be/src/storage/olap_scan_common.h

Source
// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements.  See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership.  The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License.  You may obtain a copy of the License at
//
//   http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, either express or implied.  See the License for the
// specific language governing permissions and limitations
// under the License.

#pragma once

#include <gen_cpp/Metrics_types.h>
#include <gen_cpp/PaloInternalService_types.h>
#include <glog/logging.h>
#include <stddef.h>

#include <boost/container/detail/std_fwd.hpp>
#include <boost/lexical_cast.hpp>
#include <cstdint>
#include <iterator>
#include <limits>
#include <memory>
#include <set>
#include <sstream>
#include <string>
#include <utility>
#include <variant>
#include <vector>

#include "common/compare.h"
#include "common/status.h"
#include "core/data_type/define_primitive_type.h"
#include "core/data_type/primitive_type.h"
#include "core/type_limit.h"
#include "core/types.h"
#include "core/value/ipv4_value.h"
#include "core/value/ipv6_value.h"
#include "core/value/time_value.h"
#include "core/value/vdatetime_value.h"
#include "exprs/function/cast/cast_to_string.h"
#include "runtime/runtime_profile.h"
#include "storage/olap_common.h"
#include "storage/olap_tuple.h"
#include "storage/olap_utils.h"
#include "storage/predicate/filter_olap_param.h"

namespace doris {

/**
 * @brief Column's value range
 **/
template <PrimitiveType primitive_type>
class ColumnValueRange {
public:
    using CppType = typename PrimitiveTypeTraits<primitive_type>::CppType;
    using SetType = std::set<CppType, doris::Less<CppType>>;
    using IteratorType = typename SetType::iterator;

    ColumnValueRange();

    ColumnValueRange(std::string col_name, bool is_nullable_col, int precision, int scale);

    // should add fixed value before add range
    Status add_fixed_value(const CppType& value);

    // should remove fixed value after add fixed value
    void remove_fixed_value(const CppType& value);

    Status add_range(SQLFilterOp op, CppType value);

    bool is_fixed_value_range() const;

    bool is_scope_value_range() const;

    bool is_empty_value_range() const;

    bool is_fixed_value_convertible() const;

    bool is_range_value_convertible() const;

    void convert_to_range_value();

    bool convert_to_avg_range_value(std::vector<OlapTuple>& begin_scan_keys,
                                    std::vector<OlapTuple>& end_scan_keys, bool& begin_include,
                                    bool& end_include, int32_t max_scan_key_num);

    bool convert_to_close_range(std::vector<OlapTuple>& begin_scan_keys,
                                std::vector<OlapTuple>& end_scan_keys, bool& begin_include,
                                bool& end_include);

    constexpr bool is_reject_split_type() const { return _is_reject_split_type; }

    void intersection(ColumnValueRange<primitive_type>& range);

    void set_empty_value_range() {
        _fixed_values.clear();
        _low_value = TYPE_MAX;
        _high_value = TYPE_MIN;
        _contain_null = false;
    }

    const SetType& get_fixed_value_set() const { return _fixed_values; }

    CppType get_range_max_value() const { return _high_value; }

    CppType get_range_min_value() const { return _low_value; }

    const CppType* get_range_max_value_ptr() const { return &_high_value; }

    const CppType* get_range_min_value_ptr() const { return &_low_value; }

    SQLFilterOp get_range_high_op() const { return _high_op; }

    SQLFilterOp get_range_low_op() const { return _low_op; }

    bool is_low_value_minimum() const { return Compare::equal(_low_value, TYPE_MIN); }

    bool is_high_value_maximum() const { return Compare::equal(_high_value, TYPE_MAX); }

    bool is_begin_include() const { return _low_op == SQLFilterOp::FILTER_LARGER_OR_EQUAL; }

    bool is_end_include() const { return _high_op == SQLFilterOp::FILTER_LESS_OR_EQUAL; }

    PrimitiveType type() const { return _column_type; }

    const std::string& column_name() const { return _column_name; }

    bool contain_null() const { return _contain_null; }

    size_t get_fixed_value_size() const { return _fixed_values.size(); }

    void set_whole_value_range() {
        _fixed_values.clear();
        _low_value = TYPE_MIN;
        _high_value = TYPE_MAX;
        _low_op = SQLFilterOp::FILTER_LARGER_OR_EQUAL;
        _high_op = SQLFilterOp::FILTER_LESS_OR_EQUAL;
        _contain_null = _is_nullable_col;
    }

    bool is_whole_value_range() const {
        DCHECK(_is_nullable_col || !contain_null())
                << "Non-nullable column cannot contains null value";

        return _fixed_values.empty() && Compare::equal(_low_value, TYPE_MIN) &&
               Compare::equal(_high_value, TYPE_MAX) &&
               _low_op == SQLFilterOp::FILTER_LARGER_OR_EQUAL &&
               _high_op == SQLFilterOp::FILTER_LESS_OR_EQUAL && _is_nullable_col == contain_null();
    }

