get operator to run reliably on a plain kubernetes system (without nfd running) (!3) · Merge requests · nvidia / kubernetes / gpu-operator

Rajat Chopra requested to merge rajatchopra/gpu-operator:fix_crashes into master Jul 30, 2019

GPU operator runs alright (somewhat) without crashing now on plain kubernetes. Please see the (temporary?) logic of excluding openshift specific files in the operator reconciliation loop. You have to run the pod with ENV var: KUBE_UNSUPPORTED_FLAVOR=openshift and rename all openshift specific manifest files to have 'openshift' in its name, so that we can identify. The operator can alright on both openshift and kubernetes then.

Next up: run it on a cluster with GPUs. Fingers crossed.

@renaudwastaken @Ethyling PTAL. Thanks.

get operator to run reliably on a plain kubernetes system (without nfd running)

Merge request reports