We Keep Coding

Different queues in celery

I have a project where I'm starting my FastAPI using a file (python main.py):
import uvicorn
from configuration import API_HOST, API_PORT
if __name__ == "__main__":
uvicorn.run("endpoints:app", host="localhost", port=8811, reload=True, access_log=False)
Inside endpoints.py I have:
from celery import Celery
from fastapi import FastAPI
import os
import time
# Create object for fastAPI
app = FastAPI(
title="MYFASTAPI",
description="MYDESCRIPTION",
version=1.0,
contact="ME!",
)
celery = Celery(__name__)
celery.conf.broker_url = os.environ.get("CELERY_BROKER_URL", "redis://localhost:6379")
celery.conf.result_backend = os.environ.get("CELERY_RESULT_BACKEND", "redis://localhost:6379")
celery.conf.task_track_started = True
celery.conf.task_serializer = pickle
celery.conf.result_serializer = pickle
celery.conf.accept_content = ["pickle"]
# By defaul celery can handle as many threads as CPU cores have the instance.
celery.conf.worker_concurrency = os.cpu_count()
# Start the celery worker. I start it in a separate thread, so fastapi can run in parallel
worker = celery.Worker()
def start_worker():
worker.start()
ce = threading.Thread(target=start_worker)
ce.start()
#app.post("/taskA")
def taskA():
task = ask_taskA.delay()
return {"task_id": task.id}
#celery.task(name="ask_taskA", bind=True)
def ask_taskA(self):
time.sleep(100)
#app.post("/get_results")
def get_results(task_id):
task_result = celery.AsyncResult(task_id)
return {'task_status': task_result.status}
Given this code, how can I have two different queues, assign a specific number of workers per earch queue and assign a specific task to one of these queues?
I read that people use to execute celery as:
celery -A proj worker
but there was a structure in the project that limited me because of some importings, and at the end I finished by starting the celery worker in a different thread (which works perfectly)

Based on the official celery documentation https://docs.celeryq.dev/en/stable/userguide/routing.html#manual-routing[1] you can follow this to specify different queues.
from kombu import Queue
app.conf.task_default_queue = 'default'
app.conf.task_queues = (
Queue('default', routing_key='task.#'),
Queue('feed_tasks', routing_key='feed.#'),
)
app.conf.task_default_exchange = 'tasks'
app.conf.task_default_exchange_type = 'topic'
app.conf.task_default_routing_key = 'task.default'

Celery not discovering tasks inside Docker

I have a very simple implementation.
/lib/queue/__init__.py
from celery import Celery
from os import environ
REDIS_URI = environ.get('REDIS_URI')
app = Celery('tasks',
broker=f'redis://{REDIS_URI}')
app.autodiscover_tasks([
'lib.queue.cache',
], force=True)
/lib/queue/cache/tasks.py
from lib.queue import app
#app.task
def some_task():
pass
Dockerfile
RUN git clone <my_repo> /usr/src/lib
WORKDIR /usr/src/lib
RUN python3 setup.py install
CMD ["celery", "-A", "worker:app", "worker", "--loglevel=info", "--concurrency=4"]
/worker.py
from lib.queue import app
This works just fine if I initialize command line without Docker.
celery -A worker:app worker --loglevel=info
> [tasks]
> . lib.queue.cache.tasks.some_task
However, when I run it inside Docker, the tasks remain blank:
> [tasks]
Question:
Any thoughts as to why celery would not be able to find the library and tasks inside Docker? I am using another Dockerfile with an almost identical setup to push the tasks, and it is able to access lib.queue.cache.tasks no problem.

Because I have been asked to provide my solution a couple times, here it is. HOWEVER, it may not really be helpful since what I am doing now is slightly different.
Inside my worker file, where app is defined, I have just a single task.
app = Celery("tasks", broker=f"redis://{REDIS_URI}:{REDIS_PORT}/{REDIS_DB}")
#app.task
def run_task(task_name, *args, **kwargs):
print(f"Running {task_name}. Received...")
print(f"- args: {args}")
print(f"- kwargs: {kwargs}")
module_name, method_name = task_name.split(".")
module = import_module(f".{module_name}", package="common.tasks")
task = getattr(module, method_name)
loop = asyncio.get_event_loop()
retval = loop.run_until_complete(task(*args, **kwargs))
This may not be relevant to most people since it takes a string argument to import a coroutine and execute that. This really is because my tasks are sharing some functions that also need to execute in async world.