    // only two case will set range contain null, call by temp_range in olap scan node
    // 'is null' and 'is not null'
    // 1. if the pred is 'is null' means the range should be
    // empty in fixed_range and _high_value < _low_value
    // 2. if the pred is 'is not null' means the range should be whole range and
    // 'is not null' be effective
    void set_contain_null(bool contain_null) {
        if (contain_null) {
            set_empty_value_range();
        } else {
            set_whole_value_range();
        }
        _contain_null = _is_nullable_col && contain_null;
    }

    int precision() const { return _precision; }

    int scale() const { return _scale; }

    static void add_fixed_value_range(ColumnValueRange<primitive_type>& range, SQLFilterOp op,
                                      const CppType& value) {
        static_cast<void>(range.add_fixed_value(value));
    }

    static void remove_fixed_value_range(ColumnValueRange<primitive_type>& range, SQLFilterOp op,
                                         const CppType& value) {
        range.remove_fixed_value(value);
    }

    static void empty_function(ColumnValueRange<primitive_type>& range, SQLFilterOp op,
                               const CppType& value) {}

    static void add_value_range(ColumnValueRange<primitive_type>& range, SQLFilterOp op,
                                const CppType& value) {
        static_cast<void>(range.add_range(op, value));
    }

    static ColumnValueRange<primitive_type> create_empty_column_value_range(bool is_nullable_col,
                                                                            int precision,
                                                                            int scale) {
        return ColumnValueRange<primitive_type>("", TYPE_MAX, TYPE_MIN, is_nullable_col, false,
                                                precision, scale);
    }

protected:
    bool is_in_range(const CppType& value);

private:
    ColumnValueRange(std::string col_name, const CppType& min, const CppType& max,
                     bool is_nullable_col, bool contain_null, int precision, int scale);

    const static CppType TYPE_MIN; // Column type's min value
    const static CppType TYPE_MAX; // Column type's max value

    std::string _column_name;
    PrimitiveType _column_type; // Column type (eg: TINYINT,SMALLINT,INT,BIGINT)
    CppType _low_value;         // Column's low value, closed interval at left
    CppType _high_value;        // Column's high value, open interval at right
    SQLFilterOp _low_op;
    SQLFilterOp _high_op;
    SetType _fixed_values; // Column's fixed int value

    bool _is_nullable_col;
    bool _contain_null;
    int _precision;
    int _scale;

    static constexpr bool _is_reject_split_type =
            primitive_type == PrimitiveType::TYPE_FLOAT ||
            primitive_type == PrimitiveType::TYPE_DOUBLE ||
            primitive_type == PrimitiveType::TYPE_LARGEINT ||
            primitive_type == PrimitiveType::TYPE_DECIMALV2 ||
            primitive_type == PrimitiveType::TYPE_VARCHAR ||
            primitive_type == PrimitiveType::TYPE_CHAR ||
            primitive_type == PrimitiveType::TYPE_STRING ||
            primitive_type == PrimitiveType::TYPE_BOOLEAN ||
            primitive_type == PrimitiveType::TYPE_DATETIME ||
            primitive_type == PrimitiveType::TYPE_DATETIMEV2 ||
            primitive_type == PrimitiveType::TYPE_TIMESTAMP_NS ||
            primitive_type == PrimitiveType::TYPE_TIMESTAMPTZ ||
            primitive_type == PrimitiveType::TYPE_DECIMAL256;
};
template <>
const typename ColumnValueRange<TYPE_FLOAT>::CppType ColumnValueRange<TYPE_FLOAT>::TYPE_MIN;
template <>
const typename ColumnValueRange<TYPE_FLOAT>::CppType ColumnValueRange<TYPE_FLOAT>::TYPE_MAX;
template <>
const typename ColumnValueRange<TYPE_DOUBLE>::CppType ColumnValueRange<TYPE_DOUBLE>::TYPE_MIN;
template <>
const typename ColumnValueRange<TYPE_DOUBLE>::CppType ColumnValueRange<TYPE_DOUBLE>::TYPE_MAX;

