Examples with RowSchema org.apache.hadoop.hive.ql.exec.RowSchema used on opensource projects

Example 11 with RowSchema

use of org.apache.hadoop.hive.ql.exec.RowSchema in project hive by apache.

the class HiveGBOpConvUtil method genMapSideGB.

@SuppressWarnings("unchecked")
private static OpAttr genMapSideGB(OpAttr inputOpAf, GBInfo gbAttrs) throws SemanticException {
    ArrayList<String> outputColNames = new ArrayList<String>();
    ArrayList<ColumnInfo> colInfoLst = new ArrayList<ColumnInfo>();
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    Set<String> gbKeyColsAsNamesFrmIn = new HashSet<String>();
    String colOutputName = null;
    // 1. Build GB Keys, grouping set starting position
    // 1.1 First Add original GB Keys
    ArrayList<ExprNodeDesc> gbKeys = new ArrayList<ExprNodeDesc>();
    for (int i = 0; i < gbAttrs.gbKeys.size(); i++) {
        gbKeys.add(gbAttrs.gbKeys.get(i));
        colOutputName = SemanticAnalyzer.getColumnInternalName(i);
        colInfoLst.add(new ColumnInfo(colOutputName, gbAttrs.gbKeyTypes.get(i), "", false));
        outputColNames.add(colOutputName);
        gbKeyColsAsNamesFrmIn.add(gbAttrs.gbKeyColNamesInInput.get(i));
        colExprMap.put(colOutputName, gbKeys.get(i));
    }
    // 1.2. Adjust GroupingSet Position, GBKeys for GroupingSet Position if
    // needed. NOTE: GroupingID is added to map side GB only if we don't GrpSet
    // doesn't require additional MR Jobs
    int groupingSetsPosition = -1;
    boolean inclGrpID = inclGrpSetInMapSide(gbAttrs);
    if (inclGrpID) {
        groupingSetsPosition = gbKeys.size();
        addGrpSetCol(true, null, false, gbKeys, outputColNames, colInfoLst, colExprMap);
    }
    // gen would have prevented it)
    for (int i = 0; i < gbAttrs.distExprNodes.size(); i++) {
        if (!gbKeyColsAsNamesFrmIn.contains(gbAttrs.distExprNames.get(i))) {
            gbKeys.add(gbAttrs.distExprNodes.get(i));
            colOutputName = SemanticAnalyzer.getColumnInternalName(gbKeys.size() - 1);
            colInfoLst.add(new ColumnInfo(colOutputName, gbAttrs.distExprTypes.get(i), "", false));
            outputColNames.add(colOutputName);
            gbKeyColsAsNamesFrmIn.add(gbAttrs.distExprNames.get(i));
            colExprMap.put(colOutputName, gbKeys.get(gbKeys.size() - 1));
        }
    }
    // 2. Build Aggregations
    ArrayList<AggregationDesc> aggregations = new ArrayList<AggregationDesc>();
    for (UDAFAttrs udafAttr : gbAttrs.udafAttrs) {
        Mode amode = SemanticAnalyzer.groupByDescModeToUDAFMode(GroupByDesc.Mode.HASH, udafAttr.isDistinctUDAF);
        aggregations.add(new AggregationDesc(udafAttr.udafName.toLowerCase(), udafAttr.udafEvaluator, udafAttr.udafParams, udafAttr.isDistinctUDAF, amode));
        GenericUDAFInfo udafInfo;
        try {
            udafInfo = SemanticAnalyzer.getGenericUDAFInfo(udafAttr.udafEvaluator, amode, udafAttr.udafParams);
        } catch (SemanticException e) {
            throw new RuntimeException(e);
        }
        colOutputName = SemanticAnalyzer.getColumnInternalName(gbKeys.size() + aggregations.size() - 1);
        colInfoLst.add(new ColumnInfo(colOutputName, udafInfo.returnType, "", false));
        outputColNames.add(colOutputName);
    }
    // 3. Create GB
    @SuppressWarnings("rawtypes") Operator gbOp = OperatorFactory.getAndMakeChild(new GroupByDesc(GroupByDesc.Mode.HASH, outputColNames, gbKeys, aggregations, false, gbAttrs.groupByMemoryUsage, gbAttrs.memoryThreshold, gbAttrs.grpSets, inclGrpID, groupingSetsPosition, gbAttrs.containsDistinctAggr), new RowSchema(colInfoLst), inputOpAf.inputs.get(0));
    // 5. Setup Expr Col Map
    // NOTE: UDAF is not included in ExprColMap
    gbOp.setColumnExprMap(colExprMap);
    return new OpAttr("", new HashSet<Integer>(), gbOp);
}

