Highlights of Db2 12.1.5

Vector indexes for high-performance AI search

Vector indexes extend the vector capabilities of Db2 12.1 with native, index-based acceleration for approximate similarity search on high-dimensional vectors. Building on the vector data type introduced in Db2® 12.1.2, this enhancement dramatically improves query performance for AI and machine learning workloads that operate on large-scale embedding collections. Vector indexes use a graph-based Approximate Nearest Neighbor (ANN) algorithm inspired by DiskANN, that enables millisecond-latency similarity searches across millions or billions of vectors. This feature makes Db2 ideal for semantic search, retrieval-augmented generation (RAG), recommendation systems, and other interactive AI applications. For more information, see Vector Indexes.

RPM package installation for Db2 on Linux distributions

You can now install Db2 database products natively on Linux systems by using Red Hat Package Manager (RPM) package management. RPM packages are available for all Db2 products and include all components with English interface support. You can install additional language support by installing the corresponding language RPM packages.

RPM installation provides automated dependency resolution and integrates with standard Linux package managers (rpm, yum, dnf, zypper). All RPM packages are digitally signed by using IBM Garasign GPG signing for security verification.

RPM package installation is available on the following Linux distributions:

• Red Hat Enterprise Linux® (RHEL) 9.x.
• Red Hat Enterprise Linux (RHEL) 10.x.
• SUSE Linux Enterprise Server (SLES) 15.x.
• Other RPM-based Linux distributions.

For more information, see Installing Db2 using RPM packages (Linux).

Index integrity validation

Changes to the Unicode Collation Algorithm (UCA) ordering during upgrade or update can cause existing indexes to become improperly ordered. This is especially important when upgrading to Db2 12.1, where newer UCA versions are introduced and older ones may be deprecated. When choosing to override the upgrade invalidation of indexes and MQTs by setting db2set DB2_DB_UPGRADE=SKIP_DOWNLEVEL_ICU:TRUE, it is up to the customer to validate their own indexes and MQTs and, if necessary, recreate them after upgrade.

To detect such a need, we have introduced the INSPECT INDEXORDER feature, which allows you to validate whether index entries are stored in the correct sort order. This helps identify corrupted or disordered indexes early—before they lead to query failures or data integrity issues. It can also be used more broadly as part of regular index maintenance.

For more information, see INSPECT command.

Db2 Text Search modernization

We're excited to announce a major evolution in Db2 Text Search in Db2 12.1.5 release, introducing a pluggable architecture that delivers unprecedented flexibility and scalability. Enterprise systems contain valuable text data, such as product descriptions and customer feedback, but traditional SQL queries struggle to search that data effectively. Organizations need modern search capability that scales independently from the database, deploys anywhere, and works with existing applications without requiring rewrites. This release supports two popular search engines: Elasticsearch and OpenSearch.

For more information, see Db2 Text Search

Improved metrics returned by MON_GET_TABLE table function

Four new columns are now added to the MON_GET_TABLE procedure. By including these columns in FROM TABLE queries of the MON_GET_TABLE procedure, Db2 provides you with comprehensive table status information, alongside existing monitoring metrics.

For more information, see MON_GET_TABLE table function.

New Database Manager enhancements for setting maximum heap size used by Datalake Scheduler and Datalake Metastore.

New enhancements for replacing Overlay IP addresses on Db2 HADR deployments on AWS

Pacemaker automation is now enhanced to direct Amazon Route 53 DNS records for a specified domain name after an HADR takeover. This feature replaces the need to manually configure the Pacemaker resource model to use Overlay IP addresses.

For more information, see Configuring Route 53 DNS-based traffic redirection for a Pacemaker-managed two-node Db2 HADR cluster on AWS.

Enhanced REORGCHK F2 formula for range-partitioned tables

The REORGCHK command returns table or index statistics that show whether reorganization can improve usage, disk space, and performance. The REORGCHK F2 formula now provides more accurate table partition reorganization recommendations for range-partitioned tables.

For more information, see SYSCAT.DATAPARTITIONS.

New tiebreaker policy creation capability for HADR clusters that use a quorum device

This release introduces tiebreaker policy support for HADR configurations that use a quorum device (QDevice). This support is specific to Db2 instances running on Pacemaker-managed Linux clusters. Users can now configure a tiebreaker policy by running a db2cm -set -option tiebreakerPolicy command, to ensure the node running the HADR PRIMARY database is favored during split-brain scenarios. In previous releases of Db2 this capability required manual steps to be made on each cluster node.

For more information, see db2cm command reference for HADR-configured clusters.

New disk tiebreaker available for Db2 pureScale on Pacemaker-managed Linux clusters

For Pacemaker-managed Linux cluster deployments, Db2 pureScale currently uses a QDevice quorum as a tiebreaker mechanism when a network split (split-brain scenario) breaks a cluster into two even pieces. This unfortunately is a step back in functionality from IBM® Tivoli® System Automation for Multiplatforms (SA MP) for AIX platforms, which allowed the use of a tiebreaker disk for the same functionality. In a Pacemaker-managed cluster, the customer needs to provide an extra host to be used as a QDevice, while SA MP only requires an extra disk.

Db2 12.1.5 adds a disk tiebreaker to Db2 pureScale® deployments running on Pacemaker-managed Red Hat Enterprise Linux (RHEL) clusters. This feature provides significant hardware savings to users and makes upgrading to Db2 12.1 from SA MP-managed Db2 11.x deployments much easier.

Enhancement to how row-organized tables with adaptive compression are evaluated by REORGCHK command

Enhancements to how the Db2 RUNSTATS command calculates table size now provide improved performance of REORGCHK operations.

Datalake tables

The Datalake table feature in Db2 12.1.5 enables efficient processing of large structured and semi-structured datasets stored in remote object storage (like AWS S3 or IBM COS) using open table formats such as Hive and Iceberg. The Datalake table feature leverages the massively parallel processing capability of Db2 for analytics, big data, and data warehousing integration.

A database administrator must first create a storage access alias before anyone can create or use Datalake tables. The alias holds the remote storage connection details and credentials, stored securely in an encrypted keystore. It may also define a default path for where tables are placed. Access to the alias is privilege-based and can be granted to users, groups, or database roles.

Large Language Model (LLM) integration

API key authentication support is now extended to OpenAI-compatible providers in Db2. While previously available for only watsonx.ai-hosted models, this new support provides secure access to a broader range of external AI services for vector embedding generation directly within Db2 workflows.

For more information, see Large Language Model (LLM) integration with Db2.

Enhanced AI Query Optimizer

AI join cardinality estimation has been extended to equality join predicates on string columns when MODEL_JOIN_CARD_STRING is enabled. This enhancement improves the performance of the AI Query Optimizer. For more information, see Predicate selectivity estimation prediction.

Added support for Amazon Simple Storage Service (S3) on AIX systems

Support for using Amazon Simple Storage Service (S3) compatible remote storage on AIX systems is added. This enhancement extends Db2 remote storage capabilities beyond Linux platforms, enabling AIX deployments to access data stored in S3‑based object storage for supported Db2 commands. On AIX, S3 remote storage is disabled by default and must be enabled. In addition, this feature requires libcurl version 8.14.1 or later to be installed on the system. For more information, see Remote storage requirements.