Merge pull request #666
27f76e2
blockchain_import: Build string for db type list (warptangent)1aa8a9d
blockchain_import: Add mode argument representing multiple DB flags (warptangent)cffc411
blockchain_import: Support BerkeleyDB (warptangent)19c1aaa
blockchain_import: Add database type argument (warptangent)0fedce0
blockchain_import: Make LMDB-specific names general (warptangent)
This commit is contained in:
commit
3225a7d28d
2 changed files with 160 additions and 50 deletions
|
@ -38,6 +38,7 @@
|
|||
#include "serialization/binary_utils.h" // dump_binary(), parse_binary()
|
||||
#include "serialization/json_utils.h" // dump_json()
|
||||
#include "include_base_utils.h"
|
||||
#include "blockchain_db/db_types.h"
|
||||
|
||||
#include <lmdb.h> // for db flag arguments
|
||||
|
||||
|
@ -77,12 +78,48 @@ using namespace cryptonote;
|
|||
using namespace epee;
|
||||
|
||||
|
||||
int parse_db_arguments(const std::string& db_arg_str, std::string& db_engine, int& mdb_flags)
|
||||
std::string join_set_strings(const std::unordered_set<std::string>& db_types_all, const char* delim)
|
||||
{
|
||||
std::string result;
|
||||
std::ostringstream s;
|
||||
std::copy(db_types_all.begin(), db_types_all.end(), std::ostream_iterator<std::string>(s, delim));
|
||||
result = s.str();
|
||||
if (result.length() > 0)
|
||||
result.erase(result.end()-strlen(delim), result.end());
|
||||
return result;
|
||||
}
|
||||
|
||||
// db_type: lmdb, berkeley
|
||||
// db_mode: safe, fast, fastest
|
||||
int get_db_flags_from_mode(const std::string& db_type, const std::string& db_mode)
|
||||
{
|
||||
uint64_t BDB_FAST_MODE = 0;
|
||||
uint64_t BDB_FASTEST_MODE = 0;
|
||||
uint64_t BDB_SAFE_MODE = 0;
|
||||
|
||||
#if defined(BERKELEY_DB)
|
||||
BDB_FAST_MODE = DB_TXN_WRITE_NOSYNC;
|
||||
BDB_FASTEST_MODE = DB_TXN_NOSYNC;
|
||||
BDB_SAFE_MODE = DB_TXN_SYNC;
|
||||
#endif
|
||||
|
||||
int db_flags = 0;
|
||||
bool islmdb = db_type == "lmdb";
|
||||
if (db_mode == "safe")
|
||||
db_flags = islmdb ? MDB_NORDAHEAD : BDB_SAFE_MODE;
|
||||
else if (db_mode == "fast")
|
||||
db_flags = islmdb ? MDB_NOMETASYNC | MDB_NOSYNC | MDB_NORDAHEAD : BDB_FAST_MODE;
|
||||
else if (db_mode == "fastest")
|
||||
db_flags = islmdb ? MDB_WRITEMAP | MDB_MAPASYNC | MDB_NORDAHEAD | MDB_NOMETASYNC | MDB_NOSYNC : BDB_FASTEST_MODE;
|
||||
return db_flags;
|
||||
}
|
||||
|
||||
int parse_db_arguments(const std::string& db_arg_str, std::string& db_type, int& db_flags)
|
||||
{
|
||||
std::vector<std::string> db_args;
|
||||
boost::split(db_args, db_arg_str, boost::is_any_of("#"));
|
||||
db_engine = db_args.front();
|
||||
boost::algorithm::trim(db_engine);
|
||||
db_type = db_args.front();
|
||||
boost::algorithm::trim(db_type);
|
||||
|
||||
if (db_args.size() == 1)
|
||||
{
|
||||
|
@ -94,28 +131,73 @@ int parse_db_arguments(const std::string& db_arg_str, std::string& db_engine, in
|
|||
return 1;
|
||||
}
|
||||
|
||||
#if !defined(BERKELEY_DB)
|
||||
if (db_type == "berkeley")
|
||||
{
|
||||
LOG_ERROR("BerkeleyDB support disabled.");
|
||||
return false;
|
||||
}
|
||||
#endif
|
||||
|
||||
std::string db_arg_str2 = db_args[1];
|
||||
boost::split(db_args, db_arg_str2, boost::is_any_of(","));
|
||||
for (auto& it : db_args)
|
||||
|
||||
// optionally use a composite mode instead of individual flags
