From d4acf518d09515560e1082a80b8a4d6550e20d9b Mon Sep 17 00:00:00 2001 From: Jae-Won Chung Date: Mon, 17 Nov 2025 04:54:15 -0500 Subject: [PATCH] [Metrics] Fix KV cache usage percent metric multiproc (#28792) The `vllm:kv_cache_usage_perc` Gauge metric is missing `multiprocess_mode="mostrecent"` and ends up returning ``` vllm:kv_cache_usage_perc{engine="0",model_name="Qwen/Qwen3-VL-8B-Instruct",pid="277"} 0.0 vllm:kv_cache_usage_perc{engine="0",model_name="Qwen/Qwen3-VL-8B-Instruct",pid="275"} 0.0 vllm:kv_cache_usage_perc{engine="0",model_name="Qwen/Qwen3-VL-8B-Instruct",pid="273"} 0.6530455880475035 ... ``` The deprecated `vllm:gpu_cache_usage_perc` Gauge metric has `multiprocess_mode="mostrecent"`. Signed-off-by: Jae-Won Chung --- vllm/v1/metrics/loggers.py | 1 + 1 file changed, 1 insertion(+) diff --git a/vllm/v1/metrics/loggers.py b/vllm/v1/metrics/loggers.py index 21280b9c84cf2..cb36e7973650e 100644 --- a/vllm/v1/metrics/loggers.py +++ b/vllm/v1/metrics/loggers.py @@ -494,6 +494,7 @@ class PrometheusStatLogger(AggregateStatLoggerBase): gauge_kv_cache_usage = self._gauge_cls( name="vllm:kv_cache_usage_perc", documentation="KV-cache usage. 1 means 100 percent usage.", + multiprocess_mode="mostrecent", labelnames=labelnames, ) self.gauge_kv_cache_usage = make_per_engine(