vllm:request_success

Summary

In vLLM v0.23.0, under the V1 metrics path, the PrometheusStatLogger publisher registers vllm:request_success as a Prometheus counter. In that same v0.23.0 PrometheusStatLogger/V1 metrics registration, vllm:request_success carries the label names model_name, engine, and finished_reason. With the V1 metrics PrometheusStatLogger in vLLM v0.23.0, vllm:request_success counts finished engine requests, partitioned by finished_reason. Mechanically, in vLLM v0.23.0 the counter advances by one for each FinishedRequestStats record that the logger consumes. Interpreting it in vLLM v0.23.0, vllm:request_success represents a population of completed engine requests and therefore excludes requests rejected before engine completion. As guidance for vLLM v0.23.0, sum vllm:request_success only over the finish reasons your question intends, and obtain a denominator covering every HTTP attempt from API or ingress telemetry instead.

Version history

• The only version documented here is vLLM v0.23.0, where vllm:request_success is registered as a Prometheus counter with the label names model_name, engine, and finished_reason under the PrometheusStatLogger V1 metrics publisher.
• In that same v0.23.0 snapshot, the counter measures finished engine requests partitioned by finished_reason, incrementing once per FinishedRequestStats record consumed by the logger.
• Interpreting v0.23.0 data, the metric represents a completed-engine-request population that excludes requests rejected before engine completion, so you should sum only the finish reasons relevant to your question and rely on API or ingress telemetry for a denominator covering every HTTP attempt.

Identity

Prometheus type

counter

Labels

model_nameenginefinished_reason

Unit

Not recorded — vLLM registration code has no structured unit field, and units are not inferred from name suffixes

Semantic coverage

Metric knowledge (6)

Related Use Cases (2)

• #8115: TTFT observations and completed requests count different events
• #9346: vLLM metrics do not provide an all-HTTP-request denominator

Availability