|
||||
const std::unordered_set<std::string> db_modes {"safe", "fast", "fastest"};
|
||||
std::string db_mode;
|
||||
if (db_args.size() == 1)
|
||||
{
|
||||
boost::algorithm::trim(it);
|
||||
if (it.empty())
|
||||
continue;
|
||||
LOG_PRINT_L1("LMDB flag: " << it);
|
||||
if (it == "nosync")
|
||||
mdb_flags |= MDB_NOSYNC;
|
||||
else if (it == "nometasync")
|
||||
mdb_flags |= MDB_NOMETASYNC;
|
||||
else if (it == "writemap")
|
||||
mdb_flags |= MDB_WRITEMAP;
|
||||
else if (it == "mapasync")
|
||||
mdb_flags |= MDB_MAPASYNC;
|
||||
else if (it == "nordahead")
|
||||
mdb_flags |= MDB_NORDAHEAD;
|
||||
else
|
||||
if (db_modes.count(db_args[0]) > 0)
|
||||
{
|
||||
std::cerr << "unrecognized database flag: " << it << ENDL;
|
||||
return 1;
|
||||
db_mode = db_args[0];
|
||||
}
|
||||
}
|
||||
if (! db_mode.empty())
|
||||
{
|
||||
db_flags = get_db_flags_from_mode(db_type, db_mode);
|
||||
}
|
||||
else
|
||||
{
|
||||
for (auto& it : db_args)
|
||||
{
|
||||
boost::algorithm::trim(it);
|
||||
if (it.empty())
|
||||
continue;
|
||||
if (db_type == "lmdb")
|
||||
{
|
||||
LOG_PRINT_L1("LMDB flag: " << it);
|
||||
if (it == "nosync")
|
||||
db_flags |= MDB_NOSYNC;
|
||||
else if (it == "nometasync")
|
||||
db_flags |= MDB_NOMETASYNC;
|
||||
else if (it == "writemap")
|
||||
db_flags |= MDB_WRITEMAP;
|
||||
else if (it == "mapasync")
|
||||
db_flags |= MDB_MAPASYNC;
|
||||
else if (it == "nordahead")
|
||||
db_flags |= MDB_NORDAHEAD;
|
||||
else
|
||||
{
|
||||
std::cerr << "unrecognized database flag: " << it << ENDL;
|
||||
return 1;
|
||||
}
|
||||
}
|
||||
#if defined(BERKELEY_DB)
|
||||
else if (db_type == "berkeley")
|
||||
{
|
||||
if (it == "txn_write_nosync")
|
||||
db_flags = DB_TXN_WRITE_NOSYNC;
|
||||
else if (it == "txn_nosync")
|
||||
db_flags = DB_TXN_NOSYNC;
|
||||
else if (it == "txn_sync")
|
||||
db_flags = DB_TXN_SYNC;
|
||||
else
|
||||
{
|
||||
std::cerr << "unrecognized database flag: " << it << ENDL;
|
||||
return 1;
|
||||
}
|
||||
}
|
||||
#endif
|
||||
}
|
||||
}
|
||||
return 0;
|
||||
|
@ -131,7 +213,7 @@ int pop_blocks(FakeCore& simple_core, int num_blocks)
|
|||
if (simple_core.support_batch)
|
||||
use_batch = true;
|
||||
else
|
||||
LOG_PRINT_L0("WARNING: batch transactions enabled but unsupported or unnecessary for this database engine - ignoring");
|
||||
LOG_PRINT_L0("WARNING: batch transactions enabled but unsupported or unnecessary for this database type - ignoring");
|
||||
}
|
||||
|
||||
if (use_batch)
|
||||
|
@ -176,11 +258,11 @@ template <typename FakeCore>
|
|||
int import_from_file(FakeCore& simple_core, const std::string& import_file_path, uint64_t block_stop=0)
|
||||
{
|
||||
#if !defined(BLOCKCHAIN_DB)
|
||||
static_assert(std::is_same<fake_core_memory, FakeCore>::value || std::is_same<fake_core_lmdb, FakeCore>::value,
|
||||
static_assert(std::is_same<fake_core_memory, FakeCore>::value || std::is_same<fake_core_db, FakeCore>::value,
|
||||
"FakeCore constraint error");
|
||||
#endif
|
||||
#if !defined(BLOCKCHAIN_DB) || (BLOCKCHAIN_DB == DB_LMDB)
|
||||
if (std::is_same<fake_core_lmdb, FakeCore>::value)
|
||||
if (std::is_same<fake_core_db, FakeCore>::value)
|
||||
{
|
||||
// Reset stats, in case we're using newly created db, accumulating stats
|
||||
// from addition of genesis block.
