Analytics & usage

bytetourist records every request as an event and aggregates it for usage, billing, and dashboards — without slowing the proxy down.

The event pipeline

The hot path stays fast: core publishes a request-event to Kafka asynchronously after each response, then returns to the caller. The scaler consumes that topic and writes to ClickHouse. Kafka is the immutable source of truth for billing; ClickHouse is the queryable materialised view.

core ──async──▶ Kafka (request-events) ──▶ scaler ──▶ ClickHouse ──▶ dashboards / rollups

What’s recorded per request

Each event carries: timestamp, org id, request id, node id and node IP, target host, method, URL, status code, latency (measured at the node), bytes in/out, region, country, ip_type, and any error. Events are retained for 90 days.

Usage & billing rollups

A materialised view aggregates events per org per hour (client_usage_hourly), which a periodic rollup turns into per-cycle usage for metered billing. Because the raw event log is immutable, usage is auditable and re-computable.

What you can see

In the dashboard:

Today — request count, success rate, data used, active keys.
Usage vs. quota — requests and bytes spent against your plan, with remaining quota and % utilisation.
Recent requests — a paginated, org-scoped log with status, latency, and target host.
Trends — success/error-rate time series.

Fleet metrics

Separately, the scaler exposes a metrics API (node health, in-flight counts) that core polls every few seconds to drive lowest_latency routing and the circuit breaker, and that the scaler itself uses to make autoscaling decisions.