1 | /***************************************************************************** |
2 | |
3 | Copyright (c) 2011, 2017, Oracle and/or its affiliates. All Rights Reserved. |
4 | Copyright (c) 2016, 2018, MariaDB Corporation. |
5 | |
6 | This program is free software; you can redistribute it and/or modify it under |
7 | the terms of the GNU General Public License as published by the Free Software |
8 | Foundation; version 2 of the License. |
9 | |
10 | This program is distributed in the hope that it will be useful, but WITHOUT |
11 | ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS |
12 | FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. |
13 | |
14 | You should have received a copy of the GNU General Public License along with |
15 | this program; if not, write to the Free Software Foundation, Inc., |
16 | 51 Franklin Street, Suite 500, Boston, MA 02110-1335 USA |
17 | |
18 | *****************************************************************************/ |
19 | |
20 | /**************************************************//** |
21 | @file fts/fts0fts.cc |
22 | Full Text Search interface |
23 | ***********************************************************************/ |
24 | |
25 | #include "ha_prototypes.h" |
26 | |
27 | #include "trx0roll.h" |
28 | #include "row0mysql.h" |
29 | #include "row0upd.h" |
30 | #include "dict0types.h" |
31 | #include "dict0stats_bg.h" |
32 | #include "row0sel.h" |
33 | #include "fts0fts.h" |
34 | #include "fts0priv.h" |
35 | #include "fts0types.h" |
36 | #include "fts0types.ic" |
37 | #include "fts0vlc.ic" |
38 | #include "fts0plugin.h" |
39 | #include "dict0priv.h" |
40 | #include "dict0stats.h" |
41 | #include "btr0pcur.h" |
42 | #include "sync0sync.h" |
43 | #include "ut0new.h" |
44 | |
45 | static const ulint FTS_MAX_ID_LEN = 32; |
46 | |
47 | /** Column name from the FTS config table */ |
48 | #define FTS_MAX_CACHE_SIZE_IN_MB "cache_size_in_mb" |
49 | |
50 | /** Verify if a aux table name is a obsolete table |
51 | by looking up the key word in the obsolete table names */ |
52 | #define FTS_IS_OBSOLETE_AUX_TABLE(table_name) \ |
53 | (strstr((table_name), "DOC_ID") != NULL \ |
54 | || strstr((table_name), "ADDED") != NULL \ |
55 | || strstr((table_name), "STOPWORDS") != NULL) |
56 | |
57 | /** This is maximum FTS cache for each table and would be |
58 | a configurable variable */ |
59 | ulong fts_max_cache_size; |
60 | |
61 | /** Whether the total memory used for FTS cache is exhausted, and we will |
62 | need a sync to free some memory */ |
63 | bool fts_need_sync = false; |
64 | |
65 | /** Variable specifying the total memory allocated for FTS cache */ |
66 | ulong fts_max_total_cache_size; |
67 | |
68 | /** This is FTS result cache limit for each query and would be |
69 | a configurable variable */ |
70 | ulong fts_result_cache_limit; |
71 | |
72 | /** Variable specifying the maximum FTS max token size */ |
73 | ulong fts_max_token_size; |
74 | |
75 | /** Variable specifying the minimum FTS max token size */ |
76 | ulong fts_min_token_size; |
77 | |
78 | |
79 | // FIXME: testing |
80 | static ib_time_t elapsed_time = 0; |
81 | static ulint n_nodes = 0; |
82 | |
83 | #ifdef FTS_CACHE_SIZE_DEBUG |
84 | /** The cache size permissible lower limit (1K) */ |
85 | static const ulint FTS_CACHE_SIZE_LOWER_LIMIT_IN_MB = 1; |
86 | |
87 | /** The cache size permissible upper limit (1G) */ |
88 | static const ulint FTS_CACHE_SIZE_UPPER_LIMIT_IN_MB = 1024; |
89 | #endif |
90 | |
91 | /** Time to sleep after DEADLOCK error before retrying operation. */ |
92 | static const ulint FTS_DEADLOCK_RETRY_WAIT = 100000; |
93 | |
94 | /** variable to record innodb_fts_internal_tbl_name for information |
95 | schema table INNODB_FTS_INSERTED etc. */ |
96 | char* fts_internal_tbl_name = NULL; |
97 | char* fts_internal_tbl_name2 = NULL; |
98 | |
99 | /** InnoDB default stopword list: |
100 | There are different versions of stopwords, the stop words listed |
101 | below comes from "Google Stopword" list. Reference: |
102 | http://meta.wikimedia.org/wiki/Stop_word_list/google_stop_word_list. |
103 | The final version of InnoDB default stopword list is still pending |
104 | for decision */ |
105 | const char *fts_default_stopword[] = |
106 | { |
107 | "a" , |
108 | "about" , |
109 | "an" , |
110 | "are" , |
111 | "as" , |
112 | "at" , |
113 | "be" , |
114 | "by" , |
115 | "com" , |
116 | "de" , |
117 | "en" , |
118 | "for" , |
119 | "from" , |
120 | "how" , |
121 | "i" , |
122 | "in" , |
123 | "is" , |
124 | "it" , |
125 | "la" , |
126 | "of" , |
127 | "on" , |
128 | "or" , |
129 | "that" , |
130 | "the" , |
131 | "this" , |
132 | "to" , |
133 | "was" , |
134 | "what" , |
135 | "when" , |
136 | "where" , |
137 | "who" , |
138 | "will" , |
139 | "with" , |
140 | "und" , |
141 | "the" , |
142 | "www" , |
143 | NULL |
144 | }; |
145 | |
146 | /** For storing table info when checking for orphaned tables. */ |
147 | struct fts_aux_table_t { |
148 | table_id_t id; /*!< Table id */ |
149 | table_id_t parent_id; /*!< Parent table id */ |
150 | table_id_t index_id; /*!< Table FT index id */ |
151 | char* name; /*!< Name of the table */ |
152 | }; |
153 | |
154 | /** FTS auxiliary table suffixes that are common to all FT indexes. */ |
155 | const char* fts_common_tables[] = { |
156 | "BEING_DELETED" , |
157 | "BEING_DELETED_CACHE" , |
158 | "CONFIG" , |
159 | "DELETED" , |
160 | "DELETED_CACHE" , |
161 | NULL |
162 | }; |
163 | |
164 | /** FTS auxiliary INDEX split intervals. */ |
165 | const fts_index_selector_t fts_index_selector[] = { |
166 | { 9, "INDEX_1" }, |
167 | { 65, "INDEX_2" }, |
168 | { 70, "INDEX_3" }, |
169 | { 75, "INDEX_4" }, |
170 | { 80, "INDEX_5" }, |
171 | { 85, "INDEX_6" }, |
172 | { 0 , NULL } |
173 | }; |
174 | |
175 | /** Default config values for FTS indexes on a table. */ |
176 | static const char* fts_config_table_insert_values_sql = |
177 | "BEGIN\n" |
178 | "\n" |
179 | "INSERT INTO $config_table VALUES('" |
180 | FTS_MAX_CACHE_SIZE_IN_MB "', '256');\n" |
181 | "" |
182 | "INSERT INTO $config_table VALUES('" |
183 | FTS_OPTIMIZE_LIMIT_IN_SECS "', '180');\n" |
184 | "" |
185 | "INSERT INTO $config_table VALUES ('" |
186 | FTS_SYNCED_DOC_ID "', '0');\n" |
187 | "" |
188 | "INSERT INTO $config_table VALUES ('" |
189 | FTS_TOTAL_DELETED_COUNT "', '0');\n" |
190 | "" /* Note: 0 == FTS_TABLE_STATE_RUNNING */ |
191 | "INSERT INTO $config_table VALUES ('" |
192 | FTS_TABLE_STATE "', '0');\n" ; |
193 | |
194 | /** FTS tokenize parmameter for plugin parser */ |
195 | struct fts_tokenize_param_t { |
196 | fts_doc_t* result_doc; /*!< Result doc for tokens */ |
197 | ulint add_pos; /*!< Added position for tokens */ |
198 | }; |
199 | |
200 | /** Run SYNC on the table, i.e., write out data from the cache to the |
201 | FTS auxiliary INDEX table and clear the cache at the end. |
202 | @param[in,out] sync sync state |
203 | @param[in] unlock_cache whether unlock cache lock when write node |
204 | @param[in] wait whether wait when a sync is in progress |
205 | @param[in] has_dict whether has dict operation lock |
206 | @return DB_SUCCESS if all OK */ |
207 | static |
208 | dberr_t |
209 | fts_sync( |
210 | fts_sync_t* sync, |
211 | bool unlock_cache, |
212 | bool wait, |
213 | bool has_dict); |
214 | |
215 | /****************************************************************//** |
216 | Release all resources help by the words rb tree e.g., the node ilist. */ |
217 | static |
218 | void |
219 | fts_words_free( |
220 | /*===========*/ |
221 | ib_rbt_t* words) /*!< in: rb tree of words */ |
222 | MY_ATTRIBUTE((nonnull)); |
223 | #ifdef FTS_CACHE_SIZE_DEBUG |
224 | /****************************************************************//** |
225 | Read the max cache size parameter from the config table. */ |
226 | static |
227 | void |
228 | fts_update_max_cache_size( |
229 | /*======================*/ |
230 | fts_sync_t* sync); /*!< in: sync state */ |
231 | #endif |
232 | |
233 | /*********************************************************************//** |
234 | This function fetches the document just inserted right before |
235 | we commit the transaction, and tokenize the inserted text data |
236 | and insert into FTS auxiliary table and its cache. |
237 | @return TRUE if successful */ |
238 | static |
239 | ulint |
240 | fts_add_doc_by_id( |
241 | /*==============*/ |
242 | fts_trx_table_t*ftt, /*!< in: FTS trx table */ |
243 | doc_id_t doc_id, /*!< in: doc id */ |
244 | ib_vector_t* fts_indexes MY_ATTRIBUTE((unused))); |
245 | /*!< in: affected fts indexes */ |
246 | /******************************************************************//** |
247 | Update the last document id. This function could create a new |
248 | transaction to update the last document id. |
249 | @return DB_SUCCESS if OK */ |
250 | static |
251 | dberr_t |
252 | fts_update_sync_doc_id( |
253 | /*===================*/ |
254 | const dict_table_t* table, /*!< in: table */ |
255 | const char* table_name, /*!< in: table name, or NULL */ |
256 | doc_id_t doc_id, /*!< in: last document id */ |
257 | trx_t* trx) /*!< in: update trx, or NULL */ |
258 | MY_ATTRIBUTE((nonnull(1))); |
259 | |
260 | /** Tokenize a document. |
261 | @param[in,out] doc document to tokenize |
262 | @param[out] result tokenization result |
263 | @param[in] parser pluggable parser */ |
264 | static |
265 | void |
266 | fts_tokenize_document( |
267 | fts_doc_t* doc, |
268 | fts_doc_t* result, |
269 | st_mysql_ftparser* parser); |
270 | |
271 | /** Continue to tokenize a document. |
272 | @param[in,out] doc document to tokenize |
273 | @param[in] add_pos add this position to all tokens from this tokenization |
274 | @param[out] result tokenization result |
275 | @param[in] parser pluggable parser */ |
276 | static |
277 | void |
278 | fts_tokenize_document_next( |
279 | fts_doc_t* doc, |
280 | ulint add_pos, |
281 | fts_doc_t* result, |
282 | st_mysql_ftparser* parser); |
283 | |
284 | /** Create the vector of fts_get_doc_t instances. |
285 | @param[in,out] cache fts cache |
286 | @return vector of fts_get_doc_t instances */ |
287 | static |
288 | ib_vector_t* |
289 | fts_get_docs_create( |
290 | fts_cache_t* cache); |
291 | |
292 | /** Free the FTS cache. |
293 | @param[in,out] cache to be freed */ |
294 | static |
295 | void |
296 | fts_cache_destroy(fts_cache_t* cache) |
297 | { |
298 | rw_lock_free(&cache->lock); |
299 | rw_lock_free(&cache->init_lock); |
300 | mutex_free(&cache->optimize_lock); |
301 | mutex_free(&cache->deleted_lock); |
302 | mutex_free(&cache->doc_id_lock); |
303 | os_event_destroy(cache->sync->event); |
304 | |
305 | if (cache->stopword_info.cached_stopword) { |
306 | rbt_free(cache->stopword_info.cached_stopword); |
307 | } |
308 | |
309 | if (cache->sync_heap->arg) { |
310 | mem_heap_free(static_cast<mem_heap_t*>(cache->sync_heap->arg)); |
311 | } |
312 | |
313 | mem_heap_free(cache->cache_heap); |
314 | } |
315 | |
316 | /** Get a character set based on precise type. |
317 | @param prtype precise type |
