From f2d19ebe0a6fac570bea724a21b52f1c7d94f2b5 Mon Sep 17 00:00:00 2001
From: nate stephany <nate@redhat.com>
Date: Tue, 10 Mar 2026 09:20:18 +0100
Subject: [PATCH] ocp4_workload_nvidia_gpu_operator: Wait for ClusterPolicy CRD
 before applying

The install_operator role considers installation complete when the CSV
reaches Succeeded state. However, the NVIDIA GPU Operator registers its
ClusterPolicy CRD after the CSV succeeds, causing a race condition where
the ClusterPolicy task fails to resolve nvidia.com/v1.ClusterPolicy.

Add an explicit wait for clusterpolicies.nvidia.com CRD availability
(30 retries x 10s = 5 min) before applying the ClusterPolicy resource.
---
 .../tasks/workload.yml                                 | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/roles/ocp4_workload_nvidia_gpu_operator/tasks/workload.yml b/roles/ocp4_workload_nvidia_gpu_operator/tasks/workload.yml
index 1e55735..e23b3a6 100644
--- a/roles/ocp4_workload_nvidia_gpu_operator/tasks/workload.yml
+++ b/roles/ocp4_workload_nvidia_gpu_operator/tasks/workload.yml
@@ -32,6 +32,16 @@
     install_operator_catalogsource_image: "{{ ocp4_workload_nvidia_gpu_operator_catalog_snapshot_image }}"
     install_operator_catalogsource_image_tag: "{{ ocp4_workload_nvidia_gpu_operator_catalog_snapshot_image_tag }}"
 
+- name: Wait for NVIDIA GPU Operator ClusterPolicy CRD to be available
+  kubernetes.core.k8s_info:
+    api_version: apiextensions.k8s.io/v1
+    kind: CustomResourceDefinition
+    name: clusterpolicies.nvidia.com
+  register: r_clusterpolicy_crd
+  retries: 30
+  delay: 10
+  until: r_clusterpolicy_crd.resources | length > 0
+
 - name: Setup NVIDIA GPU Cluster Policy
   kubernetes.core.k8s:
     state: present