OceanBase v4.6.0: Native AI Hybrid Search and HTAP Enhancements

by

OceanBase v4.6.0_CE represents a significant evolution of the open-source relational database, shifting focus toward AI-native data processing and high-performance HTAP (Hybrid Transactional/Analytical Processing). Released on July 17, 2026, this version consolidates vector database capabilities directly into the core SQL engine and refines HTAP routing logic.

The standout feature of this release is the introduction of a native SQL hybrid search interface. Unlike previous iterations that relied on specific system packages, v4.6.0 allows users to perform multi-modal queries—combining vectors, full-text, and scalar filters—using standard SQL syntax via HYBRID_SEARCH(TABLE table_name, DSL_STRING). This enables “RAG all in One” knowledge base solutions where users can perform complex Boolean combinations and weighted ranking directly within the database.

To support these search capabilities, the release introduces several indexing enhancements:

  • Search Index (Universal Inverted Index): A new index type for JSON, Array, and Map fields that supports path-based filtering, solving performance issues in schema-less scenarios.
  • Match Phrase: A new full-text index type that maintains token position arrays, enabling exact phrase matching with configurable slop (fuzziness) for knowledge base retrieval.
  • Vector Index Optimizations: The HNSW index now uses a segmented architecture (Active/Frozen/Incr/Base) to control memory usage during high-write scenarios, preventing memory bloat. Additionally, the system supports SINDI sparse vectors and IVF indexes on non-primary key tables.

HTAP and Column Store Upgrades

HTAP performance in v4.6.0 is improved through smarter column store routing and consistency models. The database now supports strong consistency reads on column store replicas. Previously, column stores typically accepted weak consistency reads for performance; this release allows the optimizer to route AP queries to column stores while enforcing GTS-based strong consistency when required.

Furthermore, the parallel execution framework has been overhauled:

  • PX Queuing: A new real-load-based parallel execution queue replaces the static parallel_servers_target configuration, dynamically adjusting parallelism to prevent CPU exhaustion.
  • Adaptive Task Splitting: The system now detects “long tail” granules during parallel scans and splits them further to balance load across threads.

Benchmarks cited in the release notes indicate substantial gains: follower read performance in high-concurrency scenarios improved by 14x, and single-thread memory allocation performance increased by 168% via a new lock-free allocator based on mimalloc.

Compatibility and Ecosystem

OceanBase v4.6.0 deepens its integration with the open data ecosystem:

  • Iceberg V3: Full support for Iceberg V3 format, including Time Travel queries (via snapshot ID or timestamp) and Deletion Vector support.
  • REST Catalog: A new REST Catalog protocol allows OceanBase to query Iceberg tables via standard HTTP APIs, simplifying data lake integration.
  • MySQL Compatibility: Expanded support for Recursive CTEs, Partition Exchange, and Session-level temporary tables to smooth migration paths.

Storage and Maintenance Updates

Operational efficiency is addressed through new storage models and maintenance tools:

  • Append-Only Tables: A new table model that prohibits updates and deletes, suitable for immutable logs and financial records.
  • Compaction TTL: An asynchronous data lifecycle management feature that deletes expired rows during major compaction without consuming write bandwidth.
  • Window Compaction: A new maintenance mode that limits major compaction to specific time windows, preventing resource contention during business hours.

Security is also enhanced with the addition of caching_sha2_password authentication support.