Fix label syntax;Add retries;Sort list;Housekeeping

ansible/roles/firstboot/files/ansible_payload/bootstrap/roles/metacluster/tasks/init.yml

@@ -63,14 +63,14 @@
       jq -r '.Tags[0]'
     chdir: /opt/metacluster/container-images
   register: registry_artifacts
-  loop: "{{ query('ansible.builtin.fileglob', '/opt/metacluster/container-images/*.tar') }}"
+  loop: "{{ query('ansible.builtin.fileglob', '/opt/metacluster/container-images/*.tar') | sort }}"
   loop_control:
     label: "{{ item | basename }}"
 
 - name: Get source registries of all artifacts
   ansible.builtin.set_fact:
     source_registries: "{{ (source_registries | default([]) + [(item | split('/'))[0]]) | unique | sort }}"
-  loop: "{{ registry_artifacts | json_query('results[*].stdout') | select }}"
+  loop: "{{ registry_artifacts | json_query('results[*].stdout') | select | sort }}"
 
 - name: Configure K3s node for private registry
   ansible.builtin.template:

ansible/roles/firstboot/files/ansible_payload/bootstrap/roles/metacluster/tasks/k3s.yml

@@ -71,5 +71,13 @@
     definition:
       metadata:
         labels:
-          vm_id: "{{ moref_id }}"
+          ova.airgappedk8s/moref_id: "{{ moref_id }}"
     kubeconfig: "{{ kubeconfig.path }}"
+  register: patch_result
+  until:
+    - patch_result is not failed
+  retries: "{{ playbook.retries }}"
+  delay: "{{ playbook.delay.short }}"
+
+- debug:
+    msg: "{{ patch_result }}"

ansible/roles/firstboot/files/ansible_payload/upgrade/roles/decommission/tasks/k3s.yml

@@ -1,12 +1,12 @@
 - name: Cordon node
   kubernetes.core.k8s_drain:
-    name: "{{ decom_node }}"
+    name: "{{ decom_nodename }}"
     state: cordon
     kubeconfig: "{{ kubeconfig.path }}"
 
 - name: Drain node
   kubernetes.core.k8s_drain:
-    name: "{{ decom_node }}"
+    name: "{{ decom_nodename }}"
     state: drain
     delete_options:
       ignore_daemonsets: true
@@ -17,7 +17,7 @@
 
 - name: Delete node
   kubernetes.core.k8s:
-    name: "{{ decom_node }}"
+    name: "{{ decom_nodename }}"
     kind: node
     state: absent
     wait: true

ansible/roles/firstboot/files/ansible_payload/upgrade/roles/decommission/tasks/main.yml

@@ -1,15 +1,15 @@
-- name: Lookup node name and moref id for decommissioning
+- name: Lookup node name and moref-id for decommissioning
   ansible.builtin.set_fact:
-    decom_node: >-
+    decom_nodename: >-
       {{
         lookup('kubernetes.core.k8s', kind='Node', kubeconfig=(kubeconfig.path)) |
           json_query('[? metadata.name != `' ~ ansible_facts.nodename ~ '`].metadata.name') |
           first
       }}
-    decom_vmid: >-
+    decom_morefid: >-
       {{
         lookup('kubernetes.core.k8s', kind='Node', kubeconfig=(kubeconfig.path)) |
-          json_query('[? metadata.name != `' ~ ansible_facts.nodename ~ '`].metadata.labels.vm_id') |
+          json_query('[? metadata.name != `' ~ ansible_facts.nodename ~ '`].metadata.labels."ova.airgappedk8s/moref_id"') |
           first
       }}
 

ansible/roles/firstboot/files/ansible_payload/upgrade/roles/decommission/tasks/storage.yml

@@ -1,6 +1,6 @@
 - name: Disable disk scheduling and evict replicas
   kubernetes.core.k8s:
-    name: "{{ decom_node }}"
+    name: "{{ decom_nodename }}"
     namespace: longhorn-system
     kind: nodes.longhorn.io
     state: patched

ansible/roles/firstboot/files/ansible_payload/upgrade/roles/decommission/tasks/virtualmachine.yml

@@ -2,7 +2,7 @@
 
     - name: Lookup VM name
       community.vmware.vmware_guest_info:
-        moid: "{{ decom_vmid }}"
+        moid: "{{ decom_morefid }}"
       register: virtualmachine_details
 
     - name: Power off VM

ansible/roles/firstboot/files/ansible_payload/upgrade/roles/metacluster/tasks/k3s.yml

@@ -71,5 +71,10 @@
     definition:
       metadata:
         labels:
-          vm_id: "{{ moref_id }}"
+          ova.airgappedk8s/moref_id: "{{ moref_id }}"
     kubeconfig: "{{ kubeconfig.path }}"
+  register: patch_result
+  until:
+    - patch_result is not failed
+  retries: "{{ playbook.retries }}"
+  delay: "{{ playbook.delay.short }}"