Reduce SQL queries needed to load open merge requestssh-optimize-merge-request-project-lookup

The SQL queries and memory allocation in MergeRequests::RefreshService
is dominated by queries for Project and Route loads. On staging, the
absence of an inverse relationship caused Rails to make over 1100
extraneous SQL queries for the www-gitlab-com repository.

Relates to https://gitlab.com/gitlab-org/gitlab-ce/issues/49703

5 files changed, 25 insertions, 2 deletions

diff --git a/app/models/project.rb b/app/models/project.rb
index d593cbb223a..106cb92e0cc 100644
--- a/app/models/project.rb
+++ b/app/models/project.rb
@@ -181,7 +181,7 @@ class Project < ActiveRecord::Base
   has_one :import_export_upload, dependent: :destroy # rubocop:disable Cop/ActiveRecordDependent
 
   # Merge Requests for target project should be removed with it
-  has_many :merge_requests, foreign_key: 'target_project_id'
+  has_many :merge_requests, foreign_key: 'target_project_id', inverse_of: :target_project
   has_many :source_of_merge_requests, foreign_key: 'source_project_id', class_name: 'MergeRequest'
   has_many :issues
   has_many :labels, class_name: 'ProjectLabel'
diff --git a/changelogs/unreleased/sh-optimize-merge-request-project-lookup.yml b/changelogs/unreleased/sh-optimize-merge-request-project-lookup.yml
new file mode 100644
index 00000000000..241b89c4633
--- /dev/null
+++ b/changelogs/unreleased/sh-optimize-merge-request-project-lookup.yml
@@ -0,0 +1,5 @@
+---
+title: Reduce SQL queries needed to load open merge requests
+merge_request: 22709
+author:
+type: performance
diff --git a/lib/gitlab/import/merge_request_helpers.rb b/lib/gitlab/import/merge_request_helpers.rb
index 97dc1a987c4..9215067d973 100644
--- a/lib/gitlab/import/merge_request_helpers.rb
+++ b/lib/gitlab/import/merge_request_helpers.rb
@@ -22,7 +22,7 @@ module Gitlab
           # additional work that is strictly necessary.
           merge_request_id = insert_and_return_id(attributes, project.merge_requests)
 
-          merge_request = project.merge_requests.find(merge_request_id)
+          merge_request = project.merge_requests.reload.find(merge_request_id)
 
           # We use .insert_and_return_id which effectively disables all callbacks.
           # Trigger iid logic here to make sure we track internal id values consistently.
diff --git a/spec/models/merge_request_spec.rb b/spec/models/merge_request_spec.rb
index c8943f2d86f..85a4ebac66c 100644
--- a/spec/models/merge_request_spec.rb
+++ b/spec/models/merge_request_spec.rb
@@ -13,6 +13,20 @@ describe MergeRequest do
     it { is_expected.to belong_to(:merge_user).class_name("User") }
     it { is_expected.to belong_to(:assignee) }
     it { is_expected.to have_many(:merge_request_diffs) }
+
+    context 'for forks' do
+      let!(:project) { create(:project) }
+      let!(:fork) { fork_project(project) }
+      let!(:merge_request) { create(:merge_request, target_project: project, source_project: fork) }
+
+      it 'does not load another project due to inverse relationship' do
+        expect(project.merge_requests.first.target_project.object_id).to eq(project.object_id)
+      end
+
+      it 'finds the associated merge request' do
+        expect(project.merge_requests.find(merge_request.id)).to eq(merge_request)
+      end
+    end
   end
 
   describe '#squash_in_progress?' do
diff --git a/spec/models/project_spec.rb b/spec/models/project_spec.rb
index e66838edd1a..83b3f308ec3 100644
--- a/spec/models/project_spec.rb
+++ b/spec/models/project_spec.rb
@@ -88,6 +88,10 @@ describe Project do
     it { is_expected.to have_many(:project_deploy_tokens) }
     it { is_expected.to have_many(:deploy_tokens).through(:project_deploy_tokens) }
 
+    it 'has an inverse relationship with merge requests' do
+      expect(described_class.reflect_on_association(:merge_requests).has_inverse?).to eq(:target_project)
+    end
+
     context 'after initialized' do
       it "has a project_feature" do
         expect(described_class.new.project_feature).to be_present