Issues in ray.init() functionality

1

I am using ray version 1.0.1.post1 in Linux.
I have made the ray cluster of two systems.
My member node has one GPU also.
I wanted to use both the nodes and GPU also but when I am mentioning ray.init( address = "auto") I cannot use the GPU.
and when I want to explicitly mention the num_cpus according to total number of cores combining both the nodes and num_gpus it is not able to acccess the other node.

Also I want to limit the number of processes created and divide it among two machines in cluster. This is also not possible I guess ?

Am I understanding the API wrong or is there any bug ?

2

@Mohit_Ranawat not 100% sure I’m interpreting your question correctly, but when you run ray start on a machine to add that node to the Ray cluster, you can pass in --num-cpus or --num-gpus to constrain the resources that Ray will use on that machine.