Checklist

• I have included the output of celery -A proj report in the issue.
  (if you are not able to do this, then at least specify the Celery
  version affected).

I can reproduce the issue on linux(inside a docker container) or on my own mac.
The celery report goes like this:

software -> celery:4.0.2 (latentcall) kombu:4.0.2 py:3.6.0
            billiard:3.5.0.2 py-amqp:2.1.4
platform -> system:Darwin arch:64bit imp:CPython
loader   -> celery.loaders.app.AppLoader
settings -> transport:amqp results:disabled

broker_url: 'amqp://guest:********@localhost:5672//'

• I have verified that the issue exists against the master branch of Celery.

Steps to reproduce

Running a simple script with:

from celery import Celery

from some_config import CELERY_BROKER

app = Celery('test_celery', broker='amqp://{user}:{password}@{host}:{port}/{vhost}'.format(**CELERY_BROKER))


class FunkyException(Exception):
    def __init__(self, desc=None, extra=None):
        super().__init__(desc)
        self.extra = extra


@app.task(bind=True)
def eat_memory(self):
    try:
        raise FunkyException('test', extra=bytearray(512000000))
    except Exception as exc:
        raise self.retry(exc=exc, countdown=5)


if __name__ == '__main__':
    eat_memory.delay()

Start celery with gevent:

celery worker -A test_celery:app -P gevent

Then run the script:

python test_celery.py

Expected behavior

I did't set the max_retries parameter, so the task should be executed after 15 seconds(3 times), and memory should be released after that.

Actual behavior

The memory leaks, and did not release. It might be released after like 5-10 minutes, and sometimes never released.

Actual behavior in my production

The celery worker process leaks slowly. I have to restart the worker every hour.

4.0.x and 4.1.0 are tested with this issue.