By default gpu-operator deploys toolkit with CRIO hooks dir on host as "/etc/containers/oci/hooks.d". Sometimes when host is rebooted, toolkit is not able to cleanup the nvidia runtime hook configuration and driver pod fails to run with it. Changing it to be under /run fixes this issue. CRIO by default is configured to look under this path:

cat /etc/crio/crio.conf.d/00-default

hooks_dir = [ "/etc/containers/oci/hooks.d", "/run/containers/oci/hooks.d", ]

Signed-off-by: Shiva Krishna, Merla smerla@nvidia.com

Update hooks paths for nvidia-runtime to under /run/containers/oci/hooks.d

cat /etc/crio/crio.conf.d/00-default

Merge request reports