Example 12 with RowSchema

use of org.apache.hadoop.hive.ql.exec.RowSchema in project hive by apache.

the class HiveGBOpConvUtil method genReduceGBRS.

private static OpAttr genReduceGBRS(OpAttr inputOpAf, GBInfo gbInfo) throws SemanticException {
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    ArrayList<String> outputColumnNames = new ArrayList<String>();
    ArrayList<ColumnInfo> colInfoLst = new ArrayList<ColumnInfo>();
    GroupByOperator reduceSideGB1 = (GroupByOperator) inputOpAf.inputs.get(0);
    List<ColumnInfo> gb1ColInfoLst = reduceSideGB1.getSchema().getSignature();
    ArrayList<ExprNodeDesc> reduceKeys = getReduceKeysForRS(reduceSideGB1, 0, gbInfo.gbKeys.size() - 1, outputColumnNames, false, colInfoLst, colExprMap, true, true);
    if (inclGrpSetInReduceSide(gbInfo)) {
        addGrpSetCol(false, gb1ColInfoLst.get(reduceKeys.size()).getInternalName(), true, reduceKeys, outputColumnNames, colInfoLst, colExprMap);
    }
    ArrayList<ExprNodeDesc> reduceValues = getValueKeysForRS(reduceSideGB1, reduceSideGB1.getConf().getKeys().size(), outputColumnNames, colInfoLst, colExprMap, true, true);
    ReduceSinkOperator rsOp = (ReduceSinkOperator) OperatorFactory.getAndMakeChild(PlanUtils.getReduceSinkDesc(reduceKeys, reduceValues, outputColumnNames, true, -1, getNumPartFieldsForReduceSideRS(gbInfo), getParallelismForReduceSideRS(gbInfo), AcidUtils.Operation.NOT_ACID), new RowSchema(colInfoLst), reduceSideGB1);
    rsOp.setColumnExprMap(colExprMap);
    return new OpAttr("", new HashSet<Integer>(), rsOp);
}

Example 13 with RowSchema

use of org.apache.hadoop.hive.ql.exec.RowSchema in project hive by apache.

the class GenTezWork method process.

