influxdb/meta/prometheus.yml

{%- if pillar.influxdb.server is defined %}
{%- from "influxdb/map.jinja" import server, monitoring with context %}

{%- if server.get('enabled', False) %}
server:
  alert:
{%- if server.get('http', {}).get('enabled', False) %}
    InfluxdbQDown:
      if: >-
        influxdb_up != 1
      labels:
        severity: warning
        service: influxdb
      annotations:
        summary: 'InfluxDB service down'
      {% raw %}
        description: 'InfluxDB service is down on node {{ $labels.host }}'
      {% endraw %}
    InfluxdbHTTPClientError:
      {%- set influx_http_client_error_threshold = monitoring.http_errors_percentage %}
      if: >-
        rate(influxdb_httpd_clientError[2m]) / rate(influxdb_httpd_req[2m]) * 100 > {{ influx_http_client_error_threshold }}
      {% raw %}
      labels:
        severity: warning
        service: influxdb
      annotations:
        summary: 'Influxdb number of client errors is high'
        description: '{{ printf `%.1f` $value }}% of client requests are in error on {{ $labels.host }} (threshold={%- endraw %}{{ influx_http_client_error_threshold }}).'
    InfluxdbHTTPPointsWrittenFail:
      {%- set influx_http_points_written_fail_threshold = monitoring.failed_points_percentage %}
      if: >-
        rate(influxdb_httpd_pointsWrittenFail[2m]) / rate(influxdb_httpd_pointsWrittenOK[2m]) * 100 > {{ influx_http_points_written_fail_threshold }}
      {% raw %}
      labels:
        severity: warning
        service: influxdb
      annotations:
        summary: 'Influxdb too many failed writes'
        description: '{{ printf `%.1f` $value }}% of written points have failed on {{ $labels.host }} (threshold={%- endraw %}{{ influx_http_points_written_fail_threshold }}).'
    InfluxdbHTTPPointsWrittenDropped:
      {%- set influx_http_points_written_dropped_threshold = monitoring.dropped_points_percentage %}
      if: >-
        rate(influxdb_httpd_pointsWrittenDropped[2m]) / rate(influxdb_httpd_pointsWrittenOK[2m]) * 100 > {{ influx_http_points_written_dropped_threshold }}
      {% raw %}
      labels:
        severity: warning
        service: influxdb
      annotations:
        summary: 'Influxdb too many dropped writes'
        description: '{{ printf `%.1f` $value }}% of written points have been dropped on {{ $labels.host }} (threshold={%- endraw %}{{ influx_http_points_written_dropped_threshold }}).'
{%- endif %}
{%- endif %}
{%- endif %}