| 1 | #include <Databases/DatabaseOnDisk.h> |
| 2 | |
| 3 | #include <IO/ReadBufferFromFile.h> |
| 4 | #include <IO/ReadHelpers.h> |
| 5 | #include <IO/WriteBufferFromFile.h> |
| 6 | #include <IO/WriteHelpers.h> |
| 7 | #include <Interpreters/Context.h> |
| 8 | #include <Interpreters/InterpreterCreateQuery.h> |
| 9 | #include <Parsers/ASTCreateQuery.h> |
| 10 | #include <Parsers/ParserCreateQuery.h> |
| 11 | #include <Parsers/formatAST.h> |
| 12 | #include <Parsers/parseQuery.h> |
| 13 | #include <Storages/IStorage.h> |
| 14 | #include <Storages/StorageFactory.h> |
| 15 | #include <TableFunctions/TableFunctionFactory.h> |
| 16 | #include <Common/escapeForFileName.h> |
| 17 | |
| 18 | #include <common/logger_useful.h> |
| 19 | #include <Poco/DirectoryIterator.h> |
| 20 | |
| 21 | |
| 22 | |
| 23 | namespace DB |
| 24 | { |
| 25 | |
| 26 | static constexpr size_t METADATA_FILE_BUFFER_SIZE = 32768; |
| 27 | |
| 28 | namespace ErrorCodes |
| 29 | { |
| 30 | extern const int FILE_DOESNT_EXIST; |
| 31 | extern const int INCORRECT_FILE_NAME; |
| 32 | extern const int SYNTAX_ERROR; |
| 33 | extern const int TABLE_ALREADY_EXISTS; |
| 34 | extern const int UNKNOWN_TABLE; |
| 35 | extern const int DICTIONARY_ALREADY_EXISTS; |
| 36 | extern const int EMPTY_LIST_OF_COLUMNS_PASSED; |
| 37 | } |
| 38 | |
| 39 | |
| 40 | std::pair<String, StoragePtr> createTableFromAST( |
| 41 | ASTCreateQuery ast_create_query, |
| 42 | const String & database_name, |
| 43 | const String & table_data_path_relative, |
| 44 | Context & context, |
| 45 | bool has_force_restore_data_flag) |
| 46 | { |
| 47 | ast_create_query.attach = true; |
| 48 | ast_create_query.database = database_name; |
| 49 | |
| 50 | if (ast_create_query.as_table_function) |
| 51 | { |
| 52 | const auto & table_function = ast_create_query.as_table_function->as<ASTFunction &>(); |
| 53 | const auto & factory = TableFunctionFactory::instance(); |
| 54 | StoragePtr storage = factory.get(table_function.name, context)->execute(ast_create_query.as_table_function, context, ast_create_query.table); |
| 55 | return {ast_create_query.table, storage}; |
| 56 | } |
| 57 | /// We do not directly use `InterpreterCreateQuery::execute`, because |
| 58 | /// - the database has not been loaded yet; |
| 59 | /// - the code is simpler, since the query is already brought to a suitable form. |
| 60 | if (!ast_create_query.columns_list || !ast_create_query.columns_list->columns) |
| 61 | throw Exception("Missing definition of columns." , ErrorCodes::EMPTY_LIST_OF_COLUMNS_PASSED); |
| 62 | |
| 63 | ColumnsDescription columns = InterpreterCreateQuery::getColumnsDescription(*ast_create_query.columns_list->columns, context); |
| 64 | ConstraintsDescription constraints = InterpreterCreateQuery::getConstraintsDescription(ast_create_query.columns_list->constraints); |
| 65 | |
| 66 | return |
| 67 | { |
| 68 | ast_create_query.table, |
| 69 | StorageFactory::instance().get( |
| 70 | ast_create_query, |
| 71 | table_data_path_relative, ast_create_query.table, database_name, context, context.getGlobalContext(), |
| 72 | columns, constraints, |
| 73 | true, has_force_restore_data_flag) |
| 74 | }; |
| 75 | } |
| 76 | |
| 77 | |
| 78 | String getObjectDefinitionFromCreateQuery(const ASTPtr & query) |
| 79 | { |
