How would you handle async order processing challenges with Kafka?

Focuses on building a resilient async order-processing pipeline using Kafka in a high-throughput, latency-sensitive context (e.g., quick-commerce like Blinkit).

Expect discussion across several layers:

• Kafka tuning: partition strategy, consumer group scaling, idempotent producers, exactly-once semantics vs. at-least-once trade-offs
• Backpressure & lag: monitoring consumer lag, dead-letter queues for poison pills, retry topics with exponential backoff
• Backend resilience: idempotency keys to prevent duplicate order creation, saga/outbox pattern for distributed transactions across inventory/payment/delivery services
• Frontend checks: optimistic UI updates, polling or WebSocket-based status sync, graceful degradation when order state is uncertain

Senior candidates should address failure modes (broker unavailability, duplicate messages, out-of-order events) and operational concerns (schema evolution with Avro/Protobuf, observability via consumer-lag metrics).