如果你有 **多个容器**，可能每个容器都运行一个 **单个进程**（例如，在 **Kubernetes** 集群中），那么你可能希望有一个 **单独的容器** 执行以下操作： 在单个容器中运行单个进程执行**先前步骤**，即运行复制的worker容器之前。

!!! info
    如果你使用 Kubernetes，这可能是 <a href="https://kubernetes.io/docs/concepts/workloads/pods/init-containers/" class="external-link" target="_blank">Init Container</a>。

如果在你的用例中，运行前面的步骤**并行多次**没有问题（例如，如果你没有运行数据库迁移，而只是检查数据库是否已准备好），那么你也可以将它们放在开始主进程之前在每个容器中。

### 单容器

        multiple of the number of replicas.
    *   `tf.data.Dataset` now supports automatic data distribution and sharding
        in distributed environments, including on TPU pods.
    *   Distribution policies for `tf.data.Dataset` can now be tuned with 1.
        `tf.data.experimental.AutoShardPolicy(OFF, AUTO, FILE, DATA)` 2.
        `tf.data.experimental.ExternalStatePolicy(WARN, IGNORE, FAIL)`

  // all pods within the namespace.
  // +patchStrategy=replace
  // +optional
  optional k8s.io.apimachinery.pkg.apis.meta.v1.LabelSelector selector = 2;

  // An eviction is allowed if at most "maxUnavailable" pods selected by
  // "selector" are unavailable after the eviction, i.e. even in absence of
  // the evicted pod. For example, one can prevent all voluntary evictions

  // can be scaled down further, followed by scaling up the new ReplicaSet, ensuring
  // that the total number of pods available at all times during the update is at
  // least 70% of desired pods.
  // +optional
  optional k8s.io.apimachinery.pkg.util.intstr.IntOrString maxUnavailable = 1;

  // The maximum number of pods that can be scheduled above the desired number of
  // pods.

  // resource metric across all relevant pods, represented as a percentage of
  // the requested value of the resource for the pods.
  // +optional
  optional int32 targetAverageUtilization = 2;

  // targetAverageValue is the target value of the average of the
  // resource metric across all relevant pods, as a raw value (instead of as
  // a percentage of the request), similar to the "pods" metric source type.
  // +optional

  //   * increase no more than 4 pods per 60 seconds
  //   * double the number of pods per 60 seconds
  // No stabilization is used.
  // +optional
  optional HPAScalingRules scaleUp = 1;

  // scaleDown is scaling policy for scaling Down.
  // If not set, the default value is to allow to scale down to minReplicas pods, with a
  // 300 second stabilization window (i.e., the highest recommendation for

  // A null selector selects no pods.
  // An empty selector ({}) also selects no pods, which differs from standard behavior of selecting all pods.
  // In policy/v1, an empty selector will select all pods in the namespace.
  // +optional
  optional k8s.io.apimachinery.pkg.apis.meta.v1.LabelSelector selector = 2;

  // An eviction is allowed if at most "maxUnavailable" pods selected by

  optional string completionMode = 9;

  // suspend specifies whether the Job controller should create Pods or not. If
  // a Job is created with suspend set to true, no Pods are created by the Job
  // controller. If a Job is suspended after creation (i.e. the flag goes from
  // false to true), the Job controller will delete all active Pods associated
  // with this Job. Users must design their workload to gracefully handle this.

message NetworkPolicyPeer {
  // podSelector is a label selector which selects pods. This field follows standard label
  // selector semantics; if present but empty, it selects all pods.
  //
  // If namespaceSelector is also set, then the NetworkPolicyPeer as a whole selects
  // the pods matching podSelector in the Namespaces selected by NamespaceSelector.
  // Otherwise it selects the pods matching podSelector in the policy's own namespace.
  // +optional

  // resource metric across all relevant pods, as a raw value (instead of as
  // a percentage of the request), similar to the "pods" metric source type.
  // +optional
  optional k8s.io.apimachinery.pkg.api.resource.Quantity targetAverageValue = 3;

  // container is the name of the container in the pods of the scaling target
  optional string container = 4;
}