Split backfill query to avoid cross joins

db/post_migrate/20210818185845_backfill_projects_with_coverage.rb

@@ -6,6 +6,7 @@ class BackfillProjectsWithCoverage < ActiveRecord::Migration[6.1]
   MIGRATION = 'BackfillProjectsWithCoverage'
   DELAY_INTERVAL = 2.minutes
   BATCH_SIZE = 10_000
+  SUB_BATCH_SIZE = 100
 
   disable_ddl_transaction!
 
@@ -20,7 +21,8 @@ class BackfillProjectsWithCoverage < ActiveRecord::Migration[6.1]
       CiDailyBuildGroupReportResult,
       MIGRATION,
       DELAY_INTERVAL,
-      batch_size: BATCH_SIZE
+      batch_size: BATCH_SIZE,
+      other_job_arguments: [SUB_BATCH_SIZE]
     )
   end
 

lib/gitlab/background_migration/backfill_projects_with_coverage.rb

 # frozen_string_literal: true
 
-# Backfill project_ci_feature_usages for a range of projects with coverage
-class Gitlab::BackgroundMigration::BackfillProjectsWithCoverage
-  COVERAGE_ENUM_VALUE = 1
+module Gitlab
+  module BackgroundMigration
+    # Backfill project_ci_feature_usages for a range of projects with coverage
+    class BackfillProjectsWithCoverage
+      COVERAGE_ENUM_VALUE = 1
+      INSERT_DELAY_SECONDS = 0.1
 
-  def perform(start_id, end_id)
-    ActiveRecord::Base.connection.execute <<~SQL
-      INSERT INTO project_ci_feature_usages (project_id, feature, default_branch)
-        SELECT DISTINCT project_id, #{COVERAGE_ENUM_VALUE} as feature, default_branch
-        FROM ci_daily_build_group_report_results
-        WHERE id BETWEEN #{start_id} AND #{end_id}
-      ON CONFLICT (project_id, feature, default_branch) DO NOTHING;
-    SQL
+      def perform(start_id, end_id, sub_batch_size)
+        report_results = ActiveRecord::Base.connection.execute <<~SQL
+          SELECT DISTINCT project_id, default_branch
+          FROM ci_daily_build_group_report_results
+          WHERE id BETWEEN #{start_id} AND #{end_id}
+        SQL
+
+        report_results.to_a.in_groups_of(sub_batch_size, false) do |batch|
+          ActiveRecord::Base.connection.execute <<~SQL
+            INSERT INTO project_ci_feature_usages (project_id, feature, default_branch) VALUES
+            #{build_values(batch)}
+            ON CONFLICT (project_id, feature, default_branch) DO NOTHING;
+          SQL
+
+          sleep INSERT_DELAY_SECONDS
+        end
+      end
+
+      private
+
+      def build_values(batch)
+        batch.map do |data|
+          "(#{data['project_id']}, #{COVERAGE_ENUM_VALUE}, #{data['default_branch']})"
+        end.join(', ')
+      end
+    end
   end
 end

spec/lib/gitlab/background_migration/backfill_projects_with_coverage_spec.rb

@@ -66,10 +66,12 @@ RSpec.describe Gitlab::BackgroundMigration::BackfillProjectsWithCoverage, schema
         default_branch: false,
         group_id: group.id
       )
+
+      stub_const("#{described_class}::INSERT_DELAY_SECONDS", 0)
     end
 
     it 'creates entries per project and default_branch combination in the given range', :aggregate_failures do
-      subject.perform(1, 4)
+      subject.perform(1, 4, 2)
 
       entries = project_ci_feature_usages.order('project_id ASC, default_branch DESC')
 
@@ -81,11 +83,11 @@ RSpec.describe Gitlab::BackgroundMigration::BackfillProjectsWithCoverage, schema
 
     context 'when an entry for the project and default branch combination already exists' do
       before do
-        subject.perform(1, 4)
+        subject.perform(1, 4, 2)
       end
 
       it 'does not create a new entry' do
-        expect { subject.perform(1, 4) }.not_to change { project_ci_feature_usages.count }
+        expect { subject.perform(1, 4, 2) }.not_to change { project_ci_feature_usages.count }
       end
     end
   end

spec/migrations/20210818185845_backfill_projects_with_coverage_spec.rb

@@ -17,6 +17,7 @@ RSpec.describe BackfillProjectsWithCoverage do
   describe '#up' do
     before do
       stub_const("#{described_class}::BATCH_SIZE", 2)
+      stub_const("#{described_class}::SUB_BATCH_SIZE", 1)
 
       ci_daily_build_group_report_results.create!(
         id: 1,
@@ -60,8 +61,8 @@ RSpec.describe BackfillProjectsWithCoverage do
         freeze_time do
           migrate!
 
-          expect(described_class::MIGRATION).to be_scheduled_delayed_migration(2.minutes, 1, 2)
-          expect(described_class::MIGRATION).to be_scheduled_delayed_migration(4.minutes, 3, 3)
+          expect(described_class::MIGRATION).to be_scheduled_delayed_migration(2.minutes, 1, 2, 1)
+          expect(described_class::MIGRATION).to be_scheduled_delayed_migration(4.minutes, 3, 3, 1)
           expect(BackgroundMigrationWorker.jobs.size).to eq(2)
         end
       end