From 7282e864837c3e19b58cff56ce0fc31097469ac9 Mon Sep 17 00:00:00 2001 From: nd Date: Mon, 5 Jul 2021 21:06:08 +0200 Subject: [PATCH] saner cpu load alerts --- defaults/main.yml | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/defaults/main.yml b/defaults/main.yml index c6cd612..6ed466f 100644 --- a/defaults/main.yml +++ b/defaults/main.yml @@ -98,8 +98,7 @@ prometheus_agent: alert: CPULoad enabled: True expr: | - 100 - avg without (cpu) (irate(node_cpu_seconds_total{mode="idle", job="node", instance="{{ inventory_hostname }}"}[5m])) * 100 > 80 or - node_load5{job="node", instance="{{ inventory_hostname }}"} > count without(cpu, mode) (node_cpu_seconds_total{mode="idle", job="node", instance="{{ inventory_hostname }}"}) + node_load5{job="node", instance="{{ inventory_hostname }}"} > count without(cpu, mode) (node_cpu_seconds_total{mode="idle", job="node", instance="{{ inventory_hostname }}"}) * 10 for: 5m labels: severity: warning @@ -111,8 +110,7 @@ prometheus_agent: alert: CPULoad enabled: True expr: | - 100 - avg without (cpu) (irate(node_cpu_seconds_total{mode="idle", job="node", instance="{{ inventory_hostname }}"}[5m])) * 100 > 80 or - node_load5{job="node", instance="{{ inventory_hostname }}"} > count without(cpu, mode) (node_cpu_seconds_total{mode="idle", job="node", instance="{{ inventory_hostname }}"}) + node_load5{job="node", instance="{{ inventory_hostname }}"} > count without(cpu, mode) (node_cpu_seconds_total{mode="idle", job="node", instance="{{ inventory_hostname }}"}) * 20 for: 5m labels: severity: critical