@Override
public Object process(Node nd, Stack<Node> stack, NodeProcessorCtx procContext, Object... nodeOutputs) throws SemanticException {
    GenTezProcContext context = (GenTezProcContext) procContext;
    assert context != null && context.currentTask != null && context.currentRootOperator != null;
    // Operator is a file sink or reduce sink. Something that forces
    // a new vertex.
    Operator<?> operator = (Operator<?>) nd;
    // root is the start of the operator pipeline we're currently
    // packing into a vertex, typically a table scan, union or join
    Operator<?> root = context.currentRootOperator;
    LOG.debug("Root operator: " + root);
    LOG.debug("Leaf operator: " + operator);
    if (context.clonedReduceSinks.contains(operator)) {
        // just skip and keep going
        return null;
    }
    TezWork tezWork = context.currentTask.getWork();
    // Right now the work graph is pretty simple. If there is no
    // Preceding work we have a root and will generate a map
    // vertex. If there is a preceding work we will generate
    // a reduce vertex
    BaseWork work;
    if (context.rootToWorkMap.containsKey(root)) {
        // will result into a vertex with multiple FS or RS operators.
        if (context.childToWorkMap.containsKey(operator)) {
            // if we've seen both root and child, we can bail.
            // clear out the mapjoin set. we don't need it anymore.
            context.currentMapJoinOperators.clear();
            // clear out the union set. we don't need it anymore.
            context.currentUnionOperators.clear();
            return null;
        } else {
            // At this point we don't have to do anything special. Just
            // run through the regular paces w/o creating a new task.
            work = context.rootToWorkMap.get(root);
        }
    } else {
        // create a new vertex
        if (context.preceedingWork == null) {
            work = utils.createMapWork(context, root, tezWork, null);
        } else {
            work = GenTezUtils.createReduceWork(context, root, tezWork);
        }
        context.rootToWorkMap.put(root, work);
    }
    // this is where we set the sort columns that we will be using for KeyValueInputMerge
    if (operator instanceof DummyStoreOperator) {
        work.addSortCols(root.getOpTraits().getSortCols().get(0));
    }
    if (!context.childToWorkMap.containsKey(operator)) {
        List<BaseWork> workItems = new LinkedList<BaseWork>();
        workItems.add(work);
        context.childToWorkMap.put(operator, workItems);
    } else {
        context.childToWorkMap.get(operator).add(work);
    }
    // which can affect the working of all downstream transformations.
    if (context.currentMergeJoinOperator != null) {
        // we are currently walking the big table side of the merge join. we need to create or hook up
        // merge join work.
        MergeJoinWork mergeJoinWork = null;
        if (context.opMergeJoinWorkMap.containsKey(context.currentMergeJoinOperator)) {
            // we have found a merge work corresponding to this closing operator. Hook up this work.
            mergeJoinWork = context.opMergeJoinWorkMap.get(context.currentMergeJoinOperator);
        } else {
            // we need to create the merge join work
            mergeJoinWork = new MergeJoinWork();
            mergeJoinWork.setMergeJoinOperator(context.currentMergeJoinOperator);
            tezWork.add(mergeJoinWork);
            context.opMergeJoinWorkMap.put(context.currentMergeJoinOperator, mergeJoinWork);
        }
        // connect the work correctly.
        work.addSortCols(root.getOpTraits().getSortCols().get(0));
        mergeJoinWork.addMergedWork(work, null, context.leafOperatorToFollowingWork);
        Operator<? extends OperatorDesc> parentOp = getParentFromStack(context.currentMergeJoinOperator, stack);
        // Set the big table position. Both the reduce work and merge join operator
        // should be set with the same value.
        //      int pos = context.currentMergeJoinOperator.getTagForOperator(parentOp);
        int pos = context.currentMergeJoinOperator.getConf().getBigTablePosition();
        work.setTag(pos);
        context.currentMergeJoinOperator.getConf().setBigTablePosition(pos);
        tezWork.setVertexType(work, VertexType.MULTI_INPUT_UNINITIALIZED_EDGES);
        for (BaseWork parentWork : tezWork.getParents(work)) {
            TezEdgeProperty edgeProp = tezWork.getEdgeProperty(parentWork, work);
            tezWork.disconnect(parentWork, work);
            tezWork.connect(parentWork, mergeJoinWork, edgeProp);
        }
        for (BaseWork childWork : tezWork.getChildren(work)) {
