Allow cross-db transaction for Build#drop_with_exit_code!

When dropping a build with this method, it wraps the execution with
a transaction.

If this build;
- Configured as auto-retry
- Has a pipeline related to an MR
- Has a TODO with that MR
Then Ci::RetryBuildService;
- Retries the job
- Closes the TODO

And those two updates should not be in a single transaction because
they refer to different DBs.

The next step is to remove this cross-db transaction.

app/models/ci/build.rb

@@ -1051,9 +1051,11 @@ module Ci
     end
 
     def drop_with_exit_code!(failure_reason, exit_code)
-      transaction do
-        conditionally_allow_failure!(exit_code)
-        drop!(failure_reason)
+      ::Gitlab::Database::QueryAnalyzers::PreventCrossDatabaseModification.allow_cross_database_modification_within_transaction(url: 'https://gitlab.com/gitlab-org/gitlab/-/issues/348495') do
+        transaction do
+          conditionally_allow_failure!(exit_code)
+          drop!(failure_reason)
+        end
       end
     end
 

spec/models/ci/build_spec.rb

@@ -5223,6 +5223,26 @@ RSpec.describe Ci::Build do
         it_behaves_like 'drops the build without changing allow_failure'
       end
     end
+
+    context 'when build is configured to be retried' do
+      let(:options) { { retry: 3 } }
+
+      context 'when there is an MR attached to the pipeline and a failed job todo for that MR' do
+        let!(:merge_request) { create(:merge_request, source_project: project, author: user, head_pipeline: pipeline) }
+        let!(:todo) { create(:todo, :build_failed, user: user, project: project, author: user, target: merge_request) }
+
+        before do
+          build.update!(user: user)
+          project.add_developer(user)
+        end
+
+        it 'resolves the todo for the old failed build' do
+          expect do
+            drop_with_exit_code
+          end.to change { todo.reload.state }.from('pending').to('done')
+        end
+      end
+    end
   end
 
   describe '#exit_codes_defined?' do

spec/services/ci/retry_build_service_spec.rb

@@ -188,13 +188,6 @@ RSpec.describe Ci::RetryBuildService do
         expect(new_build).to be_pending
       end
 
-      it 'resolves todos for old build that failed' do
-        expect(::MergeRequests::AddTodoWhenBuildFailsService)
-          .to receive_message_chain(:new, :close)
-
-        service.execute(build)
-      end
-
       context 'when there are subsequent processables that are skipped' do
         let!(:subsequent_build) do
           create(:ci_build, :skipped, stage_idx: 2,
@@ -272,6 +265,17 @@ RSpec.describe Ci::RetryBuildService do
           expect(bridge.reload).to be_pending
         end
       end
+
+      context 'when there is a failed job todo for the MR' do
+        let!(:merge_request) { create(:merge_request, source_project: project, author: user, head_pipeline: pipeline) }
+        let!(:todo) { create(:todo, :build_failed, user: user, project: project, author: user, target: merge_request) }
+
+        it 'resolves the todo for the old failed build' do
+          expect do
+            service.execute(build)
+          end.to change { todo.reload.state }.from('pending').to('done')
+        end
+      end
     end
 
     context 'when user does not have ability to execute build' do