit_await_fetch (lifted SOCKET_UPDATE_GRACE_MS) instead of eligible_kit, so the child process waits for the detached update check to finish writing lastCheckAt before exit. Assertions are unchanged—the unreachable host still uses a 2s fetch cap; only the join grace is relaxed to avoid reading the seeded STALE timestamp when the runner is slow.

Doc comments on the test explain why the lifted grace is required.

^{Reviewed by Cursor Bugbot for commit 25c1633. Configure here.}

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 25c1633. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Ready for review at 25c1633.

• CI: 35/35 non-skipped checks green on the head (18 skipped by path filters); mergeable, no conflicts; no main-wide failures.
• Bugbot: reviewed 25c1633, no findings.
• Reviewer focus: test-only change in crates/socket-patch-cli/tests/update/update_notifier_e2e.rs; check the dead-endpoint assertion no longer depends on the detached task finishing inside the 500 ms join grace.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief

What it does. Test-only fix for a Windows flake. dead_endpoint_never_fails_the_command asserts that a failed update check still advances lastCheckAt, but that write happens in the detached check task. Under the default 500 ms join grace, a loaded runner could exit before the write. The test now uses eligible_kit_await_fetch (SOCKET_UPDATE_GRACE_MS=30000), like the other rows that assert on the task's effect.

Risk: low. One test changed, no product code. The endpoint is 127.0.0.1:1 (connection refused), so the longer grace adds no runtime in practice.

Look here

• update_notifier_e2e.rs#L503-L517:`` the change and its rationale
• update_notifier_e2e.rs#L105-L109:``` eligible_kit_await_fetch, already used by the five other effect-asserting rows

Verified. Read the diff and the helper. The test's assertions are unchanged; only the grace is lifted, which matches the rows at L136, L203, L235, L301 and L332. CI 53/53 green (35 pass, 18 skipped) on 25c1633, Windows included; Cursor Bugbot: no new issues. CHANGELOG untouched.

Changes I made. None.

Open questions. None.

Auto-merge is armed: approving sends it straight to the merge queue.

---

Generated by Claude Code