Backpressure

TL;DR

Explanation

Without backpressure: fast producer + slow consumer → queue grows indefinitely → OOM or unbounded latency. Backpressure strategies: (1) Block producer (synchronous — producer waits until consumer is ready). (2) Drop messages (circuit breaker — shed load). (3) Buffer with size limit (bounded queue — producer blocks or errors when full). (4) Rate limiting producer. (5) Signal via protocol (TCP flow control, Reactive Streams, gRPC flow control). In PHP: Redis list maxlength, queue bounded size, Swoole channel with capacity, HTTP 429 Too Many Requests. ReactiveX / RxPHP: Observable with backpressure operators.

Common Misconception

Why It Matters

Common Mistakes

• Infinite queue — hides the problem until OOM.
• No backpressure to HTTP clients — API accepts more than it can process.
• Blocking the event loop with backpressure — use async/non-blocking where possible.

Code Examples

// Unbounded Redis list — can grow until OOM:
$redis->rpush('jobs', $job); // No limit

// Bounded with backpressure:
$len = $redis->llen('jobs');
if ($len >= 10000) {
    throw new ServiceUnavailableException('Queue full — apply backpressure');
    // HTTP 503 to caller — let them retry
}
$redis->rpush('jobs', $job);

References

• ↗ https://mechanical-sympathy.blogspot.com/2012/05/apply-back-pressure-when-overloaded.html

Tags

Related Terms