Metrics reference

Every catalogue-backed series the worker exports, and what each one measures. This is the vocabulary the alert rules are written in. Prometheus synthesises a few series of its own that no catalogue entry can cover — see Process and runtime metrics below.

Where to scrape it. The worker exposes /metrics on its own admin listener (WORKER_ADMIN_PORT, default 9101), the api on ADMIN_PORT (default 9100). Both bind to 127.0.0.1 by default and are unauthenticated — see Alert rules for the bind and scrape config Prometheus needs, and why neither port belongs in ports:.

The catalogue is closed. The worker records through a sink that accepts only the names below, so a typo cannot spawn a new series and an ad-hoc metric cannot appear without a catalogue entry. The table is generated from that catalogue, and the What it measures column is the same HELP text the scrape carries, so this page and /metrics cannot disagree.

Cardinality is bounded by your own deployment. One series per (profile, symbol) for the tick metrics, one per Redis audit stream for the audit metrics, one per account for the Binance user-data stream metrics, one per queue for the BullMQ depth gauge. The Postgres pool gauges carry no labels at all — one series each, per process.

A counter you read as a level will mislead you

Counters only go up, and a worker restart puts them back to zero. Read them with rate() or increase() over a window, never as a raw value. Several are also seeded at zero the first time a stream is touched: without that seed a first-ever incident is a series that appears already at its final value, which increase() reads as no change at all.

Gauges are last-value-wins and are the opposite trap: one that stops being written keeps exporting its last sample for the life of the process. audit_consumer_lag behaves that way for a stream that stops being probed.

Process and runtime metrics

Not in the table above, because the worker does not declare them: process_* and nodejs_* come from prom-client's default collectors, and up is synthesised by Prometheus itself, one series per scrape target. WorkerDown reads up; nothing in this repo writes it.