ci instability: DynMIG: inspect device attributes failed with `Cannot iterate over null`

[jgehrcke]

not ok 17 DynMIG: inspect device attributes in resource slice (gpu) in 1217ms
# tags: fastfeedback
# (from function `assert_attrs_equal' in file tests/bats/helpers.sh, line 165,
#  in test file tests/bats/test_gpu_dynmig.bats, line 59)
#   `assert_attrs_equal "$attrs" "${reference[@]}"' failed
# No resources found in default namespace.
# No resources found in default namespace.
# No resources found in default namespace.
# NAME                                                READY   STATUS    RESTARTS   AGE   IP               NODE                   NOMINATED NODE   READINESS GATES
# nvidia-dra-driver-gpu-controller-7f6c956fb9-hlc28   1/1     Running   0          4s    192.168.33.166   gb-nvl-156-compute11   <none>           <none>
# nvidia-dra-driver-gpu-kubelet-plugin-6p8nh          2/2     Running   0          4s    192.168.33.41    gb-nvl-156-compute12   <none>           <none>
# nvidia-dra-driver-gpu-kubelet-plugin-s244j          2/2     Running   0          4s    192.168.33.153   gb-nvl-156-compute11   <none>           <none>
# resource slice name: gb-nvl-156-compute11-gpu.nvidia.com-g8pzk
# jq: error (at <stdin>:1): Cannot iterate over null (null)
# No resources found in default namespace.

https://github.com/NVIDIA/k8s-dra-driver-gpu/actions/runs/22589654877/job/65443933706#step:3:241

[jgehrcke]

Another occurrence:

ok 16 GPUs: shared RC across 20 pods, 1 repetitions in 8648ms
not ok 17 DynMIG: inspect device attributes in resource slice (gpu) in 1207ms
# tags: fastfeedback
# (from function `assert_attrs_equal' in file tests/bats/helpers.sh, line 165,
#  in test file tests/bats/test_gpu_dynmig.bats, line 59)
#   `assert_attrs_equal "$attrs" "${reference[@]}"' failed
# No resources found in default namespace.
# No resources found in default namespace.
# No resources found in default namespace.
# NAME                                                READY   STATUS    RESTARTS   AGE   IP               NODE                   NOMINATED NODE   READINESS GATES
# nvidia-dra-driver-gpu-controller-6b8f4bd85c-lp8zs   1/1     Running   0          5s    192.168.33.167   gb-nvl-156-compute11   <none>           <none>
# nvidia-dra-driver-gpu-kubelet-plugin-rzqf9          2/2     Running   0          4s    192.168.33.45    gb-nvl-156-compute12   <none>           <none>
# nvidia-dra-driver-gpu-kubelet-plugin-x7wtj          2/2     Running   0          4s    192.168.33.168   gb-nvl-156-compute11   <none>           <none>
# resource slice name: gb-nvl-156-compute11-gpu.nvidia.com-bp94d
# jq: error (at <stdin>:1): Cannot iterate over null (null)

https://github.com/NVIDIA/k8s-dra-driver-gpu/actions/runs/22657758122/job/65671070951#step:3:228

[jgehrcke]

another one: https://github.com/NVIDIA/k8s-dra-driver-gpu/actions/runs/22673423406/job/65723367615#step:3:241

[jgehrcke]

Improvement landed via #906. Closing this for now, let's reopen when this specific symptom pops up again.

[jgehrcke]

Another instance of this happened over night:
https://github.com/NVIDIA/k8s-dra-driver-gpu/actions/runs/22705290510/job/65830932803#step:3:268

Yesterday's debuggability improvements were sadly not all-too-fruitful. Based on logs and tmpdir contents I could not yet fully understand what went wrong. Interesting. Re-opening.