Native Apache Kafka Service Is Coming Soon to StreamNative Cloud. Join the waitlist and get $1,000 in credits.
Join Waitlist >
Lakestream: The Lakehouse-Native Streaming Architecture
Streams as first-class lakehouse primitives alongside tables — no connectors, no ETL, one governed data plane.
Data Layer — Ursa
Diskless, leaderless, lakehouse-native. Data lives on open data lakes.
Metadata Layer — Catalog
Extend your existing lakehouse catalog to govern streams alongside tables.
Protocol Layer — Stateless Serving
Scale Kafka, Pulsar, REST, gRPC independently from storage.
From Streams to Lakestreams
The next paradigm in data infrastructure. Streams become first-class lakehouse primitives — no connectors, no ETL, no copies.
• ARCHITECTURE
Three Layers, One Unified Architecture
Lakestream pushes interoperability down to the storage and catalog layers — three independent layers, each scaling on its own, all sharing the same governed data.
• VISION
Why Streaming and Lakehouse Are Converging
Enterprises operate two parallel infrastructures: streaming for events, lakehouse for analytics. Bridging them requires connectors, ETL, and schema sync — adding latency, doubling storage costs, and fragmenting metadata. Lakestream dissolves this divide by treating streams as first-class lakehouse primitives alongside tables.
Stream-Table Duality, Built in
Every stream is simultaneously a table — with offset monotonicity, eventual visibility, and schema consistency guarantees built in.
Zero-ETL by Design
Data is written once in open formats on object storage. Analytics and ML query the same data that streaming apps produce — no second copy, no connectors.
Streaming and Lakehouse Converge
Rather than building better connectors, Lakestream pushes interoperability down to the storage and catalog layers, eliminating the connector entirely.
• DEEP DIVE
Inside the Lakestream Architecture
Each layer of Lakestream is designed to scale independently while sharing the same governed data plane.
Ursa — Cloud-Native Storage Engine
Cloud-native stream storage that is leaderless and diskless. A distributed Write-Ahead Log (WAL) delivers low-latency streaming while data is compacted to Parquet on object storage for analytics.
WAL for sub-millisecond write latency
Leaderless, diskless — no broker disks, no cross-AZ replication
Lakestream Catalog — Governed Streaming
A three-level namespace (catalog.namespace.stream) governs all streams and tables from a single catalog. Integrates with Unity Catalog, Snowflake Horizon, AWS S3 Tables, and Apache Polaris.
Three-level namespace for streams and tables
Unity Catalog, Snowflake Horizon, S3 Tables, Polaris
Stateless Serving — Multi-Protocol Access
Stateless protocol servers decouple compute from storage. Kafka, Pulsar, REST, and gRPC are choices of interface, not choices of data silo. Add or remove protocol servers like web servers.
Kafka, Pulsar, REST, gRPC — all stateless
No leader elections, no partition rebalancing
• KEY BENEFITS
Why Lakestream Changes Everything
Up to 95% Cost Reduction
Object storage eliminates cross-AZ replication. At 5 GB/s throughput, implementations show up to 95% cost savings versus traditional broker-based Kafka.
Leaderless & Diskless
No leader elections, no partition rebalancing. Stateless protocol servers scale horizontally — add or remove like web servers.
Open Formats
Data lives in Parquet, Iceberg, and Delta Lake on your object storage. No proprietary segment format, full query engine portability.
Stream-Table Duality
Every stream is simultaneously a table with built-in consistency guarantees: offset monotonicity, eventual visibility, and schema consistency.
Multi-Protocol, Single Data
Kafka, Pulsar, REST, and gRPC serve the same streams simultaneously. The protocol is a choice of interface, not a choice of data silo.
Up to 95% Cost Reduction
Object storage eliminates cross-AZ replication. At 5 GB/s throughput, implementations show up to 95% cost savings versus traditional broker-based Kafka.
Leaderless & Diskless
No leader elections, no partition rebalancing. Stateless protocol servers scale horizontally — add or remove like web servers.
Open Formats
Data lives in Parquet, Iceberg, and Delta Lake on your object storage. No proprietary segment format, full query engine portability.
Stream-Table Duality
Every stream is simultaneously a table with built-in consistency guarantees: offset monotonicity, eventual visibility, and schema consistency.
Multi-Protocol, Single Data
Kafka, Pulsar, REST, and gRPC serve the same streams simultaneously. The protocol is a choice of interface, not a choice of data silo.
• PROOF POINTS
Lakestream in Practice
Lakestream is not a concept — it is shipping today. Ursa and UFK prove that lakehouse-native streaming delivers on its promise.
Ursa
Cloud-Native Storage Engine
The first proof point of Lakestream. Ursa is a cloud-native storage engine with a WAL + Parquet two-tier model on object storage — the foundational data layer that makes lakehouse-native streaming possible.
VIEW DETAILS
Ursa For Kafka (UFK)
Native Apache Kafka on Lakestream
Native Apache Kafka — not compatible, native. UFK takes Apache Kafka 4.0+ and replaces local disk storage with Lakestream, delivering leaderless, diskless Kafka with up to 95% cost reduction.
VIEW DETAILSUrsa Awarded Best Industry Paper
Read our award-winning paper on Ursa, the first lakehouse-native, leaderless streaming engine for Kafka.
READ MORE
Experience Lakestream
- Spin up a Kafka or Pulsar cluster on StreamNative in minutes.
- Every topic is a lakehouse table — no connectors, no ETL.
- Up to 95% cost reduction with leaderless, diskless architecture.