docs · v0.10.14
Browse documentation
In progress · page being written
Observability
Wire LLMKube into your existing Prometheus + Grafana stack. The Helm chart ships a PodMonitor for inference pods, and the controller exposes its own Prometheus metrics.
What this page will cover
- Enabling the bundled PodMonitor with prometheus.inferencePodMonitor.enabled in values.yaml.
- Scraping the controller metrics endpoint with prometheus.serviceMonitor.enabled.
- Controller-side metrics: reconcile timing, model download duration, InferenceService phase counts.
- Pod-side metrics from llama.cpp /metrics (always enabled via the --metrics flag).