1 #ifndef PARTITION_HANDLER_INCLUDED 2 #define PARTITION_HANDLER_INCLUDED 3 4 /* 5 Copyright (c) 2005, 2021, Oracle and/or its affiliates. 6 7 This program is free software; you can redistribute it and/or modify 8 it under the terms of the GNU General Public License, version 2.0, 9 as published by the Free Software Foundation. 10 11 This program is also distributed with certain software (including 12 but not limited to OpenSSL) that is licensed under separate terms, 13 as designated in a particular file or component or in included license 14 documentation. The authors of MySQL hereby grant you an additional 15 permission to link the program and your derivative works with the 16 separately licensed software that they have included with MySQL. 17 18 This program is distributed in the hope that it will be useful, 19 but WITHOUT ANY WARRANTY; without even the implied warranty of 20 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 21 GNU General Public License, version 2.0, for more details. 22 23 You should have received a copy of the GNU General Public License 24 along with this program; if not, write to the Free Software 25 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 26 */ 27 28 #include "my_global.h" // uint etc. 29 #include "my_base.h" // ha_rows. 30 #include "handler.h" // Handler_share 31 #include "sql_partition.h" // part_id_range 32 #include "mysqld_error.h" // ER_ILLEGAL_HA 33 #include "priority_queue.h" 34 #include "key.h" // key_rec_cmp 35 #include <vector> 36 37 #define PARTITION_BYTES_IN_POS 2 38 39 /* forward declarations */ 40 typedef struct st_ha_create_information HA_CREATE_INFO; 41 typedef struct st_mem_root MEM_ROOT; 42 43 static const uint NO_CURRENT_PART_ID= UINT_MAX32; 44 45 /** 46 bits in Partition_handler::alter_flags(): 47 48 HA_PARTITION_FUNCTION_SUPPORTED indicates that the function is 49 supported at all. 50 HA_FAST_CHANGE_PARTITION means that optimized variants of the changes 51 exists but they are not necessarily done online. 52 53 HA_ONLINE_DOUBLE_WRITE means that the handler supports writing to both 54 the new partition and to the old partitions when updating through the 55 old partitioning schema while performing a change of the partitioning. 56 This means that we can support updating of the table while performing 57 the copy phase of the change. For no lock at all also a double write 58 from new to old must exist and this is not required when this flag is 59 set. 60 This is actually removed even before it was introduced the first time. 61 The new idea is that handlers will handle the lock level already in 62 store_lock for ALTER TABLE partitions. 63 TODO: Implement this via the alter-inplace api. 64 */ 65 #define HA_PARTITION_FUNCTION_SUPPORTED (1L << 0) 66 #define HA_FAST_CHANGE_PARTITION (1L << 1) 67 68 enum enum_part_operation { 69 OPTIMIZE_PARTS= 0, 70 ANALYZE_PARTS, 71 CHECK_PARTS, 72 REPAIR_PARTS, 73 ASSIGN_KEYCACHE_PARTS, 74 PRELOAD_KEYS_PARTS 75 }; 76 77 /** Struct used for partition_name_hash */ 78 typedef struct st_part_name_def 79 { 80 uchar *partition_name; 81 uint length; 82 uint32 part_id; 83 my_bool is_subpart; 84 } PART_NAME_DEF; 85 86 87 /** 88 Initialize partitioning (currently only PSI keys). 89 */ 90 void partitioning_init(); 91 92 93 /** 94 Partition specific Handler_share. 95 */ 96 class Partition_share : public Handler_share 97 { 98 public: 99 Partition_share(); 100 ~Partition_share(); 101 102 /** Set if auto increment is used an initialized. */ 103 bool auto_inc_initialized; 104 /** 105 Mutex protecting next_auto_inc_val. 106 Initialized if table uses auto increment. 107 */ 108 mysql_mutex_t *auto_inc_mutex; 109 /** First non reserved auto increment value. */ 110 ulonglong next_auto_inc_val; 111 /** 112 Hash of partition names. Initialized by the first handler instance of a 113 table_share calling populate_partition_name_hash(). 114 After that it is read-only, i.e. no locking required for reading. 115 */ 116 HASH partition_name_hash; 117 /** flag that the name hash is initialized, so it only will do it once. */ 118 bool partition_name_hash_initialized; 119 120 /** 121 Initializes and sets auto_inc_mutex. 122 Only needed to be called if the table have an auto increment. 123 Must hold TABLE_SHARE::LOCK_ha_data when calling. 124 */ 125 bool init_auto_inc_mutex(TABLE_SHARE *table_share); 126 /** 127 Release reserved auto increment values not used. 128 @param thd Thread. 129 @param table_share Table Share 130 @param next_insert_id Next insert id (first non used auto inc value). 131 @param max_reserved End of reserved auto inc range. 132 */ 133 void release_auto_inc_if_possible(THD *thd, TABLE_SHARE *table_share, 134 const ulonglong next_insert_id, 135 const ulonglong max_reserved); 136 137 /** lock mutex protecting auto increment value next_auto_inc_val. */ lock_auto_inc()138 inline void lock_auto_inc() 139 { 140 assert(auto_inc_mutex); 141 mysql_mutex_lock(auto_inc_mutex); 142 } 143 /** unlock mutex protecting auto increment value next_auto_inc_val. */ unlock_auto_inc()144 inline void unlock_auto_inc() 145 { 146 assert(auto_inc_mutex); 147 mysql_mutex_unlock(auto_inc_mutex); 148 } 149 /** 150 Populate partition_name_hash with partition and subpartition names 151 from part_info. 152 @param part_info Partition info containing all partitions metadata. 153 154 @return Operation status. 155 @retval false Success. 