diff --git a/CMakeLists.txt b/CMakeLists.txt index 6790b5ec27..d41254db83 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -2161,7 +2161,6 @@ else () src/ripple/nodestore/backend/NuDBFactory.cpp src/ripple/nodestore/backend/NullFactory.cpp src/ripple/nodestore/backend/RocksDBFactory.cpp - src/ripple/nodestore/backend/RocksDBQuickFactory.cpp src/ripple/nodestore/impl/BatchWriter.cpp src/ripple/nodestore/impl/Database.cpp src/ripple/nodestore/impl/DatabaseNodeImp.cpp diff --git a/cfg/rippled-example.cfg b/cfg/rippled-example.cfg index b4a0aa7630..285b82c91b 100644 --- a/cfg/rippled-example.cfg +++ b/cfg/rippled-example.cfg @@ -828,13 +828,7 @@ # RocksDB is an alternative backend for systems that don't use solid-state # drives. Because RocksDB's performance degrades as it stores more data, # keeping full history is not advised, and using online delete is -# recommended. RocksDB is not available on Windows. -# -# The RocksDB backend also provides these optional parameters: -# -# compression 0 for none, 1 for Snappy compression -# -# +# recommended. # # Required keys: # path Location to store the database (all types) @@ -884,10 +878,6 @@ # # type = RocksDB # -# The RocksDB backend also provides these optional parameters: -# -# compression 0 for none, 1 for Snappy compression -# # Required keys: # path Location to store the database (all types) # diff --git a/src/ripple/nodestore/Benchmarks.md b/src/ripple/nodestore/Benchmarks.md index 1b5473a69d..ed32fb0661 100644 --- a/src/ripple/nodestore/Benchmarks.md +++ b/src/ripple/nodestore/Benchmarks.md @@ -1,21 +1,16 @@ #Benchmarks -``` -$rippled --unittest=NodeStoreTiming --unittest-arg="type=rocksdb,num_objects=2000000,open_files=2000,filter_bits=12,cache_mb=256,file_size_mb=8,file_size_mult=2;type=rocksdbquick,num_objects=2000000" -2014-Nov-01 21:49:02 Validators:NFO Validators constructed (info) -ripple.bench.NodeStoreTiming repeatableObject - Config Run Inserts Batch Insert Fetch 50/50 Ordered Fetch Fetch Random Fetch Missing - 0 0 160.57 699.08 50.88 51.17 29.99 14.05 - 0 1 406.70 797.47 32.53 60.18 46.63 14.94 - 0 2 408.81 743.89 42.79 72.99 49.03 14.93 - 1 0 111.03 151.06 28.89 53.44 31.88 18.46 - 1 1 92.63 160.75 19.64 41.60 28.17 10.40 - 1 2 101.31 122.83 30.66 55.65 32.69 16.15 +The `NodeStore.Timing` test is used to execute a set of read/write workloads to compare current available nodestore backends. It can be executed with: -Configs: - 0: type=rocksdb,num_objects=2000000,open_files=2000,filter_bits=12,cache_mb=256,file_size_mb=8,file_size_mult=2 - 1: type=rocksdbquick,num_objects=2000000 ``` +$rippled --unittest=NodeStoreTiming +``` + +It is also possible to use alternate DB config params by passing config strings as `--unittest-arg`. + +##Addendum + +The discussion below refers to a `RocksDBQuick` backend that has since been removed from the code as it was not working and not maintained. That backend primarily used one of the several rocks `Optimize*` methods to setup the majority of the DB options/params, whereas the primary RocksDB backend exposes many of the available config options directly. The code for RocksDBQuick can be found in versions of this repo 1.2 and earlier if you need to refer back to it. The conclusions below date from about 2014 and may need revisiting based on newer versions of RocksDB (TBD). ##Discussion diff --git a/src/ripple/nodestore/backend/RocksDBFactory.cpp b/src/ripple/nodestore/backend/RocksDBFactory.cpp index 2315ca30fc..fb2a6bd99c 100644 --- a/src/ripple/nodestore/backend/RocksDBFactory.cpp +++ b/src/ripple/nodestore/backend/RocksDBFactory.cpp @@ -156,11 +156,7 @@ public: m_options.max_background_flushes = highThreads; } - if (keyValues.exists ("compression") && - (get(keyValues, "compression") == 0)) - { - m_options.compression = rocksdb::kNoCompression; - } + m_options.compression = rocksdb::kSnappyCompression; get_if_exists (keyValues, "block_size", table_options.block_size); diff --git a/src/ripple/nodestore/backend/RocksDBQuickFactory.cpp b/src/ripple/nodestore/backend/RocksDBQuickFactory.cpp deleted file mode 100644 index a8d1f1cf3d..0000000000 --- a/src/ripple/nodestore/backend/RocksDBQuickFactory.cpp +++ /dev/null @@ -1,428 +0,0 @@ -//------------------------------------------------------------------------------ -/* - This file is part of rippled: https://github.com/ripple/rippled - Copyright (c) 2012, 2013 Ripple Labs Inc. - - Permission to