IDDQueue

Documentation · Step-by-step setup

A PostgreSQL broker and Results backend for Dramatiq, built on synchronous Psycopg 3. Tasks, results and coordination live in PostgreSQL; no Redis or ORM is required by IDDQueue.

Current release: 0.13.0rc4 — a release candidate for evaluation. GitHub assets include the wheel, sdist and SHA256 checksums.

IDDQueue is a fork of DALIBO's dramatiq-pg (original package). It preserves the PostgreSQL license and original contributor credits.

What it provides

• Durable JSONB task/results storage, delayed tasks and crash recovery.
• Transactional publication alongside business writes; atomic batches and deduplication.
• PostgreSQL rate limiters, barriers and Dramatiq group callbacks.
• Pause/resume, cooperative cancellation, failed-task inspection/retry and attempt history.
• Fixed-interval scheduling and optional Prometheus queue metrics.
• Schema/prefix isolation and a tested FastAPI example.

Delivery is at least once: actors and callbacks must be idempotent. LISTEN/NOTIFY carries message IDs as wakeup hints; consumers claim authoritative SQL rows with session advisory locks. See behavior and limits.

Install

Requires Python 3.10+, Dramatiq 2.2.1+ and Psycopg 3.3.6+. CI tests Python 3.10 / 3.13 / 3.14 with PostgreSQL 14 / 18. See compatibility and API stability.

uv venv
uv pip install "iddqueue[binary]==0.13.0rc4"

The binary extra supplies libpq. Base installs require system libpq; [binary,monitoring] also enables Prometheus support.

Migrating from dramatiq-pg? Follow the migration guide before changing imports, pools or database storage. Stop all participants, back up data, run the upgrade and restart consistently. Fresh initialization and an existing-storage upgrade are different operations.

Quickstart

Configure PostgreSQL using PGHOST, PGPORT, PGUSER, PGDATABASE and your normal credential mechanism. Initialize fresh storage:

uv run --no-sync iddqueue init

For existing storage, use iddqueue upgrade after following the migration guide. Both commands manage the complete feature schema.

Save as tasks.py:

import dramatiq
from iddqueue import PostgresBroker

broker = PostgresBroker()
dramatiq.set_broker(broker)

@dramatiq.actor(broker=broker, store_results=True)
def add(left: int, right: int):
    return left + right

Start a separate worker from the same directory/environment:

uv run --no-sync dramatiq --use-spawn --processes=1 --threads=1 tasks

In another terminal, publish and read the result:

uv run --no-sync python - <<'PY'
from tasks import add, broker

try:
    message = add.send(2, 3)
    print(message.message_id)
    print(message.get_result(backend=broker.backend, block=True, timeout=10000))
finally:
    broker.close()
PY

The result is 5; timeouts are milliseconds. Each process creates its own pool. Close broker-owned pools on shutdown; callers close pools they supply.

Guides and examples

Domain actors and one worker bootstrap are included since RC3.

• Documentation index, user guide, API.
• Detailed recipes: transactions, middleware, controls, metrics and schedules.
• FastAPI: lifespan, async endpoint thread offload and separate worker.
• Deployment and retention, why PostgreSQL.
• RC notes, changelog, release checks.

The broker and result APIs are synchronous. Async actors use Dramatiq's AsyncIO middleware; FastAPI async endpoints offload publication to a thread. Transactional enqueue does not accept a Psycopg AsyncConnection.

Session-bound locks/listeners require persistent database sessions; PgBouncer transaction pooling is unsuitable. Namespace separation is not access control. The legacy django-dramatiq-pg integration has been archived since September 3, 2024; IDDQueue compatibility is unverified. No tested Django integration is currently provided.

Compared with dramatiq-pg

Baseline: dramatiq-pg 0.12.0, compared with the published IDDQueue 0.13.0rc4 prerelease. This describes the published version, not every future upstream revision. Both projects provide a PostgreSQL Dramatiq broker.

See migration from dramatiq-pg for dependency, pool, import, CLI and database changes, including backup and rollback. No throughput advantage is claimed here. The 0.13.0rc4 prerelease is available on PyPI; the legacy Django integration is archived and has not been verified with IDDQueue.

Development and support

Use CONTRIBUTING for locked uv setup and the full release gate. Functional tests require a dedicated PostgreSQL instance: they terminate sessions and crash/restart workers. Project planning and evidence live in OpenSpec. Report reproducible issues on GitHub.

License and credits

Released under the PostgreSQL license, preserving Copyright (c) 2019, DALIBO and the original author Étienne BERSAC.

Thanks to upstream contributors Andy Freeland, Curtis Maloney (Django support), Federico Caselli, Giuseppe Papallo and Rafal Kwasny. The historical upstream logo was created by Damien CAZEILS. Original changelog remains available as historical context.