156 @retval true Failure. 157 */ 158 bool populate_partition_name_hash(partition_info *part_info); 159 /** Get partition name. 160 161 @param part_id Partition id (for subpartitioned table only subpartition 162 names will be returned.) 163 164 @return partition name or NULL if error. 165 */ 166 const char *get_partition_name(size_t part_id) const; 167 private: 168 const uchar **partition_names; 169 /** 170 Insert [sub]partition name into partition_name_hash 171 @param name Partition name. 172 @param part_id Partition id. 173 @param is_subpart True if subpartition else partition. 174 175 @return Operation status. 176 @retval false Success. 177 @retval true Failure. 178 */ 179 bool insert_partition_name_in_hash(const char *name, 180 uint part_id, 181 bool is_subpart); 182 }; 183 184 185 /** 186 Class for partitioning specific operations. 187 188 Returned from handler::get_partition_handler(). 189 */ 190 class Partition_handler :public Sql_alloc 191 { 192 public: Partition_handler()193 Partition_handler() {} ~Partition_handler()194 ~Partition_handler() {} 195 196 /** 197 Get dynamic table information from partition. 198 199 @param[out] stat_info Statistics struct to fill in. 200 @param[out] check_sum Check sum value to fill in if supported. 201 @param[in] part_id Partition to report for. 202 203 @note stat_info and check_sum are initialized by caller. 204 check_sum is only expected to be updated if HA_HAS_CHECKSUM. 205 */ 206 virtual void get_dynamic_partition_info(ha_statistics *stat_info, 207 ha_checksum *check_sum, 208 uint part_id) = 0; 209 /** 210 Get default number of partitions. 211 212 Used during creating a partitioned table. 213 214 @param info Create info. 215 @return Number of default partitions. 216 */ get_default_num_partitions(HA_CREATE_INFO * info)217 virtual int get_default_num_partitions(HA_CREATE_INFO *info) { return 1;} 218 /** 219 Setup auto partitioning. 220 221 Called for engines with HA_USE_AUTO_PARTITION to setup the partition info 222 object 223 224 @param[in,out] part_info Partition object to setup. 225 */ set_auto_partitions(partition_info * part_info)226 virtual void set_auto_partitions(partition_info *part_info) { return; } 227 /** 228 Get number of partitions for table in SE 229 230 @param name normalized path(same as open) to the table 231 232 @param[out] num_parts Number of partitions 233 234 @retval false for success 235 @retval true for failure, for example table didn't exist in engine 236 */ get_num_parts(const char * name,uint * num_parts)237 virtual bool get_num_parts(const char *name, 238 uint *num_parts) 239 { 240 *num_parts= 0; 241 return false; 242 } 243 /** 244 Set the partition info object to be used by the handler. 245 246 @param part_info Partition info to be used by the handler. 247 @param early True if called when part_info only created and parsed, 248 but not setup, checked or fixed. 249 */ 250 virtual void set_part_info(partition_info *part_info, bool early) = 0; 251 /** 252 Initialize partition. 253 254 @param mem_root Memory root for memory allocations. 255 256 @return Operation status 257 @retval false Success. 258 @retval true Failure. 259 */ initialize_partition(MEM_ROOT * mem_root)260 virtual bool initialize_partition(MEM_ROOT *mem_root) {return false;} 261 262 263 /** 264 Truncate partitions. 265 266 Truncate all partitions matching table->part_info->read_partitions. 267 Handler level wrapper for truncating partitions, will ensure that 268 mark_trx_read_write() is called and also checks locking assertions. 269 270 @return Operation status. 271 @retval 0 Success. 272 @retval != 0 Error code. 273 */ truncate_partition()274 int truncate_partition() 275 { 276 handler *file= get_handler(); 277 if (!file) 278 { 279 return HA_ERR_WRONG_COMMAND; 280 } 281 assert(file->table_share->tmp_table != NO_TMP_TABLE || 282 file->m_lock_type == F_WRLCK); 283 file->mark_trx_read_write(); 284 return truncate_partition_low(); 285 } 286 /** 287 Change partitions. 288 289 Change partitions according to their partition_element::part_state set up 290 in prep_alter_part_table(). Will create new partitions and copy requested 291 partitions there. Also updating part_state to reflect current state. 292 293 Handler level wrapper for changing partitions. 294 This is the reason for having Partition_handler a friend class of handler, 295 mark_trx_read_write() is called and also checks locking assertions. 296 to ensure that mark_trx_read_write() is called and checking the asserts. 297 298 @param[in] create_info Table create info. 299 @param[in] path Path including table name. 300 @param[out] copied Number of rows copied. 301 @param[out] deleted Number of rows deleted. 302 */ change_partitions(HA_CREATE_INFO * create_info,const char * path,ulonglong * const copied,ulonglong * const deleted)303 int change_partitions(HA_CREATE_INFO *create_info, 304 const char *path, 305 ulonglong * const copied, 306 ulonglong * const deleted) 307 { 308 handler *file= get_handler(); 309 if (!file) 310 { 311 my_error(ER_ILLEGAL_HA, MYF(0), create_info->alias); 312 return HA_ERR_WRONG_COMMAND; 313 } 314 assert(file->table_share->tmp_table != NO_TMP_TABLE || 315 file->m_lock_type != F_UNLCK); 316 file->mark_trx_read_write(); 317 return change_partitions_low(create_info, path, copied, deleted); 318 } 319 /** 320 Alter flags. 321 322 Given a set of alter table flags, return which is supported. 