From 08fe22119882d24ef23a454df5e34c05b1bcc7f6 Mon Sep 17 00:00:00 2001
From: Herve Nicol <12008875+hervenicol@users.noreply.github.com>
Date: Mon, 21 Oct 2024 17:24:57 +0200
Subject: [PATCH] Add InhibitionClusterWithoutWorkerNodes for CAPA

---
 CHANGELOG.md                                  |  4 ++
 .../alerting-rules/capa.inhibition.rules.yml  | 34 +++++++++++++++++
 .../capa.inhibition.rules.test.yml            | 38 +++++++++++++++++++
 3 files changed, 76 insertions(+)
 create mode 100644 helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/capa.inhibition.rules.yml
 create mode 100644 test/tests/providers/capi/capa-mimir/kaas/phoenix/alerting-rules/capa.inhibition.rules.test.yml

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 9d8c2850..6d5540b3 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+
+- Added InhibitionClusterWithoutWorkerNodes for CAPA
+
 ### Changed
 
 - Modify `KyvernoWebhookHasNoAvailableReplicas` to check specifically for Kyverno resource webhook.
diff --git a/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/capa.inhibition.rules.yml b/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/capa.inhibition.rules.yml
new file mode 100644
index 00000000..b5767549
--- /dev/null
+++ b/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/capa.inhibition.rules.yml
@@ -0,0 +1,34 @@
+{{- if eq .Values.managementCluster.provider.kind "capa" }}
+apiVersion: monitoring.coreos.com/v1
+kind: PrometheusRule
+metadata:
+  creationTimestamp: null
+  labels:
+    {{- include "labels.common" . | nindent 4 }}
+    cluster_type: "management_cluster"
+  name: capa.inhibitions.rules
+  namespace: {{ .Values.namespace  }}
+spec:
+  groups:
+  - name: capa.inhibitions
+    rules:
+    - alert: InhibitionClusterWithoutWorkerNodes
+      annotations:
+        description: '{{`Cluster ({{ $labels.cluster_id }}) has no worker nodes.`}}'
+      expr: |-
+        label_replace(
+            capi_cluster_status_condition{type="ControlPlaneReady", status="True"},
+                "cluster_id",
+                "$1",
+                "name",
+                "(.*)"
+            ) == 1
+        unless on (cluster_id) (
+            sum(capi_machinepool_spec_replicas{} > 0) by (cluster_id)
+        )
+      labels:
+        area: kaas
+        has_worker_nodes: "false"
+        team: phoenix
+        topic: status
+{{- end }}
diff --git a/test/tests/providers/capi/capa-mimir/kaas/phoenix/alerting-rules/capa.inhibition.rules.test.yml b/test/tests/providers/capi/capa-mimir/kaas/phoenix/alerting-rules/capa.inhibition.rules.test.yml
new file mode 100644
index 00000000..02a12a05
--- /dev/null
+++ b/test/tests/providers/capi/capa-mimir/kaas/phoenix/alerting-rules/capa.inhibition.rules.test.yml
@@ -0,0 +1,38 @@
+---
+rule_files:
+- capa.inhibition.rules.yml
+
+tests:
+  # Tests for `InhibitionClusterWithoutWorkerNodes` inhibition alert
+  - interval: 1m
+    input_series:
+      - series: 'capi_cluster_status_condition{cluster_id="golem", cluster_type="management_cluster", name="golem", pipeline="testing", status="True", type="ControlPlaneReady"}'
+        values: "1+0x300"
+      - series: 'capi_machinepool_spec_replicas{cluster_id="golem", cluster_name="golem", cluster_type="management_cluster", customer="giantswarm", installation="golem", organization="giantswarm", pipeline="testing", provider="capa"}'
+        values: "_x60 0x60 3x60"
+    alert_rule_test:
+      - alertname: InhibitionClusterWithoutWorkerNodes
+        eval_time: 30m
+      - alertname: InhibitionClusterWithoutWorkerNodes
+        eval_time: 90m
+      - alertname: InhibitionClusterWithoutWorkerNodes
+        eval_time: 150m
+      - alertname: PrometheusAgentFailing
+        eval_time: 30m
+        exp_alerts:
+          - exp_labels:
+              area: platform
+              severity: page
+              team: atlas
+              topic: observability
+              inhibit_prometheus_agent_down: "true"
+              instance: prometheus-agent
+              cancel_if_cluster_has_no_workers: "true"
+              cancel_if_cluster_is_not_running_prometheus_agent: "true"
+              cancel_if_cluster_status_creating: "true"
+              cancel_if_cluster_status_deleting: "true"
+            exp_annotations:
+              dashboard: "promRW001/prometheus-remote-write"
+              description: "Prometheus agent remote write is failing."
+              opsrecipe: "prometheus-agent/"
+              summary: "Prometheus agent fails to send samples to remote write endpoint."