Celery task routing doesn't work when using current_app.send_task

I have an issue with Celery queue routing when using current_app.send_task
I have two workers (each one for each queue)
python manage.py celery worker -E -Q priority --concurrency=8 --loglevel=DEBUG
python manage.py celery worker -Q low --concurrency=8 -E -B --loglevel=DEBUG
I have two queues defined in celeryconfig.py file:
# -*- coding: utf-8 -*-
from __future__ import unicode_literals
from django.core.exceptions import ImproperlyConfigured
from celery import Celery
from django.conf import settings
try:
app = Celery('proj', broker=getattr(settings, 'BROKER_URL', 'redis://'))
except ImproperlyConfigured:
app = Celery('proj', broker='redis://')
app.conf.update(
CELERY_TASK_SERIALIZER='json',
CELERY_ACCEPT_CONTENT=['json'],
CELERY_RESULT_SERIALIZER='json',
CELERY_RESULT_BACKEND='djcelery.backends.database:DatabaseBackend',
CELERY_DEFAULT_EXCHANGE='tasks',
CELERY_DEFAULT_EXCHANGE_TYPE='topic',
CELERY_DEFAULT_ROUTING_KEY='task.priority',
CELERY_QUEUES=(
Queue('priority',routing_key='priority.#'),
Queue('low', routing_key='low.#'),
),
CELERY_DEFAULT_EXCHANGE='priority',
CELERY_IMPORTS=('mymodule.tasks',)
CELERY_ENABLE_UTC = True
CELERY_TIMEZONE = 'UTC'
if __name__ == '__main__':
app.start()
In the definition of tasks, we use decorator to explicit the queue:
#task(name='mymodule.mytask', routing_key='low.mytask', queue='low')
def mytask():
# does something
pass
This task is run indeed in the low queue when this task is run using:
from mymodule.tasks import mytask
mytask.delay()
But it's not the case when it's run using: (it's run in the default queue: "priority")
from celery import current_app
current_app.send_task('mymodule.mytask')
I wonder why this later way doesn't route the task to the "low" queue!
p.s: I use redis.

send_task is a low-level method. It sends directly to the broker the task signature without going through your task decorator.
With this method, you can even send a task without loading the task code/module.
To solve your problem, you can fetch the routing_key/queue from configuration directly:
route = celery.amqp.routes[0].route_for_task("mymodule.mytask")
Out[10]: {'queue': 'low', 'routing_key': 'low.mytask'}
celery.send_task("myodule.mytask", queue=route['queue'], routing_key=route['routing_key']`

Celery auto reload on ANY changes

I could make celery reload itself automatically when there is changes on modules in CELERY_IMPORTS in settings.py.
I tried to give mother modules to detect changes even on child modules but it did not detect changes in child modules. That make me understand that detecting is not done recursively by celery. I searched it in the documentation but I did not meet any response for my problem.
It is really bothering me to add everything related celery part of my project to CELERY_IMPORTS to detect changes.
Is there a way to tell celery that "auto reload yourself when there is any changes in anywhere of project".
Thank You!

Celery --autoreload doesn't work and it is deprecated.
Since you are using django, you can write a management command for that.
Django has autoreload utility which is used by runserver to restart WSGI server when code changes.
The same functionality can be used to reload celery workers. Create a seperate management command called celery. Write a function to kill existing worker and start a new worker. Now hook this function to autoreload as follows.
import shlex
import subprocess
from django.core.management.base import BaseCommand
from django.utils import autoreload
def restart_celery():
cmd = 'pkill celery'
subprocess.call(shlex.split(cmd))
cmd = 'celery worker -l info -A foo'
subprocess.call(shlex.split(cmd))
class Command(BaseCommand):
def handle(self, *args, **options):
print('Starting celery worker with autoreload...')
# For Django>=2.2
autoreload.run_with_reloader(restart_celery)
# For django<2.1
# autoreload.main(restart_celery)
Now you can run celery worker with python manage.py celery which will autoreload when codebase changes.
This is only for development purposes and do not use it in production. Code taken from my other answer here.