| 80 | ASTPtr query_clone = query->clone(); |
| 81 | auto * create = query_clone->as<ASTCreateQuery>(); |
| 82 | |
| 83 | if (!create) |
| 84 | { |
| 85 | std::ostringstream query_stream; |
| 86 | formatAST(*query, query_stream, true); |
| 87 | throw Exception("Query '" + query_stream.str() + "' is not CREATE query" , ErrorCodes::LOGICAL_ERROR); |
| 88 | } |
| 89 | |
| 90 | if (!create->is_dictionary) |
| 91 | create->attach = true; |
| 92 | |
| 93 | /// We remove everything that is not needed for ATTACH from the query. |
| 94 | create->database.clear(); |
| 95 | create->as_database.clear(); |
| 96 | create->as_table.clear(); |
| 97 | create->if_not_exists = false; |
| 98 | create->is_populate = false; |
| 99 | create->replace_view = false; |
| 100 | |
| 101 | /// For views it is necessary to save the SELECT query itself, for the rest - on the contrary |
| 102 | if (!create->is_view && !create->is_materialized_view && !create->is_live_view) |
| 103 | create->select = nullptr; |
| 104 | |
| 105 | create->format = nullptr; |
| 106 | create->out_file = nullptr; |
| 107 | |
| 108 | std::ostringstream statement_stream; |
| 109 | formatAST(*create, statement_stream, false); |
| 110 | statement_stream << '\n'; |
| 111 | return statement_stream.str(); |
| 112 | } |
| 113 | |
| 114 | void DatabaseOnDisk::createTable( |
| 115 | const Context & context, |
| 116 | const String & table_name, |
| 117 | const StoragePtr & table, |
| 118 | const ASTPtr & query) |
| 119 | { |
| 120 | const auto & settings = context.getSettingsRef(); |
| 121 | |
| 122 | /// Create a file with metadata if necessary - if the query is not ATTACH. |
| 123 | /// Write the query of `ATTACH table` to it. |
| 124 | |
| 125 | /** The code is based on the assumption that all threads share the same order of operations |
| 126 | * - creating the .sql.tmp file; |
| 127 | * - adding a table to `tables`; |
| 128 | * - rename .sql.tmp to .sql. |
| 129 | */ |
| 130 | |
| 131 | /// A race condition would be possible if a table with the same name is simultaneously created using CREATE and using ATTACH. |
| 132 | /// But there is protection from it - see using DDLGuard in InterpreterCreateQuery. |
| 133 | |
| 134 | if (isDictionaryExist(context, table_name)) |
| 135 | throw Exception("Dictionary " + backQuote(getDatabaseName()) + "." + backQuote(table_name) + " already exists." , |
| 136 | ErrorCodes::DICTIONARY_ALREADY_EXISTS); |
| 137 | |
| 138 | if (isTableExist(context, table_name)) |
| 139 | throw Exception("Table " + backQuote(getDatabaseName()) + "." + backQuote(table_name) + " already exists." , ErrorCodes::TABLE_ALREADY_EXISTS); |
| 140 | |
| 141 | String table_metadata_path = getObjectMetadataPath(table_name); |
| 142 | String table_metadata_tmp_path = table_metadata_path + ".tmp" ; |
| 143 | String statement; |
| 144 | |
| 145 | { |
| 146 | statement = getObjectDefinitionFromCreateQuery(query); |
| 147 | |
| 148 | /// Exclusive flags guarantees, that table is not created right now in another thread. Otherwise, exception will be thrown. |
| 149 | WriteBufferFromFile out(table_metadata_tmp_path, statement.size(), O_WRONLY | O_CREAT | O_EXCL); |
| 150 | writeString(statement, out); |
| 151 | out.next(); |
| 152 | if (settings.fsync_metadata) |
| 153 | out.sync(); |
