1 /* Copyright (c) 2001, 2010, Oracle and/or its affiliates. 2 Copyright (c) 2010, 2015, MariaDB 3 4 This program is free software; you can redistribute it and/or modify 5 it under the terms of the GNU General Public License as published by 6 the Free Software Foundation; version 2 of the License. 7 8 This program is distributed in the hope that it will be useful, 9 but WITHOUT ANY WARRANTY; without even the implied warranty of 10 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 11 GNU General Public License for more details. 12 13 You should have received a copy of the GNU General Public License 14 along with this program; if not, write to the Free Software 15 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1335 USA */ 16 17 /* 18 Function to handle quick removal of duplicates 19 This code is used when doing multi-table deletes to find the rows in 20 reference tables that needs to be deleted. 21 22 The basic idea is as follows: 23 24 Store first all strings in a binary tree, ignoring duplicates. 25 When the tree uses more memory than 'max_heap_table_size', 26 write the tree (in sorted order) out to disk and start with a new tree. 27 When all data has been generated, merge the trees (removing any found 28 duplicates). 29 30 The unique entries will be returned in sort order, to ensure that we do the 31 deletes in disk order. 32 */ 33 34 #include "mariadb.h" 35 #include "sql_priv.h" 36 #include "unireg.h" 37 #include "sql_sort.h" 38 #include "queues.h" // QUEUE 39 #include "my_tree.h" // element_count 40 #include "uniques.h" // Unique 41 #include "sql_sort.h" 42 #include "myisamchk.h" // BUFFPEK 43 44 int unique_write_to_file(uchar* key, element_count count, Unique *unique) 45 { 46 /* 47 Use unique->size (size of element stored in the tree) and not 48 unique->tree.size_of_element. The latter is different from unique->size 49 when tree implementation chooses to store pointer to key in TREE_ELEMENT 50 (instead of storing the element itself there) 51 */ 52 return my_b_write(&unique->file, key, unique->size) ? 1 : 0; 53 } 54 55 int unique_write_to_file_with_count(uchar* key, element_count count, Unique *unique) 56 { 57 return my_b_write(&unique->file, key, unique->size) || 58 my_b_write(&unique->file, (uchar*)&count, sizeof(element_count)) ? 1 : 0; 59 } 60 61 int unique_write_to_ptrs(uchar* key, element_count count, Unique *unique) 62 { 63 memcpy(unique->sort.record_pointers, key, unique->size); 64 unique->sort.record_pointers+=unique->size; 65 return 0; 66 } 67 68 int unique_intersect_write_to_ptrs(uchar* key, element_count count, Unique *unique) 69 { 70 if (count >= unique->min_dupl_count) 71 { 72 memcpy(unique->sort.record_pointers, key, unique->size); 73 unique->sort.record_pointers+=unique->size; 74 } 75 else 76 unique->filtered_out_elems++; 77 return 0; 78 } 79 80 81 Unique::Unique(qsort_cmp2 comp_func, void * comp_func_fixed_arg, 82 uint size_arg, size_t max_in_memory_size_arg, 83 uint min_dupl_count_arg) 84 :max_in_memory_size(max_in_memory_size_arg), 85 size(size_arg), 86 elements(0) 87 { 88 my_b_clear(&file); 89 min_dupl_count= min_dupl_count_arg; 90 full_size= size; 91 if (min_dupl_count_arg) 92 full_size+= sizeof(element_count); 93 with_counters= MY_TEST(min_dupl_count_arg); 94 init_tree(&tree, (max_in_memory_size / 16), 0, size, comp_func, 95 NULL, comp_func_fixed_arg, MYF(MY_THREAD_SPECIFIC)); 96 /* If the following fail's the next add will also fail */ 97 my_init_dynamic_array(&file_ptrs, sizeof(BUFFPEK), 16, 16, 98 MYF(MY_THREAD_SPECIFIC)); 99 /* 100 If you change the following, change it in get_max_elements function, too. 101 */ 102 max_elements= (ulong) (max_in_memory_size / 103 ALIGN_SIZE(sizeof(TREE_ELEMENT)+size)); 104 if (!max_elements) 105 max_elements= 1; 106 107 (void) open_cached_file(&file, mysql_tmpdir,TEMP_PREFIX, DISK_BUFFER_SIZE, 108 MYF(MY_WME)); 109 } 110 111 112 /* 113 Calculate log2(n!) 114 115 NOTES 116 Stirling's approximate formula is used: 117 118 n! ~= sqrt(2*M_PI*n) * (n/M_E)^n 119 120 Derivation of formula used for calculations is as follows: 121 122 log2(n!) = log(n!)