Question

A retail company ingests clickstream and order events into Amazon Kinesis Data Streams. The downstream consumer must process all events for a single customer strictly in the order they occurred, but events across different customers can be processed in any order. The data engineer notices that events for the same customer are sometimes processed out of order. Which change will guarantee per-customer ordering?

Accepted Answer

Use the customer ID as the partition key when calling PutRecord so all of a customer's events map to the same shard.

Answer

Enable enhanced fan-out for the consumer to increase read throughput per shard.

Answer

Use a random UUID as the partition key to distribute records evenly across all shards.

Answer

Switch to Kinesis Data Firehose and buffer records for 60 seconds before delivery.