Implementing Order Auto‑Close with Delayed Tasks: Best Practices and Pitfalls

In e‑commerce and payment systems, orders that are placed but not paid need to be automatically closed after a specific time window; major platforms achieve this with high precision, often within a one‑second margin.

Common ways to implement such delayed tasks include using the delayed delivery features of message queues like RocketMQ, RabbitMQ, or Pulsar, and using Redisson's DelayedQueue built on Redis sorted sets.

Some widely circulated solutions are problematic: relying on Redis key‑space expiration notifications is unreliable because expiration events are generated only when the server actually deletes the key, which may occur minutes after the intended time, and the fire‑and‑forget nature of key‑space notifications can cause lost events during client disconnections.

RabbitMQ's dead‑letter mechanism can also be misused for delayed tasks; a message becomes a dead letter when it is negatively acknowledged without requeue, exceeds its TTL, or the queue exceeds its maximum length. While dead‑letter queues collect undelivered messages, they do not guarantee timely delivery, prompting the use of the official rabbitmq-delayed-message-exchange plugin for proper delayed messaging.

Time wheels offer an efficient in‑memory data structure for scheduling, but most implementations lack persistence; a process crash will lose all scheduled tasks, making them unsuitable without additional safeguards.

Redisson's delay queue uses a Redis sorted set where each element's score is the delivery timestamp; a background scan with zrangebyscore moves due messages to a ready list, providing durability as long as Redis remains operational.

Conclusion: Prefer message queues with native delayed delivery (e.g., RocketMQ, Pulsar). If unavailable, Redisson delay queue is a viable Redis‑based alternative, but must include compensation mechanisms for Redis failures. Time wheels can be considered only with robust backup strategies, and Redis expiration listening should be avoided for scheduling delayed tasks.