How to use p2p over PCIe or NVLink between different containers on the same node?

Accelerated Computing NGC GPU Cloud Docker and NVIDIA Docker
1

There are two GPUs on my machine that are linked via NVLink and in the same PCIe RC. I run two docker containers which only a single GPU is visible by --gpus deivec=xx parameter.
Later in the test, I was found that the two containers use cudaDeviceCanPeerAccess return can parameter is 0, and cudaDeviceEnabledPeerAccess will return an error. They are unable use p2p.
But when I run the two GPUs in the same container, they are able to use p2p.
Could someone please tell me how to solve this problem? This is very important to me