Examples with GenericUDF org.apache.hadoop.hive.ql.udf.generic.GenericUDF used on opensource projects

Example 11 with GenericUDF

use of org.apache.hadoop.hive.ql.udf.generic.GenericUDF in project phoenix by apache.

the class IndexPredicateAnalyzer method analyzeExpr.

private ExprNodeDesc analyzeExpr(ExprNodeGenericFuncDesc expr, List<IndexSearchCondition> searchConditions, Object... nodeOutputs) throws SemanticException {
    if (FunctionRegistry.isOpAnd(expr)) {
        assert (nodeOutputs.length == 2);
        ExprNodeDesc residual1 = (ExprNodeDesc) nodeOutputs[0];
        ExprNodeDesc residual2 = (ExprNodeDesc) nodeOutputs[1];
        if (residual1 == null) {
            return residual2;
        }
        if (residual2 == null) {
            return residual1;
        }
        List<ExprNodeDesc> residuals = new ArrayList<ExprNodeDesc>();
        residuals.add(residual1);
        residuals.add(residual2);
        return new ExprNodeGenericFuncDesc(TypeInfoFactory.booleanTypeInfo, FunctionRegistry.getGenericUDFForAnd(), residuals);
    }
    GenericUDF genericUDF = expr.getGenericUDF();
    if (!(genericUDF instanceof GenericUDFBaseCompare)) {
        // 2015-10-22 Added by JeongMin Ju : Processing Between/In Operator
        if (genericUDF instanceof GenericUDFBetween) {
            // In case of not between, The value of first element of nodeOutputs is true.
            // otherwise false.
            processingBetweenOperator(expr, searchConditions, nodeOutputs);
            return expr;
        } else if (genericUDF instanceof GenericUDFIn) {
            // In case of not in operator, in operator exist as child of not operator.
            processingInOperator(expr, searchConditions, false, nodeOutputs);
            return expr;
        } else if (genericUDF instanceof GenericUDFOPNot && ((ExprNodeGenericFuncDesc) expr.getChildren().get(0)).getGenericUDF() instanceof GenericUDFIn) {
            // In case of not in operator, in operator exist as child of not operator.
            processingInOperator((ExprNodeGenericFuncDesc) expr.getChildren().get(0), searchConditions, true, ((ExprNodeGenericFuncDesc) nodeOutputs[0]).getChildren().toArray());
            return expr;
        } else if (genericUDF instanceof GenericUDFOPNull) {
            processingNullOperator(expr, searchConditions, nodeOutputs);
            return expr;
        } else if (genericUDF instanceof GenericUDFOPNotNull) {
            processingNotNullOperator(expr, searchConditions, nodeOutputs);
            return expr;
        } else {
            return expr;
        }
    }
    ExprNodeDesc expr1 = (ExprNodeDesc) nodeOutputs[0];
    ExprNodeDesc expr2 = (ExprNodeDesc) nodeOutputs[1];
    // user
    if (expr1.getTypeInfo().equals(expr2.getTypeInfo())) {
        expr1 = getColumnExpr(expr1);
        expr2 = getColumnExpr(expr2);
    }
    ExprNodeDesc[] extracted = ExprNodeDescUtils.extractComparePair(expr1, expr2);
    if (extracted == null || (extracted.length > 2 && !acceptsFields)) {
        return expr;
    }
    ExprNodeColumnDesc columnDesc;
    ExprNodeConstantDesc constantDesc;
    if (extracted[0] instanceof ExprNodeConstantDesc) {
        genericUDF = genericUDF.flip();
        columnDesc = (ExprNodeColumnDesc) extracted[1];
        constantDesc = (ExprNodeConstantDesc) extracted[0];
    } else {
        columnDesc = (ExprNodeColumnDesc) extracted[0];
        constantDesc = (ExprNodeConstantDesc) extracted[1];
    }
    Set<String> allowed = columnToUDFs.get(columnDesc.getColumn());
    if (allowed == null) {
        return expr;
    }
    String udfName = genericUDF.getUdfName();
    if (!allowed.contains(genericUDF.getUdfName())) {
        return expr;
    }
    String[] fields = null;
    if (extracted.length > 2) {
        ExprNodeFieldDesc fieldDesc = (ExprNodeFieldDesc) extracted[2];
        if (!isValidField(fieldDesc)) {
            return expr;
        }
        fields = ExprNodeDescUtils.extractFields(fieldDesc);
    }
    // We also need to update the expr so that the index query can be
    // generated.
    // Note that, hive does not support UDFToDouble etc in the query text.
    List<ExprNodeDesc> list = new ArrayList<ExprNodeDesc>();
    list.add(expr1);
    list.add(expr2);
    expr = new ExprNodeGenericFuncDesc(expr.getTypeInfo(), expr.getGenericUDF(), list);
    searchConditions.add(new IndexSearchCondition(columnDesc, udfName, constantDesc, expr, fields));
    // remove it from the residual predicate
    return fields == null ? null : expr;
}

