---
name: aemet-madrid
id: 0002
status: pending
created: 2026-05-16
updated: 2026-05-16
priority: medium
risk: low
related_issues: [0097]
apps:
  - dag_engine
  - data_factory
  - footprint_geo_stack
trigger: cron
schedule: "0 * * * *"
expected_runtime_s: 10
tags: [api, weather, geo, http-only]
---

## Goal

Probar path HTTP-only (sin Chrome/CDP). Extractor REST -> data_factory -> sink geo (PostGIS via footprint_geo_stack). Demuestra que el stack tambien sirve para APIs publicas + datos georeferenciados.

## Pre-requisitos

- API key AEMET (gratis, signup). Guardar en `pass insert aemet/api-key`.
- `footprint_geo_stack` corriendo (PostGIS :5432 + Martin tiles :3000).
- dag_engine activo.

## Flow

1. Crear funcion del registry `aemet_get_weather_py_infra` (o usar `http_get_json_py_infra` directamente si la API responde JSON simple).
2. Endpoint AEMET observacion convencional: `GET https://opendata.aemet.es/opendata/api/observacion/convencional/datos/estacion/<id>` con header `api_key`.
3. Schema esperado: `[{ts, temp, humidity, pressure, wind_speed, lat, lon}, ...]`.
4. Sink: INSERT en PostGIS tabla `weather_madrid (ts, temp, humidity, pressure, geom geometry(Point, 4326))`.
5. Crear node en data_factory: `{kind: 'database', label: 'postgis_weather'}` (sink declarado).
6. DAG `aemet_madrid_hourly.yaml`:
   ```yaml
   name: aemet-madrid
   schedule: "0 * * * *"
   steps:
     - name: extract
       function: aemet_get_weather_py_infra
       args: ["--station", "3195", "--out", "/tmp/aemet.json"]
     - name: load_postgis
       function: db_insert_row_go_infra
       args: ["--db", "postgres://...", "--table", "weather_madrid", "--from-json", "/tmp/aemet.json"]
       depends: [extract]
   ```
7. Verificar Martin tiles renderiza overlay (opcional).

## Acceptance

- [ ] Funcion AEMET extractor existe (creada o reusada `http_get_json_*`).
- [ ] DAG corre 2x consecutivas via scheduler.
- [ ] PostGIS tabla `weather_madrid` tiene >=2 filas.
- [ ] data_factory muestra node `aemet_madrid` kind=extractor + node `postgis_weather` kind=database con `last_seen_at` reciente.
- [ ] Martin tile server sirve overlay weather (opcional).

## Telemetria esperada

- `function_stats.http_get_json_py_infra` o `aemet_get_weather_py_infra`: calls_24h += 24 (1/hora).
- `data_factory.runs`: 24 entries/dia.
- `data_factory.databases.last_seen_at` actualizado por sink.

## Notas

- Sin LLM/CDP. Mas barato que flow 0001.
- Caso minimal para servicios geo. Si funciona, sirve de plantilla para mas extractores API.