Apache Kafka - Real-time Stream Processing (Master Class) - Caution with States

The video tutorial addresses a problem in Kafka stream processing where invoices from different stores are partitioned based on store ID, leading to issues with local state stores. The solution involves ensuring all invoices for a customer are sent to the same partition, either by changing the message key to customer ID or using a custom partitioner. The tutorial explains how to implement these solutions using Kafka's API, emphasizing the importance of repartitioning and its impact on performance.

Explain how changing the message key can resolve the partitioning issue.

What is the role of a custom partitioner in the context of Kafka streams?

Why is repartitioning considered an expensive activity in streaming applications?