SOFTWARE / SYSTEMS / AIEngineering news. Technical depth.
NEWS / Data · 2 MIN READ

Kinesis adds service-managed partition keys for on-demand streams

AWS adds opt-in record distribution without producer-supplied partition keys for Kinesis On-Demand Standard and On-Demand Advantage streams.

Announcement: · From AWS

Availability and supported stream modes

AWS announced service-managed partition keys for Amazon Kinesis Data Streams on September 23, 2026. The opt-in feature lets producers publish records without specifying partition keys on On-Demand Standard and On-Demand Advantage streams. AWS says the feature is available in all AWS commercial regions, at no additional charge for the feature. AWS announcement

AWS positions the capability for workloads that do not require record ordering, including log aggregation, metrics collection and IoT telemetry. AWS announcement

Why record distribution matters

According to AWS, customers using random partition keys, such as UUIDs, can still encounter uneven throughput across shards. That imbalance can cause throttling for some partition keys even when the stream has sufficient aggregate capacity. With service-managed partition keys, AWS says Kinesis instead distributes records across shards based on available warm capacity. AWS announcement

For engineering teams, the meaningful change is moving record-distribution responsibility out of producer code. Generating a random identifier and placing traffic according to available capacity are different tasks; the feature addresses the latter rather than merely offering another way to generate keys.

The capacity qualifier matters. AWS describes distribution against available warm capacity, so teams should not interpret the announcement as a promise of unlimited throughput or a reason to skip capacity planning.

What to check before adopting

AWS directs customers to upgrade to the latest AWS SDK or Kinesis Producer Library version to send records without specifying partition keys. AWS announcement

Before opting in, check whether downstream processing depends on ordering associated with existing keys. Workloads that treat records independently are the clearest fit with AWS’s stated use case. An application that relies on ordered events needs a design review, not simply the deletion of a partition-key argument.

For rollout, verify client compatibility and monitor throttling and throughput under representative traffic. Treat the change as a shift in distribution responsibility, not as evidence that every workload will use capacity more efficiently.

SOURCES & CONTEXT

See the original announcement for availability and release details.