// 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.

#include "format_v2/parquet/parquet_reader.h"

#include <algorithm>

#include <map>

#include <memory>

#include <optional>

#include <ranges>

#include <unordered_set>

#include <utility>

#include <vector>

#include "common/cast_set.h"

#include "core/assert_cast.h"

#include "core/block/block.h"

#include "core/data_type/data_type_array.h"

#include "core/data_type/data_type_factory.hpp"

#include "core/data_type/data_type_map.h"

#include "core/data_type/data_type_nullable.h"

#include "core/data_type/data_type_struct.h"

#include "format_v2/column_mapper.h"

#include "format_v2/parquet/parquet_column_schema.h"

#include "format_v2/parquet/parquet_file_context.h"

#include "format_v2/parquet/parquet_scan.h"

#include "format_v2/parquet/parquet_statistics.h"

#include "format_v2/parquet/reader/column_reader.h"

#include "io/io_common.h"

#include "runtime/runtime_state.h"

#include "util/timezone_utils.h"

namespace doris::format::parquet {

struct ParquetReaderScanState {

    ParquetFileContext file_context;

    std::vector<std::unique_ptr<ParquetColumnSchema>> file_schema;

    RowGroupScanPlan scan_plan;

    ParquetScanScheduler scheduler;

    const RuntimeState* runtime_state = nullptr;

    const cctz::time_zone* timezone = nullptr;

    std::optional<cctz::time_zone> int96_timezone;

    bool enable_bloom_filter = false;

    bool enable_page_cache = false;

    bool enable_strict_mode = false;

};

int64_t column_chunk_start_offset(const ::parquet::ColumnChunkMetaData& column_metadata) {

    return column_metadata.has_dictionary_page()

                   ? cast_set<int64_t>(column_metadata.dictionary_page_offset())

                   : cast_set<int64_t>(column_metadata.data_page_offset());

}

void collect_all_leaf_column_ids(const ParquetColumnSchema& column_schema,

                                 std::unordered_set<int>* leaf_column_ids) {

    DORIS_CHECK(leaf_column_ids != nullptr);

    if (column_schema.kind == ParquetColumnSchemaKind::PRIMITIVE) {

        if (column_schema.leaf_column_id >= 0) {

            leaf_column_ids->insert(column_schema.leaf_column_id);

        }

        return;

    }

    for (const auto& child : column_schema.children) {

        DORIS_CHECK(child != nullptr);

        collect_all_leaf_column_ids(*child, leaf_column_ids);

    }

}

void collect_projected_leaf_column_ids(const ParquetColumnSchema& column_schema,

                                       const format::LocalColumnIndex& projection,

                                       std::unordered_set<int>* leaf_column_ids) {

    DORIS_CHECK(leaf_column_ids != nullptr);

    if (projection.project_all_children || projection.children.empty()) {

        collect_all_leaf_column_ids(column_schema, leaf_column_ids);

        return;

    }

    for (const auto& child_projection : projection.children) {

        const auto child_it =

                std::ranges::find_if(column_schema.children, [&](const auto& child_schema) {

                    return child_schema->local_id == child_projection.local_id();

                });

        DORIS_CHECK(child_it != column_schema.children.end());

        collect_projected_leaf_column_ids(**child_it, child_projection, leaf_column_ids);

    }

}

void collect_request_leaf_column_ids(

        const std::vector<std::unique_ptr<ParquetColumnSchema>>& file_schema,

        const format::FileScanRequest& request, std::unordered_set<int>* leaf_column_ids) {

    DORIS_CHECK(leaf_column_ids != nullptr);

    auto collect_scan_column = [&](const format::LocalColumnIndex& projection) {

        const auto local_id = projection.local_id();

        if (local_id == format::ROW_POSITION_COLUMN_ID ||

            local_id == format::GLOBAL_ROWID_COLUMN_ID) {

            return;

        }

        DORIS_CHECK(local_id >= 0 && local_id < static_cast<int32_t>(file_schema.size()));

        DORIS_CHECK(file_schema[local_id] != nullptr);

        collect_projected_leaf_column_ids(*file_schema[local_id], projection, leaf_column_ids);