            TezEdgeProperty edgeProp = tezWork.getEdgeProperty(work, childWork);
            tezWork.disconnect(work, childWork);
            tezWork.connect(mergeJoinWork, childWork, edgeProp);
        }
        tezWork.remove(work);
        context.rootToWorkMap.put(root, mergeJoinWork);
        context.childToWorkMap.get(operator).remove(work);
        context.childToWorkMap.get(operator).add(mergeJoinWork);
        work = mergeJoinWork;
        context.currentMergeJoinOperator = null;
    }
    // remember which mapjoin operator links with which work
    if (!context.currentMapJoinOperators.isEmpty()) {
        for (MapJoinOperator mj : context.currentMapJoinOperators) {
            // so we can later run the same logic that is run in ReduceSinkMapJoinProc.
            if (mj.getConf().isDynamicPartitionHashJoin()) {
                // Since this is a dynamic partitioned hash join, the work for this join should be a ReduceWork
                ReduceWork reduceWork = (ReduceWork) work;
                int bigTablePosition = mj.getConf().getPosBigTable();
                reduceWork.setTag(bigTablePosition);
                // Use context.mapJoinParentMap to get the original RS parents, because
                // the MapJoin's parents may have been replaced by dummy operator.
                List<Operator<?>> mapJoinOriginalParents = context.mapJoinParentMap.get(mj);
                if (mapJoinOriginalParents == null) {
                    throw new SemanticException("Unexpected error - context.mapJoinParentMap did not have an entry for " + mj);
                }
                for (int pos = 0; pos < mapJoinOriginalParents.size(); ++pos) {
                    // This processing only needs to happen for the small tables
                    if (pos == bigTablePosition) {
                        continue;
                    }
                    Operator<?> parentOp = mapJoinOriginalParents.get(pos);
                    context.smallTableParentToMapJoinMap.put(parentOp, mj);
                    ReduceSinkOperator parentRS = (ReduceSinkOperator) parentOp;
                    // TableDesc needed for dynamic partitioned hash join
                    GenMapRedUtils.setKeyAndValueDesc(reduceWork, parentRS);
                    // has its ReduceSink parent removed.
                    if (!context.mapJoinToUnprocessedSmallTableReduceSinks.get(mj).contains(parentRS)) {
                        // This reduce sink has been processed already, so the work for the parentRS exists
                        BaseWork parentWork = ReduceSinkMapJoinProc.getMapJoinParentWork(context, parentRS);
                        int tag = parentRS.getConf().getTag();
                        tag = (tag == -1 ? 0 : tag);
                        reduceWork.getTagToInput().put(tag, parentWork.getName());
                    }
                }
            }
            LOG.debug("Processing map join: " + mj);
            // mapjoin later
            if (!context.mapJoinWorkMap.containsKey(mj)) {
                List<BaseWork> workItems = new LinkedList<BaseWork>();
                workItems.add(work);
                context.mapJoinWorkMap.put(mj, workItems);
            } else {
                context.mapJoinWorkMap.get(mj).add(work);
            }
            /*
         * this happens in case of map join operations.
         * The tree looks like this:
         *
         *        RS <--- we are here perhaps
         *        |
         *     MapJoin
         *     /     \
         *   RS       TS
         *  /
         * TS
         *
         * If we are at the RS pointed above, and we may have already visited the
         * RS following the TS, we have already generated work for the TS-RS.
         * We need to hook the current work to this generated work.
         */
            if (context.linkOpWithWorkMap.containsKey(mj)) {
                Map<BaseWork, TezEdgeProperty> linkWorkMap = context.linkOpWithWorkMap.get(mj);
                if (linkWorkMap != null) {
                    // Note: it's not quite clear why this is done inside this if. Seems like it should be on the top level.
                    if (context.linkChildOpWithDummyOp.containsKey(mj)) {
                        if (LOG.isDebugEnabled()) {
                            LOG.debug("Adding dummy ops to work: " + work.getName() + ": " + context.linkChildOpWithDummyOp.get(mj));
                        }
                        for (Operator<?> dummy : context.linkChildOpWithDummyOp.get(mj)) {
                            work.addDummyOp((HashTableDummyOperator) dummy);
                        }
                    }
                    for (Entry<BaseWork, TezEdgeProperty> parentWorkMap : linkWorkMap.entrySet()) {
                        BaseWork parentWork = parentWorkMap.getKey();
