1 #ifndef PARTITION_HANDLER_INCLUDED 2 #define PARTITION_HANDLER_INCLUDED 3 4 /* 5 Copyright (c) 2005, 2019, Oracle and/or its affiliates. All rights reserved. 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 DBUG_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 DBUG_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 DBUG_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 DBUG_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 DBUG_OFF 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. 498 499 @return Operation status. 500 @retval 0 Success 501 @retval != 0 Error code 502 */ 503 int ph_update_row(const uchar *old_data, uchar *new_data); 504 /** 505 Delete an existing row in the partitioned table. 506 507 This will delete a row. buf will contain a copy of the row to be deleted. 508 The server will call this right after the current row has been read 509 (from either a previous rnd_xxx() or index_xxx() call). 510 If you keep a pointer to the last row or can access a primary key it will 511 make doing the deletion quite a bit easier. 512 Keep in mind that the server does no guarantee consecutive deletions. 513 ORDER BY clauses can be used. 514 515 buf is either record[0] or record[1] 516 517 @param buf The record in MySQL Row Format. 518 519 @return Operation status. 520 @retval 0 Success 521 @retval != 0 Error code 522 */ 523 int ph_delete_row(const uchar *buf); 524 525 /** @} */ 526 527 /** Release unused auto increment values. */ 528 void ph_release_auto_increment(); 529 /** 530 Calculate key hash value from an null terminated array of fields. 531 Support function for KEY partitioning. 532 533 @param field_array An array of the fields in KEY partitioning 534 535 @return hash_value calculated 536 537 @note Uses the hash function on the character set of the field. 538 Integer and floating point fields use the binary character set by default. 539 */ 540 static uint32 ph_calculate_key_hash_value(Field **field_array); 541 /** Get checksum for table. 542 @return Checksum or 0 if not supported (which also may be a correct checksum!). 543 */ 544 ha_checksum ph_checksum() const; 545 546 /** 547 MODULE full table scan 548 549 This module is used for the most basic access method for any table 550 handler. This is to fetch all data through a full table scan. No 551 indexes are needed to implement this part. 552 It contains one method to start the scan (rnd_init) that can also be 553 called multiple times (typical in a nested loop join). Then proceeding 554 to the next record (rnd_next) and closing the scan (rnd_end). 555 To remember a record for later access there is a method (position) 556 and there is a method used to retrieve the record based on the stored 557 position. 558 The position can be a file position, a primary key, a ROWID dependent 559 on the handler below. 560 561 unlike index_init(), rnd_init() can be called two times 562 without rnd_end() in between (it only makes sense if scan=1). 563 then the second call should prepare for the new table scan 564 (e.g if rnd_init allocates the cursor, second call should 565 position it to the start of the table, no need to deallocate 566 and allocate it again. 567 @see handler.h 568 @{ 569 */ 570 571 int ph_rnd_init(bool scan); 572 int ph_rnd_end(); 573 int ph_rnd_next(uchar *buf); 574 void ph_position(const uchar *record); 575 int ph_rnd_pos(uchar *buf, uchar *pos); 576 577 /** @} */ 578 579 /** 580 MODULE index scan 581 582 This part of the handler interface is used to perform access through 583 indexes. The interface is defined as a scan interface but the handler 584 can also use key lookup if the index is a unique index or a primary 585 key index. 586 Index scans are mostly useful for SELECT queries but are an important 587 part also of UPDATE, DELETE, REPLACE and CREATE TABLE table AS SELECT 588 and so forth. 