318 | @return the corresponding character set */ |
319 | UNIV_INLINE |
320 | CHARSET_INFO* |
321 | fts_get_charset(ulint prtype) |
322 | { |
323 | #ifdef UNIV_DEBUG |
324 | switch (prtype & DATA_MYSQL_TYPE_MASK) { |
325 | case MYSQL_TYPE_BIT: |
326 | case MYSQL_TYPE_STRING: |
327 | case MYSQL_TYPE_VAR_STRING: |
328 | case MYSQL_TYPE_TINY_BLOB: |
329 | case MYSQL_TYPE_MEDIUM_BLOB: |
330 | case MYSQL_TYPE_BLOB: |
331 | case MYSQL_TYPE_LONG_BLOB: |
332 | case MYSQL_TYPE_VARCHAR: |
333 | break; |
334 | default: |
335 | ut_error; |
336 | } |
337 | #endif /* UNIV_DEBUG */ |
338 | |
339 | uint cs_num = (uint) dtype_get_charset_coll(prtype); |
340 | |
341 | if (CHARSET_INFO* cs = get_charset(cs_num, MYF(MY_WME))) { |
342 | return(cs); |
343 | } |
344 | |
345 | ib::fatal() << "Unable to find charset-collation " << cs_num; |
346 | return(NULL); |
347 | } |
348 | |
349 | /****************************************************************//** |
350 | This function loads the default InnoDB stopword list */ |
351 | static |
352 | void |
353 | fts_load_default_stopword( |
354 | /*======================*/ |
355 | fts_stopword_t* stopword_info) /*!< in: stopword info */ |
356 | { |
357 | fts_string_t str; |
358 | mem_heap_t* heap; |
359 | ib_alloc_t* allocator; |
360 | ib_rbt_t* stop_words; |
361 | |
362 | allocator = stopword_info->heap; |
363 | heap = static_cast<mem_heap_t*>(allocator->arg); |
364 | |
365 | if (!stopword_info->cached_stopword) { |
366 | stopword_info->cached_stopword = rbt_create_arg_cmp( |
367 | sizeof(fts_tokenizer_word_t), innobase_fts_text_cmp, |
368 | &my_charset_latin1); |
369 | } |
370 | |
371 | stop_words = stopword_info->cached_stopword; |
372 | |
373 | str.f_n_char = 0; |
374 | |
375 | for (ulint i = 0; fts_default_stopword[i]; ++i) { |
376 | char* word; |
377 | fts_tokenizer_word_t new_word; |
378 | |
379 | /* We are going to duplicate the value below. */ |
380 | word = const_cast<char*>(fts_default_stopword[i]); |
381 | |
382 | new_word.nodes = ib_vector_create( |
383 | allocator, sizeof(fts_node_t), 4); |
384 | |
385 | str.f_len = ut_strlen(word); |
386 | str.f_str = reinterpret_cast<byte*>(word); |
387 | |
388 | fts_string_dup(&new_word.text, &str, heap); |
389 | |
390 | rbt_insert(stop_words, &new_word, &new_word); |
391 | } |
392 | |
393 | stopword_info->status = STOPWORD_FROM_DEFAULT; |
394 | } |
395 | |
396 | /****************************************************************//** |
397 | Callback function to read a single stopword value. |
398 | @return Always return TRUE */ |
399 | static |
400 | ibool |
401 | fts_read_stopword( |
402 | /*==============*/ |
403 | void* row, /*!< in: sel_node_t* */ |
404 | void* user_arg) /*!< in: pointer to ib_vector_t */ |
405 | { |
406 | ib_alloc_t* allocator; |
407 | fts_stopword_t* stopword_info; |
408 | sel_node_t* sel_node; |
409 | que_node_t* exp; |
410 | ib_rbt_t* stop_words; |
411 | dfield_t* dfield; |
412 | fts_string_t str; |
413 | mem_heap_t* heap; |
414 | ib_rbt_bound_t parent; |
415 | |
416 | sel_node = static_cast<sel_node_t*>(row); |
417 | stopword_info = static_cast<fts_stopword_t*>(user_arg); |
418 | |
419 | stop_words = stopword_info->cached_stopword; |
420 | allocator = static_cast<ib_alloc_t*>(stopword_info->heap); |
421 | heap = static_cast<mem_heap_t*>(allocator->arg); |
422 | |
423 | exp = sel_node->select_list; |
424 | |
425 | /* We only need to read the first column */ |
426 | dfield = que_node_get_val(exp); |
427 | |
428 | str.f_n_char = 0; |
429 | str.f_str = static_cast<byte*>(dfield_get_data(dfield)); |
430 | str.f_len = dfield_get_len(dfield); |
431 | |
432 | /* Only create new node if it is a value not already existed */ |
433 | if (str.f_len != UNIV_SQL_NULL |
434 | && rbt_search(stop_words, &parent, &str) != 0) { |
435 | |
436 | fts_tokenizer_word_t new_word; |
437 | |
438 | new_word.nodes = ib_vector_create( |
439 | allocator, sizeof(fts_node_t), 4); |
440 | |
441 | new_word.text.f_str = static_cast<byte*>( |
442 | mem_heap_alloc(heap, str.f_len + 1)); |
443 | |
444 | memcpy(new_word.text.f_str, str.f_str, str.f_len); |
445 | |
446 | new_word.text.f_n_char = 0; |
447 | new_word.text.f_len = str.f_len; |
448 | new_word.text.f_str[str.f_len] = 0; |
449 | |
450 | rbt_insert(stop_words, &new_word, &new_word); |
451 | } |
452 | |
453 | return(TRUE); |
454 | } |
455 | |
456 | /******************************************************************//** |
457 | Load user defined stopword from designated user table |
458 | @return TRUE if load operation is successful */ |
459 | static |
460 | ibool |
461 | fts_load_user_stopword( |
462 | /*===================*/ |
463 | fts_t* fts, /*!< in: FTS struct */ |
464 | const char* stopword_table_name, /*!< in: Stopword table |
465 | name */ |
466 | fts_stopword_t* stopword_info) /*!< in: Stopword info */ |
467 | { |
468 | pars_info_t* info; |
469 | que_t* graph; |
470 | dberr_t error = DB_SUCCESS; |
471 | ibool ret = TRUE; |
472 | trx_t* trx; |
473 | ibool has_lock = fts->fts_status & TABLE_DICT_LOCKED; |
474 | |
475 | trx = trx_create(); |
476 | trx->op_info = "Load user stopword table into FTS cache" ; |
477 | |
478 | if (!has_lock) { |
479 | mutex_enter(&dict_sys->mutex); |
480 | } |
481 | |
482 | /* Validate the user table existence and in the right |
483 | format */ |
484 | stopword_info->charset = fts_valid_stopword_table(stopword_table_name); |
485 | if (!stopword_info->charset) { |
486 | ret = FALSE; |
487 | goto cleanup; |
488 | } else if (!stopword_info->cached_stopword) { |
489 | /* Create the stopword RB tree with the stopword column |
490 | charset. All comparison will use this charset */ |
491 | stopword_info->cached_stopword = rbt_create_arg_cmp( |
492 | sizeof(fts_tokenizer_word_t), innobase_fts_text_cmp, |
493 | (void*)stopword_info->charset); |
494 | |
495 | } |
496 | |
497 | info = pars_info_create(); |
498 | |
499 | pars_info_bind_id(info, TRUE, "table_stopword" , stopword_table_name); |
500 | |
501 | pars_info_bind_function(info, "my_func" , fts_read_stopword, |
502 | stopword_info); |
503 | |
504 | graph = fts_parse_sql_no_dict_lock( |
505 | info, |
506 | "DECLARE FUNCTION my_func;\n" |
507 | "DECLARE CURSOR c IS" |
508 | " SELECT value" |
509 | " FROM $table_stopword;\n" |
510 | "BEGIN\n" |
511 | "\n" |
512 | "OPEN c;\n" |
513 | "WHILE 1 = 1 LOOP\n" |
514 | " FETCH c INTO my_func();\n" |
515 | " IF c % NOTFOUND THEN\n" |
516 | " EXIT;\n" |
517 | " END IF;\n" |
518 | "END LOOP;\n" |
519 | "CLOSE c;" ); |
520 | |
521 | for (;;) { |
522 | error = fts_eval_sql(trx, graph); |
523 | |
524 | if (error == DB_SUCCESS) { |
525 | fts_sql_commit(trx); |
526 | stopword_info->status = STOPWORD_USER_TABLE; |
527 | break; |
528 | } else { |
529 | |
530 | fts_sql_rollback(trx); |
531 | |
532 | if (error == DB_LOCK_WAIT_TIMEOUT) { |
533 | ib::warn() << "Lock wait timeout reading user" |
534 | " stopword table. Retrying!" ; |
535 | |
536 | trx->error_state = DB_SUCCESS; |
537 | } else { |
538 | ib::error() << "Error '" << ut_strerr(error) |
539 | << "' while reading user stopword" |
540 | " table." ; |
541 | ret = FALSE; |
542 | break; |
543 | } |
544 | } |
545 | } |
546 | |
547 | que_graph_free(graph); |
548 | |
549 | cleanup: |
550 | if (!has_lock) { |
551 | mutex_exit(&dict_sys->mutex); |
552 | } |
553 | |
554 | trx_free(trx); |
555 | return(ret); |
556 | } |
557 | |
558 | /******************************************************************//** |
559 | Initialize the index cache. */ |
560 | static |
561 | void |
562 | fts_index_cache_init( |
563 | /*=================*/ |
564 | ib_alloc_t* allocator, /*!< in: the allocator to use */ |
565 | fts_index_cache_t* index_cache) /*!< in: index cache */ |
566 | { |
567 | ulint i; |
568 | |
569 | ut_a(index_cache->words == NULL); |
570 | |
571 | index_cache->words = rbt_create_arg_cmp( |
572 | sizeof(fts_tokenizer_word_t), innobase_fts_text_cmp, |
573 | (void*) index_cache->charset); |
574 | |
575 | ut_a(index_cache->doc_stats == NULL); |
576 | |
577 | index_cache->doc_stats = ib_vector_create( |
578 | allocator, sizeof(fts_doc_stats_t), 4); |
579 | |
580 | for (i = 0; i < FTS_NUM_AUX_INDEX; ++i) { |
581 | ut_a(index_cache->ins_graph[i] == NULL); |
582 | ut_a(index_cache->sel_graph[i] == NULL); |
583 | } |
584 | } |
585 | |
586 | /*********************************************************************//** |
587 | Initialize FTS cache. */ |
588 | void |
589 | fts_cache_init( |
590 | /*===========*/ |
591 | fts_cache_t* cache) /*!< in: cache to initialize */ |
592 | { |
593 | ulint i; |
594 | |
595 | /* Just to make sure */ |
596 | ut_a(cache->sync_heap->arg == NULL); |
597 | |
598 | cache->sync_heap->arg = mem_heap_create(1024); |
599 | |
600 | cache->total_size = 0; |
601 | |
602 | mutex_enter((ib_mutex_t*) &cache->deleted_lock); |
603 | cache->deleted_doc_ids = ib_vector_create( |
604 | cache->sync_heap, sizeof(fts_update_t), 4); |
605 | mutex_exit((ib_mutex_t*) &cache->deleted_lock); |
606 | |
607 | /* Reset the cache data for all the FTS indexes. */ |
608 | for (i = 0; i < ib_vector_size(cache->indexes); ++i) { |
609 | fts_index_cache_t* index_cache; |
610 | |
611 | index_cache = static_cast<fts_index_cache_t*>( |
612 | ib_vector_get(cache->indexes, i)); |
613 | |
614 | fts_index_cache_init(cache->sync_heap, index_cache); |
615 | } |
616 | } |
617 | |
618 | /****************************************************************//** |
619 | Create a FTS cache. */ |
620 | fts_cache_t* |
621 | fts_cache_create( |
622 | /*=============*/ |
623 | dict_table_t* table) /*!< in: table owns the FTS cache */ |
624 | { |
625 | mem_heap_t* heap; |
626 | fts_cache_t* cache; |
627 | |
628 | heap = static_cast<mem_heap_t*>(mem_heap_create(512)); |
629 | |
630 | cache = static_cast<fts_cache_t*>( |
631 | mem_heap_zalloc(heap, sizeof(*cache))); |
632 | |
633 | cache->cache_heap = heap; |
634 | |
635 | rw_lock_create(fts_cache_rw_lock_key, &cache->lock, SYNC_FTS_CACHE); |
636 | |
637 | rw_lock_create( |
638 | fts_cache_init_rw_lock_key, &cache->init_lock, |
639 | SYNC_FTS_CACHE_INIT); |
640 | |
641 | mutex_create(LATCH_ID_FTS_DELETE, &cache->deleted_lock); |
642 | |
643 | mutex_create(LATCH_ID_FTS_OPTIMIZE, &cache->optimize_lock); |
644 | |
645 | mutex_create(LATCH_ID_FTS_DOC_ID, &cache->doc_id_lock); |
646 | |
647 | /* This is the heap used to create the cache itself. */ |
648 | cache->self_heap = ib_heap_allocator_create(heap); |
649 | |
650 | /* This is a transient heap, used for storing sync data. */ |
651 | cache->sync_heap = ib_heap_allocator_create(heap); |
652 | cache->sync_heap->arg = NULL; |
653 | |
654 | cache->sync = static_cast<fts_sync_t*>( |
655 | mem_heap_zalloc(heap, sizeof(fts_sync_t))); |
656 | |
657 | cache->sync->table = table; |
658 | cache->sync->event = os_event_create(0); |
659 | |
660 | /* Create the index cache vector that will hold the inverted indexes. */ |
661 | cache->indexes = ib_vector_create( |
662 | cache->self_heap, sizeof(fts_index_cache_t), 2); |
663 | |
664 | fts_cache_init(cache); |
665 | |
666 | cache->stopword_info.cached_stopword = NULL; |
