Added experimental support for MariaDB 11.7

.github/workflows/build.yml

@@ -19,7 +19,7 @@ jobs:
       BUNDLE_GEMFILE: ${{ matrix.gemfile }}
     services:
       mariadb:
-        image: quay.io/mariadb-foundation/mariadb-devel:11.6-vector-preview
+        image: mariadb:11.7-rc
         env:
           MARIADB_ALLOW_EMPTY_ROOT_PASSWORD: 1
           MARIADB_DATABASE: neighbor_test

CHANGELOG.md

@@ -1,3 +1,8 @@
+## 0.5.1 (unreleased)
+
+- Added experimental support for MariaDB 11.7
+- Dropped experimental support for MariaDB 11.6 Vector
+
 ## 0.5.0 (2024-10-07)
 
 - Added experimental support for SQLite (sqlite-vec)

README.md

@@ -6,7 +6,7 @@ Supports:
 
 - Postgres (cube and pgvector)
 - SQLite (sqlite-vec) - experimental
-- MariaDB 11.6 Vector - experimental
+- MariaDB 11.7 - experimental
 - MySQL 9 (searching requires HeatWave) - experimental
 
 [![Build Status](https://github.com/ankane/neighbor/actions/workflows/build.yml/badge.svg)](https://github.com/ankane/neighbor/actions)
@@ -61,10 +61,10 @@ class AddEmbeddingToItems < ActiveRecord::Migration[8.0]
     # cube
     add_column :items, :embedding, :cube
 
-    # pgvector and MySQL
+    # pgvector, MariaDB, and MySQL
     add_column :items, :embedding, :vector, limit: 3 # dimensions
 
-    # sqlite-vec and MariaDB
+    # sqlite-vec
     add_column :items, :embedding, :binary
   end
 end
@@ -387,14 +387,6 @@ Supported values are:
 - `cosine`
 - `hamming`
 
-For cosine distance with MariaDB, vectors must be normalized before being stored.
-
-```ruby
-class Item < ApplicationRecord
-  has_neighbors :embedding, normalize: true
-end
-```
-
 ### Indexing
 
 Vector columns must use `null: false` to add a vector index
@@ -403,7 +395,7 @@ Vector columns must use `null: false` to add a vector index
 class CreateItems < ActiveRecord::Migration[8.0]
   def change
     create_table :items do |t|
-      t.binary :embedding, null: false
+      t.vector :embedding, limit: 3, null: false
       t.index :embedding, type: :vector
     end
   end
@@ -892,7 +884,7 @@ bundle exec rake test:postgresql
 bundle exec rake test:sqlite
 
 # MariaDB
-docker run -e MARIADB_ALLOW_EMPTY_ROOT_PASSWORD=1 -e MARIADB_DATABASE=neighbor_test -p 3307:3306 quay.io/mariadb-foundation/mariadb-devel:11.6-vector-preview
+docker run -e MARIADB_ALLOW_EMPTY_ROOT_PASSWORD=1 -e MARIADB_DATABASE=neighbor_test -p 3307:3306 mariadb:11.7-rc
 bundle exec rake test:mariadb
 
 # MySQL

lib/neighbor/utils.rb

@@ -33,7 +33,7 @@ def self.validate(value, dimensions:, type:, adapter:)
     def self.normalize(value, column_info:)
       return nil if value.nil?
 
-      raise Error, "Normalize not supported for type" unless [:cube, :vector, :halfvec, :binary].include?(column_info&.type)
+      raise Error, "Normalize not supported for type" unless [:cube, :vector, :halfvec].include?(column_info&.type)
 
       norm = Math.sqrt(value.sum { |v| v * v })
 
@@ -86,10 +86,12 @@ def self.operator(adapter, column_type, distance)
         end
       when :mariadb
         case column_type
-        when :binary
+        when :vector
           case distance
-          when "euclidean", "cosine"
-            "VEC_DISTANCE"
+          when "euclidean"
+            "VEC_DISTANCE_EUCLIDEAN"
+          when "cosine"
+            "VEC_DISTANCE_COSINE"
           end
         when :integer
           case distance
@@ -168,7 +170,7 @@ def self.order(adapter, type, operator, quoted_attribute, query)
         if operator == "BIT_COUNT"
           "BIT_COUNT(#{quoted_attribute} ^ #{query})"
         else
-          "VEC_DISTANCE(#{quoted_attribute}, #{query})"
+          "#{operator}(#{quoted_attribute}, #{query})"
         end
       when :mysql
         if operator == "BIT_COUNT"
@@ -191,8 +193,6 @@ def self.normalize_required?(adapter, column_type)
       case adapter
       when :postgresql
         column_type == :cube
-      when :mariadb
-        true
       else
         false
       end

test/mariadb_test.rb

@@ -13,14 +13,6 @@ def test_cosine
     assert_elements_in_delta [0, 0.05719095841050148], result.map(&:neighbor_distance)
   end
 
-  def test_cosine_no_normalize
-    create_items(MariadbItem, :embedding)
-    error = assert_raises(Neighbor::Error) do
-      MariadbItem.find(1).nearest_neighbors(:embedding, distance: "cosine").first(3)
-    end
-    assert_equal "Set normalize for cosine distance with cube", error.message
-  end
-
   def test_euclidean
     create_items(MariadbItem, :embedding)
     result = MariadbItem.find(1).nearest_neighbors(:embedding, distance: "euclidean").first(3)
@@ -41,7 +33,7 @@ def test_create
 
   def test_vec_totext
     MariadbItem.create!(embedding: [1, 2, 3])
-    assert_equal "[1.000000,2.000000,3.000000]", MariadbItem.pluck("VEC_ToText(embedding)").last
+    assert_equal "[1,2,3]", MariadbItem.pluck("VEC_ToText(embedding)").last
   end
 
   def test_vec_fromtext

test/support/mariadb.rb

@@ -20,7 +20,7 @@ class MariadbRecord < ActiveRecord::Base
 
 MariadbRecord.connection.instance_eval do
   create_table :mariadb_items, force: true do |t|
-    t.binary :embedding, null: false
+    t.vector :embedding, limit: 3, null: false
     t.index :embedding, type: :vector
   end