[BUG] Segmentation fault (core dumped) #2632
Description
Description
When I use multithreading or asyncio to make requests concurrently, I eventually get a "Segmentation fault (core dumped)" error. There aren’t any other logs — it just crashes. It usually happens after the different crews have been running in parallel for a while. The same problem for two approaches. Tried with 5+ threads / semaphores.
I'm using Python 3.10, 32 GB of RAM. For the LLM, I’m connecting to a private API where the model is hosted by vLLM.
Steps to Reproduce
use concurrency in Python for the Crew
Expected behavior
no crash + appropriate logging of errors / warnings if possible
Screenshots/Code snippets
1. Threads
from concurrent.futures import ThreadPoolExecutor, as_completed
def generate_response(payload: dict):
try:
output = MyAgent().crew().kickoff(inputs=payload['inputs'])
except Exception as e:
output = None
return output
def main():
results = []
with ThreadPoolExecutor(max_workers=5) as executor:
futures = [executor.submit(generate_response_draft(payload) for payload in payloads]
for future in as_completed(futures):
results.append(future.result())
2. Coroutines
import asyncio
sem = asyncio.Semaphore(5)
async def generate_response(payloads: dict):
with sem:
try:
output = await MyAgent().crew().kickoff_async(inputs=payloads['inputs'])
except Exception as e:
output = None
return output
async def main():
tasks = [generate_response_draft(payload) for payload in payloads]
results = await asyncio.gather(*tasks)
Operating System
Ubuntu 20.04
Python Version
3.10
crewAI Version
0.108.0
crewAI Tools Version
0.38.1
Virtual Environment
Venv
Evidence
Crash with a message: Segmentation fault (core dumped). No other logs.
Possible Solution
None
Additional context
None