FIX: More resilient sentiment backfill query (#998)

lib/sentiment/post_classification.rb

@@ -5,11 +5,10 @@ module DiscourseAi
     class PostClassification
       def self.backfill_query(from_post_id: nil, max_age_days: nil)
         available_classifier_names =
-          DiscourseAi::Sentiment::SentimentSiteSettingJsonSchema
+          DiscourseAi::Sentiment::SentimentSiteSettingJsonSchema.values.map { _1.model_name }
-            .values
-            .map { |mc| mc.model_name.downcase }
-            .sort
 
+        queries =
+          available_classifier_names.map do |classifier_name|
             base_query =
               Post
                 .includes(:sentiment_classifications)
@@ -23,16 +22,12 @@ module DiscourseAi
                   ON crs.target_id = posts.id
                   AND crs.target_type = 'Post'
                   AND crs.classification_type = 'sentiment'
+                  AND crs.model_used = '#{classifier_name}'
               SQL
-            .group("posts.id")
+                .where("crs.id IS NULL")
-            .having(<<~SQL, available_classifier_names)
-              COUNT(crs.model_used) = 0
-              OR array_agg(
-                  DISTINCT LOWER(crs.model_used) ORDER BY LOWER(crs.model_used)
-                )::text[] IS DISTINCT FROM array[?]
-            SQL
 
-        base_query = base_query.where("posts.id >= ?", from_post_id.to_i) if from_post_id.present?
+            base_query =
+              base_query.where("posts.id >= ?", from_post_id.to_i) if from_post_id.present?
 
             if max_age_days.present?
               base_query =
@@ -44,6 +39,11 @@ module DiscourseAi
             base_query
           end
 
+        unioned_queries = queries.map(&:to_sql).join(" UNION ")
+
+        Post.from(Arel.sql("(#{unioned_queries}) as posts"))
+      end
+
       def bulk_classify!(relation)
         http_pool_size = 100
         pool =