Operating an orchestrator means keeping the node online, processing jobs correctly, calling rewards each round, and redeeming tickets. Monitoring helps you spot failures before they cost you income or reputation.
Built-in metrics
With -monitor=true, go-livepeer exposes Prometheus-compatible metrics at:
http://localhost:7935/metrics
What to track
| Layer | What to monitor |
|---|
| Hardware | GPU utilisation, VRAM, temperature (nvidia-smi) |
| Application | go-livepeer health, segment/job success rate |
| Network | Latency, packet loss |
| On-chain | Bonded stake, active set status, reward calls |
| Economics | ETH fees, LPT rewards |
livepeer_segment_processed_total, livepeer_segment_errors_total, livepeer_transcode_latency_seconds, and (for AI) inference job counts and latency. High error rates reduce selection probability.
If you do not call reward() every round, you stop earning inflation rewards. Automate reward calls or use a service that does.
Explorer
Use the Livepeer Explorer to check:
- Active set status
- Bonded and delegated LPT
- Reward calls and fee earnings
- Historical performance
Prometheus and Grafana
For production, run Prometheus with a scrape config for your node(s), and use or build Grafana dashboards. Add Node exporter and NVIDIA DCGM exporter for host and GPU metrics.
See also
