Module org.elasticsearch.server
Package org.elasticsearch.cluster

Interface ClusterStateTaskExecutor<T extends ClusterStateTaskListener>

Type Parameters:
T - The type of tasks to execute.
All Known Implementing Classes:
NodeJoinExecutor, NodeLeftExecutor, ReservedStateUpdateTaskExecutor, ShardStateAction.ShardFailedClusterStateTaskExecutor, ShardStateAction.ShardStartedClusterStateTaskExecutor, SimpleBatchedAckListenerTaskExecutor, SimpleBatchedExecutor
public interface ClusterStateTaskExecutor<T extends ClusterStateTaskListener>
An executor for batches of cluster state update tasks.

  • Method Details

    • execute

      ClusterState execute(ClusterStateTaskExecutor.BatchExecutionContext<T> batchExecutionContext) throws Exception
      Update the cluster state based on the current state and the given tasks. Return batchExecutionContext.initialState() to avoid publishing any update.

      If this method throws an exception then the cluster state is unchanged and every task's ClusterStateTaskListener.onFailure(java.lang.Exception) method is called.

      A common implementation pattern is to iterate through the tasks, constructing a new and updated ClusterState for each one. This works ok but beware that constructing a whole new ClusterState can be somewhat expensive, and there may sometimes be surprisingly many tasks to process in the batch. If it's possible to accumulate the effects of the tasks at a lower level then you should do that instead.

      Returning batchExecutionContext.initialState() is an important and useful optimisation in most cases, but note that this fast-path exposes APIs to the risk of stale reads in the vicinity of a master failover: a node N that handles such a no-op task batch does not verify with its peers that it's still the master, and if it's not the master then another node M may already have become master and updated the state in a way that would be inconsistent with the response that N sends back to clients.

      Returns:
      The resulting cluster state after executing all the tasks. If {code batchExecutionContext.initialState()} is returned then no update is published.
      Throws:
      Exception

    • runOnlyOnMaster

      default boolean runOnlyOnMaster()
      Returns:
      true iff this executor should only run on the elected master.

    • clusterStatePublished

      default void clusterStatePublished(ClusterState newClusterState)
      Callback invoked after new cluster state is published. Note that this method is not invoked if the cluster state was not updated. Note that this method will be executed using system context.
      Parameters:
      newClusterState - The new state which was published.

    • describeTasks

      default String describeTasks(List<T> tasks)
      Builds a concise description of a list of tasks (to be used in logging etc.). Note that the tasks given are not necessarily the same as those that will be passed to execute(org.elasticsearch.cluster.ClusterStateTaskExecutor.BatchExecutionContext<T>) but are guaranteed to be a subset of them. This method can be called multiple times with different lists before execution.
      Parameters:
      tasks - the tasks to describe.
      Returns:
      A string which describes the batch of tasks.