Package org.apache.flink.compiler.plantranslate

Source Code of org.apache.flink.compiler.plantranslate.NepheleJobGraphGenerator

/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements.  See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership.  The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License.  You may obtain a copy of the License at
*
*     http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/

package org.apache.flink.compiler.plantranslate;

import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.HashMap;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.Map.Entry;

import org.apache.flink.api.common.aggregators.AggregatorRegistry;
import org.apache.flink.api.common.aggregators.AggregatorWithName;
import org.apache.flink.api.common.aggregators.ConvergenceCriterion;
import org.apache.flink.api.common.aggregators.LongSumAggregator;
import org.apache.flink.api.common.cache.DistributedCache;
import org.apache.flink.api.common.cache.DistributedCache.DistributedCacheEntry;
import org.apache.flink.api.common.distributions.DataDistribution;
import org.apache.flink.api.common.typeutils.TypeSerializerFactory;
import org.apache.flink.compiler.CompilerException;
import org.apache.flink.compiler.dag.TempMode;
import org.apache.flink.compiler.plan.BulkIterationPlanNode;
import org.apache.flink.compiler.plan.BulkPartialSolutionPlanNode;
import org.apache.flink.compiler.plan.Channel;
import org.apache.flink.compiler.plan.DualInputPlanNode;
import org.apache.flink.compiler.plan.IterationPlanNode;
import org.apache.flink.compiler.plan.NAryUnionPlanNode;
import org.apache.flink.compiler.plan.NamedChannel;
import org.apache.flink.compiler.plan.OptimizedPlan;
import org.apache.flink.compiler.plan.PlanNode;
import org.apache.flink.compiler.plan.SingleInputPlanNode;
import org.apache.flink.compiler.plan.SinkPlanNode;
import org.apache.flink.compiler.plan.SolutionSetPlanNode;
import org.apache.flink.compiler.plan.SourcePlanNode;
import org.apache.flink.compiler.plan.WorksetIterationPlanNode;
import org.apache.flink.compiler.plan.WorksetPlanNode;
import org.apache.flink.configuration.ConfigConstants;
import org.apache.flink.configuration.Configuration;
import org.apache.flink.configuration.GlobalConfiguration;
import org.apache.flink.runtime.iterative.convergence.WorksetEmptyConvergenceCriterion;
import org.apache.flink.runtime.iterative.task.IterationHeadPactTask;
import org.apache.flink.runtime.iterative.task.IterationIntermediatePactTask;
import org.apache.flink.runtime.iterative.task.IterationSynchronizationSinkTask;
import org.apache.flink.runtime.iterative.task.IterationTailPactTask;
import org.apache.flink.runtime.jobgraph.AbstractJobVertex;
import org.apache.flink.runtime.jobgraph.DistributionPattern;
import org.apache.flink.runtime.jobgraph.InputFormatVertex;
import org.apache.flink.runtime.jobgraph.JobGraph;
import org.apache.flink.runtime.jobgraph.OutputFormatVertex;
import org.apache.flink.runtime.jobmanager.scheduler.SlotSharingGroup;
import org.apache.flink.runtime.operators.CoGroupDriver;
import org.apache.flink.runtime.operators.CoGroupWithSolutionSetFirstDriver;
import org.apache.flink.runtime.operators.CoGroupWithSolutionSetSecondDriver;
import org.apache.flink.runtime.operators.DataSinkTask;
import org.apache.flink.runtime.operators.DataSourceTask;
import org.apache.flink.runtime.operators.DriverStrategy;
import org.apache.flink.runtime.operators.JoinWithSolutionSetFirstDriver;
import org.apache.flink.runtime.operators.JoinWithSolutionSetSecondDriver;
import org.apache.flink.runtime.operators.MatchDriver;
import org.apache.flink.runtime.operators.NoOpDriver;
import org.apache.flink.runtime.operators.RegularPactTask;
import org.apache.flink.runtime.operators.chaining.ChainedDriver;
import org.apache.flink.runtime.operators.shipping.ShipStrategyType;
import org.apache.flink.runtime.operators.util.LocalStrategy;
import org.apache.flink.runtime.operators.util.TaskConfig;
import org.apache.flink.util.Visitor;

/**
* This component translates the optimizer's resulting plan a nephele job graph. The
* translation is a one to one mapping. All decisions are made by the optimizer, this class
* simply creates nephele data structures and descriptions corresponding to the optimizer's
* result.
* <p>
* The basic method of operation is a top down traversal over the plan graph. On the way down, tasks are created
* for the plan nodes, on the way back up, the nodes connect their predecessor.
*/
public class NepheleJobGraphGenerator implements Visitor<PlanNode> {
 
  public static final String MERGE_ITERATION_AUX_TASKS_KEY = "compiler.merge-iteration-aux";
 
  private static final boolean mergeIterationAuxTasks = GlobalConfiguration.getBoolean(MERGE_ITERATION_AUX_TASKS_KEY, false);
 
//  private static final Logger LOG = LoggerFactory.getLogger(NepheleJobGraphGenerator.class);
 
  private static final TaskInChain ALREADY_VISITED_PLACEHOLDER = new TaskInChain(null, null, null);
 
  // ------------------------------------------------------------------------

  private Map<PlanNode, AbstractJobVertex> vertices; // a map from optimizer nodes to nephele vertices
 
  private Map<PlanNode, TaskInChain> chainedTasks; // a map from optimizer nodes to nephele vertices
 
  private Map<IterationPlanNode, IterationDescriptor> iterations;
 
  private List<TaskInChain> chainedTasksInSequence;
 
  private List<AbstractJobVertex> auxVertices; // auxiliary vertices which are added during job graph generation
 
  private final int defaultMaxFan;
 
  private final float defaultSortSpillingThreshold;
 
  private int iterationIdEnumerator = 1;
 
  private IterationPlanNode currentIteration;  // hack: as long as no nesting is possible, remember the enclosing iteration
 
  private SlotSharingGroup sharingGroup;
 
  // ------------------------------------------------------------------------

  /**
   * Creates a new job graph generator that uses the default values for its resource configuration.
   */
  public NepheleJobGraphGenerator() {
    this.defaultMaxFan = ConfigConstants.DEFAULT_SPILLING_MAX_FAN;
    this.defaultSortSpillingThreshold = ConfigConstants.DEFAULT_SORT_SPILLING_THRESHOLD;
  }
 
  public NepheleJobGraphGenerator(Configuration config) {
    this.defaultMaxFan = config.getInteger(ConfigConstants.DEFAULT_SPILLING_MAX_FAN_KEY,
        ConfigConstants.DEFAULT_SPILLING_MAX_FAN);
    this.defaultSortSpillingThreshold = config.getFloat(ConfigConstants.DEFAULT_SORT_SPILLING_THRESHOLD_KEY,
      ConfigConstants.DEFAULT_SORT_SPILLING_THRESHOLD);
  }

  /**
   * Translates a {@link org.apache.flink.compiler.plan.OptimizedPlan} into a
   * {@link org.apache.flink.runtime.jobgraph.JobGraph}.
   * This is an 1-to-1 mapping. No optimization whatsoever is applied.
   *
   * @param program
   *        Optimized PACT plan that is translated into a JobGraph.
   * @return JobGraph generated from PACT plan.
   */
  public JobGraph compileJobGraph(OptimizedPlan program) {
    this.vertices = new HashMap<PlanNode, AbstractJobVertex>();
    this.chainedTasks = new HashMap<PlanNode, TaskInChain>();
    this.chainedTasksInSequence = new ArrayList<TaskInChain>();
    this.auxVertices = new ArrayList<AbstractJobVertex>();
    this.iterations = new HashMap<IterationPlanNode, IterationDescriptor>();
   
    this.sharingGroup = new SlotSharingGroup();
   
    // generate Nephele job graph
    program.accept(this);
   
    // finalize the iterations
    for (IterationDescriptor iteration : this.iterations.values()) {
      if (iteration.getIterationNode() instanceof BulkIterationPlanNode) {
        finalizeBulkIteration(iteration);
      } else if (iteration.getIterationNode() instanceof WorksetIterationPlanNode) {
        finalizeWorksetIteration(iteration);
      } else {
        throw new CompilerException();
      }
    }
   
