redis_task_queue 0.2.0

A small Redis-backed task queue for server-side Dart. Enqueue jobs and process them in a worker with retries, a dead-letter list, and weighted queues.

redis_task_queue

A small Redis-backed task queue for server-side Dart. Enqueue work from your request path and process it in a separate worker — with retries, a dead-letter list, and weighted queues so one noisy queue can't starve the others.

If you've used Asynq in Go or Sidekiq in Ruby, the model will feel familiar. Dart server frameworks (Serverpod, Dart Frog, Shelf) don't have a maintained equivalent, so this fills that gap with a deliberately small surface.

The path a task takes, end to end:

---
config:
  look: handDrawn
---
flowchart LR
    P["Producer<br/>client.enqueue(task, queue, maxRetries)"] -->|LPUSH| Q
    subgraph Q["Redis lists — drained by weight"]
      direction TB
      C["critical &times;6"]
      D["default &times;3"]
      L["low &times;1"]
    end
    Q -->|"weighted BRPOP (1s block)"| W["Worker poll loop"]
    W --> H["Handler for task.type"]
    H -->|returns normally| DONE(["done"])
    H -->|throws / no handler| R{"attempts left?"}
    R -->|"yes — ZADD with backoff, attempt++"| Z[["delayed set<br/>&lt;prefix&gt;:&lt;queue&gt;:delayed"]]
    Z -->|"due-mover: score &le; now &rarr; LPUSH"| Q
    R -->|no| DL[["dead-letter list<br/>&lt;prefix&gt;:dead"]]

Why

Anything slow or retryable — sending email, processing an upload, calling a flaky third-party API — shouldn't run inside the request. It should go on a queue and be handled out of band, where a failure can be retried instead of turning into a 500 the user sees.

That's all this does: a producer drops a task onto Redis and returns immediately; a worker picks it up, runs it, and retries on failure until it either succeeds or lands in the dead-letter list.

Install

dependencies:
  redis_task_queue: ^0.2.0

Enqueue (from your request path)

final client = await QueueClient.connect(); // localhost:6379 by default

await client.enqueue(
  Task('email:welcome', {'user_id': '42'}),
  queue: 'default',
  maxRetries: 5,
);

enqueue is a single LPUSH — keep one client around and reuse it.

Process (in a separate worker process)

final worker = await Worker.connect(
  queues: {'critical': 6, 'default': 3, 'low': 1},
  // Optional — these are the defaults. The first retry waits backoffBase,
  // each further retry doubles it up to backoffCap, plus a bit of jitter.
  backoffBase: const Duration(seconds: 1),
  backoffCap: const Duration(seconds: 60),
  backoffJitter: 0.1, // 0..1; fraction of the delay added at random
);

worker.handle('email:welcome', (task) async {
  // Real work. Throwing triggers a retry; returning marks the task done.
  await sendWelcomeEmail(task.payload['user_id'] as String);
});

await worker.run();

How it behaves

A task moves through a small set of states — it either lands on done or, once retries are exhausted, on the dead-letter list:

---
config:
  look: handDrawn
---
stateDiagram-v2
    [*] --> pending: enqueue / LPUSH
    pending --> processing: worker weighted BRPOP
    processing --> done: handler returns
    processing --> retry: handler throws
    retry --> delayed: ZADD with backoff (attempt++)
    retry --> deadLetter: attempt reached maxRetries
    delayed --> pending: due-mover promotes (score ≤ now)
    done --> [*]
    deadLetter --> [*]

• Weighted queues. With {'critical': 6, 'default': 3, 'low': 1} the worker polls critical about six times as often as low, so a flood of low-priority jobs can't starve important ones.
• Retries with exponential backoff. A handler that throws is retried up to the task's maxRetries. Retries aren't immediate: the envelope goes into a per-queue delayed sorted set (<prefix>:<queue>:delayed) scored with the time it becomes due. The wait grows min(cap, base * 2^(retry-1)) — the first retry waits backoffBase (default 1s), each further one doubles up to backoffCap (default 60s) — plus a little jitter so a burst of failures doesn't re-fire in lockstep. All three are configurable on Worker.connect.
• Due-mover. Each poll-loop pass, before it blocks on BRPOP, the worker promotes any delayed tasks whose score has passed back onto their pending list. The move runs inside a single Redis Lua script (ZRANGEBYSCORE + ZREM + LPUSH), so it's atomic — a task can't be lost or duplicated, even if several workers run the mover at once. Because the pop uses a short (1s) timeout, a due task waits at most about a second past its scheduled time.
• Dead-letter list. Once retries are exhausted, the envelope moves to a dead-letter list (<prefix>:dead) instead of looping forever, so you can inspect what failed.
• Missing handler = failure. A task with no registered handler is retried, not silently dropped, so a wiring mistake surfaces loudly.

What this version keeps small (on purpose)

• No in-flight tracking. A task is popped with BRPOP, so if a worker crashes between the pop and finishing (or re-scheduling) the task, that task is lost — there's no processing set or visibility timeout to recover it. The delayed-retry mover itself is atomic and safe across workers; this caveat is about the pop/handle step, and it's out of scope for this version.
• No scheduler / cron, no unique-task dedup, no web UI. The goal is the enqueue → process → backoff-retry → dead-letter core, done clearly.

Requirements

• Dart 3.5+
• A running Redis instance

Running the example

# terminal 1
dart run example/redis_task_queue_example.dart worker
# terminal 2
dart run example/redis_task_queue_example.dart enqueue

License

MIT © Yusuf İhsan Görgel