/log(2) = log(sqrt(2*M_PI*n)*(n/M_E)^n) / log(2) = 123 124 = (log(2*M_PI*n)/2 + n*log(n/M_E)) / log(2). 125 */ 126 127 inline double log2_n_fact(double x) 128 { 129 return (log(2*M_PI*x)/2 + x*log(x/M_E)) / M_LN2; 130 } 131 132 133 /* 134 Calculate cost of merge_buffers function call for given sequence of 135 input stream lengths and store the number of rows in result stream in *last. 136 137 SYNOPSIS 138 get_merge_buffers_cost() 139 buff_elems Array of #s of elements in buffers 140 elem_size Size of element stored in buffer 141 first Pointer to first merged element size 142 last Pointer to last merged element size 143 144 RETURN 145 Cost of merge_buffers operation in disk seeks. 146 147 NOTES 148 It is assumed that no rows are eliminated during merge. 149 The cost is calculated as 150 151 cost(read_and_write) + cost(merge_comparisons). 152 153 All bytes in the sequences is read and written back during merge so cost 154 of disk io is 2*elem_size*total_buf_elems/IO_SIZE (2 is for read + write) 155 156 For comparisons cost calculations we assume that all merged sequences have 157 the same length, so each of total_buf_size elements will be added to a sort 158 heap with (n_buffers-1) elements. This gives the comparison cost: 159 160 total_buf_elems* log2(n_buffers) / TIME_FOR_COMPARE_ROWID; 161 */ 162 163 static double get_merge_buffers_cost(uint *buff_elems, uint elem_size, 164 uint *first, uint *last, 165 uint compare_factor) 166 { 167 uint total_buf_elems= 0; 168 for (uint *pbuf= first; pbuf <= last; pbuf++) 169 total_buf_elems+= *pbuf; 170 *last= total_buf_elems; 171 172 size_t n_buffers= last - first + 1; 173 174 /* Using log2(n)=log(n)/log(2) formula */ 175 return 2*((double)total_buf_elems*elem_size) / IO_SIZE + 176 total_buf_elems*log((double) n_buffers) / (compare_factor * M_LN2); 177 } 178 179 180 /* 181 Calculate cost of merging buffers into one in Unique::get, i.e. calculate 182 how long (in terms of disk seeks) the two calls 183 merge_many_buffs(...); 184 merge_buffers(...); 185 will take. 186 187 SYNOPSIS 188 get_merge_many_buffs_cost() 189 buffer buffer space for temporary data, at least 190 Unique::get_cost_calc_buff_size bytes 191 maxbuffer # of full buffers 192 max_n_elems # of elements in first maxbuffer buffers 193 last_n_elems # of elements in last buffer 194 elem_size size of buffer element 195 196 NOTES 197 maxbuffer+1 buffers are merged, where first maxbuffer buffers contain 198 max_n_elems elements each and last buffer contains last_n_elems elements. 199 200 The current implementation does a dumb simulation of merge_many_buffs 201 function actions. 202 203 RETURN 204 Cost of merge in disk seeks. 205 */ 206 207 static double get_merge_many_buffs_cost(uint *buffer, 208 uint maxbuffer, uint max_n_elems, 209 uint last_n_elems, int elem_size, 210 uint compare_factor) 211 { 212 int i; 213 double total_cost= 0.0; 214 uint *buff_elems= buffer; /* #s of elements in each of merged sequences */ 215 216 /* 217 Set initial state: first maxbuffer sequences contain max_n_elems elements 218 each, last sequence contains last_n_elems elements. 219 */ 220 for (i = 0; i < (int)maxbuffer; i++) 221 buff_elems[i]= max_n_elems; 222 buff_elems[maxbuffer]= last_n_elems; 223 224 /* 225 Do it exactly as merge_many_buff function does, calling 226 get_merge_buffers_cost to get cost of merge_buffers. 