prometheus-operator.libsonnet: Add PrometheusOperatorWatchErrors alert

2020-07-13 17:30:56 +02:00
parent a87f322edc
commit dfe9184c9b
1 changed files with 13 additions and 0 deletions
--- a/jsonnet/kube-prometheus/alerts/prometheus-operator.libsonnet
+++ b/jsonnet/kube-prometheus/alerts/prometheus-operator.libsonnet
@@ -4,6 +4,19 @@
      {
        name: 'prometheus-operator',
        rules: [
+          {
+            alert: 'PrometheusOperatorWatchErrors',
+            expr: |||
+              (sum by (controller,namespace) (rate(prometheus_operator_watch_operations_failed_total{%(prometheusOperatorSelector)s}[1h])) / sum by (controller,namespace) (rate(prometheus_operator_watch_operations_total{%(prometheusOperatorSelector)s}[1h]))) > 0.1
+            ||| % $._config,
+            labels: {
+              severity: 'warning',
+            },
+            annotations: {
+              message: 'Errors while performing watch operations in controller {{$labels.controller}} in {{$labels.namespace}} namespace.',
+            },
+            'for': '15m',
+          },
          {
            alert: 'PrometheusOperatorReconcileErrors',
            expr: |||