Fix duration for builds failed via Ci::Build#doom!

The duration of a build is `finished_at - started_at`. This method was
marking a build as failed without setting `finished_at`, resulting in
its duration appearing ever-growing.

Changelog: fixed

app/models/ci/build.rb

@@ -921,13 +921,25 @@ def report_artifacts
     # Consider this object to have a structural integrity problems
     def doom!
       transaction do
-        update_columns(status: :failed, failure_reason: :data_integrity_failure)
+        now = Time.current
+        attributes = {
+          status: :failed,
+          failure_reason: :data_integrity_failure,
+          updated_at: now
+        }
+        attributes[:finished_at] = now unless finished_at.present?
+
+        update_columns(attributes)
         all_queuing_entries.delete_all
         all_runtime_metadata.delete_all
       end
 
       deployment&.sync_status_with(self)
 
+      ::Gitlab::Ci::Pipeline::Metrics
+        .job_failure_reason_counter
+        .increment(reason: :data_integrity_failure)
+
       Gitlab::AppLogger.info(
         message: 'Build doomed',
         class: self.class.name,

spec/models/ci/build_spec.rb

@@ -5229,16 +5229,34 @@ def run_job_without_exception
     subject { build.doom! }
 
     let(:traits) { [] }
-    let(:build) { create(:ci_build, *traits, pipeline: pipeline) }
+    let(:build) do
+      travel(-1.minute) do
+        create(:ci_build, *traits, pipeline: pipeline)
+      end
+    end
 
-    it 'updates status and failure_reason', :aggregate_failures do
-      subject
+    it 'updates status, failure_reason, finished_at and updated_at', :aggregate_failures do
+      old_timestamp = build.updated_at
 
+      new_timestamp = \
+        freeze_time do
+          Time.current.tap do
+            subject
+          end
+        end
+
+      expect(old_timestamp).not_to eq(new_timestamp)
+      expect(build.updated_at).to eq(new_timestamp)
+      expect(build.finished_at).to eq(new_timestamp)
       expect(build.status).to eq("failed")
       expect(build.failure_reason).to eq("data_integrity_failure")
     end
 
-    it 'logs a message' do
+    it 'logs a message and increments the job failure counter', :aggregate_failures do
+      expect(::Gitlab::Ci::Pipeline::Metrics.job_failure_reason_counter)
+        .to(receive(:increment))
+        .with(reason: :data_integrity_failure)
+
       expect(Gitlab::AppLogger)
         .to receive(:info)
         .with(a_hash_including(message: 'Build doomed', class: build.class.name, build_id: build.id))
@@ -5273,12 +5291,20 @@ def run_job_without_exception
     context 'with running builds' do
       let(:traits) { [:picked] }
 
-      it 'drops associated runtime metadata' do
+      it 'drops associated runtime metadata', :aggregate_failures do
         subject
 
         expect(build.reload.runtime_metadata).not_to be_present
       end
     end
+
+    context 'finished builds' do
+      let(:traits) { [:finished] }
+
+      it 'does not update finished_at' do
+        expect { subject }.not_to change { build.reload.finished_at }
+      end
+    end
   end
 
   it 'does not generate cross DB queries when a record is created via FactoryBot' do