cpp/src/arrow/dataset/file_parquet.h

// 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.

// This API is EXPERIMENTAL.

#pragma once

#include <memory>
#include <string>
#include <unordered_set>
#include <utility>
#include <vector>

#include "arrow/dataset/discovery.h"
#include "arrow/dataset/file_base.h"
#include "arrow/dataset/type_fwd.h"
#include "arrow/dataset/visibility.h"
#include "arrow/io/caching.h"
#include "arrow/util/optional.h"

namespace parquet {
class ParquetFileReader;
class Statistics;
class ColumnChunkMetaData;
class RowGroupMetaData;
class FileMetaData;
class FileDecryptionProperties;
class FileEncryptionProperties;

class ReaderProperties;
class ArrowReaderProperties;

class WriterProperties;
class ArrowWriterProperties;

namespace arrow {
class FileReader;
class FileWriter;
struct SchemaManifest;
}  // namespace arrow
}  // namespace parquet

namespace arrow {
namespace dataset {

/// \brief A FileFormat implementation that reads from Parquet files
class ARROW_DS_EXPORT ParquetFileFormat : public FileFormat {
 public:
  ParquetFileFormat() = default;

  /// Convenience constructor which copies properties from a parquet::ReaderProperties.
  /// memory_pool will be ignored.
  explicit ParquetFileFormat(const parquet::ReaderProperties& reader_properties);

  std::string type_name() const override { return "parquet"; }

  bool splittable() const override { return true; }

  bool Equals(const FileFormat& other) const override;

  struct ReaderOptions {
    /// \defgroup parquet-file-format-reader-properties properties which correspond to
    /// members of parquet::ReaderProperties.
    ///
    /// We don't embed parquet::ReaderProperties directly because we get memory_pool from
    /// ScanOptions at scan time and provide differing defaults.
    ///
    /// @{
    bool use_buffered_stream = false;
    int64_t buffer_size = 1 << 13;
    std::shared_ptr<parquet::FileDecryptionProperties> file_decryption_properties;
    /// @}

    /// \defgroup parquet-file-format-arrow-reader-properties properties which correspond
    /// to members of parquet::ArrowReaderProperties.
    ///
    /// We don't embed parquet::ReaderProperties directly because we get batch_size from
    /// ScanOptions at scan time, and we will never pass use_threads == true (since we
    /// defer parallelization of the scan). Additionally column names (rather than
    /// indices) are used to indicate dictionary columns.
    ///
    /// @{
    std::unordered_set<std::string> dict_columns;
    bool pre_buffer = false;
    arrow::io::CacheOptions cache_options = arrow::io::CacheOptions::Defaults();
    arrow::io::IOContext io_context;
    /// @}

    /// EXPERIMENTAL: Parallelize conversion across columns. This option is ignored if a
    /// scan is already parallelized across input files to avoid thread contention. This
    /// option will be removed after support is added for simultaneous parallelization
    /// across files and columns.
    bool enable_parallel_column_conversion = false;
  } reader_options;

  Result<bool> IsSupported(const FileSource& source) const override;

  /// \brief Return the schema of the file if possible.
  Result<std::shared_ptr<Schema>> Inspect(const FileSource& source) const override;

  /// \brief Open a file for scanning
  Result<ScanTaskIterator> ScanFile(
      std::shared_ptr<ScanOptions> options,
      const std::shared_ptr<FileFragment>& file) const override;

  using FileFormat::MakeFragment;

  /// \brief Create a Fragment targeting all RowGroups.
  Result<std::shared_ptr<FileFragment>> MakeFragment(
      FileSource source, Expression partition_expression,
      std::shared_ptr<Schema> physical_schema) override;

  /// \brief Create a Fragment, restricted to the specified row groups.
  Result<std::shared_ptr<ParquetFileFragment>> MakeFragment(
      FileSource source, Expression partition_expression,
      std::shared_ptr<Schema> physical_schema, std::vector<int> row_groups);

  /// \brief Return a FileReader on the given source.
  Result<std::unique_ptr<parquet::arrow::FileReader>> GetReader(
      const FileSource& source, ScanOptions* = NULLPTR) const;

  Result<std::shared_ptr<FileWriter>> MakeWriter(
      std::shared_ptr<io::OutputStream> destination, std::shared_ptr<Schema> schema,
      std::shared_ptr<FileWriteOptions> options) const override;

  std::shared_ptr<FileWriteOptions> DefaultWriteOptions() override;
};