667 | cache->stopword_info.charset = NULL; |
668 | |
669 | cache->stopword_info.heap = cache->self_heap; |
670 | |
671 | cache->stopword_info.status = STOPWORD_NOT_INIT; |
672 | |
673 | return(cache); |
674 | } |
675 | |
676 | /*******************************************************************//** |
677 | Add a newly create index into FTS cache */ |
678 | void |
679 | fts_add_index( |
680 | /*==========*/ |
681 | dict_index_t* index, /*!< FTS index to be added */ |
682 | dict_table_t* table) /*!< table */ |
683 | { |
684 | fts_t* fts = table->fts; |
685 | fts_cache_t* cache; |
686 | fts_index_cache_t* index_cache; |
687 | |
688 | ut_ad(fts); |
689 | cache = table->fts->cache; |
690 | |
691 | rw_lock_x_lock(&cache->init_lock); |
692 | |
693 | ib_vector_push(fts->indexes, &index); |
694 | |
695 | index_cache = fts_find_index_cache(cache, index); |
696 | |
697 | if (!index_cache) { |
698 | /* Add new index cache structure */ |
699 | index_cache = fts_cache_index_cache_create(table, index); |
700 | } |
701 | |
702 | rw_lock_x_unlock(&cache->init_lock); |
703 | } |
704 | |
705 | /*******************************************************************//** |
706 | recalibrate get_doc structure after index_cache in cache->indexes changed */ |
707 | static |
708 | void |
709 | fts_reset_get_doc( |
710 | /*==============*/ |
711 | fts_cache_t* cache) /*!< in: FTS index cache */ |
712 | { |
713 | fts_get_doc_t* get_doc; |
714 | ulint i; |
715 | |
716 | ut_ad(rw_lock_own(&cache->init_lock, RW_LOCK_X)); |
717 | |
718 | ib_vector_reset(cache->get_docs); |
719 | |
720 | for (i = 0; i < ib_vector_size(cache->indexes); i++) { |
721 | fts_index_cache_t* ind_cache; |
722 | |
723 | ind_cache = static_cast<fts_index_cache_t*>( |
724 | ib_vector_get(cache->indexes, i)); |
725 | |
726 | get_doc = static_cast<fts_get_doc_t*>( |
727 | ib_vector_push(cache->get_docs, NULL)); |
728 | |
729 | memset(get_doc, 0x0, sizeof(*get_doc)); |
730 | |
731 | get_doc->index_cache = ind_cache; |
732 | } |
733 | |
734 | ut_ad(ib_vector_size(cache->get_docs) |
735 | == ib_vector_size(cache->indexes)); |
736 | } |
737 | |
738 | /*******************************************************************//** |
739 | Check an index is in the table->indexes list |
740 | @return TRUE if it exists */ |
741 | static |
742 | ibool |
743 | fts_in_dict_index( |
744 | /*==============*/ |
745 | dict_table_t* table, /*!< in: Table */ |
746 | dict_index_t* index_check) /*!< in: index to be checked */ |
747 | { |
748 | dict_index_t* index; |
749 | |
750 | for (index = dict_table_get_first_index(table); |
751 | index != NULL; |
752 | index = dict_table_get_next_index(index)) { |
753 | |
754 | if (index == index_check) { |
755 | return(TRUE); |
756 | } |
757 | } |
758 | |
759 | return(FALSE); |
760 | } |
761 | |
762 | /*******************************************************************//** |
763 | Check an index is in the fts->cache->indexes list |
764 | @return TRUE if it exists */ |
765 | static |
766 | ibool |
767 | fts_in_index_cache( |
768 | /*===============*/ |
769 | dict_table_t* table, /*!< in: Table */ |
770 | dict_index_t* index) /*!< in: index to be checked */ |
771 | { |
772 | ulint i; |
773 | |
774 | for (i = 0; i < ib_vector_size(table->fts->cache->indexes); i++) { |
775 | fts_index_cache_t* index_cache; |
776 | |
777 | index_cache = static_cast<fts_index_cache_t*>( |
778 | ib_vector_get(table->fts->cache->indexes, i)); |
779 | |
780 | if (index_cache->index == index) { |
781 | return(TRUE); |
782 | } |
783 | } |
784 | |
785 | return(FALSE); |
786 | } |
787 | |
788 | /*******************************************************************//** |
789 | Check indexes in the fts->indexes is also present in index cache and |
790 | table->indexes list |
791 | @return TRUE if all indexes match */ |
792 | ibool |
793 | fts_check_cached_index( |
794 | /*===================*/ |
795 | dict_table_t* table) /*!< in: Table where indexes are dropped */ |
796 | { |
797 | ulint i; |
798 | |
799 | if (!table->fts || !table->fts->cache) { |
800 | return(TRUE); |
801 | } |
802 | |
803 | ut_a(ib_vector_size(table->fts->indexes) |
804 | == ib_vector_size(table->fts->cache->indexes)); |
805 | |
806 | for (i = 0; i < ib_vector_size(table->fts->indexes); i++) { |
807 | dict_index_t* index; |
808 | |
809 | index = static_cast<dict_index_t*>( |
810 | ib_vector_getp(table->fts->indexes, i)); |
811 | |
812 | if (!fts_in_index_cache(table, index)) { |
813 | return(FALSE); |
814 | } |
815 | |
816 | if (!fts_in_dict_index(table, index)) { |
817 | return(FALSE); |
818 | } |
819 | } |
820 | |
821 | return(TRUE); |
822 | } |
823 | |
824 | /*******************************************************************//** |
825 | Drop auxiliary tables related to an FTS index |
826 | @return DB_SUCCESS or error number */ |
827 | dberr_t |
828 | fts_drop_index( |
829 | /*===========*/ |
830 | dict_table_t* table, /*!< in: Table where indexes are dropped */ |
831 | dict_index_t* index, /*!< in: Index to be dropped */ |
832 | trx_t* trx) /*!< in: Transaction for the drop */ |
833 | { |
834 | ib_vector_t* indexes = table->fts->indexes; |
835 | dberr_t err = DB_SUCCESS; |
836 | |
837 | ut_a(indexes); |
838 | |
839 | if ((ib_vector_size(indexes) == 1 |
840 | && (index == static_cast<dict_index_t*>( |
841 | ib_vector_getp(table->fts->indexes, 0)))) |
842 | || ib_vector_is_empty(indexes)) { |
843 | doc_id_t current_doc_id; |
844 | doc_id_t first_doc_id; |
845 | |
846 | /* If we are dropping the only FTS index of the table, |
847 | remove it from optimize thread */ |
848 | fts_optimize_remove_table(table); |
849 | |
850 | DICT_TF2_FLAG_UNSET(table, DICT_TF2_FTS); |
851 | |
852 | /* If Doc ID column is not added internally by FTS index, |
853 | we can drop all FTS auxiliary tables. Otherwise, we will |
854 | need to keep some common table such as CONFIG table, so |
855 | as to keep track of incrementing Doc IDs */ |
856 | if (!DICT_TF2_FLAG_IS_SET( |
857 | table, DICT_TF2_FTS_HAS_DOC_ID)) { |
858 | |
859 | err = fts_drop_tables(trx, table); |
860 | |
861 | err = fts_drop_index_tables(trx, index); |
862 | |
863 | for(;;) { |
864 | bool retry = false; |
865 | if (index->index_fts_syncing) { |
866 | retry = true; |
867 | } |
868 | if (!retry){ |
869 | fts_free(table); |
870 | break; |
871 | } |
872 | DICT_BG_YIELD(trx); |
873 | } |
874 | return(err); |
875 | } |
876 | |
877 | for(;;) { |
878 | bool retry = false; |
879 | if (index->index_fts_syncing) { |
880 | retry = true; |
881 | } |
882 | if (!retry){ |
883 | current_doc_id = table->fts->cache->next_doc_id; |
884 | first_doc_id = table->fts->cache->first_doc_id; |
885 | fts_cache_clear(table->fts->cache); |
886 | fts_cache_destroy(table->fts->cache); |
887 | table->fts->cache = fts_cache_create(table); |
888 | table->fts->cache->next_doc_id = current_doc_id; |
889 | table->fts->cache->first_doc_id = first_doc_id; |
890 | break; |
891 | } |
892 | DICT_BG_YIELD(trx); |
893 | } |
894 | } else { |
895 | fts_cache_t* cache = table->fts->cache; |
896 | fts_index_cache_t* index_cache; |
897 | |
898 | rw_lock_x_lock(&cache->init_lock); |
899 | |
900 | index_cache = fts_find_index_cache(cache, index); |
901 | |
902 | if (index_cache != NULL) { |
903 | for(;;) { |
904 | bool retry = false; |
905 | if (index->index_fts_syncing) { |
906 | retry = true; |
907 | } |
908 | if (!retry && index_cache->words) { |
909 | fts_words_free(index_cache->words); |
910 | rbt_free(index_cache->words); |
911 | break; |
912 | } |
913 | DICT_BG_YIELD(trx); |
914 | } |
915 | |
916 | ib_vector_remove(cache->indexes, *(void**) index_cache); |
917 | } |
918 | |
919 | if (cache->get_docs) { |
920 | fts_reset_get_doc(cache); |
921 | } |
922 | |
923 | rw_lock_x_unlock(&cache->init_lock); |
924 | } |
925 | |
926 | err = fts_drop_index_tables(trx, index); |
927 | |
928 | ib_vector_remove(indexes, (const void*) index); |
929 | |
930 | return(err); |
931 | } |
932 | |
933 | /****************************************************************//** |
934 | Free the query graph but check whether dict_sys->mutex is already |
935 | held */ |
936 | void |
937 | fts_que_graph_free_check_lock( |
938 | /*==========================*/ |
939 | fts_table_t* fts_table, /*!< in: FTS table */ |
940 | const fts_index_cache_t*index_cache, /*!< in: FTS index cache */ |
941 | que_t* graph) /*!< in: query graph */ |
942 | { |
943 | ibool has_dict = FALSE; |
944 | |
945 | if (fts_table && fts_table->table) { |
946 | ut_ad(fts_table->table->fts); |
947 | |
948 | has_dict = fts_table->table->fts->fts_status |
949 | & TABLE_DICT_LOCKED; |
950 | } else if (index_cache) { |
951 | ut_ad(index_cache->index->table->fts); |
952 | |
953 | has_dict = index_cache->index->table->fts->fts_status |
954 | & TABLE_DICT_LOCKED; |
955 | } |
956 | |
957 | if (!has_dict) { |
958 | mutex_enter(&dict_sys->mutex); |
959 | } |
960 | |
961 | ut_ad(mutex_own(&dict_sys->mutex)); |
962 | |
963 | que_graph_free(graph); |
964 | |
965 | if (!has_dict) { |
966 | mutex_exit(&dict_sys->mutex); |
967 | } |
968 | } |
969 | |
970 | /****************************************************************//** |
971 | Create an FTS index cache. */ |
972 | CHARSET_INFO* |
973 | fts_index_get_charset( |
974 | /*==================*/ |
975 | dict_index_t* index) /*!< in: FTS index */ |
976 | { |
977 | CHARSET_INFO* charset = NULL; |
978 | dict_field_t* field; |
979 | ulint prtype; |
980 | |
981 | field = dict_index_get_nth_field(index, 0); |
982 | prtype = field->col->prtype; |
983 | |
984 | charset = fts_get_charset(prtype); |
985 | |
986 | #ifdef FTS_DEBUG |
987 | /* Set up charset info for this index. Please note all |
988 | field of the FTS index should have the same charset */ |
989 | for (i = 1; i < index->n_fields; i++) { |
990 | CHARSET_INFO* fld_charset; |
991 | |
992 | field = dict_index_get_nth_field(index, i); |
993 | prtype = field->col->prtype; |
994 | |
995 | fld_charset = fts_get_charset(prtype); |
996 | |
997 | /* All FTS columns should have the same charset */ |
998 | if (charset) { |
999 | ut_a(charset == fld_charset); |
1000 | } else { |
1001 | charset = fld_charset; |
1002 | } |
1003 | } |
1004 | #endif |
1005 | |
1006 | return(charset); |
1007 | |
1008 | } |
1009 | /****************************************************************//** |
1010 | Create an FTS index cache. |
1011 | @return Index Cache */ |
1012 | fts_index_cache_t* |
1013 | fts_cache_index_cache_create( |
1014 | /*=========================*/ |
1015 | dict_table_t* table, /*!< in: table with FTS index */ |
1016 | dict_index_t* index) /*!< in: FTS index */ |
1017 | { |
1018 | ulint n_bytes; |
1019 | fts_index_cache_t* index_cache; |
1020 | fts_cache_t* cache = table->fts->cache; |
1021 | |
1022 | ut_a(cache != NULL); |
1023 | |
1024 | ut_ad(rw_lock_own(&cache->init_lock, RW_LOCK_X)); |
1025 | |
1026 | /* Must not already exist in the cache vector. */ |
1027 | ut_a(fts_find_index_cache(cache, index) == NULL); |
1028 | |
1029 | index_cache = static_cast<fts_index_cache_t*>( |
