Merge branch 'sh-fix-pipeline-retry-dag' into 'master'

Fix pipeline retry in a CI DAG

Closes #36068

See merge request gitlab-org/gitlab!21296

app/models/ci/pipeline.rb

@@ -812,6 +812,10 @@ module Ci
       @persistent_ref ||= PersistentRef.new(pipeline: self)
     end
 
+    def find_successful_build_ids_by_names(names)
+      statuses.latest.success.where(name: names).pluck(:id)
+    end
+
     private
 
     def pipeline_data

app/models/concerns/ci/processable.rb

@@ -14,6 +14,8 @@ module Ci
       has_many :needs, class_name: 'Ci::BuildNeed', foreign_key: :build_id, inverse_of: :build
 
       accepts_nested_attributes_for :needs
+
+      scope :preload_needs, -> { preload(:needs) }
     end
 
     def schedulable?

app/services/ci/retry_pipeline_service.rb

@@ -9,13 +9,23 @@ module Ci
         raise Gitlab::Access::AccessDeniedError
       end
 
-      pipeline.retryable_builds.find_each do |build|
+      needs = Set.new
+
+      pipeline.retryable_builds.preload_needs.find_each do |build|
         next unless can?(current_user, :update_build, build)
 
         Ci::RetryBuildService.new(project, current_user)
           .reprocess!(build)
+
+        needs += build.needs.map(&:name)
       end
 
+      # In a DAG, the dependencies may have already completed. Figure out
+      # which builds have succeeded and use them to update the pipeline. If we don't
+      # do this, then builds will be stuck in the created state since their dependencies
+      # will never run.
+      completed_build_ids = pipeline.find_successful_build_ids_by_names(needs) if needs.any?
+
       pipeline.builds.latest.skipped.find_each do |skipped|
         retry_optimistic_lock(skipped) { |build| build.process }
       end
@@ -26,7 +36,7 @@ module Ci
 
       Ci::ProcessPipelineService
         .new(pipeline)
-        .execute
+        .execute(completed_build_ids)
     end
   end
 end

changelogs/unreleased/sh-fix-pipeline-retry-dag.yml

+---
+title: Fix pipeline retry in a CI DAG
+merge_request: 21296
+author:
+type: fixed

spec/services/ci/retry_pipeline_service_spec.rb

@@ -91,6 +91,25 @@ describe Ci::RetryPipelineService, '#execute' do
       end
     end
 
+    context 'when there is a failed test in a DAG' do
+      before do
+        create_build('build', :success, 0)
+        create_build('build2', :success, 0)
+        test_build = create_build('test', :failed, 1)
+        create(:ci_build_need, build: test_build, name: 'build')
+        create(:ci_build_need, build: test_build, name: 'build2')
+      end
+
+      it 'retries the test' do
+        service.execute(pipeline)
+
+        expect(build('build')).to be_success
+        expect(build('build2')).to be_success
+        expect(build('test')).to be_pending
+        expect(build('test').needs.map(&:name)).to match_array(%w(build build2))
+      end
+    end
+
     context 'when the last stage was skipepd' do
       before do
         create_build('build 1', :success, 0)