From 4cefd7df40c80501b37377092aa2f8be34351832 Mon Sep 17 00:00:00 2001
From: MichaelFisher1997 <contact@michaelfisher.tech>
Date: Sun, 1 Mar 2026 18:42:03 +0000
Subject: [PATCH] fix: surface CSI rollout diagnostics before failing

---
 ansible/roles/csi/tasks/main.yml | 27 ++++++++++++++++++++++-----
 1 file changed, 22 insertions(+), 5 deletions(-)

diff --git a/ansible/roles/csi/tasks/main.yml b/ansible/roles/csi/tasks/main.yml
index 0884901..57c7ab3 100644
--- a/ansible/roles/csi/tasks/main.yml
+++ b/ansible/roles/csi/tasks/main.yml
@@ -19,6 +19,7 @@
   until: csi_controller_rollout.rc == 0
   retries: "{{ csi_rollout_retries | int }}"
   delay: "{{ csi_rollout_delay_seconds | int }}"
+  failed_when: false
   changed_when: false
 
 - name: Show CSI controller status on failure
@@ -26,28 +27,35 @@
   register: csi_controller_deploy_status
   changed_when: false
   failed_when: false
-  when: csi_controller_rollout is failed
+  when: csi_controller_rollout.rc != 0
 
 - name: Show CSI controller pods on failure
   command: kubectl -n kube-system get pods -l app=hcloud-csi-controller -o wide
   register: csi_controller_pods_status
   changed_when: false
   failed_when: false
-  when: csi_controller_rollout is failed
+  when: csi_controller_rollout.rc != 0
+
+- name: Describe CSI controller deployment on failure
+  command: kubectl -n kube-system describe deployment hcloud-csi-controller
+  register: csi_controller_deploy_describe
+  changed_when: false
+  failed_when: false
+  when: csi_controller_rollout.rc != 0
 
 - name: Show CSI driver logs on failure
   command: kubectl -n kube-system logs deployment/hcloud-csi-controller -c hcloud-csi-driver --tail={{ csi_failure_log_tail_lines }}
   register: csi_driver_logs
   changed_when: false
   failed_when: false
-  when: csi_controller_rollout is failed
+  when: csi_controller_rollout.rc != 0
 
 - name: Show recent kube-system events on failure
   command: kubectl -n kube-system get events --sort-by=.lastTimestamp
   register: csi_recent_events
   changed_when: false
   failed_when: false
-  when: csi_controller_rollout is failed
+  when: csi_controller_rollout.rc != 0
 
 - name: Fail with CSI controller diagnostics
   fail:
@@ -59,12 +67,15 @@
       Pods status:
       {{ csi_controller_pods_status.stdout | default('n/a') }}
 
+      Deployment describe:
+      {{ csi_controller_deploy_describe.stdout | default('n/a') }}
+
       hcloud-csi-driver logs:
       {{ csi_driver_logs.stdout | default('n/a') }}
 
       Recent kube-system events:
       {{ csi_recent_events.stdout | default('n/a') }}
-  when: csi_controller_rollout is failed
+  when: csi_controller_rollout.rc != 0
 
 - name: Wait for CSI node daemonset rollout
   command: kubectl rollout status daemonset/hcloud-csi-node -n kube-system --timeout={{ csi_rollout_timeout_seconds }}s
@@ -72,4 +83,10 @@
   until: csi_node_rollout.rc == 0
   retries: "{{ csi_rollout_retries | int }}"
   delay: "{{ csi_rollout_delay_seconds | int }}"
+  failed_when: false
   changed_when: false
+
+- name: Fail when CSI node daemonset rollout does not complete
+  fail:
+    msg: "CSI node daemonset rollout failed: {{ csi_node_rollout.stdout | default('') }} {{ csi_node_rollout.stderr | default('') }}"
+  when: csi_node_rollout.rc != 0