1030 | ib_vector_push(cache->indexes, NULL)); |
1031 | |
1032 | memset(index_cache, 0x0, sizeof(*index_cache)); |
1033 | |
1034 | index_cache->index = index; |
1035 | |
1036 | index_cache->charset = fts_index_get_charset(index); |
1037 | |
1038 | n_bytes = sizeof(que_t*) * FTS_NUM_AUX_INDEX; |
1039 | |
1040 | index_cache->ins_graph = static_cast<que_t**>( |
1041 | mem_heap_zalloc(static_cast<mem_heap_t*>( |
1042 | cache->self_heap->arg), n_bytes)); |
1043 | |
1044 | index_cache->sel_graph = static_cast<que_t**>( |
1045 | mem_heap_zalloc(static_cast<mem_heap_t*>( |
1046 | cache->self_heap->arg), n_bytes)); |
1047 | |
1048 | fts_index_cache_init(cache->sync_heap, index_cache); |
1049 | |
1050 | if (cache->get_docs) { |
1051 | fts_reset_get_doc(cache); |
1052 | } |
1053 | |
1054 | return(index_cache); |
1055 | } |
1056 | |
1057 | /****************************************************************//** |
1058 | Release all resources help by the words rb tree e.g., the node ilist. */ |
1059 | static |
1060 | void |
1061 | fts_words_free( |
1062 | /*===========*/ |
1063 | ib_rbt_t* words) /*!< in: rb tree of words */ |
1064 | { |
1065 | const ib_rbt_node_t* rbt_node; |
1066 | |
1067 | /* Free the resources held by a word. */ |
1068 | for (rbt_node = rbt_first(words); |
1069 | rbt_node != NULL; |
1070 | rbt_node = rbt_first(words)) { |
1071 | |
1072 | ulint i; |
1073 | fts_tokenizer_word_t* word; |
1074 | |
1075 | word = rbt_value(fts_tokenizer_word_t, rbt_node); |
1076 | |
1077 | /* Free the ilists of this word. */ |
1078 | for (i = 0; i < ib_vector_size(word->nodes); ++i) { |
1079 | |
1080 | fts_node_t* fts_node = static_cast<fts_node_t*>( |
1081 | ib_vector_get(word->nodes, i)); |
1082 | |
1083 | ut_free(fts_node->ilist); |
1084 | fts_node->ilist = NULL; |
1085 | } |
1086 | |
1087 | /* NOTE: We are responsible for free'ing the node */ |
1088 | ut_free(rbt_remove_node(words, rbt_node)); |
1089 | } |
1090 | } |
1091 | |
1092 | /** Clear cache. |
1093 | @param[in,out] cache fts cache */ |
1094 | void |
1095 | fts_cache_clear( |
1096 | fts_cache_t* cache) |
1097 | { |
1098 | ulint i; |
1099 | |
1100 | for (i = 0; i < ib_vector_size(cache->indexes); ++i) { |
1101 | ulint j; |
1102 | fts_index_cache_t* index_cache; |
1103 | |
1104 | index_cache = static_cast<fts_index_cache_t*>( |
1105 | ib_vector_get(cache->indexes, i)); |
1106 | |
1107 | fts_words_free(index_cache->words); |
1108 | |
1109 | rbt_free(index_cache->words); |
1110 | |
1111 | index_cache->words = NULL; |
1112 | |
1113 | for (j = 0; j < FTS_NUM_AUX_INDEX; ++j) { |
1114 | |
1115 | if (index_cache->ins_graph[j] != NULL) { |
1116 | |
1117 | fts_que_graph_free_check_lock( |
1118 | NULL, index_cache, |
1119 | index_cache->ins_graph[j]); |
1120 | |
1121 | index_cache->ins_graph[j] = NULL; |
1122 | } |
1123 | |
1124 | if (index_cache->sel_graph[j] != NULL) { |
1125 | |
1126 | fts_que_graph_free_check_lock( |
1127 | NULL, index_cache, |
1128 | index_cache->sel_graph[j]); |
1129 | |
1130 | index_cache->sel_graph[j] = NULL; |
1131 | } |
1132 | } |
1133 | |
1134 | index_cache->doc_stats = NULL; |
1135 | } |
1136 | |
1137 | mem_heap_free(static_cast<mem_heap_t*>(cache->sync_heap->arg)); |
1138 | cache->sync_heap->arg = NULL; |
1139 | |
1140 | fts_need_sync = false; |
1141 | |
1142 | cache->total_size = 0; |
1143 | |
1144 | mutex_enter((ib_mutex_t*) &cache->deleted_lock); |
1145 | cache->deleted_doc_ids = NULL; |
1146 | mutex_exit((ib_mutex_t*) &cache->deleted_lock); |
1147 | } |
1148 | |
1149 | /*********************************************************************//** |
1150 | Search the index specific cache for a particular FTS index. |
1151 | @return the index cache else NULL */ |
1152 | UNIV_INLINE |
1153 | fts_index_cache_t* |
1154 | fts_get_index_cache( |
1155 | /*================*/ |
1156 | fts_cache_t* cache, /*!< in: cache to search */ |
1157 | const dict_index_t* index) /*!< in: index to search for */ |
1158 | { |
1159 | ulint i; |
1160 | |
1161 | ut_ad(rw_lock_own((rw_lock_t*) &cache->lock, RW_LOCK_X) |
1162 | || rw_lock_own((rw_lock_t*) &cache->init_lock, RW_LOCK_X)); |
1163 | |
1164 | for (i = 0; i < ib_vector_size(cache->indexes); ++i) { |
1165 | fts_index_cache_t* index_cache; |
1166 | |
1167 | index_cache = static_cast<fts_index_cache_t*>( |
1168 | ib_vector_get(cache->indexes, i)); |
1169 | |
1170 | if (index_cache->index == index) { |
1171 | |
1172 | return(index_cache); |
1173 | } |
1174 | } |
1175 | |
1176 | return(NULL); |
1177 | } |
1178 | |
1179 | #ifdef FTS_DEBUG |
1180 | /*********************************************************************//** |
1181 | Search the index cache for a get_doc structure. |
1182 | @return the fts_get_doc_t item else NULL */ |
1183 | static |
1184 | fts_get_doc_t* |
1185 | fts_get_index_get_doc( |
1186 | /*==================*/ |
1187 | fts_cache_t* cache, /*!< in: cache to search */ |
1188 | const dict_index_t* index) /*!< in: index to search for */ |
1189 | { |
1190 | ulint i; |
1191 | |
1192 | ut_ad(rw_lock_own((rw_lock_t*) &cache->init_lock, RW_LOCK_X)); |
1193 | |
1194 | for (i = 0; i < ib_vector_size(cache->get_docs); ++i) { |
1195 | fts_get_doc_t* get_doc; |
1196 | |
1197 | get_doc = static_cast<fts_get_doc_t*>( |
1198 | ib_vector_get(cache->get_docs, i)); |
1199 | |
1200 | if (get_doc->index_cache->index == index) { |
1201 | |
1202 | return(get_doc); |
1203 | } |
1204 | } |
1205 | |
1206 | return(NULL); |
1207 | } |
1208 | #endif |
1209 | |
1210 | /**********************************************************************//** |
1211 | Find an existing word, or if not found, create one and return it. |
1212 | @return specified word token */ |
1213 | static |
1214 | fts_tokenizer_word_t* |
1215 | fts_tokenizer_word_get( |
1216 | /*===================*/ |
1217 | fts_cache_t* cache, /*!< in: cache */ |
1218 | fts_index_cache_t* |
1219 | index_cache, /*!< in: index cache */ |
1220 | fts_string_t* text) /*!< in: node text */ |
1221 | { |
1222 | fts_tokenizer_word_t* word; |
1223 | ib_rbt_bound_t parent; |
1224 | |
1225 | ut_ad(rw_lock_own(&cache->lock, RW_LOCK_X)); |
1226 | |
1227 | /* If it is a stopword, do not index it */ |
1228 | if (!fts_check_token(text, |
1229 | cache->stopword_info.cached_stopword, |
1230 | index_cache->charset)) { |
1231 | |
1232 | return(NULL); |
1233 | } |
1234 | |
1235 | /* Check if we found a match, if not then add word to tree. */ |
1236 | if (rbt_search(index_cache->words, &parent, text) != 0) { |
1237 | mem_heap_t* heap; |
1238 | fts_tokenizer_word_t new_word; |
1239 | |
1240 | heap = static_cast<mem_heap_t*>(cache->sync_heap->arg); |
1241 | |
1242 | new_word.nodes = ib_vector_create( |
1243 | cache->sync_heap, sizeof(fts_node_t), 4); |
1244 | |
1245 | fts_string_dup(&new_word.text, text, heap); |
1246 | |
1247 | parent.last = rbt_add_node( |
1248 | index_cache->words, &parent, &new_word); |
1249 | |
1250 | /* Take into account the RB tree memory use and the vector. */ |
1251 | cache->total_size += sizeof(new_word) |
1252 | + sizeof(ib_rbt_node_t) |
1253 | + text->f_len |
1254 | + (sizeof(fts_node_t) * 4) |
1255 | + sizeof(*new_word.nodes); |
1256 | |
1257 | ut_ad(rbt_validate(index_cache->words)); |
1258 | } |
1259 | |
1260 | word = rbt_value(fts_tokenizer_word_t, parent.last); |
1261 | |
1262 | return(word); |
1263 | } |
1264 | |
1265 | /**********************************************************************//** |
1266 | Add the given doc_id/word positions to the given node's ilist. */ |
1267 | void |
1268 | fts_cache_node_add_positions( |
1269 | /*=========================*/ |
1270 | fts_cache_t* cache, /*!< in: cache */ |
1271 | fts_node_t* node, /*!< in: word node */ |
1272 | doc_id_t doc_id, /*!< in: doc id */ |
1273 | ib_vector_t* positions) /*!< in: fts_token_t::positions */ |
1274 | { |
1275 | ulint i; |
1276 | byte* ptr; |
1277 | byte* ilist; |
1278 | ulint enc_len; |
1279 | ulint last_pos; |
1280 | byte* ptr_start; |
1281 | ulint doc_id_delta; |
1282 | |
1283 | #ifdef UNIV_DEBUG |
1284 | if (cache) { |
1285 | ut_ad(rw_lock_own(&cache->lock, RW_LOCK_X)); |
1286 | } |
1287 | #endif /* UNIV_DEBUG */ |
1288 | |
1289 | ut_ad(doc_id >= node->last_doc_id); |
1290 | |
1291 | /* Calculate the space required to store the ilist. */ |
1292 | doc_id_delta = (ulint)(doc_id - node->last_doc_id); |
1293 | enc_len = fts_get_encoded_len(doc_id_delta); |
1294 | |
1295 | last_pos = 0; |
1296 | for (i = 0; i < ib_vector_size(positions); i++) { |
1297 | ulint pos = *(static_cast<ulint*>( |
1298 | ib_vector_get(positions, i))); |
1299 | |
1300 | ut_ad(last_pos == 0 || pos > last_pos); |
1301 | |
1302 | enc_len += fts_get_encoded_len(pos - last_pos); |
1303 | last_pos = pos; |
1304 | } |
1305 | |
1306 | /* The 0x00 byte at the end of the token positions list. */ |
1307 | enc_len++; |
1308 | |
1309 | if ((node->ilist_size_alloc - node->ilist_size) >= enc_len) { |
1310 | /* No need to allocate more space, we can fit in the new |
1311 | data at the end of the old one. */ |
1312 | ilist = NULL; |
1313 | ptr = node->ilist + node->ilist_size; |
1314 | } else { |
1315 | ulint new_size = node->ilist_size + enc_len; |
1316 | |
1317 | /* Over-reserve space by a fixed size for small lengths and |
1318 | by 20% for lengths >= 48 bytes. */ |
1319 | if (new_size < 16) { |
1320 | new_size = 16; |
1321 | } else if (new_size < 32) { |
1322 | new_size = 32; |
1323 | } else if (new_size < 48) { |
1324 | new_size = 48; |
1325 | } else { |
1326 | new_size = (ulint)(1.2 * new_size); |
1327 | } |
1328 | |
1329 | ilist = static_cast<byte*>(ut_malloc_nokey(new_size)); |
1330 | ptr = ilist + node->ilist_size; |
1331 | |
1332 | node->ilist_size_alloc = new_size; |
1333 | } |
1334 | |
1335 | ptr_start = ptr; |
1336 | |
1337 | /* Encode the new fragment. */ |
1338 | ptr += fts_encode_int(doc_id_delta, ptr); |
1339 | |
1340 | last_pos = 0; |
1341 | for (i = 0; i < ib_vector_size(positions); i++) { |
1342 | ulint pos = *(static_cast<ulint*>( |
1343 | ib_vector_get(positions, i))); |
1344 | |
1345 | ptr += fts_encode_int(pos - last_pos, ptr); |
1346 | last_pos = pos; |
1347 | } |
1348 | |
1349 | *ptr++ = 0; |
1350 | |
1351 | ut_a(enc_len == (ulint)(ptr - ptr_start)); |
1352 | |
1353 | if (ilist) { |
1354 | /* Copy old ilist to the start of the new one and switch the |
1355 | new one into place in the node. */ |
1356 | if (node->ilist_size > 0) { |
1357 | memcpy(ilist, node->ilist, node->ilist_size); |
1358 | ut_free(node->ilist); |
1359 | } |
1360 | |
1361 | node->ilist = ilist; |
1362 | } |
1363 | |
1364 | node->ilist_size += enc_len; |
1365 | |
1366 | if (cache) { |
1367 | cache->total_size += enc_len; |
1368 | } |
1369 | |
1370 | if (node->first_doc_id == FTS_NULL_DOC_ID) { |
1371 | node->first_doc_id = doc_id; |
1372 | } |
1373 | |
1374 | node->last_doc_id = doc_id; |
1375 | ++node->doc_count; |
1376 | } |
1377 | |
1378 | /**********************************************************************//** |
1379 | Add document to the cache. */ |
1380 | static |
1381 | void |
1382 | fts_cache_add_doc( |
1383 | /*==============*/ |