You can manually include additional modules with -I|--include. Combine this with GNU tools like find and awk and you'll be able to find all .py files and include them.
$ celery -A app worker --autoreload --include=$(find . -name "*.py" -type f | awk '{sub("\./",""); gsub("/", "."); sub(".py",""); print}' ORS=',' | sed 's/.$//')
Lets explain it:
find . -name "*.py" -type f
find searches recursively for all files containing .py. The output looks something like this:
./app.py
./some_package/foopy
./some_package/bar.py
Then:
awk '{sub("\./",""); gsub("/", "."); sub(".py",""); print}' ORS=','
This line takes output of find as input and removes all occurences of ./. Then it replaces all / with a .. The last sub() removes replaces .py with an empty string. ORS replaces all newlines with ,. This outputs:
app,some_package.foo,some_package.bar,
The last command, sed removes the last ,.
So the command that is being executed looks like:
$ celery -A app worker --autoreload --include=app,some_package.foo,some_package.bar
If you have a virtualenv inside your source you can exclude it by adding -path .path_to_your_env -prune -o:
$ celery -A app worker --autoreload --include=$(find . -path .path_to_your_env -prune -o -name "*.py" -type f | awk '{sub("\./",""); gsub("/", "."); sub(".py",""); print}' ORS=',' | sed 's/.$//')

You can use watchmedo
pip install watchdog
Start celery worker indirectly via watchmedo
watchmedo auto-restart --directory=./ --pattern=*.py --recursive -- celery worker --app=worker.app --concurrency=1 --loglevel=INFO
More detailed

I used watchdog watchdemo utility, it works great but for some reason the PyCharm debugger was not able to debug the subprocess spawned by watchdemo.
So if your project has werkzeug as dependency, you can use the werkzeug._reloader.run_with_reloader function to autoreload celery worker on code change. Plus it works with PyCharm debugger.
"""
Filename: celery_dev.py
"""
import sys
from werkzeug._reloader import run_with_reloader
# this is the celery app path in my application, change it according to your project
from web.app import celery_app
def run():
# create copy of "argv" and remove script name
argv = sys.argv.copy()
argv.pop(0)
# start the celery worker
celery_app.worker_main(argv)
if __name__ == '__main__':
run_with_reloader(run)
Sample PyCharm debug configuration.
NOTE:
This is a private werkzeug API and is working as of Werkzeug==2.0.3. It may stop working in future versions. Use at you own risk.

OrangeTux's solution didn't work out for me, so I wrote a little Python script to achieve more or less the same. It monitors file changes using inotify, and triggers a celery restart if it detects a IN_MODIFY, IN_ATTRIB, or IN_DELETE.
#!/usr/bin/env python
"""Runs a celery worker, and reloads on a file change. Run as ./run_celery [directory]. If
directory is not given, default to cwd."""
import os
import sys
import signal
import time
import multiprocessing
import subprocess
import threading
import inotify.adapters
CELERY_CMD = tuple("celery -A amcat.amcatcelery worker -l info -Q amcat".split())
CHANGE_EVENTS = ("IN_MODIFY", "IN_ATTRIB", "IN_DELETE")
WATCH_EXTENSIONS = (".py",)
def watch_tree(stop, path, event):
"""
#type stop: multiprocessing.Event
#type event: multiprocessing.Event
"""
path = os.path.abspath(path)
for e in inotify.adapters.InotifyTree(path).event_gen():
if stop.is_set():
break
if e is not None:
_, attrs, path, filename = e
if filename is None:
continue
if any(filename.endswith(ename) for ename in WATCH_EXTENSIONS):
continue
if any(ename in attrs for ename in CHANGE_EVENTS):
event.set()
class Watcher(threading.Thread):
def __init__(self, path):
super(Watcher, self).__init__()
self.celery = subprocess.Popen(CELERY_CMD)
self.stop_event_wtree = multiprocessing.Event()
self.event_triggered_wtree = multiprocessing.Event()
self.wtree = multiprocessing.Process(target=watch_tree, args=(self.stop_event_wtree, path, self.event_triggered_wtree))
self.wtree.start()
self.running = True
def run(self):
while self.running:
if self.event_triggered_wtree.is_set():
self.event_triggered_wtree.clear()
self.restart_celery()
time.sleep(1)
def join(self, timeout=None):
self.running = False
self.stop_event_wtree.set()
self.celery.terminate()
self.wtree.join()
self.celery.wait()
super(Watcher, self).join(timeout=timeout)
def restart_celery(self):
self.celery.terminate()
self.celery.wait()
self.celery = subprocess.Popen(CELERY_CMD)
if __name__ == '__main__':
watcher = Watcher(sys.argv[1] if len(sys.argv) > 1 else ".")
watcher.start()
signal.signal(signal.SIGINT, lambda signal, frame: watcher.join())
signal.pause()
You should probably change CELERY_CMD, or any other global variables.

