Package org.elasticsearch.tasks

Class TaskManager

java.lang.Object
org.elasticsearch.tasks.TaskManager
All Implemented Interfaces:
ClusterStateApplier
public class TaskManager
extends java.lang.Object
implements ClusterStateApplier
Task Manager service for keeping track of currently running tasks on the nodes

  • Constructor Details

  • Method Details

    • setTaskResultsService

      public void setTaskResultsService​(TaskResultsService taskResultsService)

    • register

      public Task register​(java.lang.String type, java.lang.String action, TaskAwareRequest request)
      Registers a task without parent task

    • cancel

      public void cancel​(CancellableTask task, java.lang.String reason, java.lang.Runnable listener)
      Cancels a task

      After starting cancellation on the parent task, the task manager tries to cancel all children tasks of the current task. Once cancellation of the children tasks is done, the listener is triggered. If the task is completed or unregistered from TaskManager, then the listener is called immediately.

    • unregister

      public Task unregister​(Task task)
      Unregister the task

    • registerChildNode

      public Releasable registerChildNode​(long taskId, DiscoveryNode node)
      Register a node on which a child task will execute. The returned Releasable must be called to unregister the child node once the child task is completed or failed.

    • localNode

      public DiscoveryNode localNode()

    • storeResult

      public <Response extends ActionResponse> void storeResult​(Task task, java.lang.Exception error, ActionListener<Response> listener)
      Stores the task failure

    • storeResult

      public <Response extends ActionResponse> void storeResult​(Task task, Response response, ActionListener<Response> listener)
      Stores the task result

    • getTasks

      public java.util.Map<java.lang.Long,​Task> getTasks()
      Returns the list of currently running tasks on the node

    • getCancellableTasks

      public java.util.Map<java.lang.Long,​CancellableTask> getCancellableTasks()
      Returns the list of currently running tasks on the node that can be cancelled

    • getTask

      public Task getTask​(long id)
      Returns a task with given id, or null if the task is not found.

    • getCancellableTask

      public CancellableTask getCancellableTask​(long id)
      Returns a cancellable task with given id, or null if the task is not found.

    • getBanCount

      public int getBanCount()
      Returns the number of currently banned tasks.

      Will be used in task manager stats and for debugging.

    • setBan

      public java.util.List<CancellableTask> setBan​(TaskId parentTaskId, java.lang.String reason)
      Bans all tasks with the specified parent task from execution, cancels all tasks that are currently executing.

      This method is called when a parent task that has children is cancelled.

      Returns:
      a list of pending cancellable child tasks

    • removeBan

      public void removeBan​(TaskId parentTaskId)
      Removes the ban for the specified parent task.

      This method is called when a previously banned task finally cancelled

    • getBannedTaskIds

      public java.util.Set<TaskId> getBannedTaskIds()

    • startBanOnChildrenNodes

      public java.util.Collection<DiscoveryNode> startBanOnChildrenNodes​(long taskId, java.lang.Runnable onChildTasksCompleted)
      Start rejecting new child requests as the parent task was cancelled.
      Parameters:
      taskId - the parent task id
      onChildTasksCompleted - called when all child tasks are completed or failed
      Returns:
      the set of current nodes that have outstanding child tasks

    • applyClusterState

      public void applyClusterState​(ClusterChangedEvent event)
      Description copied from interface: ClusterStateApplier
      Called when a new cluster state (ClusterChangedEvent.state() needs to be applied. The cluster state to be applied is already committed when this method is called, so an applier must therefore be prepared to deal with any state it receives without throwing an exception. Throwing an exception from an applier is very bad because it will stop the application of this state before it has reached all the other appliers, and will likely result in another attempt to apply the same (or very similar) cluster state which might continue until this node is removed from the cluster.
      Specified by:
      applyClusterState in interface ClusterStateApplier

    • waitForTaskCompletion

      public void waitForTaskCompletion​(Task task, long untilInNanos)
      Blocks the calling thread, waiting for the task to vanish from the TaskManager.