specification: API Commons Rate Limits
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits
provider: AWS Kinesis
providerId: kinesis
created: '2026-05-04'
# Provenance stamped 2026-08-11: this artifact was written by the API Evangelist
# bulk sweep dated 2026-05-04, not harvested from the provider. See roadmap#35.
method: generated
modified: '2026-05-05'
reconciled: true
tags:
  - Rate Limiting
  - Streaming
  - AWS
  - Real-Time
description: Kinesis Data Streams enforces a mix of per-account/per-region control-plane API TPS limits
  and per-shard data-plane throughput limits (1 MB/s or 1000 records/s write per shard, 2 MB/s or 5 GetRecords/s
  read per shard). Limits are scoped per AWS account per region; many control-plane limits are adjustable
  via Service Quotas. Exceeding throughput returns ProvisionedThroughputExceededException.
sources:
  - https://docs.aws.amazon.com/streams/latest/dev/service-sizes-and-limits.html
  - https://aws.amazon.com/kinesis/data-streams/pricing/
responseCodes:
  throttled: 400
  provisionedThroughputExceeded: 400
  limitExceeded: 400
limits:
  - name: PutRecord (per shard)
    scope: shard
    metric: requests_per_second
    limit: 1000
    timeFrame: second
    notes: Each shard supports up to 1000 records/s and 1 MB/s write.
  - name: PutRecords (per shard)
    scope: shard
    metric: bytes_per_second
    limit: 1048576
    timeFrame: second
    notes: 500 records max per request; 10 MiB request limit; 10 MiB max per record (burst).
  - name: GetRecords (per shard)
    scope: shard
    metric: requests_per_second
    limit: 5
    timeFrame: second
    notes: Returns ProvisionedThroughputExceededException for 5 seconds after a 10 MB response.
  - name: GetRecords data read rate (per shard)
    scope: shard
    metric: bytes_per_second
    limit: 2097152
    timeFrame: second
  - name: SubscribeToShard (Enhanced Fan-Out)
    scope: consumer-shard
    metric: requests_per_second
    limit: 1
    timeFrame: second
    notes: One call per second per registered consumer per shard.
  - name: GetShardIterator
    scope: shard
    metric: requests_per_second
    limit: 5
    timeFrame: second
  - name: ListShards
    scope: stream
    metric: requests_per_second
    limit: 1000
    timeFrame: second
  - name: DescribeStream
    scope: account
    metric: requests_per_second
    limit: 10
    timeFrame: second
  - name: DescribeStreamSummary
    scope: account
    metric: requests_per_second
    limit: 20
    timeFrame: second
  - name: CreateStream / DeleteStream / ListStreams
    scope: account
    metric: requests_per_second
    limit: 5
    timeFrame: second
  - name: On-Demand Standard default throughput
    scope: stream
    metric: bytes_per_second
    limit: 4194304
    timeFrame: second
    notes: 4 MB/s write / 8 MB/s read default; auto-scales to 200 MB/s write or higher in select regions.
  - name: On-Demand max throughput (US-East-1, US-West-2, EU-West-1)
    scope: stream
    metric: bytes_per_second
    limit: 10737418240
    timeFrame: second
    notes: 10 GB/s write, 20 GB/s read with support-ticket increase.
  - name: Provisioned default shard quota
    scope: account/region
    metric: shards
    limit: 'see Service Quotas - defaults 20000 / 6000 / 1000 by region'
  - name: On-demand streams default
    scope: account
    metric: streams
    limit: 50
  - name: Stream retention
    scope: stream
    metric: hours
    limit: 8760
    notes: Default 24 hours; max 365 days (8760 hours).
  - name: Registered consumers per stream
    scope: stream
    metric: consumers
    limit: 50
    notes: 50 on On-Demand Advantage, 20 on Standard / Provisioned.
policies:
  - name: Provisioned throughput exceptions
    description: When per-shard write or read throughput is exceeded, the API returns ProvisionedThroughputExceededException;
      AWS SDKs implement automatic retry with exponential backoff.
  - name: Backoff and retry
    description: Use exponential backoff with jitter on throttling. KCL and KPL libraries handle this automatically
      for typical workloads.
  - name: Capacity-mode switching
    description: A stream can switch between On-Demand and Provisioned modes only twice in any 24-hour
      window.
  - name: Service Quotas adjustments
    description: Most control-plane and shard-count limits are adjustable via the AWS Service Quotas console;
      some (like UpdateShardCount) require an AWS Support case.
  - name: Resharding
    description: Use SplitShard / MergeShards (provisioned) or rely on Kinesis auto-scaling (on-demand)
      to keep per-shard limits ahead of producer/consumer load.
maintainers:
  - FN: Kin Lane
    email: kin@apievangelist.com