                        LOG.debug("connecting " + parentWork.getName() + " with " + work.getName());
                        TezEdgeProperty edgeProp = parentWorkMap.getValue();
                        tezWork.connect(parentWork, work, edgeProp);
                        if (edgeProp.getEdgeType() == EdgeType.CUSTOM_EDGE) {
                            tezWork.setVertexType(work, VertexType.INITIALIZED_EDGES);
                        }
                        // of the downstream work
                        for (ReduceSinkOperator r : context.linkWorkWithReduceSinkMap.get(parentWork)) {
                            if (r.getConf().getOutputName() != null) {
                                LOG.debug("Cloning reduce sink for multi-child broadcast edge");
                                // we've already set this one up. Need to clone for the next work.
                                r = (ReduceSinkOperator) OperatorFactory.getAndMakeChild(r.getCompilationOpContext(), (ReduceSinkDesc) r.getConf().clone(), new RowSchema(r.getSchema()), r.getParentOperators());
                                context.clonedReduceSinks.add(r);
                            }
                            r.getConf().setOutputName(work.getName());
                            context.connectedReduceSinks.add(r);
                        }
                    }
                }
            }
        }
        // clear out the set. we don't need it anymore.
        context.currentMapJoinOperators.clear();
    }
    // we might have to connect parent work with this work later.
    for (Operator<?> parent : new ArrayList<Operator<?>>(root.getParentOperators())) {
        if (LOG.isDebugEnabled()) {
            LOG.debug("Removing " + parent + " as parent from " + root);
        }
        context.leafOperatorToFollowingWork.remove(parent);
        context.leafOperatorToFollowingWork.put(parent, work);
        root.removeParent(parent);
    }
    if (!context.currentUnionOperators.isEmpty()) {
        // if there are union all operators, it means that the walking context contains union all operators.
        // please see more details of context.currentUnionOperator in GenTezWorkWalker
        UnionWork unionWork;
        if (context.unionWorkMap.containsKey(operator)) {
            // since we've passed this operator before.
            assert operator.getChildOperators().isEmpty();
            unionWork = (UnionWork) context.unionWorkMap.get(operator);
            // finally connect the union work with work
            connectUnionWorkWithWork(unionWork, work, tezWork, context);
        } else {
            // we've not seen this terminal before. we need to check
            // rootUnionWorkMap which contains the information of mapping the root
            // operator of a union work to a union work
            unionWork = context.rootUnionWorkMap.get(root);
            if (unionWork == null) {
                // if unionWork is null, it means it is the first time. we need to
                // create a union work object and add this work to it. Subsequent
                // work should reference the union and not the actual work.
                unionWork = GenTezUtils.createUnionWork(context, root, operator, tezWork);
                // finally connect the union work with work
                connectUnionWorkWithWork(unionWork, work, tezWork, context);
            }
        }
        context.currentUnionOperators.clear();
        work = unionWork;
    }
    // reasons. Roots are data sources, leaves are data sinks. I know.
    if (context.leafOperatorToFollowingWork.containsKey(operator)) {
        BaseWork followingWork = context.leafOperatorToFollowingWork.get(operator);
        long bytesPerReducer = context.conf.getLongVar(HiveConf.ConfVars.BYTESPERREDUCER);
        LOG.debug("Second pass. Leaf operator: " + operator + " has common downstream work:" + followingWork);
        if (operator instanceof DummyStoreOperator) {
            // this is the small table side.
            assert (followingWork instanceof MergeJoinWork);
            MergeJoinWork mergeJoinWork = (MergeJoinWork) followingWork;
            CommonMergeJoinOperator mergeJoinOp = mergeJoinWork.getMergeJoinOperator();
            work.setTag(mergeJoinOp.getTagForOperator(operator));
            mergeJoinWork.addMergedWork(null, work, context.leafOperatorToFollowingWork);
            tezWork.setVertexType(mergeJoinWork, VertexType.MULTI_INPUT_UNINITIALIZED_EDGES);
            for (BaseWork parentWork : tezWork.getParents(work)) {
                TezEdgeProperty edgeProp = tezWork.getEdgeProperty(parentWork, work);
                tezWork.disconnect(parentWork, work);
