```Python hl_lines="21  26"
{!../../docs_src/metadata/tutorial004.py!}
```

/// info

Lesen Sie mehr zu Tags unter [Pfadoperation-Konfiguration](path-operation-configuration.md#tags){.internal-link target=_blank}.

///

### Die Dokumentation anschauen

Wenn Sie nun die Dokumentation ansehen, werden dort alle zusätzlichen Metadaten angezeigt:

///

### 使用你的标签

将 `tags` 参数和*路径操作*（以及 `APIRouter`）一起使用，将其分配给不同的标签：

```Python hl_lines="21  26"
{!../../docs_src/metadata/tutorial004.py!}
```

/// info | "信息"

阅读更多关于标签的信息[路径操作配置](path-operation-configuration.md#tags){.internal-link target=_blank}。

///

### 查看文档

如果你现在查看文档，它们会显示所有附加的元数据：

<img src="/img/tutorial/metadata/image02.png">

### 标签顺序

每个标签元数据字典的顺序也定义了在文档用户界面显示的顺序。

```Python hl_lines="21  26"
{!../../docs_src/metadata/tutorial004.py!}
```

/// info  | Informação

Leia mais sobre tags em [Configuração de Operação de Caminho](path-operation-configuration.md#tags){.internal-link target=_blank}.

///

### Cheque os documentos

Agora, se você verificar a documentação, ela exibirá todos os metadados adicionais:

struct LoggingDevice {
  tensorflow::string device_name;
  tensorflow::string underlying_device;
  // Set to true whenever a TensorHandle is copied onto the device
  bool* arrived_flag;
  // Set to true whenever an operation is executed
  bool* executed_flag;
  // If true, only explicit op placements are accepted. If false, uses
  // type-based dispatch.
  bool strict_scope_placement;
};

struct LoggedTensor {

                                    }
                                  },
                                  "decorator": {
                                    "operation": ":9080/*"
                                  },
                                  "name": "default"
                                }
                              ]
                            }

   *
   * <p>If the mapped keys or values contain duplicates (according to {@link
   * Object#equals(Object)}), an {@code IllegalArgumentException} is thrown when the collection
   * operation is performed. (This differs from the {@code Collector} returned by {@link
   * Collectors#toMap(Function, Function)}, which throws an {@code IllegalStateException}.)
   *
   * @since 33.2.0 (available since 21.0 in guava-jre)

```

////

/// tip

We'll discuss the `@lru_cache` in a bit.

For now you can assume `get_settings()` is a normal function.

///

And then we can require it from the *path operation function* as a dependency and use it anywhere we need it.

//// tab | Python 3.9+

```Python hl_lines="17  19-21"
{!> ../../docs_src/settings/app02_an_py39/main.py!}
```

////

        NT_STATUS_CONNECTION_REFUSED, NT_STATUS_PATH_NOT_COVERED, NT_STATUS_IO_REPARSE_TAG_NOT_HANDLED, NT_STATUS_NO_MORE_FILES,
    };

    static final String[] NT_STATUS_MESSAGES = {
        "The operation completed successfully.", "Request is pending", "A notify change request is being completed.",

  // so there is no object state put at risk by stopping the threads abruptly. In some cases a test
  // may put a thread into an uninterruptible operation intentionally, so there is no other way to
  // clean up these threads.
  @SuppressWarnings("deprecation")
  @Override
  public void tearDown() throws Exception {
    stop();
    join();

Let's start with an example and then see it in detail.

We create an async function `lifespan()` with `yield` like this:

{* ../../docs_src/events/tutorial003.py hl[16,19] *}

Here we are simulating the expensive *startup* operation of loading the model by putting the (fake) model function in the dictionary with machine learning models before the `yield`. This code will be executed **before** the application **starts taking requests**, during the *startup*.