@green-tea/metrics

Count the lifecycle. Render Prometheus.

Reads the event stream green-tea already emits, so nothing in your
handlers has to know it is being measured.

Subscribes to green-tea’s lifecycle stream and counts what happened. You expose it on a route you own, with the transformer this package ships.

Runtimes

Runtime	Supported	Why not
Node	✅
Deno	✅
Bun	✅
workerd (edge)	❌	Counters live in one isolate, and an edge deployment has many, so a scrape reports whichever isolate answered it. Supported once the counters can be shared.

Install

npx jsr add @green-tea/metrics

Or deno add jsr:@green-tea/metrics, or bunx jsr add @green-tea/metrics. Under npm, yarn and pnpm jsr add installs through npm.jsr.io, so the import below stays @green-tea/metrics. There is no npmjs.org package — npm i @green-tea/metrics will not find one.

Usage

import { createApp } from '@green-tea/core';
import { metrics } from '@green-tea/metrics';

const app = createApp({ modules: [ApiModule], plugins: [metrics()] });

import { prometheus, type Metrics } from '@green-tea/metrics';

@Get('/metrics')
@Transformer(prometheus)
scrape(@needs('metrics') metrics: Metrics) {
  return metrics.render();
}

The route is yours, so its path, its auth and whether it is exposed at all stay your decisions.

What it counts, and what it refuses to

One failed request emits three lifecycle events, and a collector that counts them as three outcomes reports one request as several. The division this package follows:

Series	From	Meaning
`green_tea_requests_total`	`request:end`	Every dispatched request. Terminal and universal — this is the request counter
`green_tea_request_errors_total`	`request:failed`	Handler code threw. Additional to the request above, never a second request
`green_tea_request_duration_seconds`	`request:end`	Sum and count, so an average is available without buckets

Labels are route, method and status — never the event’s name. route is the matched pattern and is bounded by your route table (<unmatched> when nothing matched); name carries the path that arrived, which is bounded by nothing, so a scanner walking /aaa, /aab, /aac would mint a label per URL. That is a memory leak with a metrics backend attached.