Fleet is a generic distributed task distribution framework based on a distributed file system. Task distribution frameworks like Ray and Celery require network connections for communication, which makes them difficult to use in clusters with poor network conditions. Fleet is a distributed framework based on a shared file system, independent of any network communication, allowing for task distribution among nodes without any network connections.
- Distributed task distribution based on a shared file system
- Supports dynamic scaling
- Supports worker node heartbeat
- Supports manager node restart
- Supports set timeout for each task
- Supports set max_job and max_work_time for worker
- Pure Python implementation
pip install open-fleetSee more examples in ./examples.
# run_manager.py
from fleet.manager import Manager
from fleet.config.config import get_args
base_dir = "./share_dir"
def main():
job_list = [1,2,3,4]
args = get_args(f"--base_dir {base_dir}")
manager = Manager(args=args, job_list=job_list)
manager.run()
if __name__ == '__main__':
main()# run_worker.py
from fleet.worker import Worker
from fleet.config.config import get_args
from run_manager import base_dir
def add_one(x, info):
return {"status": "success", "result": x + 1}
def main():
args = get_args(f"--base_dir {base_dir}")
worker = Worker(args=args, job_func=add_one)
worker.run()
if __name__ == '__main__':
main()