use, copy, modify, and/or distribute this software for any - purpose with or without fee is hereby granted, provided that the above - copyright notice and this permission notice appear in all copies. - - THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES - WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF - MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR - ANY SPECIAL , DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES - WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN - ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF - OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. -*/ -//============================================================================== - - -#include - -#if RIPPLE_ROCKSDB_AVAILABLE - -#include -#include -#include // VFALCO Bad dependency -#include -#include -#include -#include -#include -#include -#include - -namespace ripple { -namespace NodeStore { - -class RocksDBQuickEnv : public rocksdb::EnvWrapper -{ -public: - RocksDBQuickEnv () - : EnvWrapper (rocksdb::Env::Default()) - { - } - - struct ThreadParams - { - ThreadParams (void (*f_)(void*), void* a_) - : f (f_) - , a (a_) - { - } - - void (*f)(void*); - void* a; - }; - - static - void - thread_entry (void* ptr) - { - ThreadParams* const p (reinterpret_cast (ptr)); - void (*f)(void*) = p->f; - void* a (p->a); - delete p; - - static std::atomic n; - std::size_t const id (++n); - std::stringstream ss; - ss << "rocksdb #" << id; - beast::setCurrentThreadName (ss.str()); - - (*f)(a); - } - - void - StartThread (void (*f)(void*), void* a) override - { - ThreadParams* const p (new ThreadParams (f, a)); - EnvWrapper::StartThread (&RocksDBQuickEnv::thread_entry, p); - } -}; - -//------------------------------------------------------------------------------ - -class RocksDBQuickBackend - : public Backend -{ -private: - std::atomic m_deletePath; - -public: - beast::Journal m_journal; - size_t const m_keyBytes; - std::string m_name; - std::unique_ptr m_db; - int fdlimit_ = 2048; - rocksdb::Options m_options; - - RocksDBQuickBackend (int keyBytes, Section const& keyValues, - Scheduler& scheduler, beast::Journal journal, RocksDBQuickEnv* env) - : m_deletePath (false) - , m_journal (journal) - , m_keyBytes (keyBytes) - , m_name (get(keyValues, "path")) - { - if (m_name.empty()) - Throw ( - "Missing path in RocksDBQuickFactory backend"); - - // Defaults - std::uint64_t budget = megabytes(512); - std::string style("level"); - std::uint64_t threads=4; - - get_if_exists (keyValues, "budget", budget); - get_if_exists (keyValues, "style", style); - get_if_exists (keyValues, "threads", threads); - - // Set options - m_options.create_if_missing = true; - m_options.env = env; - - if (style == "level") - m_options.OptimizeLevelStyleCompaction(budget); - - if (style == "universal") - m_options.OptimizeUniversalStyleCompaction(budget); - - if (style == "point") - m_options.OptimizeForPointLookup(budget / megabytes(1)); // In MB - - m_options.IncreaseParallelism(threads); - - // Allows hash indexes in blocks - m_options.prefix_extractor.reset(rocksdb::NewNoopTransform()); - - // overrride OptimizeLevelStyleCompaction - m_options.min_write_buffer_number_to_merge = 1; - - rocksdb::BlockBasedTableOptions table_options; - // Use hash index - table_options.index_type = - rocksdb::BlockBasedTableOptions::kHashSearch; - table_options.filter_policy.reset( - rocksdb::NewBloomFilterPolicy(10)); - m_options.table_factory.reset( - NewBlockBasedTableFactory(table_options)); - - // Higher values make reads slower - // table_options.block_size = 4096; - - // No point when DatabaseImp has a cache - // table_options.block_cache = - // rocksdb::NewLRUCache(64 * 1024 * 1024); - - m_options.memtable_factory.reset(rocksdb::NewHashSkipListRepFactory()); - // Alternative: - // m_options.memtable_factory.reset( - // rocksdb::NewHashCuckooRepFactory(m_options.write_buffer_size)); - - if (get_if_exists (keyValues, "open_files", m_options.max_open_files)) - fdlimit_ = m_options.max_open_files; - - if (keyValues.exists ("compression") && - (get(keyValues, "compression") == 0)) - m_options.compression = rocksdb::kNoCompression; - } - - ~RocksDBQuickBackend () override - { - close(); - } - - std::string - getName() override - { - return m_name; - } - - void - open(bool createIfMissing) override - { - if (m_db) - { - assert(false); - JLOG(m_journal.error()) << - "database is already