323 324 @param flags Alter table operation flags. 325 326 @return Supported alter table flags. 327 */ alter_flags(uint flags)328 virtual uint alter_flags(uint flags) const 329 { return 0; } 330 331 /** 332 Get partition row type from SE 333 @param part_id Id of partition for which row type to be retrieved 334 @return Partition row type. 335 */ get_partition_row_type(uint part_id)336 virtual enum row_type get_partition_row_type(uint part_id) { 337 return ROW_TYPE_NOT_USED; 338 } 339 340 private: 341 /** 342 Truncate partition. 343 344 Low-level primitive for handler, implementing 345 Partition_handler::truncate_partition(). 346 347 @return Operation status 348 @retval 0 Success. 349 @retval != 0 Error code. 350 */ truncate_partition_low()351 virtual int truncate_partition_low() 352 { return HA_ERR_WRONG_COMMAND; } 353 /** 354 Truncate partition. 355 356 Low-level primitive for handler, implementing 357 Partition_handler::change_partitions(). 358 359 @param[in] create_info Table create info. 360 @param[in] path Path including table name. 361 @param[out] copied Number of rows copied. 362 @param[out] deleted Number of rows deleted. 363 364 @return Operation status 365 @retval 0 Success. 366 @retval != 0 Error code. 367 */ change_partitions_low(HA_CREATE_INFO * create_info,const char * path,ulonglong * const copied,ulonglong * const deleted)368 virtual int change_partitions_low(HA_CREATE_INFO *create_info, 369 const char *path, 370 ulonglong * const copied, 371 ulonglong * const deleted) 372 { 373 my_error(ER_ILLEGAL_HA, MYF(0), create_info->alias); 374 return HA_ERR_WRONG_COMMAND; 375 } 376 /** 377 Return the table handler. 378 379 For some partitioning specific functions it is still needed to access 380 the handler directly for transaction handling (mark_trx_read_write()) 381 and to assert correct locking. 382 383 @return handler or NULL if not supported. 384 */ get_handler()385 virtual handler *get_handler() 386 { return NULL; } 387 }; 388 389 390 /// Maps compare function to strict weak ordering required by Priority_queue. 391 struct Key_rec_less 392 { 393 typedef int (*key_compare_fun)(KEY**, uchar *, uchar *); 394 Key_rec_lessKey_rec_less395 explicit Key_rec_less(KEY **keys) 396 : m_keys(keys), m_fun(key_rec_cmp), m_max_at_top(false) 397 { 398 } 399 operatorKey_rec_less400 bool operator()(uchar *first, uchar *second) 401 { 402 const int cmpval= 403 (*m_fun)(m_keys, first + m_rec_offset, second + m_rec_offset); 404 return m_max_at_top ? cmpval < 0 : cmpval > 0; 405 } 406 407 KEY **m_keys; 408 key_compare_fun m_fun; 409 uint m_rec_offset; 410 bool m_max_at_top; 411 }; 412 413 414 /** 415 Partition_helper is a helper class that implements most generic partitioning 416 functionality such as: 417 table scan, index scan (both ordered and non-ordered), 418 insert (write_row()), delete and update. 419 And includes ALTER TABLE ... ADD/COALESCE/DROP/REORGANIZE/... PARTITION 420 support. 421 It also implements a cache for the auto increment value and check/repair for 422 rows in wrong partition. 423 424 How to use it: 425 Inherit it and implement: 426 - *_in_part() functions for row operations. 427 - prepare_for_new_partitions(), create_new_partition(), close_new_partitions() 428 write_row_in_new_part() for handling 'fast' alter partition. 429 */ 430 class Partition_helper : public Sql_alloc 431 { 432 typedef Priority_queue<uchar *, std::vector<uchar*>, Key_rec_less> Prio_queue; 433 public: 434 Partition_helper(handler *main_handler); 435 ~Partition_helper(); 436 437 /** 438 Set partition info. 439 440 To be called from Partition_handler. 441 442 @param part_info Partition info to use. 443 @param early True if called when part_info only created and parsed, 444 but not setup, checked or fixed. 445 */ 446 virtual void set_part_info_low(partition_info *part_info, bool early); 447 /** 448 Initialize variables used before the table is opened. 449 450 @param mem_root Memory root to allocate things from (not yet used). 451 452 @return Operation status. 453 @retval false success. 454 @retval true failure. 455 */ init_partitioning(MEM_ROOT * mem_root)456 inline bool init_partitioning(MEM_ROOT *mem_root) 457 { 458 #ifndef NDEBUG 459 m_key_not_found_partitions.bitmap= NULL; 460 #endif 461 return false; 462 } 463 464 465 /** 466 INSERT/UPDATE/DELETE functions. 467 @see handler.h 468 @{ 469 */ 470 471 /** 472 Insert a row to the partitioned table. 473 474 @param buf The row in MySQL Row Format. 475 476 @return Operation status. 477 @retval 0 Success 478 @retval != 0 Error code 479 */ 480 int ph_write_row(uchar *buf); 481 /** 482 Update an existing row in the partitioned table. 483 484 Yes, update_row() does what you expect, it updates a row. old_data will 485 have the previous row record in it, while new_data will have the newest 486 data in it. 487 Keep in mind that the server can do updates based on ordering if an 488 ORDER BY clause was used. Consecutive ordering is not guaranteed. 489 490 If the new record belongs to a different partition than the old record 491 then it will be inserted into the new partition and deleted from the old. 492 493 new_data is always record[0] 494 old_data is always record[1] 495 496 @param old_data The old record in MySQL Row Format. 497 @param new_data The new record in MySQL Row Format. 498 @param lookup_rows Indicator for TokuDB read free replication. 