scylladb/sstables/compress.cc

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

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

#include <stdexcept>
#include <cstdlib>

#include <boost/range/algorithm/find_if.hpp>
#include <seastar/core/align.hh>
#include <seastar/core/bitops.hh>
#include <seastar/core/byteorder.hh>
#include <seastar/core/fstream.hh>

#include "../compress.hh"
#include "compress.hh"
#include "exceptions.hh"
#include "unimplemented.hh"
#include "segmented_compress_params.hh"
#include "utils/class_registrator.hh"

namespace sstables {

extern logging::logger sstlog;

enum class mask_type : uint8_t {
    set,
    clear
};

// size_bits cannot be >= 64
static inline uint64_t make_mask(uint8_t size_bits, uint8_t offset, mask_type t) noexcept {
    const uint64_t mask = ((1 << size_bits) - 1) << offset;
    return t == mask_type::set ? mask : ~mask;
}

/*
 * ----> memory addresses
 * MSB           LSB
 * | | | | |3|2|1|0| CPU integer (big or little endian byte order)
 *          -------
 *             |
 *       +-----+ << shift = prefix bits
 *       |
 *    -------
 *  7 6 5 4 3 2 1 0  index*
 * | |3|2|1|0| | | | raw storage (unaligned, little endian byte order)
 *  = ------- =====
 *  |           |
 *  |           +-> prefix bits
 *  +-> suffix bits
 *
 * |0|1|1|1|1|0|0|0| read/write mask
 *
 * * On big endian systems the indices in storage are reversed and
 *   run left to right: 0 1 .. 6 7. To avoid differences in the
 *   encoding logic on machines with different native byte orders
 *   reads and writes to storage must be explicitly little endian.
 */
struct bit_displacement {
    uint64_t shift;
    uint64_t mask;
};

inline bit_displacement displacement_for(uint64_t prefix_bits, uint8_t size_bits, mask_type t) {
    return {prefix_bits, make_mask(size_bits, prefix_bits, t)};
}

std::pair<bucket_info, segment_info> params_for_chunk_size(uint32_t chunk_size) {
    const uint8_t chunk_size_log2 = log2ceil(chunk_size);

    auto it = boost::find_if(bucket_infos, [&] (const bucket_info& bi) {
        return bi.chunk_size_log2 == chunk_size_log2;
    });

    // This scenario should be so rare that we only fall back to a safe
    // set of parameters, not optimal ones.
    if (it == bucket_infos.end()) {
        const uint8_t data_size = bucket_infos.front().best_data_size_log2;
        return {{chunk_size_log2, data_size, (8 * bucket_size - 56) / data_size},
            {chunk_size_log2, data_size, uint8_t(1)}};
    }

    auto b = *it;
    auto s = *boost::find_if(segment_infos, [&] (const segment_info& si) {
        return si.data_size_log2 == b.best_data_size_log2 && si.chunk_size_log2 == b.chunk_size_log2;
    });

    return {std::move(b), std::move(s)};
}

uint64_t compression::segmented_offsets::read(uint64_t bucket_index, uint64_t offset_bits, uint64_t size_bits) const {
    const uint64_t offset_byte = offset_bits / 8;
    uint64_t value = seastar::read_le<uint64_t>(_storage[bucket_index].storage.get() + offset_byte);

    const auto displacement = displacement_for(offset_bits % 8, size_bits, mask_type::set);

    value &= displacement.mask;
    value >>= displacement.shift;

    return value;
}

void compression::segmented_offsets::write(uint64_t bucket_index, uint64_t offset_bits, uint64_t size_bits, uint64_t value) {
    const uint64_t offset_byte = offset_bits / 8;

    uint64_t old_value = seastar::read_le<uint64_t>(_storage[bucket_index].storage.get() + offset_byte);

    const auto displacement = displacement_for(offset_bits % 8, size_bits, mask_type::clear);

    value <<= displacement.shift;

    if ((~displacement.mask | value) != ~displacement.mask) {
        throw std::invalid_argument(format("{}: to-be-written value would overflow the allocated bits", __FUNCTION__));
    }

    old_value &= displacement.mask;
    value |= old_value;

    seastar::write_le(_storage[bucket_index].storage.get() + offset_byte, value);
}

void compression::segmented_offsets::state::update_position_trackers(std::size_t index, uint16_t segment_size_bits,
        uint32_t segments_per_bucket, uint8_t grouped_offsets) {
    if (_current_index != index - 1) {
        _current_index = index;
        const uint64_t current_segment_index = _current_index / grouped_offsets;
        _current_bucket_segment_index = current_segment_index % segments_per_bucket;
        _current_segment_relative_index = _current_index % grouped_offsets;
        _current_bucket_index = current_segment_index / segments_per_bucket;
        _current_segment_offset_bits = (_current_bucket_segment_index % segments_per_bucket) * segment_size_bits;
    } else {
        ++_current_index;
        ++_current_segment_relative_index;

