Follow-up from #120.
DaemonWatchSetCache currently has zero observability — a user who enables the cache has no way to tell whether it actually short-circuited the fp-watches-collect walk on their project. For tuning the freshness window (#122) and diagnosing stale-hit confusion, we should emit counters.
Scope:
- On
get: emit tracing::debug!(target: "fbuild_daemon::watch_set_cache", hit = true|false, key_hash, age_ms, "watch-set cache lookup").
- Aggregate cumulative hit / miss counters on
DaemonContext exposed via /api/daemon/info (already the home for similar aggregate stats).
- Add an integration test that runs two back-to-back builds and asserts the second one bumps the hit counter.
Non-goals:
- No Prometheus / OTLP export. Plain tracing + JSON field on
/api/daemon/info is enough for now.
Related: #120 (the cache), #91 (warm-build instrumentation).
Follow-up from #120.
DaemonWatchSetCachecurrently has zero observability — a user who enables the cache has no way to tell whether it actually short-circuited thefp-watches-collectwalk on their project. For tuning the freshness window (#122) and diagnosing stale-hit confusion, we should emit counters.Scope:
get: emittracing::debug!(target: "fbuild_daemon::watch_set_cache", hit = true|false, key_hash, age_ms, "watch-set cache lookup").DaemonContextexposed via/api/daemon/info(already the home for similar aggregate stats).Non-goals:
/api/daemon/infois enough for now.Related: #120 (the cache), #91 (warm-build instrumentation).