499 500 @return Operation status. 501 @retval 0 Success 502 @retval != 0 Error code 503 */ 504 int ph_update_row(const uchar *old_data, uchar *new_data, 505 bool lookup_rows = true); 506 /** 507 Delete an existing row in the partitioned table. 508 509 This will delete a row. buf will contain a copy of the row to be deleted. 510 The server will call this right after the current row has been read 511 (from either a previous rnd_xxx() or index_xxx() call). 512 If you keep a pointer to the last row or can access a primary key it will 513 make doing the deletion quite a bit easier. 514 Keep in mind that the server does no guarantee consecutive deletions. 515 ORDER BY clauses can be used. 516 517 buf is either record[0] or record[1] 518 519 @param buf The record in MySQL Row Format. 520 @param lookup_rows Indicator for TokuDB read free replication. 521 522 @return Operation status. 523 @retval 0 Success 524 @retval != 0 Error code 525 */ 526 int ph_delete_row(const uchar *buf, bool lookup_rows = true); 527 528 /** @} */ 529 530 /** Release unused auto increment values. */ 531 void ph_release_auto_increment(); 532 /** 533 Calculate key hash value from an null terminated array of fields. 534 Support function for KEY partitioning. 535 536 @param field_array An array of the fields in KEY partitioning 537 538 @return hash_value calculated 539 540 @note Uses the hash function on the character set of the field. 541 Integer and floating point fields use the binary character set by default. 542 */ 543 static uint32 ph_calculate_key_hash_value(Field **field_array); 544 /** Get checksum for table. 545 @return Checksum or 0 if not supported (which also may be a correct checksum!). 546 */ 547 ha_checksum ph_checksum() const; 548 549 /** 550 MODULE full table scan 551 552 This module is used for the most basic access method for any table 553 handler. This is to fetch all data through a full table scan. No 554 indexes are needed to implement this part. 555 It contains one method to start the scan (rnd_init) that can also be 556 called multiple times (typical in a nested loop join). Then proceeding 557 to the next record (rnd_next) and closing the scan (rnd_end). 558 To remember a record for later access there is a method (position) 559 and there is a method used to retrieve the record based on the stored 560 position. 561 The position can be a file position, a primary key, a ROWID dependent 562 on the handler below. 563 564 unlike index_init(), rnd_init() can be called two times 565 without rnd_end() in between (it only makes sense if scan=1). 566 then the second call should prepare for the new table scan 567 (e.g if rnd_init allocates the cursor, second call should 568 position it to the start of the table, no need to deallocate 569 and allocate it again. 570 @see handler.h 571 @{ 572 */ 573 574 int ph_rnd_init(bool scan); 575 int ph_rnd_end(); 576 int ph_rnd_next(uchar *buf); 577 void ph_position(const uchar *record); 578 int ph_rnd_pos(uchar *buf, uchar *pos); 579 580 /** @} */ 581 582 /** 583 MODULE index scan 584 585 This part of the handler interface is used to perform access through 586 indexes. The interface is defined as a scan interface but the handler 587 can also use key lookup if the index is a unique index or a primary 588 key index. 589 Index scans are mostly useful for SELECT queries but are an important 590 part also of UPDATE, DELETE, REPLACE and CREATE TABLE table AS SELECT 591 and so forth. 592 Naturally an index is needed for an index scan and indexes can either 593 be ordered, hash based. Some ordered indexes can return data in order 594 but not necessarily all of them. 595 There are many flags that define the behavior of indexes in the 596 various handlers. These methods are found in the optimizer module. 597 ------------------------------------------------------------------------- 598 599 index_read is called to start a scan of an index. The find_flag defines 600 the semantics of the scan. These flags are defined in 601 include/my_base.h 602 index_read_idx is the same but also initializes index before calling doing 603 the same thing as index_read. Thus it is similar to index_init followed 604 by index_read. This is also how we implement it. 605 606 index_read/index_read_idx does also return the first row. Thus for 607 key lookups, the index_read will be the only call to the handler in 608 the index scan. 609 610 index_init initializes an index before using it and index_end does 611 any end processing needed. 612 @{ 613 */ 614 615 int ph_index_init_setup(uint key_nr, bool sorted); 616 int ph_index_init(uint key_nr, bool sorted); 617 int ph_index_end(); 618 /* 619 These methods are used to jump to next or previous entry in the index 620 scan. There are also methods to jump to first and last entry. 621 */ 622 int ph_index_first(uchar *buf); 623 int ph_index_last(uchar *buf); 624 int ph_index_next(uchar *buf); 625 int ph_index_next_same(uchar *buf, const uchar *key, uint keylen); 626 int ph_index_prev(uchar *buf); 627 int ph_index_read_map(uchar *buf, 628 const uchar *key, 629 key_part_map keypart_map, 630 enum ha_rkey_function find_flag); 631 int ph_index_read_last_map(uchar *buf, 632 const uchar *key, 633 key_part_map keypart_map); 634 int ph_index_read_idx_map(uchar *buf, 635 uint index, 636 const uchar *key, 637 key_part_map keypart_map, 638 enum ha_rkey_function find_flag); 639 int ph_read_range_first(const key_range *start_key, 640 const key_range *end_key, 641 bool eq_range_arg, 642 bool sorted); 643 int ph_read_range_next(); 644 /** @} */ 645 646 /** 647 Functions matching Partition_handler API. 648 @{ 649 */ 650 651 /** 652 Get statistics from a specific partition. 