Monitoring support for K8s build cluster

kubetest2-tf/data/k8s-ansible/install-k8s-monitoring.yml

@@ -0,0 +1,27 @@
+---
+- name: Deploy kube-prometheus-stack to Kubernetes
+  hosts: masters
+  gather_facts: no
+  become: yes
+
+  pre_tasks:
+    - name: Ensure kubernetes.core collection is installed
+      command: ansible-galaxy collection install kubernetes.core
+      changed_when: false
+      delegate_to: "{{ groups['masters'][0] }}"
+      become: no
+
+    - name: Download Helm install script
+      ansible.builtin.get_url:
+        url: https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3
+        dest: /tmp/get_helm.sh
+        mode: '0700'
+      delegate_to: "{{ groups['masters'][0] }}"
+
+    - name: Run Helm install script
+      command: /tmp/get_helm.sh
+      delegate_to: "{{ groups['masters'][0] }}"
+
+  roles:
+    - role: install-monitoring
+      name: Deploy kube-prometheus-stack

kubetest2-tf/data/k8s-ansible/roles/install-monitoring/defaults/main.yml

@@ -0,0 +1,9 @@
+#Default vars for install-monitoring role
+---
+helm_release_name: k8s-monitoring
+helm_chart_name: prometheus-community/kube-prometheus-stack
+helm_chart_version: "78.4.0"
+helm_namespace: monitoring
+helm_repo_name: prometheus-community
+helm_repo_url: https://prometheus-community.github.io/helm-charts
+helm_custom_values: "power-custom-values.yaml"

kubetest2-tf/data/k8s-ansible/roles/install-monitoring/files/power-custom-values.yaml

@@ -0,0 +1,61 @@
+grafana:
+  image: 
+    registry: quay.io
+    repository: powercloud/grafana
+    tag: v12.1.0
+    pullPolicy: IfNotPresent
+
+  service:
+    type: NodePort
+    portName: http-web
+    ipFamilies: []
+    ipFamilyPolicy: ""
+    port: 3000
+    targetPort: 3000
+
+  adminUser: admin
+  adminPassword: prom-operator
+
+  nodeSelector: 
+    kubernetes.io/role: infra
+
+prometheusOperator:
+  admissionWebhooks:
+    deployment:
+      nodeSelector: 
+        kubernetes.io/role: infra
+
+    patch:
+      enabled: true
+      image:
+        registry: quay.io
+        repository: powercloud/kube-webhook-certgen
+        tag: v1.6.3
+
+  nodeSelector: 
+    kubernetes.io/role: infra
+
+prometheus:
+  prometheusSpec:
+    retention: 14d
+    nodeSelector: 
+      kubernetes.io/role: infra
+
+crds:
+  upgradeJob:
+    nodeSelector: 
+      kubernetes.io/role: infra
+
+alertmanager:
+  alertmanagerSpec:
+    nodeSelector: 
+      kubernetes.io/role: infra
+
+kube-state-metrics:
+  nodeSelector: 
+    kubernetes.io/role: infra
+
+thanosRuler:
+  thanosRulerSpec:
+    nodeSelector: 
+      kubernetes.io/role: infra

kubetest2-tf/data/k8s-ansible/roles/install-monitoring/tasks/main.yml

@@ -0,0 +1,35 @@
+---
+- name: Ensure namespace exists
+  command: kubectl create namespace {{ helm_namespace }}
+  ignore_errors: yes # Ignore error if namespace already exists
+
+- name: Ensure Helm repository is added
+  kubernetes.core.helm_repository:
+    name: "{{ helm_repo_name }}"
+    repo_url: "{{ helm_repo_url }}"
+    state: present
+  delegate_to: "{{ groups['masters'][0] }}"
+
+- name: Copy IBM Power specific helm chart values files
+  copy:
+    src: "{{ helm_custom_values }}"
+    dest: /tmp/
+
+- name: Install or upgrade kube-prometheus-stack chart
+  kubernetes.core.helm:
+    update_repo_cache: yes
+    name: "{{ helm_release_name }}"
+    chart_ref: "{{ helm_chart_name }}"
+    namespace: "{{ helm_namespace }}"
+    chart_version: "{{ helm_chart_version }}"
+    values_files: 
+      - "/tmp/power-custom-values.yaml"
+    timeout: 10m
+    set_values:
+      - value: grafana.adminUser={{ grafana_admin_user }}
+        value_type: string
+      - value: grafana.adminPassword={{ grafana_admin_password }}
+        value_type: string
+    state: present
+  delegate_to: "{{ groups['masters'][0] }}"
+