#9346: vLLM metrics do not provide an all-HTTP-request denominator

Reported environment

API-serving deployment seeking a Prometheus error-rate SLI.

The nuance

request_success counts finished engine requests; running requests is a gauge; neither counts every HTTP attempt.

Expected behavior

Reject invalid denominators and require ingress/API request telemetry for an all-attempt SLI.

What Schema establishes

Lifecycle semantics rule out proposed denominators; complete SLI needs external API-layer telemetry.

Runtime evidence still needed

• API or ingress request counter by stable outcome
• Retry disconnect validation and engine-failure accounting

Try the questions (2)

Each question runs live against the product.

Grounding claims

• In vLLM v0.23.0, vllm:request_success is registered as a Prometheus counter.
• In vLLM v0.23.0, vllm:num_requests_running is a Prometheus gauge for the number of requests in model execution batches.

Topics, entities, and gaps