    };

    for (const auto& column : request.predicate_columns) {

        collect_scan_column(column);

    }

    for (const auto& column : request.non_predicate_columns) {

        collect_scan_column(column);

    }

}

std::vector<ParquetPageCacheRange> build_page_cache_ranges(

        const ::parquet::FileMetaData& metadata,

        const std::vector<std::unique_ptr<ParquetColumnSchema>>& file_schema,

        const format::FileScanRequest& request, const RowGroupScanPlan& row_group_plan) {

    std::unordered_set<int> leaf_column_ids;

    collect_request_leaf_column_ids(file_schema, request, &leaf_column_ids);

    std::vector<ParquetPageCacheRange> ranges;

    ranges.reserve(row_group_plan.row_groups.size() * leaf_column_ids.size());

    for (const auto& row_group_plan_item : row_group_plan.row_groups) {

        auto row_group_metadata = metadata.RowGroup(row_group_plan_item.row_group_id);

        DORIS_CHECK(row_group_metadata != nullptr);

        for (const auto leaf_column_id : leaf_column_ids) {

            DORIS_CHECK(leaf_column_id >= 0 && leaf_column_id < row_group_metadata->num_columns());

            auto column_metadata = row_group_metadata->ColumnChunk(leaf_column_id);

            DORIS_CHECK(column_metadata != nullptr);

            const int64_t offset = column_chunk_start_offset(*column_metadata);

            const int64_t size = column_metadata->total_compressed_size();

            DORIS_CHECK(offset >= 0);

            DORIS_CHECK(size >= 0);

            if (size > 0) {

                ranges.push_back(ParquetPageCacheRange {.offset = offset, .size = size});

            }

        }

    }

    return ranges;

}

const ParquetColumnSchema& projected_root_schema(

        const std::vector<std::unique_ptr<ParquetColumnSchema>>& file_schema,

        const format::LocalColumnIndex& projection) {

    const auto local_id = projection.local_id();

    DORIS_CHECK(local_id >= 0 && local_id < static_cast<int32_t>(file_schema.size()));

    DORIS_CHECK(file_schema[local_id] != nullptr);

    return *file_schema[local_id];

}

int64_t count_loaded_non_null_values(const ParquetColumnSchema& root_schema,

                                     const ParquetColumnReader& shape_reader,

                                     int64_t expected_rows) {

    const auto& def_levels = shape_reader.nested_definition_levels();

    const auto& rep_levels = shape_reader.nested_repetition_levels();

    const int64_t levels_written = shape_reader.nested_levels_written();

    DORIS_CHECK(levels_written >= expected_rows);

    if (root_schema.max_repetition_level == 0) {

        DORIS_CHECK(levels_written == expected_rows);

        const int16_t non_null_definition_level = root_schema.nullable_definition_level;

        int64_t count = 0;

        for (int64_t level_idx = 0; level_idx < levels_written; ++level_idx) {

            count += def_levels[level_idx] >= non_null_definition_level ? 1 : 0;

        }

        return count;

    }

    // For repeated encodings, repetition level zero starts a top-level row. Empty MAP/LIST rows

    // have no entries but still carry a level slot; they are non-NULL and must be counted by

    // count(col). The root nullable level distinguishes a NULL top-level value from a non-NULL

    // value regardless of which repeated leaf represents its shape.

    const int16_t non_null_definition_level = root_schema.nullable_definition_level;

    int64_t counted_rows = 0;

    int64_t non_null_rows = 0;

    for (int64_t level_idx = 0; level_idx < levels_written && counted_rows < expected_rows;

         ++level_idx) {

        if (rep_levels[level_idx] != 0) {

            continue;

        }

        ++counted_rows;

        non_null_rows += def_levels[level_idx] >= non_null_definition_level ? 1 : 0;

    }

    DORIS_CHECK(counted_rows == expected_rows);

    return non_null_rows;

}

DataTypePtr nullable_like_original(const DataTypePtr& type, DataTypePtr nested_type) {

    return type != nullptr && type->is_nullable() ? make_nullable(nested_type) : nested_type;

