Gramine supports for parallel computation?

[Chiao-Yu]

Hi there, I have a problem about parallelize some python computation using Gramine. We have some encrypted algorithm (based on python) inside some SGX enclave, and another party has some encrypted data to be computed by the algo. The problem consists of N similar subproblems and would greatly benefit from parallel computation. However, it seems like anything using fork & exec (multiprocessing, Ray, etc) would be forbidden, and we are looking for an alternative way to complete the task without breaking any data or algo privacy.

I wonder if there exists some good solution to the problem. Many thanks in advance.

[mkow]

Hi,

fork+exec should work out-of-the box in Gramine, but it's quite complex and slow, so ideally you'd want to use threads instead. They are also supported out of the box.

[Chiao-Yu]

thx! we have tried fork+exec but it was forbidden, and the only workaround we have found so far is to break external data (which is supposed to be encrypted all the time, not read by anyone including us) into different components, send them outside the enclave (encrypted) and then set up multiple processes (by multiple enclaves). Do you think this is the right way to go? Or is there a more clever way to use fork+exec inside the enclave?

[mkow]

I'd rather investigate why you got "forbidden", it should work. Please check the Gramine documentation about the logging and look what's happening around the failure.

[Chiao-Yu]

Not sure if we screwed up some basic things, but please see the following example

import time
import multiprocessing


def square_number(number):
    time.sleep(1) 
    print(f"Process {multiprocessing.current_process().name} is computing {number}...")
    return number * number

if __name__ == "__main__":
    numbers = [1, 2, 3, 4, 5, 6, 7, 8]
    
    print("Starting the main process...")
    
    with multiprocessing.Pool(processes=4) as pool:
        results = pool.map(square_number, numbers)
        
    print("Tasks completed.")
    print("Final result:", results)
    # output: [1, 4, 9, 16, 25, 36, 49, 64]

with an error message:

Starting the main process...
Traceback (most recent call last):
  File "//app/pool-multiprocessing.py", line 17, in <module>
    with multiprocessing.Pool(processes=4) as pool:
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.11/multiprocessing/context.py", line 119, in Pool
    return Pool(processes, initializer, initargs, maxtasksperchild,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.11/multiprocessing/pool.py", line 191, in __init__
    self._setup_queues()
  File "/usr/lib/python3.11/multiprocessing/pool.py", line 346, in _setup_queues
    self._inqueue = self._ctx.SimpleQueue()
                    ^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.11/multiprocessing/context.py", line 113, in SimpleQueue
    return SimpleQueue(ctx=self.get_context())
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.11/multiprocessing/queues.py", line 341, in __init__
    self._rlock = ctx.Lock()
                  ^^^^^^^^^^
  File "/usr/lib/python3.11/multiprocessing/context.py", line 68, in Lock
    return Lock(ctx=self.get_context())
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.11/multiprocessing/synchronize.py", line 169, in __init__
    SemLock.__init__(self, SEMAPHORE, 1, 1, ctx=ctx)
  File "/usr/lib/python3.11/multiprocessing/synchronize.py", line 57, in __init__
    sl = self._semlock = _multiprocessing.SemLock(
                         ^^^^^^^^^^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory

[mkow]

Ok, but that's unrelated to multiprocessing itself, it's just that Python's multiprocessing library uses SemLock, which I think use something that Gramine doesn't support. You'd need to look at Gramine logs (see my previous message) and see which syscall exactly is used by SemLock.
https://gramine.readthedocs.io/en/stable/manifest-syntax.html#log-level

[Chiao-Yu]

Thanks. Are you aware of any solution that supports multiple processing in python? I think there are something like Pipe but not sure if it really works.