1384 | fts_cache_t* cache, /*!< in: cache */ |
1385 | fts_index_cache_t* |
1386 | index_cache, /*!< in: index cache */ |
1387 | doc_id_t doc_id, /*!< in: doc id to add */ |
1388 | ib_rbt_t* tokens) /*!< in: document tokens */ |
1389 | { |
1390 | const ib_rbt_node_t* node; |
1391 | ulint n_words; |
1392 | fts_doc_stats_t* doc_stats; |
1393 | |
1394 | if (!tokens) { |
1395 | return; |
1396 | } |
1397 | |
1398 | ut_ad(rw_lock_own(&cache->lock, RW_LOCK_X)); |
1399 | |
1400 | n_words = rbt_size(tokens); |
1401 | |
1402 | for (node = rbt_first(tokens); node; node = rbt_first(tokens)) { |
1403 | |
1404 | fts_tokenizer_word_t* word; |
1405 | fts_node_t* fts_node = NULL; |
1406 | fts_token_t* token = rbt_value(fts_token_t, node); |
1407 | |
1408 | /* Find and/or add token to the cache. */ |
1409 | word = fts_tokenizer_word_get( |
1410 | cache, index_cache, &token->text); |
1411 | |
1412 | if (!word) { |
1413 | ut_free(rbt_remove_node(tokens, node)); |
1414 | continue; |
1415 | } |
1416 | |
1417 | if (ib_vector_size(word->nodes) > 0) { |
1418 | fts_node = static_cast<fts_node_t*>( |
1419 | ib_vector_last(word->nodes)); |
1420 | } |
1421 | |
1422 | if (fts_node == NULL || fts_node->synced |
1423 | || fts_node->ilist_size > FTS_ILIST_MAX_SIZE |
1424 | || doc_id < fts_node->last_doc_id) { |
1425 | |
1426 | fts_node = static_cast<fts_node_t*>( |
1427 | ib_vector_push(word->nodes, NULL)); |
1428 | |
1429 | memset(fts_node, 0x0, sizeof(*fts_node)); |
1430 | |
1431 | cache->total_size += sizeof(*fts_node); |
1432 | } |
1433 | |
1434 | fts_cache_node_add_positions( |
1435 | cache, fts_node, doc_id, token->positions); |
1436 | |
1437 | ut_free(rbt_remove_node(tokens, node)); |
1438 | } |
1439 | |
1440 | ut_a(rbt_empty(tokens)); |
1441 | |
1442 | /* Add to doc ids processed so far. */ |
1443 | doc_stats = static_cast<fts_doc_stats_t*>( |
1444 | ib_vector_push(index_cache->doc_stats, NULL)); |
1445 | |
1446 | doc_stats->doc_id = doc_id; |
1447 | doc_stats->word_count = n_words; |
1448 | |
1449 | /* Add the doc stats memory usage too. */ |
1450 | cache->total_size += sizeof(*doc_stats); |
1451 | |
1452 | if (doc_id > cache->sync->max_doc_id) { |
1453 | cache->sync->max_doc_id = doc_id; |
1454 | } |
1455 | } |
1456 | |
1457 | /****************************************************************//** |
1458 | Drops a table. If the table can't be found we return a SUCCESS code. |
1459 | @return DB_SUCCESS or error code */ |
1460 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
1461 | dberr_t |
1462 | fts_drop_table( |
1463 | /*===========*/ |
1464 | trx_t* trx, /*!< in: transaction */ |
1465 | const char* table_name) /*!< in: table to drop */ |
1466 | { |
1467 | dict_table_t* table; |
1468 | dberr_t error = DB_SUCCESS; |
1469 | |
1470 | /* Check that the table exists in our data dictionary. |
1471 | Similar to regular drop table case, we will open table with |
1472 | DICT_ERR_IGNORE_INDEX_ROOT and DICT_ERR_IGNORE_CORRUPT option */ |
1473 | table = dict_table_open_on_name( |
1474 | table_name, TRUE, FALSE, |
1475 | static_cast<dict_err_ignore_t>( |
1476 | DICT_ERR_IGNORE_INDEX_ROOT | DICT_ERR_IGNORE_CORRUPT)); |
1477 | |
1478 | if (table != 0) { |
1479 | |
1480 | dict_table_close(table, TRUE, FALSE); |
1481 | |
1482 | /* Pass nonatomic=false (dont allow data dict unlock), |
1483 | because the transaction may hold locks on SYS_* tables from |
1484 | previous calls to fts_drop_table(). */ |
1485 | error = row_drop_table_for_mysql(table_name, trx, true, false, false); |
1486 | |
1487 | if (error != DB_SUCCESS) { |
1488 | ib::error() << "Unable to drop FTS index aux table " |
1489 | << table_name << ": " << ut_strerr(error); |
1490 | } |
1491 | } else { |
1492 | error = DB_FAIL; |
1493 | } |
1494 | |
1495 | return(error); |
1496 | } |
1497 | |
1498 | /****************************************************************//** |
1499 | Rename a single auxiliary table due to database name change. |
1500 | @return DB_SUCCESS or error code */ |
1501 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
1502 | dberr_t |
1503 | fts_rename_one_aux_table( |
1504 | /*=====================*/ |
1505 | const char* new_name, /*!< in: new parent tbl name */ |
1506 | const char* fts_table_old_name, /*!< in: old aux tbl name */ |
1507 | trx_t* trx) /*!< in: transaction */ |
1508 | { |
1509 | char fts_table_new_name[MAX_TABLE_NAME_LEN]; |
1510 | ulint new_db_name_len = dict_get_db_name_len(new_name); |
1511 | ulint old_db_name_len = dict_get_db_name_len(fts_table_old_name); |
1512 | ulint table_new_name_len = strlen(fts_table_old_name) |
1513 | + new_db_name_len - old_db_name_len; |
1514 | |
1515 | /* Check if the new and old database names are the same, if so, |
1516 | nothing to do */ |
1517 | ut_ad((new_db_name_len != old_db_name_len) |
1518 | || strncmp(new_name, fts_table_old_name, old_db_name_len) != 0); |
1519 | |
1520 | /* Get the database name from "new_name", and table name |
1521 | from the fts_table_old_name */ |
1522 | strncpy(fts_table_new_name, new_name, new_db_name_len); |
1523 | strncpy(fts_table_new_name + new_db_name_len, |
1524 | strchr(fts_table_old_name, '/'), |
1525 | table_new_name_len - new_db_name_len); |
1526 | fts_table_new_name[table_new_name_len] = 0; |
1527 | |
1528 | return(row_rename_table_for_mysql( |
1529 | fts_table_old_name, fts_table_new_name, trx, false)); |
1530 | } |
1531 | |
1532 | /****************************************************************//** |
1533 | Rename auxiliary tables for all fts index for a table. This(rename) |
1534 | is due to database name change |
1535 | @return DB_SUCCESS or error code */ |
1536 | dberr_t |
1537 | fts_rename_aux_tables( |
1538 | /*==================*/ |
1539 | dict_table_t* table, /*!< in: user Table */ |
1540 | const char* new_name, /*!< in: new table name */ |
1541 | trx_t* trx) /*!< in: transaction */ |
1542 | { |
1543 | ulint i; |
1544 | fts_table_t fts_table; |
1545 | |
1546 | FTS_INIT_FTS_TABLE(&fts_table, NULL, FTS_COMMON_TABLE, table); |
1547 | |
1548 | /* Rename common auxiliary tables */ |
1549 | for (i = 0; fts_common_tables[i] != NULL; ++i) { |
1550 | char old_table_name[MAX_FULL_NAME_LEN]; |
1551 | dberr_t err = DB_SUCCESS; |
1552 | |
1553 | fts_table.suffix = fts_common_tables[i]; |
1554 | |
1555 | fts_get_table_name(&fts_table, old_table_name); |
1556 | |
1557 | err = fts_rename_one_aux_table(new_name, old_table_name, trx); |
1558 | |
1559 | if (err != DB_SUCCESS) { |
1560 | return(err); |
1561 | } |
1562 | } |
1563 | |
1564 | fts_t* fts = table->fts; |
1565 | |
1566 | /* Rename index specific auxiliary tables */ |
1567 | for (i = 0; fts->indexes != 0 && i < ib_vector_size(fts->indexes); |
1568 | ++i) { |
1569 | dict_index_t* index; |
1570 | |
1571 | index = static_cast<dict_index_t*>( |
1572 | ib_vector_getp(fts->indexes, i)); |
1573 | |
1574 | FTS_INIT_INDEX_TABLE(&fts_table, NULL, FTS_INDEX_TABLE, index); |
1575 | |
1576 | for (ulint j = 0; j < FTS_NUM_AUX_INDEX; ++j) { |
1577 | dberr_t err; |
1578 | char old_table_name[MAX_FULL_NAME_LEN]; |
1579 | |
1580 | fts_table.suffix = fts_get_suffix(j); |
1581 | |
1582 | fts_get_table_name(&fts_table, old_table_name); |
1583 | |
1584 | err = fts_rename_one_aux_table( |
1585 | new_name, old_table_name, trx); |
1586 | |
1587 | DBUG_EXECUTE_IF("fts_rename_failure" , |
1588 | err = DB_DEADLOCK; |
1589 | fts_sql_rollback(trx);); |
1590 | |
1591 | if (err != DB_SUCCESS) { |
1592 | return(err); |
1593 | } |
1594 | } |
1595 | } |
1596 | |
1597 | return(DB_SUCCESS); |
1598 | } |
1599 | |
1600 | /****************************************************************//** |
1601 | Drops the common ancillary tables needed for supporting an FTS index |
1602 | on the given table. row_mysql_lock_data_dictionary must have been called |
1603 | before this. |
1604 | @return DB_SUCCESS or error code */ |
1605 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
1606 | dberr_t |
1607 | fts_drop_common_tables( |
1608 | /*===================*/ |
1609 | trx_t* trx, /*!< in: transaction */ |
1610 | fts_table_t* fts_table) /*!< in: table with an FTS |
1611 | index */ |
1612 | { |
1613 | ulint i; |
1614 | dberr_t error = DB_SUCCESS; |
1615 | |
1616 | for (i = 0; fts_common_tables[i] != NULL; ++i) { |
1617 | dberr_t err; |
1618 | char table_name[MAX_FULL_NAME_LEN]; |
1619 | |
1620 | fts_table->suffix = fts_common_tables[i]; |
1621 | |
1622 | fts_get_table_name(fts_table, table_name); |
1623 | |
1624 | err = fts_drop_table(trx, table_name); |
1625 | |
1626 | /* We only return the status of the last error. */ |
1627 | if (err != DB_SUCCESS && err != DB_FAIL) { |
1628 | error = err; |
1629 | } |
1630 | } |
1631 | |
1632 | return(error); |
1633 | } |
1634 | |
1635 | /****************************************************************//** |
1636 | Since we do a horizontal split on the index table, we need to drop |
1637 | all the split tables. |
1638 | @return DB_SUCCESS or error code */ |
1639 | static |
1640 | dberr_t |
1641 | fts_drop_index_split_tables( |
1642 | /*========================*/ |
1643 | trx_t* trx, /*!< in: transaction */ |
1644 | dict_index_t* index) /*!< in: fts instance */ |
1645 | |
1646 | { |
1647 | ulint i; |
1648 | fts_table_t fts_table; |
1649 | dberr_t error = DB_SUCCESS; |
1650 | |
1651 | FTS_INIT_INDEX_TABLE(&fts_table, NULL, FTS_INDEX_TABLE, index); |
1652 | |
1653 | for (i = 0; i < FTS_NUM_AUX_INDEX; ++i) { |
1654 | dberr_t err; |
1655 | char table_name[MAX_FULL_NAME_LEN]; |
1656 | |
1657 | fts_table.suffix = fts_get_suffix(i); |
1658 | |
1659 | fts_get_table_name(&fts_table, table_name); |
1660 | |
1661 | err = fts_drop_table(trx, table_name); |
1662 | |
1663 | /* We only return the status of the last error. */ |
1664 | if (err != DB_SUCCESS && err != DB_FAIL) { |
1665 | error = err; |
1666 | } |
1667 | } |
1668 | |
1669 | return(error); |
1670 | } |
1671 | |
1672 | /****************************************************************//** |
1673 | Drops FTS auxiliary tables for an FTS index |
1674 | @return DB_SUCCESS or error code */ |
1675 | dberr_t |
1676 | fts_drop_index_tables( |
1677 | /*==================*/ |
1678 | trx_t* trx, /*!< in: transaction */ |
1679 | dict_index_t* index) /*!< in: Index to drop */ |
1680 | { |
1681 | return(fts_drop_index_split_tables(trx, index)); |
1682 | } |
1683 | |
1684 | /****************************************************************//** |
1685 | Drops FTS ancillary tables needed for supporting an FTS index |
1686 | on the given table. row_mysql_lock_data_dictionary must have been called |
1687 | before this. |
1688 | @return DB_SUCCESS or error code */ |
1689 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
1690 | dberr_t |
1691 | fts_drop_all_index_tables( |
1692 | /*======================*/ |
1693 | trx_t* trx, /*!< in: transaction */ |
1694 | fts_t* fts) /*!< in: fts instance */ |
1695 | { |
1696 | dberr_t error = DB_SUCCESS; |
1697 | |
1698 | for (ulint i = 0; |
1699 | fts->indexes != 0 && i < ib_vector_size(fts->indexes); |
1700 | ++i) { |
1701 | |
1702 | dberr_t err; |
