"crossVersionTests": false
  },
  {
    "name": "worker-processes",
    "path": "platforms/core-execution/worker-processes",
    "unitTests": true,
    "functionalTests": false,
    "crossVersionTests": false
  },
  {
    "name": "worker-services",
    "path": "platforms/core-runtime/worker-services",
    "unitTests": false,
    "functionalTests": false,

    listOf(
        // We pass the 'maxParallelForks' setting as 'workers.max' to limit the maximum number of executers even
        // if multiple test tasks run in parallel. We also pass it to the Gradle build as a maximum (maxParallelForks)
        // for each test task, such that we are independent of whatever default value is defined in the build itself.
        "-Dorg.gradle.workers.max=$maxParallelForks",
        "-PmaxParallelForks=$maxParallelForks",

      "coverage" : {
        "per_week" : [ "linux" ]
      }
    } ]
  }, {
    "testId" : "org.gradle.performance.experiment.java.ParallelBuildPerformanceTest.clean assemble with 4 parallel workers",
    "groups" : [ {
      "testProject" : "largeJavaMultiProject",
      "coverage" : {
        "per_week" : [ "linux" ]
      }
    }, {
      "testProject" : "largeMonolithicJavaProject",
      "coverage" : {

| `minio_cluster_replication_current_active_workers`         | Total number of active replication workers                                                               |
| `minio_cluster_replication_average_active_workers`         | Average number of active replication workers                                                             |

可以通过多种方法来实现这一目标，我将在接下来的章节中向您详细介绍具体策略，例如在谈论 Docker 和容器时。

要考虑的主要限制是必须有一个**单个**组件来处理**公共IP**中的**端口**。 然后它必须有一种方法将通信**传输**到复制的**进程/worker**。

以下是一些可能的组合和策略：

* **Gunicorn** 管理 **Uvicorn workers**
     * Gunicorn 将是监听 **IP** 和 **端口** 的 **进程管理器**，复制将通过 **多个 Uvicorn 工作进程** 进行
* **Uvicorn** 管理 **Uvicorn workers**
     * 一个 Uvicorn **进程管理器** 将监听 **IP** 和 **端口**，并且它将启动 **多个 Uvicorn 工作进程**
* **Kubernetes** 和其他分布式 **容器系统**

TF_CAPI_EXPORT extern TFE_ContextDevicePlacementPolicy
TFE_ContextGetDevicePlacementPolicy(TFE_Context* ctx);

// A tensorflow.ServerDef specifies remote workers (in addition to the current
// workers name). Operations created in this context can then be executed on
// any of these remote workers by setting an appropriate device.
//
// If the following is set, all servers identified by the
// ServerDef must be up when the context is created.

  server_def.set_task_index(1);
  std::unique_ptr<tensorflow::GrpcServer> worker_server1;
  ASSERT_TRUE(tensorflow::GrpcServer::Create(
                  server_def, tensorflow::Env::Default(), &worker_server1)
                  .ok());
  ASSERT_TRUE(worker_server1->Start().ok());

  server_def.set_task_index(2);
  std::unique_ptr<tensorflow::GrpcServer> worker_server2;
  ASSERT_TRUE(tensorflow::GrpcServer::Create(

## Imagem Oficial do Docker com Gunicorn - Uvicorn

Há uma imagem oficial do Docker que inclui o Gunicorn executando com trabalhadores Uvicorn, conforme detalhado em um capítulo anterior: [Server Workers - Gunicorn com Uvicorn](server-workers.md){.internal-link target=_blank}.

Tenha certeza de ajustar o número de _workers_ etc.

  /**
   * This counter prevents an ABA issue where a thread may successfully schedule the worker, the
   * worker runs and exhausts the queue, another thread enqueues a task and fails to schedule the
   * worker, and then the first thread's call to delegate.execute() returns. Without this counter,
   * it would observe the QUEUING state and set it to QUEUED, and the worker would never be
   * scheduled again for future submissions.
   */
  @GuardedBy("queue")