| 154 | out.close(); |
| 155 | } |
| 156 | |
| 157 | try |
| 158 | { |
| 159 | /// Add a table to the map of known tables. |
| 160 | attachTable(table_name, table); |
| 161 | |
| 162 | /// If it was ATTACH query and file with table metadata already exist |
| 163 | /// (so, ATTACH is done after DETACH), then rename atomically replaces old file with new one. |
| 164 | Poco::File(table_metadata_tmp_path).renameTo(table_metadata_path); |
| 165 | } |
| 166 | catch (...) |
| 167 | { |
| 168 | Poco::File(table_metadata_tmp_path).remove(); |
| 169 | throw; |
| 170 | } |
| 171 | } |
| 172 | |
| 173 | void DatabaseOnDisk::removeTable(const Context & /* context */, const String & table_name) |
| 174 | { |
| 175 | StoragePtr res = detachTable(table_name); |
| 176 | |
| 177 | String table_metadata_path = getObjectMetadataPath(table_name); |
| 178 | |
| 179 | try |
| 180 | { |
| 181 | Poco::File(table_metadata_path).remove(); |
| 182 | } |
| 183 | catch (...) |
| 184 | { |
| 185 | try |
| 186 | { |
| 187 | Poco::File(table_metadata_path + ".tmp_drop" ).remove(); |
| 188 | return; |
| 189 | } |
| 190 | catch (...) |
| 191 | { |
| 192 | LOG_WARNING(log, getCurrentExceptionMessage(__PRETTY_FUNCTION__)); |
| 193 | } |
| 194 | attachTable(table_name, res); |
| 195 | throw; |
| 196 | } |
| 197 | } |
| 198 | |
| 199 | void DatabaseOnDisk::renameTable( |
| 200 | const Context & context, |
| 201 | const String & table_name, |
| 202 | IDatabase & to_database, |
| 203 | const String & to_table_name, |
| 204 | TableStructureWriteLockHolder & lock) |
| 205 | { |
| 206 | if (typeid(*this) != typeid(to_database)) |
| 207 | throw Exception("Moving tables between databases of different engines is not supported" , ErrorCodes::NOT_IMPLEMENTED); |
| 208 | |
| 209 | StoragePtr table = tryGetTable(context, table_name); |
| 210 | |
| 211 | if (!table) |
| 212 | throw Exception("Table " + backQuote(getDatabaseName()) + "." + backQuote(table_name) + " doesn't exist." , ErrorCodes::UNKNOWN_TABLE); |
| 213 | |
| 214 | ASTPtr ast = parseQueryFromMetadata(getObjectMetadataPath(table_name)); |
| 215 | if (!ast) |
| 216 | throw Exception("There is no metadata file for table " + backQuote(table_name) + "." , ErrorCodes::FILE_DOESNT_EXIST); |
| 217 | auto & create = ast->as<ASTCreateQuery &>(); |
| 218 | create.table = to_table_name; |
| 219 | |
| 220 | /// Notify the table that it is renamed. If the table does not support renaming, exception is thrown. |
| 221 | try |
| 222 | { |
| 223 | table->rename(to_database.getTableDataPath(create), |
| 224 | to_database.getDatabaseName(), |
| 225 | to_table_name, lock); |
| 226 | } |
| 227 | catch (const Exception &) |
| 228 | { |
| 229 | throw; |
| 230 | } |
| 231 | catch (const Poco::Exception & e) |
| 232 | { |
| 233 | /// Better diagnostics. |
| 234 | throw Exception{Exception::CreateFromPoco, e}; |
| 235 | } |
| 236 | |
| 237 | /// NOTE Non-atomic. |
| 238 | to_database.createTable(context, to_table_name, table, ast); |
| 239 | removeTable(context, table_name); |
| 240 | } |
| 241 | |
| 242 | ASTPtr DatabaseOnDisk::getCreateTableQueryImpl(const Context & context, const String & table_name, bool throw_on_error) const |
| 243 | { |
| 244 | ASTPtr ast; |
| 245 | |
| 246 | auto table_metadata_path = getObjectMetadataPath(table_name); |