1703 | dict_index_t* index; |
1704 | |
1705 | index = static_cast<dict_index_t*>( |
1706 | ib_vector_getp(fts->indexes, i)); |
1707 | |
1708 | err = fts_drop_index_tables(trx, index); |
1709 | |
1710 | if (err != DB_SUCCESS) { |
1711 | error = err; |
1712 | } |
1713 | } |
1714 | |
1715 | return(error); |
1716 | } |
1717 | |
1718 | /*********************************************************************//** |
1719 | Drops the ancillary tables needed for supporting an FTS index on a |
1720 | given table. row_mysql_lock_data_dictionary must have been called before |
1721 | this. |
1722 | @return DB_SUCCESS or error code */ |
1723 | dberr_t |
1724 | fts_drop_tables( |
1725 | /*============*/ |
1726 | trx_t* trx, /*!< in: transaction */ |
1727 | dict_table_t* table) /*!< in: table has the FTS index */ |
1728 | { |
1729 | dberr_t error; |
1730 | fts_table_t fts_table; |
1731 | |
1732 | FTS_INIT_FTS_TABLE(&fts_table, NULL, FTS_COMMON_TABLE, table); |
1733 | |
1734 | /* TODO: This is not atomic and can cause problems during recovery. */ |
1735 | |
1736 | error = fts_drop_common_tables(trx, &fts_table); |
1737 | |
1738 | if (error == DB_SUCCESS) { |
1739 | error = fts_drop_all_index_tables(trx, table->fts); |
1740 | } |
1741 | |
1742 | return(error); |
1743 | } |
1744 | |
1745 | /** Create dict_table_t object for FTS Aux tables. |
1746 | @param[in] aux_table_name FTS Aux table name |
1747 | @param[in] table table object of FTS Index |
1748 | @param[in] n_cols number of columns for FTS Aux table |
1749 | @return table object for FTS Aux table */ |
1750 | static |
1751 | dict_table_t* |
1752 | fts_create_in_mem_aux_table( |
1753 | const char* aux_table_name, |
1754 | const dict_table_t* table, |
1755 | ulint n_cols) |
1756 | { |
1757 | dict_table_t* new_table = dict_mem_table_create( |
1758 | aux_table_name, NULL, n_cols, 0, table->flags, |
1759 | table->space->id == TRX_SYS_SPACE |
1760 | ? 0 : table->space->purpose == FIL_TYPE_TEMPORARY |
1761 | ? DICT_TF2_TEMPORARY : DICT_TF2_USE_FILE_PER_TABLE); |
1762 | |
1763 | if (DICT_TF_HAS_DATA_DIR(table->flags)) { |
1764 | ut_ad(table->data_dir_path != NULL); |
1765 | new_table->data_dir_path = mem_heap_strdup( |
1766 | new_table->heap, table->data_dir_path); |
1767 | } |
1768 | |
1769 | return(new_table); |
1770 | } |
1771 | |
1772 | /** Function to create on FTS common table. |
1773 | @param[in,out] trx InnoDB transaction |
1774 | @param[in] table Table that has FTS Index |
1775 | @param[in] fts_table_name FTS AUX table name |
1776 | @param[in] fts_suffix FTS AUX table suffix |
1777 | @param[in,out] heap temporary memory heap |
1778 | @return table object if created, else NULL */ |
1779 | static |
1780 | dict_table_t* |
1781 | fts_create_one_common_table( |
1782 | trx_t* trx, |
1783 | const dict_table_t* table, |
1784 | const char* fts_table_name, |
1785 | const char* fts_suffix, |
1786 | mem_heap_t* heap) |
1787 | { |
1788 | dict_table_t* new_table = NULL; |
1789 | dberr_t error; |
1790 | bool is_config = strcmp(fts_suffix, "CONFIG" ) == 0; |
1791 | |
1792 | if (!is_config) { |
1793 | |
1794 | new_table = fts_create_in_mem_aux_table( |
1795 | fts_table_name, table, FTS_DELETED_TABLE_NUM_COLS); |
1796 | |
1797 | dict_mem_table_add_col( |
1798 | new_table, heap, "doc_id" , DATA_INT, DATA_UNSIGNED, |
1799 | FTS_DELETED_TABLE_COL_LEN); |
1800 | } else { |
1801 | /* Config table has different schema. */ |
1802 | new_table = fts_create_in_mem_aux_table( |
1803 | fts_table_name, table, FTS_CONFIG_TABLE_NUM_COLS); |
1804 | |
1805 | dict_mem_table_add_col( |
1806 | new_table, heap, "key" , DATA_VARCHAR, 0, |
1807 | FTS_CONFIG_TABLE_KEY_COL_LEN); |
1808 | |
1809 | dict_mem_table_add_col( |
1810 | new_table, heap, "value" , DATA_VARCHAR, DATA_NOT_NULL, |
1811 | FTS_CONFIG_TABLE_VALUE_COL_LEN); |
1812 | } |
1813 | |
1814 | dict_table_add_system_columns(new_table, heap); |
1815 | error = row_create_table_for_mysql(new_table, trx, |
1816 | FIL_ENCRYPTION_DEFAULT, FIL_DEFAULT_ENCRYPTION_KEY); |
1817 | |
1818 | if (error == DB_SUCCESS) { |
1819 | |
1820 | dict_index_t* index = dict_mem_index_create( |
1821 | new_table, "FTS_COMMON_TABLE_IND" , |
1822 | DICT_UNIQUE|DICT_CLUSTERED, 1); |
1823 | |
1824 | if (!is_config) { |
1825 | dict_mem_index_add_field(index, "doc_id" , 0); |
1826 | } else { |
1827 | dict_mem_index_add_field(index, "key" , 0); |
1828 | } |
1829 | |
1830 | /* We save and restore trx->dict_operation because |
1831 | row_create_index_for_mysql() changes the operation to |
1832 | TRX_DICT_OP_TABLE. */ |
1833 | trx_dict_op_t op = trx_get_dict_operation(trx); |
1834 | |
1835 | error = row_create_index_for_mysql(index, trx, NULL); |
1836 | |
1837 | trx->dict_operation = op; |
1838 | } |
1839 | |
1840 | if (error != DB_SUCCESS) { |
1841 | trx->error_state = error; |
1842 | dict_mem_table_free(new_table); |
1843 | new_table = NULL; |
1844 | ib::warn() << "Failed to create FTS common table " |
1845 | << fts_table_name; |
1846 | } |
1847 | return(new_table); |
1848 | } |
1849 | |
1850 | /** Creates the common auxiliary tables needed for supporting an FTS index |
1851 | on the given table. row_mysql_lock_data_dictionary must have been called |
1852 | before this. |
1853 | The following tables are created. |
1854 | CREATE TABLE $FTS_PREFIX_DELETED |
1855 | (doc_id BIGINT UNSIGNED, UNIQUE CLUSTERED INDEX on doc_id) |
1856 | CREATE TABLE $FTS_PREFIX_DELETED_CACHE |
1857 | (doc_id BIGINT UNSIGNED, UNIQUE CLUSTERED INDEX on doc_id) |
1858 | CREATE TABLE $FTS_PREFIX_BEING_DELETED |
1859 | (doc_id BIGINT UNSIGNED, UNIQUE CLUSTERED INDEX on doc_id) |
1860 | CREATE TABLE $FTS_PREFIX_BEING_DELETED_CACHE |
1861 | (doc_id BIGINT UNSIGNED, UNIQUE CLUSTERED INDEX on doc_id) |
1862 | CREATE TABLE $FTS_PREFIX_CONFIG |
1863 | (key CHAR(50), value CHAR(200), UNIQUE CLUSTERED INDEX on key) |
1864 | @param[in,out] trx transaction |
1865 | @param[in,out] table table with FTS index |
1866 | @param[in] skip_doc_id_index Skip index on doc id |
1867 | @return DB_SUCCESS if succeed */ |
1868 | dberr_t |
1869 | fts_create_common_tables( |
1870 | trx_t* trx, |
1871 | dict_table_t* table, |
1872 | bool skip_doc_id_index) |
1873 | { |
1874 | dberr_t error; |
1875 | que_t* graph; |
1876 | fts_table_t fts_table; |
1877 | mem_heap_t* heap = mem_heap_create(1024); |
1878 | pars_info_t* info; |
1879 | char fts_name[MAX_FULL_NAME_LEN]; |
1880 | char full_name[sizeof(fts_common_tables) / sizeof(char*)] |
1881 | [MAX_FULL_NAME_LEN]; |
1882 | |
1883 | dict_index_t* index = NULL; |
1884 | trx_dict_op_t op; |
1885 | /* common_tables vector is used for dropping FTS common tables |
1886 | on error condition. */ |
1887 | std::vector<dict_table_t*> common_tables; |
1888 | std::vector<dict_table_t*>::const_iterator it; |
1889 | |
1890 | FTS_INIT_FTS_TABLE(&fts_table, NULL, FTS_COMMON_TABLE, table); |
1891 | |
1892 | error = fts_drop_common_tables(trx, &fts_table); |
1893 | |
1894 | if (error != DB_SUCCESS) { |
1895 | |
1896 | goto func_exit; |
1897 | } |
1898 | |
1899 | /* Create the FTS tables that are common to an FTS index. */ |
1900 | for (ulint i = 0; fts_common_tables[i] != NULL; ++i) { |
1901 | |
1902 | fts_table.suffix = fts_common_tables[i]; |
1903 | fts_get_table_name(&fts_table, full_name[i]); |
1904 | dict_table_t* common_table = fts_create_one_common_table( |
1905 | trx, table, full_name[i], fts_table.suffix, heap); |
1906 | |
1907 | if (common_table == NULL) { |
1908 | error = DB_ERROR; |
1909 | goto func_exit; |
1910 | } else { |
1911 | common_tables.push_back(common_table); |
1912 | } |
1913 | |
1914 | mem_heap_empty(heap); |
1915 | |
1916 | DBUG_EXECUTE_IF("ib_fts_aux_table_error" , |
1917 | /* Return error after creating FTS_AUX_CONFIG table. */ |
1918 | if (i == 4) { |
1919 | error = DB_ERROR; |
1920 | goto func_exit; |
1921 | } |
1922 | ); |
1923 | |
1924 | } |
1925 | |
1926 | /* Write the default settings to the config table. */ |
1927 | info = pars_info_create(); |
1928 | |
1929 | fts_table.suffix = "CONFIG" ; |
1930 | fts_get_table_name(&fts_table, fts_name); |
1931 | pars_info_bind_id(info, true, "config_table" , fts_name); |
1932 | |
1933 | graph = fts_parse_sql_no_dict_lock( |
1934 | info, fts_config_table_insert_values_sql); |
1935 | |
1936 | error = fts_eval_sql(trx, graph); |
1937 | |
1938 | que_graph_free(graph); |
1939 | |
1940 | if (error != DB_SUCCESS || skip_doc_id_index) { |
1941 | |
1942 | goto func_exit; |
1943 | } |
1944 | |
1945 | index = dict_mem_index_create(table, FTS_DOC_ID_INDEX_NAME, |
1946 | DICT_UNIQUE, 1); |
1947 | dict_mem_index_add_field(index, FTS_DOC_ID_COL_NAME, 0); |
1948 | |
1949 | op = trx_get_dict_operation(trx); |
1950 | |
1951 | error = row_create_index_for_mysql(index, trx, NULL); |
1952 | |
1953 | trx->dict_operation = op; |
1954 | |
1955 | func_exit: |
1956 | if (error != DB_SUCCESS) { |
1957 | for (it = common_tables.begin(); it != common_tables.end(); |
1958 | ++it) { |
1959 | row_drop_table_for_mysql( |
1960 | (*it)->name.m_name, trx, true, FALSE); |
1961 | } |
1962 | } |
1963 | |
1964 | common_tables.clear(); |
1965 | mem_heap_free(heap); |
1966 | |
1967 | return(error); |
1968 | } |
1969 | |
1970 | /** Create one FTS auxiliary index table for an FTS index. |
1971 | @param[in,out] trx transaction |
1972 | @param[in] index the index instance |
1973 | @param[in] fts_table fts_table structure |
1974 | @param[in,out] heap temporary memory heap |
1975 | @see row_merge_create_fts_sort_index() |
1976 | @return DB_SUCCESS or error code */ |
1977 | static |
1978 | dict_table_t* |
1979 | fts_create_one_index_table( |
1980 | trx_t* trx, |
1981 | const dict_index_t* index, |
1982 | const fts_table_t* fts_table, |
1983 | mem_heap_t* heap) |
1984 | { |
1985 | dict_field_t* field; |
1986 | dict_table_t* new_table = NULL; |
1987 | char table_name[MAX_FULL_NAME_LEN]; |
1988 | dberr_t error; |
1989 | CHARSET_INFO* charset; |
1990 | |
1991 | ut_ad(index->type & DICT_FTS); |
1992 | |
1993 | fts_get_table_name(fts_table, table_name); |
1994 | |
1995 | new_table = fts_create_in_mem_aux_table( |
1996 | table_name, fts_table->table, |
1997 | FTS_AUX_INDEX_TABLE_NUM_COLS); |
1998 | |
1999 | field = dict_index_get_nth_field(index, 0); |
2000 | charset = fts_get_charset(field->col->prtype); |
2001 | |
2002 | dict_mem_table_add_col(new_table, heap, "word" , |
2003 | charset == &my_charset_latin1 |
2004 | ? DATA_VARCHAR : DATA_VARMYSQL, |
2005 | field->col->prtype, |
2006 | FTS_MAX_WORD_LEN_IN_CHAR |
2007 | * unsigned(field->col->mbmaxlen)); |
2008 | |
2009 | dict_mem_table_add_col(new_table, heap, "first_doc_id" , DATA_INT, |
2010 | DATA_NOT_NULL | DATA_UNSIGNED, |
2011 | FTS_INDEX_FIRST_DOC_ID_LEN); |
2012 | |
2013 | dict_mem_table_add_col(new_table, heap, "last_doc_id" , DATA_INT, |
2014 | DATA_NOT_NULL | DATA_UNSIGNED, |
2015 | FTS_INDEX_LAST_DOC_ID_LEN); |
2016 | |
2017 | dict_mem_table_add_col(new_table, heap, "doc_count" , DATA_INT, |
2018 | DATA_NOT_NULL | DATA_UNSIGNED, |
2019 | FTS_INDEX_DOC_COUNT_LEN); |
2020 | |
2021 | /* The precise type calculation is as follows: |