}

int timestamp_tz_scale(const ParquetTypeDescriptor& type_descriptor) {

    switch (type_descriptor.time_unit) {

    case ParquetTimeUnit::MILLIS:

        return 3;

    case ParquetTimeUnit::MICROS:

    case ParquetTimeUnit::UNKNOWN:

    default:

        return 6;

    }

}

bool should_map_to_timestamp_tz(const ParquetColumnSchema& column_schema) {

    const auto& type_descriptor = column_schema.type_descriptor;

    return type_descriptor.physical_type == ::parquet::Type::INT96 ||

           (type_descriptor.is_timestamp && type_descriptor.timestamp_is_adjusted_to_utc);

}

DataTypePtr apply_timestamp_tz_mapping(ParquetColumnSchema* column_schema) {

    DORIS_CHECK(column_schema != nullptr);

    if (column_schema->kind == ParquetColumnSchemaKind::PRIMITIVE) {

        if (should_map_to_timestamp_tz(*column_schema)) {

            const bool nullable =

                    column_schema->type != nullptr && column_schema->type->is_nullable();

            const auto scale = timestamp_tz_scale(column_schema->type_descriptor);

            column_schema->type = DataTypeFactory::instance().create_data_type(TYPE_TIMESTAMPTZ,

                                                                               nullable, 0, scale);

            column_schema->type_descriptor.doris_type = column_schema->type;

        }

        return column_schema->type;

    }

    std::vector<DataTypePtr> child_types;

    child_types.reserve(column_schema->children.size());

    for (auto& child : column_schema->children) {

        child_types.push_back(apply_timestamp_tz_mapping(child.get()));

    }

    if (column_schema->kind == ParquetColumnSchemaKind::LIST) {

        DORIS_CHECK(child_types.size() == 1);

        column_schema->type = nullable_like_original(

                column_schema->type, std::make_shared<DataTypeArray>(child_types[0]));

    } else if (column_schema->kind == ParquetColumnSchemaKind::MAP) {

        DORIS_CHECK(child_types.size() == 2);

        column_schema->type = nullable_like_original(

                column_schema->type, std::make_shared<DataTypeMap>(make_nullable(child_types[0]),

                                                                   make_nullable(child_types[1])));

    } else if (column_schema->kind == ParquetColumnSchemaKind::STRUCT) {

        Strings child_names;

        child_names.reserve(column_schema->children.size());

        for (const auto& child : column_schema->children) {

            child_names.push_back(child->name);

        }

        column_schema->type = nullable_like_original(

                column_schema->type, std::make_shared<DataTypeStruct>(child_types, child_names));

    }

    return column_schema->type;

}

static Status find_projected_minmax_leaf(const ParquetColumnSchema& column_schema,

                                         const format::LocalColumnIndex& projection,

                                         const ParquetColumnSchema** leaf_schema) {

    DORIS_CHECK(leaf_schema != nullptr);

    if (projection.project_all_children || projection.children.empty()) {

        if (column_schema.leaf_column_id < 0) {

            return Status::NotSupported(

                    "Parquet aggregate pushdown only supports primitive column {}",

                    column_schema.name);

        }

        if (column_schema.max_repetition_level > 0) {

            return Status::NotSupported(

                    "Parquet aggregate pushdown does not support repeated column {}",

                    column_schema.name);

        }

        *leaf_schema = &column_schema;

        return Status::OK();

    }

    if (projection.children.size() != 1) {

        return Status::NotSupported(

                "Parquet aggregate pushdown only supports a single nested leaf under column {}",

                column_schema.name);

    }

    const auto& child_projection = projection.children[0];

    const auto child_schema_it =

            std::ranges::find_if(column_schema.children, [&](const auto& child_schema) {

                return child_schema->local_id == child_projection.local_id();

            });

    if (child_schema_it != column_schema.children.end()) {

        return find_projected_minmax_leaf(**child_schema_it, child_projection, leaf_schema);

    }

    return Status::InvalidArgument("Invalid parquet aggregate projection local id {} for column {}",

                                   child_projection.local_id(), column_schema.name);

}

static Status validate_minmax_aggregate_statistics(const ParquetColumnSchema& column_schema) {

