StarRocks

StarRocks is a next-generation MPP analytical database licensed under Apache 2.0 with over 12,000 GitHub stars, governed under the Linux Foundation LF Projects Incubating program, and deployed by Microsoft, Pinterest, Shopee, Shein, Trip.com, DiDi, Tencent, Lenovo, Expedia, miHoYo, Fanatics, Demandbase, FLYR, Innogames, and Naver for sub-second analytics on petabyte-scale data. The core execution engine uses a fully pipelined query processor that eliminates intermediate materialization between operators, combined with a cost-based optimizer that automatically reorders multi-table joins and selects between hash join, shuffle join, and broadcast join based on table statistics and data distribution. Storage employs a columnar format with three index types: prefix index for sorted prefix columns, bitmap index for low-cardinality string filtering, and bloom filter index for high-cardinality lookups. Global dictionary encoding compresses string columns and accelerates aggregation by performing integer operations instead of string operations. The primary key table model supports real-time UPSERT and DELETE with second-level data freshness through Routine Load from Kafka and Stream Load from application SDKs, without impacting concurrent query performance. StarRocks 3.3 introduced shared-data architecture that separates compute and storage, enabling stateless BE nodes that scale independently from object storage backends on AWS S3, Google GCS, and Azure Blob. The external catalog feature queries Apache Iceberg, Hudi, Delta Lake, and Hive tables directly without data import, accelerated by a multi-tier data cache that automatically populates hot data from remote storage. Materialized views support automatic query rewrite for pre-computed aggregations. On the SSB Flat 100GB benchmark, StarRocks achieves 2 to 5x higher throughput than ClickHouse and Apache Druid. Shopee reported 3x query performance improvement after migrating from Apache Hive.