Package org.apache.tez.mapreduce.hadoop

Examples of org.apache.tez.mapreduce.hadoop.InputSplitInfoMem

113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
    // Read all credentials into the credentials instance stored in JobConf.
    JobConf jobConf = new JobConf(conf);
    jobConf.getCredentials().mergeAll(UserGroupInformation.getCurrentUser().getCredentials());

    InputSplitInfoMem inputSplitInfo = null;
    boolean groupSplits = userPayloadProto.getGroupingEnabled();
    if (groupSplits) {
      LOG.info("Grouping input splits");
      inputSplitInfo = MRInputHelpers.generateInputSplitsToMem(jobConf, true, numTasks);
    } else {
      inputSplitInfo = MRInputHelpers.generateInputSplitsToMem(jobConf, false, 0);
    }
    if (LOG.isDebugEnabled()) {
      sw.stop();
      LOG.debug("Time to create splits to mem: " + sw.elapsedMillis());
    }

    List<Event> events = Lists.newArrayListWithCapacity(inputSplitInfo
        .getNumTasks() + 1);
   
    InputConfigureVertexTasksEvent configureVertexEvent = InputConfigureVertexTasksEvent.create(
        inputSplitInfo.getNumTasks(),
        VertexLocationHint.create(inputSplitInfo.getTaskLocationHints()),
        InputSpecUpdate.getDefaultSinglePhysicalInputSpecUpdate());
    events.add(configureVertexEvent);

    if (sendSerializedEvents) {
      MRSplitsProto splitsProto = inputSplitInfo.getSplitsProto();
      int count = 0;
      for (MRSplitProto mrSplit : splitsProto.getSplitsList()) {
        // Unnecessary array copy, can be avoided by using ByteBuffer instead of a raw array.
        InputDataInformationEvent diEvent = InputDataInformationEvent.createWithSerializedPayload(
            count++,
            mrSplit.toByteString().asReadOnlyByteBuffer());
        events.add(diEvent);
      }
    } else {
      int count = 0;
      if (inputSplitInfo.holdsNewFormatSplits()) {
        for (org.apache.hadoop.mapreduce.InputSplit split : inputSplitInfo.getNewFormatSplits()) {
          InputDataInformationEvent diEvent = InputDataInformationEvent.createWithObjectPayload(
              count++, split);
          events.add(diEvent);
        }
      } else {
        for (org.apache.hadoop.mapred.InputSplit split : inputSplitInfo.getOldFormatSplits()) {
          InputDataInformationEvent diEvent = InputDataInformationEvent.createWithObjectPayload(
              count++, split);
          events.add(diEvent);
        }
      }
View Full Code Here
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
    LOG.info("Input " + rootInputContext.getInputName() + " asking for " + numTasks
        + " tasks. Headroom: " + totalResource + " Task Resource: "
        + taskResource + " waves: " + waves);


    InputSplitInfoMem inputSplitInfo = null;
    String realInputFormatName = userPayloadProto.getInputFormatName();
    if ( realInputFormatName != null && !realInputFormatName.isEmpty()) {
      // split grouping on the AM
      JobConf jobConf = new JobConf(conf);
      if (jobConf.getUseNewMapper()) {
        LOG.info("Grouping mapreduce api input splits");
        Job job = Job.getInstance(conf);
        org.apache.hadoop.mapreduce.InputSplit[] splits = MRHelpers
            .generateNewSplits(job, realInputFormatName, numTasks);