653 @param[out] stat_info Area to report values into. 654 @param[out] check_sum Check sum of partition. 655 @param[in] part_id Partition to report from. 656 */ 657 virtual void get_dynamic_partition_info_low(ha_statistics *stat_info, 658 ha_checksum *check_sum, 659 uint part_id); 660 661 /** 662 Implement the partition changes defined by ALTER TABLE of partitions. 663 664 Add and copy if needed a number of partitions, during this operation 665 only read operation is ongoing in the server. This is used by 666 ADD PARTITION all types as well as by REORGANIZE PARTITION. For 667 one-phased implementations it is used also by DROP and COALESCE 668 PARTITIONs. 669 One-phased implementation needs the new frm file, other handlers will 670 get zero length and a NULL reference here. 671 672 @param[in] create_info HA_CREATE_INFO object describing all 673 fields and indexes in table 674 @param[in] path Complete path of db and table name 675 @param[out] copied Output parameter where number of copied 676 records are added 677 @param[out] deleted Output parameter where number of deleted 678 records are added 679 680 @return Operation status 681 @retval 0 Success 682 @retval != 0 Failure 683 */ 684 virtual int change_partitions(HA_CREATE_INFO *create_info, 685 const char *path, 686 ulonglong * const copied, 687 ulonglong * const deleted); 688 /** @} */ 689 690 protected: 691 /* Common helper functions to be used by inheriting engines. */ 692 693 /* 694 open/close functions. 695 */ 696 697 /** 698 Set m_part_share, Allocate internal bitmaps etc. used by open tables. 699 700 @param mem_root Memory root to allocate things from (not yet used). 701 702 @return Operation status. 703 @retval false success. 704 @retval true failure. 705 */ 706 bool open_partitioning(Partition_share *part_share); 707 /** 708 Close partitioning for a table. 709 710 Frees memory and release other resources. 711 */ 712 void close_partitioning(); 713 714 /** 715 Lock auto increment value if needed. 716 */ lock_auto_increment()717 inline void lock_auto_increment() 718 { 719 /* lock already taken */ 720 if (m_auto_increment_safe_stmt_log_lock) 721 return; 722 assert(!m_auto_increment_lock); 723 if(m_table->s->tmp_table == NO_TMP_TABLE) 724 { 725 m_auto_increment_lock= true; 726 m_part_share->lock_auto_inc(); 727 } 728 } 729 /** 730 unlock auto increment. 731 */ unlock_auto_increment()732 inline void unlock_auto_increment() 733 { 734 /* 735 If m_auto_increment_safe_stmt_log_lock is true, we have to keep the lock. 736 It will be set to false and thus unlocked at the end of the statement by 737 ha_partition::release_auto_increment. 738 */ 739 if(m_auto_increment_lock && !m_auto_increment_safe_stmt_log_lock) 740 { 741 m_part_share->unlock_auto_inc(); 742 m_auto_increment_lock= false; 743 } 744 } 745 /** 746 Get auto increment. 747 748 Only to be used for auto increment values that are the first field in 749 an unique index. 750 751 @param[in] increment Increment between generated numbers. 752 @param[in] nb_desired_values Number of values requested. 753 @param[out] first_value First reserved value (ULLONG_MAX on error). 754 @param[out] nb_reserved_values Number of values reserved. 755 */ 756 void get_auto_increment_first_field(ulonglong increment, 757 ulonglong nb_desired_values, 758 ulonglong *first_value, 759 ulonglong *nb_reserved_values); 760 761 /** 762 Initialize the record priority queue used for sorted index scans. 763 @return Operation status. 764 @retval 0 Success. 765 @retval != 0 Error code. 766 */ 767 int init_record_priority_queue(); 768 /** 769 Destroy the record priority queue used for sorted index scans. 770 */ 771 void destroy_record_priority_queue(); 772 /* 773 Administrative support functions. 774 */ 775 776 /** Print partitioning specific error. 777 @param error Error code. 778 @param errflag Error flag. 779 @return false if error is printed else true. 780 */ 781 bool print_partition_error(int error, myf errflag); 782 /** 783 Print a message row formatted for ANALYZE/CHECK/OPTIMIZE/REPAIR TABLE. 784 785 Modeled after mi_check_print_msg. 786 787 @param thd Thread context. 788 @param len Needed length for message buffer. 789 @param msg_type Message type. 790 @param db_name Database name. 791 @param table_name Table name. 792 @param op_name Operation name. 793 @param fmt Message (in printf format with additional arguments). 794 795 @return Operation status. 796 @retval false for success else true. 797 */ 798 bool print_admin_msg(THD *thd, 799 uint len, 800 const char *msg_type, 801 const char *db_name, 802 const char *table_name, 803 const char *op_name, 804 const char *fmt, 805 ...); 806 /** 807 Check/fix misplaced rows. 808 809 @param part_id Partition to check/fix. 810 @param repair If true, move misplaced rows to correct partition. 811 812 @return Operation status. 