OceanBase 4.4.2 LTS: HTAP Optimizations, Heterogeneous Zones, and OBCDC Improvements

by

OceanBase Database v4.4.2_CE is now available as a Long Term Support (LTS) release. Positioned as a “TP/AP fusion” version, it integrates transactional capabilities from the 4.2.5 LTS branch with analytical capabilities from the 4.3.5 LTS branch. The update aims to provide a more robust foundation for critical transactional workloads and complex analytical queries.

Heterogeneous Zones and Scalability

A significant architectural shift in this release is the support for heterogeneous zones. OceanBase 4.0 traditionally enforced homogeneous zones, which simplified deployment but limited flexibility during scaling or node failures. The new heterogeneous architecture allows a tenant to configure different UNIT NUM values across zones. This means a single node failure no longer requires shrinking all zones, reducing resource waste and operational costs. It also ensures that traffic rebalancing remains smooth during failures.

Transactional and Replication Enhancements

In the transactional (TP) domain, the release introduces Session-level private temporary tables, restoring functionality that was disabled in earlier versions. For data replication and synchronization, OBCDC (OceanBase Change Data Capture) has been improved:

  • Table-level recovery sync: OBCDC can now identify and sync tables recovered via the offline DDL process by tracking them in the __all_ddl_operation table.
  • Virtual generated columns: The CDC tool now supports synchronizing data for virtual generated columns, addressing a gap where these columns were previously invisible to downstream consumers.
  • Partition management: Users can now exchange partitions between secondary partition tables and primary partition tables, simplifying data reorganization.

Key-Value and Time-Series Optimizations

The release continues to evolve the OceanBase Key-Value (OBKV) capabilities:

  • Weak read and locality: OBKV now supports weak read (allowing reads from non-leader replicas) and proximity read (prioritizing data from the nearest data center).
  • TTL with index scanning: Time-To-Live (TTL) data cleanup can now use indexes to reduce full-table scans, improving performance for large datasets.
  • HBase monitoring: Monitoring granularity has been enhanced, distinguishing between HBase get and scan operations, as well as individual put and delete operations within batch calls.
  • Time-series model: A new put operation has been added for time-series models to reduce contention and write latency caused by primary key conflict checks.

Performance Benchmarks

OceanBase reports targeted performance improvements across various hardware configurations:

  • Serial table creation: Optimizations in schema generation and writing have resulted in a ~60% performance increase for serial table creation compared to V4.3.5.
  • Large-scale CPUs: On 256-core domestic CPUs, Parallel DML (PDML) performance has improved by approximately 25%.
  • Replication queries: Query performance on follower replicas for replication tables has improved by 14x, narrowing the gap with leader queries (approaching 30w QPS).
  • Parallel Query (PX): An adaptive task splitting mechanism has been introduced to automatically redistribute long-tail tasks, improving load balancing.

Observability and Operations

Diagnostic capabilities have been strengthened to help DBAs identify bottlenecks:

  • ASH Data Integrity: The Active Session History (ASH) buffer has been enhanced with a weight column to prevent data loss during queue backlogs.
  • SQLSTAT Refactoring: Statistics collection has been moved out of the library cache to a tenant hashmap to prevent cache exhaustion and ensure consistency across replicas.
  • Real-time statistics: New metrics for micro-block access and filter operations have been added to monitoring views.

Compatibility and Upgrade Path

Upgrading to v4.4.2 is supported from V4.2.5 BP7 and V4.4.1 via a smooth upgrade path. However, upgrades from V4.3.x versions (4.3.0 through 4.3.5 BP2) are only recommended for POC and testing environments due to potential risks.

Key compatibility notes:

  • Clusters using vector capabilities cannot be upgraded to this version.
  • Several system views have been added (e.g., for LOB consistency checks) and modified (e.g., V$ACTIVE_SESSION_HISTORY).
  • SQL syntax has been extended to support CAST AS INTEGER/ARRAY, SPLIT, and recursive union distinct.

Tooling and Dependencies

Version 4.4.2 recommends specific companion tool versions, including ODP 4.3.6, OCP 4.4.1, OBD 4.2.0, and OBCDC 4.4.2.