Merge branch 'ml-update-stale-test' into 'master'

Update stale E2E test - automatic replication

See merge request gitlab-org/gitlab!54776

qa/qa/service/praefect_manager.rb

@@ -353,15 +353,47 @@ module QA
         Support::Waiter.wait_until(sleep_interval: 1) { replication_queue_incomplete_count == 0 && replicated?(project_id) }
       end
 
+      def replication_pending?
+        result = []
+        shell sql_to_docker_exec_cmd(
+          <<~SQL
+            select job from replication_queue
+            where state = 'ready'
+              and job ->> 'change' = 'update'
+              and job ->> 'source_node_storage' = '#{current_primary_node}'
+              and job ->> 'target_node_storage' = '#{@primary_node}';
+          SQL
+        ) do |line|
+          result << line
+        end
+
+        # The result looks like:
+        #
+        #  job
+        #  -----------
+        #   {"change": "update", "params": null, "relative_path": "@hashed/4b/22/4b227777d4dd1fc61c6f884f48641d02b4d121d3fd328cb08b5531fcacdabf8a.git", "virtual_storage": "default", "source_node_storage": "gitaly3", "target_node_storage": "gitaly1"}
+        #  (1 row)
+        #  <blank row>
+        #
+        # Therefore when replication is pending there is at least 1 row of data plus 4 rows of metadata/layout
+
+        result.size >= 5
+      end
+
       private
 
       def current_primary_node
-        shell dataloss_command do |line|
-          QA::Runtime::Logger.debug(line.chomp)
-
-          match = line.match(/Primary: (.*)/)
-          break match[1] if match
+        result = []
+        shell sql_to_docker_exec_cmd("select node_name from shard_primaries where shard_name = '#{@virtual_storage}';") do |line|
+          result << line
         end
+        # The result looks like:
+        #  node_name
+        #  -----------
+        #   gitaly1
+        #  (1 row)
+
+        result[2].strip
       end
 
       def dataloss_command

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

@@ -3,7 +3,7 @@
 module QA
   RSpec.describe 'Create' do
     context 'Gitaly' do
-      describe 'Backend node recovery', :orchestrated, :gitaly_cluster, :skip_live_env, quarantine: { issue: 'https://gitlab.com/gitlab-org/gitlab/-/issues/238186', type: :investigating } do
+      describe 'Backend node recovery', :orchestrated, :gitaly_cluster, :skip_live_env do
         let(:praefect_manager) { Service::PraefectManager.new }
         let(:project) do
           Resource::Project.fabricate! do |project|
@@ -50,18 +50,17 @@ module QA
             push.file_content = 'new file'
           end
 
+          # Confirm that the commit is waiting to be replicated
+          expect(praefect_manager).to be_replication_pending
+
           # Start the old primary node again
           praefect_manager.start_primary_node
-          praefect_manager.wait_for_health_check_current_primary_node
-
-          # Confirm dataloss (i.e., inconsistent nodes)
-          expect(praefect_manager.replicated?(project.id)).to be false
+          praefect_manager.wait_for_health_check_all_nodes
 
-          # Reconcile nodes to recover from dataloss
-          praefect_manager.reconcile_nodes
+          # Wait for automatic replication
           praefect_manager.wait_for_replication(project.id)
 
-          # Confirm that both commits are available after reconciliation
+          # Confirm that both commits are available
           expect(project.commits.map { |commit| commit[:message].chomp })
             .to include("Initial commit").and include("pushed after failover")
         end