OceanBase v4.3.5_CE_BP5: Enhanced Vector Search, Materialized Views, and Performance Optimizations

OceanBase v4.3.5_CE_BP5: Enhanced Vector Search, Materialized Views, and Performance Optimizations

OceanBase has released version v4.3.5_CE_BP5 (RPM: oceanbase-ce-4.3.5.5-105000012025111711) on November 19, 2025. This build introduces substantial upgrades to vector search, materialized view management, and query performance for complex joins and recursive Common Table Expressions (CTEs).

Upgrade Path and Compatibility

Administrators planning to upgrade should note the specific compatibility constraints:

  • Supported Path: The release supports upgrades from any V4.3.x version to V4.3.5 BP5.
  • Unsupported Path: Upgrades from V4.2.x or lower versions are not supported.
  • Migration Considerations: Since V4.3.2, the multi-source data persistence format was refactored. Upgrading requires conversion between old and new formats, which can be time-consuming in environments with a high number of partitions. Additionally, continuous write vector scenarios require index rebuilding.
  • Known Issue Resolved: The upgrade process for V4.3.0 through V4.3.2 previously caused transfer failures when tenant Leaders remained on old nodes. This is resolved in BP5.

Advanced Vector Search

The release significantly expands vector search capabilities, focusing on similarity scoring and index efficiency.

  • Similarity Functions: New similarity functions inner_product_similarity, cosine_similarity, and l2_similarity have been added. These return values in the [0,1] range to indicate similarity. Queries can now use parameters (similarity=x) in the LIMIT clause to filter results below a specific threshold.
  • IVF Index Optimization: Performance for IVF_FLAT and IVF_PQ indexes has been improved across four dimensions: build, query, DML, and usability. Key improvements include Kmeans parallelization for faster builds, an iterative filter algorithm for smarter query strategies, and caching mechanisms for incremental writes. Usability has been enhanced with monitoring tools and dynamic configuration of the nprobes parameter.

Materialized Views and Query Optimization

OceanBase has introduced a wide array of enhancements for materialized views and complex query execution.

  • Materialized View Features: New capabilities include support for incremental refresh of materialized views with Left Join, nested materialized views based on non-primary key incremental views, and the use of non-basic columns in MIN/MAX aggregations. The release also supports full refresh referencing User Defined Functions (UDFs) and a new Minimal refresh mode that reduces CLOG content.
  • Delete-Insert Optimization: For tables using the Delete-Insert pattern, the system now supports pre-generating Skip Indexes for incremental data. This allows query processing to prune unnecessary data blocks, significantly improving efficiency.
  • Cartesian Product Performance: Significant speedups have been achieved for Cartesian product calculations. Benchmarks show performance improvements of 15.8x for 1 join, 7.6x for 10 joins, 3.8x for 100 joins, and 1.2x for 1,000 joins.
  • Recursive CTEs (RCTE): The database now fully supports MySQL 8.0's Recursive Union Distinct, ensuring output uniqueness. The Recursive Union All functionality has also been strengthened to support data spilling to disk when memory is insufficient.

SQL, Compatibility, and Observability

Several changes improve MySQL compatibility and system observability.

  • MySQL Syntax: New functions cast as integer, cast as array, and split (equivalent to string_to_array) have been added. The CONTAINS function behavior is controlled by the function_contains_dialect config. Other additions include array<> syntax, JSON-to-text casting, and support for the into keyword in insert overwrite.
  • DBLink Timeout: A behavior change affects DBLink timeouts. Previously, a hard 120-second limit applied to nested SQLs (triggers/UDFs) if no hint was present. The new version uses the local session's ob_query_timeout instead of the hard limit.
  • Observability: The GV$OB_SQLSTAT view now includes columns for multi_query, multi_query_batch_execute, full_table_scan, and error_count. The generation of ASH reports has been accelerated.
  • Reliability: Log stream migration now selects only non-bad-disk replicas as sources.

Configuration and Bug Fixes

Administrators should review new configuration options and the extensive list of bug fixes.

  • New Configs: enable_mv_binlog_minimal_mode (for MV refresh optimization), default_skip_index_level (controls Skip Index scope), and function_contains_dialect (controls CONTAINS semantics).
  • System Variables: sql_transpiler is now available to expand simple UDF expressions for performance.
  • Bug Fixes: The release addresses a wide range of issues including memory leaks, transaction error codes, vector index pruning bugs, LOB handling, and specific fixes for recursive queries and partition management.