Django Q2

Django Q2 Background Jobs

Use this before touching task enqueueing, schedules, worker deployment, Q_CLUSTER, or code imported by Django Q2 workers.

Project Defaults

• The dependency is django-q2; the Python import path is django_q.
• django_q is in INSTALLED_APPS; its migrations provide task result, schedule, and broker models.
• Q_CLUSTER lives in the generated Django settings module and uses Redis through REDIS_URL by default.
• The worker process is uv run --no-sync python manage.py qcluster; the local workers Docker Compose service and deployment APP_PROCESS_TYPE=worker path use that command.
• Redis also backs the default Django cache. If you remove Redis as the broker, review cache, health check, Docker, deployment, and documentation references separately.

Mental Model

• Web code calls async_task(...) or creates Schedule rows.
• The broker stores queued task packages until a qcluster process reserves work.
• Worker processes execute importable Python functions and save results or failures.
• The scheduler runs inside the cluster. Schedules are database rows; they do nothing unless qcluster is running.

Adding Tasks

1. Put the task function in the app that owns the behavior, usually apps/core/tasks.py for shared product work.
2. Keep the function importable at module import time. Do not rely on request objects, local closures, or process-local state.
3. Pass durable identifiers such as primary keys, not model instances, open files, connections, or large payloads.
4. Make the task idempotent. Redis does not give exactly-once execution guarantees, and receipt-based brokers can re-run work.
5. If the task depends on a just-saved database row, enqueue it from transaction.on_commit(...).

# apps/core/tasks.py
def send_welcome_email(user_id: int) -> None:
    from django.contrib.auth import get_user_model

    user = get_user_model().objects.get(pk=user_id)
    ...

from django.db import transaction
from django_q.tasks import async_task

transaction.on_commit(
    lambda: async_task("apps.core.tasks.send_welcome_email", user.pk)
)

Use q_options when Django Q2 options would collide with task kwargs:

async_task(
    "apps.core.tasks.rebuild_report",
    report_id,
    q_options={"timeout": 300, "group": "reports"},
)

Scheduling Work

Prefer named, idempotent schedules created by a migration, admin action, or setup command. Avoid creating schedules unconditionally at import time or app startup.

from django_q.models import Schedule

Schedule.objects.get_or_create(
    name="clear-expired-sessions",
    defaults={
        "func": "django.core.management.call_command",
        "args": "'clearsessions'",
        "schedule_type": Schedule.HOURLY,
    },
)

Use Schedule.objects.get_or_create(name=..., defaults={...}) when seeding schedules so repeated setup does not duplicate jobs. Cron schedules require the optional croniter dependency; do not use Schedule.CRON unless the generated project includes it.

Missed schedules catch up by default. Set Q_CLUSTER["catch_up"] = False when a job should run once after downtime instead of replaying every missed interval.

Broker Choices

Redis Broker

This template should normally use Redis:

Q_CLUSTER = {
    "name": "...",
    "timeout": 3600,
    "workers": 4,
    "redis": REDIS_URL,
}

Redis is fast and matches local Docker, deployment workers, and the default cache configuration. The default Redis broker does not support delivery receipts. If a worker host dies catastrophically while executing a task, the in-flight package can be lost; if task code raises, Django Q2 records a failure. Use idempotent task design, explicit retries in task code where needed, and monitoring for failures.

ORM Broker

Use the Django database broker only for low-throughput deployments, local simplicity, or environments where Redis is unavailable:

Q_CLUSTER = {
    "name": "...",
    "timeout": 3600,
    "retry": 4800,
    "workers": 4,
    "max_attempts": 2,
    "orm": "default",
}

When switching to ORM:

• Remove the "redis" broker key; configure one broker per cluster unless you intentionally use custom clusters.
• Run migrations for django_q. If the broker uses a non-default database alias, run migrations with --database <alias>.
• Increase "poll" above the default 0.2 seconds, for example "poll": 2.0, when you need lower database polling pressure and can tolerate higher queue pickup latency.
• The ORM broker enables the Queued Tasks admin table.
• Review Redis-dependent cache, health check, Docker, and deployment settings separately. Schedules are always database rows; the broker setting controls queued task packages, not the schedule table.

Testing

• Test task business logic by calling the function directly.
• Test enqueueing with synchronous mode:
  • per call: async_task("apps.core.tasks.fn", arg, sync=True)
  • per test: override Q_CLUSTER["sync"] = True
• For worker/broker integration, run uv run --no-sync python manage.py qcluster in a separate process and wait for result(task_id, 200) or a similar bounded wait; do not rely on arbitrary sleeps.
• Use pytest.mark.django_db(transaction=True) when a real worker process must observe committed database rows.

Debugging Checklist

• Is a qcluster process running with the same settings module, SECRET_KEY, broker URL, and cluster name as the web process?
• Can the worker import the dotted task path?
• Did database migrations run, including django_q migrations?
• Is Redis reachable from both web and worker containers, or is the ORM broker polling the expected database?
• Did a scheduled task duplicate because setup created another Schedule row with no stable name?
• Did downtime trigger schedule catch-up?
• Is task failure visible in Django admin, logs, or the configured error reporter?

References

• Official docs: https://django-q2.readthedocs.io/en/master/
• Upstream repo: https://github.com/django-q2/django-q2