227 */ 228 if (maxbuffer >= MERGEBUFF2) 229 { 230 while (maxbuffer >= MERGEBUFF2) 231 { 232 uint lastbuff= 0; 233 for (i = 0; i <= (int) maxbuffer - MERGEBUFF*3/2; i += MERGEBUFF) 234 { 235 total_cost+=get_merge_buffers_cost(buff_elems, elem_size, 236 buff_elems + i, 237 buff_elems + i + MERGEBUFF-1, 238 compare_factor); 239 lastbuff++; 240 } 241 total_cost+=get_merge_buffers_cost(buff_elems, elem_size, 242 buff_elems + i, 243 buff_elems + maxbuffer, 244 compare_factor); 245 maxbuffer= lastbuff; 246 } 247 } 248 249 /* Simulate final merge_buff call. */ 250 total_cost += get_merge_buffers_cost(buff_elems, elem_size, 251 buff_elems, buff_elems + maxbuffer, 252 compare_factor); 253 return total_cost; 254 } 255 256 257 /* 258 Calculate cost of using Unique for processing nkeys elements of size 259 key_size using max_in_memory_size memory. 260 261 SYNOPSIS 262 Unique::get_use_cost() 263 buffer space for temporary data, use Unique::get_cost_calc_buff_size 264 to get # bytes needed. 265 nkeys #of elements in Unique 266 key_size size of each elements in bytes 267 max_in_memory_size amount of memory Unique will be allowed to use 268 compare_factor used to calculate cost of one comparison 269 write_fl if the result must be saved written to disk 270 in_memory_elems OUT estimate of the number of elements in memory 271 if disk is not used 272 273 RETURN 274 Cost in disk seeks. 275 276 NOTES 277 cost(using_unqiue) = 278 cost(create_trees) + (see #1) 279 cost(merge) + (see #2) 280 cost(read_result) (see #3) 281 282 1. Cost of trees creation 283 For each Unique::put operation there will be 2*log2(n+1) elements 284 comparisons, where n runs from 1 tree_size (we assume that all added 285 elements are different). Together this gives: 286 287 n_compares = 2*(log2(2) + log2(3) + ... + log2(N+1)) = 2*log2((N+1)!) 288 289 then cost(tree_creation) = n_compares*ROWID_COMPARE_COST; 290 291 Total cost of creating trees: 292 (n_trees - 1)*max_size_tree_cost + non_max_size_tree_cost. 293 294 Approximate value of log2(N!) is calculated by log2_n_fact function. 295 296 2. Cost of merging. 297 If only one tree is created by Unique no merging will be necessary. 298 Otherwise, we model execution of merge_many_buff function and count 299 #of merges. (The reason behind this is that number of buffers is small, 300 while size of buffers is big and we don't want to loose precision with 301 O(x)-style formula) 302 303 3. If only one tree is created by Unique no disk io will happen. 304 Otherwise, ceil(key_len*n_keys) disk seeks are necessary. We assume 305 these will be random seeks. 306 */ 307 308 double Unique::get_use_cost(uint *buffer, size_t nkeys, uint key_size, 309 size_t max_in_memory_size, 310 uint compare_factor, 311 bool intersect_fl, bool *in_memory) 312 { 313 size_t max_elements_in_tree; 314 size_t last_tree_elems; 315 size_t n_full_trees; /* number of trees in unique - 1 */ 316 double result; 317 318 max_elements_in_tree= ((size_t) max_in_memory_size / 319 ALIGN_SIZE(sizeof(TREE_ELEMENT)+key_size)); 320 321 if (max_elements_in_tree == 0) 322 max_elements_in_tree= 1; 323 324 n_full_trees= nkeys / max_elements_in_tree; 325 last_tree_elems= nkeys % max_elements_in_tree; 326 327 /* Calculate cost of creating trees */ 328 result= 2*log2_n_fact(last_tree_elems + 1.0); 329 if (n_full_trees) 330 result+= n_full_trees * log2_n_fact(max_elements_in_tree + 1.0); 331 result /= compare_factor; 332 333 DBUG_PRINT("info",("unique trees sizes: %u=%u*%u + %u", (uint)nkeys, 334 (uint)n_full_trees, 335 (uint)(n_full_trees?max_elements_in_tree:0), 336 (uint)last_tree_elems)); 337 338 if (in_memory) 339 *in_memory= !n_full_trees; 340 341 if (!n_full_trees) 342 return result; 343 344 /* 345 There is more then one tree and merging is necessary. 