Skip to content

Commit 48007bd

Browse files
mattia-eleuterimattia-eleuteri
committed
test(kubernetes): pin the remaining host-driver and CDI settings
Review feedback on the addon default: the suite checked DEVICE_LIST_STRATEGY and NVIDIA_CTK_PATH but not NVIDIA_DRIVER_ROOT, DEVICE_ID_STRATEGY or CDI_ANNOTATION_PREFIX, so a regression in any of those three could break CDI device resolution while the assertions still passed. CDI device names key on the UUID strategy, the plugin resolves the driver from NVIDIA_DRIVER_ROOT, and the annotation prefix has to be the one containerd reads. Pin all of them, plus the validator's own NVIDIA_DRIVER_ROOT alongside DRIVER_ROOT. Assisted-By: Claude <[email protected]> Signed-off-by: Mattia Eleuteri <[email protected]>
1 parent 10137b8 commit 48007bd

1 file changed

Lines changed: 27 additions & 0 deletions

File tree

‎packages/apps/kubernetes/tests/gpu_operator_hami_test.yaml‎

Lines changed: 27 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -64,6 +64,33 @@ tests:
6464
content:
6565
name: NVIDIA_CTK_PATH
6666
value: /usr/local/bin/nvidia-ctk
67+
# The remaining three are just as load-bearing as the two above: CDI device
68+
# names key on the UUID strategy, the plugin resolves the driver from
69+
# NVIDIA_DRIVER_ROOT, and the annotation prefix has to be the one containerd
70+
# reads. A regression in any of them breaks device resolution while the two
71+
# assertions above still pass.
72+
- contains:
73+
path: spec.values.gpu-operator.devicePlugin.env
74+
content:
75+
name: DEVICE_ID_STRATEGY
76+
value: uuid
77+
- contains:
78+
path: spec.values.gpu-operator.devicePlugin.env
79+
content:
80+
name: NVIDIA_DRIVER_ROOT
81+
value: "/"
82+
- contains:
83+
path: spec.values.gpu-operator.devicePlugin.env
84+
content:
85+
name: CDI_ANNOTATION_PREFIX
86+
value: cdi.k8s.io/
87+
# Same for the validator: NVIDIA_DRIVER_ROOT alongside DRIVER_ROOT is what
88+
# the driver-validation init container reads.
89+
- contains:
90+
path: spec.values.gpu-operator.validator.driver.env
91+
content:
92+
name: NVIDIA_DRIVER_ROOT
93+
value: "/"
6794

6895
# The previous default fed NVreg_NvLinkDisable=1 to the operator's driver
6996
# container through the nvidia-kernel-module-params ConfigMap. With the driver

0 commit comments

Comments
 (0)