diff --git a/tempest/api/volume/test_volumes_extend.py b/tempest/api/volume/test_volumes_extend.py
index 1eb76a0..de28a30 100644
--- a/tempest/api/volume/test_volumes_extend.py
+++ b/tempest/api/volume/test_volumes_extend.py
@@ -13,12 +13,15 @@
 #    License for the specific language governing permissions and limitations
 #    under the License.
 
+import time
+
 import testtools
 
 from tempest.api.volume import base
 from tempest.common import waiters
 from tempest import config
 from tempest.lib import decorators
+from tempest.lib import exceptions as lib_exc
 
 CONF = config.CONF
 
@@ -53,3 +56,129 @@
         resized_volume = self.volumes_client.show_volume(
             volume['id'])['volume']
         self.assertEqual(extend_size, resized_volume['size'])
+
+
+class VolumesExtendAttachedTest(base.BaseVolumeTest):
+    """Tests extending the size of an attached volume."""
+
+    # We need admin credentials for getting instance action event details. By
+    # default a non-admin can list and show instance actions if they own the
+    # server instance, but since the event details can contain error messages
+    # and tracebacks, like an instance fault, those are not viewable by
+    # non-admins. This is obviously not a great user experience since the user
+    # may not know when the operation is actually complete. A microversion in
+    # the compute API will be added so that non-admins can see instance action
+    # events but will continue to hide the traceback field.
+    # TODO(mriedem): Change this to not rely on the admin user to get the event
+    # details once that microversion is available in Nova.
+    credentials = ['primary', 'admin']
+
+    _api_version = 3
+    # NOTE(mriedem): The minimum required volume API version is 3.42 and the
+    # minimum required compute API microversion is 2.51, but the compute call
+    # is implicit - Cinder calls Nova at that microversion, Tempest does not.
+    min_microversion = '3.42'
+
+    @classmethod
+    def setup_clients(cls):
+        super(VolumesExtendAttachedTest, cls).setup_clients()
+        cls.admin_servers_client = cls.os_admin.servers_client
+
+    def _find_extend_volume_instance_action(self, server_id):
+        actions = self.servers_client.list_instance_actions(
+            server_id)['instanceActions']
+        for action in actions:
+            if action['action'] == 'extend_volume':
+                return action
+
+    def _find_extend_volume_instance_action_finish_event(self, action):
+        # This has to be called by an admin client otherwise
+        # the events don't show up.
+        action = self.admin_servers_client.show_instance_action(
+            action['instance_uuid'], action['request_id'])['instanceAction']
+        for event in action['events']:
+            if (event['event'] == 'compute_extend_volume' and
+                    event['finish_time']):
+                return event
+
+    @decorators.idempotent_id('301f5a30-1c6f-4ea0-be1a-91fd28d44354')
+    @testtools.skipUnless(CONF.volume_feature_enabled.extend_attached_volume,
+                          "Attached volume extend is disabled.")
+    def test_extend_attached_volume(self):
+        """This is a happy path test which does the following:
+
+        * Create a volume at the configured volume_size.
+        * Create a server instance.
+        * Attach the volume to the server.
+        * Wait for the volume status to be "in-use".
+        * Extend the size of the volume and wait for the volume status to go
+          back to "in-use".
+        * Assert the volume size change is reflected in the volume API.
+        * Wait for the "compute_extend_volume" instance action event to show
+          up in the compute API with the success or failure status. We fail
+          if we timeout waiting for the instance action event to show up, or
+          if the action on the server fails.
+        """
+        # Create a test volume. Will be automatically cleaned up on teardown.
+        volume = self.create_volume()
+        # Create a test server. Will be automatically cleaned up on teardown.
+        server = self.create_server()
+        # Attach the volume to the server and wait for the volume status to be
+        # "in-use".
+        self.attach_volume(server['id'], volume['id'])
+        # Extend the size of the volume. If this is successful, the volume API
+        # will change the status on the volume to "extending" before doing an
+        # RPC cast to the volume manager on the backend. Note that we multiply
+        # the size of the volume since certain Cinder backends, e.g. ScaleIO,
+        # require multiples of 8GB.
+        extend_size = volume['size'] * 2
+        self.volumes_client.extend_volume(volume['id'], new_size=extend_size)
+        # The volume status should go back to in-use since it is still attached
+        # to the server instance.
+        waiters.wait_for_volume_resource_status(self.volumes_client,
+                                                volume['id'], 'in-use')
+        # Assert that the volume size has changed in the volume API.
+        volume = self.volumes_client.show_volume(volume['id'])['volume']
+        self.assertEqual(extend_size, volume['size'])
+        # Now we wait for the "compute_extend_volume" instance action event
+        # to show up for the server instance. This is our indication that the
+        # asynchronous operation is complete on the compute side.
+        start_time = int(time.time())
+        timeout = self.servers_client.build_timeout
+        action = self._find_extend_volume_instance_action(server['id'])
+        while action is None and int(time.time()) - start_time < timeout:
+            time.sleep(self.servers_client.build_interval)
+            action = self._find_extend_volume_instance_action(server['id'])
+
+        if action is None:
+            msg = ("Timed out waiting to get 'extend_volume' instance action "
+                   "record for server %(server)s after %(timeout)s seconds." %
+                   {'server': server['id'], 'timeout': timeout})
+            raise lib_exc.TimeoutException(msg)
+
+        # Now that we found the extend_volume instance action, we can wait for
+        # the compute_extend_volume instance action event to show up to
+        # indicate the operation is complete.
+        start_time = int(time.time())
+        event = self._find_extend_volume_instance_action_finish_event(action)
+        while event is None and int(time.time()) - start_time < timeout:
+            time.sleep(self.servers_client.build_interval)
+            event = self._find_extend_volume_instance_action_finish_event(
+                action)
+
+        if event is None:
+            msg = ("Timed out waiting to get 'compute_extend_volume' instance "
+                   "action event record for server %(server)s and request "
+                   "%(request_id)s after %(timeout)s seconds." %
+                   {'server': server['id'],
+                    'request_id': action['request_id'],
+                    'timeout': timeout})
+            raise lib_exc.TimeoutException(msg)
+
+        # Finally, assert that the action completed successfully.
+        self.assertTrue(
+            event['result'].lower() == 'success',
+            "Unexpected compute_extend_volume result '%(result)s' for request "
+            "%(request_id)s." %
+            {'result': event['result'],
+             'request_id': action['request_id']})
diff --git a/tempest/config.py b/tempest/config.py
index af9eefc..e6fa579 100644
--- a/tempest/config.py
+++ b/tempest/config.py
@@ -833,7 +833,14 @@
                 help="Is the v2 volume API enabled"),
     cfg.BoolOpt('api_v3',
                 default=True,
-                help="Is the v3 volume API enabled")
+                help="Is the v3 volume API enabled"),
+    cfg.BoolOpt('extend_attached_volume',
+                default=False,
+                help='Does the cloud support extending the size of a volume '
+                     'which is currently attached to a server instance? This '
+                     'depends on the 3.42 volume API microversion and the '
+                     '2.51 compute API microversion. Also, not all volume or '
+                     'compute backends support this operation.')
 ]