    // now that the traversal is done, we have the chained tasks write their configs into their
    // parents' configurations
    for (int i = 0; i < this.chainedTasksInSequence.size(); i++) {
      TaskInChain tic = this.chainedTasksInSequence.get(i);
      TaskConfig t = new TaskConfig(tic.getContainingVertex().getConfiguration());
      t.addChainedTask(tic.getChainedTask(), tic.getTaskConfig(), tic.getTaskName());
    }
   
    // create the jobgraph object
    JobGraph graph = new JobGraph(program.getJobName());
    graph.setNumberOfExecutionRetries(program.getOriginalPactPlan().getNumberOfExecutionRetries());
    graph.setAllowQueuedScheduling(false);
   
    // add vertices to the graph
    for (AbstractJobVertex vertex : this.vertices.values()) {
      graph.addVertex(vertex);
    }
   
    for (AbstractJobVertex vertex : this.auxVertices) {
      graph.addVertex(vertex);
      vertex.setSlotSharingGroup(sharingGroup);
    }
   
    // add registered cache file into job configuration
    for (Entry<String, DistributedCacheEntry> e : program.getOriginalPactPlan().getCachedFiles()) {
      DistributedCache.writeFileInfoToConfig(e.getKey(), e.getValue(), graph.getJobConfiguration());
    }
   
    // release all references again
    this.vertices = null;
    this.chainedTasks = null;
    this.chainedTasksInSequence = null;
    this.auxVertices = null;
    this.iterations = null;

    // return job graph
    return graph;
  }
 
  /**
   * This methods implements the pre-visiting during a depth-first traversal. It create the job vertex and
   * sets local strategy.
   *
   * @param node
   *        The node that is currently processed.
   * @return True, if the visitor should descend to the node's children, false if not.
   * @see org.apache.flink.util.Visitor#preVisit(org.apache.flink.util.Visitable)
   */
  @Override
  public boolean preVisit(PlanNode node) {
    // check if we have visited this node before. in non-tree graphs, this happens
    if (this.vertices.containsKey(node) || this.chainedTasks.containsKey(node) || this.iterations.containsKey(node)) {
      // return false to prevent further descend
      return false;
    }

    // the vertex to be created for the current node
    final AbstractJobVertex vertex;
    try {
      if (node instanceof SinkPlanNode) {
        vertex = createDataSinkVertex((SinkPlanNode) node);
      }
      else if (node instanceof SourcePlanNode) {
        vertex = createDataSourceVertex((SourcePlanNode) node);
      }
      else if (node instanceof BulkIterationPlanNode) {
        BulkIterationPlanNode iterationNode = (BulkIterationPlanNode) node;
        // for the bulk iteration, we skip creating anything for now. we create the graph
        // for the step function in the post visit.
       
        // check that the root of the step function has the same DOP as the iteration.
        // because the tail must have the same DOP as the head, we can only merge the last
        // operator with the tail, if they have the same DOP. not merging is currently not
        // implemented
        PlanNode root = iterationNode.getRootOfStepFunction();
        if (root.getDegreeOfParallelism() != node.getDegreeOfParallelism())
        {
          throw new CompilerException("Error: The final operator of the step " +
              "function has a different degree of parallelism than the iteration operator itself.");
        }
       
        IterationDescriptor descr = new IterationDescriptor(iterationNode, this.iterationIdEnumerator++);
        this.iterations.put(iterationNode, descr);
        vertex = null;
      }
      else if (node instanceof WorksetIterationPlanNode) {
        WorksetIterationPlanNode iterationNode = (WorksetIterationPlanNode) node;

        // we have the same constraints as for the bulk iteration
        PlanNode nextWorkSet = iterationNode.getNextWorkSetPlanNode();
        PlanNode solutionSetDelta  = iterationNode.getSolutionSetDeltaPlanNode();
       
        if (nextWorkSet.getDegreeOfParallelism() != node.getDegreeOfParallelism())
        {
          throw new CompilerException("It is currently not supported that the final operator of the step " +
              "function has a different degree of parallelism than the iteration operator itself.");
        }
        if (solutionSetDelta.getDegreeOfParallelism() != node.getDegreeOfParallelism())
        {
          throw new CompilerException("It is currently not supported that the final operator of the step " +
              "function has a different degree of parallelism than the iteration operator itself.");
        }
       
        IterationDescriptor descr = new IterationDescriptor(iterationNode, this.iterationIdEnumerator++);
        this.iterations.put(iterationNode, descr);
        vertex = null;
      }
      else if (node instanceof SingleInputPlanNode) {
        vertex = createSingleInputVertex((SingleInputPlanNode) node);
      }
      else if (node instanceof DualInputPlanNode) {
        vertex = createDualInputVertex((DualInputPlanNode) node);
      }
      else if (node instanceof NAryUnionPlanNode) {
        // skip the union for now
        vertex = null;
      }
      else if (node instanceof BulkPartialSolutionPlanNode) {
        // create a head node (or not, if it is merged into its successor)
        vertex = createBulkIterationHead((BulkPartialSolutionPlanNode) node);
      }
      else if (node instanceof SolutionSetPlanNode) {
        // this represents an access into the solution set index.
        // we do not create a vertex for the solution set here (we create the head at the workset place holder)
       
        // we adjust the joins / cogroups that go into the solution set here
        for (Channel c : node.getOutgoingChannels()) {
          DualInputPlanNode target = (DualInputPlanNode) c.getTarget();
          AbstractJobVertex accessingVertex = this.vertices.get(target);
          TaskConfig conf = new TaskConfig(accessingVertex.getConfiguration());
          int inputNum = c == target.getInput1() ? 0 : c == target.getInput2() ? 1 : -1;
         
          // sanity checks
          if (inputNum == -1) {
            throw new CompilerException();
          }
         
          // adjust the driver
          if (conf.getDriver().equals(MatchDriver.class)) {
            conf.setDriver(inputNum == 0 ? JoinWithSolutionSetFirstDriver.class : JoinWithSolutionSetSecondDriver.class);
          }
          else if (conf.getDriver().equals(CoGroupDriver.class)) {
            conf.setDriver(inputNum == 0 ? CoGroupWithSolutionSetFirstDriver.class : CoGroupWithSolutionSetSecondDriver.class);
          }
          else {
            throw new CompilerException("Found join with solution set using incompatible operator (only Join/CoGroup are valid).");
          }
        }
       
        // make sure we do not visit this node again. for that, we add a 'already seen' entry into one of the sets
        this.chainedTasks.put(node, ALREADY_VISITED_PLACEHOLDER);
       
        vertex = null;
      }
      else if (node instanceof WorksetPlanNode) {
        // create the iteration head here
        vertex = createWorksetIterationHead((WorksetPlanNode) node);
      }
      else {
        throw new CompilerException("Unrecognized node type: " + node.getClass().getName());
      }
    }
    catch (Exception e) {
      throw new CompilerException("Error translating node '" + node + "': " + e.getMessage(), e);
    }
   
    // check if a vertex was created, or if it was chained or skipped
    if (vertex != null) {
      // set degree of parallelism
      int pd = node.getDegreeOfParallelism();
      vertex.setParallelism(pd);
     
      vertex.setSlotSharingGroup(sharingGroup);
     
      // check whether this vertex is part of an iteration step function
      if (this.currentIteration != null) {
        // check that the task has the same DOP as the iteration as such
        PlanNode iterationNode = (PlanNode) this.currentIteration;
        if (iterationNode.getDegreeOfParallelism() < pd) {
          throw new CompilerException("Error: All functions that are part of an iteration must have the same, or a lower, degree-of-parallelism than the iteration operator.");
        }

        // store the id of the iterations the step functions participate in
        IterationDescriptor descr = this.iterations.get(this.currentIteration);
        new TaskConfig(vertex.getConfiguration()).setIterationId(descr.getId());
      }
 
