Package org.apache.hadoop.hive.ql.exec

Examples of org.apache.hadoop.hive.ql.exec.FileSinkOperator

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
168
169
170
171
172
173
174
    ParseContext parseCtx = ctx.getParseCtx();
    boolean chDir = false;
    Task<? extends Serializable> currTask = ctx.getCurrTask();
    ctx.addRootIfPossible(currTask);

    FileSinkOperator fsOp = (FileSinkOperator) nd;
    boolean isInsertTable = // is INSERT OVERWRITE TABLE
    fsOp.getConf().getTableInfo().getTableName() != null &&
        parseCtx.getQB().getParseInfo().isInsertToTable();
    HiveConf hconf = parseCtx.getConf();

    // Mark this task as a final map reduce task (ignoring the optional merge task)
    ((MapredWork)currTask.getWork()).setFinalMapRed(true);

    // If this file sink desc has been processed due to a linked file sink desc,
    // use that task
    Map<FileSinkDesc, Task<? extends Serializable>> fileSinkDescs = ctx.getLinkedFileDescTasks();
    if (fileSinkDescs != null) {
      Task<? extends Serializable> childTask = fileSinkDescs.get(fsOp.getConf());
      processLinkedFileDesc(ctx, childTask);
      return true;
    }

    // Has the user enabled merging of files for map-only jobs or for all jobs
    if ((ctx.getMvTask() != null) && (!ctx.getMvTask().isEmpty())) {
      List<Task<MoveWork>> mvTasks = ctx.getMvTask();

      // In case of unions or map-joins, it is possible that the file has
      // already been seen.
      // So, no need to attempt to merge the files again.
      if ((ctx.getSeenFileSinkOps() == null)
          || (!ctx.getSeenFileSinkOps().contains(nd))) {

        // no need of merging if the move is to a local file system
        MoveTask mvTask = (MoveTask) findMoveTask(mvTasks, fsOp);

        if (mvTask != null && isInsertTable && hconf.getBoolVar(ConfVars.HIVESTATSAUTOGATHER)) {
          addStatsTask(fsOp, mvTask, currTask, parseCtx.getConf());
        }

        if ((mvTask != null) && !mvTask.isLocal() && fsOp.getConf().canBeMerged()) {
          if (fsOp.getConf().isLinkedFileSink()) {
            // If the user has HIVEMERGEMAPREDFILES set to false, the idea was the
            // number of reducers are few, so the number of files anyway are small.
            // However, with this optimization, we are increasing the number of files
            // possibly by a big margin. So, merge aggresively.
            if (hconf.getBoolVar(ConfVars.HIVEMERGEMAPFILES) ||
                hconf.getBoolVar(ConfVars.HIVEMERGEMAPREDFILES)) {
              chDir = true;
            }
          } else {
              // There are separate configuration parameters to control whether to
              // merge for a map-only job
              // or for a map-reduce job
              MapredWork currWork = (MapredWork) currTask.getWork();
              boolean mergeMapOnly =
                  hconf.getBoolVar(ConfVars.HIVEMERGEMAPFILES) && currWork.getReduceWork() == null;
              boolean mergeMapRed =
                  hconf.getBoolVar(ConfVars.HIVEMERGEMAPREDFILES) &&
                      currWork.getReduceWork() != null;
              if (mergeMapOnly || mergeMapRed) {
                chDir = true;
              }
          }
        }
      }
    }

    String finalName = processFS(fsOp, stack, opProcCtx, chDir);

    if (chDir) {
      // Merge the files in the destination table/partitions by creating Map-only merge job
      // If underlying data is RCFile a RCFileBlockMerge task would be created.
      LOG.info("using CombineHiveInputformat for the merge job");
      createMRWorkForMergingFiles(fsOp, ctx, finalName);
    }

    FileSinkDesc fileSinkDesc = fsOp.getConf();
    if (fileSinkDesc.isLinkedFileSink()) {
      Map<FileSinkDesc, Task<? extends Serializable>> linkedFileDescTasks =
        ctx.getLinkedFileDescTasks();
      if (linkedFileDescTasks == null) {
        linkedFileDescTasks = new HashMap<FileSinkDesc, Task<? extends Serializable>>();
View Full Code Here
303
304
305
306
307
308
309
310
311
312
313
    // Create a FileSink operator
    TableDesc ts = (TableDesc) fsInputDesc.getTableInfo().clone();
    FileSinkDesc fsOutputDesc = new FileSinkDesc(finalName, ts,
      conf.getBoolVar(ConfVars.COMPRESSRESULT));
    FileSinkOperator fsOutput = (FileSinkOperator) OperatorFactory.getAndMakeChild(
      fsOutputDesc, inputRS, tsMerge);

