Add healthcheck;Improve console healthchecks;Increase default retries

ansible/roles/firstboot/files/ansible_payload/bootstrap/roles/workloadcluster/tasks/clusterapi.yml

@@ -99,7 +99,7 @@
         --kubeconfig {{ kubeconfig.path }}
     chdir: /opt/metacluster/cluster-api
 
-- name: Ensure CAPI/CAPV controller availability
+- name: Ensure controller availability
   kubernetes.core.k8s_info:
     kind: Deployment
     name: "{{ item.name }}"
@@ -107,6 +107,8 @@
     wait: true
     kubeconfig: "{{ kubeconfig.path }}"
   loop:
+    - name: caip-in-cluster-controller-manager
+      namespace: caip-in-cluster-system
     - name: capi-controller-manager
       namespace: capi-system
     - name: capv-controller-manager

ansible/roles/firstboot/files/ansible_payload/common/roles/tty/tasks/main.yml

@@ -19,6 +19,25 @@
         executable: /opt/firstboot/tty.sh
         workingdir: /tmp/
       metacluster:
+        components:
+          - name: ArgoCD
+            url: https://gitops.${FQDN}
+            healthcheck: https://gitops.${FQDN}
+          - name: Gitea
+            url: https://git.${FQDN}
+            healthcheck: https://git.${FQDN}
+          - name: Harbor
+            url: https://registry.${FQDN}
+            healthcheck: https://registry.${FQDN}
+          - name: Longhorn
+            url: https://storage.${FQDN}
+            healthcheck: https://storage.${FQDN}
+          - name: StepCA
+            url: ''
+            healthcheck: https://ca.${FQDN}/health
+          - name: Traefik
+            url: https://ingress.${FQDN}
+            healthcheck: https://ingress.${FQDN}
         fqdn: "{{ vapp['metacluster.fqdn'] }}"
         vip: "{{ vapp['metacluster.vip'] }}"
   loop:

ansible/roles/firstboot/files/ansible_payload/common/templates/tty.j2

@@ -12,11 +12,14 @@ DFLT='\033[0m'        # Reset colour
 LCLR='\033[K'         # Clear to end of line
 PRST='\033[0;0H'      # Reset cursor position
 
-# COMPONENTS=('ca' 'ingress' 'storage' 'registry' 'git' 'gitops')
-COMPONENTS=('storage' 'registry' 'git' 'gitops')
 FQDN='{{ _template.metacluster.fqdn }}'
 IPADDRESS='{{ _template.metacluster.vip }}'
 
+declare -A COMPONENTS
+{% for component in _template.metacluster.components %}
+COMPONENTS["{{ component.name }}\t({{ component.url }})"]="{{ component.healthcheck }}"
+{% endfor %}
+
 I=0
 
 while /bin/true; do
@@ -30,13 +33,13 @@ while /bin/true; do
   echo -e "${PRST}" > /dev/tty1
   echo -e "\n\n\t${DFLT}To manage this appliance, please connect to one of the following:${LCLR}\n" > /dev/tty1
 
-  for c in "${COMPONENTS[@]}"; do
+  for c in "${!COMPONENTS[@]}"; do
-    STATUS=$(curl -ks "https://${c}.${FQDN}" -o /dev/null -w '%{http_code}')
+    STATUS=$(curl -kLs "${COMPONENTS[${c}]}" -o /dev/null -w '%{http_code}')
 
     if [[ "${STATUS}" -eq "200" ]]; then
-      echo -e "\t [${BGRN}+${DFLT}] ${BBLU}https://${c}.${FQDN}${DFLT}${LCLR}" > /dev/tty1
+      echo -e "\t [${BGRN}+${DFLT}] ${BBLU}${c}${DFLT}${LCLR}" > /dev/tty1
     else
-      echo -e "\t [${BRED}-${DFLT}] ${BBLU}https://${c}.${FQDN}${DFLT}${LCLR}" > /dev/tty1
+      echo -e "\t [${BRED}-${DFLT}] ${BBLU}${c}${DFLT}${LCLR}" > /dev/tty1
     fi
   done
 

ansible/roles/firstboot/files/ansible_payload/common/vars/defaults.yml

@@ -1,5 +1,5 @@
 playbook:
-  retries: 5
+  retries: 10
   delay:
     # These values are multiplied with the value of `storage_benchmark`
     long: 2