      // store in the map
      this.vertices.put(node, vertex);
    }

    // returning true causes deeper descend
    return true;
  }

  /**
   * This method implements the post-visit during the depth-first traversal. When the post visit happens,
   * all of the descendants have been processed, so this method connects all of the current node's
   * predecessors to the current node.
   *
   * @param node
   *        The node currently processed during the post-visit.
   * @see org.apache.flink.util.Visitor#postVisit(org.apache.flink.util.Visitable) t
   */
  @Override
  public void postVisit(PlanNode node) {
    try {
      // --------- check special cases for which we handle post visit differently ----------
     
      // skip data source node (they have no inputs)
      // also, do nothing for union nodes, we connect them later when gathering the inputs for a task
      // solution sets have no input. the initial solution set input is connected when the iteration node is in its postVisit
      if (node instanceof SourcePlanNode || node instanceof NAryUnionPlanNode || node instanceof SolutionSetPlanNode) {
        return;
      }
     
      // check if we have an iteration. in that case, translate the step function now
      if (node instanceof IterationPlanNode) {
        // for now, prevent nested iterations
        if (this.currentIteration != null) {
          throw new CompilerException("Nested Iterations are not possible at the moment!");
        }
        this.currentIteration = (IterationPlanNode) node;
        this.currentIteration.acceptForStepFunction(this);
        this.currentIteration = null;
       
        // inputs for initial bulk partial solution or initial workset are already connected to the iteration head in the head's post visit.
        // connect the initial solution set now.
        if (node instanceof WorksetIterationPlanNode) {
          // connect the initial solution set
          WorksetIterationPlanNode wsNode = (WorksetIterationPlanNode) node;
          AbstractJobVertex headVertex = this.iterations.get(wsNode).getHeadTask();
          TaskConfig headConfig = new TaskConfig(headVertex.getConfiguration());
          int inputIndex = headConfig.getDriverStrategy().getNumInputs();
          headConfig.setIterationHeadSolutionSetInputIndex(inputIndex);
          translateChannel(wsNode.getInitialSolutionSetInput(), inputIndex, headVertex, headConfig, false);
        }
       
        return;
      }
     
      final AbstractJobVertex targetVertex = this.vertices.get(node);
     
     
      // --------- Main Path: Translation of channels ----------
      //
      // There are two paths of translation: One for chained tasks (or merged tasks in general),
      // which do not have their own task vertex. The other for tasks that have their own vertex,
      // or are the primary task in a vertex (to which the others are chained).
     
      // check whether this node has its own task, or is merged with another one
      if (targetVertex == null) {
        // node's task is merged with another task. it is either chained, of a merged head vertex
        // from an iteration
        final TaskInChain chainedTask;
        if ((chainedTask = this.chainedTasks.get(node)) != null) {
          // Chained Task. Sanity check first...
          final Iterator<Channel> inConns = node.getInputs().iterator();
          if (!inConns.hasNext()) {
            throw new CompilerException("Bug: Found chained task with no input.");
          }
          final Channel inConn = inConns.next();
         
          if (inConns.hasNext()) {
            throw new CompilerException("Bug: Found a chained task with more than one input!");
          }
          if (inConn.getLocalStrategy() != null && inConn.getLocalStrategy() != LocalStrategy.NONE) {
            throw new CompilerException("Bug: Found a chained task with an input local strategy.");
          }
          if (inConn.getShipStrategy() != null && inConn.getShipStrategy() != ShipStrategyType.FORWARD) {
            throw new CompilerException("Bug: Found a chained task with an input ship strategy other than FORWARD.");
          }
 
          AbstractJobVertex container = chainedTask.getContainingVertex();
         
          if (container == null) {
            final PlanNode sourceNode = inConn.getSource();
            container = this.vertices.get(sourceNode);
            if (container == null) {
              // predecessor is itself chained
              container = this.chainedTasks.get(sourceNode).getContainingVertex();
              if (container == null) {
                throw new IllegalStateException("Bug: Chained task predecessor has not been assigned its containing vertex.");
              }
            } else {
              // predecessor is a proper task job vertex and this is the first chained task. add a forward connection entry.
              new TaskConfig(container.getConfiguration()).addOutputShipStrategy(ShipStrategyType.FORWARD);
            }
            chainedTask.setContainingVertex(container);
          }
         
          // add info about the input serializer type
          chainedTask.getTaskConfig().setInputSerializer(inConn.getSerializer(), 0);
         
          // update name of container task
          String containerTaskName = container.getName();
          if(containerTaskName.startsWith("CHAIN ")) {
            container.setName(containerTaskName+" -> "+chainedTask.getTaskName());
          } else {
            container.setName("CHAIN "+containerTaskName+" -> "+chainedTask.getTaskName());
          }
         
          this.chainedTasksInSequence.add(chainedTask);
          return;
        }
        else if (node instanceof BulkPartialSolutionPlanNode ||
            node instanceof WorksetPlanNode)
        {
          // merged iteration head task. the task that the head is merged with will take care of it
          return;
        } else {
          throw new CompilerException("Bug: Unrecognized merged task vertex.");
        }
      }
     
      // -------- Here, we translate non-chained tasks -------------
     
     
      if (this.currentIteration != null) {
        AbstractJobVertex head = this.iterations.get(this.currentIteration).getHeadTask();
        // the head may still be null if we descend into the static parts first
        if (head != null) {
          targetVertex.setStrictlyCoLocatedWith(head);
        }
      }
     
     
      // create the config that will contain all the description of the inputs
      final TaskConfig targetVertexConfig = new TaskConfig(targetVertex.getConfiguration());
           
      // get the inputs. if this node is the head of an iteration, we obtain the inputs from the
      // enclosing iteration node, because the inputs are the initial inputs to the iteration.
      final Iterator<Channel> inConns;
      if (node instanceof BulkPartialSolutionPlanNode) {
        inConns = ((BulkPartialSolutionPlanNode) node).getContainingIterationNode().getInputs().iterator();
        // because the partial solution has its own vertex, is has only one (logical) input.
        // note this in the task configuration
        targetVertexConfig.setIterationHeadPartialSolutionOrWorksetInputIndex(0);
      } else if (node instanceof WorksetPlanNode) {
        WorksetPlanNode wspn = (WorksetPlanNode) node;
        // input that is the initial workset
        inConns = Collections.singleton(wspn.getContainingIterationNode().getInput2()).iterator();
       
        // because we have a stand-alone (non-merged) workset iteration head, the initial workset will
        // be input 0 and the solution set will be input 1
        targetVertexConfig.setIterationHeadPartialSolutionOrWorksetInputIndex(0);
        targetVertexConfig.setIterationHeadSolutionSetInputIndex(1);
      } else {
        inConns = node.getInputs().iterator();
      }
      if (!inConns.hasNext()) {
        throw new CompilerException("Bug: Found a non-source task with no input.");
      }
     
      int inputIndex = 0;
      while (inConns.hasNext()) {
        Channel input = inConns.next();
        inputIndex += translateChannel(input, inputIndex, targetVertex, targetVertexConfig, false);
      }
      // broadcast variables
      int broadcastInputIndex = 0;
      for (NamedChannel broadcastInput: node.getBroadcastInputs()) {
        int broadcastInputIndexDelta = translateChannel(broadcastInput, broadcastInputIndex, targetVertex, targetVertexConfig, true);
        targetVertexConfig.setBroadcastInputName(broadcastInput.getName(), broadcastInputIndex);
        targetVertexConfig.setBroadcastInputSerializer(broadcastInput.getSerializer(), broadcastInputIndex);
        broadcastInputIndex += broadcastInputIndexDelta;
      }
    } catch (Exception e) {
      throw new CompilerException(
        "An error occurred while translating the optimized plan to a nephele JobGraph: " + e.getMessage(), e);
    }
  }
 
  private int translateChannel(Channel input, int inputIndex, AbstractJobVertex targetVertex,
      TaskConfig targetVertexConfig, boolean isBroadcast) throws Exception
  {
    final PlanNode inputPlanNode = input.getSource();
    final Iterator<Channel> allInChannels;
   
    if (inputPlanNode instanceof NAryUnionPlanNode) {
      allInChannels = ((NAryUnionPlanNode) inputPlanNode).getListOfInputs().iterator();
    }
    else if (inputPlanNode instanceof BulkPartialSolutionPlanNode) {
      if (this.vertices.get(inputPlanNode) == null) {
        // merged iteration head
        final BulkPartialSolutionPlanNode pspn = (BulkPartialSolutionPlanNode) inputPlanNode;
        final BulkIterationPlanNode iterationNode = pspn.getContainingIterationNode();
       
