Bump terminationGracePeriodSeconds to 120 for vGPU Manager daemonset

This ensures the vGPU Manager has enough time to finish its cleanup logic
when the pod is terminated. This fixes an issue observed on a HGX node
(with 8 GPUs) where the vGPU Manager was killed before it finished disabling
VFs on all 8 GPUs.

Signed-off-by: Christopher Desiniotis <[email protected]>

Author: cdesiniotis

assets/state-vgpu-manager/0500_daemonset.yaml

@@ -18,6 +18,7 @@ spec:
       labels:
         app: nvidia-vgpu-manager-daemonset
     spec:
+      terminationGracePeriodSeconds: 120
       nodeSelector:
         nvidia.com/gpu.deploy.vgpu-manager: "true"
       tolerations:

internal/state/testdata/golden/driver-vgpu-host-manager-openshift.yaml

@@ -330,6 +330,7 @@ spec:
         nvidia.com/gpu.deploy.vgpu-manager: "true"
       priorityClassName: system-node-critical
       serviceAccountName: nvidia-vgpu-manager-openshift
+      terminationGracePeriodSeconds: 120
       tolerations:
       - effect: NoSchedule
         key: nvidia.com/gpu

internal/state/testdata/golden/driver-vgpu-host-manager.yaml

@@ -238,6 +238,7 @@ spec:
         nvidia.com/gpu.deploy.vgpu-manager: "true"
       priorityClassName: system-node-critical
       serviceAccountName: nvidia-vgpu-manager-ubuntu22.04
+      terminationGracePeriodSeconds: 120
       tolerations:
       - effect: NoSchedule
         key: nvidia.com/gpu

manifests/state-driver/0500_daemonset.yaml

@@ -59,6 +59,9 @@ spec:
         {{- .Driver.Spec.Labels | yaml | nindent 8 }}
         {{- end }}
     spec:
+      {{- if eq .Driver.Spec.DriverType "vgpu-host-manager" }}
+      terminationGracePeriodSeconds: 120
+      {{- end }}
       nodeSelector:
         {{- if eq .Driver.Spec.DriverType "vgpu-host-manager" }}
         nvidia.com/gpu.deploy.vgpu-manager: "true"