[Bugfix] Fix incorrect kv cache metrics in grafana.json (vllm-project#27133)

Signed-off-by: Fangping Shi <[email protected]>
Co-authored-by: Fangping Shi <[email protected]>

Author: 2 people

examples/online_serving/dashboards/perses/performance_statistics.yaml

@@ -530,7 +530,7 @@ spec:
                     name: accelerators-thanos-querier-datasource
                   # Multiply by 100 so we can read it as a percentage without setting a unit (avoids CUE unit conflicts)
                   query: >
-                    100 * avg(vllm:gpu_cache_usage_perc)
+                    100 * avg(vllm:kv_cache_usage_perc)
 
     "18":
       kind: Panel

examples/online_serving/dashboards/perses/query_statistics.yaml

@@ -98,7 +98,7 @@ spec:
                 kind: PrometheusTimeSeriesQuery
                 spec:
                   datasource: { kind: PrometheusDatasource, name: accelerators-thanos-querier-datasource }
-                  query: avg(vllm:gpu_cache_usage_perc{namespace="$NS",service="$SVC"}) or vector(0)
+                  query: avg(vllm:kv_cache_usage_perc{namespace="$NS",service="$SVC"}) or vector(0)
                   minStep: "15s"
 
     core_running_ts:
@@ -168,7 +168,7 @@ spec:
                 spec:
                   datasource: { kind: PrometheusDatasource, name: accelerators-thanos-querier-datasource }
                   # multiply by 100 to present percentage; omit format.unit to avoid schema conflicts
-                  query: (avg(vllm:gpu_cache_usage_perc{namespace="$NS",service="$SVC"}) * 100) or vector(0)
+                  query: (avg(vllm:kv_cache_usage_perc{namespace="$NS",service="$SVC"}) * 100) or vector(0)
                   minStep: "15s"
 
     core_kv_usage_pct_ts:
@@ -187,7 +187,7 @@ spec:
                 kind: PrometheusTimeSeriesQuery
                 spec:
                   datasource: { kind: PrometheusDatasource, name: accelerators-thanos-querier-datasource }
-                  query: (avg by (service) (vllm:gpu_cache_usage_perc{namespace="$NS",service="$SVC"}) * 100) or vector(0)
+                  query: (avg by (service) (vllm:kv_cache_usage_perc{namespace="$NS",service="$SVC"}) * 100) or vector(0)
                   minStep: "15s"
 
     # --- Per-Pod breakdowns (works on Simulator & Real) ---
@@ -246,7 +246,7 @@ spec:
                 spec:
                   datasource: { kind: PrometheusDatasource, name: accelerators-thanos-querier-datasource }
                   # if your exporter labels kv metric with pod (the sim does), this works; otherwise it will just return empty
-                  query: (avg by (pod) (vllm:gpu_cache_usage_perc{namespace="$NS",service="$SVC"}) * 100) or vector(0)
+                  query: (avg by (pod) (vllm:kv_cache_usage_perc{namespace="$NS",service="$SVC"}) * 100) or vector(0)
                   minStep: "15s"
 
     # --- Real vLLM only (zeros on simulator) ---

examples/online_serving/prometheus_grafana/grafana.json

@@ -852,7 +852,7 @@
             "uid": "${DS_PROMETHEUS}"
           },
           "editorMode": "code",
-          "expr": "vllm:gpu_cache_usage_perc{model_name=\"$model_name\"}",
+          "expr": "vllm:kv_cache_usage_perc{model_name=\"$model_name\"}",
           "instant": false,
           "legendFormat": "GPU Cache Usage",
           "range": true,