Message Queues & Async Communication

Synchronous coupling creates chains of failure: if the email service is slow, the order service is slow. Message queues decouple producers from consumers — the order service publishes an event and returns immediately; the email service processes it when it can. This improves availability, absorbs traffic spikes, and enables independent scaling of each service.

Kafka is the dominant choice for high-throughput event streaming. AWS SQS and RabbitMQ handle task queues — background jobs, email sending, file processing. The key design decision is whether consumers need to replay events (Kafka retains messages for days) or just process once (SQS deletes after acknowledgement). Dead-letter queues catch repeatedly failing messages so they can be inspected without blocking the main queue.

Before

Synchronous — slow email blocks order response

1// User waits for ALL of these before getting a response2async function placeOrder(order: Order) {3  await db.saveOrder(order);          // 20ms4  await emailService.sendConfirm();   // 800ms (if slow)5  await inventoryService.reserve();   // 150ms6  await analyticsService.track();     // 300ms7  return { success: true };           // total: ~1,270ms8}

After

Async — order confirmed in 25ms, rest runs in background

1async function placeOrder(order: Order) {2  await db.saveOrder(order);                    // 20ms3  await queue.publish('order.placed', order);   // 5ms4  return { success: true };                     // total: 25ms5}6 7// These run independently, in parallel, at their own pace:8queue.subscribe('order.placed', emailService.sendConfirm);9queue.subscribe('order.placed', inventoryService.reserve);10queue.subscribe('order.placed', analyticsService.track);

Check your understanding

What does a queue buy the caller?Show answerHide answer
Answer
It can return after publish; slow consumers no longer block the request path.
Kafka vs SQS-style queues — one distinction?Show answerHide answer
Answer
Kafka retains a stream for replay/consumers at their pace; many task queues delete after ack.
What are dead-letter queues for?Show answerHide answer
Answer
Parking repeatedly failing messages for inspection without blocking the main queue.