    DORIS_CHECK(column_schema.descriptor != nullptr);

    switch (column_schema.descriptor->physical_type()) {

    case ::parquet::Type::BYTE_ARRAY:

    case ::parquet::Type::FIXED_LEN_BYTE_ARRAY:

        // Arrow 17 does not expose Parquet's min/max exactness flags. Binary statistics may be

        // truncated bounds rather than values present in the file, so they are safe for pruning

        // but cannot be returned as exact aggregate results.

        return Status::NotSupported(

                "Parquet MIN/MAX aggregate pushdown requires exact statistics for column {}",

                column_schema.name);

    default:

        return Status::OK();

    }

}

void ParquetReader::_fill_column_definition(const ParquetColumnSchema& column_schema,

                                            format::ColumnDefinition* field) const {

    if (column_schema.parquet_field_id >= 0) {

        field->identifier = Field::create_field<TYPE_INT>(column_schema.parquet_field_id);

    } else {

        field->identifier = Field::create_field<TYPE_STRING>(column_schema.name);

    }

    field->local_id = column_schema.local_id;

    field->name = column_schema.name;

    field->type = column_schema.type != nullptr && !column_schema.type->is_nullable()

                          ? make_nullable(column_schema.type)

                          : column_schema.type;

    field->children.clear();

    field->children.reserve(column_schema.children.size());

    for (const auto& child : column_schema.children) {

        format::ColumnDefinition child_field;

        _fill_column_definition(*child, &child_field);

        field->children.push_back(std::move(child_field));

    }

}

ParquetReader::ParquetReader(std::shared_ptr<io::FileSystemProperties>& system_properties,

                             std::unique_ptr<io::FileDescription>& file_description,

                             std::shared_ptr<io::IOContext> io_ctx, RuntimeProfile* profile,

                             std::optional<format::GlobalRowIdContext> global_rowid_context,

                             bool enable_mapping_timestamp_tz, std::string hive_parquet_time_zone)

        : FileReader(system_properties, file_description, io_ctx, profile),

          _global_rowid_context(global_rowid_context),

          _enable_mapping_timestamp_tz(enable_mapping_timestamp_tz),

          _hive_parquet_time_zone(std::move(hive_parquet_time_zone)) {}

ParquetReader::~ParquetReader() = default;

Status ParquetReader::init(RuntimeState* state) {

    if (_io_ctx != nullptr && _io_ctx->should_stop) {

        return Status::EndOfFile("stop");

    }

    RETURN_IF_ERROR(format::FileReader::init(state));

    if (_profile != nullptr) {

        COUNTER_UPDATE(_parquet_profile.file_reader_create_time,

                       _reader_statistics.file_reader_create_time);

        COUNTER_UPDATE(_parquet_profile.open_file_num, _reader_statistics.open_file_num);

    }

    _state = std::make_unique<ParquetReaderScanState>();

    _state->enable_bloom_filter =

            state != nullptr && state->query_options().enable_parquet_filter_by_bloom_filter;

    _state->enable_page_cache =

            state != nullptr && state->query_options().enable_parquet_file_page_cache;

    if (!_hive_parquet_time_zone.empty()) {

        cctz::time_zone int96_timezone;

        if (!TimezoneUtils::find_cctz_time_zone(_hive_parquet_time_zone, int96_timezone)) {

            return Status::InvalidArgument("Invalid hive.parquet.time-zone: {}",

                                           _hive_parquet_time_zone);

        }

        _state->int96_timezone = int96_timezone;

        _state->scheduler.set_int96_timezone(&*_state->int96_timezone);

    }

    if (state != nullptr) {

        _state->runtime_state = state;

        _state->timezone = &state->timezone_obj();

        _state->enable_strict_mode = state->enable_strict_mode();

        _state->scheduler.set_timezone(&state->timezone_obj());

        _state->scheduler.set_enable_strict_mode(_state->enable_strict_mode);

    }

    int64_t merge_read_slice_size = -1;

    if (state != nullptr && state->query_options().__isset.merge_read_slice_size) {

        merge_read_slice_size = state->query_options().merge_read_slice_size;

    }

    _state->scheduler.set_merge_read_options(_profile, merge_read_slice_size);

    _state->scheduler.set_batch_size(_batch_size);

    // Open parquet file and parse metadata to get file schema.

