Backpressure

Definition

Backpressure is a control mechanism through which a slower consumer makes upstream producers reduce, pause or reject incoming work. It keeps buffers, queues and memory bounded when arrival exceeds processing capacity. The signal may be explicit demand, a queue limit, a rate adjustment or an admission refusal.

In plain terms

If water enters a pipe faster than it can leave, pressure rises. A job system has the same problem: accepting every request while workers are saturated only moves the wait into a larger queue. Backpressure carries the capacity problem upstream, where new work can slow down, wait elsewhere or receive an honest rejection.

This is different from silently dropping work. A clear signal tells the producer what happened and, where appropriate, when to try again.

Example

A queue is limited to 1,000 waiting jobs. Workers complete 50 jobs per minute, while producers suddenly submit 120 per minute.

The bound protects the scheduler and gives operators a visible overload signal. It does not create capacity; the system still needs to reduce arrivals, add safe capacity or shed low-value work.

Common misunderstandings

• “A large queue solves overload.” It delays the symptom while increasing latency and recovery time.
• “Backpressure means failure.” It is controlled refusal or slowing that protects the rest of the system.
• “Retries are the answer.” Immediate retries add more load; use a clear retry policy with Exponential Backoff where retry is appropriate.