/// OlapScanKeys accumulates multi-column prefix scan keys from per-column ColumnValueRange
/// constraints, and converts them into OlapScanRange objects for the storage layer.
///
/// Overall pipeline (with examples for table t(k1 INT, k2 INT, v INT)):
///
///  1. _normalize_conjuncts()  (scan_operator.cpp)
///     Parses SQL WHERE conjuncts into per-column ColumnValueRange objects.
///       e.g. "WHERE k1 IN (1,2) AND k2 = 10"
///         => ColumnValueRange<k1>: fixed_values = {1, 2}
///         => ColumnValueRange<k2>: fixed_values = {10}
///
///  2. _build_key_ranges_and_filters()  (olap_scan_operator.cpp)
///     Iterates key columns in schema order, calling extend_scan_key() for each column
///     to expand internal _begin_scan_keys / _end_scan_keys.
///
///  3. extend_scan_key()  (this class)
///     Appends one more column dimension to existing scan keys (Cartesian product for
///     fixed values, or min/max for range values).
///       After k1: _begin_scan_keys = [(1), (2)]     _end_scan_keys = [(1), (2)]
///       After k2: _begin_scan_keys = [(1,10), (2,10)] _end_scan_keys = [(1,10), (2,10)]
///
///  4. get_key_range()  (olap_scan_common.cpp)
///     Converts each (_begin_scan_keys[i], _end_scan_keys[i]) pair into an OlapScanRange.
///       => OlapScanRange{ begin=(1,10), end=(1,10), has_lower_bound=true, ... }
///       => OlapScanRange{ begin=(2,10), end=(2,10), has_lower_bound=true, ... }
///
///  5. If no key predicates exist, get_key_range returns empty; the caller creates a single
///     default OlapScanRange with has_lower_bound=false (represents full table scan).
///
class OlapScanKeys {
public:
    /// Extend internal scan key pairs with the next key column's ColumnValueRange.
    ///
    /// - If the range has fixed values, produces a Cartesian product of existing keys
    ///   and the fixed values (subject to max_scan_key_num limit).
    /// - If the range is a scope (min..max), appends min to begin keys and max to end keys,
    ///   and sets _has_range_value=true (no further columns can be appended).
    ///
    /// @param exact_value [out]: true if the range covers the column's values exactly
    ///                           (can be erased from residual predicates).
    /// @param eos         [out]: true if the range is provably empty (no rows to scan).
    /// @param should_break[out]: true if the range cannot be encoded and we should stop.
    template <PrimitiveType primitive_type>
    Status extend_scan_key(ColumnValueRange<primitive_type>& range, int32_t max_scan_key_num,
                           bool* exact_value, bool* eos, bool* should_break);

    /// Convert accumulated scan key pairs into OlapScanRange objects for the storage layer.
    Status get_key_range(std::vector<std::unique_ptr<OlapScanRange>>* key_range);

    bool has_range_value() const { return _has_range_value; }

    void clear() {
        _has_range_value = false;
        _begin_scan_keys.clear();
        _end_scan_keys.clear();
    }

    std::string debug_string() {
        std::stringstream ss;
        DCHECK(_begin_scan_keys.size() == _end_scan_keys.size());
        ss << "ScanKeys:";

        for (int i = 0; i < _begin_scan_keys.size(); ++i) {
            ss << "ScanKey=" << (_begin_include ? "[" : "(") << _begin_scan_keys[i].debug_string()
               << " : " << _end_scan_keys[i].debug_string() << (_end_include ? "]" : ")");
        }
        return ss.str();
    }

    size_t size() {
        DCHECK(_begin_scan_keys.size() == _end_scan_keys.size());
        return _begin_scan_keys.size();
    }

    void set_is_convertible(bool is_convertible) { _is_convertible = is_convertible; }

private:
    std::vector<OlapTuple> _begin_scan_keys;
    std::vector<OlapTuple> _end_scan_keys;
    bool _has_range_value = false;
    bool _begin_include = false;
    bool _end_include = false;
    bool _is_convertible = false;
};

using ColumnValueRangeType = std::variant<
        ColumnValueRange<TYPE_TINYINT>, ColumnValueRange<TYPE_SMALLINT>, ColumnValueRange<TYPE_INT>,
        ColumnValueRange<TYPE_BIGINT>, ColumnValueRange<TYPE_LARGEINT>,
        ColumnValueRange<TYPE_FLOAT>, ColumnValueRange<TYPE_DOUBLE>, ColumnValueRange<TYPE_IPV4>,
        ColumnValueRange<TYPE_IPV6>, ColumnValueRange<TYPE_CHAR>, ColumnValueRange<TYPE_VARCHAR>,
        ColumnValueRange<TYPE_STRING>, ColumnValueRange<TYPE_DATE>, ColumnValueRange<TYPE_DATEV2>,
        ColumnValueRange<TYPE_DATETIME>, ColumnValueRange<TYPE_DATETIMEV2>,
        ColumnValueRange<TYPE_TIMESTAMP_NS>, ColumnValueRange<TYPE_TIMESTAMPTZ>,
        ColumnValueRange<TYPE_UUID>, ColumnValueRange<TYPE_DECIMALV2>,
        ColumnValueRange<TYPE_BOOLEAN>, ColumnValueRange<TYPE_DECIMAL32>,
        ColumnValueRange<TYPE_DECIMAL64>, ColumnValueRange<TYPE_DECIMAL128I>,
        ColumnValueRange<TYPE_DECIMAL256>>;

template <PrimitiveType primitive_type>
const typename ColumnValueRange<primitive_type>::CppType
        ColumnValueRange<primitive_type>::TYPE_MIN =
                type_limit<typename ColumnValueRange<primitive_type>::CppType>::min();
template <PrimitiveType primitive_type>
const typename ColumnValueRange<primitive_type>::CppType
        ColumnValueRange<primitive_type>::TYPE_MAX =
                type_limit<typename ColumnValueRange<primitive_type>::CppType>::max();

template <PrimitiveType primitive_type>
ColumnValueRange<primitive_type>::ColumnValueRange()
        : _column_type(INVALID_TYPE), _precision(-1), _scale(-1) {}

