Amazon Kinesis Data Streams

Use it when

• Your stack is already on AWS and you want durable streaming with IAM, Lambda, Firehose, and Managed Flink wired in natively.
• You want capacity to be a billing decision, not an operational one: on-demand mode splits shards automatically as traffic grows.
• Multiple applications need to replay the same stream independently, with retention configurable from 24 hours to 365 days.
• Nobody on the team wants to run brokers; AWS runs everything and shards are the only knob.

Think twice when

• You want the Kafka protocol or its tooling; Kinesis speaks its own API, so clients, Connect pipelines, and skills do not transfer (Amazon MSK is the AWS answer for that).
• Your partition keys are uneven: a single hot key is capped at 1 MB/s and 1,000 records per second per shard, and on-demand mode cannot absorb that.
• You may leave AWS later; the workload and its tooling are tied to the platform.

How it runs

Fully managed and serverless: create a stream, choose provisioned shards or on-demand capacity, and pay per shard hour and payload (or per GB on-demand). There is no connector framework; Firehose, Lambda, and Managed Service for Apache Flink fill that role. Cross-region replication is your own consumers' job.