formula-prometheus/prometheus/meta/prometheus.yml
Olivier Bourdon 5b6b583c42 Add Prometheus alerts
Change-Id: I4ad10555d728d62c8e6504659d30558f95b410ac
2017-07-26 13:45:51 +02:00

41 lines
1.9 KiB
YAML

{%- if pillar.prometheus is defined %}
{%- from "prometheus/map.jinja" import server, remote_storage_adapter, monitoring with context %}
server:
alert:
{%- if server.get('enabled', False) %}
{% raw %}
PrometheusTargetDown:
if: 'up != 1'
labels:
severity: critical
service: prometheus
annotations:
summary: 'Prometheus endpoint {{ $labels.instance }} is down'
description: 'Prometheus endpoint {{ $labels.instance }} is down for job {{ $labels.job }}'
{% endraw %}
{%- endif %}
{%- if remote_storage_adapter.get('enabled', False) %}
RemoteStorageAdapterSendingTooSlow:
{%- set threshold = monitoring.remote_storage_adapter.sent_vs_received_ratio|float %}
if: >-
100.0 - (100.0 * sent_samples_total{job="remote_storage_adapter"} / on (job, instance) received_samples_total) > {{ threshold }}
{% raw %}
labels:
severity: warning
service: remote_storage_adapter
annotations:
summary: 'Remote storage adapter too slow on {{ $labels.instance }}'
description: 'Remote storage adapter can not ingest samples fast enough on {{ $labels.instance }} (current value={{ $value }}%, threshold={%- endraw %}{{ threshold }}%).'
RemoteStorageAdapterIgnoredTooHigh:
{%- set threshold = monitoring.remote_storage_adapter.ignored_vs_sent_ratio|float %}
if: >-
100.0 * prometheus_influxdb_ignored_samples_total{job="remote_storage_adapter"} / on (job, instance) sent_samples_total > {{ threshold }}
{% raw %}
labels:
severity: warning
service: remote_storage_adapter
annotations:
summary: 'Remote storage adapter receiving too many invalid metrics on {{ $labels.instance }}'
description: 'Remote storage adapter is receiving too many invalid metrics on {{ $labels.instance }} (current value={{ $value }}%, threshold={%- endraw %}{{ threshold }}%).'
{%- endif %}
{%- endif %}