[zackees]

Why

Fixes the remaining ask in #1627, the flake that failed two release dry runs in a row with:

ERROR fbuild::output: daemon error: daemon did not become healthy after 3 spawn attempts (10s each)
running-process broker: direct daemon fallback (failed to connect to broker: No such file or directory)

Two earlier PRs already landed the diagnosis halves of that issue: #1732 reports the startup phase and each failed attempt, and #1734 gave CI a 30 s readiness budget. The third ask — "if the cause is a leftover daemon from an earlier step, attach to it or retire it rather than failing" — was still open. Three failed attempts remained terminal, so a runner with a leftover daemon (or a wedged one, #1360) could only be fixed by a manual rerun.

What changed

daemon_stop.rs gains a decision plus the I/O that acts on it, following the file's existing split (StopPlan = decision, run_daemon_stop = I/O):

• SpawnRecovery / plan_spawn_recovery(endpoint_healthy, live_daemon_pid)
  • endpoint answers → Adopt (the failed spawn already achieved its goal; spawn nothing)
  • silent + a verified live fbuild-daemon PID → RetireAndRetry(pid)
  • silent + nothing alive → GiveUp (report the failure, as today)
• retire_unresponsive_for_recovery — on RetireAndRetry it terminates that PID with the same terminate_and_confirm escalation fbuild daemon stop uses, then clears the records.

daemon_client::start_daemon calls it after the three attempts are exhausted and, if a daemon was retired, makes exactly one final spawn + readiness wait. On success the build continues; on failure the original error is reported with one added line naming the retired PID.

Safety

• Identity is fail-closed: the PID comes from recorded_daemon_pid(), which requires the process to be alive and its image stem to be fbuild-daemon. A recycled PID is never signalled.
• A daemon that answers is never retired — that is the Adopt branch, and it is what protects a parallel build on the same machine.
• Endpoint-scoped, like daemon stop: only this endpoint's recorded PID is considered, never a scan for every daemon (another checkout's daemon is a neighbour, not debris).
• Strictly additive: this runs only on a path that is a hard failure today, so no working build can take it.

Tests

Three new unit tests in daemon_stop_tests.rs (decision, no live daemon needed, matching the existing pattern for plan_stop):

running 8 tests
test cli::daemon_cmd::stop::tests::a_failed_spawn_retires_a_verified_unresponsive_daemon ... ok
test cli::daemon_cmd::stop::tests::a_failed_spawn_with_nothing_alive_to_retire_gives_up ... ok
test cli::daemon_cmd::stop::tests::a_healthy_endpoint_is_adopted_never_retired ... ok
test cli::daemon_cmd::stop::tests::a_healthy_endpoint_is_asked_to_stop ... ok
test cli::daemon_cmd::stop::tests::a_silent_endpoint_with_a_live_process_is_terminated_not_declared_absent ... ok
test cli::daemon_cmd::stop::tests::a_silent_endpoint_with_no_live_process_is_nothing_to_stop ... ok
test cli::daemon_cmd::stop::tests::a_live_non_daemon_pid_fails_the_identity_gate ... ok
test cli::daemon_cmd::stop::tests::wait_for_process_exit_returns_immediately_for_a_dead_pid ... ok

test result: ok. 8 passed; 0 failed; 0 ignored; 0 measured; 388 filtered out

Gates, all run in this worktree:

• soldr cargo clippy --workspace --all-targets -- -D warnings — clean
• soldr cargo fmt --all -- --check — clean
• uv run --no-project python -m ci.run_dylint --package fbuild-cli — exit 0
• ci/enforce_platform_boundary.py --print-totals — rows=44; dylint_rows=43 (44/0)
• ci/check_dylint_allowlists.py — current · ci/check_test_layout.py — ok

Not claimed

The retirement branch is exercised at the decision level; the terminating I/O is the already-shipped terminate_and_confirm used by fbuild daemon stop, which keeps one killer in the codebase rather than adding a second. Forcing a real daemon into the silent-but-alive state inside an integration test needs a stopped process, which I did not add here — say the word and it can follow up as a #[cfg(unix)] it test that parks a daemon and asserts the retirement.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

Run configuration

• Source mode: Auto
• Repository: FastLED/fbuild
• Branch: agent/daemon-retire-leftover
• File: .coderabbit.yaml
• Commit: 7a29f1b
• Layers: Repository YAML
• Review profile: CHILL
• Plan: Advanced
• Run ID: 288cbac5-1135-43b7-b725-b3d1b4cabf5c

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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.}