The disaggregated device plugin was initially designed for the Falcon 4005 GPU chassis. To make it available to users without the specific device, we provide this framework, and the Falcon resource is substituted with our simulated resource pool.
make # compile locally
make buildImage # build the docker image
make pushImage # push the image to the kind cluster
make deploy # deploy in K8S
make remove # remove from K8S
written in charts/values.yaml
- api_endpoint: the endpoint to get the resource
- local_ips: internal IP of Kubernetes nodes, which can be figured out by
kubectl get node -o wide
- host_ports: the ports that the Kubernetes nodes connected to
For example, the definition below indicates that the node with IP 172.18.0.5 is connected to host port 1.
local_ips: 172.18.0.5,172.18.0.2,172.18.0.3
host_ports: 1,2,3
If the Disaggregated Device Plugin is successfully deployed, falcon.com/gpu
can be found in nodes' Capacity and Allocatable.