Merge pull request #1237 from raptorsun/test/AggAPIDNAlert

AggregatedAPIDown alert threshold set back to 85%
openshift · Jun 29, 2021 · f4b1311 · f4b1311
2 parents 625f934 + 66086e9
commit f4b1311
Show file tree

Hide file tree

Showing 2 changed files with 1 addition and 14 deletions.
diff --git a/assets/control-plane/prometheus-rule.yaml b/assets/control-plane/prometheus-rule.yaml
@@ -400,7 +400,7 @@ spec:
           has been only {{ $value | humanize }}% available over the last 10m.
         summary: An aggregated API is down.
       expr: |
-        (1 - max by(name, namespace)(avg_over_time(aggregator_unavailable_apiservice[10m]))) * 100 < 70
+        (1 - max by(name, namespace)(avg_over_time(aggregator_unavailable_apiservice[10m]))) * 100 < 85
       for: 5m
       labels:
         severity: warning

diff --git a/jsonnet/patch-rules.libsonnet b/jsonnet/patch-rules.libsonnet
@@ -81,19 +81,6 @@ local patchedRules = [
       },
     ],
   },
-  {
-    name: 'kubernetes-system-apiserver',
-    rules: [
-      {
-        // Lower treshold to be resilient to DNS rollouts and CA rotations.
-        // https://bugzilla.redhat.com/show_bug.cgi?id=1970624
-        alert: 'AggregatedAPIDown',
-        expr: |||
-          (1 - max by(name, namespace)(avg_over_time(aggregator_unavailable_apiservice[10m]))) * 100 < 70
-        |||,
-      },
-    ],
-  },
   {
     name: 'prometheus',
     rules: [