deployment.yaml

# =============================================================================
# src/config/deployment.yaml — Deployment configuration
# =============================================================================
# Controls how the prediction service identifies and loads models,
# and how the server is configured for local and containerized deployment.
# =============================================================================

server:
  host: "0.0.0.0"
  port: 8000
  log_level: "info"

model:
  # Only serve models in this MLflow stage.
  allowed_stage: "Production"
  # Fail startup if no Production model is found.
  require_production_model: true
  # Maximum seconds to wait for model to load at startup.
  startup_timeout_seconds: 120

healthcheck:
  # Include model metadata (version, run_id, algorithm) in health response.
  include_model_info: true

reload:
  # Opt-in: after a successful pipeline run, POST the reload endpoint so a
  # running API picks up a freshly promoted model with no restart.
  # Disabled by default — keeps the pipeline side-effect-free and CI-safe.
  enabled: false
  # Reload endpoint of the running prediction API.
  url: "http://localhost:8000/admin/reload"
  # Max seconds to wait for the reload HTTP call before giving up.
  timeout_seconds: 5