Kinesis

Managed real-time streaming — the AWS Kafka alternative. Data Streams (ingest + fan-out) and Firehose (delivery) are the two that matter; Managed Apache Flink does stateful stream analytics on top.

flowchart LR
  P[Producers] --> DS[Data Streams]
  DS --> C[Consumers / KCL / Lambda]
  DS --> FH[Firehose]
  FH --> S3[(S3 / Redshift / OpenSearch / Splunk)]
  DS --> FL[Managed Apache Flink]

Data Streams

Ordered shards; ordering is per-shard by partition key. Records ≤ 1 MB, immutable once written, replayable. Retention 24 h default → 365 days. Many independent consumers per stream.

	Provisioned	On-Demand
Capacity	You pick shard count	Auto (starts 4 MB/s, tracks 30-day peak, up to 200 MB/s)
Per shard	1 MB/s or 1000 rec/s in; 2 MB/s out	—
Over limit	ProvisionedThroughputException	—
Pay	Shard-hour	Stream-hour + data
Use	Predictable	Spiky / unknown

Consumers

• Shared fan-out — 2 MB/s per shard shared by all consumers; 5 GetRecords/s per shard.
• Enhanced fan-out — dedicated 2 MB/s per consumer per shard, push over HTTP/2.
• KCL (consumer; checkpoints in DynamoDB), KPL (producer; batching/aggregation), or a Lambda event source mapping.

Firehose

Near-real-time delivery, zero code: S3, Redshift, OpenSearch, Splunk, HTTP endpoints, Datadog/New Relic. Buffers by size or time; optional Lambda transform, Parquet/ORC conversion, compression, failed-record backup to S3.

Managed Service for Apache Flink

Formerly Kinesis Data Analytics (SQL apps deprecated). Windowing, joins, stateful processing over streams.

Choosing

Need	Service
Custom processing, replay, many consumers	Data Streams
Zero-code load into S3/Redshift/OpenSearch	Firehose
Stateful analytics / windowing	Managed Apache Flink
Video ingest for ML	Kinesis Video Streams