2022 | least signficiant byte: MySQL type code (not applicable for sys cols) |
2023 | second least : DATA_NOT_NULL | DATA_BINARY_TYPE |
2024 | third least : the MySQL charset-collation code (DATA_MTYPE_MAX) */ |
2025 | |
2026 | dict_mem_table_add_col( |
2027 | new_table, heap, "ilist" , DATA_BLOB, |
2028 | (DATA_MTYPE_MAX << 16) | DATA_UNSIGNED | DATA_NOT_NULL, |
2029 | FTS_INDEX_ILIST_LEN); |
2030 | |
2031 | dict_table_add_system_columns(new_table, heap); |
2032 | error = row_create_table_for_mysql(new_table, trx, |
2033 | FIL_ENCRYPTION_DEFAULT, FIL_DEFAULT_ENCRYPTION_KEY); |
2034 | |
2035 | if (error == DB_SUCCESS) { |
2036 | dict_index_t* index = dict_mem_index_create( |
2037 | new_table, "FTS_INDEX_TABLE_IND" , |
2038 | DICT_UNIQUE|DICT_CLUSTERED, 2); |
2039 | dict_mem_index_add_field(index, "word" , 0); |
2040 | dict_mem_index_add_field(index, "first_doc_id" , 0); |
2041 | |
2042 | trx_dict_op_t op = trx_get_dict_operation(trx); |
2043 | |
2044 | error = row_create_index_for_mysql(index, trx, NULL); |
2045 | |
2046 | trx->dict_operation = op; |
2047 | } |
2048 | |
2049 | if (error != DB_SUCCESS) { |
2050 | trx->error_state = error; |
2051 | dict_mem_table_free(new_table); |
2052 | new_table = NULL; |
2053 | ib::warn() << "Failed to create FTS index table " |
2054 | << table_name; |
2055 | } |
2056 | |
2057 | return(new_table); |
2058 | } |
2059 | |
2060 | /** Creates the column specific ancillary tables needed for supporting an |
2061 | FTS index on the given table. row_mysql_lock_data_dictionary must have |
2062 | been called before this. |
2063 | |
2064 | All FTS AUX Index tables have the following schema. |
2065 | CREAT TABLE $FTS_PREFIX_INDEX_[1-6]( |
2066 | word VARCHAR(FTS_MAX_WORD_LEN), |
2067 | first_doc_id INT NOT NULL, |
2068 | last_doc_id UNSIGNED NOT NULL, |
2069 | doc_count UNSIGNED INT NOT NULL, |
2070 | ilist VARBINARY NOT NULL, |
2071 | UNIQUE CLUSTERED INDEX ON (word, first_doc_id)) |
2072 | @param[in,out] trx dictionary transaction |
2073 | @param[in] index fulltext index |
2074 | @param[in] id table id |
2075 | @return DB_SUCCESS or error code */ |
2076 | dberr_t |
2077 | fts_create_index_tables(trx_t* trx, const dict_index_t* index, table_id_t id) |
2078 | { |
2079 | ulint i; |
2080 | fts_table_t fts_table; |
2081 | dberr_t error = DB_SUCCESS; |
2082 | mem_heap_t* heap = mem_heap_create(1024); |
2083 | |
2084 | fts_table.type = FTS_INDEX_TABLE; |
2085 | fts_table.index_id = index->id; |
2086 | fts_table.table_id = id; |
2087 | fts_table.parent = index->table->name.m_name; |
2088 | fts_table.table = index->table; |
2089 | |
2090 | /* aux_idx_tables vector is used for dropping FTS AUX INDEX |
2091 | tables on error condition. */ |
2092 | std::vector<dict_table_t*> aux_idx_tables; |
2093 | std::vector<dict_table_t*>::const_iterator it; |
2094 | |
2095 | for (i = 0; i < FTS_NUM_AUX_INDEX && error == DB_SUCCESS; ++i) { |
2096 | dict_table_t* new_table; |
2097 | |
2098 | /* Create the FTS auxiliary tables that are specific |
2099 | to an FTS index. We need to preserve the table_id %s |
2100 | which fts_parse_sql_no_dict_lock() will fill in for us. */ |
2101 | fts_table.suffix = fts_get_suffix(i); |
2102 | |
2103 | new_table = fts_create_one_index_table( |
2104 | trx, index, &fts_table, heap); |
2105 | |
2106 | if (new_table == NULL) { |
2107 | error = DB_FAIL; |
2108 | break; |
2109 | } else { |
2110 | aux_idx_tables.push_back(new_table); |
2111 | } |
2112 | |
2113 | mem_heap_empty(heap); |
2114 | |
2115 | DBUG_EXECUTE_IF("ib_fts_index_table_error" , |
2116 | /* Return error after creating FTS_INDEX_5 |
2117 | aux table. */ |
2118 | if (i == 4) { |
2119 | error = DB_FAIL; |
2120 | break; |
2121 | } |
2122 | ); |
2123 | } |
2124 | |
2125 | if (error != DB_SUCCESS) { |
2126 | |
2127 | for (it = aux_idx_tables.begin(); it != aux_idx_tables.end(); |
2128 | ++it) { |
2129 | row_drop_table_for_mysql( |
2130 | (*it)->name.m_name, trx, true, FALSE); |
2131 | } |
2132 | } |
2133 | |
2134 | aux_idx_tables.clear(); |
2135 | mem_heap_free(heap); |
2136 | |
2137 | return(error); |
2138 | } |
2139 | |
2140 | #if 0 |
2141 | /******************************************************************//** |
2142 | Return string representation of state. */ |
2143 | static |
2144 | const char* |
2145 | fts_get_state_str( |
2146 | /*==============*/ |
2147 | /* out: string representation of state */ |
2148 | fts_row_state state) /*!< in: state */ |
2149 | { |
2150 | switch (state) { |
2151 | case FTS_INSERT: |
2152 | return("INSERT" ); |
2153 | |
2154 | case FTS_MODIFY: |
2155 | return("MODIFY" ); |
2156 | |
2157 | case FTS_DELETE: |
2158 | return("DELETE" ); |
2159 | |
2160 | case FTS_NOTHING: |
2161 | return("NOTHING" ); |
2162 | |
2163 | case FTS_INVALID: |
2164 | return("INVALID" ); |
2165 | |
2166 | default: |
2167 | return("UNKNOWN" ); |
2168 | } |
2169 | } |
2170 | #endif |
2171 | |
2172 | /******************************************************************//** |
2173 | Calculate the new state of a row given the existing state and a new event. |
2174 | @return new state of row */ |
2175 | static |
2176 | fts_row_state |
2177 | fts_trx_row_get_new_state( |
2178 | /*======================*/ |
2179 | fts_row_state old_state, /*!< in: existing state of row */ |
2180 | fts_row_state event) /*!< in: new event */ |
2181 | { |
2182 | /* The rules for transforming states: |
2183 | |
2184 | I = inserted |
2185 | M = modified |
2186 | D = deleted |
2187 | N = nothing |
2188 | |
2189 | M+D -> D: |
2190 | |
2191 | If the row existed before the transaction started and it is modified |
2192 | during the transaction, followed by a deletion of the row, only the |
2193 | deletion will be signaled. |
2194 | |
2195 | M+ -> M: |
2196 | |
2197 | If the row existed before the transaction started and it is modified |
2198 | more than once during the transaction, only the last modification |
2199 | will be signaled. |
2200 | |
2201 | IM*D -> N: |
2202 | |
2203 | If a new row is added during the transaction (and possibly modified |
2204 | after its initial insertion) but it is deleted before the end of the |
2205 | transaction, nothing will be signaled. |
2206 | |
2207 | IM* -> I: |
2208 | |
2209 | If a new row is added during the transaction and modified after its |
2210 | initial insertion, only the addition will be signaled. |
2211 | |
2212 | M*DI -> M: |
2213 | |
2214 | If the row existed before the transaction started and it is deleted, |
2215 | then re-inserted, only a modification will be signaled. Note that |
2216 | this case is only possible if the table is using the row's primary |
2217 | key for FTS row ids, since those can be re-inserted by the user, |
2218 | which is not true for InnoDB generated row ids. |
2219 | |
2220 | It is easily seen that the above rules decompose such that we do not |
2221 | need to store the row's entire history of events. Instead, we can |
2222 | store just one state for the row and update that when new events |
2223 | arrive. Then we can implement the above rules as a two-dimensional |
2224 | look-up table, and get checking of invalid combinations "for free" |
2225 | in the process. */ |
2226 | |
2227 | /* The lookup table for transforming states. old_state is the |
2228 | Y-axis, event is the X-axis. */ |
2229 | static const fts_row_state table[4][4] = { |
2230 | /* I M D N */ |
2231 | /* I */ { FTS_INVALID, FTS_INSERT, FTS_NOTHING, FTS_INVALID }, |
2232 | /* M */ { FTS_INVALID, FTS_MODIFY, FTS_DELETE, FTS_INVALID }, |
2233 | /* D */ { FTS_MODIFY, FTS_INVALID, FTS_INVALID, FTS_INVALID }, |
2234 | /* N */ { FTS_INVALID, FTS_INVALID, FTS_INVALID, FTS_INVALID } |
2235 | }; |
2236 | |
2237 | fts_row_state result; |
2238 | |
2239 | ut_a(old_state < FTS_INVALID); |
2240 | ut_a(event < FTS_INVALID); |
2241 | |
2242 | result = table[(int) old_state][(int) event]; |
2243 | ut_a(result != FTS_INVALID); |
2244 | |
2245 | return(result); |
2246 | } |
2247 | |
2248 | /******************************************************************//** |
2249 | Create a savepoint instance. |
2250 | @return savepoint instance */ |
2251 | static |
2252 | fts_savepoint_t* |
2253 | fts_savepoint_create( |
2254 | /*=================*/ |
2255 | ib_vector_t* savepoints, /*!< out: InnoDB transaction */ |
2256 | const char* name, /*!< in: savepoint name */ |
2257 | mem_heap_t* heap) /*!< in: heap */ |
2258 | { |
2259 | fts_savepoint_t* savepoint; |
2260 | |
2261 | savepoint = static_cast<fts_savepoint_t*>( |
2262 | ib_vector_push(savepoints, NULL)); |
2263 | |
2264 | memset(savepoint, 0x0, sizeof(*savepoint)); |
2265 | |
2266 | if (name) { |
2267 | savepoint->name = mem_heap_strdup(heap, name); |
2268 | } |
2269 | |
2270 | savepoint->tables = rbt_create( |
2271 | sizeof(fts_trx_table_t*), fts_trx_table_cmp); |
2272 | |
2273 | return(savepoint); |
2274 | } |
2275 | |
2276 | /******************************************************************//** |
2277 | Create an FTS trx. |
2278 | @return FTS trx */ |
2279 | fts_trx_t* |
2280 | fts_trx_create( |
2281 | /*===========*/ |
2282 | trx_t* trx) /*!< in/out: InnoDB |
2283 | transaction */ |
2284 | { |
2285 | fts_trx_t* ftt; |
2286 | ib_alloc_t* heap_alloc; |
2287 | mem_heap_t* heap = mem_heap_create(1024); |
2288 | trx_named_savept_t* savep; |
2289 | |
2290 | ut_a(trx->fts_trx == NULL); |
2291 | |
2292 | ftt = static_cast<fts_trx_t*>(mem_heap_alloc(heap, sizeof(fts_trx_t))); |
2293 | ftt->trx = trx; |
2294 | ftt->heap = heap; |
2295 | |
2296 | heap_alloc = ib_heap_allocator_create(heap); |
2297 | |
2298 | ftt->savepoints = static_cast<ib_vector_t*>(ib_vector_create( |
2299 | heap_alloc, sizeof(fts_savepoint_t), 4)); |
2300 | |
2301 | ftt->last_stmt = static_cast<ib_vector_t*>(ib_vector_create( |
2302 | heap_alloc, sizeof(fts_savepoint_t), 4)); |
2303 | |
2304 | /* Default instance has no name and no heap. */ |
2305 | fts_savepoint_create(ftt->savepoints, NULL, NULL); |
2306 | fts_savepoint_create(ftt->last_stmt, NULL, NULL); |
2307 | |
2308 | /* Copy savepoints that already set before. */ |
2309 | for (savep = UT_LIST_GET_FIRST(trx->trx_savepoints); |
2310 | savep != NULL; |
2311 | savep = UT_LIST_GET_NEXT(trx_savepoints, savep)) { |
2312 | |
2313 | fts_savepoint_take(ftt, savep->name); |
2314 | } |
2315 | |
2316 | return(ftt); |
2317 | } |
2318 | |
2319 | /******************************************************************//** |
2320 | Create an FTS trx table. |
2321 | @return FTS trx table */ |
2322 | static |
2323 | fts_trx_table_t* |
2324 | fts_trx_table_create( |
2325 | /*=================*/ |
2326 | fts_trx_t* fts_trx, /*!< in: FTS trx */ |
2327 | dict_table_t* table) /*!< in: table */ |
2328 | { |
2329 | fts_trx_table_t* ftt; |
2330 | |
2331 | ftt = static_cast<fts_trx_table_t*>( |
2332 | mem_heap_alloc(fts_trx->heap, sizeof(*ftt))); |
2333 | |
2334 | memset(ftt, 0x0, sizeof(*ftt)); |
2335 | |
2336 | ftt->table = table; |
2337 | ftt->fts_trx = fts_trx; |
2338 | |
2339 | ftt->rows = rbt_create(sizeof(fts_trx_row_t), fts_trx_row_doc_id_cmp); |
2340 | |
