huginn_datatable 0.4.1 → 0.4.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 93c77acb3179b06e77712420af3deb4c8f595a715c45e3c4f00c03a5f4e8a73c
-  data.tar.gz: a0dba33d0fffd18d0df9303d9253ec9652ab5620d6e77fbc0b017a3c8000b335
+  metadata.gz: 69da28c529c40a8b48b25dec0b0d2dbae158b9b070b4744ffd880c7600d5b915
+  data.tar.gz: 15fb1354d48685d12e1e8f0c5f61f1ec93a9bf8053292a8fff527e9287af6a07
 SHA512:
-  metadata.gz: 56699ffb963d5c809dc3664fbbc7456f39a634cd6952ad9472b8eb0f9f891016291baa859203a32d3024e7b496adc003c2f55ace6a384ce691c54b6adb37c715
-  data.tar.gz: 2437df93a69e3328ae63b6be1fbb623c7a5784e0232d0bb1fa757a74698f9bad211da135459f27b5a879d9f0cbca02388372fa169f0acb7e6c947ed0693b9805
+  metadata.gz: 7cb1bfff379be1ee47ab9fd593a6fe974a90bb7a1bef70c0bae7d9b9b39583f96ac16ddbf0c43c27ddf7e10462eb541d6bd4aca35741c01ffd1132403dfcf5b3
+  data.tar.gz: 16b019669e29974017f5c8a77d33036d5f8bee2d20449024d061bb695316cc9747631ec44e1921a13ad97044109cbcbe4182368d5f0df255ae94b2ca9620ded8

data/CHANGELOG.md

@@ -2,6 +2,10 @@
 
 ### Unreleased
 
+#### 0.4.2
+
+- **Performance (PostgreSQL):** association matches in `search` are written as `fk = ANY(ARRAY(SELECT DISTINCT ...))` instead of `fk IN (SELECT DISTINCT ...)`. The `IN` form was planned as a hashed SubPlan, which can't take part in a BitmapOr: a single association column (e.g. `searchable_columns :title, company: [:name]`) made the whole OR fall back to filtering every row, ignoring the GIN trigram/tsvector indexes. `ANY(ARRAY(...))` runs once as an InitPlan and keeps the index plan (measured on 20k-30k rows: ~500 ms → ~1 ms for a selective term). Results are unchanged; other adapters keep `IN`.
+
 #### 0.4.1
 
 - **Fix:** the specs no longer use the Ruby 3.1+ shorthand hash syntax (`f(allowed_paths:)`), so the suite loads again on Ruby 3.0 (the gem's minimum). No library code changed from 0.4.0.

data/lib/huginn/searchable/query.rb

@@ -108,6 +108,7 @@ module Huginn
       #   belongs_to -> base[fk]  IN (SELECT DISTINCT t1[join_pk]  FROM t1 ...)
       #   has_many   -> base[pk]   IN (SELECT DISTINCT t1[fk]      FROM t1 ...)
       #
+      # On PostgreSQL the membership is written as `= ANY(ARRAY(...))` (see #member_of).
       # Walks the chain from the first target table outwards so the base table
       # is never scanned inside the subquery.
       def anchored_predicate(path, attrs)
@@ -139,7 +140,20 @@ module Huginn
         manager.project(projection)
         manager.distinct
 
-        outer.in(manager)
+        member_of(outer, manager)
+      end
+
+      # `column IN (subquery)` becomes a hashed SubPlan on PostgreSQL, which can't
+      # take part in a BitmapOr: one association arm then forces the whole search
+      # (every column OR'ed together) into a filter over every row of the table,
+      # ignoring the GIN indexes. `column = ANY(ARRAY(subquery))` is evaluated
+      # once as an InitPlan and the column's index is used, so the OR stays a
+      # BitmapOr. Other adapters keep the plain IN.
+      def member_of(column, subquery)
+        return column.in(subquery) unless postgresql?
+
+        array = Arel::Nodes::NamedFunction.new("ARRAY", [subquery.ast])
+        Arel::Nodes::Equality.new(column, Arel::Nodes::NamedFunction.new("ANY", [array]))
       end
 
       # Whether the first hop goes through a join model; the FK anchoring is
@@ -161,7 +175,7 @@ module Huginn
                         .distinct
 
         predicate = fuzzy_predicates(attrs).reduce(&:or)
-        arel_table[primary_key].in(base.joins(spec).where(predicate).arel)
+        member_of(arel_table[primary_key], base.joins(spec).where(predicate).arel)
       end
 
       def fuzzy_predicate(attr)

data/lib/huginn/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Huginn
-  VERSION = "0.4.1"
+  VERSION = "0.4.2"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: huginn_datatable
 version: !ruby/object:Gem::Version
-  version: 0.4.1
+  version: 0.4.2
 platform: ruby
 authors:
 - Kayky Marcelo