Question

A media company ingests clickstream events into a single Amazon Kinesis Data Stream. Two independent teams need to process every record in near real time: one team runs analytics aggregation, and the other performs fraud detection. Currently both teams' Lambda consumers share the stream, and as they add more consumers the per-consumer read latency keeps increasing because they compete for the shard's 2 MB/second read throughput. Which approach best resolves the throughput contention while keeping both consumers reading all records?

Accepted Answer

Enable enhanced fan-out consumers so each consumer gets a dedicated 2 MB/second per shard throughput via a registered stream consumer

Answer

Increase the stream retention period to 168 hours so consumers can spread their reads over time

Answer

Switch both Lambda functions to a single shared consumer that writes to an SQS queue the teams poll separately

Answer

Reduce the Lambda batch size so each consumer reads fewer records per GetRecords call