fix: reduce rerun bootstrap pre-pull delays

ansible/roles/kube-vip-deploy/tasks/main.yml

@@ -1,12 +1,37 @@
 ---
 - name: Pre-pull kube-vip bootstrap images into containerd
-  command: /usr/local/bin/ctr -n k8s.io images pull {{ item }}
+  shell: |
+    if /usr/local/bin/ctr -n k8s.io images ls -q | grep -Fx -- "{{ item }}" >/dev/null; then
+      echo "already present"
+      exit 0
+    fi
+
+    for attempt in 1 2 3; do
+      if timeout 120s /usr/local/bin/ctr -n k8s.io images pull "{{ item }}"; then
+        echo "pulled image"
+        exit 0
+      fi
+
+      sleep 10
+    done
+
+    exit 1
+  args:
+    executable: /bin/bash
   register: kube_vip_image_pull
-  retries: 12
-  delay: 15
-  until: kube_vip_image_pull.rc == 0
   loop: "{{ kube_vip_prepull_images }}"
-  changed_when: true
+  changed_when: "'pulled image' in kube_vip_image_pull.stdout"
+  failed_when: false
+
+- name: Report kube-vip images that did not pre-pull after retries
+  debug:
+    msg: >-
+      Best-effort kube-vip image pre-pull did not complete for {{ item.item }} after
+      3 attempt(s): {{ item.stderr | default('no stderr') }}
+  loop: "{{ kube_vip_image_pull.results | default([]) }}"
+  loop_control:
+    label: "{{ item.item }}"
+  when: item.rc is defined and item.rc != 0
 
 - name: Render kube-vip control plane manifest
   template:

ansible/site.yml

@@ -110,11 +110,12 @@
     - bootstrap-image-prepull
 
 - name: Pre-pull Rancher bootstrap images
-  hosts: cluster
+  hosts: workers
   become: true
 
   roles:
-    - rancher-image-prepull
+    - role: rancher-image-prepull
+      when: rancher_image_prepull_enabled | default(false) | bool
 
 - name: Deploy observability stack
   hosts: control_plane[0]