Backpressure occurs when producers outpace consumers. Without handling, systems crash from memory exhaustion.
Strategies:
Control producer: Slow down based on consumer pace (reactive streams, request(n))
Buffer: Accumulate with size limits. Drop oldest or newest when full.
Drop: Discard messages during overload (acceptable for metrics, not orders)
Sample: Keep every Nth message during overload
Kafka uses pull-based consumption. Consumers control their own pace. Natural backpressure without explicit signaling.