db.cpp

#include <daemon/backend/metadata/db.hpp>
#include <daemon/backend/metadata/merge.hpp>
#include <daemon/backend/exceptions.hpp>

#include <extern/spdlog/spdlog.h>

MetadataDB::MetadataDB(const std::string& path): path(path) {
    // Optimize RocksDB. This is the easiest way to get RocksDB to perform well
    options.IncreaseParallelism();
    options.OptimizeLevelStyleCompaction();
    // create the DB if it's not already present
    options.create_if_missing = true;
    options.merge_operator.reset(new MetadataMergeOperator);
    MetadataDB::optimize_rocksdb_options(options);

#if !defined(KV_WOL)
    write_opts.disableWAL = true;
#endif
    rdb::DB * rdb_ptr;
    auto s = rocksdb::DB::Open(options, path, &rdb_ptr);
    if (!s.ok()) {
        std::runtime_error("Failed to opend RocksDB: " + s.ToString());
    }
    this->db.reset(rdb_ptr);
}

void MetadataDB::throw_rdb_status_excpt(const rdb::Status& s){
    assert(!s.ok());

    if(s.IsNotFound()){
        throw NotFoundException(s.ToString());
    } else {
        throw DBException(s.ToString());
    }
}

std::string MetadataDB::get(const std::string& key) const {
    std::string val;
    auto s = db->Get(rdb::ReadOptions(), key, &val);
    if(!s.ok()){
        MetadataDB::throw_rdb_status_excpt(s);
    }
    return val;
}

void MetadataDB::put(const std::string& key, const std::string& val) {
    auto cop = CreateOperand(val);
    auto s = db->Merge(write_opts, key, cop.serialize());
    if(!s.ok()){
        MetadataDB::throw_rdb_status_excpt(s);
    }
}

void MetadataDB::remove(const std::string& key) {
    auto s = db->Delete(write_opts, key);
    if(!s.ok()){
        MetadataDB::throw_rdb_status_excpt(s);
    }
}

bool MetadataDB::exists(const std::string& key) {
    std::string val;
    auto s = db->Get(rdb::ReadOptions(), key, &val);
    if(!s.ok()){
        if(s.IsNotFound()){
            return false;
        } else {
            MetadataDB::throw_rdb_status_excpt(s);
        }
    }
    return true;
}

/**
 * Updates a metadentry atomically and also allows to change keys
 * @param old_key
 * @param new_key
 * @param val
 * @return
 */
void MetadataDB::update(const std::string& old_key, const std::string& new_key, const std::string& val) {
    //TODO use rdb::Put() method
    rdb::WriteBatch batch;
    batch.Delete(old_key);
    batch.Put(new_key, val);
    auto s = db->Write(write_opts, &batch);
    if(!s.ok()){
        MetadataDB::throw_rdb_status_excpt(s);
    }
}

void MetadataDB::update_size(const std::string& key, size_t size, bool append){
    auto uop = IncreaseSizeOperand(size, append);
    auto s = db->Merge(write_opts, key, uop.serialize());
    if(!s.ok()){
        MetadataDB::throw_rdb_status_excpt(s);
    }
}

void MetadataDB::iterate_all() {
    std::string key;
    std::string val;
    // Do RangeScan on parent inode
    auto iter = db->NewIterator(rdb::ReadOptions());
    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
        key = iter->key().ToString();
        val = iter->value().ToString();
        //TODO ADAFS_DATA->spdlogger()->trace("key '{}' value '{}'", key, val);
    }
}

void MetadataDB::optimize_rocksdb_options(rdb::Options& options) {
    options.max_successive_merges = 128;

#if defined(KV_OPTIMIZE_RAMDISK)
    // as described at https://github.com/facebook/rocksdb/wiki/RocksDB-Tuning-Guide
    // use mmap read
    options.allow_mmap_reads = true;
    // disable block cache, enable blook filters and reduce the delta encoding restart interval
    rocksdb::BlockBasedTableOptions table_options{};
    table_options.filter_policy.reset(rocksdb::NewBloomFilterPolicy(10, true));
    table_options.no_block_cache = true;
    table_options.block_restart_interval = 4;
    options.table_factory.reset(NewBlockBasedTableFactory(table_options));
    // enable lightweight compression (snappy or lz4). We use lz4 for now
    options.compression = rocksdb::CompressionType::kLZ4Compression;
    // set up compression more aggressively and allocate more threads for flush and compaction
    options.level0_file_num_compaction_trigger = 1;
    options.max_background_flushes = 8;
    options.max_background_compactions = 8;
    options.max_subcompactions = 4;
    // keep all the files open
    options.max_open_files = -1;
#elif defined(KV_OPTIMIZE)
    // rocksdb::BlockBasedTableOptions block_options{};
    // block_options.block_size = 16384 * 2;
    // options.table_factory.reset(rocksdb::NewBlockBasedTableFactory(block_options));
    // experimental settings
    // options.write_buffer_size = 512;
    // options.max_write_buffer_number = 16;
    // options.min_write_buffer_number_to_merge = 4;
    // These 4 below have the most impact
    options.max_bytes_for_level_base = 2048;
    options.max_bytes_for_level_multiplier = 10;
    options.target_file_size_base = 256;
    options.target_file_size_multiplier = 1;

    options.max_background_flushes = 1;
    options.max_background_compactions = 48;
    options.level0_file_num_compaction_trigger = 1;
    options.level0_slowdown_writes_trigger = 48;
    options.level0_stop_writes_trigger = 56;
    // options.arena_block_size = 1024 * 8;
    // options.compression = rocksdb::kNoCompression; // doesnt do anything
#endif

#if defined(KV_WRITE_BUFFER)
    // write_buffer_size is multiplied by the write_buffer_number to get the amount of data hold in memory.
    // at min_write_buffer_number_to_merge rocksdb starts to flush entries out to disk
    options.write_buffer_size = KV_WRITE_BUFFER << 20;
    // XXX experimental values. We only want one buffer, which is held in memory
    options.max_write_buffer_number = 1;
    options.min_write_buffer_number_to_merge = 1;
#endif
}