docker-compose.yml

services:

  postgres:
    image: postgres:15
    environment:
      POSTGRES_USER: tagger
      POSTGRES_PASSWORD: tagger
      POSTGRES_DB: tagger
    ports:
      - "5432:5432"
    volumes:
      - ./data/postgres:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U tagger"]
      interval: 10s
      timeout: 5s
      retries: 5

  mlflow:
    image: ghcr.io/mlflow/mlflow:v2.13.0
    ports:
      - "5001:5000"
    volumes:
      - ./data/mlflow_artifacts:/mlflow/artifacts
    depends_on:
      postgres:
        condition: service_healthy
    entrypoint: ["sh", "-c", "pip install psycopg2-binary -q && mlflow server --backend-store-uri postgresql://tagger:tagger@postgres:5432/tagger --default-artifact-root /mlflow/artifacts --host 0.0.0.0 --port 5000"]
    healthcheck:
      test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:5000/health')"]
      interval: 15s
      timeout: 5s
      retries: 5

  serving:
    build:
      context: ./serving
      dockerfile: Dockerfile
    ports:
      - "8000:8000"
    environment:
      DB_URL: postgresql://tagger:tagger@postgres:5432/tagger
      MLFLOW_TRACKING_URI: http://mlflow:5000
      USE_ONNX: "false"
      USE_STUB_MODEL: "false"
      BUNDLE_PATH: /models/model_bundle.joblib
      PROMETHEUS_MULTIPROC_DIR: /tmp/prometheus_multiproc
    volumes:
      # Shared model store — training writes here, serving reads from here
      # This allows hot-reload via /admin/load-model without rebuilding the image
      - ./serving/models:/models
    depends_on:
      postgres:
        condition: service_healthy
      mlflow:
        condition: service_healthy
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3

  nextcloud:
    image: nextcloud:27
    ports:
      - "8080:80"
    volumes:
      - ./data/nextcloud:/var/www/html
      - ./nextcloud-app:/var/www/html/custom_apps/smartfiletagger:ro
    environment:
      NEXTCLOUD_ADMIN_USER: admin
      NEXTCLOUD_ADMIN_PASSWORD: admin
      SQLITE_DATABASE: nextcloud
      NEXTCLOUD_TRUSTED_DOMAINS: localhost nextcloud
      SIDECAR_URL: http://serving:8000
      NEXTCLOUD_INTERNAL_URL: http://nextcloud
    depends_on:
      serving:
        condition: service_healthy
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost/status.php"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 60s

  training:
    build:
      context: ./training
      dockerfile: Dockerfile
    environment:
      DB_URL: postgresql://tagger:tagger@postgres:5432/tagger
      MLFLOW_TRACKING_URI: http://mlflow:5000
      ARTIFACT_DIR: /app/artifacts
      SERVING_MODEL_BUNDLE_PATH: /app/serving_models/model_bundle.joblib
      FEEDBACK_THRESHOLD: "300"
    volumes:
      - ./training/artifacts:/app/artifacts
      - ./data/artifacts:/app/data_artifacts:ro
      # Same host directory as serving's /models mount
      - ./serving/models:/app/serving_models
    depends_on:
      postgres:
        condition: service_healthy
      mlflow:
        condition: service_healthy
    profiles:
      - training

  retrain-cron:
    build:
      context: ./training
      dockerfile: Dockerfile.retrain
    environment:
      DB_URL: postgresql://tagger:tagger@postgres:5432/tagger
      MLFLOW_TRACKING_URI: http://mlflow:5000
      ARTIFACT_DIR: /app/artifacts
      SERVING_MODEL_BUNDLE_PATH: /app/serving_models/model_bundle.joblib
      FEEDBACK_THRESHOLD: "300"
    volumes:
      - ./training/artifacts:/app/artifacts
      - ./data/artifacts:/app/data_artifacts:ro
      - ./serving/models:/app/serving_models
      - ./data:/app/data_pipeline:ro
      - ./data/mlflow_artifacts:/mlflow/artifacts
    depends_on:
      postgres:
        condition: service_healthy
      mlflow:
        condition: service_healthy
    entrypoint: >
      sh -c "while true; do
        echo '[retrain-cron] Checking feedback trigger at $$(date)';
        python retrain_trigger.py;
        echo '[retrain-cron] Sleeping 1h.';
        sleep 3600;
      done"