Example 12 with GenericUDF

use of org.apache.hadoop.hive.ql.udf.generic.GenericUDF in project hive by apache.

the class ConstantPropagateProcFactory method foldExprShortcut.

/**
   * Fold input expression desc, only performing short-cutting.
   *
   * Unnecessary AND/OR operations involving a constant true/false value will be eliminated.
   *
   * @param desc folding expression
   * @param constants current propagated constant map
   * @param cppCtx
   * @param op processing operator
   * @param propagate if true, assignment expressions will be added to constants.
   * @return fold expression
   * @throws UDFArgumentException
   */
private static ExprNodeDesc foldExprShortcut(ExprNodeDesc desc, Map<ColumnInfo, ExprNodeDesc> constants, ConstantPropagateProcCtx cppCtx, Operator<? extends Serializable> op, int tag, boolean propagate) throws UDFArgumentException {
    // Combine NOT operator with the child operator. Otherwise, the following optimization
    // from bottom up could lead to incorrect result, such as not(x > 3 and x is not null),
    // should not be optimized to not(x > 3), but (x <=3 or x is null).
    desc = foldNegative(desc);
    if (desc instanceof ExprNodeGenericFuncDesc) {
        ExprNodeGenericFuncDesc funcDesc = (ExprNodeGenericFuncDesc) desc;
        GenericUDF udf = funcDesc.getGenericUDF();
        boolean propagateNext = propagate && propagatableUdfs.contains(udf.getClass());
        List<ExprNodeDesc> newExprs = new ArrayList<ExprNodeDesc>();
        for (ExprNodeDesc childExpr : desc.getChildren()) {
            newExprs.add(foldExpr(childExpr, constants, cppCtx, op, tag, propagateNext));
        }
        // Don't evaluate nondeterministic function since the value can only calculate during runtime.
        if (!isDeterministicUdf(udf, newExprs)) {
            if (LOG.isDebugEnabled()) {
                LOG.debug("Function " + udf.getClass() + " is undeterministic. Don't evaluate immediately.");
            }
            ((ExprNodeGenericFuncDesc) desc).setChildren(newExprs);
            return desc;
        }
        // Check if the function can be short cut.
        ExprNodeDesc shortcut = shortcutFunction(udf, newExprs, op);
        if (shortcut != null) {
            if (LOG.isDebugEnabled()) {
                LOG.debug("Folding expression:" + desc + " -> " + shortcut);
            }
            return shortcut;
        }
        ((ExprNodeGenericFuncDesc) desc).setChildren(newExprs);
    }
    return desc;
}

Example 13 with GenericUDF

use of org.apache.hadoop.hive.ql.udf.generic.GenericUDF in project hive by apache.

the class ConstantPropagateProcFactory method shortcutFunction.

