* expect a newly-started thread to call [Runnable.run]. We shouldn't request new threads until
   * the already-requested ones are in service, otherwise we might create more threads than we need.
   *
   * We use [executeCallCount] and [runCallCount] to defend against starting more threads than we
   * need. Both fields are guarded by [lock].
   */
  private var executeCallCount = 0
  private var runCallCount = 0

For discovery, the JWT token will be read directly from a file and sent as is. For CA, this logic is a bit more complex,
as the support for external CAs is more mature than external discovery servers. This supports some additional
configuration, a `CredentialFetcher` which allows fetching a token from places other than a file (for example, a local
metadata server), and a `TokenExchanger` which allows exchanging a token for another form to match the CA server requirements.

      description: "MinIO instance {{ $labels.server }} of job {{ $labels.job }} has lost quorum on pool {{ $labels.pool }} on set {{ $labels.set }} for more than 5 minutes."
```

## Verify the configuration and alerts
To verify the above sample alert follow below steps

1. Start a distributed MinIO instance (4 nodes setup)
2. Start Prometheus server and AlertManager

使用 `Query` 和 `Path`（以及你将在后面看到的其他类）可以声明字符串约束，但也可以声明数值约束。

像下面这样，添加 `ge=1` 后，`item_id` 将必须是一个大于（`g`reater than）或等于（`e`qual）`1` 的整数。

```Python hl_lines="8"
{!../../../docs_src/path_params_numeric_validations/tutorial004.py!}
```

## 数值校验：大于和小于等于

同样的规则适用于：

* `gt`：大于（`g`reater `t`han）
* `le`：小于等于（`l`ess than or `e`qual）

```Python hl_lines="9"
{!../../../docs_src/path_params_numeric_validations/tutorial005.py!}

 *
 * This class may be used to stream very large responses. For example, it is possible to use this
 * class to read a response that is larger than the entire memory allocated to the current process.
 * It can even stream a response larger than the total storage on the current device, which is a
 * common requirement for video streaming applications.
 *

  //      implementations prior to v1.22)
  //   2. Signer whose configured maximum is shorter than the requested duration
  //   3. Signer whose configured minimum is longer than the requested duration
  //
  // The minimum valid value for expirationSeconds is 600, i.e. 10 minutes.
  //
  // +optional
  optional int32 expirationSeconds = 8;

Import the `Response` class (sub-class) you want to use and declare it in the *path operation decorator*.

For large responses, returning a `Response` directly is much faster than returning a dictionary.

And then they can try again knowing that it's probably something more similar to `stanleyjobsox` than to `johndoe`.

#### A "professional" attack

  optional string name = 5;

  // Namespace is the namespace associated with the request (if any).
  // +optional
  optional string namespace = 6;

  // Operation is the operation being performed. This may be different than the operation
  // requested. e.g. a patch can result in either a CREATE or UPDATE Operation.
  optional string operation = 7;

  // UserInfo is information about the requesting user

And as you can have parallelism and asynchronicity at the same time, you get higher performance than most of the tested NodeJS frameworks and on par with Go, which is a compiled language closer to C <a href="https://www.techempower.com/benchmarks/#section=data-r17&hw=ph&test=query&l=zijmkf-1" class="external-link" target="_blank">(all thanks to Starlette)</a>.

### Is concurrency better than parallelism?

Nope! That's not the moral of the story.