scylladb/sstables/compress.hh

/*
 * Copyright (C) 2015-present ScyllaDB
 */

/*
 * SPDX-License-Identifier: AGPL-3.0-or-later
 */

#pragma once

// This is an implementation of a random-access compressed file compatible
// with Cassandra's org.apache.cassandra.io.compress compressed files.
//
// To allow reasonably-efficient seeking in the compressed file, the file
// is not compressed as a whole, but rather divided into chunks of a known
// size (by default, 64 KB), where each chunk is compressed individually.
// The compressed size of each chunk is different, so for allowing seeking
// to a particular position in the uncompressed data, we need to also know
// the position of each chunk. This offset vector is supplied externally as
// a "compression_metadata" object, which also contains additional information
// needed from decompression - such as the chunk size and compressor type.
//
// Cassandra supports four different compression algorithms for the chunks,
// LZ4, Snappy, Deflate, and Zstd - the default (and therefore most important) is
// LZ4. Each compressor is an implementation of the "compressor" class.
//
// Each compressed chunk is followed by a 4-byte checksum of the compressed
// data, using the Adler32 or CRC32 algorithm. In Cassandra, there is a parameter
// "crc_check_chance" (defaulting to 1.0) which determines the probability
// of us verifying the checksum of each chunk we read.
//
// This implementation does not cache the compressed disk blocks (which
// are read using O_DIRECT), nor uncompressed data. We intend to cache high-
// level Cassandra rows, not disk blocks.

#include "utils/assert.hh"
#include <vector>
#include <cstdint>
#include <iterator>
#include <deque>

#include <seastar/core/file.hh>
#include <seastar/core/seastar.hh>
#include <seastar/core/shared_ptr.hh>
#include <seastar/core/fstream.hh>

#include "types/types.hh"
#include "sstables/types.hh"
#include "checksum_utils.hh"
#include "../compress.hh"

class reader_permit;

class compression_parameters;
class compressor;
using compressor_ptr = shared_ptr<compressor>;

namespace sstables {

struct compression {
    // To reduce the memory footpring of compression-info, n offsets are grouped
    // together into segments, where each segment stores a base absolute offset
    // into the file, the other offsets in the segments being relative offsets
    // (and thus of reduced size). Also offsets are allocated only just enough
    // bits to store their maximum value. The offsets are thus packed in a
    // buffer like so:
    //      arrrarrrarrr...
    // where n is 4, a is an absolute offset and r are offsets relative to a.
    // Segments are stored in buckets, where each bucket has its own base offset.
    // Segments in a buckets are optimized to address as large of a chunk of the
    // data as possible for a given chunk size and bucket size.
    //
    // This is not a general purpose container. There are limitations:
    // * Can't be used before init() is called.
    // * at() is best called incrementally, although random lookups are
    // perfectly valid as well.
    // * The iterator and at() can't provide references to the elements.
    // * No point insert is available.
    class segmented_offsets {
    public:
        class state {
            std::size_t _current_index{0};
            std::size_t _current_bucket_index{0};
            uint64_t _current_bucket_segment_index{0};
            uint64_t _current_segment_relative_index{0};
            uint64_t _current_segment_offset_bits{0};

            void update_position_trackers(std::size_t index, uint16_t segment_size_bits,
                uint32_t segments_per_bucket, uint8_t grouped_offsets);

            friend class segmented_offsets;
        };

        class accessor {
            const segmented_offsets& _offsets;
            mutable state _state;
        public:
            accessor(const segmented_offsets& offsets) : _offsets(offsets) { }

            uint64_t at(std::size_t i) const {
                return _offsets.at(i, _state);
            }
        };

        class writer {
            segmented_offsets& _offsets;
            state _state;
        public:
            writer(segmented_offsets& offsets) : _offsets(offsets) { }

            void push_back(uint64_t offset) {
                return _offsets.push_back(offset, _state);
            }
        };

        accessor get_accessor() const {
            return accessor(*this);
        }

        writer get_writer() {
            return writer(*this);
        }
    private:
        struct bucket {
            uint64_t base_offset;
            std::unique_ptr<char[]> storage;
        };

        uint32_t _chunk_size{0};
        uint8_t _segment_base_offset_size_bits{0};
        uint8_t _segmented_offset_size_bits{0};
        uint16_t _segment_size_bits{0};
        uint32_t _segments_per_bucket{0};
        uint8_t _grouped_offsets{0};

        uint64_t _last_written_offset{0};

        std::size_t _size{0};
        std::deque<bucket> _storage;

        uint64_t read(uint64_t bucket_index, uint64_t offset_bits, uint64_t size_bits) const;
        void write(uint64_t bucket_index, uint64_t offset_bits, uint64_t size_bits, uint64_t value);

        uint64_t at(std::size_t i, state& s) const;
        void push_back(uint64_t offset, state& s);
    public:
        class const_iterator {
        public:
            using iterator_category = std::random_access_iterator_tag;
            using value_type = uint64_t;
            using difference_type = std::ptrdiff_t;
            using pointer = const uint64_t*;
            using reference = const uint64_t&;
        private:
            friend class segmented_offsets;
            struct end_tag {};

            segmented_offsets::accessor _offsets;
            std::size_t _index;

            const_iterator(const segmented_offsets& offsets)
                : _offsets(offsets.get_accessor())
                , _index(0) {
            }

            const_iterator(const segmented_offsets& offsets, end_tag)
                : _offsets(offsets.get_accessor())
                , _index(offsets.size()) {
            }

        public:
            const_iterator(const const_iterator& other) = default;