    // If the input FileSinkOperator is a dynamic partition enabled, the tsMerge input schema
    // needs to include the partition column, and the fsOutput should have
    // a DynamicPartitionCtx to indicate that it needs to dynamically partitioned.
View Full Code Here
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
      this.pctx = pctx;
    }

    public Object process(Node nd, Stack<Node> stack, NodeProcessorCtx procCtx,
        Object... nodeOutputs) throws SemanticException {
      FileSinkOperator FS = (FileSinkOperator) nd;
      GroupByOperator cGBY = (GroupByOperator) stack.get(stack.size() - 3);
      ReduceSinkOperator RS = (ReduceSinkOperator) stack.get(stack.size() - 4);
      if (RS.getConf().getNumReducers() != 1 || !RS.getConf().getKeyCols().isEmpty()) {
        return null;
      }
      GroupByOperator pGBY = (GroupByOperator) stack.get(stack.size() - 5);

      String fileName = FS.getConf().getFinalDirName();
      TableDesc tsDesc = createIntermediateFS(pGBY, fileName);

      for (AggregationDesc aggregation : cGBY.getConf().getAggregators()) {
        List<ExprNodeDesc> parameters = aggregation.getParameters();
        aggregation.setParameters(ExprNodeDescUtils.backtrack(parameters, cGBY, pGBY));
View Full Code Here
116
117
118
119
120
121
122
123
124
125
126
127
128
129
      TableDesc tsDesc = PlanUtils.getIntermediateFileTableDesc(PlanUtils
          .getFieldSchemasFromRowSchema(parent.getSchema(), "temporarycol"));

      // Create a file sink operator for this file name
      FileSinkDesc desc = new FileSinkDesc(fileName, tsDesc, false);
      FileSinkOperator newFS = (FileSinkOperator) OperatorFactory.get(desc, parent.getSchema());

      newFS.setParentOperators(new ArrayList<Operator<? extends OperatorDesc>>());
      newFS.getParentOperators().add(parent);

      parent.getChildOperators().clear();
      parent.getChildOperators().add(newFS);

      return tsDesc;
View Full Code Here
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
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
      if (!(op instanceof FileSinkOperator)) {
        return;
      }

      FileSinkOperator fop = (FileSinkOperator)op;
      String workDir = fop.getConf().getDirName();

      Map<String, ArrayList<String>> childPathToAliases = childWork.getPathToAliases();
      if (childPathToAliases.size() > 1) {
        return;
      }

      // The filesink writes to a different directory
      if (!childPathToAliases.keySet().iterator().next().equals(workDir)) {
        return;
      }

      // Either of them should not be bucketed
      if ((localWork.getBucketMapjoinContext() != null) ||
          (childLocalWork.getBucketMapjoinContext() != null)) {
        return;
      }

      // Merge the trees
      if (childWork.getAliasToWork().size() > 1) {
        return;
      }
      long mapJoinSize = HiveConf.getLongVar(conf,
          HiveConf.ConfVars.HIVECONVERTJOINNOCONDITIONALTASKTHRESHOLD);
      long localTableTotalSize = 0;
      for (String alias : localWork.getAliasToWork().keySet()) {
        Long tabSize = aliasToSize.get(alias);
        if (tabSize == null) {
          /* if the size is unavailable, we need to assume a size 1 greater than mapJoinSize
           * this implies that merge cannot happen so we can return.
           */
          return;
        }
        localTableTotalSize += tabSize;
      }

      for (String alias : childLocalWork.getAliasToWork().keySet()) {
        Long tabSize = aliasToSize.get(alias);
        if (tabSize == null) {
          /* if the size is unavailable, we need to assume a size 1 greater than mapJoinSize
           * this implies that merge cannot happen so we can return.
           */
          return;
        }
        localTableTotalSize += tabSize;
        if (localTableTotalSize > mapJoinSize) {
          return;
        }
      }

      Operator<? extends Serializable> childAliasOp =
          childWork.getAliasToWork().values().iterator().next();
      if (fop.getParentOperators().size() > 1) {
        return;
      }

      // Merge the 2 trees - remove the FileSinkOperator from the first tree pass it to the
      // top of the second
      Operator<? extends Serializable> parentFOp = fop.getParentOperators().get(0);
      parentFOp.getChildOperators().remove(fop);
      parentFOp.getChildOperators().add(childAliasOp);
      List<Operator<? extends OperatorDesc>> parentOps =
          new ArrayList<Operator<? extends OperatorDesc>>();
      parentOps.add(parentFOp);
View Full Code Here
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
      if (!(mapJoinLeafOperator instanceof FileSinkOperator)) {
        // Sanity check, shouldn't happen.
        return;
      }

      FileSinkOperator mapJoinTaskFileSinkOperator = (FileSinkOperator) mapJoinLeafOperator;

