Reword Prometheus alert messages
Change-Id: I54e02e0741d53ec7b2335145dc968b7b8c8f5e00
diff --git a/linux/meta/prometheus.yml b/linux/meta/prometheus.yml
index 35eab02..0c3c41d 100644
--- a/linux/meta/prometheus.yml
+++ b/linux/meta/prometheus.yml
@@ -1,15 +1,15 @@
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) }}
+ {%- set cpu_idle_threshold = prometheus_server.get('alert', {}).get('AvgCPUUsageIdle', {}).get('var', {}).get('threshold', 10) %}
+ if: avg_over_time(cpu_usage_idle{cpu="cpu-total"}[5m]) < {{ cpu_idle_threshold }}
{% 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 %}
+ summary: 'Idle CPU usage too low on {{ $labels.host }}'
+ description: 'The average idle CPU usage is too low on node {{ $labels.host }} (current value={{ $value }}, threshold={% endraw %}{{ cpu_idle_threshold}})'
PredictLinearDiskFree:
if: 'predict_linear(disk_free[1h], 8*3600) < 0'
{% raw %}
@@ -17,8 +17,8 @@
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 }}'
+ summary: 'Free space for {{ $labels.path }} too low on {{ $labels.host }}'
+ description: 'The disk partition ({{ $labels.path }}) will be full in less than 8 hours on {{ $labels.host }}'
{% endraw %}
PredictLinearDiskInodesFree:
if: 'predict_linear(disk_inodes_free[1h], 8*3600) < 0'
@@ -27,19 +27,19 @@
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 }}'
+ summary: 'Free inodes for {{ $labels.path }} too low on {{ $labels.host }}'
+ description: 'The disk inodes ({{ $labels.path }}) will be full in less than 8 hours on {{ $labels.host }}'
{% endraw %}
AvgMemAvailablePercent:
- if: avg_over_time(mem_available_percent[5m]) < {{ prometheus_server.get('alert', {}).get('AvgMemAvailablePercent', {}).get('var', {}).get('threshold', 10) }}
+ {%- set mem_avail_threshold = prometheus_server.get('alert', {}).get('AvgMemAvailablePercent', {}).get('var', {}).get('threshold', 10) %}
+ if: avg_over_time(mem_available_percent[5m]) < {{ mem_avail_threshold }}
{% 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 %}
+ summary: 'Free memory too low on {{ $labels.host }}'
+ description: 'The percentage of free memory is too low on node {{ $labels.host }} (current value={{ $value }}, threshold={% endraw %}{{ mem_avail_threshold }})'
SystemLoad5:
if: system_load5 / system_n_cpus > {{ prometheus_server.get('alert', {}).get('SystemLoad5', {}).get('var', {}).get('threshold', 3) }}
{% raw %}