/*

  * 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.giraph.comm.messages.primitives.long_id;

 import it.unimi.dsi.fastutil.ints.Int2ObjectOpenHashMap;

 import it.unimi.dsi.fastutil.longs.Long2ObjectOpenHashMap;

 import java.util.List;

 import org.apache.giraph.comm.messages.PartitionSplitInfo;

 import org.apache.giraph.conf.ImmutableClassesGiraphConfiguration;

 import org.apache.giraph.factories.MessageValueFactory;

 import org.apache.giraph.graph.Vertex;

 import org.apache.giraph.partition.Partition;

 import org.apache.hadoop.io.LongWritable;

 import org.apache.hadoop.io.Writable;

 /**

  * Special message store to be used when ids are LongWritable and no combiner

  * is used.

  * Uses fastutil primitive maps in order to decrease number of objects and

  * get better performance.

  *

  * @param <M> message type

  * @param <L> list type

  */

 public abstract class LongAbstractListStore<M extends Writable,

   L extends List> extends LongAbstractStore<M, L> {

   /**

    * Map used to store messages for nascent vertices i.e., ones

    * that did not exist at the start of current superstep but will get

    * created because of sending message to a non-existent vertex id

    */

   private final

   Int2ObjectOpenHashMap<Long2ObjectOpenHashMap<L>> nascentMap;

   /**

    * Constructor

    *

    * @param messageValueFactory Factory for creating message values

    * @param partitionInfo       Partition split info

    * @param config              Hadoop configuration

    */

   public LongAbstractListStore(

       MessageValueFactory<M> messageValueFactory,

       PartitionSplitInfo<LongWritable> partitionInfo,

       ImmutableClassesGiraphConfiguration<LongWritable,

           Writable, Writable> config) {

     super(messageValueFactory, partitionInfo, config);

     populateMap();

     // create map for vertex ids (i.e., nascent vertices) not known yet

     nascentMap = new Int2ObjectOpenHashMap<>();

     for (int partitionId : partitionInfo.getPartitionIds()) {

       nascentMap.put(partitionId, new Long2ObjectOpenHashMap<L>());

     }

   }

   /**

    * Populate the map with all vertexIds for each partition

    */

   private void populateMap() { // TODO - can parallelize?

     // populate with vertex ids already known

     partitionInfo.startIteration();

     while (true) {

       Partition partition = partitionInfo.getNextPartition();

       if (partition == null) {

         break;

       }

       Long2ObjectOpenHashMap<L> partitionMap = map.get(partition.getId());

       for (Object obj : partition) {

         Vertex vertex = (Vertex) obj;

         LongWritable vertexId = (LongWritable) vertex.getId();

         partitionMap.put(vertexId.get(), createList());

       }

       partitionInfo.putPartition(partition);

     }

   }

   /**

    * Create an instance of L

    * @return instance of L

    */

   protected abstract L createList();

   /**

    * Get list for the current vertexId

    *

    * @param vertexId vertex id

    * @return list for current vertexId

    */

   protected L getList(LongWritable vertexId) {

     long id = vertexId.get();

     int partitionId = partitionInfo.getPartitionId(vertexId);

     Long2ObjectOpenHashMap<L> partitionMap = map.get(partitionId);

     L list = partitionMap.get(id);

     if (list == null) {

       Long2ObjectOpenHashMap<L> nascentPartitionMap =

         nascentMap.get(partitionId);

       // assumption: not many nascent vertices are created

       // so overall synchronization is negligible

       synchronized (nascentPartitionMap) {

         list = nascentPartitionMap.get(id);

         if (list == null) {

           list = createList();

           nascentPartitionMap.put(id, list);

         }

         return list;

       }

     }

     return list;

   }

   @Override

   public void finalizeStore() {

     for (int partitionId : nascentMap.keySet()) {

       // nascent vertices are present only in nascent map

       map.get(partitionId).putAll(nascentMap.get(partitionId));

     }

     nascentMap.clear();

   }

   @Override

   public boolean hasMessagesForVertex(LongWritable vertexId) {

     int partitionId = partitionInfo.getPartitionId(vertexId);

     Long2ObjectOpenHashMap<L> partitionMap = map.get(partitionId);

     L list = partitionMap.get(vertexId.get());

     if (list != null && !list.isEmpty()) {

       return true;

     }

     Long2ObjectOpenHashMap<L> nascentMessages = nascentMap.get(partitionId);

     return nascentMessages != null &&

            nascentMessages.containsKey(vertexId.get());

   }

   // TODO - discussion

   /*

   some approaches for ensuring correctness with parallel inserts

   - current approach: uses a small extra bit of memory by pre-populating

   map & pushes everything map cannot handle to nascentMap

   at the beginning of next superstep compute a single threaded finalizeStore is

   called (so little extra memory + 1 sequential finish ops)

   - used striped parallel fast utils instead (unsure of perf)

   - use concurrent map (every get gets far slower)

   - use reader writer locks (unsure of perf)

   (code looks something like underneath)

       private final ReadWriteLock rwl = new ReentrantReadWriteLock();

       rwl.readLock().lock();

       L list = partitionMap.get(vertexId);

       if (list == null) {

         rwl.readLock().unlock();

         rwl.writeLock().lock();

         if (partitionMap.get(vertexId) == null) {

           list = createList();

           partitionMap.put(vertexId, list);

         }

         rwl.readLock().lock();

         rwl.writeLock().unlock();

       }

       rwl.readLock().unlock();

   - adopted from the article

     http://docs.oracle.com/javase/1.5.0/docs/api/java/util/concurrent/locks/\

     ReentrantReadWriteLock.html

    */

 }