There was an issue in #AlexTT answer, I don't know if I should comment on his answer of put this as an answer.
You can use watchmedo
pip install watchdog
Start celery worker indirectly via watchmedo
watchmedo auto-restart --directory=./ --pattern=*.py --recursive -- celery -A <app> worker --concurrency=1 --loglevel=INFO

This is the way I made it work in Django:
# worker_dev.py (put it next to manage.py)
from django.utils import autoreload
def run_celery():
from projectname import celery_app
celery_app.worker_main(["-Aprojectname", "-linfo", "-Psolo"])
print("Starting celery worker with autoreload...")
autoreload.run_with_reloader(run_celery)
Then run python worker_dev.py. This has an advantage of working inside docker container.

This is a huge adaptation from Suor's code.
I made a custom Django command which can be called like this:
python manage.py runcelery
So, every time the code changes, celery's main process is gracefully killed and then executed again.
Change the CELERY_COMMAND variable as you wish.
# File: runcelery.py
import os
import signal
import subprocess
import time
import psutil
from django.core.management.base import BaseCommand
from django.utils import autoreload
DELAY_UNTIL_START = 5.0
CELERY_COMMAND = 'celery --config my_project.celeryconfig worker --loglevel=INFO'
class Command(BaseCommand):
help = ''
def kill_celery(self, parent_pid):
os.kill(parent_pid, signal.SIGTERM)
def run_celery(self):
time.sleep(DELAY_UNTIL_START)
subprocess.run(CELERY_COMMAND.split(' '))
def get_main_process(self):
for process in psutil.process_iter():
if process.ppid() == 0: # PID 0 has no parent
continue
parent = psutil.Process(process.ppid())
if process.name() == 'celery' and parent.name() == 'celery':
return parent
return
def reload_celery(self):
parent = self.get_main_process()
if parent is not None:
self.stdout.write('[*] Killing Celery process gracefully..')
self.kill_celery(parent.pid)
self.stdout.write('[*] Starting Celery...')
self.run_celery()
def handle(self, *args, **options):
autoreload.run_with_reloader(self.reload_celery)

How to pass settings to rq task?

I wonder what's the best approach to pass settings from a RQ worker to task function?
I want to keep all of my settings outside of the app in text file, then read it while worker is starting and then have those values for performing the actual task.
Right now I'm using os.environ.
Thanks!

It appears from the documentation from the section "Custom worker classes" that the way to do this is to write a custom worker class. Here is a pattern that I've been using which loads a section from a ConfigParser file into a virtual python module named config:
import configparser
import rq.worker
import sys
import types
class ConfigurableWorker(rq.worker.Worker):
"""An RQ Worker class which loads configuration from config.ini"""
def work(self, *args, **kwargs):
"""Perform RQ work"""
# Store some configuration on a virtual module
sys.modules['config'] = types.ModuleType('config')
import config
config_reader = configparser.ConfigParser()
config_reader.read('config.ini')
for key, value in config_reader['rq'].items():
setattr(config, key, value)
# Do work
super().work(*args, **kwargs)
And then from the job function:
def my_bunch_of_work():
import config
con = db_connection(url=config.my_db_connection)
Running the rq worker with a customer worker class can be done with the --worker-class argument.
$ rq worker --worker-class my.prog.ConfigurableWorker
It might be nice to do some kind of connection pooling setup in the ConfigurableWorker.work method, but I'm unsure of the safety of that without knowing too much about RQ worker fork model with os.fork()

Assuming you are taking about python-rq, you may want to read this page here and for config file here.

Try using the os module.
#IN WORKER
import os
os.environ['TOKEN'] = 'xxx'
# IN JOB
#Open child processes via os.system(), popen() or fork() and execv()
someVariable = os.environ['TOKEN']