TopolearnReal projects keep settings in one place, and a few defaults (serializer, timezone, prefetch) are worth setting deliberately.
Celery settings are lowercase names like task_serializer or broker_url. Set them with app.conf.update(...), or keep them in a module and load it with app.config_from_object("celeryconfig"). With Django, the usual setup is app.config_from_object("django.conf:settings", namespace="CELERY"), so CELERY_BROKER_URL in settings.py maps to broker_url, plus app.autodiscover_tasks() to import each app's tasks.py.
Settings worth knowing early:
broker_url,result_backend.task_serializer,result_serializer,accept_content: JSON by default. Pickle is disabled by default because unpickling messages can execute arbitrary code.timezoneandenable_utc(on by default) for ETAs and beat schedules.result_expiresfor result cleanup.task_acks_late,worker_prefetch_multiplier: delivery behavior, covered later.task_default_queue(defaultcelery).
A typical project layout is a package with celery.py defining the app and a tasks.py per area; the worker is started with celery -A proj worker, where proj is a package or module containing an attribute named app or celery.
Configuration must be loaded identically by clients and workers; mismatched serializers or task names are a frequent cause of tasks that "never run". celery -A proj report prints the effective configuration.
Example
# proj/celery.py
from celery import Celery
app = Celery("proj")
app.conf.update(
broker_url="redis://localhost:6379/0",
result_backend="redis://localhost:6379/1",
task_serializer="json",
accept_content=["json"],
result_expires=3600, # seconds
timezone="Europe/Berlin",
enable_utc=True,
include=["proj.tasks"], # modules the worker imports
)
# Django alternative:
# app.config_from_object("django.conf:settings", namespace="CELERY")
# app.autodiscover_tasks()
# $ celery -A proj worker -l INFORemember
- Settings are lowercase names set via app.conf or config_from_object
- JSON is the default and safest serializer
- With Django, namespace='CELERY' maps CELERY_* settings
- Clients and workers must share the same configuration and task modules
Watch out for
- Enabling pickle to pass complex objects, opening a code-execution hole if the broker is reachable.