            const_iterator& operator=(const const_iterator& other) {
                SCYLLA_ASSERT(&_offsets == &other._offsets);
                _index = other._index;
                return *this;
            }

            const_iterator operator++(int) {
                const_iterator it{*this};
                return ++it;
            }

            const_iterator& operator++() {
                *this += 1;
                return *this;
            }

            const_iterator operator+(ssize_t i) const {
                const_iterator it{*this};
                it += i;
                return it;
            }

            const_iterator& operator+=(ssize_t i) {
                _index += i;

                return *this;
            }

            const_iterator operator--(int) {
                const_iterator it{*this};
                return --it;
            }

            const_iterator& operator--() {
                *this -= 1;
                return *this;
            }

            const_iterator operator-(ssize_t i) const {
                const_iterator it{*this};
                it -= i;
                return it;
            }

            const_iterator& operator-=(ssize_t i) {
                _index -= i;
                return *this;
            }

            value_type operator*() const {
                return _offsets.at(_index);
            }

            value_type operator[](ssize_t i) const {
                return _offsets.at(_index + i);
            }

            bool operator==(const const_iterator& other) const {
                return _index == other._index;
            }

            bool operator<(const const_iterator& other) const {
                return _index < other._index;
            }

            bool operator<=(const const_iterator& other) const {

                return _index <= other._index;
            }

            bool operator>(const const_iterator& other) const {
                return _index > other._index;
            }

            bool operator>=(const const_iterator& other) const {
                return _index >= other._index;
            }
        };

        segmented_offsets() = default;

        segmented_offsets(const segmented_offsets&) = delete;
        segmented_offsets& operator=(const segmented_offsets&) = delete;

        segmented_offsets(segmented_offsets&&) = default;
        segmented_offsets& operator=(segmented_offsets&&) = default;

        // Has to be called before using the class. Doing otherwise
        // results in undefined behaviour! Don't call more than once!
        // TODO: fold into constructor, once the parse() et. al. code
        // allows it.
        void init(uint32_t chunk_size);

        uint32_t chunk_size() const noexcept {
            return _chunk_size;
        }

        std::size_t size() const noexcept {
            return _size;
        }

        const_iterator begin() const {
            return const_iterator(*this);
        }

        const_iterator end() const {
            return const_iterator(*this, const_iterator::end_tag{});
        }

        const_iterator cbegin() const {
            return const_iterator(*this);
        }

        const_iterator cend() const {
            return const_iterator(*this, const_iterator::end_tag{});
        }
    };

    disk_string<uint16_t> name;
    disk_array<uint32_t, option> options;
    uint32_t chunk_len = 0;
    uint64_t data_len = 0;
    segmented_offsets offsets;

private:
    // Variables *not* found in the "Compression Info" file (added by update()):
    uint64_t _compressed_file_length = 0;
    uint32_t _full_checksum = 0;
public:
    // Set the compressor algorithm, please check the definition of enum compressor.
    void set_compressor(compressor_ptr c);
    // After changing _compression, update() must be called to update
    // additional variables depending on it.    
    void update(uint64_t compressed_file_length);
    operator bool() const {
        return !name.value.empty();
    }
    // locate() locates in the compressed file the given byte position of
    // the uncompressed data:
    //   1. The byte range containing the appropriate compressed chunk, and
    //   2. the offset into the uncompressed chunk.
    // Note that the last 4 bytes of the returned chunk are not the actual
    // compressed data, but rather the checksum of the compressed data.
    // locate() throws an out-of-range exception if the position is beyond
    // the last chunk.
    struct chunk_and_offset {
        uint64_t chunk_start;
        uint64_t chunk_len; // variable size of compressed chunk
        unsigned offset; // offset into chunk after uncompressing it
    };
    chunk_and_offset locate(uint64_t position, const compression::segmented_offsets::accessor& accessor);

    unsigned uncompressed_chunk_length() const noexcept {
        return chunk_len;
    }

    void set_uncompressed_chunk_length(uint32_t cl) {
        chunk_len = cl;

        offsets.init(chunk_len);
    }

    uint64_t uncompressed_file_length() const noexcept {
        return data_len;
    }

    void set_uncompressed_file_length(uint64_t fl) {
        data_len = fl;
    }

    uint64_t compressed_file_length() const {
        return _compressed_file_length;
    }
    void set_compressed_file_length(uint64_t compressed_file_length) {
        _compressed_file_length = compressed_file_length;
    }

    uint32_t get_full_checksum() const {
        return _full_checksum;
    }

    void set_full_checksum(uint32_t checksum) {
        _full_checksum = checksum;
    }

    friend class sstable;
};

// for API query only. Free function just to distinguish it from an accessor in compression
compressor_ptr get_sstable_compressor(const compression&);

// Note: compression_metadata is passed by reference; The caller is
// responsible for keeping the compression_metadata alive as long as there
// are open streams on it. This should happen naturally on a higher level -
// as long as we have *sstables* work in progress, we need to keep the whole
// sstable alive, and the compression metadata is only a part of it.
input_stream<char> make_compressed_file_k_l_format_input_stream(file f,
                sstables::compression* cm, uint64_t offset, size_t len,
                class file_input_stream_options options, reader_permit permit);

input_stream<char> make_compressed_file_m_format_input_stream(file f,
                sstables::compression* cm, uint64_t offset, size_t len,
                class file_input_stream_options options, reader_permit permit);

output_stream<char> make_compressed_file_m_format_output_stream(output_stream<char> out,
                sstables::compression* cm,
                const compression_parameters& cp);

}