/// \brief A FileFragment with parquet logic.
///
/// ParquetFileFragment provides a lazy (with respect to IO) interface to
/// scan parquet files. Any heavy IO calls are deferred to the Scan() method.
///
/// The caller can provide an optional list of selected RowGroups to limit the
/// number of scanned RowGroups, or to partition the scans across multiple
/// threads.
///
/// Metadata can be explicitly provided, enabling pushdown predicate benefits without
/// the potentially heavy IO of loading Metadata from the file system. This can induce
/// significant performance boost when scanning high latency file systems.
class ARROW_DS_EXPORT ParquetFileFragment : public FileFragment {
 public:
  Result<FragmentVector> SplitByRowGroup(Expression predicate);

  /// \brief Return the RowGroups selected by this fragment.
  const std::vector<int>& row_groups() const {
    if (row_groups_) return *row_groups_;
    static std::vector<int> empty;
    return empty;
  }

  /// \brief Return the FileMetaData associated with this fragment.
  const std::shared_ptr<parquet::FileMetaData>& metadata() const { return metadata_; }

  /// \brief Ensure this fragment's FileMetaData is in memory.
  Status EnsureCompleteMetadata(parquet::arrow::FileReader* reader = NULLPTR);

  /// \brief Return fragment which selects a filtered subset of this fragment's RowGroups.
  Result<std::shared_ptr<Fragment>> Subset(Expression predicate);
  Result<std::shared_ptr<Fragment>> Subset(std::vector<int> row_group_ids);

 private:
  ParquetFileFragment(FileSource source, std::shared_ptr<FileFormat> format,
                      Expression partition_expression,
                      std::shared_ptr<Schema> physical_schema,
                      util::optional<std::vector<int>> row_groups);

  Status SetMetadata(std::shared_ptr<parquet::FileMetaData> metadata,
                     std::shared_ptr<parquet::arrow::SchemaManifest> manifest);

  // Overridden to opportunistically set metadata since a reader must be opened anyway.
  Result<std::shared_ptr<Schema>> ReadPhysicalSchemaImpl() override {
    ARROW_RETURN_NOT_OK(EnsureCompleteMetadata());
    return physical_schema_;
  }

  // Return a filtered subset of row group indices.
  Result<std::vector<int>> FilterRowGroups(Expression predicate);

  ParquetFileFormat& parquet_format_;

  // Indices of row groups selected by this fragment,
  // or util::nullopt if all row groups are selected.
  util::optional<std::vector<int>> row_groups_;

  std::vector<Expression> statistics_expressions_;
  std::vector<bool> statistics_expressions_complete_;
  std::shared_ptr<parquet::FileMetaData> metadata_;
  std::shared_ptr<parquet::arrow::SchemaManifest> manifest_;

  friend class ParquetFileFormat;
  friend class ParquetDatasetFactory;
};

class ARROW_DS_EXPORT ParquetFileWriteOptions : public FileWriteOptions {
 public:
  std::shared_ptr<parquet::WriterProperties> writer_properties;

  std::shared_ptr<parquet::ArrowWriterProperties> arrow_writer_properties;

 protected:
  using FileWriteOptions::FileWriteOptions;

  friend class ParquetFileFormat;
};

class ARROW_DS_EXPORT ParquetFileWriter : public FileWriter {
 public:
  const std::shared_ptr<parquet::arrow::FileWriter>& parquet_writer() const {
    return parquet_writer_;
  }

  Status Write(const std::shared_ptr<RecordBatch>& batch) override;

 private:
  ParquetFileWriter(std::shared_ptr<io::OutputStream> destination,
                    std::shared_ptr<parquet::arrow::FileWriter> writer,
                    std::shared_ptr<ParquetFileWriteOptions> options);

  Status FinishInternal() override;

  std::shared_ptr<parquet::arrow::FileWriter> parquet_writer_;

  friend class ParquetFileFormat;
};

struct ParquetFactoryOptions {
  // Either an explicit Partitioning or a PartitioningFactory to discover one.
  //
  // If a factory is provided, it will be used to infer a schema for partition fields
  // based on file and directory paths then construct a Partitioning. The default
  // is a Partitioning which will yield no partition information.
  //
  // The (explicit or discovered) partitioning will be applied to discovered files
  // and the resulting partition information embedded in the Dataset.
  PartitioningOrFactory partitioning{Partitioning::Default()};

