agent.yaml

# DGPU Agent Configuration

agent:
  # Agent ID (unique identifier)
  # If empty, will be auto-generated from hostname
  id: ""
  # Heartbeat interval in seconds
  heartbeat_interval: 5

scheduler:
  # Primary scheduler address
  master_address: "scheduler-master:9090"
  # Standby scheduler address
  standby_address: "scheduler-standby:9090"
  # Connection timeout in seconds
  connection_timeout: 10
  # Reconnection retry interval in seconds
  retry_interval: 2
  # Maximum retry interval in seconds (exponential backoff)
  max_retry_interval: 30

gpu:
  # GPU detection method: nvml or nvidia-smi
  detection_method: "nvml"
  # Health check interval in seconds
  health_check_interval: 30

executor:
  # Task execution method: docker or process
  execution_method: "docker"
  # Working directory for tasks
  work_dir: "/var/lib/dgpu-agent/tasks"
  # Docker configuration (if execution_method is docker)
  docker:
    # Docker socket path
    socket: "/var/run/docker.sock"
    # Default image
    default_image: "nvidia/cuda:12.0-runtime-ubuntu22.04"

logging:
  # Log level: debug, info, warn, error
  level: "info"
  # Log format: json, text
  format: "json"
  # Log output: stdout, stderr, or file path
  output: "stdout"