|
||||
|
@ -250,7 +332,7 @@ int import_from_file(FakeCore& simple_core, const std::string& import_file_path,
|
|||
if (simple_core.support_batch)
|
||||
use_batch = true;
|
||||
else
|
||||
LOG_PRINT_L0("WARNING: batch transactions enabled but unsupported or unnecessary for this database engine - ignoring");
|
||||
LOG_PRINT_L0("WARNING: batch transactions enabled but unsupported or unnecessary for this database type - ignoring");
|
||||
}
|
||||
|
||||
if (use_batch)
|
||||
|
@ -530,11 +612,19 @@ int import_from_file(FakeCore& simple_core, const std::string& import_file_path,
|
|||
int main(int argc, char* argv[])
|
||||
{
|
||||
#if defined(BLOCKCHAIN_DB) && (BLOCKCHAIN_DB == DB_MEMORY)
|
||||
std::string default_db_engine = "memory";
|
||||
std::string default_db_type = "memory";
|
||||
std::string default_db_engine_compiled = "memory";
|
||||
#else
|
||||
std::string default_db_engine = "lmdb";
|
||||
std::string default_db_type = "lmdb";
|
||||
std::string default_db_engine_compiled = "blockchain_db";
|
||||
#endif
|
||||
|
||||
std::unordered_set<std::string> db_types_all = cryptonote::blockchain_db_types;
|
||||
db_types_all.insert("memory");
|
||||
|
||||
std::string available_dbs = join_set_strings(db_types_all, ", ");
|
||||
available_dbs = "available: " + available_dbs;
|
||||
|
||||
uint32_t log_level = LOG_LEVEL_0;
|
||||
uint64_t num_blocks = 0;
|
||||
uint64_t block_stop = 0;
|
||||
|
@ -566,8 +656,7 @@ int main(int argc, char* argv[])
|
|||
, false
|
||||
};
|
||||
const command_line::arg_descriptor<std::string> arg_database = {
|
||||
"database", "available: memory, lmdb"
|
||||
, default_db_engine
|
||||
"database", available_dbs.c_str(), default_db_type
|
||||
};
|
||||
const command_line::arg_descriptor<bool> arg_verify = {"verify",
|
||||
"Verify blocks and transactions during import", true};
|
||||
|
@ -648,8 +737,6 @@ int main(int argc, char* argv[])
|
|||
}
|
||||
}
|
||||
|
||||
std::vector<std::string> db_engines {"memory", "lmdb"};
|
||||
|
||||
opt_testnet = command_line::get_arg(vm, arg_testnet_on);
|
||||
auto data_dir_arg = opt_testnet ? command_line::arg_testnet_data_dir : command_line::arg_data_dir;
|
||||
m_config_folder = command_line::get_arg(vm, data_dir_arg);
|
||||
|
@ -677,23 +764,34 @@ int main(int argc, char* argv[])
|
|||
}
|
||||
|
||||
|
||||
std::string db_engine;
|
||||
int mdb_flags = 0;
|
||||
std::string db_type;
|
||||
std::string db_engine_compiled;
|
||||
int db_flags = 0;
|
||||
int res = 0;
|
||||
res = parse_db_arguments(db_arg_str, db_engine, mdb_flags);
|
||||
res = parse_db_arguments(db_arg_str, db_type, db_flags);
|
||||
if (res)
|
||||
{
|
||||
std::cerr << "Error parsing database argument(s)" << ENDL;
|
||||
return 1;
|
||||
}
|
||||
|
||||
if (std::find(db_engines.begin(), db_engines.end(), db_engine) == db_engines.end())
|
||||
if (db_types_all.count(db_type) == 0)
|
||||
{
|
||||
std::cerr << "Invalid database engine: " << db_engine << std::endl;
|
||||
std::cerr << "Invalid database type: " << db_type << std::endl;
|
||||
return 1;
|
||||
}
|
||||
|
||||
LOG_PRINT_L0("database: " << db_engine);
|
||||
if ((db_type == "lmdb") || (db_type == "berkeley"))
|
||||
{
|
||||
db_engine_compiled = "blockchain_db";
|
||||
}
|
||||
else
|
||||
{
|
||||
db_engine_compiled = "memory";
|
||||
}
|
||||
|
||||
LOG_PRINT_L0("database: " << db_type);
|
||||
LOG_PRINT_L0("database flags: " << db_flags);
|
||||
LOG_PRINT_L0("verify: " << std::boolalpha << opt_verify << std::noboolalpha);
|
||||
if (opt_batch)
|
||||
{
|
||||
|
@ -722,31 +820,31 @@ int main(int argc, char* argv[])
|
|||
|
||||
// for multi_db_runtime:
|
||||
#if !defined(BLOCKCHAIN_DB)
|
||||
if (db_engine == "lmdb")
|
||||
if (db_type == "lmdb" || db_type == "berkeley")
|
||||
{
|
||||
fake_core_lmdb simple_core(m_config_folder, opt_testnet, opt_batch, mdb_flags);
|
||||