private static ExprNodeDesc shortcutFunction(GenericUDF udf, List<ExprNodeDesc> newExprs, Operator<? extends Serializable> op) throws UDFArgumentException {
    if (udf instanceof GenericUDFOPEqual) {
        assert newExprs.size() == 2;
        boolean foundUDFInFirst = false;
        ExprNodeGenericFuncDesc caseOrWhenexpr = null;
        if (newExprs.get(0) instanceof ExprNodeGenericFuncDesc) {
            caseOrWhenexpr = (ExprNodeGenericFuncDesc) newExprs.get(0);
            if (caseOrWhenexpr.getGenericUDF() instanceof GenericUDFWhen || caseOrWhenexpr.getGenericUDF() instanceof GenericUDFCase) {
                foundUDFInFirst = true;
            }
        }
        if (!foundUDFInFirst && newExprs.get(1) instanceof ExprNodeGenericFuncDesc) {
            caseOrWhenexpr = (ExprNodeGenericFuncDesc) newExprs.get(1);
            if (!(caseOrWhenexpr.getGenericUDF() instanceof GenericUDFWhen || caseOrWhenexpr.getGenericUDF() instanceof GenericUDFCase)) {
                return null;
            }
        }
        if (null == caseOrWhenexpr) {
            // we didn't find case or when udf
            return null;
        }
        GenericUDF childUDF = caseOrWhenexpr.getGenericUDF();
        List<ExprNodeDesc> children = caseOrWhenexpr.getChildren();
        int i;
        if (childUDF instanceof GenericUDFWhen) {
            for (i = 1; i < children.size(); i += 2) {
                children.set(i, ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPEqual(), Lists.newArrayList(children.get(i), newExprs.get(foundUDFInFirst ? 1 : 0))));
            }
            if (children.size() % 2 == 1) {
                i = children.size() - 1;
                children.set(i, ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPEqual(), Lists.newArrayList(children.get(i), newExprs.get(foundUDFInFirst ? 1 : 0))));
            }
            // after constant folding of child expression the return type of UDFWhen might have changed,
            // so recreate the expression
            ExprNodeGenericFuncDesc newCaseOrWhenExpr = ExprNodeGenericFuncDesc.newInstance(childUDF, caseOrWhenexpr.getFuncText(), children);
            return newCaseOrWhenExpr;
        } else if (childUDF instanceof GenericUDFCase) {
            for (i = 2; i < children.size(); i += 2) {
                children.set(i, ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPEqual(), Lists.newArrayList(children.get(i), newExprs.get(foundUDFInFirst ? 1 : 0))));
            }
            if (children.size() % 2 == 0) {
                i = children.size() - 1;
                children.set(i, ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPEqual(), Lists.newArrayList(children.get(i), newExprs.get(foundUDFInFirst ? 1 : 0))));
            }
            // after constant folding of child expression the return type of UDFCase might have changed,
            // so recreate the expression
            ExprNodeGenericFuncDesc newCaseOrWhenExpr = ExprNodeGenericFuncDesc.newInstance(childUDF, caseOrWhenexpr.getFuncText(), children);
            return newCaseOrWhenExpr;
        } else {
            // cant happen
            return null;
        }
    }
    if (udf instanceof GenericUDFOPAnd) {
        final BitSet positionsToRemove = new BitSet();
        final List<ExprNodeDesc> notNullExprs = new ArrayList<ExprNodeDesc>();
        final List<Integer> notNullExprsPositions = new ArrayList<Integer>();
        final List<ExprNodeDesc> compareExprs = new ArrayList<ExprNodeDesc>();
        for (int i = 0; i < newExprs.size(); i++) {
            ExprNodeDesc childExpr = newExprs.get(i);
            if (childExpr instanceof ExprNodeConstantDesc) {
                ExprNodeConstantDesc c = (ExprNodeConstantDesc) childExpr;
                if (Boolean.TRUE.equals(c.getValue())) {
                    // if true, prune it
                    positionsToRemove.set(i);
                } else {
                    // if false, return false
                    return childExpr;
                }
            } else if (childExpr instanceof ExprNodeGenericFuncDesc && ((ExprNodeGenericFuncDesc) childExpr).getGenericUDF() instanceof GenericUDFOPNotNull && childExpr.getChildren().get(0) instanceof ExprNodeColumnDesc) {
