device plugin allocate judge error

[angao]

github.com/AliyunContainerService/gpushare-device-plugin/pkg/gpu/nvidia/allocate.go:79

// podReqGPU = uint(0)
for _, req := range reqs.ContainerRequests {
	podReqGPU += uint(len(req.DevicesIDs))
}
...
if getGPUMemoryFromPodResource(pod) == podReqGPU {
...
}

getGPUMemoryFromPodResource() returns pod memory usage, but podReqGPU is pod request gpu count

[angao]

@cheyang

[cheyang]

podReqGPU is the pod's GPU memory, is aliyun.com/gpu-mem. It's because the DevicesIDs here is not the real GPU device ID, it's the fake DeviceIDs and the total number of the DeviceID equals GPU memory.

Maybe reading the design doc is helpful to you. https://github.com/AliyunContainerService/gpushare-scheduler-extender/blob/master/docs/designs/designs.md#design

[angao]

thx