        // Crossed segment boundary.
        if (_current_segment_relative_index == grouped_offsets) {
            ++_current_bucket_segment_index;
            _current_segment_relative_index = 0;

            // Crossed bucket boundary.
            if (_current_bucket_segment_index == segments_per_bucket) {
                ++_current_bucket_index;
                _current_bucket_segment_index = 0;
                _current_segment_offset_bits = 0;
            } else {
                _current_segment_offset_bits += segment_size_bits;
            }
        }
    }
}

void compression::segmented_offsets::init(uint32_t chunk_size) {
    assert(chunk_size != 0);

    _chunk_size = chunk_size;

    const auto params = params_for_chunk_size(chunk_size);

    sstlog.trace(
            "{} {}(): chunk size {} (log2)",
            fmt::ptr(this),
            __FUNCTION__,
            static_cast<int>(params.first.chunk_size_log2));

    _grouped_offsets = params.second.grouped_offsets;
    _segment_base_offset_size_bits = params.second.data_size_log2;
    _segmented_offset_size_bits = static_cast<uint64_t>(log2ceil((_chunk_size + 64) * (_grouped_offsets - 1)));
    _segment_size_bits = _segment_base_offset_size_bits + (_grouped_offsets - 1) * _segmented_offset_size_bits;
    _segments_per_bucket = params.first.segments_per_bucket;
}

uint64_t compression::segmented_offsets::at(std::size_t i, compression::segmented_offsets::state& s) const {
    if (i >= _size) {
        throw std::out_of_range(format("{}: index {} is out of range", __FUNCTION__, i));
    }

    s.update_position_trackers(i, _segment_size_bits, _segments_per_bucket, _grouped_offsets);
    const uint64_t bucket_base_offset = _storage[s._current_bucket_index].base_offset;
    const uint64_t segment_base_offset = bucket_base_offset + read(s._current_bucket_index, s._current_segment_offset_bits, _segment_base_offset_size_bits);

    if (s._current_segment_relative_index == 0) {
        return segment_base_offset;
    }

    return segment_base_offset
        + read(s._current_bucket_index,
                s._current_segment_offset_bits + _segment_base_offset_size_bits + (s._current_segment_relative_index - 1) * _segmented_offset_size_bits,
                _segmented_offset_size_bits);
}

void compression::segmented_offsets::push_back(uint64_t offset, compression::segmented_offsets::state& s) {
    s.update_position_trackers(_size, _segment_size_bits, _segments_per_bucket, _grouped_offsets);

    if (s._current_bucket_index == _storage.size()) {
        _storage.push_back(bucket{_last_written_offset, std::unique_ptr<char[]>(new char[bucket_size])});
    }

    const uint64_t bucket_base_offset = _storage[s._current_bucket_index].base_offset;

    if (s._current_segment_relative_index == 0) {
        write(s._current_bucket_index, s._current_segment_offset_bits, _segment_base_offset_size_bits, offset - bucket_base_offset);
    } else {
        const uint64_t segment_base_offset = bucket_base_offset + read(s._current_bucket_index, s._current_segment_offset_bits, _segment_base_offset_size_bits);
        write(s._current_bucket_index,
                s._current_segment_offset_bits + _segment_base_offset_size_bits + (s._current_segment_relative_index - 1) * _segmented_offset_size_bits,
                _segmented_offset_size_bits,
                offset - segment_base_offset);
    }
    _last_written_offset = offset;
    ++_size;
}

/**
 * Thin wrapper type around a "compressor" object.
 * This is the non-shared part of sstable compression,
 * since the compressor might have both dependents and
 * semi-state that cannot be shared across shards.
 *
 * The local state is instantiated in either
 * reader/writer object for the sstable on demand,
 * typically based on the shared compression
 * info/table schema.
 */
class local_compression {
    compressor_ptr _compressor;
public:
    local_compression()= default;
    local_compression(const compression&);
    local_compression(compressor_ptr);

    size_t uncompress(const char* input, size_t input_len, char* output,
                    size_t output_len) const;
    size_t compress(const char* input, size_t input_len, char* output,
                    size_t output_len) const;
    size_t compress_max_size(size_t input_len) const;