        // check if the iteration's input is a union
        if (iterationNode.getInput().getSource() instanceof NAryUnionPlanNode) {
          allInChannels = ((NAryUnionPlanNode) iterationNode.getInput().getSource()).getInputs().iterator();
        } else {
          allInChannels = Collections.singletonList(iterationNode.getInput()).iterator();
        }
       
        // also, set the index of the gate with the partial solution
        targetVertexConfig.setIterationHeadPartialSolutionOrWorksetInputIndex(inputIndex);
      } else {
        // standalone iteration head
        allInChannels = Collections.singletonList(input).iterator();
      }
    } else if (inputPlanNode instanceof WorksetPlanNode) {
      if (this.vertices.get(inputPlanNode) == null) {
        // merged iteration head
        final WorksetPlanNode wspn = (WorksetPlanNode) inputPlanNode;
        final WorksetIterationPlanNode iterationNode = wspn.getContainingIterationNode();
       
        // check if the iteration's input is a union
        if (iterationNode.getInput2().getSource() instanceof NAryUnionPlanNode) {
          allInChannels = ((NAryUnionPlanNode) iterationNode.getInput2().getSource()).getInputs().iterator();
        } else {
          allInChannels = Collections.singletonList(iterationNode.getInput2()).iterator();
        }
       
        // also, set the index of the gate with the partial solution
        targetVertexConfig.setIterationHeadPartialSolutionOrWorksetInputIndex(inputIndex);
      } else {
        // standalone iteration head
        allInChannels = Collections.singletonList(input).iterator();
      }
    } else if (inputPlanNode instanceof SolutionSetPlanNode) {
      // for now, skip connections with the solution set node, as this is a local index access (later to be parameterized here)
      // rather than a vertex connection
      return 0;
    } else {
      allInChannels = Collections.singletonList(input).iterator();
    }
   
    // check that the type serializer is consistent
    TypeSerializerFactory<?> typeSerFact = null;
   
    // accounting for channels on the dynamic path
    int numChannelsTotal = 0;
    int numChannelsDynamicPath = 0;
    int numDynamicSenderTasksTotal = 0;
   

    // expand the channel to all the union channels, in case there is a union operator at its source
    while (allInChannels.hasNext()) {
      final Channel inConn = allInChannels.next();
     
      // sanity check the common serializer
      if (typeSerFact == null) {
        typeSerFact = inConn.getSerializer();
      } else if (!typeSerFact.equals(inConn.getSerializer())) {
        throw new CompilerException("Conflicting types in union operator.");
      }
     
      final PlanNode sourceNode = inConn.getSource();
      AbstractJobVertex sourceVertex = this.vertices.get(sourceNode);
      TaskConfig sourceVertexConfig;

      if (sourceVertex == null) {
        // this predecessor is chained to another task or an iteration
        final TaskInChain chainedTask;
        final IterationDescriptor iteration;
        if ((chainedTask = this.chainedTasks.get(sourceNode)) != null) {
          // push chained task
          if (chainedTask.getContainingVertex() == null) {
            throw new IllegalStateException("Bug: Chained task has not been assigned its containing vertex when connecting.");
          }
          sourceVertex = chainedTask.getContainingVertex();
          sourceVertexConfig = chainedTask.getTaskConfig();
        } else if ((iteration = this.iterations.get(sourceNode)) != null) {
          // predecessor is an iteration
          sourceVertex = iteration.getHeadTask();
          sourceVertexConfig = iteration.getHeadFinalResultConfig();
        } else {
          throw new CompilerException("Bug: Could not resolve source node for a channel.");
        }
      } else {
        // predecessor is its own vertex
        sourceVertexConfig = new TaskConfig(sourceVertex.getConfiguration());
      }
      DistributionPattern pattern = connectJobVertices(
        inConn, inputIndex, sourceVertex, sourceVertexConfig, targetVertex, targetVertexConfig, isBroadcast);
     
      // accounting on channels and senders
      numChannelsTotal++;
      if (inConn.isOnDynamicPath()) {
        numChannelsDynamicPath++;
        numDynamicSenderTasksTotal += getNumberOfSendersPerReceiver(pattern,
          sourceVertex.getParallelism(), targetVertex.getParallelism());
      }
    }
   
    // for the iterations, check that the number of dynamic channels is the same as the number
    // of channels for this logical input. this condition is violated at the moment, if there
    // is a union between nodes on the static and nodes on the dynamic path
    if (numChannelsDynamicPath > 0 && numChannelsTotal != numChannelsDynamicPath) {
      throw new CompilerException("Error: It is currently not supported to union between dynamic and static path in an iteration.");
    }
    if (numDynamicSenderTasksTotal > 0) {
      if (isBroadcast) {
        targetVertexConfig.setBroadcastGateIterativeWithNumberOfEventsUntilInterrupt(inputIndex, numDynamicSenderTasksTotal);
      } else {
        targetVertexConfig.setGateIterativeWithNumberOfEventsUntilInterrupt(inputIndex, numDynamicSenderTasksTotal);
      }
    }
   
    // the local strategy is added only once. in non-union case that is the actual edge,
    // in the union case, it is the edge between union and the target node
    addLocalInfoFromChannelToConfig(input, targetVertexConfig, inputIndex, isBroadcast);
    return 1;
  }
 
  private int getNumberOfSendersPerReceiver(DistributionPattern pattern, int numSenders, int numReceivers) {
    if (pattern == DistributionPattern.BIPARTITE) {
      return numSenders;
    } else if (pattern == DistributionPattern.POINTWISE) {
      if (numSenders != numReceivers) {
        if (numReceivers == 1) {
          return numSenders;
        }
        else if (numSenders == 1) {
          return 1;
        }
        else {
          throw new CompilerException("Error: A changing degree of parallelism is currently " +
              "not supported between tasks within an iteration.");
        }
      } else {
        return 1;
      }
    } else {
      throw new CompilerException("Unknown distribution pattern for channels: " + pattern);
    }
  }
 
  // ------------------------------------------------------------------------
  // Methods for creating individual vertices
  // ------------------------------------------------------------------------
 
  private AbstractJobVertex createSingleInputVertex(SingleInputPlanNode node) throws CompilerException {
    final String taskName = node.getNodeName();
    final DriverStrategy ds = node.getDriverStrategy();
   
    // check, whether chaining is possible
    boolean chaining = false;
    {
      Channel inConn = node.getInput();
      PlanNode pred = inConn.getSource();
      chaining = ds.getPushChainDriverClass() != null &&
          !(pred instanceof NAryUnionPlanNode) &&  // first op after union is stand-alone, because union is merged
          !(pred instanceof BulkPartialSolutionPlanNode) &&  // partial solution merges anyways
          !(pred instanceof WorksetPlanNode) &&  // workset merges anyways
          !(pred instanceof IterationPlanNode) && // cannot chain with iteration heads currently
          inConn.getShipStrategy() == ShipStrategyType.FORWARD &&
          inConn.getLocalStrategy() == LocalStrategy.NONE &&
          pred.getOutgoingChannels().size() == 1 &&
          node.getDegreeOfParallelism() == pred.getDegreeOfParallelism() &&
          node.getBroadcastInputs().isEmpty();
     