589 Naturally an index is needed for an index scan and indexes can either 590 be ordered, hash based. Some ordered indexes can return data in order 591 but not necessarily all of them. 592 There are many flags that define the behavior of indexes in the 593 various handlers. These methods are found in the optimizer module. 594 ------------------------------------------------------------------------- 595 596 index_read is called to start a scan of an index. The find_flag defines 597 the semantics of the scan. These flags are defined in 598 include/my_base.h 599 index_read_idx is the same but also initializes index before calling doing 600 the same thing as index_read. Thus it is similar to index_init followed 601 by index_read. This is also how we implement it. 602 603 index_read/index_read_idx does also return the first row. Thus for 604 key lookups, the index_read will be the only call to the handler in 605 the index scan. 606 607 index_init initializes an index before using it and index_end does 608 any end processing needed. 609 @{ 610 */ 611 612 int ph_index_init_setup(uint key_nr, bool sorted); 613 int ph_index_init(uint key_nr, bool sorted); 614 int ph_index_end(); 615 /* 616 These methods are used to jump to next or previous entry in the index 617 scan. There are also methods to jump to first and last entry. 618 */ 619 int ph_index_first(uchar *buf); 620 int ph_index_last(uchar *buf); 621 int ph_index_next(uchar *buf); 622 int ph_index_next_same(uchar *buf, const uchar *key, uint keylen); 623 int ph_index_prev(uchar *buf); 624 int ph_index_read_map(uchar *buf, 625 const uchar *key, 626 key_part_map keypart_map, 627 enum ha_rkey_function find_flag); 628 int ph_index_read_last_map(uchar *buf, 629 const uchar *key, 630 key_part_map keypart_map); 631 int ph_index_read_idx_map(uchar *buf, 632 uint index, 633 const uchar *key, 634 key_part_map keypart_map, 635 enum ha_rkey_function find_flag); 636 int ph_read_range_first(const key_range *start_key, 637 const key_range *end_key, 638 bool eq_range_arg, 639 bool sorted); 640 int ph_read_range_next(); 641 /** @} */ 642 643 /** 644 Functions matching Partition_handler API. 645 @{ 646 */ 647 648 /** 649 Get statistics from a specific partition. 650 @param[out] stat_info Area to report values into. 651 @param[out] check_sum Check sum of partition. 652 @param[in] part_id Partition to report from. 653 */ 654 virtual void get_dynamic_partition_info_low(ha_statistics *stat_info, 655 ha_checksum *check_sum, 656 uint part_id); 657 658 /** 659 Implement the partition changes defined by ALTER TABLE of partitions. 660 661 Add and copy if needed a number of partitions, during this operation 662 only read operation is ongoing in the server. This is used by 663 ADD PARTITION all types as well as by REORGANIZE PARTITION. For 664 one-phased implementations it is used also by DROP and COALESCE 665 PARTITIONs. 666 One-phased implementation needs the new frm file, other handlers will 667 get zero length and a NULL reference here. 668 669 @param[in] create_info HA_CREATE_INFO object describing all 670 fields and indexes in table 671 @param[in] path Complete path of db and table name 672 @param[out] copied Output parameter where number of copied 673 records are added 674 @param[out] deleted Output parameter where number of deleted 675 records are added 676 677 @return Operation status 678 @retval 0 Success 679 @retval != 0 Failure 680 */ 681 virtual int change_partitions(HA_CREATE_INFO *create_info, 682 const char *path, 683 ulonglong * const copied, 684 ulonglong * const deleted); 685 /** @} */ 686 687 protected: 688 /* Common helper functions to be used by inheriting engines. */ 689 690 /* 691 open/close functions. 692 */ 693 694 /** 695 Set m_part_share, Allocate internal bitmaps etc. used by open tables. 696 697 @param mem_root Memory root to allocate things from (not yet used). 698 699 @return Operation status. 700 @retval false success. 