How do I specify num_returns="dynamic" for an Actor?

Ray Core
1

I can write a Ray function that returns a generator like so

from typing import List, Iterable

import ray

@ray.remote(num_returns="dynamic")
def capitalize(strings: List[str]) -> Iterable[str]:
    for string in strings:
        yield string.upper()

def main():
    strings = ["The", "man", "is", "named", "John"]
    g = capitalize.remote(strings)
    for item in ray.get(g):
        print(ray.get(item))

if __name__ == "__main__":
    main()

This produces the expected output.

2023-04-14 16:11:55,103	INFO worker.py:1364 -- Connecting to existing Ray cluster at address: 127.0.0.1:6379...
2023-04-14 16:11:55,111	INFO worker.py:1544 -- Connected to Ray cluster. View the dashboard at 127.0.0.1:8265 
THE
MAN
IS
NAMED
JOHN

I’d like to do the same thing with an Actor class.

from typing import List, Iterable

import ray

@ray.remote
class Capitalizer:
    def capitalize(self, strings: List[str]) -> Iterable[str]:
        for string in strings:
            yield string.upper()

def main():
    strings = ["The", "man", "is", "named", "John"]
    c = Capitalizer.remote()
    g = c.capitalize.remote(strings)
    for item in ray.get(g):
        print(ray.get(item))

if __name__ == "__main__":
    main()

When I run the Actor class version I get the following error:

Traceback (most recent call last):
  File "/Users/bill.mcneill/Src/mine/simple.py", line 35, in <module>
    main()
  File "/Users/bill.mcneill/Src/mine/simple.py", line 31, in main
    print(ray.get(item))
  File "/usr/local/anaconda3/envs/mine/lib/python3.9/site-packages/ray/_private/client_mode_hook.py", line 105, in wrapper
    return func(*args, **kwargs)
  File "/usr/local/anaconda3/envs/mine/lib/python3.9/site-packages/ray/_private/worker.py", line 2369, in get
    raise ValueError(
ValueError: 'object_refs' must either be an ObjectRef or a list of ObjectRefs.
(Capitalizer pid=11434) 2023-04-14 16:05:33,556	ERROR worker.py:772 -- Unhandled error: Task threw exception, but all 1 return values already created. This should only occur when using generator tasks.
(Capitalizer pid=11434) See https://github.com/ray-project/ray/issues/28689.
(Capitalizer pid=11434) Traceback (most recent call last):
(Capitalizer pid=11434)   File "python/ray/_raylet.pyx", line 931, in ray._raylet.execute_task
(Capitalizer pid=11434)   File "python/ray/_raylet.pyx", line 968, in ray._raylet.execute_task
(Capitalizer pid=11434)   File "python/ray/_raylet.pyx", line 2500, in ray._raylet.CoreWorker.store_task_outputs
(Capitalizer pid=11434) ValueError: Task returned more than num_returns=1 objects.

I think the problem here is that I haven’t specified num_returns="dynamic" anywhere in my Actor class. But I can’t figure out where to specify this.

How do I get the Actor version of this algorithm to work?

2

You can do

import ray

ray.init()

@ray.remote
class Actor:
    @ray.method(num_returns="dynamic")
    def generator(self):
        for i in range(10):
            yield i

actor = Actor.remote()
print(list(ray.get(actor.generator.remote())))
3

@wpm
I hope that answers your question. Thanks @jjyao for the quick response. If that solves the problem, @wpm can close this issue as resolved. Thanks for using Ray!

HTH

4

That’s the answer. Thanks.