    operator bool() const {
        return _compressor != nullptr;
    }
    const compressor_ptr& compressor() const {
        return _compressor;
    }
};

local_compression::local_compression(compressor_ptr p)
    : _compressor(std::move(p))
{}

local_compression::local_compression(const compression& c)
    : _compressor([&c] {
        sstring n(c.name.value.begin(), c.name.value.end());
        return compressor::create(n, [&c, &n](const sstring& key) -> compressor::opt_string {
            if (key == compression_parameters::CHUNK_LENGTH_KB || key == compression_parameters::CHUNK_LENGTH_KB_ERR) {
                return to_sstring(c.chunk_len / 1024);
            }
            if (key == compression_parameters::SSTABLE_COMPRESSION) {
                return n;
            }
            for (auto& o : c.options.elements) {
                if (key == sstring(o.key.value.begin(), o.key.value.end())) {
                    return sstring(o.value.value.begin(), o.value.value.end());
                }
            }
            return std::nullopt;
        });
    }())
{}

size_t local_compression::uncompress(const char* input,
                size_t input_len, char* output, size_t output_len) const {
    if (!_compressor) {
        throw std::runtime_error("uncompress is not supported");
    }
    return _compressor->uncompress(input, input_len, output, output_len);
}
size_t local_compression::compress(const char* input, size_t input_len,
                char* output, size_t output_len) const {
    if (!_compressor) {
        throw std::runtime_error("compress is not supported");
    }
    return _compressor->compress(input, input_len, output, output_len);
}
size_t local_compression::compress_max_size(size_t input_len) const {
    return _compressor ? _compressor->compress_max_size(input_len) : 0;
}

void compression::set_compressor(compressor_ptr c) {
    if (c) {
        unqualified_name uqn(compressor::namespace_prefix, c->name());
        const sstring& cn = uqn;
        name.value = bytes(cn.begin(), cn.end());
        for (auto& p : c->options()) {
            if (p.first != compression_parameters::SSTABLE_COMPRESSION) {
                auto& k = p.first;
                auto& v = p.second;
                options.elements.push_back({bytes(k.begin(), k.end()), bytes(v.begin(), v.end())});
            }
        }
    }
}

void compression::update(uint64_t compressed_file_length) {
    _compressed_file_length = compressed_file_length;
}

compressor_ptr get_sstable_compressor(const compression& c) {
    return local_compression(c).compressor();
}

// locate() takes a byte position in the uncompressed stream, and finds the
// the location of the compressed chunk on disk which contains it, and the
// offset in this chunk.
// locate() may only be used for offsets of actual bytes, and in particular
// the end-of-file position (one past the last byte) MUST not be used. If the
// caller wants to read from the end of file, it should simply read nothing.
compression::chunk_and_offset
compression::locate(uint64_t position, const compression::segmented_offsets::accessor& accessor) {
    auto ucl = uncompressed_chunk_length();
    auto chunk_index = position / ucl;
    decltype(ucl) chunk_offset = position % ucl;
    auto chunk_start = accessor.at(chunk_index);
    auto chunk_end = (chunk_index + 1 == offsets.size())
            ? _compressed_file_length
            : accessor.at(chunk_index + 1);
    return { chunk_start, chunk_end - chunk_start, chunk_offset };
}