                tezWork.connect(parentWork, mergeJoinWork, edgeProp);
            }
            work = mergeJoinWork;
        } else {
            // need to add this branch to the key + value info
            assert operator instanceof ReduceSinkOperator && ((followingWork instanceof ReduceWork) || (followingWork instanceof MergeJoinWork) || followingWork instanceof UnionWork);
            ReduceSinkOperator rs = (ReduceSinkOperator) operator;
            ReduceWork rWork = null;
            if (followingWork instanceof MergeJoinWork) {
                MergeJoinWork mergeJoinWork = (MergeJoinWork) followingWork;
                rWork = (ReduceWork) mergeJoinWork.getMainWork();
            } else if (followingWork instanceof UnionWork) {
                // this can only be possible if there is merge work followed by the union
                UnionWork unionWork = (UnionWork) followingWork;
                int index = getFollowingWorkIndex(tezWork, unionWork, rs);
                BaseWork baseWork = tezWork.getChildren(unionWork).get(index);
                if (baseWork instanceof MergeJoinWork) {
                    MergeJoinWork mergeJoinWork = (MergeJoinWork) baseWork;
                    // disconnect the connection to union work and connect to merge work
                    followingWork = mergeJoinWork;
                    rWork = (ReduceWork) mergeJoinWork.getMainWork();
                } else {
                    rWork = (ReduceWork) baseWork;
                }
            } else {
                rWork = (ReduceWork) followingWork;
            }
            GenMapRedUtils.setKeyAndValueDesc(rWork, rs);
            // remember which parent belongs to which tag
            int tag = rs.getConf().getTag();
            rWork.getTagToInput().put(tag == -1 ? 0 : tag, work.getName());
            // remember the output name of the reduce sink
            rs.getConf().setOutputName(rWork.getName());
            // For dynamic partitioned hash join, run the ReduceSinkMapJoinProc logic for any
            // ReduceSink parents that we missed.
            MapJoinOperator mj = context.smallTableParentToMapJoinMap.get(rs);
            if (mj != null) {
                // Only need to run the logic for tables we missed
                if (context.mapJoinToUnprocessedSmallTableReduceSinks.get(mj).contains(rs)) {
                    // ReduceSinkMapJoinProc logic does not work unless the ReduceSink is connected as
                    // a parent of the MapJoin, but at this point we have already removed all of the
                    // parents from the MapJoin.
                    // Try temporarily adding the RS as a parent
                    ArrayList<Operator<?>> tempMJParents = new ArrayList<Operator<?>>();
                    tempMJParents.add(rs);
                    mj.setParentOperators(tempMJParents);
                    // ReduceSink also needs MapJoin as child
                    List<Operator<?>> rsChildren = rs.getChildOperators();
                    rsChildren.add(mj);
                    // Since the MapJoin has had all of its other parents removed at this point,
                    // it would be bad here if processReduceSinkToHashJoin() tries to do anything
                    // with the RS parent based on its position in the list of parents.
                    ReduceSinkMapJoinProc.processReduceSinkToHashJoin(rs, mj, context);
                    // Remove any parents from MapJoin again
                    mj.removeParents();
                // TODO: do we also need to remove the MapJoin from the list of RS's children?
                }
            }
            if (!context.connectedReduceSinks.contains(rs)) {
                // add dependency between the two work items
                TezEdgeProperty edgeProp;
                EdgeType edgeType = GenTezUtils.determineEdgeType(work, followingWork, rs);
                if (rWork.isAutoReduceParallelism()) {
                    edgeProp = new TezEdgeProperty(context.conf, edgeType, true, rWork.getMinReduceTasks(), rWork.getMaxReduceTasks(), bytesPerReducer);
                } else {
                    edgeProp = new TezEdgeProperty(edgeType);
                }
                tezWork.connect(work, followingWork, edgeProp);
                context.connectedReduceSinks.add(rs);
            }
        }
    } else {
        LOG.debug("First pass. Leaf operator: " + operator);
    }
    // the next item will be a new root.
    if (!operator.getChildOperators().isEmpty()) {
        assert operator.getChildOperators().size() == 1;
        context.parentOfRoot = operator;
        context.currentRootOperator = operator.getChildOperators().get(0);
        context.preceedingWork = work;
    }
    return null;
}

