1 {%- if pillar.influxdb.server is defined %}
2 {%- from "influxdb/map.jinja" import server, monitoring with context %}
4 {%- if server.get('enabled', False) %}
7 ProcstatRunningInfluxdb:
9 procstat_running{process_name="influxdb"} == 0
15 summary: 'Influxdb service is down'
16 description: 'Influxdb service is down on node {{ $labels.host }}'
18 {%- if server.get('http', {}).get('enabled', False) %}
19 InfluxdbHTTPClientError:
20 {%- set influx_http_client_error_threshold = monitoring.http_errors_percentage %}
22 rate(influxdb_httpd_clientError[2m]) / rate(influxdb_httpd_req[2m]) * 100 > {{ influx_http_client_error_threshold }}
28 summary: 'Influxdb number of client errors is high'
29 description: '{{ printf `%.1f` $value }}% of client requests are in error on {{ $labels.host }} (threshold={%- endraw %}{{ influx_http_client_error_threshold }}).'
30 InfluxdbHTTPPointsWrittenFail:
31 {%- set influx_http_points_written_fail_threshold = monitoring.failed_points_percentage %}
33 rate(influxdb_httpd_pointsWrittenFail[2m]) / rate(influxdb_httpd_pointsWrittenOK[2m]) * 100 > {{ influx_http_points_written_fail_threshold }}
39 summary: 'Influxdb too many failed writes'
40 description: '{{ printf `%.1f` $value }}% of written points have failed on {{ $labels.host }} (threshold={%- endraw %}{{ influx_http_points_written_fail_threshold }}).'
41 InfluxdbHTTPPointsWrittenDropped:
42 {%- set influx_http_points_written_dropped_threshold = monitoring.dropped_points_percentage %}
44 rate(influxdb_httpd_pointsWrittenDropped[2m]) / rate(influxdb_httpd_pointsWrittenOK[2m]) * 100 > {{ influx_http_points_written_dropped_threshold }}
50 summary: 'Influxdb too many dropped writes'
51 description: '{{ printf `%.1f` $value }}% of written points have been dropped on {{ $labels.host }} (threshold={%- endraw %}{{ influx_http_points_written_dropped_threshold }}).'