Nested Remote Functions#

Remote functions can call other remote functions, resulting in nested tasks. For example, consider the following.

import ray

def f():
    return 1

def g():
    # Call f 4 times and return the resulting object refs.
    return [f.remote() for _ in range(4)]

def h():
    # Call f 4 times, block until those 4 tasks finish,
    # retrieve the results, and return the values.
    return ray.get([f.remote() for _ in range(4)])

Then calling g and h produces the following behavior.

>>> ray.get(g.remote())

>>> ray.get(h.remote())
[1, 1, 1, 1]

One limitation is that the definition of f must come before the definitions of g and h because as soon as g is defined, it will be pickled and shipped to the workers, and so if f hasn’t been defined yet, the definition will be incomplete.

Yielding Resources While Blocked#

Ray will release CPU resources when being blocked. This prevents deadlock cases where the nested tasks are waiting for the CPU resources held by the parent task. Consider the following remote function.

@ray.remote(num_cpus=1, num_gpus=1)
def g():
    return ray.get(f.remote())

When a g task is executing, it will release its CPU resources when it gets blocked in the call to ray.get. It will reacquire the CPU resources when ray.get returns. It will retain its GPU resources throughout the lifetime of the task because the task will most likely continue to use GPU memory.