Metrics, explain, and events

Use explain to inspect the access plan, then explain analyze for a value-free execution profile. High examined-to-returned ratios suggest indexing or earlier filtering. High working memory suggests smaller pages/batches or fewer blocking stages. Index-entry counts confirm whether a composite index is useful.

ConcurrentEngine::metrics_snapshot() exposes versioned, cardinality-bounded request, error, latency, concurrency, connection, and receipt metrics. The optional metrics feature only renders Prometheus text and never opens an endpoint. Protect the endpoint and avoid table, query, parameter, or request-ID labels.

ConcurrentEngine::with_observer emits exactly one value-free terminal event per core request and sampled slow-query events. Request IDs are absent by default; an explicit HMAC key enables irreversible hmac1: correlation.

Observer callbacks are synchronous. Copy the small event to a bounded channel and return. Applications own retention, queue, disk, and label bounds. storage_outcome_uncertain requires reopen and check.

Troubleshoot from aggregate metrics, to explain/analyze, to temporary slow-event sampling. Never enrich diagnostics with business values.