Kinesis Shard Scaling to Improve Consumer Performance

Quick Answer

This is a 'choose two' question, and the two correct actions address consumer throughput from different angles, one about how many total shards exist, and one about how each consumer reads from those shards. Increasing the number of shards raises the total read capacity of the stream, since each shard carries its own fixed throughput ceiling, roughly 2 MB per second, or 5 read transactions per second under the standard consumer model, so more shards give the consumer fleet more parallel partitions to divide the workload across, directly reducing the backlog behind a falling-behind consumer. Enabling enhanced fan-out changes how that per-shard throughput is shared: under the standard model, all consumers reading a shard share that shard's throughput, but enhanced fan-out gives each registered consumer its own dedicated throughput per shard using a push-based delivery mechanism, which removes the contention that occurs when multiple consumer applications compete for the same shard's limited read capacity. Together, more shards address the total capacity of the stream, while enhanced fan-out addresses how efficiently each individual consumer can pull from the capacity that already exists, which is why they complement rather than duplicate each other. Whenever a question describes a Kinesis consumer falling behind with rising iterator age and asks for two ways to fix it, expect the answer to combine a capacity-side change, like shard count, with a delivery-model change, like enhanced fan-out, rather than two variations of the same fix.