OceanBase V5.0.1: Parallel DDL, AI Image Support, and Database-Level Resource Isolation

by

Overview

OceanBase Database V5.0.1, released on August 6, 2026, is a continuation of the V4.6.0 series. The update focuses on DDL efficiency, storage flexibility, and AI integration, alongside significant performance optimizations for ARM and high-concurrency environments.

DDL and Storage Enhancements

OceanBase has expanded its Data Definition Language (DDL) capabilities to reduce downtime and improve manageability:

  • Parallel DDL: The update introduces parallel execution for DDL operations. Users can now run ALTER TABLE ADD COLUMN and DROP COLUMN across multiple tables simultaneously, as well as parallel partition operations like ADD PARTITION and TRUNCATE PARTITION. Note: Parallel DDL is currently a hidden feature and requires validation in a test environment before production use.
  • Storage Format Conversion: The database now supports online conversion between row-oriented, column-oriented, and mixed storage formats. Users can switch formats using the delayed keyword, allowing schema changes without blocking data writes.
  • Random Distribution Tables: A new PARTITION BY RANDOM syntax allows data to be automatically distributed evenly across cluster nodes. This eliminates the need for manual partitioning and helps avoid hotspots in high-throughput, keyless wide-table scenarios.
  • Table Mode Switching: Users can now switch between partial_update, delete_insert, and append_only modes online, facilitating schema evolution without full table rebuilds.

Performance Optimizations

V5.0.1 delivers measurable performance gains across various hardware and workload types:

  • ARM and Parallel Optimization: The update adds ARM NEON SIMD vectorization for integer operations and string comparisons. In ARM environments, core operator performance improves by 20–40%, bringing TPC-H performance closer to x86 levels. For medium-scale parallelism (DOP 2000–6000), AP query throughput improves by 15–30%.
  • Parquet External Table Scanning: Parquet Page-level prefetching reduces IO amplification by 40–60%, and high-selectivity queries see end-to-end latency drops of over 30%.
  • SQL Execution Improvements: The system now optimizes HashDistinct with adaptive state machines, improves Window Function vectorization, and enhances aggregation pushdown. Foreign key GTS (Global Transaction Snapshot) optimization reduces RPC overhead, boosting write throughput by 15–25% in scenarios with foreign keys.

AI and Full-Text Search

OceanBase is integrating AI capabilities directly into the database engine:

  • AI Function Image Support: The AI_EMBED, AI_PROMPT, and AI_COMPLETE functions now support image data (Base64 or URL). Users can generate vectors from images, which are stored and searched alongside text vectors. Concurrency is controlled via min_concurrency and max_concurrency parameters.
  • IK Tokenizer: The built-in IK tokenizer now supports loading custom dictionaries, improving the accuracy of full-text indexes.

Resource Management and Isolation

To support multi-tenant SaaS scenarios, OceanBase introduces granular resource controls:

  • Database-Level Resource Isolation: A new feature allows mapping individual databases to independent resource groups, enabling CPU and IO isolation at the database level. This requires OBProxy V4.4.1 and is available for tenants with a minimum spec of 16C64G.
  • Zone-Level Table Groups: New SCOPE attributes allow table groups to concentrate partitions within a specific Zone, improving leader distribution and performance control.

Compatibility and Infrastructure

The release broadens platform support and compatibility:

  • LoongArch64 Support: OceanBase V5.0.1 adds native support for Loongson processors (3A6000 and above), targeting domestic innovation scenarios.
  • Cgroup V2: The database now supports Linux Cgroup V2, automatically detecting and using the appropriate interface (requires kernel ≥ 5.10).
  • MySQL 8.0 Compatibility: Setting ob_compatibility_control = 'MYSQL8.0' changes default collation behavior for utf8mb4 to utf8mb4_0900_ai_ci when collation is not specified.

Security and Reliability

Enterprise-grade reliability features have been added:

  • Double Password Management: The ALTER USER ... RETAIN CURRENT PASSWORD command allows new passwords to take effect while retaining the old one during a rolling upgrade. This prevents application disconnections caused by bulk password changes.
  • Enhanced Audit: Audit logs now support encryption, and filters have been refined to command and object levels.

Breaking Changes and Configuration

Several syntax changes and configuration updates are required:

  • New Functions: window_funnel() for funnel analysis, DBMS_MVIEW.REFRESH_REPORT() for refresh diagnostics, and DBMS_STATS.GATHER_CATALOG_TABLE_STATS() for external table statistics.
  • Configuration: New parameters include mv_refresh_concurrency (max refresh tasks) and ai_function_max_concurrency. The schema_history_expire_time is now tenant-level.
  • Views: New system views such as V$OB_KEYWORDS and DBA_MVIEW_DEPS provide deeper visibility into system state.

Limitations and Upgrade Notes

Users should be aware of the following limitations:

  • Shared Storage: V5.0.1 does not support shared storage mode.
  • OBProxy Dependency: Database-level resource isolation is dependent on OBProxy V4.4.1.
  • Upgrade Paths: Upgrades from V4.2.x or lower require an intermediate step to V4.3.x. Shared storage mode cannot be upgraded to V5.0.1.