template <PrimitiveType primitive_type>
ColumnValueRange<primitive_type>::ColumnValueRange(std::string col_name, const CppType& min,
                                                   const CppType& max, bool is_nullable_col,
                                                   bool contain_null, int precision, int scale)
        : _column_name(std::move(col_name)),
          _column_type(primitive_type),
          _low_value(min),
          _high_value(max),
          _low_op(SQLFilterOp::FILTER_LARGER_OR_EQUAL),
          _high_op(SQLFilterOp::FILTER_LESS_OR_EQUAL),
          _is_nullable_col(is_nullable_col),
          _contain_null(is_nullable_col && contain_null),
          _precision(precision),
          _scale(scale) {}

template <PrimitiveType primitive_type>
ColumnValueRange<primitive_type>::ColumnValueRange(std::string col_name, bool is_nullable_col,
                                                   int precision, int scale)
        : ColumnValueRange(std::move(col_name), TYPE_MIN, TYPE_MAX, is_nullable_col,
                           is_nullable_col, precision, scale) {}

template <PrimitiveType primitive_type>
Status ColumnValueRange<primitive_type>::add_fixed_value(const CppType& value) {
    if (INVALID_TYPE == _column_type) {
        return Status::InternalError("AddFixedValue failed, Invalid type");
    }

    _fixed_values.insert(value);
    _contain_null = false;

    _high_value = TYPE_MIN;
    _low_value = TYPE_MAX;

    return Status::OK();
}

template <PrimitiveType primitive_type>
void ColumnValueRange<primitive_type>::remove_fixed_value(const CppType& value) {
    _fixed_values.erase(value);
}

template <PrimitiveType primitive_type>
bool ColumnValueRange<primitive_type>::is_fixed_value_range() const {
    return _fixed_values.size() != 0;
}

template <PrimitiveType primitive_type>
bool ColumnValueRange<primitive_type>::is_scope_value_range() const {
    return Compare::greater(_high_value, _low_value);
}

template <PrimitiveType primitive_type>
bool ColumnValueRange<primitive_type>::is_empty_value_range() const {
    if (INVALID_TYPE == _column_type) {
        return true;
    }

    return (!is_fixed_value_range() && !is_scope_value_range() && !contain_null());
}

template <PrimitiveType primitive_type>
bool ColumnValueRange<primitive_type>::is_fixed_value_convertible() const {
    if (is_fixed_value_range()) {
        return false;
    }

    if (!is_enumeration_type(_column_type)) {
        return false;
    }

    return true;
}

template <PrimitiveType primitive_type>
bool ColumnValueRange<primitive_type>::is_range_value_convertible() const {
    if (!is_fixed_value_range()) {
        return false;
    }

    if (TYPE_NULL == _column_type || TYPE_BOOLEAN == _column_type) {
        return false;
    }

    return true;
}

// The return value indicates whether eos.
template <PrimitiveType primitive_type>
bool ColumnValueRange<primitive_type>::convert_to_close_range(
        std::vector<OlapTuple>& begin_scan_keys, std::vector<OlapTuple>& end_scan_keys,
        bool& begin_include, bool& end_include) {
    if constexpr (!_is_reject_split_type) {
        begin_include = true;
        end_include = true;

        bool is_empty = false;

        if (!is_begin_include()) {
            if (Compare::equal(_low_value, TYPE_MAX)) {
                is_empty = true;
            } else {
                ++_low_value;
            }
        }

        if (!is_end_include()) {
            if (Compare::equal(_high_value, TYPE_MIN)) {
                is_empty = true;
            } else {
                --_high_value;
            }
        }

        if (Compare::less(_high_value, _low_value)) {
            is_empty = true;
        }

        if (is_empty && !contain_null()) {
            begin_scan_keys.clear();
            end_scan_keys.clear();
            return true;
        }
    }
    return false;
}

// The return value indicates whether the split result is range or fixed value.
template <PrimitiveType primitive_type>
bool ColumnValueRange<primitive_type>::convert_to_avg_range_value(
        std::vector<OlapTuple>& begin_scan_keys, std::vector<OlapTuple>& end_scan_keys,
        bool& begin_include, bool& end_include, int32_t max_scan_key_num) {
    if constexpr (!_is_reject_split_type) {
        CppType min_value = get_range_min_value();
        CppType max_value = get_range_max_value();
        if constexpr (primitive_type == PrimitiveType::TYPE_DATE) {
            min_value.set_type(TimeType::TIME_DATE);
            max_value.set_type(TimeType::TIME_DATE);
        }
        auto empty_range_only_null = Compare::greater(min_value, max_value);
        if (empty_range_only_null) {
            // Not contain null will be disposed in `convert_to_close_range`, return eos.
            DCHECK(contain_null());
        }

        auto no_split = [&]() -> bool {
            begin_scan_keys.emplace_back();
            if (contain_null()) {
                begin_scan_keys.back().add_null();
            } else {
                begin_scan_keys.back().add_field(
                        Field::create_field<primitive_type>(get_range_min_value()));
            }
            end_scan_keys.emplace_back();
            if (empty_range_only_null) {
                end_scan_keys.back().add_null();
            } else {
                end_scan_keys.back().add_field(
                        Field::create_field<primitive_type>(get_range_max_value()));
            }
            return true;
        };
        if (empty_range_only_null || max_scan_key_num == 1) {
            return no_split();
        }

        auto cast = [](const CppType& value) {
            if constexpr (primitive_type == PrimitiveType::TYPE_DATE ||
                          primitive_type == PrimitiveType::TYPE_DATEV2) {
                return value;
            } else {
                return (int128_t)value;
            }
        };

