Fix some more handler issues

* Only notify handlers when the task is changed
* Don't run handlers on hosts which have failed

lib/ansible/executor/process/result.py

@@ -123,7 +123,7 @@ class ResultProcess(multiprocessing.Process):
                     self._send_result(('host_task_skipped', result))
                 else:
                     # if this task is notifying a handler, do it now
-                    if result._task.notify:
+                    if result._task.notify and result._result.get('changed', False):
                         # The shared dictionary for notified handlers is a proxy, which
                         # does not detect when sub-objects within the proxy are modified.
                         # So, per the docs, we reassign the list so the proxy picks up and

lib/ansible/plugins/strategies/__init__.py

@@ -380,7 +380,7 @@ class StrategyBase:
                         break
                     self._tqm.send_callback('v2_playbook_on_handler_task_start', handler)
                     for host in self._notified_handlers[handler_name]:
-                        if not handler.has_triggered(host):
+                        if not handler.has_triggered(host) and host.name not in self._tqm._failed_hosts:
                             task_vars = self._variable_manager.get_vars(loader=self._loader, play=iterator._play, host=host, task=handler)
                             task_vars = self.add_tqm_variables(task_vars, play=iterator._play)
                             self._queue_task(host, handler, task_vars, connection_info)