"format": "time",
                            }
                        ),
                        "process_after": IsDict(
                            {
                                "title": "Process After",
                                "type": "string",
                                "format": "duration",
                            }
                        )

    # TODO: remove after upgrading SQLAlchemy to a version that includes the following changes
    # https://github.com/sqlalchemy/sqlalchemy/commit/59521abcc0676e936b31a523bd968fc157fef0c2
    'ignore:datetime\.datetime\.utcfromtimestamp\(\) is deprecated and scheduled for removal in a future version\..*:DeprecationWarning:sqlalchemy',

And it will work and will be useful **during development**.

But if your connection to the server is lost, the **running process** will probably die.

And if the server is restarted (for example after updates, or migrations from the cloud provider) you probably **won't notice it**. And because of that, you won't even know that you have to restart the process manually. So, your API will just stay dead. 😱

participant handler as Exception handler
participant dep as Dep with yield
participant operation as Path Operation
participant tasks as Background tasks

    Note over client,tasks: Can raise exception for dependency, handled after response is sent
    Note over client,operation: Can raise HTTPException and can change the response
    client ->> dep: Start request
    Note over dep: Run code up to yield
    opt raise

Doing this, our `GzipRequest` will take care of decompressing the data (if necessary) before passing it to our *path operations*.

After that, all of the processing logic is the same.

But because of our changes in `GzipRequest.body`, the request body will be automatically decompressed when it is loaded by **FastAPI** when needed.

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

### Order of tags

The order of each tag metadata dictionary also defines the order shown in the docs UI.

For example, even though `users` would go after `items` in alphabetical order, it is shown before them, because we added their metadata as the first dictionary in the list.

## OpenAPI URL

By default, the OpenAPI schema is served at `/openapi.json`.

    item = data[item_id]
    if item["owner"] != username:
        raise OwnerError(username)
    return item
```

---

Before FastAPI 0.106.0, raising exceptions after `yield` was not possible, the exit code in dependencies with `yield` was executed *after* the response was sent, so [Exception Handlers](https://fastapi.tiangolo.com/tutorial/handling-errors/#install-custom-exception-handlers) would have already run.

    cwd = os.getcwd()
    os.chdir(tmp_path)
    test_db = Path("./sql_app.db")
    if test_db.is_file():  # pragma: nocover
        test_db.unlink()
    # Import while creating the client to create the DB after starting the test session
    from docs_src.sql_databases.sql_app import main

    # Ensure import side effects are re-executed
    importlib.reload(main)
    with TestClient(main.app) as c:
        yield c

    cwd = os.getcwd()
    os.chdir(tmp_path)
    test_db = Path("./sql_app.db")
    if test_db.is_file():  # pragma: nocover
        test_db.unlink()
    # Import while creating the client to create the DB after starting the test session
    from docs_src.sql_databases.sql_app_py310 import alt_main

    # Ensure import side effects are re-executed
    importlib.reload(alt_main)

    with TestClient(alt_main.app) as c:

    * A "token" is just a string with some content that we can use later to verify this user.
    * Normally, a token is set to expire after some time.
        * So, the user will have to log in again at some point later.
        * And if the token is stolen, the risk is less. It is not like a permanent key that will work forever (in most of the cases).