Stream & Batch Processing
Replay after bug
A deployed stream processor miscounted refunds for three days before a fix shipped. How do you rebuild correct aggregates without losing Kafka messages, and what idempotency does the sink need?
Advanced drill question.