StarRocks in the Modern Data Stack: Architecture Evolution, Typical Applications, and Performance Insights

The presentation begins with an overview of the MPP (Massively Parallel Processing) architecture evolution, illustrating how data management progressed from early GB‑scale offline reporting to modern TB/PB‑scale interactive analytics, real‑time processing, and the convergence of data warehouses and data lakes.

Typical industry applications are introduced, highlighting four major StarRocks scenarios: ad‑hoc analysis, real‑time analytics, user‑profile building, and fixed‑report OLAP, each leveraging StarRocks' vectorized query engine, CBO optimizer, and federated lake‑warehouse capabilities.

Core features of StarRocks—ultra‑fast queries, flexible modeling, and real‑time data ingestion—are summarized, followed by performance test results that show StarRocks outperforming ClickHouse in both single‑table and multi‑table join queries.

Three practical approaches for building a real‑time data warehouse are detailed: (1) micro‑batch scheduling using Flink‑CDC‑StarRocks and external schedulers; (2) incremental construction and aggregation where Flink and Kafka handle most processing before loading into StarRocks; and (3) StarRocks view‑based solution that supports upsert/delete operations for low‑latency analytics.

The CDP (Customer Data Platform) use case is explored, describing how RoaringBitmap and global dictionary features enable efficient user segmentation, ID‑mapping via Hive, and behavior analysis with retention and funnel functions.

Lakehouse analytics are discussed, noting StarRocks' support for external tables on Apache Hive, Hudi, and Iceberg, and its transparent acceleration that automatically moves hot data to local SSD while keeping cold data in external storage.

A short‑term roadmap outlines upcoming features such as Resource Group fine‑grained isolation, Partial Update for reducing wide tables, multi‑table materialized views (MTMV) for both offline and real‑time scenarios, and a SaaS offering for BYOC deployments.

The Q&A section confirms that StarRocks already supports Iceberg v1 and will soon support v2, with external table performance roughly three times faster than internal queries, and notes that stored procedures are not currently available.