Amazon Kinesis Data Streams
- What it is
- Real-time streaming data ingestion. Captures gigabytes per second from thousands of sources and makes the stream available to consumers in milliseconds.
- Key traits
- You write the consumer application — Lambda, Kinesis Client Library, or Apache Flink.
- Data is retained (24 hours by default, up to 365 days), so multiple applications can read the same stream independently and you can replay it.
- Capacity modes: provisioned (you manage shards) or on-demand (scales automatically).
- Ordering is guaranteed within a shard.
- Not to be confused with
- Amazon Data Firehose — fully managed delivery to a destination, near real time (buffered, ~60 seconds), no consumer code, no replay.
- Amazon Simple Queue Service (SQS) — a queue: one consumer per message, no replay, no ordered multi-reader stream.
- Exam signal
- “Real-time, custom processing, multiple consumers, replay” → Data Streams.