microsoft--agent-lightning/docker/compose.prometheus-mongo-store.yml

services:
  mongo:
    extends:
      file: compose.mongo.yml
      service: mongo

    hostname: mongo
    volumes:
      - ./data/mongo-container:/data/db
      - ../scripts/mongodb_init_rs_profiling.js:/docker-entrypoint-initdb.d/init-rs.js:ro

    # This forces "mongo" to resolve to localhost ONLY inside this container.
    # This allows rs.initiate to succeed using the hostname "mongo".
    extra_hosts:
      - "mongo:127.0.0.1"

  app:
    extends:
      file: compose.store.yml
      service: app

    depends_on:
      - mongo
      - app-exporter  # Wait for the exporter to be ready first

    command:
      - /bin/bash
      - -c
      - |
          agl store --host 0.0.0.0 --port 4747 \
            --tracker console prometheus --backend mongo \
            --mongo-uri mongodb://mongo:27017/?replicaSet=rs0 \
            --n-workers ${AGL_STORE_N_WORKERS:-32}
    environment:
      - PROMETHEUS_MULTIPROC_DIR=/tmp/prometheus_multiproc
    volumes:
      - prometheus_multiproc:/tmp/prometheus_multiproc

  app-exporter:
    build:
      context: ../
      dockerfile: docker/Dockerfile.dev

    command: agl prometheus --host 0.0.0.0 --port 4748
    ports:
      - "4748:4748"
    environment:
      - PROMETHEUS_MULTIPROC_DIR=/tmp/prometheus_multiproc
    volumes:
      - prometheus_multiproc:/tmp/prometheus_multiproc

  mongodb-exporter:
    image: percona/mongodb_exporter:0.47.1
    command:
      - "--mongodb.uri=mongodb://mongo:27017/"
      - "--collect-all"
      - "--mongodb.collstats-colls=agentlightning.rollouts,agentlightning.attempts,agentlightning.spans,agentlightning.resources,agentlightning.workers,agentlightning.rollout_queue,agentlightning.span_sequence_ids"
    depends_on:
      - mongo
    ports:
      - "9216:9216"

  node-exporter:
    image: prom/node-exporter:latest
    # In CI you might not have full /proc, but this is OK for container-level stats
    pid: "host"
    command:
      - "--path.rootfs=/host"
    volumes:
      - "/:/host:ro,rslave"

  prometheus:
    image: prom/prometheus:latest
    command:
      - "--config.file=/etc/prometheus/prometheus.yml"
      - "--storage.tsdb.path=/prometheus"
      - "--storage.tsdb.retention.time=1h"
    volumes:
      - ./prometheus/prometheus.mongo.yml:/etc/prometheus/prometheus.yml:ro
      - ./data/prometheus:/prometheus
    depends_on:
      - app
      - app-exporter
      - mongodb-exporter
      - node-exporter
    ports:
      - "9090:9090"

  grafana:
    image: grafana/grafana:latest
    ports:
      - "9091:3000"
    depends_on:
      - prometheus
    volumes:
      - ./data/grafana:/var/lib/grafana
      # 1. Mount the Datasource Config
      - ./grafana/datasource.yml:/etc/grafana/provisioning/datasources/datasource.yml
      # 2. Mount the Dashboard Provider Config
      - ./grafana/dashboard-provider.yml:/etc/grafana/provisioning/dashboards/provider.yml
      # 3. Mount the folder containing the actual JSON files
      - ./grafana/dashboards:/var/lib/grafana/dashboards

    environment:
      - GF_INSTALL_PLUGINS=grafana-piechart-panel
      - GF_AUTH_ANONYMOUS_ENABLED=true
      - GF_AUTH_ANONYMOUS_ORG_ROLE=Admin
      - GF_AUTH_DISABLE_LOGIN_FORM=true
      - GF_DASHBOARDS_DEFAULT_HOME_DASHBOARD_PATH=/var/lib/grafana/dashboards/agentlightning.json

volumes:
  prometheus_multiproc:
    driver: local
    driver_opts:
      type: tmpfs
      device: tmpfs