Changes to ManyEdgeConnectTestsFailing alert
Change-Id: I86262e2eccac1a59d5489ef6c7584d2f58c3dda3
diff --git a/monitoring/edge-monitoring-server/templates/prometheusrule.yaml b/monitoring/edge-monitoring-server/templates/prometheusrule.yaml
index ecc6acf..d554431 100644
--- a/monitoring/edge-monitoring-server/templates/prometheusrule.yaml
+++ b/monitoring/edge-monitoring-server/templates/prometheusrule.yaml
@@ -43,12 +43,14 @@
for: 10m
labels:
severity: critical
+{{- if .Values.alerts.manyEdgeConnectTestsFailing }}
- alert: ManyEdgeConnectTestsFailing
annotations:
message: |
- {{`{{ $value | humanizePercentage }}`}} of the clusters are reporting UE connect failures.
+ Over half of the clusters are reporting UE connect failures.
expr: avg(clamp_max(aetheredge_connect_test_ok{endpoint="metrics80"} + aetheredge_in_maintenance_window{endpoint="metrics80"}, 1)) < 0.5
for: 10m
labels:
severity: critical
{{- end }}
+{{- end }}