6.2 Release Notes


  • The size limit for primary keys has been increased to 320 bytes [1]

  • Dictionary encoding performance enhancements

  • Jobs can now be performed asynchronously

  • The Kinetica File System (KiFS) has been introduced to greatly ease ML workflows centered around file processing and file generation

  • Non-HA multi-head primary key lookup is now possible using the RecordRetriever object available in the Java and Python APIs

  • HA support for multi-head primary key lookup and multi-head ingest

  • /alter/table jobs are now cancellable

  • A Host Manager-controlled alerting system for application-level significant events and hardware resource usage. Alerts can be managed via gpudb.conf, and you can view the most recent alerts in Kinetica Administration Application (GAdmin)

  • Improvements to aggregation:

    • Aggregates listed in a HAVING expression no longer have to exist in the column name list
    • Aggregates can now be listed before grouping attributes in the column name list
    • Additional grouping functions (for both SQL and the native APIs):
  • Full Materialized View support via SQL or the native APIs--any number of source tables and intermediary tables & operations can be involved in creating a materialized view

  • Various improvements to the GAdmin user interface

  • Support for rank, partition, and window functionality in both SQL and the native APIs

  • The Python API has been updated to include an extension that enables increased speed when inserting and retrieving records from a GPUdbTable object

  • Joins can now be created using derived columns, e.g.,

      column_names=["a.x as ax","b.y as by","a.x+b.y as c"],
      expressions=["a.x = b.x"]
  • The /aggregate/statistics endpoint can now tune the behavior of the percentile() function using a second, comma-separated resolution value. The higher the resolution, the more accurate the estimation is but the longer the calculation takes, e.g., a 50th percentile resolution of 200:



  • Improved symbology scaling


  • Passwords now have a character limit of 1024, and user names and role names now have a character limit of 64


User-Defined Functions (UDFs)

  • Improved performance of non-distributed UDF read and write
  • Per-node concurrency limit setting available in GAdmin or the max_concurrency_per_node option in the /create/proc endpoint
  • Each UDF API has access to a status field under ProcData that helps convey status information during UDF execution


[1]Primary key max length increased from 160 to 320 in version