logo

Backpressure

Backpressure is a flow control mechanism in systems that process streams of work. When a consumer (like a worker or API) can’t keep up with the rate of incoming tasks, backpressure signals the producer to slow down — preventing memory exhaustion, dropped messages, and cascading failures.

The Problem Without Backpressure

If producers generate work faster than consumers process it, the queue grows unbounded:

Producer: 1,000 tasks/sec → Queue: [growing...] → Worker: 100 tasks/sec
                                     ↑
                           Queue grows by 900/sec
                           Memory exhausted in minutes

Without backpressure, the system crashes once the queue fills available memory.

How Backpressure Works

Backpressure mechanisms limit how fast work enters the system:

  • Bounded queues: Reject or block new tasks when the queue reaches a maximum size
  • Rate limiting at ingress: Throttle producers to match consumer throughput
  • Dynamic scaling: Spin up more workers when queue depth increases
  • Load shedding: Drop low-priority work to protect high-priority tasks

Backpressure in Task Queues

Task queues like AsyncQueue handle backpressure naturally:

  1. Tasks are persisted to durable storage (Redis), not held in memory
  2. Workers pull tasks at their own pace — they’re never force-fed
  3. Queue depth is monitored and can trigger auto-scaling
  4. Rate limiting can cap how fast tasks are created
// Workers control their own concurrency
// If processing is slow, the queue buffers — no data is lost
await aq.tasks.create({
  callbackUrl: 'https://your-app.com/api/process',
  payload: data,
  retries: 3,
});

The queue acts as a buffer between producers and consumers, absorbing traffic spikes without overloading either end.

Monitoring for Backpressure

Key metrics to watch:

  • Queue depth: Number of pending tasks — growing means consumers are falling behind
  • Processing latency: Time from task creation to completion — increasing means backpressure is building
  • Worker utilization: If workers are at 100%, you need more capacity or need to throttle producers