diff --git a/resources/chaosknoten/grafana/docker_compose/prometheus_alerts-fux.rules.yaml b/resources/chaosknoten/grafana/docker_compose/prometheus_alerts-fux.rules.yaml index d6210c7..6d1187c 100644 --- a/resources/chaosknoten/grafana/docker_compose/prometheus_alerts-fux.rules.yaml +++ b/resources/chaosknoten/grafana/docker_compose/prometheus_alerts-fux.rules.yaml @@ -2,20 +2,22 @@ groups: - name: Fux-Generic rules: - alert: HostJobFlaky - expr: group by(host, job) (changes(up{org="fux"}[24h]) > 7) + expr: group by(instance, job) (changes(up{org="fux"}[24h]) > 7) for: 0m labels: severity: info + org: fux annotations: - summary: Job {{ $labels.job }} flaky on (host {{ $labels.instance }}) - description: "The job {{ $labels.job }} on target: {{ $labels.host }} has been flaky over the last 24 hours." + summary: Job {{ $labels.job }} flaky on (instance {{ $labels.instance }}) + description: "The job {{ $labels.job }} on target: {{ $labels.instance }} has been flaky over the last 24 hours." - name: Fux-SNMP rules: - alert: SnmpTargetMissing - expr: up{job=~".*snmp|SNMP.*"} == 0 + expr: up{job=~".*snmp.*", org="fux"} == 0 for: 15m labels: severity: critical + org: fux annotations: summary: SNMP target missing (instance {{ $labels.instance }}) description: "SNMP target: {{ $labels.instance }} has disappeared for more the 15 min."