      // cannot chain the nodes that produce the next workset or the next solution set, if they are not the
      // in a tail
      if (this.currentIteration != null && this.currentIteration instanceof WorksetIterationPlanNode &&
          node.getOutgoingChannels().size() > 0)
      {
        WorksetIterationPlanNode wspn = (WorksetIterationPlanNode) this.currentIteration;
        if (wspn.getSolutionSetDeltaPlanNode() == pred || wspn.getNextWorkSetPlanNode() == pred) {
          chaining = false;
        }
      }
      // cannot chain the nodes that produce the next workset in a bulk iteration if a termination criterion follows
      if (this.currentIteration != null && this.currentIteration instanceof BulkIterationPlanNode)
      {
        BulkIterationPlanNode wspn = (BulkIterationPlanNode) this.currentIteration;
        if (node == wspn.getRootOfTerminationCriterion() && wspn.getRootOfStepFunction() == pred){
          chaining = false;
        }else if(node.getOutgoingChannels().size() > 0 &&(wspn.getRootOfStepFunction() == pred ||
            wspn.getRootOfTerminationCriterion() == pred)) {
          chaining = false;
        }
      }
    }
   
    final AbstractJobVertex vertex;
    final TaskConfig config;
   
    if (chaining) {
      vertex = null;
      config = new TaskConfig(new Configuration());
      this.chainedTasks.put(node, new TaskInChain(ds.getPushChainDriverClass(), config, taskName));
    } else {
      // create task vertex
      vertex = new AbstractJobVertex(taskName);
      vertex.setInvokableClass((this.currentIteration != null && node.isOnDynamicPath()) ? IterationIntermediatePactTask.class : RegularPactTask.class);
     
      config = new TaskConfig(vertex.getConfiguration());
      config.setDriver(ds.getDriverClass());
    }
   
    // set user code
    config.setStubWrapper(node.getPactContract().getUserCodeWrapper());
    config.setStubParameters(node.getPactContract().getParameters());
   
    // set the driver strategy
    config.setDriverStrategy(ds);
    for(int i=0;i<ds.getNumRequiredComparators();i++) {
      config.setDriverComparator(node.getComparator(i), i);
    }
    // assign memory, file-handles, etc.
    assignDriverResources(node, config);
    return vertex;
  }

  private AbstractJobVertex createDualInputVertex(DualInputPlanNode node) throws CompilerException {
    final String taskName = node.getNodeName();
    final DriverStrategy ds = node.getDriverStrategy();
    final AbstractJobVertex vertex = new AbstractJobVertex(taskName);
    final TaskConfig config = new TaskConfig(vertex.getConfiguration());
    vertex.setInvokableClass( (this.currentIteration != null && node.isOnDynamicPath()) ? IterationIntermediatePactTask.class : RegularPactTask.class);
   
    // set user code
    config.setStubWrapper(node.getPactContract().getUserCodeWrapper());
    config.setStubParameters(node.getPactContract().getParameters());
   
    // set the driver strategy
    config.setDriver(ds.getDriverClass());
    config.setDriverStrategy(ds);
    if (node.getComparator1() != null) {
      config.setDriverComparator(node.getComparator1(), 0);
    }
    if (node.getComparator2() != null) {
      config.setDriverComparator(node.getComparator2(), 1);
    }
    if (node.getPairComparator() != null) {
      config.setDriverPairComparator(node.getPairComparator());
    }
   
    // assign memory, file-handles, etc.
    assignDriverResources(node, config);
    return vertex;
  }

  private InputFormatVertex createDataSourceVertex(SourcePlanNode node) throws CompilerException {
    final InputFormatVertex vertex = new InputFormatVertex(node.getNodeName());
    final TaskConfig config = new TaskConfig(vertex.getConfiguration());

    vertex.setInvokableClass(DataSourceTask.class);

    // set user code
    config.setStubWrapper(node.getPactContract().getUserCodeWrapper());
    config.setStubParameters(node.getPactContract().getParameters());

    config.setOutputSerializer(node.getSerializer());
    return vertex;
  }

  private AbstractJobVertex createDataSinkVertex(SinkPlanNode node) throws CompilerException {
    final OutputFormatVertex vertex = new OutputFormatVertex(node.getNodeName());
    final TaskConfig config = new TaskConfig(vertex.getConfiguration());

    vertex.setInvokableClass(DataSinkTask.class);
    vertex.getConfiguration().setInteger(DataSinkTask.DEGREE_OF_PARALLELISM_KEY, node.getDegreeOfParallelism());

    // set user code
    config.setStubWrapper(node.getPactContract().getUserCodeWrapper());
    config.setStubParameters(node.getPactContract().getParameters());

    return vertex;
  }
 
  private AbstractJobVertex createBulkIterationHead(BulkPartialSolutionPlanNode pspn) {
    // get the bulk iteration that corresponds to this partial solution node
    final BulkIterationPlanNode iteration = pspn.getContainingIterationNode();
   
    // check whether we need an individual vertex for the partial solution, or whether we
    // attach ourselves to the vertex of the parent node. We can combine the head with a node of
    // the step function, if
    // 1) There is one parent that the partial solution connects to via a forward pattern and no
    //    local strategy
    // 2) DOP and the number of subtasks per instance does not change
    // 3) That successor is not a union
    // 4) That successor is not itself the last node of the step function
    // 5) There is no local strategy on the edge for the initial partial solution, as
    //    this translates to a local strategy that would only be executed in the first iteration
   
    final boolean merge;
    if (mergeIterationAuxTasks && pspn.getOutgoingChannels().size() == 1) {
      final Channel c = pspn.getOutgoingChannels().get(0);
      final PlanNode successor = c.getTarget();
      merge = c.getShipStrategy() == ShipStrategyType.FORWARD &&
          c.getLocalStrategy() == LocalStrategy.NONE &&
          c.getTempMode() == TempMode.NONE &&
          successor.getDegreeOfParallelism() == pspn.getDegreeOfParallelism() &&
          !(successor instanceof NAryUnionPlanNode) &&
          successor != iteration.getRootOfStepFunction() &&
          iteration.getInput().getLocalStrategy() == LocalStrategy.NONE;
    } else {
      merge = false;
    }
   
    // create or adopt the head vertex
    final AbstractJobVertex toReturn;
    final AbstractJobVertex headVertex;
    final TaskConfig headConfig;
    if (merge) {
      final PlanNode successor = pspn.getOutgoingChannels().get(0).getTarget();
      headVertex = (AbstractJobVertex) this.vertices.get(successor);
     
      if (headVertex == null) {
        throw new CompilerException(
          "Bug: Trying to merge solution set with its sucessor, but successor has not been created.");
      }
     
      // reset the vertex type to iteration head
      headVertex.setInvokableClass(IterationHeadPactTask.class);
      headConfig = new TaskConfig(headVertex.getConfiguration());
      toReturn = null;
    } else {
      // instantiate the head vertex and give it a no-op driver as the driver strategy.
      // everything else happens in the post visit, after the input (the initial partial solution)
      // is connected.
      headVertex = new AbstractJobVertex("PartialSolution ("+iteration.getNodeName()+")");
      headVertex.setInvokableClass(IterationHeadPactTask.class);
      headConfig = new TaskConfig(headVertex.getConfiguration());
      headConfig.setDriver(NoOpDriver.class);
      toReturn = headVertex;
    }
   
    // create the iteration descriptor and the iteration to it
    IterationDescriptor descr = this.iterations.get(iteration);
    if (descr == null) {
      throw new CompilerException("Bug: Iteration descriptor was not created at when translating the iteration node.");
    }
    descr.setHeadTask(headVertex, headConfig);
   
    return toReturn;
  }
 
  private AbstractJobVertex createWorksetIterationHead(WorksetPlanNode wspn) {
    // get the bulk iteration that corresponds to this partial solution node
    final WorksetIterationPlanNode iteration = wspn.getContainingIterationNode();
   
    // check whether we need an individual vertex for the partial solution, or whether we
    // attach ourselves to the vertex of the parent node. We can combine the head with a node of
    // the step function, if
    // 1) There is one parent that the partial solution connects to via a forward pattern and no
    //    local strategy
    // 2) DOP and the number of subtasks per instance does not change
    // 3) That successor is not a union
    // 4) That successor is not itself the last node of the step function
    // 5) There is no local strategy on the edge for the initial workset, as
    //    this translates to a local strategy that would only be executed in the first superstep
   
    final boolean merge;
    if (mergeIterationAuxTasks && wspn.getOutgoingChannels().size() == 1) {
      final Channel c = wspn.getOutgoingChannels().get(0);
      final PlanNode successor = c.getTarget();
      merge = c.getShipStrategy() == ShipStrategyType.FORWARD &&
          c.getLocalStrategy() == LocalStrategy.NONE &&
          c.getTempMode() == TempMode.NONE &&
          successor.getDegreeOfParallelism() == wspn.getDegreeOfParallelism() &&
          !(successor instanceof NAryUnionPlanNode) &&
          successor != iteration.getNextWorkSetPlanNode() &&
          iteration.getInitialWorksetInput().getLocalStrategy() == LocalStrategy.NONE;
    } else {
      merge = false;
    }
   
