7.2 Release Notes

Overview

Kinetica 7.2.3 delivers a major step forward in performance, resilience, and developer productivity. This release introduces on-disk column compression, enterprise-grade backup orchestration, expanded support for graph querying through PGQ/GQL, and native Kafka integration for high availability.

On-Disk Column Compression

• Improved storage efficiency with column compression algorithms.
• Reduces on-disk footprint and accelerates query performance, especially for large analytic workloads.
• Transparent to applications--no schema or query changes required.

SQL-Orchestrated Full & Incremental Hot Backup

• New SQL-orchestrated commands to manage both full and incremental backups.
• Supports hot backups, allowing users to back up active systems without downtime.
• Enables fine-grained scheduling and recovery workflows directly from SQL, simplifying administration.

PGQ/GQL Graph Query Language Support

• Added Kinetica SQL/PGQ.
• Support for Property Graph Queries (PGQ) and Cypher-compatible Graph Query Language (GQL).
• Unlocks advanced graph analytics directly within Kinetica’s GPU-accelerated engine.
• Supports use cases such as fraud detection, network analysis, supply chain optimization, and entity resolution.

Native Kafka Support for HA Data Replication

• Built-in integration with Apache Kafka to replicate data streams across clusters for high availability (HA).
• Enables continuous, fault-tolerant replication between primary and standby sites.
• Improves recovery objectives (RPO/RTO) for mission-critical workloads.

SQL/OLAP

• HNSW index support for vector similarity search

Infrastructure

• Parallelized group-by output handling
• Added support for DICT property on unrestricted-length string columns to enable dictionary encoding on them

UI

• Single sign-on (SSO) for Reveal, Workbench, and GAdmin with OAuth/OIDC

  Note

  Requires KAgent 7.2.2.0+ for configuration

SQL/OLAP

• Added support for remote models and table functions to generate embeddings using those models

• New UDF is included to automatically perform RAG across multiple SQL-GPT contexts to choose the relevant contexts and tables when generating SQL

• Added KI_HINT_RECURSIVE_EXPLAIN to show operations on materialized view base tables and recursively do the same for base tables that themselves are materialized views in any explain plan. This supports getting an explain plan for a materialized view by executing:

  EXPLAIN /* KI_HINT_RECURSIVE_EXPLAIN */ SELECT * FROM my_mview
  

• Added new system-wide, directory, and UDF CREATE permission, as well as CREATE permission for all object types within schemas

• Added KI_HINT_NO_SHADOW_CUBE to disable use of the Shadow Cube for caching queries within worker processes

• Added support for the following scoped hints:

  • KI_HINT_MATERIALIZE
  • KI_HINT_GROUPBY_PK
  • KI_HINT_NO_VIRTUAL_UNION

  Examples:

  SELECT /*+ KI_HINT_MATERIALIZE */ *
  FROM iter t1 , iter t2
  WHERE t1.i < 100 AND t2.i < 100
  
  SELECT /*+ KI_HINT_GROUP_BY_PK / i, COUNT()
  FROM iter
  GROUP BY 1
  
  SELECT /*+ KI_HINT_NO_VIRTUAL_UNION */ *
  FROM
  (
      SELECT i FROM iter WHERE i < 100
      UNION ALL
      SELECT i FROM iter WHERE i > 1000 AND i < 2000
  )
  

• Added support for KI_HINT_JOB_USER_DATA(user_text) hint for user to provide the query tag that will appear in the Jobs list

Infrastructure

• Reduced memory usage in head rank for unsharded group-by operations
• Optimized processing for very high cardinality group-by operations
• Improved tiered storage flexibility by allowing foreign key indexes to evict from RAM

UI

• Added support in Workbench for viewing query status and metrics after a query is executed in a SQL block.