#8115: TTFT observations and completed requests count different events

Reported environment

A scrape where TTFT histogram count exceeded the sum of request-success series.

The nuance

TTFT is observed on first-token production; request_success increments from finished requests, so lifecycle and failure paths can separate them.

Expected behavior

Trace both update events and require reconciliation of lifecycle, labels, process scope, and abnormal outcomes.

What Schema establishes

Schema proves different update events; a particular delta needs deployment observations.

Runtime evidence still needed

• Counts by finish reason process and restart window
• Lifecycle test including abort error streaming and fan-out

Try the questions (2)

Each question runs live against the product.

Grounding claims

• In vLLM v0.23.0, vllm:time_to_first_token_seconds is registered as a Prometheus histogram.
• In vLLM v0.23.0, vllm:request_success is registered as a Prometheus counter.

Topics, entities, and gaps