Backpressure

Backpressure signals that downstream cannot keep up — rising queue depth, consumer lag, or broker disk use. Responses include slowing producers, limiting prefetch, scaling consumers, shedding load, or rejecting publishes when buffers exceed thresholds.

Trade-off: buffering smooths spikes but hides overload until catastrophic failure; early shedding protects the system but drops work.

On interviews: diagnose lag growing linearly; choose between scale consumers and throttle producers.

Common pitfalls: infinite buffer pretending capacity is unlimited; scaling consumers past partition count; no alerts on lag SLO.

Checklist:

• Define lag and depth SLOs with alerts.
• Cap prefetch and in-flight messages per consumer.
• Implement producer throttling or circuit breaking.
• Load test with sustained overload scenarios.