        // Move all this into a function
        List<TaskLocationHint> locationHints = Lists
            .newArrayListWithCapacity(splits.length);
        for (org.apache.hadoop.mapreduce.InputSplit split : splits) {
          String rack =
              ((org.apache.hadoop.mapreduce.split.TezGroupedSplit) split).getRack();
          if (rack == null) {
            if (split.getLocations() != null) {
              locationHints.add(new TaskLocationHint(new HashSet<String>(Arrays
                  .asList(split.getLocations())), null));
            } else {
              locationHints.add(new TaskLocationHint(null, null));
            }
          } else {
            locationHints.add(new TaskLocationHint(null,
                Collections.singleton(rack)));
          }
        }
        inputSplitInfo = new InputSplitInfoMem(splits, locationHints, splits.length, null, conf);
      } else {
        LOG.info("Grouping mapred api input splits");
        org.apache.hadoop.mapred.InputSplit[] splits = MRHelpers
            .generateOldSplits(jobConf, realInputFormatName, numTasks);
        List<TaskLocationHint> locationHints = Lists
            .newArrayListWithCapacity(splits.length);
        for (org.apache.hadoop.mapred.InputSplit split : splits) {
          String rack =
              ((org.apache.hadoop.mapred.split.TezGroupedSplit) split).getRack();
          if (rack == null) {
            if (split.getLocations() != null) {
              locationHints.add(new TaskLocationHint(new HashSet<String>(Arrays
                  .asList(split.getLocations())), null));
            } else {
              locationHints.add(new TaskLocationHint(null, null));
            }
          } else {
            locationHints.add(new TaskLocationHint(null,
                Collections.singleton(rack)));
          }
        }
        inputSplitInfo = new InputSplitInfoMem(splits, locationHints, splits.length, null, conf);
      }
    } else {
      inputSplitInfo = MRHelpers.generateInputSplitsToMem(conf);
    }
    if (LOG.isDebugEnabled()) {
      sw.stop();
      LOG.debug("Time to create splits to mem: " + sw.elapsedMillis());
    }

    List<Event> events = Lists.newArrayListWithCapacity(inputSplitInfo
        .getNumTasks() + 1);
   
    RootInputConfigureVertexTasksEvent configureVertexEvent = new RootInputConfigureVertexTasksEvent(
        inputSplitInfo.getNumTasks(), inputSplitInfo.getTaskLocationHints());
    events.add(configureVertexEvent);

    if (sendSerializedEvents) {
      MRSplitsProto splitsProto = inputSplitInfo.getSplitsProto();
      int count = 0;
      for (MRSplitProto mrSplit : splitsProto.getSplitsList()) {
        // Unnecessary array copy, can be avoided by using ByteBuffer instead of a raw array.
        RootInputDataInformationEvent diEvent = new RootInputDataInformationEvent(count++,
            mrSplit.toByteArray());
        events.add(diEvent);
      }
    } else {
      int count = 0;
      if (inputSplitInfo.holdsNewFormatSplits()) {
        for (org.apache.hadoop.mapreduce.InputSplit split : inputSplitInfo.getNewFormatSplits()) {
          RootInputDataInformationEvent diEvent = new RootInputDataInformationEvent(count++, split);
          events.add(diEvent);
        }
      } else {
        for (org.apache.hadoop.mapred.InputSplit split : inputSplitInfo.getOldFormatSplits()) {
          RootInputDataInformationEvent diEvent = new RootInputDataInformationEvent(count++, split);
          events.add(diEvent);
        }
      }
    }
View Full Code Here
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
    if (LOG.isDebugEnabled()) {
      sw.reset().start();
    }

    InputSplitInfoMem inputSplitInfo = null;
    String realInputFormatName = userPayloadProto.getInputFormatName();
    if ( realInputFormatName != null && !realInputFormatName.isEmpty()) {
      // split grouping on the AM
      JobConf jobConf = new JobConf(conf);
      if (jobConf.getUseNewMapper()) {
        LOG.info("Grouping mapreduce api input splits");
        Job job = Job.getInstance(conf);
        org.apache.hadoop.mapreduce.InputSplit[] splits = MRHelpers
            .generateNewSplits(job, realInputFormatName,
                rootInputContext.getNumTasks());
        SerializationFactory serializationFactory = new SerializationFactory(
            job.getConfiguration());

        MRSplitsProto.Builder splitsBuilder = MRSplitsProto.newBuilder();

