* Se você quer fazer comparações com o Starlette, compare com Sanic, Flask, Django, etc. _Frameworks Web_ (ou _microframeworks_).
* **FastAPI**:
    * Do mesmo modo que Starlette utiliza Uvicorn e não pode ser mais rápido que ele, **FastAPI** utiliza o Starlette, então não tem como ser mais rápido do que o Starlette.

        client.get("/broken-session-stream")


def test_broken_session_stream_no_raise():
    """
    When a dependency with yield raises after the streaming response already started
    the 200 status code is already sent, but there's still an error in the server
    afterwards, an exception is raised and captured or shown in the server logs.
    """

    * 你不會直接在 Uvicorn 中編寫應用程式。這意味著你的程式碼必須或多或少地包含 Starlette（或 **FastAPI**）提供的所有程式碼。如果你這樣做，你的最終應用程式將具有與使用框架相同的開銷，且無法像使用框架那樣減少應用程式程式碼與錯誤。
    * 如果你要比較 Uvicorn，請將其與 Daphne、Hypercorn、uWSGI 等應用程式伺服器進行比較。
* **Starlette**：
    * 繼 Uvicorn 之後的次佳表現。事實上，Starlette 使用 Uvicorn 來運行。因此它將可能只透過執行更多程式碼而變得比 Uvicorn「慢」。
    * 但它為你提供了建立簡單網頁應用程式的工具，以及基於路徑的路由等。
    * 如果你要比較 Starlette，請將其與 Sanic、Flask、Django 等網頁框架（或微框架）進行比較。
* **FastAPI**：

    MINIO_KMS_SECRET_KEY: "my-minio-key:OSMM+vkKUTCvQs9YL/CVMIMt43HFhkUpqJxTmGl6rYw="
  healthcheck:
    test: ["CMD", "mc", "ready", "local"]
    interval: 5s
    timeout: 5s
    retries: 5

# starts 4 docker containers running minio server instances.
# using nginx reverse proxy, load balancing, you can access
# it through port 9000.
services:
  minio1:
    <<: *minio-common
    hostname: minio1
    volumes:

    MINIO_DRIVE_MAX_TIMEOUT: "5s"
  healthcheck:
    test: ["CMD", "mc", "ready", "local"]
    interval: 5s
    timeout: 5s
    retries: 5

# starts 4 docker containers running minio server instances.
# using nginx reverse proxy, load balancing, you can access
# it through port 9000.
services:
  minio1:
    <<: *minio-common
    hostname: minio1
    volumes:

x-minio-common: &minio-common
  image: quay.io/minio/minio:${RELEASE}
  command: server http://site1-minio{1...4}/data{1...2}
  environment:
    - MINIO_PROMETHEUS_AUTH_TYPE=public
    - CI=true

# starts 4 docker containers running minio server instances.
# using nginx reverse proxy, load balancing, you can access
# it through port 9000.
services:
  site1-minio1:
    <<: *minio-common
    hostname: site1-minio1

## Antigo `GraphQLApp` do Starlette { #older-graphqlapp-from-starlette }

Versões anteriores do Starlette incluiam uma classe `GraphQLApp` para integrar com [Graphene](https://graphene-python.org/).

* 等

**FastAPI** 使用这种方式验证数据、转换数据，并自动生成 API 文档。

但有时，我们也需要直接访问 `Request` 对象。

## `Request` 对象的细节 { #details-about-the-request-object }

实际上，**FastAPI** 的底层是 **Starlette**，**FastAPI** 只不过是在 **Starlette** 顶层提供了一些工具，所以能直接使用 Starlette 的 [`Request`](https://www.starlette.dev/requests/) 对象。

但直接从 `Request` 对象提取数据时（例如，读取请求体），这些数据不会被 **FastAPI** 验证、转换或文档化（使用 OpenAPI，为自动的 API 用户界面）。

不过，仍可以验证、转换与注释（使用 Pydantic 模型的请求体等）其它正常声明的参数。

        serializer.submit(blockingCallable::isRunning, directExecutor());

    // Wait for the first task to be started in the background. It will block until we explicitly
    // stop it.
    blockingCallable.waitForStart();

    // Give the second task a chance to (incorrectly) start up while the first task is running.
    assertThat(future2.isDone()).isFalse();

Таким образом, вы можете продолжать вызывать `HTTPException` от **FastAPI** как обычно в своем коде.

Но когда вы регистрируете обработчик исключений, вы должны зарегистрировать его для `HTTPException` от Starlette.