701 @retval true failure. 702 */ 703 bool open_partitioning(Partition_share *part_share); 704 /** 705 Close partitioning for a table. 706 707 Frees memory and release other resources. 708 */ 709 void close_partitioning(); 710 711 /** 712 Lock auto increment value if needed. 713 */ lock_auto_increment()714 inline void lock_auto_increment() 715 { 716 /* lock already taken */ 717 if (m_auto_increment_safe_stmt_log_lock) 718 return; 719 DBUG_ASSERT(!m_auto_increment_lock); 720 if(m_table->s->tmp_table == NO_TMP_TABLE) 721 { 722 m_auto_increment_lock= true; 723 m_part_share->lock_auto_inc(); 724 } 725 } 726 /** 727 unlock auto increment. 728 */ unlock_auto_increment()729 inline void unlock_auto_increment() 730 { 731 /* 732 If m_auto_increment_safe_stmt_log_lock is true, we have to keep the lock. 733 It will be set to false and thus unlocked at the end of the statement by 734 ha_partition::release_auto_increment. 735 */ 736 if(m_auto_increment_lock && !m_auto_increment_safe_stmt_log_lock) 737 { 738 m_part_share->unlock_auto_inc(); 739 m_auto_increment_lock= false; 740 } 741 } 742 /** 743 Get auto increment. 744 745 Only to be used for auto increment values that are the first field in 746 an unique index. 747 748 @param[in] increment Increment between generated numbers. 749 @param[in] nb_desired_values Number of values requested. 750 @param[out] first_value First reserved value (ULLONG_MAX on error). 751 @param[out] nb_reserved_values Number of values reserved. 752 */ 753 void get_auto_increment_first_field(ulonglong increment, 754 ulonglong nb_desired_values, 755 ulonglong *first_value, 756 ulonglong *nb_reserved_values); 757 758 /** 759 Initialize the record priority queue used for sorted index scans. 760 @return Operation status. 761 @retval 0 Success. 762 @retval != 0 Error code. 763 */ 764 int init_record_priority_queue(); 765 /** 766 Destroy the record priority queue used for sorted index scans. 767 */ 768 void destroy_record_priority_queue(); 769 /* 770 Administrative support functions. 771 */ 772 773 /** Print partitioning specific error. 774 @param error Error code. 775 @param errflag Error flag. 776 @return false if error is printed else true. 777 */ 778 bool print_partition_error(int error, myf errflag); 779 /** 780 Print a message row formatted for ANALYZE/CHECK/OPTIMIZE/REPAIR TABLE. 781 782 Modeled after mi_check_print_msg. 783 784 @param thd Thread context. 785 @param len Needed length for message buffer. 786 @param msg_type Message type. 787 @param db_name Database name. 788 @param table_name Table name. 789 @param op_name Operation name. 790 @param fmt Message (in printf format with additional arguments). 791 792 @return Operation status. 793 @retval false for success else true. 794 */ 795 bool print_admin_msg(THD *thd, 796 uint len, 797 const char *msg_type, 798 const char *db_name, 799 const char *table_name, 800 const char *op_name, 801 const char *fmt, 802 ...); 803 /** 804 Check/fix misplaced rows. 805 806 @param part_id Partition to check/fix. 807 @param repair If true, move misplaced rows to correct partition. 808 809 @return Operation status. 810 @retval 0 Success 811 @retval != 0 Error 812 */ 813 int check_misplaced_rows(uint part_id, bool repair); 814 /** 815 Set used partitions bitmap from Alter_info. 816 817 @return false if success else true. 818 */ 819 bool set_altered_partitions(); 820 821 private: 822 enum partition_index_scan_type 823 { 824 PARTITION_INDEX_READ= 1, 825 PARTITION_INDEX_FIRST, 826 PARTITION_INDEX_FIRST_UNORDERED, 827 PARTITION_INDEX_LAST, 828 PARTITION_INDEX_READ_LAST, 829 PARTITION_READ_RANGE, 830 PARTITION_NO_INDEX_SCAN 831 }; 832 833 /** handler to use (ha_partition, ha_innopart etc.) */ 834 handler *m_handler; 835 836 /* 837 Access methods to protected areas in handler to avoid adding 838 friend class Partition_helper in class handler. 