Skip LFS updates in mirrors if repository has not changed

The GetAllLFSPointers RPC is expensive and was being called for every
mirror. If the repository has not changed, we can assume that the LFS
pointers also did not change. We can skip calling this RPC if there are
no updates.

Since we don't actually get a return code if branches or tags were
updated, we use the repository checksum as a cheaper mechanism (which
XORs the ref contents) to determine whether we run the LFS scan.

Closes https://gitlab.com/gitlab-org/gitlab/issues/38256

changelogs/unreleased/sh-skip-lfs-updates-mirroring.yml

+---
+title: Skip updating LFS objects in mirror updates if repository has not changed
+merge_request: 21744
+author:
+type: performance

ee/app/services/projects/update_mirror_service.rb

@@ -24,12 +24,17 @@ module Projects
         return error("The mirror user is not allowed to push code to all branches on this project.")
       end
 
+      checksum_before = project.repository.checksum
+
       update_tags do
         project.fetch_mirror(forced: true)
       end
 
       update_branches
-      update_lfs_objects
+
+      # Updating LFS objects is expensive since it requires scanning for blobs with pointers.
+      # Let's skip this if the repository hasn't changed.
+      update_lfs_objects if project.repository.checksum != checksum_before
 
       success
     rescue Gitlab::Shell::Error, Gitlab::Git::BaseError, UpdateError => e

ee/spec/services/projects/update_mirror_service_spec.rb

@@ -325,60 +325,74 @@ describe Projects::UpdateMirrorService do
       end
     end
 
-    context 'updating Lfs objects' do
-      before do
-        stub_fetch_mirror(project)
-      end
+    context 'updating LFS objects' do
+      context 'when repository does not change' do
+        before do
+          allow(project).to receive(:lfs_enabled?).and_return(true)
+        end
 
-      context 'when Lfs is disabled in the project' do
-        it 'does not update Lfs objects' do
-          allow(project).to receive(:lfs_enabled?).and_return(false)
-          expect(Projects::LfsPointers::LfsObjectDownloadListService).not_to receive(:new)
+        it 'does not attempt to update LFS objects' do
+          expect(Projects::LfsPointers::LfsImportService).not_to receive(:new)
 
           service.execute
         end
       end
 
-      context 'when Lfs is enabled in the project' do
+      context 'when repository changes' do
         before do
-          allow(project).to receive(:lfs_enabled?).and_return(true)
+          stub_fetch_mirror(project)
         end
 
-        it 'updates Lfs objects' do
-          expect(Projects::LfsPointers::LfsImportService).to receive(:new).and_call_original
-          expect_any_instance_of(Projects::LfsPointers::LfsObjectDownloadListService).to receive(:execute).and_return({})
+        context 'when Lfs is disabled in the project' do
+          it 'does not update LFS objects' do
+            allow(project).to receive(:lfs_enabled?).and_return(false)
+            expect(Projects::LfsPointers::LfsObjectDownloadListService).not_to receive(:new)
 
-          service.execute
+            service.execute
+          end
         end
 
-        context 'when Lfs import fails' do
-          let(:error_message) { 'error_message' }
-
+        context 'when Lfs is enabled in the project' do
           before do
-            expect_any_instance_of(Projects::LfsPointers::LfsImportService).to receive(:execute).and_return(status: :error, message: error_message)
+            allow(project).to receive(:lfs_enabled?).and_return(true)
           end
 
-          # Uncomment once https://gitlab.com/gitlab-org/gitlab-foss/issues/61834 is closed
-          # it 'fails mirror operation' do
-          #   expect_any_instance_of(Projects::LfsPointers::LfsImportService).to receive(:execute).and_return(status: :error, message: 'error message')
+          it 'updates LFS objects' do
+            expect(Projects::LfsPointers::LfsImportService).to receive(:new).and_call_original
+            expect_any_instance_of(Projects::LfsPointers::LfsObjectDownloadListService).to receive(:execute).and_return({})
 
-          #   result = subject.execute
+            service.execute
+          end
+
+          context 'when Lfs import fails' do
+            let(:error_message) { 'error_message' }
 
-          #   expect(result[:status]).to eq :error
-          #   expect(result[:message]).to eq 'error message'
-          # end
+            before do
+              expect_any_instance_of(Projects::LfsPointers::LfsImportService).to receive(:execute).and_return(status: :error, message: error_message)
+            end
 
-          # Remove once https://gitlab.com/gitlab-org/gitlab-foss/issues/61834 is closed
-          it 'does not fail mirror operation' do
-            result = subject.execute
+            # Uncomment once https://gitlab.com/gitlab-org/gitlab-foss/issues/61834 is closed
+            # it 'fails mirror operation' do
+            #   expect_any_instance_of(Projects::LfsPointers::LfsImportService).to receive(:execute).and_return(status: :error, message: 'error message')
 
-            expect(result[:status]).to eq :success
-          end
+            #   result = subject.execute
 
-          it 'logs the error' do
-            expect_any_instance_of(Gitlab::UpdateMirrorServiceJsonLogger).to receive(:error).with(hash_including(error_message: error_message))
+            #   expect(result[:status]).to eq :error
+            #   expect(result[:message]).to eq 'error message'
+            # end
 
-            subject.execute
+            # Remove once https://gitlab.com/gitlab-org/gitlab-foss/issues/61834 is closed
+            it 'does not fail mirror operation' do
+              result = subject.execute
+
+              expect(result[:status]).to eq :success
+            end
+
+            it 'logs the error' do
+              expect_any_instance_of(Gitlab::UpdateMirrorServiceJsonLogger).to receive(:error).with(hash_including(error_message: error_message))
+
+              subject.execute
+            end
           end
         end
       end