                notNullExprs.add(childExpr.getChildren().get(0));
                notNullExprsPositions.add(i);
            } else if (childExpr instanceof ExprNodeGenericFuncDesc && ((ExprNodeGenericFuncDesc) childExpr).getGenericUDF() instanceof GenericUDFBaseCompare && !(((ExprNodeGenericFuncDesc) childExpr).getGenericUDF() instanceof GenericUDFOPNotEqual) && childExpr.getChildren().size() == 2) {
                // Try to fold (key <op> 86) and (key is not null) to (key <op> 86)
                // where <op> can be "=", ">=", "<=", ">", "<".
                // Note: (key <> 86) and (key is not null) cannot be folded
                ExprNodeColumnDesc colDesc = ExprNodeDescUtils.getColumnExpr(childExpr.getChildren().get(0));
                if (null == colDesc) {
                    colDesc = ExprNodeDescUtils.getColumnExpr(childExpr.getChildren().get(1));
                }
                if (colDesc != null) {
                    compareExprs.add(colDesc);
                }
            }
        }
        // Try to fold (key = 86) and (key is not null) to (key = 86)
        for (int i = 0; i < notNullExprs.size(); i++) {
            for (ExprNodeDesc other : compareExprs) {
                if (notNullExprs.get(i).isSame(other)) {
                    positionsToRemove.set(notNullExprsPositions.get(i));
                    break;
                }
            }
        }
        // Remove unnecessary expressions
        int pos = 0;
        int removed = 0;
        while ((pos = positionsToRemove.nextSetBit(pos)) != -1) {
            newExprs.remove(pos - removed);
            pos++;
            removed++;
        }
        if (newExprs.size() == 0) {
            return new ExprNodeConstantDesc(TypeInfoFactory.booleanTypeInfo, Boolean.TRUE);
        }
        if (newExprs.size() == 1) {
            return newExprs.get(0);
        }
    }
    if (udf instanceof GenericUDFOPOr) {
        final BitSet positionsToRemove = new BitSet();
        for (int i = 0; i < newExprs.size(); i++) {
            ExprNodeDesc childExpr = newExprs.get(i);
            if (childExpr instanceof ExprNodeConstantDesc) {
                ExprNodeConstantDesc c = (ExprNodeConstantDesc) childExpr;
                if (Boolean.FALSE.equals(c.getValue())) {
                    // if false, prune it
                    positionsToRemove.set(i);
                } else if (Boolean.TRUE.equals(c.getValue())) {
                    // if true return true
                    return childExpr;
                }
            }
        }
        int pos = 0;
        int removed = 0;
        while ((pos = positionsToRemove.nextSetBit(pos)) != -1) {
            newExprs.remove(pos - removed);
            pos++;
            removed++;
        }
        if (newExprs.size() == 0) {
            return new ExprNodeConstantDesc(TypeInfoFactory.booleanTypeInfo, Boolean.FALSE);
        }
        if (newExprs.size() == 1) {
            return newExprs.get(0);
        }
    }
    if (udf instanceof GenericUDFWhen) {
        if (!(newExprs.size() == 2 || newExprs.size() == 3)) {
            // we currently only handle either 1 or 2 branch.
            return null;
        }
        ExprNodeDesc thenExpr = newExprs.get(1);
        ExprNodeDesc elseExpr = newExprs.size() == 3 ? newExprs.get(2) : new ExprNodeConstantDesc(newExprs.get(1).getTypeInfo(), null);
        ExprNodeDesc whenExpr = newExprs.get(0);
        if (whenExpr instanceof ExprNodeConstantDesc) {
            Boolean whenVal = (Boolean) ((ExprNodeConstantDesc) whenExpr).getValue();
            return (whenVal == null || Boolean.FALSE.equals(whenVal)) ? elseExpr : thenExpr;
        }
        if (thenExpr instanceof ExprNodeConstantDesc && elseExpr instanceof ExprNodeConstantDesc) {
            ExprNodeConstantDesc constThen = (ExprNodeConstantDesc) thenExpr;
            ExprNodeConstantDesc constElse = (ExprNodeConstantDesc) elseExpr;
            Object thenVal = constThen.getValue();
            Object elseVal = constElse.getValue();
            if (thenVal == null) {
                if (elseVal == null) {
                    // both branches are null.
                    return thenExpr;
                } else if (op instanceof FilterOperator) {
                    // we can still fold, since here null is equivalent to false.
