Files
auricom-home-cluster/kubernetes/apps/monitoring/loki/app/config-map.yaml
2022-12-28 17:12:57 +01:00

131 lines
4.1 KiB
YAML

---
apiVersion: v1
kind: ConfigMap
metadata:
name: loki-alerting-rules
namespace: monitoring
data:
loki-alerting-rules.yaml: |-
groups:
#
# SMART Failures
#
- name: smart-failure
rules:
- alert: SmartFailures
expr: |
sum by (hostname) (count_over_time({hostname=~".+"} | json | _SYSTEMD_UNIT = "smartmontools.service" !~ "(?i)previous self-test completed without error" !~ "(?i)Prefailure" |~ "(?i)(error|fail)"[2m])) > 0
for: 2m
labels:
severity: critical
category: logs
annotations:
summary: "SMART has reported failures on host {{ $labels.hostname }}"
#
# zigbee2mqtt
#
- name: zigbee2mqtt
rules:
- alert: ZigbeeUnableToReachMQTT
expr: |
sum(count_over_time({app="zigbee2mqtt"} |~ "(?i)not connected to mqtt server"[2m])) > 0
for: 2m
labels:
severity: critical
category: logs
annotations:
summary: "Zigbee2mqtt is unable to reach MQTT"
#
# zwavejs2mqtt
#
- name: zwavejs2mqtt
rules:
- alert: ZwaveUnableToReachMQTT
expr: |
sum(count_over_time({app="zwavejs2mqtt"} |~ "(?i)error while connecting mqtt"[2m])) > 0
for: 2m
labels:
severity: critical
category: logs
annotations:
summary: "Zwavejs2mqtt is unable to reach MQTT"
#
# frigate
#
- name: frigate
rules:
- alert: FrigateUnableToReachMQTT
expr: |
sum(count_over_time({app="frigate"} |~ "(?i)unable to connect to mqtt server"[2m])) > 0
for: 2m
labels:
severity: critical
category: logs
annotations:
summary: "Frigate is unable to reach MQTT"
#
# *arr
#
- name: arr
rules:
- alert: ArrDatabaseIsLocked
expr: |
sum by (app) (count_over_time({app=~".*arr"} |~ "(?i)database is locked"[2m])) > 0
for: 2m
labels:
severity: critical
category: logs
annotations:
summary: "{{ $labels.app }} is experiencing locked database issues"
- alert: ArrDatabaseIsMalformed
expr: |
sum by (app) (count_over_time({app=~".*arr"} |~ "(?i)database disk image is malformed"[2m])) > 0
for: 2m
labels:
severity: critical
category: logs
annotations:
summary: "{{ $labels.app }} is experiencing malformed database disk image issues"
#
# home-assistant
#
- name: home-assistant
rules:
- alert: HomeAssistantUnableToReachPostgresql
expr: |
sum by (app) (count_over_time({app="home-assistant"} |~ "(?i)error in database connectivity"[2m])) > 0
for: 2m
labels:
severity: critical
category: logs
annotations:
summary: "Home Assistant is unable to connect to postgresql"
#
# valetudo
#
- name: valetudo
rules:
- alert: ValetudoUnableToReachMQTT
expr: |
sum by (hostname) (count_over_time({hostname="valetudo"} |~ "(?i).*error.*mqtt.*"[2m])) > 0
for: 2m
labels:
severity: critical
category: logs
annotations:
summary: "Valetudo is unable to connect to mqtt"
#
# node-red
#
- name: node-red
rules:
- alert: NodeRedUnableToReachHomeAssistant
expr: |
sum by (app) (count_over_time({app="node-red"} |~ "(?i)home assistant.*connecting to undefined"[2m])) > 0
for: 2m
labels:
severity: critical
category: logs
annotations:
summary: "Node-Red is unable to connect to Home Assistant"