Merge "Prevent rescheduling on block device failure" into stable/havana

nova/compute/manager.py

@@ -1063,6 +1063,16 @@ class ComputeManager(manager.SchedulerDependentManager):
                         instance_uuid=instance['uuid'], reason=msg)
             else:
                 raise exc_info[0], exc_info[1], exc_info[2]
+        except exception.InvalidBDM:
+            with excutils.save_and_reraise_exception():
+                if network_info is not None:
+                    network_info.wait(do_raise=False)
+                try:
+                    self._deallocate_network(context, instance)
+                except Exception:
+                    msg = _('Failed to dealloc network '
+                            'for failed instance')
+                    LOG.exception(msg, instance=instance)
         except Exception:
             exc_info = sys.exc_info()
             # try to re-schedule instance:
@@ -1388,9 +1398,9 @@ class ComputeManager(manager.SchedulerDependentManager):
             return block_device_info
 
         except Exception:
-            with excutils.save_and_reraise_exception():
-                LOG.exception(_('Instance failed block device setup'),
-                              instance=instance)
+            LOG.exception(_('Instance failed block device setup'),
+                          instance=instance)
+            raise exception.InvalidBDM()
 
     def _spawn(self, context, instance, image_meta, network_info,
                block_device_info, injected_files, admin_password,

nova/tests/compute/test_compute.py

@@ -1356,11 +1356,11 @@ class ComputeTestCase(BaseTestCase):
         the instance goes to ERROR state, keeping the task state
         """
         def fake(*args, **kwargs):
-            raise test.TestingException()
+            raise exception.InvalidBDM()
         self.stubs.Set(nova.compute.manager.ComputeManager,
                        '_prep_block_device', fake)
         instance = self._create_fake_instance()
-        self.assertRaises(test.TestingException, self.compute.run_instance,
+        self.assertRaises(exception.InvalidBDM, self.compute.run_instance,
                           self.context, instance=instance)
         #check state is failed even after the periodic poll
         self._assert_state({'vm_state': vm_states.ERROR,
@@ -9003,6 +9003,20 @@ class ComputeRescheduleOrErrorTestCase(BaseTestCase):
                 self.compute._run_instance, self.context, None, {}, None, None,
                 None, False, None, self.instance, False)
 
+    def test_no_reschedule_on_block_device_fail(self):
+        self.mox.StubOutWithMock(self.compute, '_prep_block_device')
+        self.mox.StubOutWithMock(self.compute, '_reschedule_or_error')
+
+        exc = exception.InvalidBDM()
+
+        self.compute._prep_block_device(mox.IgnoreArg(), self.instance,
+                                        mox.IgnoreArg()).AndRaise(exc)
+
+        self.mox.ReplayAll()
+        self.assertRaises(exception.InvalidBDM, self.compute._run_instance,
+                          self.context, None, {}, None, None, None, False,
+                          None, self.instance, False)
+
 
 class ComputeRescheduleResizeOrReraiseTestCase(BaseTestCase):
     """Test logic and exception handling around rescheduling prep resize