From 6593adbea6889aff45e6efb74420f15b8818f2ad Mon Sep 17 00:00:00 2001
From: MichaelFisher1997 <contact@michaelfisher.tech>
Date: Sun, 1 Mar 2026 18:28:17 +0000
Subject: [PATCH] fix: make CSI rollout checks configurable and faster

---
 ansible/roles/csi/defaults/main.yml |  4 ++++
 ansible/roles/csi/tasks/main.yml    | 24 +++++++++++++++++-------
 2 files changed, 21 insertions(+), 7 deletions(-)

diff --git a/ansible/roles/csi/defaults/main.yml b/ansible/roles/csi/defaults/main.yml
index bae3050..77fa931 100644
--- a/ansible/roles/csi/defaults/main.yml
+++ b/ansible/roles/csi/defaults/main.yml
@@ -2,3 +2,7 @@
 hcloud_token: ""
 cluster_name: "k8s-cluster"
 csi_manifest_url: "https://raw.githubusercontent.com/hetznercloud/csi-driver/v2.12.0/deploy/kubernetes/hcloud-csi.yml"
+csi_rollout_timeout_seconds: 30
+csi_rollout_retries: 8
+csi_rollout_delay_seconds: 5
+csi_failure_log_tail_lines: 120
diff --git a/ansible/roles/csi/tasks/main.yml b/ansible/roles/csi/tasks/main.yml
index 430c3e8..0884901 100644
--- a/ansible/roles/csi/tasks/main.yml
+++ b/ansible/roles/csi/tasks/main.yml
@@ -14,11 +14,11 @@
   changed_when: true
 
 - name: Wait for CSI controller rollout
-  command: kubectl rollout status deployment/hcloud-csi-controller -n kube-system --timeout=30s
+  command: kubectl rollout status deployment/hcloud-csi-controller -n kube-system --timeout={{ csi_rollout_timeout_seconds }}s
   register: csi_controller_rollout
   until: csi_controller_rollout.rc == 0
-  retries: 30
-  delay: 10
+  retries: "{{ csi_rollout_retries | int }}"
+  delay: "{{ csi_rollout_delay_seconds | int }}"
   changed_when: false
 
 - name: Show CSI controller status on failure
@@ -36,12 +36,19 @@
   when: csi_controller_rollout is failed
 
 - name: Show CSI driver logs on failure
-  command: kubectl -n kube-system logs deployment/hcloud-csi-controller -c hcloud-csi-driver --tail=120
+  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
 
+- 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
+
 - name: Fail with CSI controller diagnostics
   fail:
     msg: |
@@ -54,12 +61,15 @@
 
       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
 
 - name: Wait for CSI node daemonset rollout
-  command: kubectl rollout status daemonset/hcloud-csi-node -n kube-system --timeout=30s
+  command: kubectl rollout status daemonset/hcloud-csi-node -n kube-system --timeout={{ csi_rollout_timeout_seconds }}s
   register: csi_node_rollout
   until: csi_node_rollout.rc == 0
-  retries: 30
-  delay: 10
+  retries: "{{ csi_rollout_retries | int }}"
+  delay: "{{ csi_rollout_delay_seconds | int }}"
   changed_when: false