813 @retval 0 Success 814 @retval != 0 Error 815 */ 816 int check_misplaced_rows(uint part_id, bool repair); 817 /** 818 Set used partitions bitmap from Alter_info. 819 820 @return false if success else true. 821 */ 822 bool set_altered_partitions(); 823 824 private: 825 enum partition_index_scan_type 826 { 827 PARTITION_INDEX_READ= 1, 828 PARTITION_INDEX_FIRST, 829 PARTITION_INDEX_FIRST_UNORDERED, 830 PARTITION_INDEX_LAST, 831 PARTITION_INDEX_READ_LAST, 832 PARTITION_READ_RANGE, 833 PARTITION_NO_INDEX_SCAN 834 }; 835 836 /** handler to use (ha_partition, ha_innopart etc.) */ 837 handler *m_handler; 838 839 /* 840 Access methods to protected areas in handler to avoid adding 841 friend class Partition_helper in class handler. 842 */ 843 virtual THD *get_thd() const = 0; 844 virtual TABLE *get_table() const = 0; 845 virtual bool get_eq_range() const = 0; 846 virtual void set_eq_range(bool eq_range) = 0; 847 virtual void set_range_key_part(KEY_PART_INFO *key_part) = 0; 848 849 /* 850 Implementation of per partition operation by instantiated engine. 851 These must be implemented in the 'real' partition_helper subclass. 852 */ 853 854 /** 855 Write a row in the specified partition. 856 857 @see handler::write_row(). 858 859 @param part_id Partition to write to. 860 @param buf Buffer with data to write. 861 862 @return Operation status. 863 @retval 0 Success. 864 @retval != 0 Error code. 865 */ 866 virtual int write_row_in_part(uint part_id, uchar *buf) = 0; 867 /** 868 Update a row in the specified partition. 869 870 @see handler::update_row(). 871 872 @param part_id Partition to update in. 873 @param old_data Buffer containing old row. 874 @param new_data Buffer containing new row. 875 876 @return Operation status. 877 @retval 0 Success. 878 @retval != 0 Error code. 879 */ 880 virtual int update_row_in_part(uint new_part_id, 881 const uchar *old_data, 882 uchar *new_data) = 0; 883 /** 884 Delete an existing row in the specified partition. 885 886 @see handler::delete_row(). 887 888 @param part_id Partition to delete from. 889 @param buf Buffer containing row to delete. 890 891 @return Operation status. 892 @retval 0 Success. 893 @retval != 0 Error code. 894 */ 895 virtual int delete_row_in_part(uint part_id, const uchar *buf) = 0; 896 /** 897 Initialize the shared auto increment value. 898 899 @param no_lock If HA_STATUS_NO_LOCK should be used in info(HA_STATUS_AUTO). 900 901 Also sets stats.auto_increment_value. 902 */ 903 virtual int initialize_auto_increment(bool no_lock) = 0; 904 /** Release auto_increment in all underlying partitions. */ release_auto_increment_all_parts()905 virtual void release_auto_increment_all_parts() {} 906 /** Save or persist the current max auto increment. */ save_auto_increment(ulonglong nr)907 virtual void save_auto_increment(ulonglong nr) {} 908 /** 909 Per partition equivalent of rnd_* and index_* functions. 910 911 @see class handler. 912 */ 913 virtual int rnd_init_in_part(uint part_id, bool table_scan) = 0; 914 int ph_rnd_next_in_part(uint part_id, uchar *buf); 915 virtual int rnd_next_in_part(uint part_id, uchar *buf) = 0; 916 virtual int rnd_end_in_part(uint part_id, bool scan) = 0; 917 virtual void position_in_last_part(uchar *ref, const uchar *row) = 0; 918 /* If ph_rnd_pos is used then this needs to be implemented! */ rnd_pos_in_part(uint part_id,uchar * buf,uchar * pos)919 virtual int rnd_pos_in_part(uint part_id, uchar *buf, uchar *pos) 920 { assert(0); return HA_ERR_WRONG_COMMAND; } index_init_in_part(uint part,uint keynr,bool sorted)921 virtual int index_init_in_part(uint part, uint keynr, bool sorted) 922 { assert(0); return HA_ERR_WRONG_COMMAND; } index_end_in_part(uint part)923 virtual int index_end_in_part(uint part) 924 { assert(0); return HA_ERR_WRONG_COMMAND; } 925 virtual int index_first_in_part(uint part, uchar *buf) = 0; 926 virtual int index_last_in_part(uint part, uchar *buf) = 0; 927 virtual int index_prev_in_part(uint part, uchar *buf) = 0; 928 virtual int index_next_in_part(uint part, uchar *buf) = 0; 929 virtual int index_next_same_in_part(uint part, 930 uchar *buf, 931 const uchar *key, 932 uint length) = 0; 933 virtual int index_read_map_in_part(uint part, 934 uchar *buf, 935 const uchar *key, 936 key_part_map keypart_map, 937 enum ha_rkey_function find_flag) = 0; 938 virtual int index_read_last_map_in_part(uint part, 939 uchar *buf, 940 const uchar *key, 941 key_part_map keypart_map) = 0; 942 /** 943 Do read_range_first in the specified partition. 944 If buf is set, then copy the result there instead of table->record[0]. 945 */ 946 virtual int read_range_first_in_part(uint part, 947 uchar *buf, 948 const key_range *start_key, 949 const key_range *end_key, 950 bool eq_range, 951 bool sorted) = 0; 952 /** 953 Do read_range_next in the specified partition. 954 If buf is set, then copy the result there instead of table->record[0]. 955 */ 956 virtual int read_range_next_in_part(uint part, uchar *buf) = 0; 957 virtual int index_read_idx_map_in_part(uint part, 958 uchar *buf, 959 uint index, 960 const uchar *key, 961 key_part_map keypart_map, 962 enum ha_rkey_function find_flag) = 0; 963 /** 964 Initialize engine specific resources for the record priority queue 965 used duing ordered index reads for multiple partitions. 966 967 @param used_parts Number of partitions used in query 968 (number of set bits in m_part_info->read_partitions). 969 970 @return Operation status. 971 @retval 0 Success. 972 @retval != 0 Error code. 973 */ init_record_priority_queue_for_parts(uint used_parts)974 virtual int init_record_priority_queue_for_parts(uint used_parts) 975 { 976 return 0; 977 } 978 /** 979 Destroy and release engine specific resources used by the record 980 priority queue. 