Last updated 2026-05-18
Observability
Health
The service implements standard gRPC health checking. Health status is
refreshed while the process is running and reports NOT_SERVING
if Redis becomes unreachable or required Lua scripts cannot be loaded.
Missing Redis scripts are reloaded automatically after Redis restart,
failover, or SCRIPT FLUSH.
grpc.health.v1.HealthReflection
Reflection is enabled so tools such as grpcurl can inspect the service.
grpc.reflection.v1alpha.ServerReflectionMetrics
quota_requests_total{rpc,action,product,allowed,reason}quota_request_duration_ms{rpc,action,product}quota_denials_total{action,product,limit_id}quota_redis_errors_totalquota_idempotency_hits_totalquota_reservations_activequota_reservations_expired_totalquota_leases_activequota_overages_totalquota_event_emit_errors_totalEvents
Event emission is best-effort and append-only. Postgres event writes use their own short timeout instead of inheriting the caller RPC context, so a completed request does not cancel asynchronous event delivery.
Reservation Expiry
An in-process sweeper expires due reservations from a Redis sorted-set
index. Expiry uses stored impact keys, applies
REFUND_FULL or CHARGE_FULL policy, and increments
quota_reservations_expired_total.