fix: add kube-prometheus-stack as gpu-operator dependency

GPU operator with dcgmExporter.serviceMonitor.enabled=true tries to
create a ServiceMonitor CR during reconciliation. If kube-prometheus-stack
(which provides the ServiceMonitor CRD) is not yet installed, the
operator reports ClusterPolicy as notReady with "couldn't find
ServiceMonitor CRD".

Add kube-prometheus-stack as a dependency of gpu-operator in base.yaml
to ensure the monitoring stack and its CRDs are installed first.

Fixes: #165

Signed-off-by: Yuan Chen <yuanchen97@gmail.com>

Author: yuanchen8911

examples/recipes/eks-gb200-ubuntu-training-with-validation.yaml

@@ -152,6 +152,7 @@ componentRefs:
 
 deploymentOrder:
   - cert-manager
+  - kube-prometheus-stack
   - gpu-operator
   - nvidia-dra-driver-gpu
   - nvsentinel

examples/recipes/eks-training.yaml

@@ -63,6 +63,7 @@ componentRefs:
     valuesFile: components/skyhook-operator/values.yaml
 deploymentOrder:
   - cert-manager
+  - kube-prometheus-stack
   - gpu-operator
   - nvsentinel
   - skyhook-operator

recipes/overlays/base.yaml

@@ -50,6 +50,7 @@ spec:
           namespace: gpu-operator
       dependencyRefs:
         - cert-manager
+        - kube-prometheus-stack
 
     - name: nvsentinel
       type: Helm

tests/chainsaw/ai-conformance/offline/assert-recipe.yaml

@@ -57,11 +57,11 @@ deploymentOrder:
   - aws-efa
   - cert-manager
   - dynamo-crds
-  - gpu-operator
-  - kai-scheduler
   - kgateway-crds
   - kgateway
   - kube-prometheus-stack
+  - gpu-operator
+  - kai-scheduler
   - dynamo-platform
   - k8s-ephemeral-storage-metrics
   - nvidia-dra-driver-gpu

tests/chainsaw/cli/cuj1-training/assert-recipe.yaml

@@ -52,9 +52,9 @@ deploymentOrder:
   - aws-ebs-csi-driver
   - aws-efa
   - cert-manager
+  - kube-prometheus-stack
   - gpu-operator
   - kai-scheduler
-  - kube-prometheus-stack
   - k8s-ephemeral-storage-metrics
   - kubeflow-trainer
   - nvidia-dra-driver-gpu