    RETURN_IF_ERROR(_state->file_context.open(_tracing_file_reader, _io_ctx.get(),

                                              _state->enable_page_cache, *_file_description));

    // Build file schema from parquet metadata.

    // A file reader may expose raw file identifiers, such as Parquet field_id, through ColumnDefinition::identifier

    RETURN_IF_ERROR(

            build_parquet_column_schema(*_state->file_context.schema, &_state->file_schema));

    if (_enable_mapping_timestamp_tz) {

        for (auto& column_schema : _state->file_schema) {

            apply_timestamp_tz_mapping(column_schema.get());

        }

    }

    return Status::OK();

}

void ParquetReader::set_batch_size(size_t batch_size) {

    _batch_size = std::max<size_t>(1, batch_size);

    if (_state != nullptr) {

        _state->scheduler.set_batch_size(_batch_size);

    }

}

Status ParquetReader::get_schema(std::vector<format::ColumnDefinition>* file_schema) const {

    if (file_schema == nullptr) {

        return Status::InvalidArgument("file_schema is null");

    }

    file_schema->clear();

    if (_state == nullptr || _state->file_context.schema == nullptr) {

        return Status::Uninitialized("ParquetReader is not open");

    }

    file_schema->reserve(_state->file_schema.size());

    for (size_t column_idx = 0; column_idx < _state->file_schema.size(); ++column_idx) {

        format::ColumnDefinition field;

        _fill_column_definition(*_state->file_schema[column_idx], &field);

        DORIS_CHECK(field.local_id == static_cast<int32_t>(column_idx));

        file_schema->push_back(std::move(field));

    }

    if (_global_rowid_context.has_value()) {

        file_schema->push_back(format::global_rowid_column_definition());

    }

    return Status::OK();

}

std::unique_ptr<format::TableColumnMapper> ParquetReader::create_column_mapper(

        format::TableColumnMapperOptions options) const {

    return std::make_unique<format::ParquetColumnMapper>(std::move(options));

}

Status ParquetReader::open(std::shared_ptr<format::FileScanRequest> request) {

    if (_state == nullptr || _state->file_context.metadata == nullptr ||

        _state->file_context.schema == nullptr) {

        return Status::Uninitialized("ParquetReader is not open");

    }

    auto request_snapshot = request;

    DORIS_CHECK(request_snapshot != nullptr);

    RETURN_IF_ERROR(format::FileReader::open(std::move(request)));

    // `local_positions.empty()` means all columns are needed by table reader

    // TODO(gabriel): It will happen only for TVF `select *` query.

    if (request_snapshot->local_positions.empty()) {

        for (const auto& col : request_snapshot->predicate_columns) {

            request_snapshot->local_positions.emplace(col.column_id(),

                                                      format::LocalIndex(col.column_id().value()));

        }

        for (const auto& col : request_snapshot->non_predicate_columns) {

            request_snapshot->local_positions.emplace(col.column_id(),

                                                      format::LocalIndex(col.column_id().value()));

        }

    }

    const auto num_fields = static_cast<int32_t>(_state->file_schema.size());

    for (const auto& col : request_snapshot->predicate_columns) {

        DORIS_CHECK(request_snapshot->local_positions.count(col.column_id()) > 0);

        const auto local_id = col.local_id();

        if (local_id == format::ROW_POSITION_COLUMN_ID ||

            local_id == format::GLOBAL_ROWID_COLUMN_ID) {

            continue;

        }

        DORIS_CHECK(local_id >= 0 && local_id < num_fields);

    }

    for (const auto& col : request_snapshot->non_predicate_columns) {

        DORIS_CHECK(request_snapshot->local_positions.count(col.column_id()) > 0);

        const auto local_id = col.local_id();

        if (local_id == format::ROW_POSITION_COLUMN_ID ||

            local_id == format::GLOBAL_ROWID_COLUMN_ID) {

            continue;

        }

        DORIS_CHECK(local_id >= 0 && local_id < num_fields);

    }

    RowGroupScanPlan row_group_plan;

    ParquetScanRange scan_range;

    scan_range.start_offset = _file_description->range_start_offset;

    scan_range.size = _file_description->range_size;

    scan_range.file_size = _file_description->file_size;

    // Get selected ranges in row groups according to metadata (Row-Group level index and Page Index including Zonemap, Dictionary, Bloom Filter).