}

template <typename ChecksumType>
requires ChecksumUtils<ChecksumType>
class compressed_file_data_source_impl : public data_source_impl {
    std::optional<input_stream<char>> _input_stream;
    sstables::compression* _compression_metadata;
    sstables::compression::segmented_offsets::accessor _offsets;
    sstables::local_compression _compression;
    uint64_t _underlying_pos;
    uint64_t _pos;
    uint64_t _beg_pos;
    uint64_t _end_pos;
public:
    compressed_file_data_source_impl(file f, sstables::compression* cm,
                uint64_t pos, size_t len, file_input_stream_options options)
            : _compression_metadata(cm)
            , _offsets(_compression_metadata->offsets.get_accessor())
            , _compression(*cm)
    {
        _beg_pos = pos;
        if (pos > _compression_metadata->uncompressed_file_length()) {
            throw std::runtime_error("attempt to uncompress beyond end");
        }
        if (len == 0 || pos == _compression_metadata->uncompressed_file_length()) {
            // Nothing to read
            _end_pos = _pos = _beg_pos;
            return;
        }
        if (len <= _compression_metadata->uncompressed_file_length() - pos) {
            _end_pos = pos + len;
        } else {
            _end_pos = _compression_metadata->uncompressed_file_length();
        }
        // _beg_pos and _end_pos specify positions in the compressed stream.
        // We need to translate them into a range of uncompressed chunks,
        // and open a file_input_stream to read that range.
        auto start = _compression_metadata->locate(_beg_pos, _offsets);
        auto end = _compression_metadata->locate(_end_pos - 1, _offsets);
        _input_stream = make_file_input_stream(std::move(f),
                start.chunk_start,
                end.chunk_start + end.chunk_len - start.chunk_start,
                std::move(options));
        _underlying_pos = start.chunk_start;
        _pos = _beg_pos;
    }
    virtual future<temporary_buffer<char>> get() override {
        if (_pos >= _end_pos) {
            return make_ready_future<temporary_buffer<char>>();
        }
        auto addr = _compression_metadata->locate(_pos, _offsets);
        // Uncompress the next chunk. We need to skip part of the first
        // chunk, but then continue to read from beginning of chunks.
        if (_pos != _beg_pos && addr.offset != 0) {
            throw std::runtime_error("compressed reader out of sync");
        }
        return _input_stream->read_exactly(addr.chunk_len).
            then([this, addr](temporary_buffer<char> buf) {
                // The last 4 bytes of the chunk are the adler32/crc32 checksum
                // of the rest of the (compressed) chunk.
                auto compressed_len = addr.chunk_len - 4;
                // FIXME: Do not always calculate checksum - Cassandra has a
                // probability (defaulting to 1.0, but still...)
                auto expected_checksum = read_be<uint32_t>(buf.get() + compressed_len);
                auto actual_checksum = ChecksumType::checksum(buf.get(), compressed_len);
                if (expected_checksum != actual_checksum) {
                    throw sstables::malformed_sstable_exception(format("compressed chunk of size {} at file offset {} failed checksum, expected={}, actual={}", addr.chunk_len, _underlying_pos, expected_checksum, actual_checksum));
                }

                // We know that the uncompressed data will take exactly
                // chunk_length bytes (or less, if reading the last chunk).
                temporary_buffer<char> out(
                        _compression_metadata->uncompressed_chunk_length());
                // The compressed data is the whole chunk, minus the last 4
                // bytes (which contain the checksum verified above).

                auto len = _compression.uncompress(buf.get(), compressed_len, out.get_write(), out.size());

                out.trim(len);
                out.trim_front(addr.offset);
                _pos += out.size();
                _underlying_pos += addr.chunk_len;

                return out;
        });
    }

    virtual future<> close() override {
        if (!_input_stream) {
            return make_ready_future<>();
        }
        return _input_stream->close();
    }

    virtual future<temporary_buffer<char>> skip(uint64_t n) override {
        _pos += n;
        assert(_pos <= _end_pos);
        if (_pos == _end_pos) {
            return make_ready_future<temporary_buffer<char>>();
        }
        auto addr = _compression_metadata->locate(_pos, _offsets);
        auto underlying_n = addr.chunk_start - _underlying_pos;
        _underlying_pos = addr.chunk_start;
        _beg_pos = _pos;
        return _input_stream->skip(underlying_n).then([] {
            return make_ready_future<temporary_buffer<char>>();
        });
    }
};

template <typename ChecksumType>
requires ChecksumUtils<ChecksumType>
class compressed_file_data_source : public data_source {
public:
    compressed_file_data_source(file f, sstables::compression* cm,
            uint64_t offset, size_t len, file_input_stream_options options)
        : data_source(std::make_unique<compressed_file_data_source_impl<ChecksumType>>(
                std::move(f), cm, offset, len, std::move(options)))
        {}
};

template <typename ChecksumType>
requires ChecksumUtils<ChecksumType>
inline input_stream<char> make_compressed_file_input_stream(
        file f, sstables::compression *cm, uint64_t offset, size_t len,
        file_input_stream_options options)
{
    return input_stream<char>(compressed_file_data_source<ChecksumType>(
            std::move(f), cm, offset, len, std::move(options)));
}

// For SSTables 2.x (formats 'ka' and 'la'), the full checksum is a combination of checksums of compressed chunks.
// For SSTables 3.x (format 'mc'), however, it is supposed to contain the full checksum of the file written so
// the per-chunk checksums also count.
enum class compressed_checksum_mode {
    checksum_chunks_only,
    checksum_all,
};