                    return Boolean.TRUE.equals(elseVal) ? ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPNot(), newExprs.subList(0, 1)) : Boolean.FALSE.equals(elseVal) ? elseExpr : null;
                } else {
                    // can't do much, expression is not in context of filter, so we can't treat null as equivalent to false here.
                    return null;
                }
            } else if (elseVal == null && op instanceof FilterOperator) {
                return Boolean.TRUE.equals(thenVal) ? whenExpr : Boolean.FALSE.equals(thenVal) ? thenExpr : null;
            } else if (thenVal.equals(elseVal)) {
                return thenExpr;
            } else if (thenVal instanceof Boolean && elseVal instanceof Boolean) {
                List<ExprNodeDesc> children = new ArrayList<>();
                children.add(whenExpr);
                children.add(new ExprNodeConstantDesc(false));
                ExprNodeGenericFuncDesc func = ExprNodeGenericFuncDesc.newInstance(new GenericUDFNvl(), children);
                if (Boolean.TRUE.equals(thenVal)) {
                    return func;
                } else {
                    List<ExprNodeDesc> exprs = new ArrayList<>();
                    exprs.add(func);
                    return ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPNot(), exprs);
                }
            } else {
                return null;
            }
        }
    }
    if (udf instanceof GenericUDFCase) {
        // where ss_sold_date= '1998-01-01' ;
        if (!(newExprs.size() == 3 || newExprs.size() == 4)) {
            // we currently only handle either 1 or 2 branch.
            return null;
        }
        ExprNodeDesc thenExpr = newExprs.get(2);
        ExprNodeDesc elseExpr = newExprs.size() == 4 ? newExprs.get(3) : new ExprNodeConstantDesc(newExprs.get(2).getTypeInfo(), null);
        if (thenExpr instanceof ExprNodeConstantDesc && elseExpr instanceof ExprNodeConstantDesc) {
            ExprNodeConstantDesc constThen = (ExprNodeConstantDesc) thenExpr;
            ExprNodeConstantDesc constElse = (ExprNodeConstantDesc) elseExpr;
            Object thenVal = constThen.getValue();
            Object elseVal = constElse.getValue();
            if (thenVal == null) {
                if (null == elseVal) {
                    return thenExpr;
                } else if (op instanceof FilterOperator) {
                    return Boolean.TRUE.equals(elseVal) ? ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPNotEqual(), newExprs.subList(0, 2)) : Boolean.FALSE.equals(elseVal) ? elseExpr : null;
                } else {
                    return null;
                }
            } else if (null == elseVal && op instanceof FilterOperator) {
                return Boolean.TRUE.equals(thenVal) ? ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPEqual(), newExprs.subList(0, 2)) : Boolean.FALSE.equals(thenVal) ? thenExpr : null;
            } else if (thenVal.equals(elseVal)) {
                return thenExpr;
            } else if (thenVal instanceof Boolean && elseVal instanceof Boolean) {
                ExprNodeGenericFuncDesc equal = ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPEqual(), newExprs.subList(0, 2));
                List<ExprNodeDesc> children = new ArrayList<>();
                children.add(equal);
                children.add(new ExprNodeConstantDesc(false));
                ExprNodeGenericFuncDesc func = ExprNodeGenericFuncDesc.newInstance(new GenericUDFNvl(), children);
                if (Boolean.TRUE.equals(thenVal)) {
                    return func;
                } else {
                    List<ExprNodeDesc> exprs = new ArrayList<>();
                    exprs.add(func);
                    return ExprNodeGenericFuncDesc.newInstance(new GenericUDFOPNot(), exprs);
                }
            } else {
                return null;
            }
        }
    }
    if (udf instanceof GenericUDFUnixTimeStamp) {
        if (newExprs.size() >= 1) {
            // unix_timestamp(args) -> to_unix_timestamp(args)
            return ExprNodeGenericFuncDesc.newInstance(new GenericUDFToUnixTimeStamp(), newExprs);
        }
    }
    return null;
}

