ARTICLE AD BOX
Use DLX + TTL parking queue. Instead of requeuing, you ack the failed message and republish it to a "wait" queue that has a TTL. When the TTL expires, RabbitMQ dead-letters it back to your main queue automatically. No sleep, no blocking.
Set up the wait queue with its DLX pointing back to your main exchange. Then on failure, republish with a per-message TTL for backoff:
} catch (Exception e) { int retries = getRetryCount(delivery.getProperties()); long delay = (long) Math.pow(2, retries) * 1000; // 1s, 2s, 4s... Map<String, Object> headers = new HashMap<>( delivery.getProperties().getHeaders() != null ? delivery.getProperties().getHeaders() : Map.of()); headers.put("x-retry-count", retries + 1); AMQP.BasicProperties props = new AMQP.BasicProperties.Builder() .headers(headers) .expiration(String.valueOf(delay)) .deliveryMode(2) .build(); channel.basicPublish("wait.exchange", "task", props, delivery.getBody()); channel.basicAck(tag, false); // ack original, not nack }Track retries via a custom header and bail out after a max count.
RabbitMQ expires messages only from the queue head. A 16s message blocks a 1s message behind it. If that's a concern, use separate queues per delay tier with fixed x-message-ttl instead of per-message expiration.