        // When CppType is date, we can not convert it to integer number and calculate distance.
        // In other case, we convert element to int128 to avoit overflow.
        int128_t step_size_may_overflow = (cast(max_value) - min_value) / max_scan_key_num;

        constexpr size_t MAX_STEP_SIZE = 1 << 20;
        // When the step size is too large, the range is easy to not really contain data.
        if (step_size_may_overflow > MAX_STEP_SIZE) {
            return no_split();
        }
        int step_size = cast_set<int>(step_size_may_overflow);
        int real_step_size = 0;

        // Add null key if contain null, must do after no_split check
        if (contain_null()) {
            begin_scan_keys.emplace_back();
            begin_scan_keys.back().add_null();
            end_scan_keys.emplace_back();
            end_scan_keys.back().add_null();
        }
        while (true) {
            begin_scan_keys.emplace_back();
            begin_scan_keys.back().add_field(Field::create_field<primitive_type>(min_value));

            if (cast(max_value) - min_value < step_size) {
                min_value = max_value;
            } else {
                min_value += step_size;
            }

            end_scan_keys.emplace_back();
            end_scan_keys.back().add_field(Field::create_field<primitive_type>(min_value));

            if (Compare::equal(min_value, max_value)) {
                break;
            }
            ++min_value;
            ++real_step_size;
            if (real_step_size > MAX_STEP_SIZE) {
                throw Exception(Status::InternalError(
                        "convert_to_avg_range_value meet error. type={}, step_size={}",
                        int(primitive_type), step_size));
            }
        }

        return step_size != 0;
    }
    return false;
}

template <PrimitiveType primitive_type>
void ColumnValueRange<primitive_type>::convert_to_range_value() {
    if (!is_range_value_convertible()) {
        return;
    }

    if (!_fixed_values.empty()) {
        _low_value = *_fixed_values.begin();
        _low_op = SQLFilterOp::FILTER_LARGER_OR_EQUAL;
        _high_value = *_fixed_values.rbegin();
        _high_op = SQLFilterOp::FILTER_LESS_OR_EQUAL;
        _fixed_values.clear();
    }
}

template <PrimitiveType primitive_type>
Status ColumnValueRange<primitive_type>::add_range(SQLFilterOp op, const CppType value) {
    if (INVALID_TYPE == _column_type) {
        return Status::InternalError("AddRange failed, Invalid type");
    }

    // add range means range should not contain null
    _contain_null = false;

    if (is_fixed_value_range()) {
        std::pair<IteratorType, IteratorType> bound_pair = _fixed_values.equal_range(value);

        switch (op) {
        case SQLFilterOp::FILTER_LARGER: {
            _fixed_values.erase(_fixed_values.begin(), bound_pair.second);
            break;
        }

        case SQLFilterOp::FILTER_LARGER_OR_EQUAL: {
            _fixed_values.erase(_fixed_values.begin(), bound_pair.first);
            break;
        }

        case SQLFilterOp::FILTER_LESS: {
            if (bound_pair.first == _fixed_values.find(value)) {
                _fixed_values.erase(bound_pair.first, _fixed_values.end());
            } else {
                _fixed_values.erase(bound_pair.second, _fixed_values.end());
            }

            break;
        }

        case SQLFilterOp::FILTER_LESS_OR_EQUAL: {
            _fixed_values.erase(bound_pair.second, _fixed_values.end());
            break;
        }

        default: {
            return Status::InternalError("Add Range fail! Unsupported SQLFilterOp.");
        }
        }

        _high_value = TYPE_MIN;
        _low_value = TYPE_MAX;
    } else {
        if (Compare::greater(_high_value, _low_value)) {
            switch (op) {
            case SQLFilterOp::FILTER_LARGER: {
                if (Compare::greater_equal(value, _low_value)) {
                    _low_value = value;
                    _low_op = op;
                }

                break;
            }

            case SQLFilterOp::FILTER_LARGER_OR_EQUAL: {
                if (Compare::greater(value, _low_value)) {
                    _low_value = value;
                    _low_op = op;
                }

                break;
            }

            case SQLFilterOp::FILTER_LESS: {
                if (Compare::less_equal(value, _high_value)) {
                    _high_value = value;
                    _high_op = op;
                }

                break;
            }

            case SQLFilterOp::FILTER_LESS_OR_EQUAL: {
                if (Compare::less(value, _high_value)) {
                    _high_value = value;
                    _high_op = op;
                }

                break;
            }

            default: {
                return Status::InternalError("Add Range fail! Unsupported SQLFilterOp.");
            }
            }
        }