839 */ 840 virtual THD *get_thd() const = 0; 841 virtual TABLE *get_table() const = 0; 842 virtual bool get_eq_range() const = 0; 843 virtual void set_eq_range(bool eq_range) = 0; 844 virtual void set_range_key_part(KEY_PART_INFO *key_part) = 0; 845 846 /* 847 Implementation of per partition operation by instantiated engine. 848 These must be implemented in the 'real' partition_helper subclass. 849 */ 850 851 /** 852 Write a row in the specified partition. 853 854 @see handler::write_row(). 855 856 @param part_id Partition to write to. 857 @param buf Buffer with data to write. 858 859 @return Operation status. 860 @retval 0 Success. 861 @retval != 0 Error code. 862 */ 863 virtual int write_row_in_part(uint part_id, uchar *buf) = 0; 864 /** 865 Update a row in the specified partition. 866 867 @see handler::update_row(). 868 869 @param part_id Partition to update in. 870 @param old_data Buffer containing old row. 871 @param new_data Buffer containing new row. 872 873 @return Operation status. 874 @retval 0 Success. 875 @retval != 0 Error code. 876 */ 877 virtual int update_row_in_part(uint new_part_id, 878 const uchar *old_data, 879 uchar *new_data) = 0; 880 /** 881 Delete an existing row in the specified partition. 882 883 @see handler::delete_row(). 884 885 @param part_id Partition to delete from. 886 @param buf Buffer containing row to delete. 887 888 @return Operation status. 889 @retval 0 Success. 890 @retval != 0 Error code. 891 */ 892 virtual int delete_row_in_part(uint part_id, const uchar *buf) = 0; 893 /** 894 Initialize the shared auto increment value. 895 896 @param no_lock If HA_STATUS_NO_LOCK should be used in info(HA_STATUS_AUTO). 897 898 Also sets stats.auto_increment_value. 899 */ 900 virtual int initialize_auto_increment(bool no_lock) = 0; 901 /** Release auto_increment in all underlying partitions. */ release_auto_increment_all_parts()902 virtual void release_auto_increment_all_parts() {} 903 /** Save or persist the current max auto increment. */ save_auto_increment(ulonglong nr)904 virtual void save_auto_increment(ulonglong nr) {} 905 /** 906 Per partition equivalent of rnd_* and index_* functions. 907 908 @see class handler. 909 */ 910 virtual int rnd_init_in_part(uint part_id, bool table_scan) = 0; 911 int ph_rnd_next_in_part(uint part_id, uchar *buf); 912 virtual int rnd_next_in_part(uint part_id, uchar *buf) = 0; 913 virtual int rnd_end_in_part(uint part_id, bool scan) = 0; 914 virtual void position_in_last_part(uchar *ref, const uchar *row) = 0; 915 /* If ph_rnd_pos is used then this needs to be implemented! */ rnd_pos_in_part(uint part_id,uchar * buf,uchar * pos)916 virtual int rnd_pos_in_part(uint part_id, uchar *buf, uchar *pos) 917 { DBUG_ASSERT(0); return HA_ERR_WRONG_COMMAND; } index_init_in_part(uint part,uint keynr,bool sorted)918 virtual int index_init_in_part(uint part, uint keynr, bool sorted) 919 { DBUG_ASSERT(0); return HA_ERR_WRONG_COMMAND; } index_end_in_part(uint part)920 virtual int index_end_in_part(uint part) 921 { DBUG_ASSERT(0); return HA_ERR_WRONG_COMMAND; } 922 virtual int index_first_in_part(uint part, uchar *buf) = 0; 923 virtual int index_last_in_part(uint part, uchar *buf) = 0; 924 virtual int index_prev_in_part(uint part, uchar *buf) = 0; 925 virtual int index_next_in_part(uint part, uchar *buf) = 0; 926 virtual int index_next_same_in_part(uint part, 927 uchar *buf, 928 const uchar *key, 929 uint length) = 0; 930 virtual int index_read_map_in_part(uint part, 931 uchar *buf, 932 const uchar *key, 933 key_part_map keypart_map, 934 enum ha_rkey_function find_flag) = 0; 935 virtual int index_read_last_map_in_part(uint part, 936 uchar *buf, 937 const uchar *key, 938 key_part_map keypart_map) = 0; 939 /** 940 Do read_range_first in the specified partition. 941 If buf is set, then copy the result there instead of table->record[0]. 