      // The filesink writes to a different directory
      String workDir = mapJoinTaskFileSinkOperator.getConf().getDirName();
      if (!childPathToAliases.keySet().iterator().next().equals(workDir)) {
        return;
      }

      MapredLocalWork mapJoinLocalWork = mapJoinWork.getMapLocalWork();
      MapredLocalWork childLocalWork = childWork.getMapLocalWork();

      // Either of them should not be bucketed
      if ((mapJoinLocalWork != null && mapJoinLocalWork.getBucketMapjoinContext() != null) ||
          (childLocalWork != null && childLocalWork.getBucketMapjoinContext() != null)) {
        return;
      }

      if (childWork.getAliasToWork().size() > 1) {
        return;
      }

      Operator<? extends Serializable> childAliasOp =
          childWork.getAliasToWork().values().iterator().next();
      if (mapJoinTaskFileSinkOperator.getParentOperators().size() > 1) {
        return;
      }

      // Merge the 2 trees - remove the FileSinkOperator from the first tree pass it to the
      // top of the second
      Operator<? extends Serializable> parentFOp = mapJoinTaskFileSinkOperator
          .getParentOperators().get(0);
      parentFOp.getChildOperators().remove(mapJoinTaskFileSinkOperator);
      parentFOp.getChildOperators().add(childAliasOp);
      List<Operator<? extends OperatorDesc>> parentOps =
          new ArrayList<Operator<? extends OperatorDesc>>();
View Full Code Here
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
        // FileSink cannot be simply cloned - it requires some special processing.
        // Sub-queries for the union will be processed as independent map-reduce jobs
        // possibly running in parallel. Those sub-queries cannot write to the same
        // directory. Clone the filesink, but create a sub-directory in the final path
        // for each sub-query. Also, these different filesinks need to be linked to each other
        FileSinkOperator fileSinkOp = (FileSinkOperator)stack.get(pos);
        // For file sink operator, change the directory name
        String parentDirName = fileSinkOp.getConf().getDirName();

        // Clone the fileSinkDesc of the final fileSink and create similar fileSinks at
        // each parent
        List<FileSinkDesc> fileDescLists = new ArrayList<FileSinkDesc>();

        for (Operator<? extends OperatorDesc> parent : parents) {
          FileSinkDesc fileSinkDesc = (FileSinkDesc) fileSinkOp.getConf().clone();

          String dirName = parentDirName + Path.SEPARATOR + parent.getIdentifier() ;
          fileSinkDesc.setDirName(dirName);
          fileSinkDesc.setLinkedFileSink(true);
          fileSinkDesc.setParentDir(parentDirName);
View Full Code Here
245
246
247
248
249
250
251
252
253
254
255
256
257
258
    }

    @Override
    public Object process(Node nd, Stack<Node> stack, NodeProcessorCtx procCtx,
        Object... nodeOutputs) throws SemanticException {
      FileSinkOperator fileSinkOp   = (FileSinkOperator)nd;

      // Has this filesink already been processed
      if (fileSinkOp.getConf().isLinkedFileSink()) {
        return null;
      }

      int size = stack.size();
      int pos = size - 2;
View Full Code Here
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
    @Override
    public Object process(Node nd, Stack<Node> stack, NodeProcessorCtx procCtx,
        Object... nodeOutputs) throws SemanticException {

      BucketingSortingCtx bctx = (BucketingSortingCtx)procCtx;
      FileSinkOperator fop = (FileSinkOperator)nd;

      Operator<? extends OperatorDesc> parent = getParent(stack);
      List<BucketCol> bucketCols = bctx.getBucketedCols(parent);
      List<ColumnInfo> colInfos = fop.getSchema().getSignature();

      // Set the inferred bucket columns for the file this FileSink produces
      if (bucketCols != null) {
        List<BucketCol> newBucketCols = getNewBucketCols(bucketCols, colInfos);
        bctx.getBucketedColsByDirectory().put(fop.getConf().getDirName(), newBucketCols);
        bctx.setBucketedCols(fop, newBucketCols);
      }

      List<SortCol> sortCols = bctx.getSortedCols(parent);

      // Set the inferred sort columns for the file this FileSink produces
      if (sortCols != null) {
        List<SortCol> newSortCols = getNewSortCols(sortCols, colInfos);
        bctx.getSortedColsByDirectory().put(fop.getConf().getDirName(), newSortCols);
        bctx.setSortedCols(fop, newSortCols);
      }

      return null;
    }
View Full Code Here
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
168
169
170
171
      Object... nodeOutputs) throws SemanticException {
    GenMRProcContext ctx = (GenMRProcContext) opProcCtx;
    ParseContext parseCtx = ctx.getParseCtx();
    boolean chDir = false;
    Task<? extends Serializable> currTask = ctx.getCurrTask();
    FileSinkOperator fsOp = (FileSinkOperator) nd;
    boolean isInsertTable = // is INSERT OVERWRITE TABLE
    fsOp.getConf().getTableInfo().getTableName() != null &&
        parseCtx.getQB().getParseInfo().isInsertToTable();
    HiveConf hconf = parseCtx.getConf();

