feat: add status dimension in worker metrics

[Kaitou786]

What

Adding the status as part of the worker metrics.

Why

Without status, worker metrics aggregate all invocation outcomes together. This causes misleading data because failed/disconnected requests have vastly different performance characteristics than successful ones.

If we use cloudflare_worker_cpu_time_seconds to determine how much time the worker is taking, currently, the data can be skewed as success status takes far less time than clientDisconnected status.

Without the proper slicing/filtering of the data, it becomes difficult to measure and alert based on the above metric.

Tesing

This should be a completely safe change, especially since we are already fetching this value from the Cloudflare GraphQL API:

cloudflare-prometheus-exporter/src/cloudflare/gql/queries.ts

Lines 373 to 388 in ca74cdb

	export const WorkerTotalsQuery = graphql(`
	query WorkerTotals(
	$accountID: string!
	$mintime: Time!
	$maxtime: Time!
	$limit: uint64!
	) {
	viewer {
	accounts(filter: { accountTag: $accountID }) {
	workersInvocationsAdaptive(
	limit: $limit
	filter: { datetime_geq: $mintime, datetime_lt: $maxtime }
	) {
	dimensions {
	scriptName
	status

However, I did run it against a Cloudflare account using:

  echo "CLOUDFLARE_API_TOKEN=<TOKEN>" > .dev.vars
  npm install
  npm run dev

post which using going to http://localhost:8787/metrics I could see:

cloudflare_worker_requests_total{script_name="redacted",account="redacted",status="success"} 6
cloudflare_worker_requests_total{script_name="redacted",account="redacted",status="clientDisconnected"} 8
...

cloudflare_worker_cpu_time_seconds{script_name="redacted",account="redacted",status="success",quantile="P999"} 0.049446

cloudflare_worker_cpu_time_seconds{script_name="redacted",account="redacted",status="clientDisconnected",quantile="P999"} 0.055794