942 */ 943 virtual int read_range_first_in_part(uint part, 944 uchar *buf, 945 const key_range *start_key, 946 const key_range *end_key, 947 bool eq_range, 948 bool sorted) = 0; 949 /** 950 Do read_range_next in the specified partition. 951 If buf is set, then copy the result there instead of table->record[0]. 952 */ 953 virtual int read_range_next_in_part(uint part, uchar *buf) = 0; 954 virtual int index_read_idx_map_in_part(uint part, 955 uchar *buf, 956 uint index, 957 const uchar *key, 958 key_part_map keypart_map, 959 enum ha_rkey_function find_flag) = 0; 960 /** 961 Initialize engine specific resources for the record priority queue 962 used duing ordered index reads for multiple partitions. 963 964 @param used_parts Number of partitions used in query 965 (number of set bits in m_part_info->read_partitions). 966 967 @return Operation status. 968 @retval 0 Success. 969 @retval != 0 Error code. 970 */ init_record_priority_queue_for_parts(uint used_parts)971 virtual int init_record_priority_queue_for_parts(uint used_parts) 972 { 973 return 0; 974 } 975 /** 976 Destroy and release engine specific resources used by the record 977 priority queue. 978 */ destroy_record_priority_queue_for_parts()979 virtual void destroy_record_priority_queue_for_parts() {} 980 /** 981 Checksum for a partition. 982 983 @param part_id Partition to checksum. 984 */ checksum_in_part(uint part_id)985 virtual ha_checksum checksum_in_part(uint part_id) const 986 { DBUG_ASSERT(0); return 0; } 987 /** 988 Copy a cached row. 989 990 Used when copying a row from the record priority queue to the return buffer. 991 For some engines, like InnoDB, only marked columns must be copied, 992 to preserve non-read columns. 993 994 @param[out] to_rec Buffer to copy to. 995 @param[in] from_rec Buffer to copy from. 996 */ copy_cached_row(uchar * to_rec,const uchar * from_rec)997 virtual void copy_cached_row(uchar *to_rec, const uchar *from_rec) 998 { memcpy(to_rec, from_rec, m_rec_length); } 999 /** 1000 Prepare for creating new partitions during ALTER TABLE ... PARTITION. 1001 @param num_partitions Number of new partitions to be created. 1002 @param only_create True if only creating the partition 1003 (no open/lock is needed). 1004 1005 @return Operation status. 1006 @retval 0 Success. 1007 @retval != 0 Error code. 1008 */ 1009 virtual int prepare_for_new_partitions(uint num_partitions, 1010 bool only_create) = 0; 1011 /** 1012 Create a new partition to be filled during ALTER TABLE ... PARTITION. 1013 @param table Table to create the partition in. 1014 @param create_info Table/partition specific create info. 1015 @param part_name Partition name. 1016 @param new_part_id Partition id in new table. 1017 @param part_elem Partition element. 1018 1019 @return Operation status. 1020 @retval 0 Success. 1021 @retval != 0 Error code. 1022 */ 1023 virtual int create_new_partition(TABLE *table, 1024 HA_CREATE_INFO *create_info, 1025 const char *part_name, 1026 uint new_part_id, 1027 partition_element *part_elem) = 0; 1028 /** 1029 Close and finalize new partitions. 1030 */ 1031 virtual void close_new_partitions() = 0; 1032 /** 1033 write row to new partition. 1034 @param new_part New partition to write to. 1035 1036 @return Operation status. 1037 @retval 0 Success. 1038 @retval != 0 Error code. 1039 */ 1040 virtual int write_row_in_new_part(uint new_part) = 0; 1041 1042 /* Internal helper functions*/ 1043 /** 1044 Update auto increment value if current row contains a higher value. 1045 */ 1046 inline void set_auto_increment_if_higher(); 1047 /** 1048 Common routine to set up index scans. 1049 1050 Find out which partitions we'll need to read when scanning the specified 1051 range. 1052 1053 If we need to scan only one partition, set m_ordered_scan_ongoing=FALSE 1054 as we will not need to do merge ordering. 