Fix stale DNS records when resurrector recreates a VM

src/bosh-director/lib/bosh/director/cloudcheck_helper.rb

@@ -54,6 +54,7 @@ def recreate_vm_without_wait(instance_model)
     def recreate_vm(instance_model, wait_for_running = true)
       @logger.debug("Recreating Vm: #{instance_model})")
       delete_vm_from_cloud(instance_model)
+      delete_dns_for_instance_safely(instance_model)
 
       deployment_model = instance_model.deployment
       factory = Bosh::Director::DeploymentPlan::PlannerFactory.create(@logger)
@@ -167,6 +168,12 @@ def vm_creator(template_cache, dns_encoder, link_provider_intents)
       @vm_creator ||= VmCreator.new(@logger, template_cache, dns_encoder, agent_broadcaster, link_provider_intents)
     end
 
+    def delete_dns_for_instance_safely(instance_model)
+      LocalDnsManager.create(Config.root_domain, @logger).delete_dns_for_instance(instance_model)
+    rescue StandardError => e
+      @logger.warn("Failed to clean up DNS records for #{instance_model}, continuing: #{e.message}")
+    end
+
     def validate_spec(spec)
       handler_error('Unable to look up VM apply spec') unless spec
 

src/bosh-director/lib/bosh/director/problem_handlers/unresponsive_agent.rb

@@ -57,7 +57,7 @@ def instance_group
 
       resolution :delete_vm do
         plan { 'Delete VM' }
-        action { validate; delete_vm_from_cloud(@instance) }
+        action { validate; delete_vm_from_cloud(@instance); delete_dns_for_instance_safely(@instance) }
       end
 
       resolution :delete_vm_reference do

src/bosh-director/spec/unit/bosh/director/cloudcheck_helper_spec.rb

@@ -326,6 +326,41 @@ def expect_vm_gets_created
             end
           end
         end
+
+        context 'DNS cleanup' do
+          let(:local_dns_manager) { instance_double(LocalDnsManager) }
+
+          before do
+            fake_job_context
+            allow(Config).to receive(:root_domain).and_return('bosh')
+            allow(LocalDnsManager).to receive(:create).and_return(local_dns_manager)
+            allow(local_dns_manager).to receive(:delete_dns_for_instance)
+            allow(local_dns_manager).to receive(:update_dns_record_for_instance)
+            allow(vm_deleter).to receive(:delete_for_instance)
+            allow(vm_creator).to receive(:create_for_instance_plan)
+            allow(rendered_templates_persister).to receive(:persist)
+            allow(agent_client).to receive(:apply)
+            allow(agent_client).to receive(:run_script)
+            allow(agent_client).to receive(:start)
+            allow(DnsNameGenerator).to receive(:dns_record_name).and_return('some.dns.record')
+            allow(template_cache).to receive(:clean_cache!)
+          end
+
+          it 'deletes DNS records before recreating the VM' do
+            calls = []
+            allow(vm_deleter).to receive(:delete_for_instance) { calls << :delete_vm }
+            allow(local_dns_manager).to receive(:delete_dns_for_instance) { calls << :delete_dns }
+            allow(vm_creator).to receive(:create_for_instance_plan) { calls << :create_vm }
+            test_problem_handler.apply_resolution(:recreate_vm)
+            expect(calls).to eq(%i[delete_vm delete_dns create_vm])
+          end
+
+          it 'continues recreation when DNS cleanup raises' do
+            allow(local_dns_manager).to receive(:delete_dns_for_instance).and_raise(StandardError, 'dns error')
+            expect(vm_creator).to receive(:create_for_instance_plan)
+            test_problem_handler.apply_resolution(:recreate_vm)
+          end
+        end
       end
     end
   end

src/bosh-director/spec/unit/bosh/director/problem_handlers/unresponsive_agent_spec.rb

@@ -332,6 +332,31 @@ def expect_vm_to_be_created
           Models::Vm.where(cid: 'vm-cid').count
         }.from(1).to(0)
       end
+
+      context 'DNS cleanup' do
+        let(:local_dns_manager) { instance_double(LocalDnsManager) }
+
+        before do
+          allow(agent).to receive(:ping).and_raise(RpcTimeout)
+          allow(cloud).to receive(:delete_vm).with('vm-cid')
+          allow(Config).to receive(:root_domain).and_return('bosh')
+          allow(LocalDnsManager).to receive(:create).and_return(local_dns_manager)
+          allow(local_dns_manager).to receive(:delete_dns_for_instance)
+        end
+
+        it 'cleans up DNS records after deleting the VM' do
+          calls = []
+          allow(cloud).to receive(:delete_vm) { calls << :delete_vm }
+          allow(local_dns_manager).to receive(:delete_dns_for_instance) { calls << :delete_dns }
+          handler.apply_resolution(:delete_vm)
+          expect(calls).to eq(%i[delete_vm delete_dns])
+        end
+
+        it 'continues when DNS cleanup raises' do
+          allow(local_dns_manager).to receive(:delete_dns_for_instance).and_raise(StandardError, 'dns error')
+          expect { handler.apply_resolution(:delete_vm) }.not_to raise_error
+        end
+      end
     end
 
     describe 'delete_vm_reference resolution' do