依赖项函数的形式和结构与*路径操作函数*一样。

因此，可以把依赖项当作没有「装饰器」（即，没有 `@app.get("/some-path")` ）的路径操作函数。

依赖项可以返回各种内容。

本例中的依赖项预期接收如下参数：

* 类型为 `str` 的可选查询参数 `q`
* 类型为 `int` 的可选查询参数 `skip`，默认值是 `0`
* 类型为 `int` 的可选查询参数 `limit`，默认值是 `100`

然后，依赖项函数返回包含这些值的 `dict`。

### 导入 `Depends`

{* ../../docs_src/dependencies/tutorial001.py hl[3] *}

### 声明依赖项

与在*路径操作函数*参数中使用 `Body`、`Query` 的方式相同，声明依赖项需要使用 `Depends` 和一个新的参数：

			for k := range timeOccurrenceMap {
				if k == nano {
					// We add to ourself later
					continue
				}
				diff := k - nano
				if diff < 0 {
					diff = -diff
				}
				// We are within the limit
				if diff < groupNano {
					timeOccurrenceMap[k]++
				}
			}
		}
		// Add ourself...
		timeOccurrenceMap[nano]++
	}

	maxima = 0 // Counter for remembering max occurrence of elements.

		loggerOnce: loggerOnce,
		quitCh:     make(chan struct{}),
	}
	if target.store != nil {
		if args.BatchSize > 1 {
			target.batch = store.NewBatch[event.Event](store.BatchConfig[event.Event]{
				Limit:         args.BatchSize,
				Log:           loggerOnce,
				Store:         queueStore,
				CommitTimeout: args.BatchCommitTimeout,
			})
		}
		store.StreamItems(target.store, target, target.quitCh, target.loggerOnce)

	}
	return lockedOpenFile(path, flag, perm, lockType)
}

// fixLongPath returns the extended-length (\\?\-prefixed) form of
// path when needed, in order to avoid the default 260 character file
// path limit imposed by Windows. If path is not easily converted to
// the extended-length form (for example, if path is a relative path
// or contains .. elements), or is short enough, fixLongPath returns
// path unmodified.
//

// callers to indicate if they want to disable md5sum checksum.
func NewReaderWithOpts(ctx context.Context, src io.Reader, opts Options) (*Reader, error) {
	// return hard limited reader
	return newReader(ctx, src, opts.Size, opts.MD5Hex, opts.SHA256Hex, opts.ActualSize, opts.DisableMD5, opts.ForceMD5)
}

// NewReader returns a new Reader that wraps src and computes

В этом случае эта зависимость ожидает:

* Необязательный query-параметр `q` типа `str`.
* Необязательный query-параметр `skip` типа `int`, по умолчанию `0`.
* Необязательный query-параметр `limit` типа `int`, по умолчанию `100`.

А затем просто возвращает `dict`, содержащий эти значения.

/// info | Информация

FastAPI добавил поддержку `Annotated` (и начал рекомендовать его использование) в версии 0.95.0.

     * </ul>
     * <p>
     * The active profiles control various aspects of the build configuration including but not
     * limited to dependencies, plugins, properties, and build resources.
     *
     * @return a non-null, possibly empty list of active profiles for this project
     * @see Profile
     * @see #getEffectiveActiveProfiles()
     */

Neste caso, a dependência espera por:

* Um parâmetro de consulta opcional `q` do tipo `str`.
* Um parâmetro de consulta opcional `skip` do tipo `int`, e igual a `0` por padrão.
* Um parâmetro de consulta opcional `limit` do tipo `int`, e igual a `100` por padrão.

E então retorna um `dict` contendo esses valores.

/// info | Informação

FastAPI passou a suportar a notação `Annotated` (e começou a recomendá-la) na versão 0.95.0.

    }.also { expected ->
      assertThat(expected.message).isEqualTo("enclosed object too large")
    }
  }

  /** Object identifiers are nominally self-delimiting. Outrun the limit with one. */
  @Test fun `variable length long outruns limit`() {
    val bytes = "060229ffffff7f".decodeHex()
    assertFailsWith<ProtocolException> {
      Adapters.OBJECT_IDENTIFIER.fromDer(bytes)
    }.also { expected ->

ejecutas **un solo proceso por contenedor**, tendrás una cantidad de memoria más o menos bien definida, estable y limitada consumida por cada uno de esos contenedores (más de uno si están replicados).

Y luego puedes establecer esos mismos límites de memoria y requisitos en tus configuraciones para tu sistema de gestión de contenedores (por ejemplo, en **Kubernetes**). De esa manera, podrá **replicar los contenedores** en las **máquinas disponibles** teniendo en cuenta la cantidad de memoria...