Kubernetes scheduler plugin for multi-container nvidia gpu share.
⚠This project is for code example purpose, if you want to use it in production, I am happy to provide support, feel free to contact me.
Related project: WLBF/flex-gpu-device-plugin
This scheduler plugin enable bin-pack scheduling decisions based on nvidia.flex.com/gpu
and nvidia.flex.com/memory
resource. Allow pods exclusive or share usage of nvidia gpus. Scheduled pod will be marked with nividia.flex.com/index
annotation to denote its gpu index in running node.
Scheduler with plugin can be installed by helm chart. For development use values.dev.yaml
instead of values.pord.yaml
.
helm install flex-gpu-scheduler -f ./manifests/flexgpu/values.prod.yaml ./manifests/flexgpu