tests: pull NFD image from the ghcr.io mirror in CI

The bundled node-feature-discovery subchart pulls its image from
registry.k8s.io/nfd/node-feature-discovery:<appVersion>, which has been a
recurring source of CI image-pull failures.

Rather than hardcoding an override in the chart's values.yaml, override the
NFD image repository in the CI helm values (gha-run-k8s-common.sh) to point at
the kata-containers ghcr.io mirror. Only the repository is overridden; the tag
is left unset so the subchart keeps using its appVersion, keeping the mirror
tag in sync automatically.

Authored-by: Cursor <cursor@cursor.com>
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
This commit is contained in:
Fabiano Fidêncio
2026-06-11 20:45:55 +02:00
parent 443a999323
commit 50ee3789c4

View File

@@ -710,6 +710,14 @@ function helm_helper() {
# Enable node-feature-discovery deployment
yq -i ".node-feature-discovery.enabled = true" "${values_yaml}"
# Pull the NFD image from the kata-containers ghcr.io mirror rather than the
# upstream registry.k8s.io/nfd/node-feature-discovery, which has been a
# recurring source of CI image-pull failures. We override only the registry
# here (in CI) and leave the chart default untouched. The tag is left unset
# on purpose so the NFD subchart keeps using its appVersion, which must be
# present in the mirror.
yq -i ".node-feature-discovery.image.repository = \"${NFD_IMAGE_REPOSITORY:-ghcr.io/kata-containers/node-feature-discovery}\"" "${values_yaml}"
# Do not enable on nvidia-gpu-* tests, as it'll be deployed by the GPU operator
if [[ "${KATA_HYPERVISOR}" == *"nvidia-gpu"* ]]; then
yq -i ".node-feature-discovery.enabled = false" "${values_yaml}"