Metrics & Observability

kasas is built to be unremarkable to operate: Prometheus metrics, structured logs, and liveness/readiness probes, all out of the box.

Prometheus metrics

GET /metrics exposes the Go/process defaults plus these kasas-specific series (the endpoint is always open, so a scraper needs no token):

Metric	Type	Labels	Meaning
kasas_sync_total	counter	status	Sync runs by outcome (success/error).
kasas_sync_duration_seconds	histogram	—	Sync duration distribution.
kasas_last_successful_sync_timestamp_seconds	gauge	—	Unix time of the last successful sync.
kasas_accounts	gauge	—	Accounts seen in the most recent sync.
kasas_transactions_inserted_total	counter	—	New transactions inserted.
kasas_transactions_updated_total	counter	—	Existing transactions refreshed by a sync.
kasas_rules_applied_total	counter	—	New transactions auto-labeled by a rule.
kasas_events_emitted_total	counter	type	Events appended to the stream, by type.
kasas_events_dropped_total	counter	—	Live event subscribers dropped for lagging.
kasas_transaction_versions_total	counter	kind	History snapshots, by change kind.
kasas_webhook_deliveries_total	counter	—	Webhook deliveries (terminal outcome).
kasas_webhook_delivery_attempts_total	counter	—	Individual delivery attempts (includes retries).
kasas_webhook_deliveries_dropped_total	counter	—	Deliveries dropped (queue full).
kasas_plugin_invocations_total	counter	—	Plugin hook invocations.
kasas_plugin_errors_total	counter	—	Plugin invocations that errored or panicked.
kasas_plugin_jobs_dropped_total	counter	—	Plugin jobs dropped (per-plugin queue full).

Labeled counters start empty

Prometheus *Vec metrics emit no series until they've been observed at least once. So kasas_sync_total{status="error"} won't appear on /metrics until the first error actually happens — that's expected, not a missing metric.

Logging

Structured logging via the standard library's slog. Configure with [log]:

• log.level — debug · info · warn · error (default info).
• log.format — json (default, for log pipelines) or text (for humans).

Each HTTP request is logged at debug level with method, path, status, bytes, duration, and the request id. Notable lifecycle events (sync results, dispatcher and plugin-manager startup, shutdown, the "unsecured" warning) log at info/warn.

Health & readiness

Endpoint	Checks	Use
GET /healthz	Process is up (200 ok).	Liveness probe.
GET /readyz	Pings the database.	Readiness / dependency probe.

Both are unauthenticated. The container image's HEALTHCHECK shells out to kasas healthcheck, which probes /healthz internally — handy because the scratch-based image has no curl.