346 First, add cost of writing all trees to disk, assuming that all disk 347 writes are sequential. 348 */ 349 result += DISK_SEEK_BASE_COST * n_full_trees * 350 ceil(((double) key_size)*max_elements_in_tree / IO_SIZE); 351 result += DISK_SEEK_BASE_COST * ceil(((double) key_size)*last_tree_elems / IO_SIZE); 352 353 /* Cost of merge */ 354 if (intersect_fl) 355 key_size+= sizeof(element_count); 356 double merge_cost= get_merge_many_buffs_cost(buffer, (uint)n_full_trees, 357 (uint)max_elements_in_tree, 358 (uint)last_tree_elems, key_size, 359 compare_factor); 360 result += merge_cost; 361 /* 362 Add cost of reading the resulting sequence, assuming there were no 363 duplicate elements. 364 */ 365 result += ceil((double)key_size*nkeys/IO_SIZE); 366 367 return result; 368 } 369 370 Unique::~Unique() 371 { 372 close_cached_file(&file); 373 delete_tree(&tree, 0); 374 delete_dynamic(&file_ptrs); 375 } 376 377 378 /* Write tree to disk; clear tree */ 379 bool Unique::flush() 380 { 381 BUFFPEK file_ptr; 382 elements+= tree.elements_in_tree; 383 file_ptr.count=tree.elements_in_tree; 384 file_ptr.file_pos=my_b_tell(&file); 385 386 tree_walk_action action= min_dupl_count ? 387 (tree_walk_action) unique_write_to_file_with_count : 388 (tree_walk_action) unique_write_to_file; 389 if (tree_walk(&tree, action, 390 (void*) this, left_root_right) || 391 insert_dynamic(&file_ptrs, (uchar*) &file_ptr)) 392 return 1; 393 delete_tree(&tree, 0); 394 return 0; 395 } 396 397 398 /* 399 Clear the tree and the file. 400 You must call reset() if you want to reuse Unique after walk(). 401 */ 402 403 void 404 Unique::reset() 405 { 406 reset_tree(&tree); 407 /* 408 If elements != 0, some trees were stored in the file (see how 409 flush() works). Note, that we can not count on my_b_tell(&file) == 0 410 here, because it can return 0 right after walk(), and walk() does not 411 reset any Unique member. 412 */ 413 if (elements) 414 { 415 reset_dynamic(&file_ptrs); 416 reinit_io_cache(&file, WRITE_CACHE, 0L, 0, 1); 417 } 418 my_free(sort.record_pointers); 419 elements= 0; 420 tree.flag= 0; 421 sort.record_pointers= 0; 422 } 423 424 /* 425 The comparison function, passed to queue_init() in merge_walk() and in 426 merge_buffers() when the latter is called from Uniques::get() must 427 use comparison function of Uniques::tree, but compare members of struct 428 BUFFPEK. 429 */ 430 431 C_MODE_START 432 433 static int buffpek_compare(void *arg, uchar *key_ptr1, uchar *key_ptr2) 434 { 435 BUFFPEK_COMPARE_CONTEXT *ctx= (BUFFPEK_COMPARE_CONTEXT *) arg; 436 return ctx->key_compare(ctx->key_compare_arg, 437 *((uchar **) key_ptr1), *((uchar **)key_ptr2)); 438 } 439 440 C_MODE_END 441 442 443 inline 444 element_count get_counter_from_merged_element(void *ptr, uint ofs) 445 { 446 element_count cnt; 447 memcpy((uchar *) &cnt, (uchar *) ptr + ofs, sizeof(element_count)); 448 return cnt; 449 } 450 451 452 inline 453 void put_counter_into_merged_element(void *ptr, uint ofs, element_count cnt) 454 { 455 memcpy((uchar *) ptr + ofs, (uchar *) &cnt, sizeof(element_count)); 456 } 457 458 459 /* 460 DESCRIPTION 461 462 Function is very similar to merge_buffers, but instead of writing sorted 463 unique keys to the output file, it invokes walk_action for each key. 464 This saves I/O if you need to pass through all unique keys only once. 465 466 SYNOPSIS 467 merge_walk() 468 All params are 'IN' (but see comment for begin, end): 469 merge_buffer buffer to perform cached piece-by-piece loading 470 of trees; initially the buffer is empty 471 merge_buffer_size size of merge_buffer. Must be aligned with 472 key_length 473 key_length size of tree element; key_length * (end - begin) 474 must be less or equal than merge_buffer_size. 475 begin pointer to BUFFPEK struct for the first tree. 476 end pointer to BUFFPEK struct for the last tree; 477 end > begin and [begin, end) form a consecutive 478 range. BUFFPEKs structs in that range are used and 479 overwritten in merge_walk(). 