    RETURN_IF_ERROR(plan_parquet_row_groups(

            *_state->file_context.metadata, _state->file_context.file_reader.get(),

            _state->file_schema, *request_snapshot, scan_range, _state->enable_bloom_filter,

            &row_group_plan, _state->timezone, _state->runtime_state));

    if (_profile != nullptr) {

        _parquet_profile.update_pruning_stats(row_group_plan.pruning_stats);

    }

    if (_state->enable_page_cache) {

        _state->file_context.register_page_cache_ranges(

                build_page_cache_ranges(*_state->file_context.metadata, _state->file_schema,

                                        *request_snapshot, row_group_plan));

    }

    _state->scan_plan = row_group_plan;

    _state->scheduler.set_page_skip_profile(_parquet_profile.page_skip_profile());

    _state->scheduler.set_global_rowid_context(_global_rowid_context);

    _state->scheduler.set_scan_profile(_parquet_profile.scan_profile());

    _state->scheduler.set_plan(std::move(row_group_plan));

    _eof = _state->scheduler.empty();

    return Status::OK();

}

Status ParquetReader::get_block(Block* file_block, size_t* rows, bool* eof) {

    if (_state == nullptr || _state->file_context.file_reader == nullptr ||

        _state->file_context.schema == nullptr) {

        return Status::Uninitialized("ParquetReader is not open");

    }

    *rows = 0;

    if (_io_ctx != nullptr && _io_ctx->should_stop) {

        *eof = true;

        return Status::OK();

    }

    if (_eof) {

        *eof = true;

        return Status::OK();

    }

    auto request_snapshot = _request;

    if (request_snapshot == nullptr) {

        return Status::Cancelled("ParquetReader is closed");

    }

    const auto predicate_filtered_rows_before = _state->scheduler.predicate_filtered_rows();

    const auto raw_rows_read_before = _state->scheduler.raw_rows_read();

    Status st = _state->scheduler.read_next_batch(_state->file_context, _state->file_schema,

                                                  *request_snapshot, file_block, rows, eof);

    if (!st.ok()) {

        if (_io_ctx != nullptr && _io_ctx->should_stop) {

            *rows = 0;

            *eof = true;

            return Status::OK();

        }

        return st;

    }

    _sync_page_cache_profile();

    if (_io_ctx != nullptr) {

        _io_ctx->predicate_filtered_rows +=

                _state->scheduler.predicate_filtered_rows() - predicate_filtered_rows_before;

    }

    const auto raw_rows_read = _state->scheduler.raw_rows_read();

    DORIS_CHECK(raw_rows_read >= raw_rows_read_before);

    _record_scan_rows(raw_rows_read - raw_rows_read_before);

    _eof = *eof;

    return Status::OK();

}

bool ParquetReader::_should_stop() const {

    return _io_ctx != nullptr && _io_ctx->should_stop;

}

Status ParquetReader::_stop_status_if_requested(const Status& status) const {

    if (!status.ok() && _should_stop()) {

        return Status::EndOfFile("stop");

    }

    return status;

}

void ParquetReader::_sync_page_cache_profile() {

    if (_profile == nullptr || _state == nullptr) {

        return;

    }

    const auto stats = _state->file_context.page_cache_stats();

    COUNTER_UPDATE(_parquet_profile.page_read_counter,

                   stats.read_count - _reported_page_cache_stats.read_count);

    COUNTER_UPDATE(_parquet_profile.page_cache_write_counter,

                   stats.write_count - _reported_page_cache_stats.write_count);

    COUNTER_UPDATE(

            _parquet_profile.page_cache_compressed_write_counter,

            stats.compressed_write_count - _reported_page_cache_stats.compressed_write_count);

    COUNTER_UPDATE(_parquet_profile.page_cache_hit_counter,

                   stats.hit_count - _reported_page_cache_stats.hit_count);

