LMDeploy: any preemption metric - VERIFIED ABSENT

VERIFIED NEGATIVE. LMDeploy does not emit any preemption metric. lmdeploy_preemptions is NOT a metric this engine registers - it is recorded here under the conventional name a reader would go looking for so the absence is queryable rather than implied by a gap in a list. Turbomind DOES preempt under KV pressure and you cannot see it. This is a gap rather than a naming divergence: unlike SGLang (which measures retraction four ways) or vLLM (which has both a counter and a per-request histogram), LMDeploy's metric set has nothing to count preemptions with. Since it is a vLLM fork whose scheduler is the same shape, the omission is a porting gap rather than a design one. What was searched: grep -ci 'preempt' in lmdeploy/metrics/loggers.py returns 0, and the 33-name inventory transcribed from PrometheusStatLogger contains no preemption entry.. Do not read this record's metric_name as something the engine exports; it is the name that is missing.