    // create or adopt the head vertex
    final AbstractJobVertex toReturn;
    final AbstractJobVertex headVertex;
    final TaskConfig headConfig;
    if (merge) {
      final PlanNode successor = wspn.getOutgoingChannels().get(0).getTarget();
      headVertex = (AbstractJobVertex) this.vertices.get(successor);
     
      if (headVertex == null) {
        throw new CompilerException(
          "Bug: Trying to merge solution set with its sucessor, but successor has not been created.");
      }
     
      // reset the vertex type to iteration head
      headVertex.setInvokableClass(IterationHeadPactTask.class);
      headConfig = new TaskConfig(headVertex.getConfiguration());
      toReturn = null;
    } else {
      // instantiate the head vertex and give it a no-op driver as the driver strategy.
      // everything else happens in the post visit, after the input (the initial partial solution)
      // is connected.
      headVertex = new AbstractJobVertex("IterationHead("+iteration.getNodeName()+")");
      headVertex.setInvokableClass(IterationHeadPactTask.class);
      headConfig = new TaskConfig(headVertex.getConfiguration());
      headConfig.setDriver(NoOpDriver.class);
      toReturn = headVertex;
    }
   
    headConfig.setSolutionSetUnmanaged(iteration.getIterationNode().getIterationContract().isSolutionSetUnManaged());
   
    // create the iteration descriptor and the iteration to it
    IterationDescriptor descr = this.iterations.get(iteration);
    if (descr == null) {
      throw new CompilerException("Bug: Iteration descriptor was not created at when translating the iteration node.");
    }
    descr.setHeadTask(headVertex, headConfig);
   
    return toReturn;
  }
 
  private void assignDriverResources(PlanNode node, TaskConfig config) {
    final double relativeMem = node.getRelativeMemoryPerSubTask();
    if (relativeMem > 0) {
      config.setRelativeMemoryDriver(relativeMem);
      config.setFilehandlesDriver(this.defaultMaxFan);
      config.setSpillingThresholdDriver(this.defaultSortSpillingThreshold);
    }
  }
 
  private void assignLocalStrategyResources(Channel c, TaskConfig config, int inputNum) {
    if (c.getRelativeMemoryLocalStrategy() > 0) {
      config.setRelativeMemoryInput(inputNum, c.getRelativeMemoryLocalStrategy());
      config.setFilehandlesInput(inputNum, this.defaultMaxFan);
      config.setSpillingThresholdInput(inputNum, this.defaultSortSpillingThreshold);
    }
  }

  // ------------------------------------------------------------------------
  // Connecting Vertices
  // ------------------------------------------------------------------------

  /**
   * NOTE: The channel for global and local strategies are different if we connect a union. The global strategy
   * channel is then the channel into the union node, the local strategy channel the one from the union to the
   * actual target operator.
   *
   * @param channel
   * @param inputNumber
   * @param sourceVertex
   * @param sourceConfig
   * @param targetVertex
   * @param targetConfig
   * @param isBroadcast
   * @throws CompilerException
   */
  private DistributionPattern connectJobVertices(Channel channel, int inputNumber,
      final AbstractJobVertex sourceVertex, final TaskConfig sourceConfig,
      final AbstractJobVertex targetVertex, final TaskConfig targetConfig, boolean isBroadcast)
  throws CompilerException
  {
    // ------------ connect the vertices to the job graph --------------
    final DistributionPattern distributionPattern;

    switch (channel.getShipStrategy()) {
      case FORWARD:
        distributionPattern = DistributionPattern.POINTWISE;
        break;
      case PARTITION_RANDOM:
      case BROADCAST:
      case PARTITION_HASH:
      case PARTITION_RANGE:
      case PARTITION_FORCED_REBALANCE:
        distributionPattern = DistributionPattern.BIPARTITE;
        break;
      default:
        throw new RuntimeException("Unknown runtime ship strategy: " + channel.getShipStrategy());
    }
   
    targetVertex.connectNewDataSetAsInput(sourceVertex, distributionPattern);
   
//    sourceVertex.conn/ectTo(targetVertex, channelType, distributionPattern);

    // -------------- configure the source task's ship strategy strategies in task config --------------
    final int outputIndex = sourceConfig.getNumOutputs();
    sourceConfig.addOutputShipStrategy(channel.getShipStrategy());
    if (outputIndex == 0) {
      sourceConfig.setOutputSerializer(channel.getSerializer());
    }
    if (channel.getShipStrategyComparator() != null) {
      sourceConfig.setOutputComparator(channel.getShipStrategyComparator(), outputIndex);
    }
   
    if (channel.getShipStrategy() == ShipStrategyType.PARTITION_RANGE) {
     
      final DataDistribution dataDistribution = channel.getDataDistribution();
      if(dataDistribution != null) {
        sourceConfig.setOutputDataDistribution(dataDistribution, outputIndex);
      } else {
        throw new RuntimeException("Range partitioning requires data distribution");
        // TODO: inject code and configuration for automatic histogram generation
      }
    }
//    if (targetContract instanceof GenericDataSink) {
//      final DataDistribution distri = ((GenericDataSink) targetContract).getDataDistribution();
//      if (distri != null) {
//        configForOutputShipStrategy.setOutputDataDistribution(distri);
//      }
//    }
   
    // ---------------- configure the receiver -------------------
    if (isBroadcast) {
      targetConfig.addBroadcastInputToGroup(inputNumber);
    } else {
      targetConfig.addInputToGroup(inputNumber);
    }
    return distributionPattern;
  }
 
  private void addLocalInfoFromChannelToConfig(Channel channel, TaskConfig config, int inputNum, boolean isBroadcastChannel) {
    // serializer
    if (isBroadcastChannel) {
      config.setBroadcastInputSerializer(channel.getSerializer(), inputNum);
     
      if (channel.getLocalStrategy() != LocalStrategy.NONE || (channel.getTempMode() != null && channel.getTempMode() != TempMode.NONE)) {
        throw new CompilerException("Found local strategy or temp mode on a broadcast variable channel.");
      } else {
        return;
      }
    } else {
      config.setInputSerializer(channel.getSerializer(), inputNum);
    }
   
    // local strategy
    if (channel.getLocalStrategy() != LocalStrategy.NONE) {
      config.setInputLocalStrategy(inputNum, channel.getLocalStrategy());
      if (channel.getLocalStrategyComparator() != null) {
        config.setInputComparator(channel.getLocalStrategyComparator(), inputNum);
      }
    }
   
    assignLocalStrategyResources(channel, config, inputNum);
   
    // materialization / caching
    if (channel.getTempMode() != null) {
      final TempMode tm = channel.getTempMode();

      boolean needsMemory = false;
      if (tm.breaksPipeline()) {
        config.setInputAsynchronouslyMaterialized(inputNum, true);
        needsMemory = true;
      }
      if (tm.isCached()) {
        config.setInputCached(inputNum, true);
        needsMemory = true;
      }
     
      if (needsMemory) {
        // sanity check
        if (tm == null || tm == TempMode.NONE || channel.getRelativeTempMemory() <= 0) {
          throw new CompilerException("Bug in compiler: Inconsistent description of input materialization.");
        }
        config.setRelativeInputMaterializationMemory(inputNum, channel.getRelativeTempMemory());
      }
    }
  }
 
  private void finalizeBulkIteration(IterationDescriptor descr) {
   
    final BulkIterationPlanNode bulkNode = (BulkIterationPlanNode) descr.getIterationNode();
    final AbstractJobVertex headVertex = descr.getHeadTask();
    final TaskConfig headConfig = new TaskConfig(headVertex.getConfiguration());
    final TaskConfig headFinalOutputConfig = descr.getHeadFinalResultConfig();
   