    COUNTER_UPDATE(_parquet_profile.page_cache_missing_counter,

                   stats.miss_count - _reported_page_cache_stats.miss_count);

    COUNTER_UPDATE(_parquet_profile.page_cache_compressed_hit_counter,

                   stats.compressed_hit_count - _reported_page_cache_stats.compressed_hit_count);

    _reported_page_cache_stats = stats;

}

void ParquetReader::set_condition_cache_context(std::shared_ptr<ConditionCacheContext> ctx) {

    if (_state == nullptr) {

        return;

    }

    _state->scheduler.set_condition_cache_context(std::move(ctx));

    if (_io_ctx != nullptr) {

        // Condition-cache HIT filters row ranges before batch reading, so skipped rows never belong

        // to a later get_block() batch. Report the plan-level skipped rows at the same point where

        // the scan plan is rewritten.

        _io_ctx->condition_cache_filtered_rows += _state->scheduler.condition_cache_filtered_rows();

    }

}

int64_t ParquetReader::get_total_rows() const {

    if (_state == nullptr) {

        return 0;

    }

    int64_t rows = 0;

    for (const auto& row_group_plan : _state->scan_plan.row_groups) {

        rows += row_group_plan.row_group_rows;

    }

    return rows;

}

Status ParquetReader::get_aggregate_result(const format::FileAggregateRequest& request,

                                           format::FileAggregateResult* result) {

    DORIS_CHECK(result != nullptr);

    if (_state == nullptr || _state->file_context.metadata == nullptr ||

        _state->file_context.schema == nullptr) {

        return Status::Uninitialized("ParquetReader is not open");

    }

    if (_should_stop()) {

        return Status::EndOfFile("stop");

    }

    result->count = 0;

    result->columns.clear();

    if (request.agg_type != TPushAggOp::type::COUNT &&

        request.agg_type != TPushAggOp::type::MINMAX) {

        return Status::NotSupported("Unsupported parquet aggregate pushdown type {}",

                                    request.agg_type);

    }

    // Aggregate row count in all selected row groups. For MIN/MAX aggregate, this is used to determine whether there is no row group selected.

    for (const auto& row_group_plan : _state->scan_plan.row_groups) {

        auto row_group_metadata =

                _state->file_context.metadata->RowGroup(row_group_plan.row_group_id);

        DORIS_CHECK(row_group_metadata != nullptr);

        result->count += row_group_metadata->num_rows();

    }

    if (request.agg_type == TPushAggOp::type::COUNT) {

        if (request.columns.empty()) {

            return Status::OK();

        }

        if (request.columns.size() != 1) {

            return Status::NotSupported("Parquet COUNT pushdown only supports one count column");

        }

        const auto& count_projection = request.columns[0].projection;

        const auto& root_schema = projected_root_schema(_state->file_schema, count_projection);

        result->count = 0;

        for (const auto& row_group_plan : _state->scan_plan.row_groups) {

            std::shared_ptr<::parquet::RowGroupReader> row_group;

            try {

                row_group = _state->file_context.file_reader->RowGroup(row_group_plan.row_group_id);

            } catch (const ::parquet::ParquetException& e) {

                if (_should_stop()) {

                    return Status::EndOfFile("stop");

                }

                return Status::Corruption("Failed to open parquet row group {}: {}",

                                          row_group_plan.row_group_id, e.what());

            } catch (const std::exception& e) {

                if (_should_stop()) {

                    return Status::EndOfFile("stop");

                }

                return Status::InternalError("Failed to open parquet row group {}: {}",

                                             row_group_plan.row_group_id, e.what());

            }

            ParquetColumnReaderFactory column_reader_factory(

                    row_group, _state->file_context.schema->num_columns(),

                    &row_group_plan.page_skip_plans, _parquet_profile.page_skip_profile(),

                    _state->timezone, _state->enable_strict_mode,

                    _parquet_profile.scan_profile().column_reader_profile,

                    _state->int96_timezone ? &*_state->int96_timezone : nullptr);

            std::unique_ptr<ParquetColumnReader> shape_reader;