981 */ destroy_record_priority_queue_for_parts()982 virtual void destroy_record_priority_queue_for_parts() {} 983 /** 984 Checksum for a partition. 985 986 @param part_id Partition to checksum. 987 */ checksum_in_part(uint part_id)988 virtual ha_checksum checksum_in_part(uint part_id) const 989 { assert(0); return 0; } 990 /** 991 Copy a cached row. 992 993 Used when copying a row from the record priority queue to the return buffer. 994 For some engines, like InnoDB, only marked columns must be copied, 995 to preserve non-read columns. 996 997 @param[out] to_rec Buffer to copy to. 998 @param[in] from_rec Buffer to copy from. 999 */ copy_cached_row(uchar * to_rec,const uchar * from_rec)1000 virtual void copy_cached_row(uchar *to_rec, const uchar *from_rec) 1001 { memcpy(to_rec, from_rec, m_rec_length); } 1002 /** 1003 Prepare for creating new partitions during ALTER TABLE ... PARTITION. 1004 @param num_partitions Number of new partitions to be created. 1005 @param only_create True if only creating the partition 1006 (no open/lock is needed). 1007 1008 @return Operation status. 1009 @retval 0 Success. 1010 @retval != 0 Error code. 1011 */ 1012 virtual int prepare_for_new_partitions(uint num_partitions, 1013 bool only_create) = 0; 1014 /** 1015 Create a new partition to be filled during ALTER TABLE ... PARTITION. 1016 @param table Table to create the partition in. 1017 @param create_info Table/partition specific create info. 1018 @param part_name Partition name. 1019 @param new_part_id Partition id in new table. 1020 @param part_elem Partition element. 1021 1022 @return Operation status. 1023 @retval 0 Success. 1024 @retval != 0 Error code. 1025 */ 1026 virtual int create_new_partition(TABLE *table, 1027 HA_CREATE_INFO *create_info, 1028 const char *part_name, 1029 uint new_part_id, 1030 partition_element *part_elem) = 0; 1031 /** 1032 Close and finalize new partitions. 1033 */ 1034 virtual void close_new_partitions() = 0; 1035 /** 1036 write row to new partition. 1037 @param new_part New partition to write to. 1038 1039 @return Operation status. 1040 @retval 0 Success. 1041 @retval != 0 Error code. 1042 */ 1043 virtual int write_row_in_new_part(uint new_part) = 0; 1044 1045 /* Internal helper functions*/ 1046 /** 1047 Update auto increment value if current row contains a higher value. 1048 */ 1049 inline void set_auto_increment_if_higher(); 1050 /** 1051 Common routine to set up index scans. 1052 1053 Find out which partitions we'll need to read when scanning the specified 1054 range. 1055 1056 If we need to scan only one partition, set m_ordered_scan_ongoing=FALSE 1057 as we will not need to do merge ordering. 1058 1059 @param buf Buffer to later return record in (this function 1060 needs it to calculate partitioning function values) 1061 1062 @param idx_read_flag True <=> m_start_key has range start endpoint which 1063 probably can be used to determine the set of 1064 partitions to scan. 1065 False <=> there is no start endpoint. 1066 1067 @return Operation status. 1068 @retval 0 Success 1069 @retval !=0 Error code 1070 */ 1071 int partition_scan_set_up(uchar *buf, bool idx_read_flag); 1072 /** 1073 Common routine to handle index_next with unordered results. 1074 1075 These routines are used to scan partitions without considering order. 1076 This is performed in two situations. 1077 1) In read_multi_range this is the normal case 1078 2) When performing any type of index_read, index_first, index_last where 1079 all fields in the partition function is bound. In this case the index 1080 scan is performed on only one partition and thus it isn't necessary to 1081 perform any sort. 1082 1083 @param[out] buf Read row in MySQL Row Format. 1084 @param[in] next_same Called from index_next_same. 1085 1086 @return Operation status. 1087 @retval HA_ERR_END_OF_FILE End of scan 1088 @retval 0 Success 1089 @retval other Error code 1090 */ 1091 int handle_unordered_next(uchar *buf, bool is_next_same); 1092 /** 1093 Handle index_next when changing to new partition. 1094 1095 This routine is used to start the index scan on the next partition. 1096 Both initial start and after completing scan on one partition. 1097 1098 @param[out] buf Read row in MySQL Row Format 1099 1100 @return Operation status. 1101 @retval HA_ERR_END_OF_FILE End of scan 1102 @retval 0 Success 1103 @retval other Error code 1104 */ 1105 int handle_unordered_scan_next_partition(uchar *buf); 1106 /** 1107 Common routine to start index scan with ordered results. 1108 1109 @param[out] buf Read row in MySQL Row Format 1110 1111 @return Operation status 1112 @retval HA_ERR_END_OF_FILE End of scan 1113 @retval HA_ERR_KEY_NOT_FOUND End of scan 1114 @retval 0 Success 1115 @retval other Error code 1116 */ 1117 int handle_ordered_index_scan(uchar *buf); 1118 /** 1119 Add index_next/prev results from partitions without exact match. 1120 1121 If there where any partitions that returned HA_ERR_KEY_NOT_FOUND when 1122 ha_index_read_map was done, those partitions must be included in the 1123 following index_next/prev call. 1124 1125 @return Operation status 1126 @retval HA_ERR_END_OF_FILE End of scan 1127 @retval 0 Success 1128 @retval other Error code 1129 */ 1130 int handle_ordered_index_scan_key_not_found(); 1131 /** 1132 Common routine to handle index_prev with ordered results. 