  // For the purposes of applying the partitioning, paths will be stripped
  // of the partition_base_dir. Files not matching the partition_base_dir
  // prefix will be skipped for partition discovery. The ignored files will still
  // be part of the Dataset, but will not have partition information.
  //
  // Example:
  // partition_base_dir = "/dataset";
  //
  // - "/dataset/US/sales.csv" -> "US/sales.csv" will be given to the partitioning
  //
  // - "/home/john/late_sales.csv" -> Will be ignored for partition discovery.
  //
  // This is useful for partitioning which parses directory when ordering
  // is important, e.g. DirectoryPartitioning.
  std::string partition_base_dir;

  // Assert that all ColumnChunk paths are consistent. The parquet spec allows for
  // ColumnChunk data to be stored in multiple files, but ParquetDatasetFactory
  // supports only a single file with all ColumnChunk data. If this flag is set
  // construction of a ParquetDatasetFactory will raise an error if ColumnChunk
  // data is not resident in a single file.
  bool validate_column_chunk_paths = false;
};

/// \brief Create FileSystemDataset from custom `_metadata` cache file.
///
/// Dask and other systems will generate a cache metadata file by concatenating
/// the RowGroupMetaData of multiple parquet files into a single parquet file
/// that only contains metadata and no ColumnChunk data.
///
/// ParquetDatasetFactory creates a FileSystemDataset composed of
/// ParquetFileFragment where each fragment is pre-populated with the exact
/// number of row groups and statistics for each columns.
class ARROW_DS_EXPORT ParquetDatasetFactory : public DatasetFactory {
 public:
  /// \brief Create a ParquetDatasetFactory from a metadata path.
  ///
  /// The `metadata_path` will be read from `filesystem`. Each RowGroup
  /// contained in the metadata file will be relative to `dirname(metadata_path)`.
  ///
  /// \param[in] metadata_path path of the metadata parquet file
  /// \param[in] filesystem from which to open/read the path
  /// \param[in] format to read the file with.
  /// \param[in] options see ParquetFactoryOptions
  static Result<std::shared_ptr<DatasetFactory>> Make(
      const std::string& metadata_path, std::shared_ptr<fs::FileSystem> filesystem,
      std::shared_ptr<ParquetFileFormat> format, ParquetFactoryOptions options);

  /// \brief Create a ParquetDatasetFactory from a metadata source.
  ///
  /// Similar to the previous Make definition, but the metadata can be a Buffer
  /// and the base_path is explicited instead of inferred from the metadata
  /// path.
  ///
  /// \param[in] metadata source to open the metadata parquet file from
  /// \param[in] base_path used as the prefix of every parquet files referenced
  /// \param[in] filesystem from which to read the files referenced.
  /// \param[in] format to read the file with.
  /// \param[in] options see ParquetFactoryOptions
  static Result<std::shared_ptr<DatasetFactory>> Make(
      const FileSource& metadata, const std::string& base_path,
      std::shared_ptr<fs::FileSystem> filesystem,
      std::shared_ptr<ParquetFileFormat> format, ParquetFactoryOptions options);

  Result<std::vector<std::shared_ptr<Schema>>> InspectSchemas(
      InspectOptions options) override;

  Result<std::shared_ptr<Dataset>> Finish(FinishOptions options) override;

 protected:
  ParquetDatasetFactory(
      std::shared_ptr<fs::FileSystem> filesystem,
      std::shared_ptr<ParquetFileFormat> format,
      std::shared_ptr<parquet::FileMetaData> metadata,
      std::shared_ptr<parquet::arrow::SchemaManifest> manifest,
      std::shared_ptr<Schema> physical_schema, std::string base_path,
      ParquetFactoryOptions options,
      std::unordered_map<std::string, std::vector<int>> path_to_row_group_ids)
      : filesystem_(std::move(filesystem)),
        format_(std::move(format)),
        metadata_(std::move(metadata)),
        manifest_(std::move(manifest)),
        physical_schema_(std::move(physical_schema)),
        base_path_(std::move(base_path)),
        options_(std::move(options)),
        path_to_row_group_ids_(std::move(path_to_row_group_ids)) {}

  std::shared_ptr<fs::FileSystem> filesystem_;
  std::shared_ptr<ParquetFileFormat> format_;
  std::shared_ptr<parquet::FileMetaData> metadata_;
  std::shared_ptr<parquet::arrow::SchemaManifest> manifest_;
  std::shared_ptr<Schema> physical_schema_;
  std::string base_path_;
  ParquetFactoryOptions options_;
  std::unordered_map<std::string, std::vector<int>> path_to_row_group_ids_;

 private:
  Result<std::vector<std::shared_ptr<FileFragment>>> CollectParquetFragments(
      const Partitioning& partitioning);

  Result<std::shared_ptr<Schema>> PartitionSchema();
};

}  // namespace dataset
}  // namespace arrow