// compressed_file_data_sink_impl works as a filter for a file output stream,
// where the buffer flushed will be compressed and its checksum computed, then
// the result passed to a regular output stream.
template <typename ChecksumType, compressed_checksum_mode mode>
requires ChecksumUtils<ChecksumType>
class compressed_file_data_sink_impl : public data_sink_impl {
    output_stream<char> _out;
    sstables::compression* _compression_metadata;
    sstables::compression::segmented_offsets::writer _offsets;
    sstables::local_compression _compression;
    size_t _pos = 0;
    uint32_t _full_checksum;
public:
    compressed_file_data_sink_impl(output_stream<char> out, sstables::compression* cm, sstables::local_compression lc)
            : _out(std::move(out))
            , _compression_metadata(cm)
            , _offsets(_compression_metadata->offsets.get_writer())
            , _compression(lc)
            , _full_checksum(ChecksumType::init_checksum())
    {}

    virtual future<> put(net::packet data) override { abort(); }
    virtual future<> put(temporary_buffer<char> buf) override {
        auto output_len = _compression.compress_max_size(buf.size());

        // account space for checksum that goes after compressed data.
        temporary_buffer<char> compressed(output_len + 4);

        // compress flushed data.
        auto len = _compression.compress(buf.get(), buf.size(), compressed.get_write(), output_len);
        if (len > output_len) {
            return make_exception_future(std::runtime_error("possible overflow during compression"));
        }

        // total length of the uncompressed data.
        _compression_metadata->set_uncompressed_file_length(_compression_metadata->uncompressed_file_length() + buf.size());

        _offsets.push_back(_pos);
        // account compressed data + 32-bit checksum.
        _pos += len + 4;
        _compression_metadata->set_compressed_file_length(_pos);

        // compute 32-bit checksum for compressed data.
        uint32_t per_chunk_checksum = ChecksumType::checksum(compressed.get(), len);
        _full_checksum = checksum_combine_or_feed<ChecksumType>(_full_checksum, per_chunk_checksum, compressed.get(), len);

        // write checksum into buffer after compressed data.
        write_be<uint32_t>(compressed.get_write() + len, per_chunk_checksum);

        if constexpr (mode == compressed_checksum_mode::checksum_all) {
            uint32_t be_per_chunk_checksum = cpu_to_be(per_chunk_checksum);
            _full_checksum = ChecksumType::checksum(_full_checksum,
                reinterpret_cast<const char*>(&be_per_chunk_checksum), sizeof(be_per_chunk_checksum));
        }

        _compression_metadata->set_full_checksum(_full_checksum);

        compressed.trim(len + 4);

        auto f = _out.write(compressed.get(), compressed.size());
        return f.then([compressed = std::move(compressed)] {});
    }
    virtual future<> close() override {
        return _out.close();
    }

    virtual size_t buffer_size() const noexcept override {
        return _compression_metadata->uncompressed_chunk_length();
    }
};

template <typename ChecksumType, compressed_checksum_mode mode>
requires ChecksumUtils<ChecksumType>
class compressed_file_data_sink : public data_sink {
public:
    compressed_file_data_sink(output_stream<char> out, sstables::compression* cm, sstables::local_compression lc)
        : data_sink(std::make_unique<compressed_file_data_sink_impl<ChecksumType, mode>>(
                std::move(out), cm, std::move(lc))) {}
};

template <typename ChecksumType, compressed_checksum_mode mode>
requires ChecksumUtils<ChecksumType>
inline output_stream<char> make_compressed_file_output_stream(output_stream<char> out,
         sstables::compression* cm,
         const compression_parameters& cp) {
    // buffer of output stream is set to chunk length, because flush must
    // happen every time a chunk was filled up.

    auto p = cp.get_compressor();
    cm->set_compressor(p);
    cm->set_uncompressed_chunk_length(cp.chunk_length());
    // FIXME: crc_check_chance can be configured by the user.
    // probability to verify the checksum of a compressed chunk we read.
    // defaults to 1.0.
    cm->options.elements.push_back({"crc_check_chance", "1.0"});

    return output_stream<char>(compressed_file_data_sink<ChecksumType, mode>(std::move(out), cm, p));
}

input_stream<char> sstables::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)
{
    return make_compressed_file_input_stream<adler32_utils>(std::move(f), cm, offset, len, std::move(options));
}

input_stream<char> sstables::make_compressed_file_m_format_input_stream(file f,
        sstables::compression *cm, uint64_t offset, size_t len,
        class file_input_stream_options options) {
    return make_compressed_file_input_stream<crc32_utils>(std::move(f), cm, offset, len, std::move(options));
}

output_stream<char> sstables::make_compressed_file_m_format_output_stream(output_stream<char> out,
        sstables::compression* cm,
        const compression_parameters& cp) {
    return make_compressed_file_output_stream<crc32_utils, compressed_checksum_mode::checksum_all>(
            std::move(out), cm, cp);
}