    // ------------ finalize the head config with the final outputs and the sync gate ------------
    final int numStepFunctionOuts = headConfig.getNumOutputs();
    final int numFinalOuts = headFinalOutputConfig.getNumOutputs();
   
    if (numStepFunctionOuts == 0) {
      throw new CompilerException("The iteration has no operation inside the step function.");
    }
   
    headConfig.setIterationHeadFinalOutputConfig(headFinalOutputConfig);
    headConfig.setIterationHeadIndexOfSyncOutput(numStepFunctionOuts + numFinalOuts);
    final double relativeMemForBackChannel = bulkNode.getRelativeMemoryPerSubTask();
    if (relativeMemForBackChannel <= 0) {
      throw new CompilerException("Bug: No memory has been assigned to the iteration back channel.");
    }
    headConfig.setRelativeBackChannelMemory(relativeMemForBackChannel);
   
    // --------------------------- create the sync task ---------------------------
    final AbstractJobVertex sync = new AbstractJobVertex("Sync(" + bulkNode.getNodeName() + ")");
    sync.setInvokableClass(IterationSynchronizationSinkTask.class);
    sync.setParallelism(1);
    this.auxVertices.add(sync);
   
    final TaskConfig syncConfig = new TaskConfig(sync.getConfiguration());
    syncConfig.setGateIterativeWithNumberOfEventsUntilInterrupt(0, headVertex.getParallelism());

    // set the number of iteration / convergence criterion for the sync
    final int maxNumIterations = bulkNode.getIterationNode().getIterationContract().getMaximumNumberOfIterations();
    if (maxNumIterations < 1) {
      throw new CompilerException("Cannot create bulk iteration with unspecified maximum number of iterations.");
    }
    syncConfig.setNumberOfIterations(maxNumIterations);
   
    // connect the sync task
    sync.connectNewDataSetAsInput(headVertex, DistributionPattern.POINTWISE);
   
    // ----------------------------- create the iteration tail ------------------------------
   
    final PlanNode rootOfTerminationCriterion = bulkNode.getRootOfTerminationCriterion();
    final PlanNode rootOfStepFunction = bulkNode.getRootOfStepFunction();
    final TaskConfig tailConfig;
   
    AbstractJobVertex rootOfStepFunctionVertex = (AbstractJobVertex) this.vertices.get(rootOfStepFunction);
    if (rootOfStepFunctionVertex == null) {
      // last op is chained
      final TaskInChain taskInChain = this.chainedTasks.get(rootOfStepFunction);
      if (taskInChain == null) {
        throw new CompilerException("Bug: Tail of step function not found as vertex or chained task.");
      }
      rootOfStepFunctionVertex = (AbstractJobVertex) taskInChain.getContainingVertex();

      // the fake channel is statically typed to pact record. no data is sent over this channel anyways.
      tailConfig = taskInChain.getTaskConfig();
    } else {
      tailConfig = new TaskConfig(rootOfStepFunctionVertex.getConfiguration());
    }
   
    tailConfig.setIsWorksetUpdate();
   
    // No following termination criterion
    if (rootOfStepFunction.getOutgoingChannels().isEmpty()) {
     
      rootOfStepFunctionVertex.setInvokableClass(IterationTailPactTask.class);
     
      tailConfig.setOutputSerializer(bulkNode.getSerializerForIterationChannel());
    }
   
   
    // create the fake output task for termination criterion, if needed
    final TaskConfig tailConfigOfTerminationCriterion;
    // If we have a termination criterion and it is not an intermediate node
    if(rootOfTerminationCriterion != null && rootOfTerminationCriterion.getOutgoingChannels().isEmpty()) {
      AbstractJobVertex rootOfTerminationCriterionVertex = (AbstractJobVertex) this.vertices.get(rootOfTerminationCriterion);
     
     
      if (rootOfTerminationCriterionVertex == null) {
        // last op is chained
        final TaskInChain taskInChain = this.chainedTasks.get(rootOfTerminationCriterion);
        if (taskInChain == null) {
          throw new CompilerException("Bug: Tail of termination criterion not found as vertex or chained task.");
        }
        rootOfTerminationCriterionVertex = (AbstractJobVertex) taskInChain.getContainingVertex();

        // the fake channel is statically typed to pact record. no data is sent over this channel anyways.
        tailConfigOfTerminationCriterion = taskInChain.getTaskConfig();
      } else {
        tailConfigOfTerminationCriterion = new TaskConfig(rootOfTerminationCriterionVertex.getConfiguration());
      }
     
      rootOfTerminationCriterionVertex.setInvokableClass(IterationTailPactTask.class);
      // Hack
      tailConfigOfTerminationCriterion.setIsSolutionSetUpdate();
      tailConfigOfTerminationCriterion.setOutputSerializer(bulkNode.getSerializerForIterationChannel());
     
      // tell the head that it needs to wait for the solution set updates
      headConfig.setWaitForSolutionSetUpdate();
    }
   
    // ------------------- register the aggregators -------------------
    AggregatorRegistry aggs = bulkNode.getIterationNode().getIterationContract().getAggregators();
    Collection<AggregatorWithName<?>> allAggregators = aggs.getAllRegisteredAggregators();
   
    headConfig.addIterationAggregators(allAggregators);
    syncConfig.addIterationAggregators(allAggregators);
   
    String convAggName = aggs.getConvergenceCriterionAggregatorName();
    ConvergenceCriterion<?> convCriterion = aggs.getConvergenceCriterion();
   
    if (convCriterion != null || convAggName != null) {
      if (convCriterion == null) {
        throw new CompilerException("Error: Convergence criterion aggregator set, but criterion is null.");
      }
      if (convAggName == null) {
        throw new CompilerException("Error: Aggregator convergence criterion set, but aggregator is null.");
      }
     
      syncConfig.setConvergenceCriterion(convAggName, convCriterion);
    }
  }
 
  private void finalizeWorksetIteration(IterationDescriptor descr) {
    final WorksetIterationPlanNode iterNode = (WorksetIterationPlanNode) descr.getIterationNode();
    final AbstractJobVertex headVertex = descr.getHeadTask();
    final TaskConfig headConfig = new TaskConfig(headVertex.getConfiguration());
    final TaskConfig headFinalOutputConfig = descr.getHeadFinalResultConfig();
   
    // ------------ finalize the head config with the final outputs and the sync gate ------------
    {
      final int numStepFunctionOuts = headConfig.getNumOutputs();
      final int numFinalOuts = headFinalOutputConfig.getNumOutputs();
     
      if (numStepFunctionOuts == 0) {
        throw new CompilerException("The workset iteration has no operation on the workset inside the step function.");
      }
     
      headConfig.setIterationHeadFinalOutputConfig(headFinalOutputConfig);
      headConfig.setIterationHeadIndexOfSyncOutput(numStepFunctionOuts + numFinalOuts);
      final double relativeMemory = iterNode.getRelativeMemoryPerSubTask();
      if (relativeMemory <= 0) {
        throw new CompilerException("Bug: No memory has been assigned to the workset iteration.");
      }
     
      headConfig.setIsWorksetIteration();
      headConfig.setRelativeBackChannelMemory(relativeMemory / 2);
      headConfig.setRelativeSolutionSetMemory(relativeMemory / 2);
     
      // set the solution set serializer and comparator
      headConfig.setSolutionSetSerializer(iterNode.getSolutionSetSerializer());
      headConfig.setSolutionSetComparator(iterNode.getSolutionSetComparator());
    }
   
    // --------------------------- create the sync task ---------------------------
    final TaskConfig syncConfig;
    {
      final AbstractJobVertex sync = new AbstractJobVertex("Sync (" + iterNode.getNodeName() + ")");
      sync.setInvokableClass(IterationSynchronizationSinkTask.class);
      sync.setParallelism(1);
      this.auxVertices.add(sync);
     
      syncConfig = new TaskConfig(sync.getConfiguration());
      syncConfig.setGateIterativeWithNumberOfEventsUntilInterrupt(0, headVertex.getParallelism());
 