        if (SQLFilterOp::FILTER_LARGER_OR_EQUAL == _low_op &&
            SQLFilterOp::FILTER_LESS_OR_EQUAL == _high_op &&
            Compare::equal(_high_value, _low_value)) {
            RETURN_IF_ERROR(add_fixed_value(_high_value));
            _high_value = TYPE_MIN;
            _low_value = TYPE_MAX;
        }
    }

    return Status::OK();
}

template <PrimitiveType primitive_type>
bool ColumnValueRange<primitive_type>::is_in_range(const CppType& value) {
    switch (_high_op) {
    case SQLFilterOp::FILTER_LESS: {
        switch (_low_op) {
        case SQLFilterOp::FILTER_LARGER: {
            return Compare::less(value, _high_value) && Compare::greater(value, _low_value);
        }

        case SQLFilterOp::FILTER_LARGER_OR_EQUAL: {
            return Compare::less(value, _high_value) && Compare::greater_equal(value, _low_value);
        }

        default: {
            DCHECK(false);
        }
        }

        break;
    }

    case SQLFilterOp::FILTER_LESS_OR_EQUAL: {
        switch (_low_op) {
        case SQLFilterOp::FILTER_LARGER: {
            return Compare::less_equal(value, _high_value) && Compare::greater(value, _low_value);
        }

        case SQLFilterOp::FILTER_LARGER_OR_EQUAL: {
            return Compare::less_equal(value, _high_value) &&
                   Compare::greater_equal(value, _low_value);
        }

        default: {
            DCHECK(false);
        }
        }
    }

    default: {
        DCHECK(false);
    }
    }

    return false;
}

template <PrimitiveType primitive_type>
void ColumnValueRange<primitive_type>::intersection(ColumnValueRange<primitive_type>& range) {
    // 1. clear if column type not match
    if (_column_type != range._column_type) {
        set_empty_value_range();
    }

    // 2. clear if any range is empty
    if (is_empty_value_range() || range.is_empty_value_range()) {
        set_empty_value_range();
    }

    SetType result_values;
    // 3. fixed_value intersection, fixed value range do not contain null
    if (is_fixed_value_range() || range.is_fixed_value_range()) {
        if (is_fixed_value_range() && range.is_fixed_value_range()) {
            set_intersection(_fixed_values.begin(), _fixed_values.end(),
                             range._fixed_values.begin(), range._fixed_values.end(),
                             std::inserter(result_values, result_values.begin()));
        } else if (is_fixed_value_range() && !range.is_fixed_value_range()) {
            IteratorType iter = _fixed_values.begin();

            while (iter != _fixed_values.end()) {
                if (range.is_in_range(*iter)) {
                    result_values.insert(*iter);
                }
                ++iter;
            }
        } else if (!is_fixed_value_range() && range.is_fixed_value_range()) {
            IteratorType iter = range._fixed_values.begin();
            while (iter != range._fixed_values.end()) {
                if (this->is_in_range(*iter)) {
                    result_values.insert(*iter);
                }
                ++iter;
            }
        }

        if (!result_values.empty()) {
            _fixed_values = std::move(result_values);
            _contain_null = false;
            _high_value = TYPE_MIN;
            _low_value = TYPE_MAX;
        } else {
            set_empty_value_range();
        }
    } else {
        if (contain_null() && range.contain_null()) {
            // if both is_whole_range to keep the same, else set_contain_null
            if (!is_whole_value_range() || !range.is_whole_value_range()) {
                set_contain_null(true);
            }
        } else {
            static_cast<void>(add_range(range._high_op, range._high_value));
            static_cast<void>(add_range(range._low_op, range._low_value));
        }
    }
}