fake_core_db simple_core(m_config_folder, opt_testnet, opt_batch, db_type, db_flags);
|
||||
import_from_file(simple_core, import_file_path, block_stop);
|
||||
}
|
||||
else if (db_engine == "memory")
|
||||
else if (db_type == "memory")
|
||||
{
|
||||
fake_core_memory simple_core(m_config_folder, opt_testnet);
|
||||
import_from_file(simple_core, import_file_path, block_stop);
|
||||
}
|
||||
else
|
||||
{
|
||||
std::cerr << "database engine unrecognized" << ENDL;
|
||||
std::cerr << "database type unrecognized" << ENDL;
|
||||
return 1;
|
||||
}
|
||||
|
||||
// for multi_db_compile:
|
||||
#else
|
||||
if (db_engine != default_db_engine)
|
||||
if (db_engine_compiled != default_db_engine_compiled)
|
||||
{
|
||||
std::cerr << "Invalid database engine for compiled version: " << db_engine << std::endl;
|
||||
std::cerr << "Invalid database type for compiled version: " << db_type << std::endl;
|
||||
return 1;
|
||||
}
|
||||
#if BLOCKCHAIN_DB == DB_LMDB
|
||||
fake_core_lmdb simple_core(m_config_folder, opt_testnet, opt_batch, mdb_flags);
|
||||
fake_core_db simple_core(m_config_folder, opt_testnet, opt_batch, db_type, db_flags);
|
||||
#else
|
||||
fake_core_memory simple_core(m_config_folder, opt_testnet);
|
||||
#endif
|
||||
|
|
|
@ -33,6 +33,7 @@
|
|||
#include "cryptonote_core/blockchain_storage.h" // in-memory DB
|
||||
#include "blockchain_db/blockchain_db.h"
|
||||
#include "blockchain_db/lmdb/db_lmdb.h"
|
||||
#include "blockchain_db/berkeleydb/db_bdb.h"
|
||||
|
||||
using namespace cryptonote;
|
||||
|
||||
|
@ -47,7 +48,7 @@ namespace
|
|||
|
||||
#if !defined(BLOCKCHAIN_DB) || BLOCKCHAIN_DB == DB_LMDB
|
||||
|
||||
struct fake_core_lmdb
|
||||
struct fake_core_db
|
||||
{
|
||||
Blockchain m_storage;
|
||||
HardFork* m_hardfork = nullptr;
|
||||
|
@ -57,15 +58,26 @@ struct fake_core_lmdb
|
|||
|
||||
// for multi_db_runtime:
|
||||
#if !defined(BLOCKCHAIN_DB)
|
||||
fake_core_lmdb(const boost::filesystem::path &path, const bool use_testnet=false, const bool do_batch=true, const int mdb_flags=0) : m_pool(&m_storage), m_storage(m_pool)
|
||||
fake_core_db(const boost::filesystem::path &path, const bool use_testnet=false, const bool do_batch=true, const std::string& db_type="lmdb", const int db_flags=0) : m_pool(&m_storage), m_storage(m_pool)
|
||||
// for multi_db_compile:
|
||||
#else
|
||||
fake_core_lmdb(const boost::filesystem::path &path, const bool use_testnet=false, const bool do_batch=true, const int mdb_flags=0) : m_pool(m_storage), m_storage(m_pool)
|
||||
fake_core_db(const boost::filesystem::path &path, const bool use_testnet=false, const bool do_batch=true, const std::string& db_type="lmdb", const int db_flags=0) : m_pool(m_storage), m_storage(m_pool)
|
||||
#endif
|
||||
{
|
||||
m_pool.init(path.string());
|
||||
|
||||
BlockchainDB* db = new BlockchainLMDB();
|
||||
BlockchainDB* db = nullptr;
|
||||
if (db_type == "lmdb")
|
||||
db = new BlockchainLMDB();
|
||||
#if defined(BERKELEY_DB)
|
||||
else if (db_type == "berkeley")
|
||||
db = new BlockchainBDB();
|
||||
#endif
|
||||
else
|
||||
{
|
||||
LOG_ERROR("Attempted to use non-existent database type: " << db_type);
|
||||
throw std::runtime_error("Attempting to use non-existent database type");
|
||||
}
|
||||
|
||||
boost::filesystem::path folder(path);
|
||||
|
||||
|
@ -76,7 +88,7 @@ struct fake_core_lmdb
|
|||
const std::string filename = folder.string();
|
||||
try
|
||||
{
|
||||
db->open(filename, mdb_flags);
|
||||
db->open(filename, db_flags);
|
||||
}
|
||||
catch (const std::exception& e)
|
||||
{
|
||||
|
@ -96,7 +108,7 @@ struct fake_core_lmdb
|
|||
support_batch = true;
|
||||
support_add_block = true;
|
||||
}
|
||||
~fake_core_lmdb()
|
||||
~fake_core_db()
|
||||
{
|
||||
m_storage.get_db().check_hard_fork_info();
|
||||
m_storage.deinit();
|
||||
|
|
Loading…
Reference in a new issue