    // Mark this task as a final map reduce task (ignoring the optional merge task)
    ((MapredWork)currTask.getWork()).setFinalMapRed(true);

    // If this file sink desc has been processed due to a linked file sink desc,
    // use that task
    Map<FileSinkDesc, Task<? extends Serializable>> fileSinkDescs = ctx.getLinkedFileDescTasks();
    if (fileSinkDescs != null) {
      Task<? extends Serializable> childTask = fileSinkDescs.get(fsOp.getConf());
      processLinkedFileDesc(ctx, childTask);
      return null;
    }

    // Has the user enabled merging of files for map-only jobs or for all jobs
    if ((ctx.getMvTask() != null) && (!ctx.getMvTask().isEmpty())) {
      List<Task<MoveWork>> mvTasks = ctx.getMvTask();

      // In case of unions or map-joins, it is possible that the file has
      // already been seen.
      // So, no need to attempt to merge the files again.
      if ((ctx.getSeenFileSinkOps() == null)
          || (!ctx.getSeenFileSinkOps().contains(nd))) {

        // no need of merging if the move is to a local file system
        MoveTask mvTask = (MoveTask) findMoveTask(mvTasks, fsOp);

        if (isInsertTable && hconf.getBoolVar(ConfVars.HIVESTATSAUTOGATHER)) {
          addStatsTask(fsOp, mvTask, currTask, parseCtx.getConf());
        }

        if ((mvTask != null) && !mvTask.isLocal() && fsOp.getConf().canBeMerged()) {
          if (fsOp.getConf().isLinkedFileSink()) {
            // If the user has HIVEMERGEMAPREDFILES set to false, the idea was the
            // number of reducers are few, so the number of files anyway are small.
            // However, with this optimization, we are increasing the number of files
            // possibly by a big margin. So, merge aggresively.
            if (hconf.getBoolVar(ConfVars.HIVEMERGEMAPFILES) ||
                hconf.getBoolVar(ConfVars.HIVEMERGEMAPREDFILES)) {
              chDir = true;
            }
          } else {
              // There are separate configuration parameters to control whether to
              // merge for a map-only job
              // or for a map-reduce job
              MapredWork currWork = (MapredWork) currTask.getWork();
              boolean mergeMapOnly =
                  hconf.getBoolVar(ConfVars.HIVEMERGEMAPFILES) && currWork.getReducer() == null;
              boolean mergeMapRed =
                  hconf.getBoolVar(ConfVars.HIVEMERGEMAPREDFILES) &&
                      currWork.getReducer() != null;
              if (mergeMapOnly || mergeMapRed) {
                chDir = true;
              }
          }
        }
      }
    }

    String finalName = processFS(fsOp, stack, opProcCtx, chDir);

    if (chDir) {
      // Merge the files in the destination table/partitions by creating Map-only merge job
      // If underlying data is RCFile a RCFileBlockMerge task would be created.
      LOG.info("using CombineHiveInputformat for the merge job");
      createMRWorkForMergingFiles(fsOp, ctx, finalName);
    }

    FileSinkDesc fileSinkDesc = fsOp.getConf();
    if (fileSinkDesc.isLinkedFileSink()) {
      Map<FileSinkDesc, Task<? extends Serializable>> linkedFileDescTasks =
        ctx.getLinkedFileDescTasks();
      if (linkedFileDescTasks == null) {
        linkedFileDescTasks = new HashMap<FileSinkDesc, Task<? extends Serializable>>();
View Full Code Here
TOP

Related Classes of org.apache.hadoop.hive.ql.exec.FileSinkOperator

  • org.apache.hadoop.fs.FileStatus
  • org.apache.hadoop.fs.FileSystem
  • org.apache.hadoop.fs.Path
  • org.apache.hadoop.hive.ql.io.FSRecordWriter
  • org.apache.hadoop.hive.ql.optimizer.BucketingSortingReduceSinkOptimizer$BucketSortReduceSinkProcessor
  • org.apache.hadoop.hive.ql.optimizer.ConstantPropagateProcFactory$ConstantPropagateFileSinkProc
  • org.apache.hadoop.hive.ql.optimizer.GenMapRedUtils
  • org.apache.hadoop.hive.ql.optimizer.GenMRFileSink1
  • org.apache.hadoop.hive.ql.optimizer.physical.BucketingSortingOpProcFactory$FileSinkInferrer
  • org.apache.hadoop.hive.ql.optimizer.physical.CommonJoinResolver$CommonJoinTaskDispatcher
    • 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.