TensorRT-LLM: any preemption metric (counter or gauge) - VERIFIED ABSENT

VERIFIED NEGATIVE. TensorRT-LLM does not emit any preemption metric (counter or gauge). trtllm_num_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. The collector's own class docstring enumerates every metric it creates and preemption is not among them; the nearest signal is trtllm_num_paused_requests, a level. So a TRT-LLM deployment cannot count preemptions or the recompute they cause, which is exactly the saturation signal vLLM and SGLang both expose. What was searched: case-insensitive count of 'preempt' in tensorrt_llm/metrics/collector.py returns 0, and the docstring's own enumeration (lines 44-112) contains no preemption entry.. Do not read this record's metric_name as something the engine exports; it is the name that is missing.