Skip to content

v5.3.0

Latest
Compare
Choose a tag to compare
@MadSchemas MadSchemas released this 27 May 07:36

Core

  • [fea] Added random sorting via hash()/hash(seed) functions in sort expressions
  • [fea] Added support for exponential numbers in SQL parser
  • [fea] Allow to switch index type (from array to scalar and from scalar to array) for empty namespaces via UpdateIndex interface

Vector indexes

  • [fea] Added optional built-in auto-embedding cache with hybrid (RAM+disk) structure
  • [fea] Optimize single modification requests for HNSW indexes with enabled multithreading option. Now they have the same performance as HNSW-indexes with disabled multithreading
  • [fix] Fixed AVX512 implementation of L2 distance calculation for some specific dimensions count

Replication

  • [fix] Fixed possible hanging/timeout in set_leader_node config action

Sharding

  • [fix] Fixed incorrect sorting by expressions in distributed queries

Reindexer server

  • [fix] Fixed race in connections Listener during server termination

Reindexer tool

  • [fea] Added multithreading into dump restoration process. This allows to restore dumps with multiple namespaces 2.5-3 times faster if network is good enough
  • [fea] Added optional transactions into dump restoration process (--txsize option). This allows to speedup dumps restoration on local hosts a little bit more

Go connector

  • [fea] Improved parsing for ttl-index tags. Now expire_after=xxx option may be combined with any other options (like dense or is_no_column)
  • [fix] Fixed events stream drop on idle connections

Build

  • [fea] Added uninstall target for make

Face

  • [fea] Added default configs requests from the backend
  • [fea] Added new fields for auto-embedding cache to Statistics -> Memory
  • [fix] Fixed items view on using query with like
  • [fix] Fixed Vector values disappearing issue that appeared on items deleting
  • [fix] Fixed result value view on using query with distinct