Technology

StarRocks

An MPP analytical database with native lakehouse capabilities, able to directly query S3 data in Parquet, ORC, and Iceberg formats.

5 connections 3 resources

Summary

What it is

An MPP analytical database with native lakehouse capabilities, able to directly query S3 data in Parquet, ORC, and Iceberg formats.

Where it fits

StarRocks bridges pure lakehouse queries (Trino) and dedicated analytical databases (ClickHouse). It can query S3 data directly like Trino but also cache hot data locally for sub-second performance, making it the choice when you need low-latency analytics over lakehouse data.

Misconceptions / Traps

StarRocks' external table performance on S3 is comparable to Trino. The latency advantage comes from its local caching and materialized views — which require managing local storage.
Shared-data architecture on S3 is a newer feature. Evaluate maturity for your use case before production deployment.

Key Connections

depends_on Apache Parquet — reads Parquet files from S3
used_by Lakehouse Architecture — queries lakehouse data
constrained_by Cold Scan Latency — first-query performance limited by S3 access
scoped_to S3, Lakehouse

Definition

What it is

A massively parallel processing (MPP) analytical database with native lakehouse capabilities, able to directly query data on S3 in Parquet, ORC, and Iceberg formats.

Why it exists

Organizations want sub-second analytics without ETL. StarRocks can query S3 data directly (like Trino) but also ingest and cache hot data locally for faster performance, bridging the gap between pure lakehouse queries and dedicated analytical databases.

Primary use cases

Low-latency analytics over S3 lakehouse data, materialized views over Iceberg tables, real-time dashboards on S3-backed datasets.

Recent developments

Latest signals

StarRocks 4.1 GA (v4.1.0 Apr 13, v4.1.1 May 29, 2026) — range-based data distribution. The headline feature is multi-tenant data management via range-based data distribution with automatic tablet splitting and merging — oversized or hotspot tablets split without schema changes or re-ingestion, directly addressing data skew. Also ships Phase 1 large-capacity tablets with intra-tablet parallelism across ingest/PK-update/compaction, Fast Schema Evolution V2 (second-level DDL), and a Beta inverted index for text filtering. Per StarRocks 4.1 release notes.
Active 4.0.x patch cadence — a release every ~3 weeks. The 4.0 line shipped 4.0.6 (Feb 16), 4.0.7 (Mar 12), 4.0.8 (Mar 25), 4.0.9 (Apr 16), 4.0.10 (May 9, 2026) — a reliable patch rhythm for production teams wanting cumulative fixes without major-version risk. Per the StarRocks GitHub releases.
Coinbase and Pinterest adopted StarRocks for lakehouse-native analytics. Coinbase reports StarRocks materially outran ClickHouse on TPC-H 1TB, where ClickHouse OOM-failed 12 of 22 join-heavy queries; Coinbase runs 10 blockchains / 300+ tables / 573B rows, keeping hot data (2 weeks–3 months) in StarRocks native on S3 and federating cold history from Iceberg/Delta. Per Rill Data.
First audited TPC-H result at 10 TB scale. Alibaba Cloud EMR Serverless StarRocks 3.3 posted an official TPC-H 10000GB benchmark: 7,546,131 QphH at a price/performance of 1,280.88 CNY per kQphH. Per the TPC-H result detail.
EMR Serverless StarRocks v1.21 (Mar 2, 2026) — Compaction Service in Beta. The Alibaba Cloud managed tier added Compaction Service (Beta), monitoring-metric delivery to SLS (Log Service), and materialized-view alerts — pushing StarRocks toward a turnkey, observable "S3 lakehouse query engine" competing with managed Trino and ClickHouse Cloud. Per Alibaba Cloud release notes.