  monitor:
    build:
      context: ./serving
      dockerfile: Dockerfile.monitor
    environment:
      DB_URL: postgresql://tagger:tagger@postgres:5432/tagger
      MLFLOW_TRACKING_URI: http://mlflow:5000
      SERVING_URL: http://serving:8000
      PROMETHEUS_URL: http://prometheus:9090
      CHECK_INTERVAL_SECONDS: "300"
      ERROR_RATE_THRESHOLD: "0.05"
      CORRECTION_RATE_THRESHOLD: "0.65"
    volumes:
      - ./data/mlflow_artifacts:/mlflow/artifacts
      - ./serving/models:/serving_models
    depends_on:
      - serving
      - prometheus
      - postgres
      - mlflow
    entrypoint: ["python", "monitor.py"]
    healthcheck:
      test: ["CMD-SHELL", "python -c 'import sys; sys.exit(0)'"]
      interval: 30s
      timeout: 10s
      retries: 3

  qdrant:
    image: qdrant/qdrant:latest
    ports:
      - "6333:6333"

  qdrant-demo:
    build:
      context: ./data
      dockerfile: Dockerfile.qdrant
    entrypoint: ["python", "qdrant_demo.py"]
    volumes:
      - ./data/artifacts:/app/artifacts
    depends_on:
      - qdrant
    command:
      - --input
      - /app/artifacts/ocw_dataset.parquet
      - --qdrant-host
      - qdrant

  synthetic-expansion:
    build:
      context: ./data
    entrypoint: ["python", "synthetic_expansion.py"]
    volumes:
      - ./data/artifacts:/app/artifacts
    command:
      - --input
      - /app/artifacts/ocw_dataset.parquet
      - --output
      - /app/artifacts/ocw_dataset_expanded.parquet
      - --multiplier
      - "5"
      - --seed
      - "42"

  batch-pipeline:
    build:
      context: ./data
    entrypoint: ["python", "batch_pipeline.py"]
    environment:
      DB_URL: postgresql://tagger:tagger@postgres:5432/tagger
    volumes:
      - ./data/artifacts:/app/artifacts
    depends_on:
      postgres:
        condition: service_healthy
    command:
      - --dataset
      - /app/artifacts/ocw_dataset_expanded.parquet
      - --feedback-source
      - postgres
      - --db-url
      - postgresql://tagger:tagger@postgres:5432/tagger
      - --output-dir
      - /app/artifacts/versions
      - --version
      - v6

  drift-monitor:
    build:
      context: ./data
    entrypoint: ["python", "drift_monitor.py"]
    environment:
      DB_URL: postgresql://tagger:tagger@postgres:5432/tagger
    volumes:
      - ./data/artifacts:/app/artifacts
    depends_on:
      postgres:
        condition: service_healthy
    command:
      - --baseline
      - /app/artifacts/versions/v6/train_llm_merged_ops.parquet
      - --output
      - /app/artifacts/drift_report.json
      - --version
      - v6
      - --interval
      - "3600"

  mock-predict:
    build:
      context: ./data
    entrypoint: ["python", "mock_predict_server.py"]
    ports:
      - "8001:8000"

  data-generator:
    build:
      context: ./data
    entrypoint: ["python", "load_generator.py"]
    environment:
      PREDICT_URL: http://serving:8000/predict-text
      FEEDBACK_URL: http://serving:8000/feedback
      DATA_PATH: /app/artifacts/ocw_dataset.parquet
      DELAY_SECS: "30"
      USER_ID: load-generator
    volumes:
      - ./data/artifacts:/app/artifacts
    depends_on:
      serving:
        condition: service_healthy
    command:
      - --input
      - /app/artifacts/ocw_dataset.parquet
      - --output
      - /app/artifacts/production_feedback.jsonl
      - --predict-url
      - http://serving:8000/predict
      - --feedback-url
      - http://serving:8000/feedback

  ocw-pipeline:
    build:
      context: ./data
    volumes:
      - ./data:/app/workspace:ro
      - ./data/artifacts:/app/artifacts
    command:
      - --root
      - /app/workspace
      - --output
      - /app/artifacts/ocw_dataset.parquet
      - --summary-output
      - /app/artifacts/ocw_summary.json

  prometheus:
    image: prom/prometheus:latest
    ports:
      - "9090:9090"
    volumes:
      - ./infra/prometheus.yml:/etc/prometheus/prometheus.yml:ro
    depends_on:
      - serving

  grafana:
    image: grafana/grafana:latest
    ports:
      - "3000:3000"
    environment:
      GF_SECURITY_ADMIN_USER: proj29-grafana
      GF_SECURITY_ADMIN_PASSWORD: proj29-grafana
    volumes:
      - ./data/grafana:/var/lib/grafana
      - ./infra/grafana/provisioning:/etc/grafana/provisioning:ro
    depends_on:
      - prometheus

  adminer:
    image: adminer
    ports:
      - "8081:8080"
    restart: always
    depends_on:
      - postgres