Example 14 with RowSchema

use of org.apache.hadoop.hive.ql.exec.RowSchema in project hive by apache.

the class SemanticAnalyzer method genGroupByPlanGroupByOperator2MR.

/**
   * Generate the second GroupByOperator for the Group By Plan
   * (parseInfo.getXXX(dest)). The new GroupByOperator will do the second
   * aggregation based on the partial aggregation results.
   *
   * @param mode
   *          the mode of aggregation (FINAL)
   * @param genericUDAFEvaluators
   *          The mapping from Aggregation StringTree to the
   *          genericUDAFEvaluator.
   * @return the new GroupByOperator
   * @throws SemanticException
   */
@SuppressWarnings("nls")
private Operator genGroupByPlanGroupByOperator2MR(QBParseInfo parseInfo, String dest, Operator reduceSinkOperatorInfo2, GroupByDesc.Mode mode, Map<String, GenericUDAFEvaluator> genericUDAFEvaluators, boolean groupingSetsPresent) throws SemanticException {
    RowResolver groupByInputRowResolver2 = opParseCtx.get(reduceSinkOperatorInfo2).getRowResolver();
    RowResolver groupByOutputRowResolver2 = new RowResolver();
    groupByOutputRowResolver2.setIsExprResolver(true);
    ArrayList<ExprNodeDesc> groupByKeys = new ArrayList<ExprNodeDesc>();
    ArrayList<AggregationDesc> aggregations = new ArrayList<AggregationDesc>();
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    List<ASTNode> grpByExprs = getGroupByForClause(parseInfo, dest);
    ArrayList<String> outputColumnNames = new ArrayList<String>();
    for (int i = 0; i < grpByExprs.size(); ++i) {
        ASTNode grpbyExpr = grpByExprs.get(i);
        ColumnInfo exprInfo = groupByInputRowResolver2.getExpression(grpbyExpr);
        if (exprInfo == null) {
            throw new SemanticException(ErrorMsg.INVALID_COLUMN.getMsg(grpbyExpr));
        }
        String expression = exprInfo.getInternalName();
        groupByKeys.add(new ExprNodeColumnDesc(exprInfo.getType(), expression, exprInfo.getTabAlias(), exprInfo.getIsVirtualCol()));
        String field = getColumnInternalName(i);
        outputColumnNames.add(field);
        ColumnInfo oColInfo = new ColumnInfo(field, exprInfo.getType(), "", false);
        groupByOutputRowResolver2.putExpression(grpbyExpr, oColInfo);
        addAlternateGByKeyMappings(grpbyExpr, oColInfo, reduceSinkOperatorInfo2, groupByOutputRowResolver2);
        colExprMap.put(field, groupByKeys.get(groupByKeys.size() - 1));
    }
    int groupingSetsPosition = -1;
    // For grouping sets, add a dummy grouping key
    if (groupingSetsPresent) {
        groupingSetsPosition = groupByKeys.size();
        addGroupingSetKey(groupByKeys, groupByInputRowResolver2, groupByOutputRowResolver2, outputColumnNames, colExprMap);
    }
    HashMap<String, ASTNode> aggregationTrees = parseInfo.getAggregationExprsForClause(dest);
    boolean containsDistinctAggr = false;
    for (Map.Entry<String, ASTNode> entry : aggregationTrees.entrySet()) {
        ArrayList<ExprNodeDesc> aggParameters = new ArrayList<ExprNodeDesc>();
        ASTNode value = entry.getValue();
        ColumnInfo paraExprInfo = groupByInputRowResolver2.getExpression(value);
        if (paraExprInfo == null) {
            throw new SemanticException(ErrorMsg.INVALID_COLUMN.getMsg(value));
        }
        String paraExpression = paraExprInfo.getInternalName();
        assert (paraExpression != null);
        aggParameters.add(new ExprNodeColumnDesc(paraExprInfo.getType(), paraExpression, paraExprInfo.getTabAlias(), paraExprInfo.getIsVirtualCol()));
        String aggName = unescapeIdentifier(value.getChild(0).getText());
        boolean isDistinct = value.getType() == HiveParser.TOK_FUNCTIONDI;
        containsDistinctAggr = containsDistinctAggr || isDistinct;
        boolean isStar = value.getType() == HiveParser.TOK_FUNCTIONSTAR;
        Mode amode = groupByDescModeToUDAFMode(mode, isDistinct);
        GenericUDAFEvaluator genericUDAFEvaluator = genericUDAFEvaluators.get(entry.getKey());
        assert (genericUDAFEvaluator != null);
        GenericUDAFInfo udaf = getGenericUDAFInfo(genericUDAFEvaluator, amode, aggParameters);
        aggregations.add(new AggregationDesc(aggName.toLowerCase(), udaf.genericUDAFEvaluator, udaf.convertedParameters, (mode != GroupByDesc.Mode.FINAL && value.getToken().getType() == HiveParser.TOK_FUNCTIONDI), amode));
        String field = getColumnInternalName(groupByKeys.size() + aggregations.size() - 1);
        outputColumnNames.add(field);
        groupByOutputRowResolver2.putExpression(value, new ColumnInfo(field, udaf.returnType, "", false));
    }
    float groupByMemoryUsage = HiveConf.getFloatVar(conf, HiveConf.ConfVars.HIVEMAPAGGRHASHMEMORY);
    float memoryThreshold = HiveConf.getFloatVar(conf, HiveConf.ConfVars.HIVEMAPAGGRMEMORYTHRESHOLD);
    Operator op = putOpInsertMap(OperatorFactory.getAndMakeChild(new GroupByDesc(mode, outputColumnNames, groupByKeys, aggregations, false, groupByMemoryUsage, memoryThreshold, null, false, groupingSetsPosition, containsDistinctAggr), new RowSchema(groupByOutputRowResolver2.getColumnInfos()), reduceSinkOperatorInfo2), groupByOutputRowResolver2);
    op.setColumnExprMap(colExprMap);
    return op;
}