            RETURN_IF_ERROR(column_reader_factory.create_count_shape_reader(

                    root_schema, &count_projection, &shape_reader));

            DORIS_CHECK(shape_reader != nullptr);

            int64_t row_group_cursor = 0;

            for (const auto& selected_range : row_group_plan.selected_ranges) {

                DORIS_CHECK(selected_range.start >= row_group_cursor);

                RETURN_IF_ERROR(_stop_status_if_requested(

                        shape_reader->skip(selected_range.start - row_group_cursor)));

                row_group_cursor = selected_range.start;

                int64_t range_rows_read = 0;

                while (range_rows_read < selected_range.length) {

                    const int64_t batch_rows =

                            std::min<int64_t>(_batch_size, selected_range.length - range_rows_read);

                    // COUNT(col) only needs the top-level NULL state. The shape reader loads

                    // def/rep levels from one representative leaf and does not build value_indices

                    // or values_column. MAP chooses the key leaf; ARRAY/STRUCT may choose a string

                    // leaf, but the levels-only protocol still avoids Doris-side string

                    // materialization for that leaf.

                    RETURN_IF_ERROR(_stop_status_if_requested(

                            shape_reader->load_nested_levels_batch(batch_rows)));

                    _record_scan_rows(batch_rows);

                    result->count +=

                            count_loaded_non_null_values(root_schema, *shape_reader, batch_rows);

                    range_rows_read += batch_rows;

                    row_group_cursor += batch_rows;

                }

            }

        }

        return Status::OK();

    }

    result->columns.resize(request.columns.size());

    for (size_t request_column_idx = 0; request_column_idx < request.columns.size();

         ++request_column_idx) {

        const auto file_column_id = request.columns[request_column_idx].projection.local_id();

        if (file_column_id < 0 ||

            file_column_id >= static_cast<int32_t>(_state->file_schema.size())) {

            return Status::InvalidArgument("Invalid parquet aggregate column id {}",

                                           file_column_id);

        }

        const auto& column_schema = _state->file_schema[file_column_id];

        DORIS_CHECK(column_schema != nullptr);

        const ParquetColumnSchema* leaf_schema = nullptr;

        RETURN_IF_ERROR(find_projected_minmax_leaf(

                *column_schema, request.columns[request_column_idx].projection, &leaf_schema));

        DORIS_CHECK(leaf_schema != nullptr);

        RETURN_IF_ERROR(validate_minmax_aggregate_statistics(*leaf_schema));

        auto& aggregate_column = result->columns[request_column_idx];

        aggregate_column.projection = request.columns[request_column_idx].projection;

        for (const auto& row_group_plan : _state->scan_plan.row_groups) {

            auto row_group_metadata =

                    _state->file_context.metadata->RowGroup(row_group_plan.row_group_id);

            DORIS_CHECK(row_group_metadata != nullptr);

            auto column_chunk = row_group_metadata->ColumnChunk(leaf_schema->leaf_column_id);

            DORIS_CHECK(column_chunk != nullptr);

            const auto statistics = ParquetStatisticsUtils::TransformColumnStatistics(

                    *leaf_schema, column_chunk->statistics(), _state->timezone);

            if (!statistics.has_min_max) {

                return Status::NotSupported("Missing parquet min/max statistics for column {}",

                                            leaf_schema->name);

            }

            if (!aggregate_column.has_min || statistics.min_value < aggregate_column.min_value) {

                aggregate_column.min_value = statistics.min_value;

                aggregate_column.has_min = true;

            }

            if (!aggregate_column.has_max || aggregate_column.max_value < statistics.max_value) {

                aggregate_column.max_value = statistics.max_value;

                aggregate_column.has_max = true;

            }

        }

        if (!aggregate_column.has_min || !aggregate_column.has_max) {

            return Status::NotSupported("No parquet row group selected for min/max pushdown");

        }

    }

    return Status::OK();

}

Status ParquetReader::close() {

    if (_state != nullptr) {

        _sync_page_cache_profile();

        RETURN_IF_ERROR(_state->file_context.close());

    }

    return FileReader::close();

}

void ParquetReader::_init_profile() {

    _parquet_profile.init(_profile);

}

} // namespace doris::format::parquet