Provider health #88

	# Watches production for providers that answer nothing but errors.
	#
	# The case that motivated it: Open-Meteo returned 429 to every request from
	# Render for hours. Nothing broke — the rain nowcast fell back to two sources
	# and `lluvia.prevista` simply stopped existing — so the app quietly stopped
	# warning about coming rain and nobody found out. Every layer degraded exactly
	# as designed; that is precisely what makes it invisible.
	name: Provider health

	on:
	schedule:
	# Twice a day, inside the beach window in Madrid (UTC+2 in summer).
	- cron: '15 10,17 * * *'
	workflow_dispatch: {}

	jobs:
	salud:
	runs-on: ubuntu-latest
	defaults:
	run:
	working-directory: backend
	steps:
	- uses: actions/checkout@v4
	- uses: actions/setup-node@v4
	with:
	node-version: 20
	cache: npm
	cache-dependency-path: backend/package-lock.json
	- run: npm ci
	# The token turns the check into an actual observation: without it the
	# script can only read metrics, and an empty window has no failures in
	# it to find. It is the same secret the server reads as DIAG_PROBE_TOKEN.
	- run: npm run providers:health
	env:
	DIAG_PROBE_TOKEN: ${{ secrets.DIAG_PROBE_TOKEN }}

Navigation Menu