1055 1056 @param buf Buffer to later return record in (this function 1057 needs it to calculate partitioning function values) 1058 1059 @param idx_read_flag True <=> m_start_key has range start endpoint which 1060 probably can be used to determine the set of 1061 partitions to scan. 1062 False <=> there is no start endpoint. 1063 1064 @return Operation status. 1065 @retval 0 Success 1066 @retval !=0 Error code 1067 */ 1068 int partition_scan_set_up(uchar *buf, bool idx_read_flag); 1069 /** 1070 Common routine to handle index_next with unordered results. 1071 1072 These routines are used to scan partitions without considering order. 1073 This is performed in two situations. 1074 1) In read_multi_range this is the normal case 1075 2) When performing any type of index_read, index_first, index_last where 1076 all fields in the partition function is bound. In this case the index 1077 scan is performed on only one partition and thus it isn't necessary to 1078 perform any sort. 1079 1080 @param[out] buf Read row in MySQL Row Format. 1081 @param[in] next_same Called from index_next_same. 1082 1083 @return Operation status. 1084 @retval HA_ERR_END_OF_FILE End of scan 1085 @retval 0 Success 1086 @retval other Error code 1087 */ 1088 int handle_unordered_next(uchar *buf, bool is_next_same); 1089 /** 1090 Handle index_next when changing to new partition. 1091 1092 This routine is used to start the index scan on the next partition. 1093 Both initial start and after completing scan on one partition. 1094 1095 @param[out] buf Read row in MySQL Row Format 1096 1097 @return Operation status. 1098 @retval HA_ERR_END_OF_FILE End of scan 1099 @retval 0 Success 1100 @retval other Error code 1101 */ 1102 int handle_unordered_scan_next_partition(uchar *buf); 1103 /** 1104 Common routine to start index scan with ordered results. 1105 1106 @param[out] buf Read row in MySQL Row Format 1107 1108 @return Operation status 1109 @retval HA_ERR_END_OF_FILE End of scan 1110 @retval HA_ERR_KEY_NOT_FOUND End of scan 1111 @retval 0 Success 1112 @retval other Error code 1113 */ 1114 int handle_ordered_index_scan(uchar *buf); 1115 /** 1116 Add index_next/prev results from partitions without exact match. 1117 1118 If there where any partitions that returned HA_ERR_KEY_NOT_FOUND when 1119 ha_index_read_map was done, those partitions must be included in the 1120 following index_next/prev call. 1121 1122 @return Operation status 1123 @retval HA_ERR_END_OF_FILE End of scan 1124 @retval 0 Success 1125 @retval other Error code 1126 */ 1127 int handle_ordered_index_scan_key_not_found(); 1128 /** 1129 Common routine to handle index_prev with ordered results. 1130 1131 @param[out] buf Read row in MySQL Row Format. 1132 1133 @return Operation status. 1134 @retval HA_ERR_END_OF_FILE End of scan 1135 @retval 0 Success 1136 @retval other Error code 1137 */ 1138 int handle_ordered_prev(uchar *buf); 1139 /** 1140 Common routine to handle index_next with ordered results. 1141 1142 @param[out] buf Read row in MySQL Row Format. 1143 @param[in] next_same Called from index_next_same. 1144 1145 @return Operation status. 1146 @retval HA_ERR_END_OF_FILE End of scan 1147 @retval 0 Success 1148 @retval other Error code 1149 */ 1150 int handle_ordered_next(uchar *buf, bool is_next_same); 1151 /** 1152 Common routine for a number of index_read variants. 1153 1154 @param[out] buf Buffer where the record should be returned. 1155 @param[in] have_start_key TRUE <=> the left endpoint is available, i.e. 1156 we're in index_read call or in read_range_first 1157 call and the range has left endpoint. 1158 FALSE <=> there is no left endpoint (we're in 1159 read_range_first() call and the range has no 1160 left endpoint). 1161 1162 @return Operation status 1163 @retval 0 OK 1164 @retval HA_ERR_END_OF_FILE Whole index scanned, without finding the record. 1165 @retval HA_ERR_KEY_NOT_FOUND Record not found, but index cursor positioned. 1166 @retval other Error code. 1167 */ 1168 int common_index_read(uchar *buf, bool have_start_key); 1169 /** 1170 Common routine for index_first/index_last. 1171 1172 @param[out] buf Read row in MySQL Row Format. 