Backpressure in Go pipelines, done properly

A pipeline without backpressure is a memory leak waiting for a slow stage. In Go, the built-in answer is a bounded channel: when it is full, the producer blocks, and pressure propagates upstream naturally.

func stage(ctx context.Context, in <-chan Job, workers int) <-chan Result {
	out := make(chan Result, workers) // small, bounded buffer
	var wg sync.WaitGroup
	for i := 0; i < workers; i++ {
		wg.Add(1)
		go func() {
			defer wg.Done()
			for job := range in {
				select {
				case out <- process(job):
				case <-ctx.Done():
					return
				}
			}
		}()
	}
	go func() { wg.Wait(); close(out) }()
	return out
}

Checklist

• Every channel has a deliberate capacity, and 0 or a small number is usually right.
• Every send is paired with ctx.Done() so shutdown never deadlocks.
• At the network edge, shed load (return 429) instead of queueing forever.

Blocking is a feature: it is the cheapest form of flow control you have.