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

Example 1 with RowSchema

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

the class SyntheticJoinPredicate method createFilter.

// insert filter operator between target(child) and input(parent)
private static Operator<FilterDesc> createFilter(Operator<?> target, Operator<?> parent, RowSchema parentRS, ExprNodeDesc filterExpr) {
    FilterDesc filterDesc = new FilterDesc(filterExpr, false);
    filterDesc.setSyntheticJoinPredicate(true);
    Operator<FilterDesc> filter = OperatorFactory.get(parent.getCompilationOpContext(), filterDesc, new RowSchema(parentRS.getSignature()));
    filter.getParentOperators().add(parent);
    filter.getChildOperators().add(target);
    parent.replaceChild(target, filter);
    target.replaceParent(parent, filter);
    return filter;
}

Example 2 with RowSchema

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

the class PredicateTransitivePropagate method createFilter.

// insert filter operator between target(child) and input(parent)
private Operator<FilterDesc> createFilter(Operator<?> target, Operator<?> parent, RowSchema parentRS, ExprNodeDesc filterExpr) {
    Operator<FilterDesc> filter = OperatorFactory.get(parent.getCompilationOpContext(), new FilterDesc(filterExpr, false), new RowSchema(parentRS.getSignature()));
    filter.getParentOperators().add(parent);
    filter.getChildOperators().add(target);
    parent.replaceChild(target, filter);
    target.replaceParent(parent, filter);
    return filter;
}

Example 3 with RowSchema

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

the class PredicateTransitivePropagate method transform.

@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    pGraphContext = pctx;
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", "(" + FilterOperator.getOperatorName() + "%" + ReduceSinkOperator.getOperatorName() + "%" + JoinOperator.getOperatorName() + "%)"), new JoinTransitive());
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    TransitiveContext context = new TransitiveContext();
    Dispatcher disp = new DefaultRuleDispatcher(null, opRules, context);
    GraphWalker ogw = new LevelOrderWalker(disp, 2);
    // Create a list of topop nodes
    List<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pGraphContext.getTopOps().values());
    ogw.startWalking(topNodes, null);
    Map<ReduceSinkOperator, List<ExprNodeDesc>> newFilters = context.getNewfilters();
    // insert new filter between RS and parent of RS
    for (Map.Entry<ReduceSinkOperator, List<ExprNodeDesc>> entry : newFilters.entrySet()) {
        ReduceSinkOperator reducer = entry.getKey();
        Operator<?> parent = reducer.getParentOperators().get(0);
        List<ExprNodeDesc> exprs = entry.getValue();
        if (parent instanceof FilterOperator) {
            exprs = ExprNodeDescUtils.split(((FilterOperator) parent).getConf().getPredicate(), exprs);
            ExprNodeDesc merged = ExprNodeDescUtils.mergePredicates(exprs);
            ((FilterOperator) parent).getConf().setPredicate(merged);
        } else {
            ExprNodeDesc merged = ExprNodeDescUtils.mergePredicates(exprs);
            RowSchema parentRS = parent.getSchema();
            Operator<FilterDesc> newFilter = createFilter(reducer, parent, parentRS, merged);
        }
    }
    return pGraphContext;
}

Example 4 with RowSchema

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

the class ProjectionPusher method pushProjectionsAndFilters.