Example 15 with RowSchema

use of org.apache.hadoop.hive.ql.exec.RowSchema in project hive by apache.

the class ColumnPrunerProcCtx method handleFilterUnionChildren.

/**
   * If the input filter operator has direct child(ren) which are union operator,
   * and the filter's column is not the same as union's
   * create select operator between them. The select operator has same number of columns as
   * pruned child operator.
   *
   * @param curOp
   *          The filter operator which need to handle children.
   * @throws SemanticException
   */
public void handleFilterUnionChildren(Operator<? extends OperatorDesc> curOp) throws SemanticException {
    if (curOp.getChildOperators() == null || !(curOp instanceof FilterOperator)) {
        return;
    }
    List<FieldNode> parentPrunList = prunedColLists.get(curOp);
    if (parentPrunList == null || parentPrunList.size() == 0) {
        return;
    }
    List<FieldNode> prunList = null;
    for (Operator<? extends OperatorDesc> child : curOp.getChildOperators()) {
        if (child instanceof UnionOperator) {
            prunList = genColLists(child);
            if (prunList == null || prunList.size() == 0 || parentPrunList.size() == prunList.size()) {
                continue;
            }
            ArrayList<ExprNodeDesc> exprs = new ArrayList<ExprNodeDesc>();
            ArrayList<String> outputColNames = new ArrayList<String>();
            Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
            ArrayList<ColumnInfo> outputRS = new ArrayList<ColumnInfo>();
            for (ColumnInfo colInfo : child.getSchema().getSignature()) {
                if (lookupColumn(prunList, colInfo.getInternalName()) == null) {
                    continue;
                }
                ExprNodeDesc colDesc = new ExprNodeColumnDesc(colInfo.getType(), colInfo.getInternalName(), colInfo.getTabAlias(), colInfo.getIsVirtualCol());
                exprs.add(colDesc);
                outputColNames.add(colInfo.getInternalName());
                ColumnInfo newCol = new ColumnInfo(colInfo.getInternalName(), colInfo.getType(), colInfo.getTabAlias(), colInfo.getIsVirtualCol(), colInfo.isHiddenVirtualCol());
                newCol.setAlias(colInfo.getAlias());
                outputRS.add(newCol);
                colExprMap.put(colInfo.getInternalName(), colDesc);
            }
            SelectDesc select = new SelectDesc(exprs, outputColNames, false);
            curOp.removeChild(child);
            SelectOperator sel = (SelectOperator) OperatorFactory.getAndMakeChild(select, new RowSchema(outputRS), curOp);
            OperatorFactory.makeChild(sel, child);
            sel.setColumnExprMap(colExprMap);
        }
    }
}