480 walk_action element visitor. Action is called for each unique 481 key. 482 walk_action_arg argument to walk action. Passed to it on each call. 483 compare elements comparison function 484 compare_arg comparison function argument 485 file file with all trees dumped. Trees in the file 486 must contain sorted unique values. Cache must be 487 initialized in read mode. 488 with counters take into account counters for equal merged 489 elements 490 RETURN VALUE 491 0 ok 492 <> 0 error 493 */ 494 495 static bool merge_walk(uchar *merge_buffer, size_t merge_buffer_size, 496 uint key_length, BUFFPEK *begin, BUFFPEK *end, 497 tree_walk_action walk_action, void *walk_action_arg, 498 qsort_cmp2 compare, void *compare_arg, 499 IO_CACHE *file, bool with_counters) 500 { 501 BUFFPEK_COMPARE_CONTEXT compare_context = { compare, compare_arg }; 502 QUEUE queue; 503 if (end <= begin || 504 merge_buffer_size < (size_t) (key_length * (end - begin + 1)) || 505 init_queue(&queue, (uint) (end - begin), offsetof(BUFFPEK, key), 0, 506 buffpek_compare, &compare_context, 0, 0)) 507 return 1; 508 /* we need space for one key when a piece of merge buffer is re-read */ 509 merge_buffer_size-= key_length; 510 uchar *save_key_buff= merge_buffer + merge_buffer_size; 511 uint max_key_count_per_piece= (uint) (merge_buffer_size/(end-begin) / 512 key_length); 513 /* if piece_size is aligned reuse_freed_buffer will always hit */ 514 uint piece_size= max_key_count_per_piece * key_length; 515 ulong bytes_read; /* to hold return value of read_to_buffer */ 516 BUFFPEK *top; 517 int res= 1; 518 uint cnt_ofs= key_length - (with_counters ? sizeof(element_count) : 0); 519 element_count cnt; 520 /* 521 Invariant: queue must contain top element from each tree, until a tree 522 is not completely walked through. 523 Here we're forcing the invariant, inserting one element from each tree 524 to the queue. 525 */ 526 for (top= begin; top != end; ++top) 527 { 528 top->base= merge_buffer + (top - begin) * piece_size; 529 top->max_keys= max_key_count_per_piece; 530 bytes_read= read_to_buffer(file, top, key_length); 531 if (unlikely(bytes_read == (ulong) -1)) 532 goto end; 533 DBUG_ASSERT(bytes_read); 534 queue_insert(&queue, (uchar *) top); 535 } 536 top= (BUFFPEK *) queue_top(&queue); 537 while (queue.elements > 1) 538 { 539 /* 540 Every iteration one element is removed from the queue, and one is 541 inserted by the rules of the invariant. If two adjacent elements on 542 the top of the queue are not equal, biggest one is unique, because all 543 elements in each tree are unique. Action is applied only to unique 544 elements. 545 */ 546 void *old_key= top->key; 547 /* 548 read next key from the cache or from the file and push it to the 549 queue; this gives new top. 550 */ 551 top->key+= key_length; 552 if (--top->mem_count) 553 queue_replace_top(&queue); 554 else /* next piece should be read */ 555 { 556 /* save old_key not to overwrite it in read_to_buffer */ 557 memcpy(save_key_buff, old_key, key_length); 558 old_key= save_key_buff; 559 bytes_read= read_to_buffer(file, top, key_length); 560 if (unlikely(bytes_read == (ulong) -1)) 561 goto end; 562 else if (bytes_read) /* top->key, top->mem_count are reset */ 563 queue_replace_top(&queue); /* in read_to_buffer */ 564 else 565 { 566 /* 567 Tree for old 'top' element is empty: remove it from the queue and 568 give all its memory to the nearest tree. 569 */ 570 queue_remove_top(&queue); 571 reuse_freed_buff(&queue, top, key_length); 572 } 573 } 574 top= (BUFFPEK *) queue_top(&queue); 575 /* new top has been obtained; if old top is unique, apply the action */ 576 if (compare(compare_arg, old_key, top->key)) 577 { 578 cnt= with_counters ? 