WingKai Ho

Use nvidia-smi to ban error gpu make UnexpectedAdmissionError with failed to find gpu id, which is unexpected

5

**What happened**: If you have use nvidia-smi to ban error gpu, before add machine to k8s ,such as ``` nvidia-smi drain -p -m 0 ``` It add this machine to...

fix predicateGPUbyMemory: sort of devIDs reduce fragment gpu

2

Signed-off-by: yougjiahe Traversal key of allocatableGPUs, devIDs are not in sequence. So it will make fragment gpu use.

fix gpu shareing predictor allocate more than one gpu for pod

2

Signed-off-by: yougjiahe Predictor patch all suitable gpu ids which return from predicateGPUbyMemory to pod. But pod only need one of suitable id of gpu. Fix https://github.com/volcano-sh/volcano/issues/2491

Volcano gpu-share plugin patch all suitalbe for pod

2

**What happened**: When update volcano master, check scheduler to gpu-share pod. The volcano device-plugin patch gpu to pod fail. I found annotations `volcano.sh/gpu-index: "5,6"` **What you expected to happen**: gpu-share...