server: alert: AvgCPUUsageIdle: if: avg_over_time(cpu_usage_idle{cpu="cpu-total"}[5m]) < {{ prometheus_server.get('alert', {}).get('AvgCPUUsageIdle', {}).get('var', {}).get('threshold', 10) }} {% raw %} labels: severity: warning service: system annotations: summary: 'Avarage CPU usage (idle) for node {{ $labels.host }} is low' description: 'Avarage CPU usage (idle) for node {{ $labels.host }} is low {{ $value }}' {% endraw %} PredictLinearDiskFree: if: 'predict_linear(disk_free[1h], 8*3600) < 0' {% raw %} labels: severity: warning service: system annotations: summary: 'Disk space ({{ $labels.path }}) is filling on {{ $labels.host }}' description: 'Disk space ({{ $labels.path }}) will be full in 8h on {{ $labels.host }}' {% endraw %} PredictLinearDiskInodesFree: if: 'predict_linear(disk_inodes_free[1h], 8*3600) < 0' {% raw %} labels: severity: warning service: system annotations: summary: 'Disk inodes ({{ $labels.path }}) are filling on {{ $labels.host }}' description: 'Disk inodes ({{ $labels.path }}) will be full in 8h on {{ $labels.host }}' {% endraw %} AvgMemAvailablePercent: if: avg_over_time(mem_available_percent[5m]) < {{ prometheus_server.get('alert', {}).get('AvgMemAvailablePercent', {}).get('var', {}).get('threshold', 10) }} {% raw %} labels: severity: warning service: system annotations: summary: 'Free memory is low on {{ $labels.host }}' description: 'Free memory percent for node {{ $labels.host }} is low {{ $value }}' {% endraw %} SystemLoad5: if: system_load5 / system_n_cpus > {{ prometheus_server.get('alert', {}).get('SystemLoad5', {}).get('var', {}).get('threshold', 3) }} {% raw %} labels: severity: warning service: system annotations: summary: 'High system load (5m) on {{ $labels.host }}' description: 'High system load (5m) on node {{ $labels.host }}' {% endraw %}