Question

A ride-sharing company ingests location events into a Kinesis Data Stream using the vehicle's city name as the partition key. Most trips occur in one large city, so that partition key routes nearly all records to a single shard, causing ProvisionedThroughputExceededException errors even though the stream has 10 shards and overall traffic is well below the aggregate capacity. Producers are throttled while other shards remain idle. What is the most effective way to fix this uneven load distribution?

Accepted Answer

Change the partition key to a high-cardinality value such as the vehicle ID or a composite key, so records are distributed evenly across shards.

Answer

Increase the stream's retention period from 24 hours to 168 hours so records are buffered longer before consumption.

Answer

Enable enhanced fan-out on the consumers to give each consumer a dedicated 2 MB/s read throughput per shard.

Answer

Switch the stream to on-demand capacity mode, which removes the need to manage shards entirely.