private void pushProjectionsAndFilters(final JobConf jobConf, final String splitPath, final String splitPathWithNoSchema) {
    if (mapWork == null) {
        return;
    } else if (mapWork.getPathToAliases() == null) {
        return;
    }
    final Set<String> aliases = new HashSet<String>();
    final Iterator<Entry<Path, ArrayList<String>>> iterator = mapWork.getPathToAliases().entrySet().iterator();
    while (iterator.hasNext()) {
        final Entry<Path, ArrayList<String>> entry = iterator.next();
        final String key = entry.getKey().toUri().getPath();
        if (splitPath.equals(key) || splitPathWithNoSchema.equals(key)) {
            aliases.addAll(entry.getValue());
        }
    }
    // Collect the needed columns from all the aliases and create ORed filter
    // expression for the table.
    boolean allColumnsNeeded = false;
    boolean noFilters = false;
    Set<Integer> neededColumnIDs = new HashSet<Integer>();
    // To support nested column pruning, we need to track the path from the top to the nested
    // fields
    Set<String> neededNestedColumnPaths = new HashSet<String>();
    List<ExprNodeGenericFuncDesc> filterExprs = new ArrayList<ExprNodeGenericFuncDesc>();
    RowSchema rowSchema = null;
    for (String alias : aliases) {
        final Operator<? extends Serializable> op = mapWork.getAliasToWork().get(alias);
        if (op != null && op instanceof TableScanOperator) {
            final TableScanOperator ts = (TableScanOperator) op;
            if (ts.getNeededColumnIDs() == null) {
                allColumnsNeeded = true;
            } else {
                neededColumnIDs.addAll(ts.getNeededColumnIDs());
                neededNestedColumnPaths.addAll(ts.getNeededNestedColumnPaths());
            }
            rowSchema = ts.getSchema();
            ExprNodeGenericFuncDesc filterExpr = ts.getConf() == null ? null : ts.getConf().getFilterExpr();
            // No filter if any TS has no filter expression
            noFilters = filterExpr == null;
            filterExprs.add(filterExpr);
        }
    }
    ExprNodeGenericFuncDesc tableFilterExpr = null;
    if (!noFilters) {
        try {
            for (ExprNodeGenericFuncDesc filterExpr : filterExprs) {
                if (tableFilterExpr == null) {
                    tableFilterExpr = filterExpr;
                } else {
                    tableFilterExpr = ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPOr(), Arrays.<ExprNodeDesc>asList(tableFilterExpr, filterExpr));
                }
            }
        } catch (UDFArgumentException ex) {
            LOG.debug("Turn off filtering due to " + ex);
            tableFilterExpr = null;
        }
    }
    // push down projections
    if (!allColumnsNeeded) {
        if (!neededColumnIDs.isEmpty()) {
            ColumnProjectionUtils.appendReadColumns(jobConf, new ArrayList<Integer>(neededColumnIDs));
            ColumnProjectionUtils.appendNestedColumnPaths(jobConf, new ArrayList<String>(neededNestedColumnPaths));
        }
    } else {
        ColumnProjectionUtils.setReadAllColumns(jobConf);
    }
    pushFilters(jobConf, rowSchema, tableFilterExpr);
}

Example 5 with RowSchema

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

the class MapJoinProcessor method genSelectPlan.

protected void genSelectPlan(ParseContext pctx, MapJoinOperator input) throws SemanticException {
    List<Operator<? extends OperatorDesc>> childOps = input.getChildOperators();
    input.setChildOperators(null);
    // create a dummy select - This select is needed by the walker to split the
    // mapJoin later on
    RowSchema inputRS = input.getSchema();
    ArrayList<ExprNodeDesc> exprs = new ArrayList<ExprNodeDesc>();
    ArrayList<String> outputs = new ArrayList<String>();
    List<String> outputCols = input.getConf().getOutputColumnNames();
    ArrayList<ColumnInfo> outputRS = new ArrayList<ColumnInfo>();
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    for (int i = 0; i < outputCols.size(); i++) {
        String internalName = outputCols.get(i);
        ColumnInfo valueInfo = inputRS.getColumnInfo(internalName);
        ExprNodeDesc colDesc = new ExprNodeColumnDesc(valueInfo.getType(), valueInfo.getInternalName(), valueInfo.getTabAlias(), valueInfo.getIsVirtualCol());
        exprs.add(colDesc);
        outputs.add(internalName);
        ColumnInfo newCol = new ColumnInfo(internalName, valueInfo.getType(), valueInfo.getTabAlias(), valueInfo.getIsVirtualCol(), valueInfo.isHiddenVirtualCol());
        newCol.setAlias(valueInfo.getAlias());
        outputRS.add(newCol);
        colExprMap.put(internalName, colDesc);
    }
    SelectDesc select = new SelectDesc(exprs, outputs, false);
    SelectOperator sel = (SelectOperator) OperatorFactory.getAndMakeChild(select, new RowSchema(outputRS), input);
    sel.setColumnExprMap(colExprMap);
    // Insert the select operator in between.
    sel.setChildOperators(childOps);
    for (Operator<? extends OperatorDesc> ch : childOps) {
        ch.replaceParent(input, sel);
    }
}