        List<TaskLocationHint> locationHints = Lists
            .newArrayListWithCapacity(splits.length);
        for (org.apache.hadoop.mapreduce.InputSplit split : splits) {
          splitsBuilder.addSplits(MRHelpers.createSplitProto(split,
              serializationFactory));
          String rack =
              ((org.apache.hadoop.mapreduce.split.TezGroupedSplit) split).getRack();
          if (rack == null) {
            if (split.getLocations() != null) {
              locationHints.add(new TaskLocationHint(new HashSet<String>(Arrays
                  .asList(split.getLocations())), null));
            } else {
              locationHints.add(new TaskLocationHint(null, null));
            }
          } else {
            locationHints.add(new TaskLocationHint(null,
                Collections.singleton(rack)));
          }
        }
        inputSplitInfo = new InputSplitInfoMem(splitsBuilder.build(),
            locationHints, splits.length);
      } else {
        LOG.info("Grouping mapred api input splits");
        org.apache.hadoop.mapred.InputSplit[] splits = MRHelpers
            .generateOldSplits(jobConf, realInputFormatName,
                rootInputContext.getNumTasks());
        List<TaskLocationHint> locationHints = Lists
            .newArrayListWithCapacity(splits.length);
        MRSplitsProto.Builder splitsBuilder = MRSplitsProto.newBuilder();
        for (org.apache.hadoop.mapred.InputSplit split : splits) {
          splitsBuilder.addSplits(MRHelpers.createSplitProto(split));
          String rack =
              ((org.apache.hadoop.mapred.split.TezGroupedSplit) split).getRack();
          if (rack == null) {
            if (split.getLocations() != null) {
              locationHints.add(new TaskLocationHint(new HashSet<String>(Arrays
                  .asList(split.getLocations())), null));
            } else {
              locationHints.add(new TaskLocationHint(null, null));
            }
          } else {
            locationHints.add(new TaskLocationHint(null,
                Collections.singleton(rack)));
          }
        }
        inputSplitInfo = new InputSplitInfoMem(splitsBuilder.build(),
            locationHints, splits.length);
      }
    } else {
      inputSplitInfo = MRHelpers.generateInputSplitsToMem(conf);
    }
    if (LOG.isDebugEnabled()) {
      sw.stop();
      LOG.debug("Time to create splits to mem: " + sw.elapsedMillis());
    }

    List<Event> events = Lists.newArrayListWithCapacity(inputSplitInfo
        .getNumTasks() + 1);
   
    RootInputConfigureVertexTasksEvent configureVertexEvent = new RootInputConfigureVertexTasksEvent(
        inputSplitInfo.getNumTasks(), inputSplitInfo.getTaskLocationHints());
    events.add(configureVertexEvent);

    MRSplitsProto splitsProto = inputSplitInfo.getSplitsProto();

    int count = 0;
    for (MRSplitProto mrSplit : splitsProto.getSplitsList()) {
      // Unnecessary array copy, can be avoided by using ByteBuffer instead of a
      // raw array.
View Full Code Here
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
    // Read all credentials into the credentials instance stored in JobConf.
    JobConf jobConf = new JobConf(conf);
    jobConf.getCredentials().mergeAll(UserGroupInformation.getCurrentUser().getCredentials());

    InputSplitInfoMem inputSplitInfo = null;
    String realInputFormatName = userPayloadProto.getInputFormatName();
    if ( realInputFormatName != null && !realInputFormatName.isEmpty()) {
      // split grouping on the AM
      if (jobConf.getUseNewMapper()) {
        LOG.info("Grouping mapreduce api input splits");
        Job job = Job.getInstance(jobConf);
        org.apache.hadoop.mapreduce.InputSplit[] splits = MRHelpers
            .generateNewSplits(job, realInputFormatName, numTasks);

        // Move all this into a function
        List<TaskLocationHint> locationHints = Lists
            .newArrayListWithCapacity(splits.length);
        for (org.apache.hadoop.mapreduce.InputSplit split : splits) {
          String rack =
              ((org.apache.hadoop.mapreduce.split.TezGroupedSplit) split).getRack();
          if (rack == null) {
            if (split.getLocations() != null) {
              locationHints.add(new TaskLocationHint(new HashSet<String>(Arrays
                  .asList(split.getLocations())), null));
            } else {
              locationHints.add(new TaskLocationHint(null, null));
            }
          } else {
            locationHints.add(new TaskLocationHint(null,
                Collections.singleton(rack)));
          }
        }
        inputSplitInfo = new InputSplitInfoMem(splits, locationHints, splits.length, null, jobConf);
      } else {
        LOG.info("Grouping mapred api input splits");
        org.apache.hadoop.mapred.InputSplit[] splits = MRHelpers
            .generateOldSplits(jobConf, realInputFormatName, numTasks);
        List<TaskLocationHint> locationHints = Lists
            .newArrayListWithCapacity(splits.length);
        for (org.apache.hadoop.mapred.InputSplit split : splits) {
          String rack =
              ((org.apache.hadoop.mapred.split.TezGroupedSplit) split).getRack();
          if (rack == null) {
            if (split.getLocations() != null) {
              locationHints.add(new TaskLocationHint(new HashSet<String>(Arrays
                  .asList(split.getLocations())), null));
            } else {
              locationHints.add(new TaskLocationHint(null, null));
            }
          } else {
            locationHints.add(new TaskLocationHint(null,
                Collections.singleton(rack)));
          }
        }
        inputSplitInfo = new InputSplitInfoMem(splits, locationHints, splits.length, null, jobConf);
      }
    } else {
      inputSplitInfo = MRHelpers.generateInputSplitsToMem(jobConf);
    }
    if (LOG.isDebugEnabled()) {
      sw.stop();
      LOG.debug("Time to create splits to mem: " + sw.elapsedMillis());
    }

