Cleanup Prometheus alerts #45
2 changed files with 26 additions and 20 deletions
|
@ -2,10 +2,10 @@
|
||||||
{{ ansible_managed | comment }}
|
{{ ansible_managed | comment }}
|
||||||
|
|
||||||
{% macro raw(string) -%}
|
{% macro raw(string) -%}
|
||||||
{{ string }}
|
{{ "{{" }} {{ string }} {{ "}}" }}
|
||||||
{%- endmacro %}
|
{%- endmacro %}
|
||||||
|
|
||||||
{% set instance = '[{{ raw("$label.instance") }}]' %}
|
{% set instance = "[{{ $labels.instance }}]" %}
|
||||||
|
|
||||||
groups:
|
groups:
|
||||||
- name: alert.rules
|
- name: alert.rules
|
||||||
|
@ -15,10 +15,10 @@ groups:
|
||||||
expr: up == 0
|
expr: up == 0
|
||||||
for: 3m
|
for: 3m
|
||||||
labels:
|
labels:
|
||||||
severity: critical
|
severity: critical
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} Invisible depuis plus de 3 minutes
|
{{ instance }} Invisible depuis plus de 3 minutes
|
||||||
|
|
||||||
- alert: OutOfMemory
|
- alert: OutOfMemory
|
||||||
expr: >-
|
expr: >-
|
||||||
|
@ -32,7 +32,8 @@ groups:
|
||||||
severity: warning
|
severity: warning
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} {{ raw("$value | round") }}% de mémoire libre
|
{{ instance }} {{ raw('$value | printf "%.1f"') }}% de mémoire
|
||||||
|
libre
|
||||||
|
|
||||||
- alert: OutOfDiskSpace
|
- alert: OutOfDiskSpace
|
||||||
expr: >-
|
expr: >-
|
||||||
|
@ -42,8 +43,8 @@ groups:
|
||||||
severity: warning
|
severity: warning
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} Espace libre de {{ raw("$labels.mountpoint") }} sur
|
{{ instance }} Espace libre de {{ raw('$labels.mountpoint') }} sur
|
||||||
à {{ raw("$value | round") }}%
|
à {{ raw('$value | printf "%.1f"') }}%
|
||||||
|
|
||||||
- alert: OutOfInodes
|
- alert: OutOfInodes
|
||||||
expr: node_filesystem_files_free / node_filesystem_files * 100 < 10
|
expr: node_filesystem_files_free / node_filesystem_files * 100 < 10
|
||||||
|
@ -52,8 +53,8 @@ groups:
|
||||||
severity: warning
|
severity: warning
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} {{ raw("$value | round") }}% d'inodes restants
|
{{ instance }} {{ raw('$value | printf "%.1f"') }}% d'inodes
|
||||||
pour {{ raw("$labels.mountpoint") }}
|
restants pour {{ raw('$labels.mountpoint') }}
|
||||||
|
|
||||||
- alert: CpuUsage
|
- alert: CpuUsage
|
||||||
expr: >-
|
expr: >-
|
||||||
|
@ -67,7 +68,7 @@ groups:
|
||||||
severity: warning
|
severity: warning
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} CPU à {{ raw("$value | round") }}%
|
{{ instance }} CPU à {{ raw('$value | printf "%.1f"') }}%
|
||||||
|
|
||||||
- alert: SystemdServiceFailed
|
- alert: SystemdServiceFailed
|
||||||
expr: node_systemd_unit_state{state="failed"} == 1
|
expr: node_systemd_unit_state{state="failed"} == 1
|
||||||
|
@ -76,7 +77,7 @@ groups:
|
||||||
severity: warning
|
severity: warning
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} {{ raw("$label.name") }} a échoué
|
{{ instance }} {{ raw('$labels.name') }} a échoué
|
||||||
|
|
||||||
- alert: LoadUsage
|
- alert: LoadUsage
|
||||||
expr: node_load1 > 5
|
expr: node_load1 > 5
|
||||||
|
@ -85,7 +86,7 @@ groups:
|
||||||
severity: warning
|
severity: warning