/// Extend the accumulated scan key pairs (_begin_scan_keys / _end_scan_keys) by appending
/// one more key column's ColumnValueRange.
///
/// Called once per key column in schema order by _build_key_ranges_and_filters().
/// The function handles two kinds of ColumnValueRange:
///   (A) Fixed values  — from IN / = predicates  (begin == end for each value, point lookup)
///   (B) Scope range   — from > / >= / < / <= predicates  (begin = min, end = max)
///
/// ======== Example 1: Two fixed-value columns (IN + =) ========
/// Table t(k1 INT, k2 INT, v INT), key columns = (k1, k2).
/// WHERE k1 IN (1, 2) AND k2 = 10
///
///   Call 1: extend_scan_key(k1's range {fixed_values={1,2}})
///     _begin_scan_keys was empty, so create one pair per fixed value:
///       _begin = [(1), (2)]    _end = [(1), (2)]    include=[true, true]
///
///   Call 2: extend_scan_key(k2's range {fixed_values={10}})
///     _begin is non-empty, so do Cartesian product (existing keys × new fixed values):
///       _begin = [(1,10), (2,10)]    _end = [(1,10), (2,10)]    include=[true, true]
///
/// ======== Example 2: Fixed + range (IN + between) ========
/// WHERE k1 IN (1, 2) AND k2 >= 5 AND k2 < 10
///
///   Call 1: extend_scan_key(k1's range {fixed_values={1,2}})
///       _begin = [(1), (2)]    _end = [(1), (2)]
///
///   Call 2: extend_scan_key(k2's range {scope [5, 10)})
///     k2 is a scope range, so append min=5 to all begin keys, max=10 to all end keys:
///       _begin = [(1,5), (2,5)]    _end = [(1,10), (2,10)]
///       _begin_include = true (>=)    _end_include = false (<)
///     Set _has_range_value = true → no further columns can be appended.
///
/// ======== Example 3: Single range column ========
/// WHERE k1 >= 100 AND k1 <= 200
///
///   Call 1: extend_scan_key(k1's range {scope [100, 200]})
///     _begin was empty, so create one pair:
///       _begin = [(100)]    _end = [(200)]    include=[true, true]
///     Set _has_range_value = true.
///
/// ======== Example 4: Too many fixed values (exceeds max_scan_key_num) ========
/// WHERE k1 IN (1, 2, ..., 10000) — exceeds limit
///
///   If is_range_value_convertible(): convert fixed set {1..10000} to scope [1, 10000],
///   then extend as a range (same as Example 3), and set *exact_value = false
///   (the predicate must be kept for residual filtering).
///
///   If NOT convertible (e.g. BOOLEAN/NULL type): set *should_break = true, stop extending.
///
/// ======== Example 5: Range splitting (convert_to_avg_range_value) ========
/// WHERE k1 >= 1 AND k1 <= 100, with max_scan_key_num = 4
/// If k1 is an integer type that supports splitting:
///   convert_to_close_range: adjust to closed range [1, 100]
///   convert_to_avg_range_value: split into ~4 sub-ranges:
///       _begin = [(1), (26), (51), (76)]    _end = [(25), (50), (75), (100)]
///   Set _has_range_value = true.
///
/// @param range           [in/out] The next key column's ColumnValueRange (may be mutated
///                                  if fixed values must be converted to a range).
/// @param max_scan_key_num [in]    Upper limit on total number of scan key pairs.
/// @param exact_value      [out]   Set to true if the column's predicate is fully captured
///                                  by scan keys (can be erased from residual filters).
/// @param eos              [out]   Set to true if the range is provably empty.
/// @param should_break     [out]   Set to true if extending must stop (un-convertible overflow).
template <PrimitiveType primitive_type>
Status OlapScanKeys::extend_scan_key(ColumnValueRange<primitive_type>& range,
                                     int32_t max_scan_key_num, bool* exact_value, bool* eos,
                                     bool* should_break) {
    using ConstIterator = typename ColumnValueRange<primitive_type>::SetType::const_iterator;

    // 1. If the column's value range is empty (contradictory predicates, e.g. k1 > 10 AND k1 < 5),
    //    clear all accumulated keys — no rows can match.
    if (range.is_empty_value_range()) {
        _begin_scan_keys.clear();
        _end_scan_keys.clear();
        return Status::OK();
    }

    // 2. Once a previous column was extended as a scope range, we cannot append more columns,
    //    because the begin/end keys would have different semantics per pair.
    //    e.g. after k1 in [5, 10), appending k2 values is meaningless for short-key index.
    if (_has_range_value) {
        return Status::OK();
    }

    // 3. Overflow check: if fixed_value_count × existing_key_count > max_scan_key_num,
    //    Cartesian product would be too large.
    //    - If convertible: degrade fixed values {v1,v2,...} to scope [min(v), max(v)],
    //      set *exact_value = false (keep predicate as residual filter).
    //    - If not convertible (BOOLEAN etc.): stop extending (*should_break = true).
    auto scan_keys_size = _begin_scan_keys.empty() ? 1 : _begin_scan_keys.size();
    if (range.is_fixed_value_range()) {
        if (range.get_fixed_value_size() > max_scan_key_num / scan_keys_size) {
            if (range.is_range_value_convertible()) {
                range.convert_to_range_value();
                *exact_value = false;
            } else {
                *should_break = true;
                return Status::OK();
            }
        }
    } else {
        // 4. Range-splitting optimization: if this is the FIRST key column and it's a scope
        //    range on a splittable integer type, try to split [low, high] into multiple
        //    sub-ranges for parallel / pipelined scanning.
        //    e.g. k1 in [1, 100] with max_scan_key_num=4 → [(1,25), (26,50), (51,75), (76,100)]
        if (_begin_scan_keys.empty() && range.is_fixed_value_convertible() && _is_convertible &&
            !range.is_reject_split_type()) {
            *eos |= range.convert_to_close_range(_begin_scan_keys, _end_scan_keys, _begin_include,
                                                 _end_include);

            if (!(*eos) &&
                range.convert_to_avg_range_value(_begin_scan_keys, _end_scan_keys, _begin_include,
                                                 _end_include, max_scan_key_num)) {
                _has_range_value = true;
            }
            return Status::OK();
        }
    }

