vllm-observability

What problem does it solve?

Observability for production vLLM deployments is challenging; operators must derive triage guidance from dispersed sources, and dashboards or raw metrics alone often mislead about latency, queue depth, and GPU saturation.

Core Features & Use Cases

• Prometheus-based metrics surface at /metrics with vLLM-specific labels; dashboards and alerting recipes expedite incident triage; tracing guidance helps correlate traces with metrics for end-to-end debugging; and references provide best-practice guidance for DCGM, Grafana dashboards, and OTLP tracing.

Quick Start

Run the metrics smoke test against a vLLM endpoint to verify /health, /metrics, and basic observability surfaces.