teleport-operator specifics

giantswarm · Jan 11, 2024 · 20d15aa · 20d15aa
1 parent 031ca14
commit 20d15aa
Show file tree

Hide file tree

Showing 3 changed files with 43 additions and 3 deletions.
diff --git a/helm/prometheus-rules/templates/alerting-rules/deployment.management-cluster.rules.yml b/helm/prometheus-rules/templates/alerting-rules/deployment.management-cluster.rules.yml
@@ -215,7 +215,7 @@ spec:
       annotations:
         description: '{{`Deployment {{ $labels.namespace}}/{{ $labels.deployment }} is not satisfied.`}}'
         opsrecipe: deployment-not-satisfied/
-      expr: kube_deployment_status_replicas_unavailable{cluster_type="management_cluster", deployment=~"cert-manager-*|teleport-*|dex*|athena*|rbac-operator"} > 0
+      expr: kube_deployment_status_replicas_unavailable{cluster_type="management_cluster", deployment=~"cert-manager-*|teleport-*|dex*|athena*|rbac-operator|credentiald"} > 0
       for: 30m
       labels:
         area: kaas

diff --git a/helm/prometheus-rules/templates/alerting-rules/deployment.workload-cluster.rules.yml b/helm/prometheus-rules/templates/alerting-rules/deployment.workload-cluster.rules.yml
@@ -73,10 +73,10 @@ spec:
     - alert: WorkloadClusterDeploymentNotSatisfiedBigMac
       annotations:
         description: '{{`Deployment {{ $labels.namespace}}/{{ $labels.deployment }} has been scaled down to zero for prolonged period of time.`}}'
-      expr: kube_deployment_status_replicas_unavailable{cluster_type="workload_cluster", deployment=~"cert-manager-*|teleport-*|dex*|athena*|rbac-operator"} > 0
+      expr: kube_deployment_status_replicas_unavailable{cluster_type="workload_cluster", deployment=~"cert-manager-*|teleport-*|dex*|athena*|rbac-operator|credentiald"} > 0
       for: 30m
       labels:
-        area: kass
+        area: kaas
         cancel_if_cluster_status_creating: "true"
         cancel_if_cluster_status_deleting: "true"
         cancel_if_cluster_status_updating: "true"

diff --git a/helm/prometheus-rules/templates/alerting-rules/teleport.rules.yaml b/helm/prometheus-rules/templates/alerting-rules/teleport.rules.yaml
@@ -0,0 +1,40 @@
+apiVersion: monitoring.coreos.com/v1
+kind: PrometheusRule
+metadata:
+  creationTimestamp: null
+  labels:
+    {{- include "labels.common" . | nindent 4 }}
+  name: prometheus.rules
+  namespace: {{ .Values.namespace  }}
+spec:
+  groups:
+  - name: teleport
+    rules:
+    - alert: TeleportJoinTokenSecretMismatch
+      annotations:
+        description: '{{`Mismatch in number of teleport-join-token secrets and clusters`}}'
+      expr: count(kube_secret_created{secret=~".*-teleport-join-token"}) != count(capi_cluster_info{control_plane_reference_kind="KubeadmControlPlane"})
+      for: 30m
+      labels:
+        area: kaas
+        cancel_if_cluster_status_creating: "true"
+        cancel_if_cluster_status_updating: "true"
+        cancel_if_cluster_status_deleting: "true"
+        cancel_if_outside_working_hours: "true"
+        severity: page
+        team: bigmac
+        topic: teleport
+    - alert: TeleportKubeAgentConfigMapMismatch
+      annotations:
+        description: '{{`Teleport join token is missing`}}'
+      expr: count(kube_configmap_info{configmap=~".*-teleport-kube-agent-config"})!= count(capi_cluster_info{control_plane_reference_kind="KubeadmControlPlane"})
+      for: 30m
+      labels:
+        area: kaas
+        cancel_if_cluster_status_creating: "true"
+        cancel_if_cluster_status_updating: "true"
+        cancel_if_cluster_status_deleting: "true"
+        cancel_if_outside_working_hours: "true"
+        severity: page
+        team: bigmac
+        topic: teleport