    // ====================================================================
    // 5. Actually extend scan keys with this column's values.
    // ====================================================================

    if (range.is_fixed_value_range()) {
        // ---- 5a. Fixed values (IN / =): point lookup, begin == end per value. ----
        if (_begin_scan_keys.empty()) {
            // First column: create one key pair per fixed value.
            //   e.g. k1 IN (1, 2) → _begin=[(1),(2)]  _end=[(1),(2)]
            auto fixed_value_set = range.get_fixed_value_set();
            ConstIterator iter = fixed_value_set.begin();

            for (; iter != fixed_value_set.end(); ++iter) {
                _begin_scan_keys.emplace_back();
                _begin_scan_keys.back().add_field(Field::create_field<primitive_type>(*iter));
                _end_scan_keys.emplace_back();
                _end_scan_keys.back().add_field(Field::create_field<primitive_type>(*iter));
            }

            if (range.contain_null()) {
                _begin_scan_keys.emplace_back();
                _begin_scan_keys.back().add_null();
                _end_scan_keys.emplace_back();
                _end_scan_keys.back().add_null();
            }
        } else {
            // Subsequent column: Cartesian product of existing keys × new fixed values.
            //   e.g. existing = [(1),(2)], k2 IN (10, 20)
            //     → [(1,10),(1,20),(2,10),(2,20)]
            auto fixed_value_set = range.get_fixed_value_set();
            size_t original_key_range_size = _begin_scan_keys.size();

            for (int i = 0; i < original_key_range_size; ++i) {
                OlapTuple start_base_key_range = _begin_scan_keys[i];
                OlapTuple end_base_key_range = _end_scan_keys[i];

                ConstIterator iter = fixed_value_set.begin();

                for (; iter != fixed_value_set.end(); ++iter) {
                    // Reuse i-th slot for the first value, append new slots for the rest.
                    if (iter == fixed_value_set.begin()) {
                        _begin_scan_keys[i].add_field(Field::create_field<primitive_type>(*iter));
                        _end_scan_keys[i].add_field(Field::create_field<primitive_type>(*iter));
                    } else {
                        _begin_scan_keys.push_back(start_base_key_range);
                        _begin_scan_keys.back().add_field(
                                Field::create_field<primitive_type>(*iter));
                        _end_scan_keys.push_back(end_base_key_range);
                        _end_scan_keys.back().add_field(Field::create_field<primitive_type>(*iter));
                    }
                }

                if (range.contain_null()) {
                    _begin_scan_keys.push_back(start_base_key_range);
                    _begin_scan_keys.back().add_null();
                    _end_scan_keys.push_back(end_base_key_range);
                    _end_scan_keys.back().add_null();
                }
            }
        }

        // Fixed values are always closed intervals (begin == end, point lookup).
        _begin_include = true;
        _end_include = true;
    } else {
        // ---- 5b. Scope range (> / >= / < / <=): append min to begin, max to end. ----
        // After this, no more columns can be appended (_has_range_value = true),
        // because the range semantics only apply to the last appended column.
        //   e.g. existing = [(1),(2)], k2 >= 5 AND k2 < 10
        //     → _begin = [(1,5),(2,5)]  _end = [(1,10),(2,10)]
        //     → begin_include=true, end_include=false
        _has_range_value = true;

        // Special case: max < min means the range itself is empty,
        // but contain_null() is true, so only null values match this column.
        if (Compare::less(range.get_range_max_value(), range.get_range_min_value())) {
            CHECK(range.contain_null());
            if (_begin_scan_keys.empty()) {
                _begin_scan_keys.emplace_back();
                _begin_scan_keys.back().add_null();
                _end_scan_keys.emplace_back();
                _end_scan_keys.back().add_null();
            } else {
                for (int i = 0; i < _begin_scan_keys.size(); ++i) {
                    _begin_scan_keys[i].add_null();
                    _end_scan_keys[i].add_null();
                }
            }
        } else if (_begin_scan_keys.empty()) {
            // First column as a range:
            //   e.g. k1 >= 100 AND k1 <= 200  → _begin=[(100)]  _end=[(200)]
            _begin_scan_keys.emplace_back();
            if (range.contain_null()) {
                _begin_scan_keys.back().add_null();
            } else {
                _begin_scan_keys.back().add_field(
                        Field::create_field<primitive_type>(range.get_range_min_value()));
            }
            _end_scan_keys.emplace_back();
            _end_scan_keys.back().add_field(
                    Field::create_field<primitive_type>(range.get_range_max_value()));
        } else {
            // Subsequent column as a range: append min/max to every existing key pair.
            for (int i = 0; i < _begin_scan_keys.size(); ++i) {
                if (range.contain_null()) {
                    _begin_scan_keys[i].add_null();
                } else {
                    _begin_scan_keys[i].add_field(
                            Field::create_field<primitive_type>(range.get_range_min_value()));
                }
            }

            for (int i = 0; i < _end_scan_keys.size(); ++i) {
                _end_scan_keys[i].add_field(
                        Field::create_field<primitive_type>(range.get_range_max_value()));
            }
        }
        _begin_include = range.is_begin_include();
        _end_include = range.is_end_include();
    }

    return Status::OK();
}

} // namespace doris

Coverage Report

Created: 2026-10-23 21:13