| 247 | ast = getCreateQueryFromMetadata(table_metadata_path, throw_on_error); |
| 248 | if (!ast && throw_on_error) |
| 249 | { |
| 250 | /// Handle system.* tables for which there are no table.sql files. |
| 251 | bool has_table = tryGetTable(context, table_name) != nullptr; |
| 252 | |
| 253 | auto msg = has_table |
| 254 | ? "There is no CREATE TABLE query for table " |
| 255 | : "There is no metadata file for table " ; |
| 256 | |
| 257 | throw Exception(msg + backQuote(table_name), ErrorCodes::CANNOT_GET_CREATE_TABLE_QUERY); |
| 258 | } |
| 259 | |
| 260 | return ast; |
| 261 | } |
| 262 | |
| 263 | ASTPtr DatabaseOnDisk::getCreateDatabaseQuery() const |
| 264 | { |
| 265 | ASTPtr ast; |
| 266 | |
| 267 | auto metadata_dir_path = getMetadataPath(); |
| 268 | auto database_metadata_path = metadata_dir_path.substr(0, metadata_dir_path.size() - 1) + ".sql" ; |
| 269 | ast = getCreateQueryFromMetadata(database_metadata_path, true); |
| 270 | if (!ast) |
| 271 | { |
| 272 | /// Handle databases (such as default) for which there are no database.sql files. |
| 273 | /// If database.sql doesn't exist, then engine is Ordinary |
| 274 | String query = "CREATE DATABASE " + backQuoteIfNeed(getDatabaseName()) + " ENGINE = Ordinary" ; |
| 275 | ParserCreateQuery parser; |
| 276 | ast = parseQuery(parser, query.data(), query.data() + query.size(), "" , 0); |
| 277 | } |
| 278 | |
| 279 | return ast; |
| 280 | } |
| 281 | |
| 282 | void DatabaseOnDisk::drop(const Context & context) |
| 283 | { |
| 284 | Poco::File(context.getPath() + getDataPath()).remove(false); |
| 285 | Poco::File(getMetadataPath()).remove(false); |
| 286 | } |
| 287 | |
| 288 | String DatabaseOnDisk::getObjectMetadataPath(const String & table_name) const |
| 289 | { |
| 290 | return getMetadataPath() + escapeForFileName(table_name) + ".sql" ; |
| 291 | } |
| 292 | |
| 293 | time_t DatabaseOnDisk::getObjectMetadataModificationTime(const String & table_name) const |
| 294 | { |
| 295 | String table_metadata_path = getObjectMetadataPath(table_name); |
| 296 | Poco::File meta_file(table_metadata_path); |
| 297 | |
| 298 | if (meta_file.exists()) |
| 299 | return meta_file.getLastModified().epochTime(); |
| 300 | else |
| 301 | return static_cast<time_t>(0); |
| 302 | } |
| 303 | |
| 304 | void DatabaseOnDisk::iterateMetadataFiles(const Context & context, const IteratingFunction & iterating_function) const |
| 305 | { |
| 306 | Poco::DirectoryIterator dir_end; |
| 307 | for (Poco::DirectoryIterator dir_it(getMetadataPath()); dir_it != dir_end; ++dir_it) |
| 308 | { |
| 309 | /// For '.svn', '.gitignore' directory and similar. |
| 310 | if (dir_it.name().at(0) == '.') |
| 311 | continue; |
| 312 | |
| 313 | /// There are .sql.bak files - skip them. |
| 314 | if (endsWith(dir_it.name(), ".sql.bak" )) |
| 315 | continue; |
| 316 | |
| 317 | // There are files that we tried to delete previously |
| 318 | static const char * tmp_drop_ext = ".sql.tmp_drop" ; |
| 319 | if (endsWith(dir_it.name(), tmp_drop_ext)) |
| 320 | { |
| 321 | const std::string object_name = dir_it.name().substr(0, dir_it.name().size() - strlen(tmp_drop_ext)); |
| 322 | if (Poco::File(context.getPath() + getDataPath() + '/' + object_name).exists()) |
| 323 | { |