579 get_counter_from_merged_element(old_key, cnt_ofs) : 1; 580 if (walk_action(old_key, cnt, walk_action_arg)) 581 goto end; 582 } 583 else if (with_counters) 584 { 585 cnt= get_counter_from_merged_element(top->key, cnt_ofs); 586 cnt+= get_counter_from_merged_element(old_key, cnt_ofs); 587 put_counter_into_merged_element(top->key, cnt_ofs, cnt); 588 } 589 } 590 /* 591 Applying walk_action to the tail of the last tree: this is safe because 592 either we had only one tree in the beginning, either we work with the 593 last tree in the queue. 594 */ 595 do 596 { 597 do 598 { 599 600 cnt= with_counters ? 601 get_counter_from_merged_element(top->key, cnt_ofs) : 1; 602 if (walk_action(top->key, cnt, walk_action_arg)) 603 goto end; 604 top->key+= key_length; 605 } 606 while (--top->mem_count); 607 bytes_read= read_to_buffer(file, top, key_length); 608 if (unlikely(bytes_read == (ulong) -1)) 609 goto end; 610 } 611 while (bytes_read); 612 res= 0; 613 end: 614 delete_queue(&queue); 615 return res; 616 } 617 618 619 /* 620 DESCRIPTION 621 Walks consecutively through all unique elements: 622 if all elements are in memory, then it simply invokes 'tree_walk', else 623 all flushed trees are loaded to memory piece-by-piece, pieces are 624 sorted, and action is called for each unique value. 625 Note: so as merging resets file_ptrs state, this method can change 626 internal Unique state to undefined: if you want to reuse Unique after 627 walk() you must call reset() first! 628 SYNOPSIS 629 Unique:walk() 630 All params are 'IN': 631 table parameter for the call of the merge method 632 action function-visitor, typed in include/my_tree.h 633 function is called for each unique element 634 arg argument for visitor, which is passed to it on each call 635 RETURN VALUE 636 0 OK 637 <> 0 error 638 */ 639 640 bool Unique::walk(TABLE *table, tree_walk_action action, void *walk_action_arg) 641 { 642 int res= 0; 643 uchar *merge_buffer; 644 645 if (elements == 0) /* the whole tree is in memory */ 646 return tree_walk(&tree, action, walk_action_arg, left_root_right); 647 648 sort.return_rows= elements+tree.elements_in_tree; 649 /* flush current tree to the file to have some memory for merge buffer */ 650 if (flush()) 651 return 1; 652 if (flush_io_cache(&file) || reinit_io_cache(&file, READ_CACHE, 0L, 0, 0)) 653 return 1; 654 /* 655 merge_buffer must fit at least MERGEBUFF2 + 1 keys, because 656 merge_index() can merge that many BUFFPEKs at once. The extra space for one key 657 is needed when a piece of merge buffer is re-read, see merge_walk() 658 */ 659 size_t buff_sz= MY_MAX(MERGEBUFF2+1, max_in_memory_size/full_size+1) * full_size; 660 if (!(merge_buffer = (uchar *)my_malloc(buff_sz, MYF(MY_WME)))) 661 return 1; 662 if (buff_sz < full_size * (file_ptrs.elements + 1UL)) 663 res= merge(table, merge_buffer, buff_sz >= full_size * MERGEBUFF2) ; 664 665 if (!res) 666 { 667 res= merge_walk(merge_buffer, buff_sz, full_size, 668 (BUFFPEK *) file_ptrs.buffer, 669 (BUFFPEK *) file_ptrs.buffer + file_ptrs.elements, 670 action, walk_action_arg, 671 tree.compare, tree.custom_arg, &file, with_counters); 672 } 673 my_free(merge_buffer); 674 return res; 675 } 676 677 678 /* 679 DESCRIPTION 680 681 Perform multi-pass sort merge of the elements using the buffer buff as 682 the merge buffer. The last pass is not performed if without_last_merge is 683 TRUE. 684 685 SYNOPSIS 686 Unique:merge() 687 All params are 'IN': 688 table the parameter to access sort context 689 buff merge buffer 690 without_last_merge TRUE <=> do not