◎LightOS Fabrictelemetryd v2.4
SIMULATED FEED

API & Exporters

Slurm · Kubernetes · Docker · Linux collectors

Infrastructure

Live data source

simulated

This console runs on a built-in simulator until you point it at a running lightos-telemetryd WebSocket (e.g. wss://your-aggregator:9100/ws/telemetry/live). If the socket drops, it falls back to simulation and retries every 3 s.

Accepted messages: a full frame ({ ports: [...], nodes: [...] }) or the spec §4.1 single-port ingest payload.

Before connecting real infrastructure

  1. Start with a Linux VM or a test Kubernetes cluster; GPU hardware is only needed to validate GPU measurements.
  2. Install approved node, scheduler, and GPU exporters. Network queues and optical health require vendor-specific access.
  3. Deploy a collector and a separately built LightOS adapter that produces complete telemetry frames. A Prometheus URL cannot be pasted into this WebSocket field.
  4. Serve the adapter over WSS using a trusted certificate and access controls compatible with a browser WebSocket. Never expose a Kubernetes API or Slurm controller publicly.
  5. Provide the address, a sanitized example frame, infrastructure versions, GPU/network models, and permitted read access. Agree on any control permissions separately.

The downloadable infrastructure collector sends measured resources to Infrastructure, not to this WebSocket feed. A fabric-specific adapter is still required for optical, queue and network telemetry. Current fabric actions are local suggestions, not commands executed on hardware.

Prometheus exposition preview (:9100/metrics)

…

Endpoints served by the daemon

  • GET:9100/metricsPrometheus pull
  • WS/ws/telemetry/liveDelta-compressed frames
  • gRPC:4317 OTLPOpenTelemetry export
  • POST/v1/ingest§4.1 ingest payload

These run inside your cluster, on the daemon — not in this web app.