Patch for load times with pg_type loading in AR (#14642)

* Patch for load times with pg_type loading in AR

- See #14615
- See rails/rails#19578

NEWS.md

@@ -9,6 +9,7 @@ Development
 
 ### Bug fixes / enhancements
 - Add base URL to lockout redirection in static pages ([#14617](https://github.com/CartoDB/cartodb/pull/14617))
+- Improve the in_database operations fixing some rails behaviors that were problematic for us ([#14642]https://github.com/CartoDB/cartodb/pull/14642)
 - Makes maps listing go faster with related tables (user db size cache issue, #14165)
 - Do not redirect to /login by default when error is unknown in network interceptor ([#14616](https://github.com/CartoDB/cartodb/pull/14616))
 - Update CARTO.js to v4.1.10

config/initializers/zz_patch_activerecord_type_loading.rb

@@ -0,0 +1,40 @@
+# Every time Activerecord makes a connection, it brings all the `pg_type`
+# data into a cache in order to do mapping between pg types and rails types
+#
+# What is the problem?
+#
+# Well in our users database we have 300K type. AR picks ~142K which leads to
+# spending ~1.3s picking the data and ~0.5s in processing those records so we end
+# spending about ~2s or more without a need in all our in_database operations
+#
+# How does this fix work?
+#
+# Well, we filter all the table types (table, analysis tables, overviews) because
+# in our app they aren't going to be used and in case they're needed, Rails is going
+# to query the database for it, so no problem.
+#
+# This patch was developed for rails 4.2.10, in case of update you need to review it
+# to verify if it needs to be modified or even removed.
+module ActiveRecord
+  module ConnectionAdapters
+    module PostgreSQL
+      module OID # :nodoc:
+        class TypeMapInitializer # :nodoc:
+          def query_conditions_for_initial_load(type_map)
+            known_type_names = type_map.keys.map { |n| "'#{n}'" }
+            known_type_types = %w('r' 'e' 'd')
+            <<-SQL % [known_type_names.join(", "), known_type_types.join(", ")]
+              LEFT JOIN pg_type as tt ON (tt.typtype = 'c' AND tt.typarray = t.oid AND tt.typinput = 'record_in(cstring,oid,integer)'::regprocedure)
+              WHERE
+                tt.oid is null
+                AND (t.typname IN (%s)
+                OR t.typtype IN (%s)
+                OR t.typinput = 'array_in(cstring,oid,integer)'::regprocedure
+                OR t.typelem != 0)
+            SQL
+          end
+        end
+      end
+    end
+  end
+end