Ray Dashboard

While OpenTelemetry and Prometheus handle machine learning metrics (like Loss and Accuracy), the Ray Dashboard handles Orchestration Metrics.

If you are running a complex decentralized topology and the system hangs, the Ray Dashboard is the only way to see exactly which actor is blocked or has crashed.

Enabling the Dashboard

In your config.yaml:

ray_dashboard: true
ray_dashboard_port: 8265

Using the Dashboard

Once the make run command executes, navigate to http://localhost:8265 in your browser.

Key Features:

1. Logical View: See every VirtualNode actor spawned by app_factory.py. You can see their exact PID, IP address, and whether they are ALIVE, DEAD, or PENDING.
2. Logs: You can click on individual actors to stream their exact stdout/stderr. This is vital if a specific edge device fails to load a dataset split.
3. Hardware Utilization: View real-time CPU, RAM, and GPU memory utilization per actor, helping you tune your train_batch_size to prevent Out-Of-Memory (OOM) errors.