    List<Event> events = Lists.newArrayListWithCapacity(inputSplitInfo
        .getNumTasks() + 1);
   
    RootInputConfigureVertexTasksEvent configureVertexEvent = new RootInputConfigureVertexTasksEvent(
        inputSplitInfo.getNumTasks(), inputSplitInfo.getTaskLocationHints());
    events.add(configureVertexEvent);

    if (sendSerializedEvents) {
      MRSplitsProto splitsProto = inputSplitInfo.getSplitsProto();
      int count = 0;
      for (MRSplitProto mrSplit : splitsProto.getSplitsList()) {
        // Unnecessary array copy, can be avoided by using ByteBuffer instead of a raw array.
        RootInputDataInformationEvent diEvent = new RootInputDataInformationEvent(count++,
            mrSplit.toByteArray());
        events.add(diEvent);
      }
    } else {
      int count = 0;
      if (inputSplitInfo.holdsNewFormatSplits()) {
        for (org.apache.hadoop.mapreduce.InputSplit split : inputSplitInfo.getNewFormatSplits()) {
          RootInputDataInformationEvent diEvent = new RootInputDataInformationEvent(count++, split);
          events.add(diEvent);
        }
      } else {
        for (org.apache.hadoop.mapred.InputSplit split : inputSplitInfo.getOldFormatSplits()) {
          RootInputDataInformationEvent diEvent = new RootInputDataInformationEvent(count++, split);
          events.add(diEvent);
        }
      }
    }
View Full Code Here
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
    MapWork work = Utilities.getMapWork(jobConf);

    // perform dynamic partition pruning
    pruner.prune(work, jobConf, context);

    InputSplitInfoMem inputSplitInfo = null;
    String realInputFormatName = conf.get("mapred.input.format.class");
    boolean groupingEnabled = userPayloadProto.getGroupingEnabled();
    if (groupingEnabled) {
      // Need to instantiate the realInputFormat
      InputFormat<?, ?> inputFormat =
          (InputFormat<?, ?>) ReflectionUtils.newInstance(Class.forName(realInputFormatName),
              jobConf);

      int totalResource = rootInputContext.getTotalAvailableResource().getMemory();
      int taskResource = rootInputContext.getVertexTaskResource().getMemory();
      int availableSlots = totalResource / taskResource;

      // Create the un-grouped splits
      float waves =
          conf.getFloat(TezMapReduceSplitsGrouper.TEZ_GROUPING_SPLIT_WAVES,
              TezMapReduceSplitsGrouper.TEZ_GROUPING_SPLIT_WAVES_DEFAULT);

      InputSplit[] splits = inputFormat.getSplits(jobConf, (int) (availableSlots * waves));
      LOG.info("Number of input splits: " + splits.length + ". " + availableSlots
          + " available slots, " + waves + " waves. Input format is: " + realInputFormatName);

      Multimap<Integer, InputSplit> groupedSplits =
          generateGroupedSplits(jobConf, conf, splits, waves, availableSlots);
      // And finally return them in a flat array
      InputSplit[] flatSplits = groupedSplits.values().toArray(new InputSplit[0]);
      LOG.info("Number of grouped splits: " + flatSplits.length);

      List<TaskLocationHint> locationHints = grouper.createTaskLocationHints(flatSplits);

      Utilities.clearWork(jobConf);

      inputSplitInfo =
          new InputSplitInfoMem(flatSplits, locationHints, flatSplits.length, null, jobConf);
    } else {
      // no need for grouping and the target #of tasks.
      // This code path should never be triggered at the moment. If grouping is disabled,
      // DAGUtils uses MRInputAMSplitGenerator.
      // If this is used in the future - make sure to disable grouping in the payload, if it isn't already disabled
View Full Code Here
TOP

Related Classes of org.apache.tez.mapreduce.hadoop.InputSplitInfoMem

  • org.apache.hadoop.hive.ql.exec.tez.HiveSplitGenerator
  • org.apache.hadoop.io.serializer.SerializationFactory
  • org.apache.tez.mapreduce.common.MRInputAMSplitGenerator
    • Copyright © 2018 www.massapicom. 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.