1133 1134 @param[out] buf Read row in MySQL Row Format. 1135 1136 @return Operation status. 1137 @retval HA_ERR_END_OF_FILE End of scan 1138 @retval 0 Success 1139 @retval other Error code 1140 */ 1141 int handle_ordered_prev(uchar *buf); 1142 /** 1143 Common routine to handle index_next with ordered results. 1144 1145 @param[out] buf Read row in MySQL Row Format. 1146 @param[in] next_same Called from index_next_same. 1147 1148 @return Operation status. 1149 @retval HA_ERR_END_OF_FILE End of scan 1150 @retval 0 Success 1151 @retval other Error code 1152 */ 1153 int handle_ordered_next(uchar *buf, bool is_next_same); 1154 /** 1155 Common routine for a number of index_read variants. 1156 1157 @param[out] buf Buffer where the record should be returned. 1158 @param[in] have_start_key TRUE <=> the left endpoint is available, i.e. 1159 we're in index_read call or in read_range_first 1160 call and the range has left endpoint. 1161 FALSE <=> there is no left endpoint (we're in 1162 read_range_first() call and the range has no 1163 left endpoint). 1164 1165 @return Operation status 1166 @retval 0 OK 1167 @retval HA_ERR_END_OF_FILE Whole index scanned, without finding the record. 1168 @retval HA_ERR_KEY_NOT_FOUND Record not found, but index cursor positioned. 1169 @retval other Error code. 1170 */ 1171 int common_index_read(uchar *buf, bool have_start_key); 1172 /** 1173 Common routine for index_first/index_last. 1174 1175 @param[out] buf Read row in MySQL Row Format. 1176 1177 @return Operation status. 1178 @retval 0 Success 1179 @retval != 0 Error code 1180 */ 1181 int common_first_last(uchar *buf); 1182 /** 1183 Return the top record in sort order. 1184 1185 @param[out] buf Row returned in MySQL Row Format. 1186 */ 1187 void return_top_record(uchar *buf); 1188 /** 1189 Copy partitions as part of ALTER TABLE of partitions. 1190 1191 change_partitions has done all the preparations, now it is time to 1192 actually copy the data from the reorganized partitions to the new 1193 partitions. 1194 1195 @param[out] copied Number of records copied. 1196 @param[out] deleted Number of records deleted. 1197 1198 @return Operation status 1199 @retval 0 Success 1200 @retval >0 Error code 1201 */ 1202 virtual int copy_partitions(ulonglong * const copied, 1203 ulonglong * const deleted); 1204 1205 /** 1206 Set table->read_set taking partitioning expressions into account. 1207 */ 1208 void set_partition_read_set(); 1209 1210 /* 1211 These could be private as well, 1212 but easier to expose them to derived classes to use. 1213 */ 1214 protected: 1215 1216 /** Convenience pointer to table from m_handler (i.e. m_handler->table). */ 1217 TABLE *m_table; 1218 /** All internal partitioning data! @{ */ 1219 /** Tables partitioning info (same as table->part_info) */ 1220 partition_info *m_part_info; 1221 /** Is primary key clustered. */ 1222 bool m_pkey_is_clustered; 1223 /** Cached value of m_part_info->is_sub_partitioned(). */ 1224 bool m_is_sub_partitioned; 1225 /** Partition share for auto_inc handling. */ 1226 Partition_share *m_part_share; 1227 /** Total number of partitions. */ 1228 uint m_tot_parts; 1229 uint m_last_part; // Last accessed partition. 1230 const uchar *m_err_rec; // record which gave error. 1231 bool m_auto_increment_safe_stmt_log_lock; 1232 bool m_auto_increment_lock; 1233 part_id_range m_part_spec; // Which parts to scan 1234 uint m_scan_value; // Value passed in rnd_init 1235 // call 1236 key_range m_start_key; // index read key range 1237 enum partition_index_scan_type m_index_scan_type;// What type of index 1238 // scan 1239 uint m_rec_length; // Local copy of record length 1240 1241 bool m_ordered; // Ordered/Unordered index scan. 1242 bool m_ordered_scan_ongoing; // Ordered index scan ongoing. 1243 bool m_reverse_order; // Scanning in reverse order (prev). 1244 /** Row and key buffer for ordered index scan. */ 1245 uchar *m_ordered_rec_buffer; 1246 /** Prio queue used by sorted read. */ 1247 Prio_queue *m_queue; 1248 /** Which partition is to deliver next result. */ 1249 uint m_top_entry; 1250 /** Offset in m_ordered_rec_buffer from part buffer to its record buffer. */ 1251 uint m_rec_offset; 1252 /** 1253 Current index used for sorting. 1254 If clustered PK exists, then it will be used as secondary index to 1255 sort on if the first is equal in key_rec_cmp. 1256 So if clustered pk: m_curr_key_info[0]= current index and 1257 m_curr_key_info[1]= pk and [2]= NULL. 1258 Otherwise [0]= current index, [1]= NULL, and we will 1259 sort by rowid as secondary sort key if equal first key. 1260 */ 1261 KEY *m_curr_key_info[3]; 1262 enum enum_using_ref { 1263 /** handler::ref is not copied to the PQ. */ 1264 REF_NOT_USED= 0, 1265 /** 1266 handler::ref is copied to the PQ but does not need to be used in sorting. 1267 */ 1268 REF_STORED_IN_PQ, 1269 /** handler::ref is copied to the PQ and must be used during sorting. */ 1270 REF_USED_FOR_SORT}; 1271 /** How handler::ref is used in the priority queue. */ 1272 enum_using_ref m_ref_usage; 1273 /** Set if previous index_* call returned HA_ERR_KEY_NOT_FOUND. */ 1274 bool m_key_not_found; 1275 /** Partitions that returned HA_ERR_KEY_NOT_FOUND. */ 1276 MY_BITMAP m_key_not_found_partitions; 1277 /** @} */ 1278 }; 1279 #endif /* PARTITION_HANDLER_INCLUDED */ 1280