Merge branch 'jmd-add-praefect-dataloss-accept-e2e' into 'master'

Add praefect accept-dataloss e2e

See merge request gitlab-org/gitlab!80574

qa/qa/service/praefect_manager.rb

@@ -327,6 +327,22 @@ module QA
         end
       end
 
+      def accept_dataloss_for_project(project_id, authoritative_storage)
+        repository_hash = "#{Digest::SHA256.hexdigest(project_id.to_s)}"
+        repository = "@hashed/#{repository_hash[0, 2]}/#{repository_hash[2, 2]}/#{repository_hash}.git"
+
+        cmd = %{
+          docker exec #{@praefect} \
+          praefect \
+            -config /var/opt/gitlab/praefect/config.toml \
+            accept-dataloss \
+            --virtual-storage=default \
+            --repository=#{repository} \
+            --authoritative-storage=#{authoritative_storage}
+        }
+        shell(cmd)
+      end
+
       def wait_for_health_check_all_nodes
         wait_for_gitaly_health_check(@primary_node)
         wait_for_gitaly_health_check(@secondary_node)
@@ -415,6 +431,27 @@ module QA
         Support::Waiter.wait_until(sleep_interval: 1) { replication_queue_incomplete_count == 0 && replicated?(project_id) }
       end
 
+      def wait_for_replication_to_node(project_id, node)
+        Support::Waiter.wait_until(sleep_interval: 1) do
+          result = []
+          shell sql_to_docker_exec_cmd(%{
+            select * from replication_queue
+            where state = 'ready'
+              and job ->> 'change' = 'update'
+              and job ->> 'target_node_storage' = '#{node}'
+              and job ->> 'relative_path' = '#{Digest::SHA256.hexdigest(project_id.to_s)}.git';
+          }) do |line|
+            result << line.strip
+            QA::Runtime::Logger.debug(line.strip)
+          end
+          # The result should look like this when all items are replicated
+          #    id | state | created_at | updated_at | attempt | lock_id | job | meta
+          #   ----+-------+------------+------------+---------+---------+-----+------
+          #   (0 rows)
+          result[2] == '(0 rows)'
+        end
+      end
+
       def replication_pending?
         result = []
         shell sql_to_docker_exec_cmd(

qa/qa/specs/features/api/3_create/gitaly/praefect_dataloss_spec.rb

@@ -52,6 +52,53 @@ module QA
           expect(project_data_loss).to include('gitaly3, assigned host, unhealthy')
         end
       end
+
+      it 'allows admin resolve scenario where data cannot be recovered', testcase: 'https://gitlab.com/gitlab-org/gitlab/-/quality/test_cases/352708' do
+        # Ensure everything is in sync before begining test
+        praefect_manager.wait_for_project_synced_across_all_storages(project.id)
+
+        Resource::Repository::Commit.fabricate_via_api! do |commit|
+          commit.project = project
+          commit.commit_message = 'accept-dataloss-1'
+          commit.add_files([
+            { file_path: "new_file-#{SecureRandom.hex(8)}.txt", content: 'Add a commit to gitaly1,gitaly2,gitaly3' }
+          ])
+        end
+
+        praefect_manager.wait_for_replication_to_node(project.id, praefect_manager.primary_node)
+        praefect_manager.stop_primary_node
+        Resource::Repository::Commit.fabricate_via_api! do |commit|
+          commit.project = project
+          commit.commit_message = 'accept-dataloss-2'
+          commit.add_files([
+            { file_path: "new_file-#{SecureRandom.hex(8)}.txt", content: 'Add a commit to gitaly2,gitaly3' }
+          ])
+        end
+
+        praefect_manager.wait_for_replication_to_node(project.id, praefect_manager.secondary_node)
+        praefect_manager.stop_secondary_node
+        Resource::Repository::Commit.fabricate_via_api! do |commit|
+          commit.project = project
+          commit.commit_message = 'accept-dataloss-3'
+          commit.add_files([
+            { file_path: "new_file-#{SecureRandom.hex(8)}.txt", content: 'Add a commit to gitaly3' }
+          ])
+        end
+
+        # Confirms that they want to accept dataloss, using gitaly2 as authoritative storage to use as a base
+        praefect_manager.accept_dataloss_for_project(project.id, praefect_manager.secondary_node)
+
+        # Restart nodes, and allow replication to apply dataloss changes
+        praefect_manager.start_all_nodes
+        praefect_manager.wait_for_project_synced_across_all_storages(project.id)
+
+        # Validate that gitaly2 was accepted as the authorative storage
+        aggregate_failures "validate correct set of commits available" do
+          expect(project.commits.map { |commit| commit[:message].chomp }).to include('accept-dataloss-1')
+          expect(project.commits.map { |commit| commit[:message].chomp }).to include('accept-dataloss-2')
+          expect(project.commits.map { |commit| commit[:message].chomp }).not_to include('accept-dataloss-3')
+        end
+      end
     end
   end
 end