Package org.apache.storm.scheduler.resource.strategies.scheduling

Class BaseResourceAwareStrategy

java.lang.Object
org.apache.storm.scheduler.resource.strategies.scheduling.BaseResourceAwareStrategy
All Implemented Interfaces:
IStrategy
Direct Known Subclasses:
ConstraintSolverStrategy, DefaultResourceAwareStrategy, DefaultResourceAwareStrategyOld, GenericResourceAwareStrategy, GenericResourceAwareStrategyOld, RoundRobinResourceAwareStrategy
public abstract class BaseResourceAwareStrategy extends Object implements IStrategy

  • Field Details

  • Constructor Details

    • BaseResourceAwareStrategy

      public BaseResourceAwareStrategy()

    • BaseResourceAwareStrategy

      public BaseResourceAwareStrategy(boolean sortNodesForEachExecutor, BaseResourceAwareStrategy.NodeSortType nodeSortType)
      Initialize for the default implementation of schedule().
      Parameters:
      sortNodesForEachExecutor - Sort nodes before scheduling each executor.
      nodeSortType - type of sorting to be applied to object resource collection BaseResourceAwareStrategy.NodeSortType.

  • Method Details

    • prepare

      public void prepare(Map<String,Object> config)
      Description copied from interface: IStrategy
      Prepare the Strategy for scheduling.
      Specified by:
      prepare in interface IStrategy
      Parameters:
      config - the cluster configuration

    • schedule

      public SchedulingResult schedule(Cluster cluster, TopologyDetails td)
      Note that this method is not thread-safe. Several instance variables are generated from supplied parameters. In addition, the following instance variables are set to complete scheduling:
    • searcherState
    • execSorter to sort executors
    • nodeSorter to sort nodes

      • Scheduling consists of three main steps:

    • prepareForScheduling(Cluster, TopologyDetails)
    • checkSchedulingFeasibility(), and
    • scheduleExecutorsOnNodes(List, Iterable)

      • The executors and nodes are sorted in the order most conducive to scheduling for the strategy. Those interfaces may be overridden by subclasses using mutators:

    • setExecSorter(IExecSorter) and
    • setNodeSorter(INodeSorter)

      • Specified by:
      schedule in interface IStrategy
      Parameters:
      cluster - on which executors will be scheduled.
      td - the topology to schedule for.
      Returns:
      result of scheduling (success, failure, or null when interrupted).

    • prepareForScheduling

      protected void prepareForScheduling(Cluster cluster, TopologyDetails topologyDetails)
      Initialize instance variables as the first step in schedule(Cluster, TopologyDetails). This method may be extended by subclasses to initialize additional variables as in ConstraintSolverStrategy.prepareForScheduling(Cluster, TopologyDetails).
      Parameters:
      cluster - on which executors will be scheduled.
      topologyDetails - to be scheduled.

    • setExecSorter

      protected void setExecSorter(IExecSorter execSorter)
      Set the pluggable sorter for ExecutorDetails.
      Parameters:
      execSorter - to use for sorting executorDetails when scheduling.

    • setNodeSorter

      protected void setNodeSorter(INodeSorter nodeSorter)
      Set the pluggable sorter for Nodes.
      Parameters:
      nodeSorter - to use for sorting nodes when scheduling.

    • getMaxStateSearchFromTopoConf

      public static int getMaxStateSearchFromTopoConf(Map<String,Object> topoConf)

    • isOrderByProximity

      public static boolean isOrderByProximity(Map<String,Object> topoConf)

    • checkSchedulingFeasibility

      protected SchedulingResult checkSchedulingFeasibility()
      Check scheduling feasibility for a quick failure as the second step in schedule(Cluster, TopologyDetails). If scheduling is not possible, then return a SchedulingStatus object with a failure status. If fully scheduled then return a successful SchedulingStatus. This method can be extended by subclasses ConstraintSolverStrategy.checkSchedulingFeasibility() to check for additional failure conditions.
      Returns:
      A non-null SchedulingResult to terminate scheduling, otherwise return null to continue scheduling.

    • isExecAssignmentToWorkerValid

      protected boolean isExecAssignmentToWorkerValid(ExecutorDetails exec, WorkerSlot worker)
      Check if the assignment of the executor to the worker is valid. In simple cases, this is simply a check of RasNode.wouldFit(WorkerSlot, ExecutorDetails, TopologyDetails). This method may be extended by subclasses to add additional checks, see ConstraintSolverStrategy.isExecAssignmentToWorkerValid(ExecutorDetails, WorkerSlot).
      Parameters:
      exec - being scheduled.
      worker - on which to schedule.
      Returns:
      true if executor can be assigned to the worker, false otherwise.

    • hostnameToNodes

      public List<RasNode> hostnameToNodes(String hostname)
      hostname to Ids.
      Parameters:
      hostname - the hostname.
      Returns:
      the ids n that node.

    • idToNode

      public RasNode idToNode(String id)
      Find RASNode for specified node id.
      Parameters:
      id - the node/supervisor id to lookup
      Returns:
      a RASNode object

    • scheduleExecutorsOnNodes

      protected SchedulingResult scheduleExecutorsOnNodes(List<ExecutorDetails> orderedExecutors, Iterable<String> sortedNodesIter)
      Try to schedule till successful or till limits (backtrack count or time) have been exceeded.
      Parameters:
      orderedExecutors - Executors sorted in the preferred order cannot be null - note that ackers are isolated at the end.
      sortedNodesIter - Node iterable which may be null.
      Returns:
      SchedulingResult with success attribute set to true or false indicting whether ALL executors were assigned.

    • assignBoundAckersForNewWorkerSlot

      protected int assignBoundAckersForNewWorkerSlot(ExecutorDetails exec, RasNode node, WorkerSlot workerSlot)

      Determine how many bound ackers to put into the given workerSlot. Then try to assign the ackers one by one into this workerSlot upto the calculated maximum required. Return the number of ackers assigned. Return 0 if one of the conditions hold true: 1. No bound ackers are used. 2. This is not first exec assigned to this worker. 3. No ackers could be assigned because of space or exception.

      Parameters:
      exec - being scheduled.
      node - RasNode on which to schedule.
      workerSlot - WorkerSlot on which to schedule.
      Returns:
      Number of ackers assigned.