Package io.trino.execution.scheduler

Class UniformNodeSelector

java.lang.Object
io.trino.execution.scheduler.UniformNodeSelector
All Implemented Interfaces:
NodeSelector
public class UniformNodeSelector extends Object implements NodeSelector

  • Constructor Details

    • UniformNodeSelector

      public UniformNodeSelector(InternalNodeManager nodeManager, NodeTaskMap nodeTaskMap, boolean includeCoordinator, Supplier<NodeMap> nodeMap, int minCandidates, long maxSplitsWeightPerNode, long minPendingSplitsWeightPerTask, long maxAdjustedPendingSplitsWeightPerTask, int maxUnacknowledgedSplitsPerTask, NodeSchedulerConfig.SplitsBalancingPolicy splitsBalancingPolicy, boolean optimizedLocalScheduling)

  • Method Details

    • lockDownNodes

      public void lockDownNodes()
      Specified by:
      lockDownNodes in interface NodeSelector

    • allNodes

      public List<InternalNode> allNodes()
      Specified by:
      allNodes in interface NodeSelector

    • selectCurrentNode

      public InternalNode selectCurrentNode()
      Specified by:
      selectCurrentNode in interface NodeSelector

    • selectRandomNodes

      public List<InternalNode> selectRandomNodes(int limit, Set<InternalNode> excludedNodes)
      Specified by:
      selectRandomNodes in interface NodeSelector

    • computeAssignments

      public SplitPlacementResult computeAssignments(Set<Split> splits, List<RemoteTask> existingTasks)
      Description copied from interface: NodeSelector
      Identifies the nodes for running the specified splits.
      Specified by:
      computeAssignments in interface NodeSelector
      Parameters:
      splits - the splits that need to be assigned to nodes
      Returns:
      a multimap from node to splits only for splits for which we could identify a node to schedule on. If we cannot find an assignment for a split, it is not included in the map. Also returns a future indicating when to reattempt scheduling of this batch of splits, if some of them could not be scheduled.

    • computeAssignments

      public SplitPlacementResult computeAssignments(Set<Split> splits, List<RemoteTask> existingTasks, BucketNodeMap bucketNodeMap)
      Description copied from interface: NodeSelector
      Identifies the nodes for running the specified splits based on a precomputed fixed partitioning.
      Specified by:
      computeAssignments in interface NodeSelector
      Parameters:
      splits - the splits that need to be assigned to nodes
      Returns:
      a multimap from node to splits only for splits for which we could identify a node with free space. If we cannot find an assignment for a split, it is not included in the map. Also returns a future indicating when to reattempt scheduling of this batch of splits, if some of them could not be scheduled.

    • redistributeSplit

      public static Split redistributeSplit(com.google.common.collect.Multimap<InternalNode,Split> assignment, InternalNode fromNode, InternalNode toNode, com.google.common.collect.SetMultimap<InetAddress,InternalNode> nodesByHost)
      The method selects and removes a split from the fromNode and assigns it to the toNode. There is an attempt to redistribute a Non-local split if possible. This case is possible when there are multiple queries running simultaneously. If a Non-local split cannot be found in the maxNode, any split is selected randomly and reassigned.