open"; - return; - } - rocksdb::DB* db = nullptr; - rocksdb::Status status = rocksdb::DB::Open(m_options, m_name, &db); - if (!status.ok() || !db) - Throw( - std::string("Unable to open/create RocksDB: ") + - status.ToString()); - m_db.reset(db); - } - - void - close() override - { - if (m_db) - { - m_db.reset(); - if (m_deletePath) - { - boost::filesystem::path dir = m_name; - boost::filesystem::remove_all (dir); - } - } - } - - //-------------------------------------------------------------------------- - - Status - fetch (void const* key, std::shared_ptr* pObject) override - { - assert(m_db); - pObject->reset (); - - Status status (ok); - - rocksdb::ReadOptions const options; - rocksdb::Slice const slice (static_cast (key), m_keyBytes); - - std::string string; - - rocksdb::Status getStatus = m_db->Get (options, slice, &string); - - if (getStatus.ok ()) - { - DecodedBlob decoded (key, string.data (), string.size ()); - - if (decoded.wasOk ()) - { - *pObject = decoded.createObject (); - } - else - { - // Decoding failed, probably corrupted! - // - status = dataCorrupt; - } - } - else - { - if (getStatus.IsCorruption ()) - { - status = dataCorrupt; - } - else if (getStatus.IsNotFound ()) - { - status = notFound; - } - else - { - status = Status (customCode + getStatus.code()); - - JLOG(m_journal.error()) << getStatus.ToString (); - } - } - - return status; - } - - bool - canFetchBatch() override - { - return false; - } - - void - store (std::shared_ptr const& object) override - { - storeBatch(Batch{object}); - } - - std::vector> - fetchBatch (std::size_t n, void const* const* keys) override - { - Throw ("pure virtual called"); - return {}; - } - - void - storeBatch (Batch const& batch) override - { - assert(m_db); - rocksdb::WriteBatch wb; - - EncodedBlob encoded; - - for (auto const& e : batch) - { - encoded.prepare (e); - - wb.Put( - rocksdb::Slice(reinterpret_cast(encoded.getKey()), - m_keyBytes), - rocksdb::Slice(reinterpret_cast(encoded.getData()), - encoded.getSize())); - } - - rocksdb::WriteOptions options; - - // Crucial to ensure good write speed and non-blocking writes to memtable - options.disableWAL = true; - - auto ret = m_db->Write (options, &wb); - - if (! ret.ok ()) - Throw ("storeBatch failed: " + ret.ToString()); - } - - void - for_each (std::function )> f) override - { - assert(m_db); - rocksdb::ReadOptions const options; - - std::unique_ptr it (m_db->NewIterator (options)); - - for (it->SeekToFirst (); it->Valid (); it->Next ()) - { - if (it->key ().size () == m_keyBytes) - { - DecodedBlob decoded (it->key ().data (), - it->value ().data (), - it->value ().size ()); - - if (decoded.wasOk ()) - { - f (decoded.createObject ()); - } - else - { - // Uh oh, corrupted data! - JLOG(m_journal.fatal()) << - "Corrupt NodeObject #" << - from_hex_text(it->key ().data ()); - } - } - else - { - // VFALCO NOTE What does it mean to find an - // incorrectly sized key? Corruption? - JLOG(m_journal.fatal()) << - "Bad key size = " << it->key ().size (); - } - } - } - - int - getWriteLoad () override - { - return 0; - } - - void - setDeletePath() override - { - m_deletePath = true; - } - - //-------------------------------------------------------------------------- - - void - writeBatch (Batch const& batch) - { - storeBatch (batch); - } - - void - verify() override - { - } - - /** Returns the number of file handles the backend expects to need */ - int - fdlimit() const override - { - return fdlimit_; - } -}; - -//------------------------------------------------------------------------------ - -class RocksDBQuickFactory : public Factory -{ -public: - RocksDBQuickEnv m_env; - - RocksDBQuickFactory() - { - Manager::instance().insert(*this); - } - - ~RocksDBQuickFactory() override - { - Manager::instance().erase(*this); - } - - std::string - getName () const override - { - return "RocksDBQuick"; - } - - std::unique_ptr - createInstance ( - size_t keyBytes, - Section const& keyValues, - Scheduler& scheduler, - beast::Journal journal) override - { - return std::make_unique ( - keyBytes, keyValues, scheduler, journal, &m_env); - } -}; - -static RocksDBQuickFactory rocksDBQuickFactory; - -} -} - -#endif diff --git a/src/ripple/unity/nodestore.cpp b/src/ripple/unity/nodestore.cpp index cd9eee2779..731f04104f 100644 --- a/src/ripple/unity/nodestore.cpp +++ b/src/ripple/unity/nodestore.cpp @@ -22,7 +22,6 @@ #include #include #include -#include #include #include