Example 14 with GenericUDF

use of org.apache.hadoop.hive.ql.udf.generic.GenericUDF in project hive by apache.

the class ConstantPropagateProcFactory method isDeterministicUdf.

private static boolean isDeterministicUdf(GenericUDF udf, List<ExprNodeDesc> children) {
    UDFType udfType = udf.getClass().getAnnotation(UDFType.class);
    if (udf instanceof GenericUDFBridge) {
        udfType = ((GenericUDFBridge) udf).getUdfClass().getAnnotation(UDFType.class);
    }
    if (udfType.deterministic() == false) {
        if (udf.getClass().equals(GenericUDFUnixTimeStamp.class) && children != null && children.size() > 0) {
            // unix_timestamp is polymorphic (ignore class annotations)
            return true;
        }
        return false;
    }
    // If udf is requiring additional jars, we can't determine the result in
    // compile time.
    String[] files;
    String[] jars;
    if (udf instanceof GenericUDFBridge) {
        GenericUDFBridge bridge = (GenericUDFBridge) udf;
        String udfClassName = bridge.getUdfClassName();
        try {
            UDF udfInternal = (UDF) Class.forName(bridge.getUdfClassName(), true, Utilities.getSessionSpecifiedClassLoader()).newInstance();
            files = udfInternal.getRequiredFiles();
            jars = udfInternal.getRequiredJars();
        } catch (Exception e) {
            LOG.error("The UDF implementation class '" + udfClassName + "' is not present in the class path");
            return false;
        }
    } else {
        files = udf.getRequiredFiles();
        jars = udf.getRequiredJars();
    }
    if (files != null || jars != null) {
        return false;
    }
    return true;
}

Example 15 with GenericUDF

use of org.apache.hadoop.hive.ql.udf.generic.GenericUDF in project hive by apache.

the class ExprNodeDescUtils method foldConstant.

private static ExprNodeConstantDesc foldConstant(ExprNodeGenericFuncDesc func) {
    GenericUDF udf = func.getGenericUDF();
    if (!FunctionRegistry.isDeterministic(udf) || FunctionRegistry.isStateful(udf)) {
        return null;
    }
    try {
        // resources may not be available at compile time.
        if (udf instanceof GenericUDFBridge) {
            UDF internal = ReflectionUtils.newInstance(((GenericUDFBridge) udf).getUdfClass(), null);
            if (internal.getRequiredFiles() != null || internal.getRequiredJars() != null) {
                return null;
            }
        } else {
            if (udf.getRequiredFiles() != null || udf.getRequiredJars() != null) {
                return null;
            }
        }
        if (func.getChildren() != null) {
            for (ExprNodeDesc child : func.getChildren()) {
                if (child instanceof ExprNodeConstantDesc) {
                    continue;
                }
                if (child instanceof ExprNodeGenericFuncDesc) {
                    if (foldConstant((ExprNodeGenericFuncDesc) child) != null) {
                        continue;
                    }
                }
                return null;
            }
        }
        ExprNodeEvaluator evaluator = ExprNodeEvaluatorFactory.get(func);
        ObjectInspector output = evaluator.initialize(null);
        Object constant = evaluator.evaluate(null);
        Object java = ObjectInspectorUtils.copyToStandardJavaObject(constant, output);
        return new ExprNodeConstantDesc(java);
    } catch (Exception e) {
        return null;
    }
}