private static void pruneReduceSinkOperator(
      boolean[] retainFlags, ReduceSinkOperator reduce, ColumnPrunerProcCtx cppCtx)
      throws SemanticException {
    ReduceSinkDesc reduceConf = reduce.getConf();
    Map<String, ExprNodeDesc> oldMap = reduce.getColumnExprMap();
    LOG.info("RS " + reduce.getIdentifier() + " oldColExprMap: " + oldMap);
    RowResolver oldRR = cppCtx.getOpToParseCtxMap().get(reduce).getRowResolver();
    ArrayList<ColumnInfo> old_signature = oldRR.getRowSchema().getSignature();
    ArrayList<ColumnInfo> signature = new ArrayList<ColumnInfo>(old_signature);

    List<String> valueColNames = reduceConf.getOutputValueColumnNames();
    ArrayList<String> newValueColNames = new ArrayList<String>();

    List<ExprNodeDesc> keyExprs = reduceConf.getKeyCols();
    List<ExprNodeDesc> valueExprs = reduceConf.getValueCols();
    ArrayList<ExprNodeDesc> newValueExprs = new ArrayList<ExprNodeDesc>();

    for (int i = 0; i < retainFlags.length; i++) {
      String outputCol = valueColNames.get(i);
      ExprNodeDesc outputColExpr = valueExprs.get(i);
      if (!retainFlags[i]) {
        String[] nm = oldRR.reverseLookup(outputCol);
        if (nm == null) {
          outputCol = Utilities.ReduceField.VALUE.toString() + "." + outputCol;
          nm = oldRR.reverseLookup(outputCol);
        }

        // In case there are multiple columns referenced to the same column name, we won't
        // do row resolve once more because the ColumnInfo in row resolver is already removed
        if (nm == null) {
          continue;
        }

        // Only remove information of a column if it is not a key,
        // i.e. this column is not appearing in keyExprs of the RS
        if (ExprNodeDescUtils.indexOf(outputColExpr, keyExprs) == -1) {
          ColumnInfo colInfo = oldRR.getFieldMap(nm[0]).remove(nm[1]);
          oldRR.getInvRslvMap().remove(colInfo.getInternalName());
          oldMap.remove(outputCol);
          signature.remove(colInfo);
        }

      } else {
        newValueColNames.add(outputCol);
        newValueExprs.add(outputColExpr);
      }
    }

    oldRR.getRowSchema().setSignature(signature);
    reduce.getSchema().setSignature(signature);
    reduceConf.setOutputValueColumnNames(newValueColNames);
    reduceConf.setValueCols(newValueExprs);
    TableDesc newValueTable =
        PlanUtils.getReduceValueTableDesc(
            PlanUtils.getFieldSchemasFromColumnList(
                reduceConf.getValueCols(), newValueColNames, 0, ""));
    reduceConf.setValueSerializeInfo(newValueTable);
    LOG.info("RS " + reduce.getIdentifier() + " newColExprMap: " + oldMap);
  }
Exemplo n.º 2
0
  @Override
  protected void initializeOp(Configuration hconf) throws HiveException {
    super.initializeOp(hconf);
    try {

      numRows = 0;
      cntr = 1;
      logEveryNRows = HiveConf.getLongVar(hconf, HiveConf.ConfVars.HIVE_LOG_N_RECORDS);

      statsMap.put(getCounterName(Counter.RECORDS_OUT_INTERMEDIATE, hconf), recordCounter);

      List<ExprNodeDesc> keys = conf.getKeyCols();

      if (isLogDebugEnabled) {
        LOG.debug("keys size is " + keys.size());
        for (ExprNodeDesc k : keys) {
          LOG.debug("Key exprNodeDesc " + k.getExprString());
        }
      }

      keyEval = new ExprNodeEvaluator[keys.size()];
      int i = 0;
      for (ExprNodeDesc e : keys) {
        keyEval[i++] = ExprNodeEvaluatorFactory.get(e);
      }

      numDistributionKeys = conf.getNumDistributionKeys();
      distinctColIndices = conf.getDistinctColumnIndices();
      numDistinctExprs = distinctColIndices.size();

      valueEval = new ExprNodeEvaluator[conf.getValueCols().size()];
      i = 0;
      for (ExprNodeDesc e : conf.getValueCols()) {
        valueEval[i++] = ExprNodeEvaluatorFactory.get(e);
      }

      partitionEval = new ExprNodeEvaluator[conf.getPartitionCols().size()];
      i = 0;
      for (ExprNodeDesc e : conf.getPartitionCols()) {
        int index = ExprNodeDescUtils.indexOf(e, keys);
        partitionEval[i++] = index < 0 ? ExprNodeEvaluatorFactory.get(e) : keyEval[index];
      }

      if (conf.getBucketCols() != null && !conf.getBucketCols().isEmpty()) {
        bucketEval = new ExprNodeEvaluator[conf.getBucketCols().size()];

        i = 0;
        for (ExprNodeDesc e : conf.getBucketCols()) {
          int index = ExprNodeDescUtils.indexOf(e, keys);
          bucketEval[i++] = index < 0 ? ExprNodeEvaluatorFactory.get(e) : keyEval[index];
        }

        buckColIdxInKey = conf.getPartitionCols().size();
      }

      tag = conf.getTag();
      tagByte[0] = (byte) tag;
      skipTag = conf.getSkipTag();
      if (isLogInfoEnabled) {
        LOG.info("Using tag = " + tag);
      }

      TableDesc keyTableDesc = conf.getKeySerializeInfo();
      keySerializer = (Serializer) keyTableDesc.getDeserializerClass().newInstance();
      keySerializer.initialize(null, keyTableDesc.getProperties());
      keyIsText = keySerializer.getSerializedClass().equals(Text.class);

      TableDesc valueTableDesc = conf.getValueSerializeInfo();
      valueSerializer = (Serializer) valueTableDesc.getDeserializerClass().newInstance();
      valueSerializer.initialize(null, valueTableDesc.getProperties());

      int limit = conf.getTopN();
      float memUsage = conf.getTopNMemoryUsage();

      if (limit >= 0 && memUsage > 0) {
        reducerHash = conf.isPTFReduceSink() ? new PTFTopNHash() : reducerHash;
        reducerHash.initialize(limit, memUsage, conf.isMapGroupBy(), this);
      }

      useUniformHash = conf.getReducerTraits().contains(UNIFORM);

      firstRow = true;
    } catch (Exception e) {
      String msg = "Error initializing ReduceSinkOperator: " + e.getMessage();
      LOG.error(msg, e);
      throw new RuntimeException(e);
    }
  }