perform the last merge 691 RETURN VALUE 692 0 OK 693 <> 0 error 694 */ 695 696 bool Unique::merge(TABLE *table, uchar *buff, bool without_last_merge) 697 { 698 IO_CACHE *outfile= &sort.io_cache; 699 BUFFPEK *file_ptr= (BUFFPEK*) file_ptrs.buffer; 700 uint maxbuffer= file_ptrs.elements - 1; 701 my_off_t save_pos; 702 bool error= 1; 703 Sort_param sort_param; 704 705 /* Open cached file for table records if it isn't open */ 706 if (! my_b_inited(outfile) && 707 open_cached_file(outfile,mysql_tmpdir,TEMP_PREFIX,READ_RECORD_BUFFER, 708 MYF(MY_WME))) 709 return 1; 710 711 bzero((char*) &sort_param,sizeof(sort_param)); 712 sort_param.max_rows= elements; 713 sort_param.sort_form= table; 714 sort_param.rec_length= sort_param.sort_length= sort_param.ref_length= 715 full_size; 716 sort_param.min_dupl_count= min_dupl_count; 717 sort_param.res_length= 0; 718 sort_param.max_keys_per_buffer= 719 (uint) MY_MAX((max_in_memory_size / sort_param.sort_length), MERGEBUFF2); 720 sort_param.not_killable= 1; 721 722 sort_param.unique_buff= buff +(sort_param.max_keys_per_buffer * 723 sort_param.sort_length); 724 725 sort_param.compare= (qsort2_cmp) buffpek_compare; 726 sort_param.cmp_context.key_compare= tree.compare; 727 sort_param.cmp_context.key_compare_arg= tree.custom_arg; 728 729 /* Merge the buffers to one file, removing duplicates */ 730 if (merge_many_buff(&sort_param,buff,file_ptr,&maxbuffer,&file)) 731 goto err; 732 if (flush_io_cache(&file) || 733 reinit_io_cache(&file,READ_CACHE,0L,0,0)) 734 goto err; 735 sort_param.res_length= sort_param.rec_length- 736 (min_dupl_count ? sizeof(min_dupl_count) : 0); 737 if (without_last_merge) 738 { 739 file_ptrs.elements= maxbuffer+1; 740 return 0; 741 } 742 if (merge_index(&sort_param, buff, file_ptr, maxbuffer, &file, outfile)) 743 goto err; 744 error= 0; 745 err: 746 if (flush_io_cache(outfile)) 747 error= 1; 748 749 /* Setup io_cache for reading */ 750 save_pos= outfile->pos_in_file; 751 if (reinit_io_cache(outfile,READ_CACHE,0L,0,0)) 752 error= 1; 753 outfile->end_of_file=save_pos; 754 return error; 755 } 756 757 758 /* 759 Allocate memory that can be used with init_records() so that 760 rows will be read in priority order. 761 */ 762 763 bool Unique::get(TABLE *table) 764 { 765 bool rc= 1; 766 uchar *sort_buffer= NULL; 767 sort.return_rows= elements+tree.elements_in_tree; 768 DBUG_ENTER("Unique::get"); 769 770 if (my_b_tell(&file) == 0) 771 { 772 /* Whole tree is in memory; Don't use disk if you don't need to */ 773 if ((sort.record_pointers= (uchar*) 774 my_malloc(size * tree.elements_in_tree, MYF(MY_THREAD_SPECIFIC)))) 775 { 776 uchar *save_record_pointers= sort.record_pointers; 777 tree_walk_action action= min_dupl_count ? 778 (tree_walk_action) unique_intersect_write_to_ptrs : 779 (tree_walk_action) unique_write_to_ptrs; 780 filtered_out_elems= 0; 781 (void) tree_walk(&tree, action, 782 this, left_root_right); 783 /* Restore record_pointers that was changed in by 'action' above */ 784 sort.record_pointers= save_record_pointers; 785 sort.return_rows-= filtered_out_elems; 786 DBUG_RETURN(0); 787 } 788 } 789 /* Not enough memory; Save the result to file && free memory used by tree */ 790 if (flush()) 791 DBUG_RETURN(1); 792 /* 793 merge_buffer must fit at least MERGEBUFF2 + 1 keys, because 794 merge_index() can merge that many BUFFPEKs at once. The extra space for 795 one key for Sort_param::unique_buff 796 */ 797 size_t buff_sz= MY_MAX(MERGEBUFF2+1, max_in_memory_size/full_size+1) * full_size; 798 if (!(sort_buffer= (uchar*) my_malloc(buff_sz, 799 MYF(MY_THREAD_SPECIFIC|MY_WME)))) 800 DBUG_RETURN(1); 801 802 if (merge(table, sort_buffer, FALSE)) 803 goto err; 804 rc= 0; 805 806 err: 807 my_free(sort_buffer); 808 DBUG_RETURN(rc); 809 } 810