| 324 | /// TODO maybe complete table drop and remove all table data (including data on other volumes and metadata in ZK) |
| 325 | Poco::File(dir_it->path()).renameTo(getMetadataPath() + object_name + ".sql" ); |
| 326 | LOG_WARNING(log, "Object " << backQuote(object_name) << " was not dropped previously and will be restored" ); |
| 327 | iterating_function(object_name + ".sql" ); |
| 328 | } |
| 329 | else |
| 330 | { |
| 331 | LOG_INFO(log, "Removing file " << dir_it->path()); |
| 332 | Poco::File(dir_it->path()).remove(); |
| 333 | } |
| 334 | continue; |
| 335 | } |
| 336 | |
| 337 | /// There are files .sql.tmp - delete |
| 338 | if (endsWith(dir_it.name(), ".sql.tmp" )) |
| 339 | { |
| 340 | LOG_INFO(log, "Removing file " << dir_it->path()); |
| 341 | Poco::File(dir_it->path()).remove(); |
| 342 | continue; |
| 343 | } |
| 344 | |
| 345 | /// The required files have names like `table_name.sql` |
| 346 | if (endsWith(dir_it.name(), ".sql" )) |
| 347 | { |
| 348 | iterating_function(dir_it.name()); |
| 349 | } |
| 350 | else |
| 351 | throw Exception("Incorrect file extension: " + dir_it.name() + " in metadata directory " + getMetadataPath(), |
| 352 | ErrorCodes::INCORRECT_FILE_NAME); |
| 353 | } |
| 354 | } |
| 355 | |
| 356 | ASTPtr DatabaseOnDisk::parseQueryFromMetadata(const String & metadata_file_path, bool throw_on_error /*= true*/, bool remove_empty /*= false*/) const |
| 357 | { |
| 358 | String query; |
| 359 | |
| 360 | try |
| 361 | { |
| 362 | ReadBufferFromFile in(metadata_file_path, METADATA_FILE_BUFFER_SIZE); |
| 363 | readStringUntilEOF(query, in); |
| 364 | } |
| 365 | catch (const Exception & e) |
| 366 | { |
| 367 | if (!throw_on_error && e.code() == ErrorCodes::FILE_DOESNT_EXIST) |
| 368 | return nullptr; |
| 369 | else |
| 370 | throw; |
| 371 | } |
| 372 | |
| 373 | /** Empty files with metadata are generated after a rough restart of the server. |
| 374 | * Remove these files to slightly reduce the work of the admins on startup. |
| 375 | */ |
| 376 | if (remove_empty && query.empty()) |
| 377 | { |
| 378 | LOG_ERROR(log, "File " << metadata_file_path << " is empty. Removing." ); |
| 379 | Poco::File(metadata_file_path).remove(); |
| 380 | return nullptr; |
| 381 | } |
| 382 | |
| 383 | ParserCreateQuery parser; |
| 384 | const char * pos = query.data(); |
| 385 | std::string error_message; |
| 386 | auto ast = tryParseQuery(parser, pos, pos + query.size(), error_message, /* hilite = */ false, |
| 387 | "in file " + getMetadataPath(), /* allow_multi_statements = */ false, 0); |
| 388 | |
| 389 | if (!ast && throw_on_error) |
| 390 | throw Exception(error_message, ErrorCodes::SYNTAX_ERROR); |
| 391 | else if (!ast) |
| 392 | return nullptr; |
| 393 | |
| 394 | return ast; |
| 395 | } |
| 396 | |
| 397 | ASTPtr DatabaseOnDisk::getCreateQueryFromMetadata(const String & database_metadata_path, bool throw_on_error) const |
| 398 | { |
| 399 | ASTPtr ast = parseQueryFromMetadata(database_metadata_path, throw_on_error); |
| 400 | |
| 401 | if (ast) |
| 402 | { |
| 403 | auto & ast_create_query = ast->as<ASTCreateQuery &>(); |
| 404 | ast_create_query.attach = false; |
| 405 | ast_create_query.database = database_name; |
| 406 | } |
| 407 | |
| 408 | return ast; |
| 409 | } |
| 410 | |
| 411 | } |
| 412 | |