feat(proxmox): add k3s agents and refine VM provisioning

Signed-off-by: Tuan-Dat Tran <tuan-dat.tran@tudattr.dev>

roles/proxmox/tasks/55_create_vm.yml

@@ -12,7 +12,7 @@
     cores: "{{ vm.cores }}"
     memory: "{{ vm.memory }}"
     net: "{{ vm.net }}"
-    hostpci: "{{ vm.hostpci | default({})}}"
+    hostpci: "{{ vm.hostpci | default({}) }}"
     scsihw: "virtio-scsi-pci"
     ostype: "l26"
     tags: "{{ proxmox_tags }}"
@@ -20,8 +20,7 @@
     boot: "order=scsi0"
     cpu: "x86-64-v2-AES"
     ciuser: "{{ vm.ciuser }}"
-    # cipassword: "{{ vm_secrets[proxmox_secrets_prefix + '_' + vm.name.replace('-', '_')] }}"
-    cipassword: "flyff369"
+    cipassword: "{{ vm_secrets[proxmox_secrets_prefix + '_' + vm.name.replace('-', '_')] }}"
     ipconfig:
       ipconfig0: "ip=dhcp"
     sshkeys: "{{ vm.sshkeys }}"

roles/proxmox/tasks/56_provision_new_vm.yml

@@ -33,29 +33,12 @@
     vmid: "{{ vm.vmid }}"
     state: started
 
-- name: Stop VM
-  community.general.proxmox_kvm:
-    api_user: "{{ proxmox_api_user }}@pam"
-    api_token_id: "{{ proxmox_api_token_id }}"
-    api_token_secret: "{{ proxmox_api_token_secret }}"
-    api_host: "192.168.20.12"
-    node: "{{ vm.node }}"
-    vmid: "{{ vm.vmid }}"
-    state: stopped
-    force: true
+- name: Retry stopping VM
+  ansible.builtin.include_tasks: ./57_stop_and_verify_vm.yml
 
-- name: Wait until VM is fully stopped
-  community.general.proxmox_vm_info:
-    api_user: "{{ proxmox_api_user }}@pam"
-    api_token_id: "{{ proxmox_api_token_id }}"
-    api_token_secret: "{{ proxmox_api_token_secret }}"
-    api_host: "192.168.20.12"
-    node: "{{ vm.node }}"
-    vmid: "{{ vm.vmid }}"
-  register: vm_status_check
-  until: vm_status_check.proxmox_vms[0].status == "stopped"
-  retries: 24
-  delay: 5
+- name: Pause for 5 seconds for api
+  ansible.builtin.pause:
+    seconds: 5
 
 - name: Start VM
   community.general.proxmox_kvm:

roles/proxmox/tasks/57_stop_and_verify_vm.yml

@@ -0,0 +1,39 @@
+- name: "Wait until success"
+  block:
+    - name: Set the retry count
+      set_fact:
+        retry_count: "{{ 0 if retry_count is undefined else retry_count | int + 1 }}"
+
+    - name: Stop VM
+      community.general.proxmox_kvm:
+        api_user: "{{ proxmox_api_user }}@pam"
+        api_token_id: "{{ proxmox_api_token_id }}"
+        api_token_secret: "{{ proxmox_api_token_secret }}"
+        api_host: "192.168.20.12"
+        node: "{{ vm.node }}"
+        vmid: "{{ vm.vmid }}"
+        state: stopped
+        force: true
+
+    - name: Wait until VM is fully stopped
+      community.general.proxmox_vm_info:
+        api_user: "{{ proxmox_api_user }}@pam"
+        api_token_id: "{{ proxmox_api_token_id }}"
+        api_token_secret: "{{ proxmox_api_token_secret }}"
+        api_host: "192.168.20.12"
+        node: "{{ vm.node }}"
+        vmid: "{{ vm.vmid }}"
+      register: vm_status_check
+      failed_when: vm_status_check.proxmox_vms[0].status != "stopped"
+  rescue:
+    - name: Check for retry count
+      fail:
+        msg: Ended after 24 retries
+      when: retry_count|int == 24
+
+    - name: Wait 5s
+      ansible.builtin.pause:
+        seconds: 5
+
+    - name: "Failed to stop VM - Retrying..."
+      include_tasks: ./57_stop_and_verify_vm.yml