1173 1174 @return Operation status. 1175 @retval 0 Success 1176 @retval != 0 Error code 1177 */ 1178 int common_first_last(uchar *buf); 1179 /** 1180 Return the top record in sort order. 1181 1182 @param[out] buf Row returned in MySQL Row Format. 1183 */ 1184 void return_top_record(uchar *buf); 1185 /** 1186 Copy partitions as part of ALTER TABLE of partitions. 1187 1188 change_partitions has done all the preparations, now it is time to 1189 actually copy the data from the reorganized partitions to the new 1190 partitions. 1191 1192 @param[out] copied Number of records copied. 1193 @param[out] deleted Number of records deleted. 1194 1195 @return Operation status 1196 @retval 0 Success 1197 @retval >0 Error code 1198 */ 1199 virtual int copy_partitions(ulonglong * const copied, 1200 ulonglong * const deleted); 1201 1202 /** 1203 Set table->read_set taking partitioning expressions into account. 1204 */ 1205 void set_partition_read_set(); 1206 1207 /* 1208 These could be private as well, 1209 but easier to expose them to derived classes to use. 1210 */ 1211 protected: 1212 1213 /** Convenience pointer to table from m_handler (i.e. m_handler->table). */ 1214 TABLE *m_table; 1215 /** All internal partitioning data! @{ */ 1216 /** Tables partitioning info (same as table->part_info) */ 1217 partition_info *m_part_info; 1218 /** Is primary key clustered. */ 1219 bool m_pkey_is_clustered; 1220 /** Cached value of m_part_info->is_sub_partitioned(). */ 1221 bool m_is_sub_partitioned; 1222 /** Partition share for auto_inc handling. */ 1223 Partition_share *m_part_share; 1224 /** Total number of partitions. */ 1225 uint m_tot_parts; 1226 uint m_last_part; // Last accessed partition. 1227 const uchar *m_err_rec; // record which gave error. 1228 bool m_auto_increment_safe_stmt_log_lock; 1229 bool m_auto_increment_lock; 1230 part_id_range m_part_spec; // Which parts to scan 1231 uint m_scan_value; // Value passed in rnd_init 1232 // call 1233 key_range m_start_key; // index read key range 1234 enum partition_index_scan_type m_index_scan_type;// What type of index 1235 // scan 1236 uint m_rec_length; // Local copy of record length 1237 1238 bool m_ordered; // Ordered/Unordered index scan. 1239 bool m_ordered_scan_ongoing; // Ordered index scan ongoing. 1240 bool m_reverse_order; // Scanning in reverse order (prev). 1241 /** Row and key buffer for ordered index scan. */ 1242 uchar *m_ordered_rec_buffer; 1243 /** Prio queue used by sorted read. */ 1244 Prio_queue *m_queue; 1245 /** Which partition is to deliver next result. */ 1246 uint m_top_entry; 1247 /** Offset in m_ordered_rec_buffer from part buffer to its record buffer. */ 1248 uint m_rec_offset; 1249 /** 1250 Current index used for sorting. 1251 If clustered PK exists, then it will be used as secondary index to 1252 sort on if the first is equal in key_rec_cmp. 1253 So if clustered pk: m_curr_key_info[0]= current index and 1254 m_curr_key_info[1]= pk and [2]= NULL. 1255 Otherwise [0]= current index, [1]= NULL, and we will 1256 sort by rowid as secondary sort key if equal first key. 1257 */ 1258 KEY *m_curr_key_info[3]; 1259 enum enum_using_ref { 1260 /** handler::ref is not copied to the PQ. */ 1261 REF_NOT_USED= 0, 1262 /** 1263 handler::ref is copied to the PQ but does not need to be used in sorting. 1264 */ 1265 REF_STORED_IN_PQ, 1266 /** handler::ref is copied to the PQ and must be used during sorting. */ 1267 REF_USED_FOR_SORT}; 1268 /** How handler::ref is used in the priority queue. */ 1269 enum_using_ref m_ref_usage; 1270 /** Set if previous index_* call returned HA_ERR_KEY_NOT_FOUND. */ 1271 bool m_key_not_found; 1272 /** Partitions that returned HA_ERR_KEY_NOT_FOUND. */ 1273 MY_BITMAP m_key_not_found_partitions; 1274 /** @} */ 1275 }; 1276 #endif /* PARTITION_HANDLER_INCLUDED */ 1277