Configuration

An app is configured entirely through the Ardiq(...) constructor:

from ardiq import Ardiq

app = Ardiq(
    redis_url="redis://localhost:6379",
    queue_name="emails",
    priorities=["low", "default", "high"],
    concurrency=32,
    result_ttl_ms=600_000,
)

Options

Option	Default	Description
redis_url	redis://localhost:6379	Redis connection URL.
queue_name	"default"	Logical queue — namespaces all Redis keys for this app.
priorities	["default"]	Priority lane names, lowest-first. Higher lanes drain first.
concurrency	16	Max tasks running at once in a worker.
prefetch	concurrency * 2	Max tasks held in memory; drives backpressure against Redis.
idle_timeout_ms	60000	When an unrenewed in-flight task may be reclaimed by another worker.
result_ttl_ms	300000	How long results live. 0 drops results immediately; a negative value keeps them forever.
burst	False	Exit once the queue drains (also settable via app.burst or the CLI --burst).
serializer	msgpack	Callable[[Any], bytes] used to encode arguments and results.
deserializer	msgpack	Callable[[bytes], Any] used to decode them.
cron_poll_s	1.0	Seconds between re-checks of each @app.cron’s next fire time.

Note

redis_url, queue_name, and priorities are positional-or-keyword; serializer, deserializer and cron_poll_s are keyword-only. The remaining options (concurrency, prefetch, idle_timeout_ms, result_ttl_ms, burst) are forwarded to the Rust core as keyword arguments.

Notes on specific options

queue_name

Each app/worker pair operates on one logical queue. Workers sharing a queue_name form a Redis consumer group and split the work between them. Use distinct names to isolate unrelated workloads on the same Redis instance.

priorities

The list is lowest-priority first. With ["low", "default", "high"], a worker drains high before default before low. A task’s lane comes from its @task(priority=...) default or a per-call .options(priority=...) override.

concurrency and prefetch

concurrency caps how many tasks execute simultaneously. prefetch caps how many are pulled into memory ahead of execution; a larger prefetch smooths bursts but holds more work off Redis. The default prefetch = concurrency * 2 is a sensible starting point.

idle_timeout_ms

A running task periodically renews its claim via a heartbeat. If a worker dies, its in-flight tasks stop renewing; after idle_timeout_ms another worker reclaims them (XAUTOCLAIM). Lower it for faster recovery, raise it if tasks legitimately run long without renewing.

result_ttl_ms

Controls result retention:

• positive — keep results this many ms (default 5 minutes).
• 0 — don’t store results at all.
• negative — keep results forever (you manage cleanup yourself).

serializer / deserializer

See Serialization. Every process on a queue must use the same codec.

cron_poll_s

How often a running worker re-stages each recurring task’s next occurrence. The default of 1s is fine for minute-resolution cron; lower it only if you rely on short every= intervals.