      // set the number of iteration / convergence criterion for the sync
      final int maxNumIterations = iterNode.getIterationNode().getIterationContract().getMaximumNumberOfIterations();
      if (maxNumIterations < 1) {
        throw new CompilerException("Cannot create workset iteration with unspecified maximum number of iterations.");
      }
      syncConfig.setNumberOfIterations(maxNumIterations);
     
      // connect the sync task
      sync.connectNewDataSetAsInput(headVertex, DistributionPattern.POINTWISE);
    }
   
    // ----------------------------- create the iteration tails -----------------------------
    // ----------------------- for next workset and solution set delta-----------------------

    {
      // we have three possible cases:
      // 1) Two tails, one for workset update, one for solution set update
      // 2) One tail for workset update, solution set update happens in an intermediate task
      // 3) One tail for solution set update, workset update happens in an intermediate task
     
      final PlanNode nextWorksetNode = iterNode.getNextWorkSetPlanNode();
      final PlanNode solutionDeltaNode = iterNode.getSolutionSetDeltaPlanNode();
     
      final boolean hasWorksetTail = nextWorksetNode.getOutgoingChannels().isEmpty();
      final boolean hasSolutionSetTail = (!iterNode.isImmediateSolutionSetUpdate()) || (!hasWorksetTail);
     
      {
        // get the vertex for the workset update
        final TaskConfig worksetTailConfig;
        AbstractJobVertex nextWorksetVertex = (AbstractJobVertex) this.vertices.get(nextWorksetNode);
        if (nextWorksetVertex == null) {
          // nextWorksetVertex is chained
          TaskInChain taskInChain = this.chainedTasks.get(nextWorksetNode);
          if (taskInChain == null) {
            throw new CompilerException("Bug: Next workset node not found as vertex or chained task.");
          }
          nextWorksetVertex = (AbstractJobVertex) taskInChain.getContainingVertex();
          worksetTailConfig = taskInChain.getTaskConfig();
        } else {
          worksetTailConfig = new TaskConfig(nextWorksetVertex.getConfiguration());
        }
       
        // mark the node to perform workset updates
        worksetTailConfig.setIsWorksetIteration();
        worksetTailConfig.setIsWorksetUpdate();
       
        if (hasWorksetTail) {
          nextWorksetVertex.setInvokableClass(IterationTailPactTask.class);
         
          worksetTailConfig.setOutputSerializer(iterNode.getWorksetSerializer());
        }
      }
      {
        final TaskConfig solutionDeltaConfig;
        AbstractJobVertex solutionDeltaVertex = (AbstractJobVertex) this.vertices.get(solutionDeltaNode);
        if (solutionDeltaVertex == null) {
          // last op is chained
          TaskInChain taskInChain = this.chainedTasks.get(solutionDeltaNode);
          if (taskInChain == null) {
            throw new CompilerException("Bug: Solution Set Delta not found as vertex or chained task.");
          }
          solutionDeltaVertex = (AbstractJobVertex) taskInChain.getContainingVertex();
          solutionDeltaConfig = taskInChain.getTaskConfig();
        } else {
          solutionDeltaConfig = new TaskConfig(solutionDeltaVertex.getConfiguration());
        }
       
        solutionDeltaConfig.setIsWorksetIteration();
        solutionDeltaConfig.setIsSolutionSetUpdate();
       
        if (hasSolutionSetTail) {
          solutionDeltaVertex.setInvokableClass(IterationTailPactTask.class);
         
          solutionDeltaConfig.setOutputSerializer(iterNode.getSolutionSetSerializer());
         
          // tell the head that it needs to wait for the solution set updates
          headConfig.setWaitForSolutionSetUpdate();
        }
        else {
          // no tail, intermediate update. must be immediate update
          if (!iterNode.isImmediateSolutionSetUpdate()) {
            throw new CompilerException("A solution set update without dedicated tail is not set to perform immediate updates.");
          }
          solutionDeltaConfig.setIsSolutionSetUpdateWithoutReprobe();
        }
      }
    }
   
    // ------------------- register the aggregators -------------------
    AggregatorRegistry aggs = iterNode.getIterationNode().getIterationContract().getAggregators();
    Collection<AggregatorWithName<?>> allAggregators = aggs.getAllRegisteredAggregators();
   
    for (AggregatorWithName<?> agg : allAggregators) {
      if (agg.getName().equals(WorksetEmptyConvergenceCriterion.AGGREGATOR_NAME)) {
        throw new CompilerException("User defined aggregator used the same name as built-in workset " +
            "termination check aggregator: " + WorksetEmptyConvergenceCriterion.AGGREGATOR_NAME);
      }
    }
   
    headConfig.addIterationAggregators(allAggregators);
    syncConfig.addIterationAggregators(allAggregators);
   
    String convAggName = aggs.getConvergenceCriterionAggregatorName();
    ConvergenceCriterion<?> convCriterion = aggs.getConvergenceCriterion();
   
    if (convCriterion != null || convAggName != null) {
      throw new CompilerException("Error: Cannot use custom convergence criterion with workset iteration. Workset iterations have implicit convergence criterion where workset is empty.");
    }
   
    headConfig.addIterationAggregator(WorksetEmptyConvergenceCriterion.AGGREGATOR_NAME, new LongSumAggregator());
    syncConfig.addIterationAggregator(WorksetEmptyConvergenceCriterion.AGGREGATOR_NAME, new LongSumAggregator());
    syncConfig.setConvergenceCriterion(WorksetEmptyConvergenceCriterion.AGGREGATOR_NAME, new WorksetEmptyConvergenceCriterion());
  }

  // -------------------------------------------------------------------------------------
  // Descriptors for tasks / configurations that are chained or merged with other tasks
  // -------------------------------------------------------------------------------------
 
  /**
   * Utility class that describes a task in a sequence of chained tasks. Chained tasks are tasks that run
   * together in one thread.
   */
  private static final class TaskInChain {
   
    private final Class<? extends ChainedDriver<?, ?>> chainedTask;
   
    private final TaskConfig taskConfig;
   
    private final String taskName;
   
    private AbstractJobVertex containingVertex;

    TaskInChain(Class<? extends ChainedDriver<?, ?>> chainedTask, TaskConfig taskConfig,
          String taskName) {
      this.chainedTask = chainedTask;
      this.taskConfig = taskConfig;
      this.taskName = taskName;
    }
   
    public Class<? extends ChainedDriver<?, ?>> getChainedTask() {
      return this.chainedTask;
    }
   
    public TaskConfig getTaskConfig() {
      return this.taskConfig;
    }
   
    public String getTaskName() {
      return this.taskName;
    }
   
    public AbstractJobVertex getContainingVertex() {
      return this.containingVertex;
    }
   
    public void setContainingVertex(AbstractJobVertex containingVertex) {
      this.containingVertex = containingVertex;
    }
  }
 
  private static final class IterationDescriptor {
   
    private final IterationPlanNode iterationNode;
   
    private AbstractJobVertex headTask;
   
    private TaskConfig headConfig;
   
    private TaskConfig  headFinalResultConfig;
   
    private final int id;

    public IterationDescriptor(IterationPlanNode iterationNode, int id) {
      this.iterationNode = iterationNode;
      this.id = id;
    }
   
    public IterationPlanNode getIterationNode() {
      return iterationNode;
    }
   
    public void setHeadTask(AbstractJobVertex headTask, TaskConfig headConfig) {
      this.headTask = headTask;
      this.headFinalResultConfig = new TaskConfig(new Configuration());
     
      // check if we already had a configuration, for example if the solution set was
      if (this.headConfig != null) {
        headConfig.getConfiguration().addAll(this.headConfig.getConfiguration());
      }
     
      this.headConfig = headConfig;
    }
   
    public AbstractJobVertex getHeadTask() {
      return headTask;
    }
   
    public TaskConfig getHeadFinalResultConfig() {
      return headFinalResultConfig;
    }
   
    public int getId() {
      return this.id;
    }
  }
}
TOP

Related Classes of org.apache.flink.compiler.plantranslate.NepheleJobGraphGenerator

TOP
Copyright © 2018 www.massapi.com. All rights reserved.
All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.