|
||||||
annotations:
|
annotations:
|
||||||
summary: >
|
summary: >
|
||||||
{{ instance }} Charge à {{ raw("$value") }}
|
{{ instance }} Charge à {{ raw('$value') }}
|
||||||
|
|
||||||
- alert: UpsOutputSourceChanged
|
- alert: UpsOutputSourceChanged
|
||||||
expr: upsOutputSource != 3
|
expr: upsOutputSource != 3
|
||||||
|
@ -121,7 +122,7 @@ groups:
|
||||||
severity: critical
|
severity: critical
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} Charge de {{ raw("$value | round") }}%
|
{{ instance }} Charge de {{ raw('$value | printf "%.1f"') }}%
|
||||||
|
|
||||||
- alert: UpsWrongInputVoltage
|
- alert: UpsWrongInputVoltage
|
||||||
expr: (upsInputVoltage < 210) or (upsInputVoltage > 250)
|
expr: (upsInputVoltage < 210) or (upsInputVoltage > 250)
|
||||||
|
@ -130,7 +131,7 @@ groups:
|
||||||
severity: warning
|
severity: warning
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} Tension d'entrée de {{ raw("$value") }}V
|
{{ instance }} Tension d'entrée de {{ raw('$value') }}V
|
||||||
|
|
||||||
- alert: UpsWrongOutputVoltage
|
- alert: UpsWrongOutputVoltage
|
||||||
expr: (upsOutputVoltage < 220) or (upsOutputVoltage > 240)
|
expr: (upsOutputVoltage < 220) or (upsOutputVoltage > 240)
|
||||||
|
@ -139,7 +140,7 @@ groups:
|
||||||
severity: warning
|
severity: warning
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} Tension de sortie de {{ raw("$value") }}V
|
{{ instance }} Tension de sortie de {{ raw('$value') }}V
|
||||||
|
|
||||||
- alert: UpsTimeRemainingWarning
|
- alert: UpsTimeRemainingWarning
|
||||||
expr: upsEstimatedMinutesRemaining < 8
|
expr: upsEstimatedMinutesRemaining < 8
|
||||||
|
@ -148,7 +149,7 @@ groups:
|
||||||
severity: warning
|
severity: warning
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} Autonomie restante de {{ raw("$value") }} min
|
{{ instance }} Autonomie restante de {{ raw('$value') }} min
|
||||||
|
|
||||||
- alert: UpsTimeRemainingCritical
|
- alert: UpsTimeRemainingCritical
|
||||||
expr: upsEstimatedMinutesRemaining < 5
|
expr: upsEstimatedMinutesRemaining < 5
|
||||||
|
@ -157,5 +158,5 @@ groups:
|
||||||
severity: critical
|
severity: critical
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
{{ instance }} Autonomie restante de {{ raw("$value") }} min
|
{{ instance }} Autonomie restante de {{ raw('$value') }} min
|
||||||
...
|
...
|
||||||
|
|
|
@ -1,6 +1,12 @@
|
||||||
---
|
---
|
||||||
{{ ansible_managed | comment }}
|
{{ ansible_managed | comment }}
|
||||||
|
|
||||||
|
{% macro raw(string) -%}
|
||||||
|
{{ "{{" }} {{ string }} {{ "}}" }}
|
||||||
|
{%- endmacro %}
|
||||||
|
|
||||||
|
{% set instance = "[{{ $labels.instance }}]" %}
|
||||||
|
|
||||||
groups:
|
groups:
|
||||||
- name: alert.rules
|
- name: alert.rules
|
||||||
rules:
|
rules:
|
||||||
|
@ -11,6 +17,5 @@ groups:
|
||||||
severity: critical
|
severity: critical
|
||||||
annotations:
|
annotations:
|
||||||
summary: >-
|
summary: >-
|
||||||
Federate : {{ "{{" }} $labels.instance {{ "}}" }} est invisible
|
{{ instance }} Invisible depuis plus de 3 minutes
|
||||||
depuis plus de 3 minutes !
|
|
||||||
...
|
...
|
||||||
|
|
Loading…
Reference in a new issue