From b818a8c1fedb068e2797fabe19c1e198fb55926b Mon Sep 17 00:00:00 2001 From: Mayne0213 Date: Sat, 10 Jan 2026 17:54:55 +0900 Subject: [PATCH] fix: update CPU throttling panels to use PSI metrics with 10% threshold --- grafana/dashboards/main.json | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/grafana/dashboards/main.json b/grafana/dashboards/main.json index f5eef2d..310802a 100644 --- a/grafana/dashboards/main.json +++ b/grafana/dashboards/main.json @@ -1987,14 +1987,14 @@ }, "editorMode": "code", "exemplar": true, - "expr": "sum(rate(container_cpu_cfs_throttled_seconds_total{image!=\"\", cluster=\"$cluster\"}[$__rate_interval])) by (namespace) > 0 or vector(0)", + "expr": "sum(rate(container_pressure_cpu_waiting_seconds_total{pod!=\"\", cluster=\"$cluster\"}[$__rate_interval])) by (namespace) > 0.1", "interval": "$resolution", "legendFormat": "{{ namespace }}", "range": true, "refId": "A" } ], - "title": "CPU Throttled seconds by namespace", + "title": "CPU Pressure (waiting) by namespace", "type": "timeseries" }, { @@ -2099,14 +2099,14 @@ }, "editorMode": "code", "exemplar": true, - "expr": "sum(rate(node_cpu_core_throttles_total{cluster=\"$cluster\", job=\"$job\"}[$__rate_interval])) by (instance) or vector(0)", + "expr": "sum(rate(node_pressure_cpu_waiting_seconds_total{cluster=\"$cluster\", job=\"$job\"}[$__rate_interval])) by (instance) > 0.1", "interval": "$resolution", "legendFormat": "{{ instance }}", "range": true, "refId": "A" } ], - "title": "CPU Core Throttled by instance", + "title": "Node CPU Pressure (waiting) by instance", "type": "timeseries" }, {