2341 | return(ftt); |
2342 | } |
2343 | |
2344 | /******************************************************************//** |
2345 | Clone an FTS trx table. |
2346 | @return FTS trx table */ |
2347 | static |
2348 | fts_trx_table_t* |
2349 | fts_trx_table_clone( |
2350 | /*=================*/ |
2351 | const fts_trx_table_t* ftt_src) /*!< in: FTS trx */ |
2352 | { |
2353 | fts_trx_table_t* ftt; |
2354 | |
2355 | ftt = static_cast<fts_trx_table_t*>( |
2356 | mem_heap_alloc(ftt_src->fts_trx->heap, sizeof(*ftt))); |
2357 | |
2358 | memset(ftt, 0x0, sizeof(*ftt)); |
2359 | |
2360 | ftt->table = ftt_src->table; |
2361 | ftt->fts_trx = ftt_src->fts_trx; |
2362 | |
2363 | ftt->rows = rbt_create(sizeof(fts_trx_row_t), fts_trx_row_doc_id_cmp); |
2364 | |
2365 | /* Copy the rb tree values to the new savepoint. */ |
2366 | rbt_merge_uniq(ftt->rows, ftt_src->rows); |
2367 | |
2368 | /* These are only added on commit. At this stage we only have |
2369 | the updated row state. */ |
2370 | ut_a(ftt_src->added_doc_ids == NULL); |
2371 | |
2372 | return(ftt); |
2373 | } |
2374 | |
2375 | /******************************************************************//** |
2376 | Initialize the FTS trx instance. |
2377 | @return FTS trx instance */ |
2378 | static |
2379 | fts_trx_table_t* |
2380 | fts_trx_init( |
2381 | /*=========*/ |
2382 | trx_t* trx, /*!< in: transaction */ |
2383 | dict_table_t* table, /*!< in: FTS table instance */ |
2384 | ib_vector_t* savepoints) /*!< in: Savepoints */ |
2385 | { |
2386 | fts_trx_table_t* ftt; |
2387 | ib_rbt_bound_t parent; |
2388 | ib_rbt_t* tables; |
2389 | fts_savepoint_t* savepoint; |
2390 | |
2391 | savepoint = static_cast<fts_savepoint_t*>(ib_vector_last(savepoints)); |
2392 | |
2393 | tables = savepoint->tables; |
2394 | rbt_search_cmp(tables, &parent, &table->id, fts_trx_table_id_cmp, NULL); |
2395 | |
2396 | if (parent.result == 0) { |
2397 | fts_trx_table_t** fttp; |
2398 | |
2399 | fttp = rbt_value(fts_trx_table_t*, parent.last); |
2400 | ftt = *fttp; |
2401 | } else { |
2402 | ftt = fts_trx_table_create(trx->fts_trx, table); |
2403 | rbt_add_node(tables, &parent, &ftt); |
2404 | } |
2405 | |
2406 | ut_a(ftt->table == table); |
2407 | |
2408 | return(ftt); |
2409 | } |
2410 | |
2411 | /******************************************************************//** |
2412 | Notify the FTS system about an operation on an FTS-indexed table. */ |
2413 | static |
2414 | void |
2415 | fts_trx_table_add_op( |
2416 | /*=================*/ |
2417 | fts_trx_table_t*ftt, /*!< in: FTS trx table */ |
2418 | doc_id_t doc_id, /*!< in: doc id */ |
2419 | fts_row_state state, /*!< in: state of the row */ |
2420 | ib_vector_t* fts_indexes) /*!< in: FTS indexes affected */ |
2421 | { |
2422 | ib_rbt_t* rows; |
2423 | ib_rbt_bound_t parent; |
2424 | |
2425 | rows = ftt->rows; |
2426 | rbt_search(rows, &parent, &doc_id); |
2427 | |
2428 | /* Row id found, update state, and if new state is FTS_NOTHING, |
2429 | we delete the row from our tree. */ |
2430 | if (parent.result == 0) { |
2431 | fts_trx_row_t* row = rbt_value(fts_trx_row_t, parent.last); |
2432 | |
2433 | row->state = fts_trx_row_get_new_state(row->state, state); |
2434 | |
2435 | if (row->state == FTS_NOTHING) { |
2436 | if (row->fts_indexes) { |
2437 | ib_vector_free(row->fts_indexes); |
2438 | } |
2439 | |
2440 | ut_free(rbt_remove_node(rows, parent.last)); |
2441 | row = NULL; |
2442 | } else if (row->fts_indexes != NULL) { |
2443 | ib_vector_free(row->fts_indexes); |
2444 | row->fts_indexes = fts_indexes; |
2445 | } |
2446 | |
2447 | } else { /* Row-id not found, create a new one. */ |
2448 | fts_trx_row_t row; |
2449 | |
2450 | row.doc_id = doc_id; |
2451 | row.state = state; |
2452 | row.fts_indexes = fts_indexes; |
2453 | |
2454 | rbt_add_node(rows, &parent, &row); |
2455 | } |
2456 | } |
2457 | |
2458 | /******************************************************************//** |
2459 | Notify the FTS system about an operation on an FTS-indexed table. */ |
2460 | void |
2461 | fts_trx_add_op( |
2462 | /*===========*/ |
2463 | trx_t* trx, /*!< in: InnoDB transaction */ |
2464 | dict_table_t* table, /*!< in: table */ |
2465 | doc_id_t doc_id, /*!< in: new doc id */ |
2466 | fts_row_state state, /*!< in: state of the row */ |
2467 | ib_vector_t* fts_indexes) /*!< in: FTS indexes affected |
2468 | (NULL=all) */ |
2469 | { |
2470 | fts_trx_table_t* tran_ftt; |
2471 | fts_trx_table_t* stmt_ftt; |
2472 | |
2473 | if (!trx->fts_trx) { |
2474 | trx->fts_trx = fts_trx_create(trx); |
2475 | } |
2476 | |
2477 | tran_ftt = fts_trx_init(trx, table, trx->fts_trx->savepoints); |
2478 | stmt_ftt = fts_trx_init(trx, table, trx->fts_trx->last_stmt); |
2479 | |
2480 | fts_trx_table_add_op(tran_ftt, doc_id, state, fts_indexes); |
2481 | fts_trx_table_add_op(stmt_ftt, doc_id, state, fts_indexes); |
2482 | } |
2483 | |
2484 | /******************************************************************//** |
2485 | Fetch callback that converts a textual document id to a binary value and |
2486 | stores it in the given place. |
2487 | @return always returns NULL */ |
2488 | static |
2489 | ibool |
2490 | fts_fetch_store_doc_id( |
2491 | /*===================*/ |
2492 | void* row, /*!< in: sel_node_t* */ |
2493 | void* user_arg) /*!< in: doc_id_t* to store |
2494 | doc_id in */ |
2495 | { |
2496 | int n_parsed; |
2497 | sel_node_t* node = static_cast<sel_node_t*>(row); |
2498 | doc_id_t* doc_id = static_cast<doc_id_t*>(user_arg); |
2499 | dfield_t* dfield = que_node_get_val(node->select_list); |
2500 | dtype_t* type = dfield_get_type(dfield); |
2501 | ulint len = dfield_get_len(dfield); |
2502 | |
2503 | char buf[32]; |
2504 | |
2505 | ut_a(dtype_get_mtype(type) == DATA_VARCHAR); |
2506 | ut_a(len > 0 && len < sizeof(buf)); |
2507 | |
2508 | memcpy(buf, dfield_get_data(dfield), len); |
2509 | buf[len] = '\0'; |
2510 | |
2511 | n_parsed = sscanf(buf, FTS_DOC_ID_FORMAT, doc_id); |
2512 | ut_a(n_parsed == 1); |
2513 | |
2514 | return(FALSE); |
2515 | } |
2516 | |
2517 | #ifdef FTS_CACHE_SIZE_DEBUG |
2518 | /******************************************************************//** |
2519 | Get the max cache size in bytes. If there is an error reading the |
2520 | value we simply print an error message here and return the default |
2521 | value to the caller. |
2522 | @return max cache size in bytes */ |
2523 | static |
2524 | ulint |
2525 | fts_get_max_cache_size( |
2526 | /*===================*/ |
2527 | trx_t* trx, /*!< in: transaction */ |
2528 | fts_table_t* fts_table) /*!< in: table instance */ |
2529 | { |
2530 | dberr_t error; |
2531 | fts_string_t value; |
2532 | ulong cache_size_in_mb; |
2533 | |
2534 | /* Set to the default value. */ |
2535 | cache_size_in_mb = FTS_CACHE_SIZE_LOWER_LIMIT_IN_MB; |
2536 | |
2537 | /* We set the length of value to the max bytes it can hold. This |
2538 | information is used by the callback that reads the value. */ |
2539 | value.f_n_char = 0; |
2540 | value.f_len = FTS_MAX_CONFIG_VALUE_LEN; |
2541 | value.f_str = ut_malloc_nokey(value.f_len + 1); |
2542 | |
2543 | error = fts_config_get_value( |
2544 | trx, fts_table, FTS_MAX_CACHE_SIZE_IN_MB, &value); |
2545 | |
2546 | if (error == DB_SUCCESS) { |
2547 | |
2548 | value.f_str[value.f_len] = 0; |
2549 | cache_size_in_mb = strtoul((char*) value.f_str, NULL, 10); |
2550 | |
2551 | if (cache_size_in_mb > FTS_CACHE_SIZE_UPPER_LIMIT_IN_MB) { |
2552 | |
2553 | ib::warn() << "FTS max cache size (" |
2554 | << cache_size_in_mb << ") out of range." |
2555 | " Minimum value is " |
2556 | << FTS_CACHE_SIZE_LOWER_LIMIT_IN_MB |
2557 | << "MB and the maximum value is " |
2558 | << FTS_CACHE_SIZE_UPPER_LIMIT_IN_MB |
2559 | << "MB, setting cache size to upper limit" ; |
2560 | |
2561 | cache_size_in_mb = FTS_CACHE_SIZE_UPPER_LIMIT_IN_MB; |
2562 | |
2563 | } else if (cache_size_in_mb |
2564 | < FTS_CACHE_SIZE_LOWER_LIMIT_IN_MB) { |
2565 | |
2566 | ib::warn() << "FTS max cache size (" |
2567 | << cache_size_in_mb << ") out of range." |
2568 | " Minimum value is " |
2569 | << FTS_CACHE_SIZE_LOWER_LIMIT_IN_MB |
2570 | << "MB and the maximum value is" |
2571 | << FTS_CACHE_SIZE_UPPER_LIMIT_IN_MB |
2572 | << "MB, setting cache size to lower limit" ; |
2573 | |
2574 | cache_size_in_mb = FTS_CACHE_SIZE_LOWER_LIMIT_IN_MB; |
2575 | } |
2576 | } else { |
2577 | ib::error() << "(" << ut_strerr(error) << ") reading max" |
2578 | " cache config value from config table" ; |
2579 | } |
2580 | |
2581 | ut_free(value.f_str); |
2582 | |
2583 | return(cache_size_in_mb * 1024 * 1024); |
2584 | } |
2585 | #endif |
2586 | |
2587 | /*********************************************************************//** |
2588 | Update the next and last Doc ID in the CONFIG table to be the input |
2589 | "doc_id" value (+ 1). We would do so after each FTS index build or |
2590 | table truncate */ |
2591 | void |
2592 | fts_update_next_doc_id( |
2593 | /*===================*/ |
2594 | trx_t* trx, /*!< in/out: transaction */ |
2595 | const dict_table_t* table, /*!< in: table */ |
2596 | const char* table_name, /*!< in: table name, or NULL */ |
2597 | doc_id_t doc_id) /*!< in: DOC ID to set */ |
2598 | { |
2599 | table->fts->cache->synced_doc_id = doc_id; |
2600 | table->fts->cache->next_doc_id = doc_id + 1; |
2601 | |
2602 | table->fts->cache->first_doc_id = table->fts->cache->next_doc_id; |
2603 | |
2604 | fts_update_sync_doc_id( |
2605 | table, table_name, table->fts->cache->synced_doc_id, trx); |
2606 | |
2607 | } |
2608 | |
2609 | /*********************************************************************//** |
2610 | Get the next available document id. |
2611 | @return DB_SUCCESS if OK */ |
2612 | dberr_t |
2613 | fts_get_next_doc_id( |
2614 | /*================*/ |
2615 | const dict_table_t* table, /*!< in: table */ |
2616 | doc_id_t* doc_id) /*!< out: new document id */ |
2617 | { |
2618 | fts_cache_t* cache = table->fts->cache; |
2619 | |
2620 | /* If the Doc ID system has not yet been initialized, we |
2621 | will consult the CONFIG table and user table to re-establish |
2622 | the initial value of the Doc ID */ |
2623 | if (cache->first_doc_id == FTS_NULL_DOC_ID) { |
2624 | fts_init_doc_id(table); |
2625 | } |
2626 | |
2627 | if (!DICT_TF2_FLAG_IS_SET(table, DICT_TF2_FTS_HAS_DOC_ID)) { |
2628 | *doc_id = FTS_NULL_DOC_ID; |
2629 | return(DB_SUCCESS); |
2630 | } |
2631 | |
2632 | DEBUG_SYNC_C("get_next_FTS_DOC_ID" ); |
2633 | mutex_enter(&cache->doc_id_lock); |
2634 | *doc_id = cache->next_doc_id++; |
2635 | mutex_exit(&cache->doc_id_lock); |
2636 | |
2637 | return(DB_SUCCESS); |
2638 | } |
2639 | |
2640 | /*********************************************************************//** |
2641 | This function fetch the Doc ID from CONFIG table, and compare with |
2642 | the Doc ID supplied. And store the larger one to the CONFIG table. |
2643 | @return DB_SUCCESS if OK */ |
2644 | static MY_ATTRIBUTE((nonnull)) |
2645 | dberr_t |
2646 | fts_cmp_set_sync_doc_id( |
2647 | /*====================*/ |
2648 | const dict_table_t* table, /*!< in: table */ |
2649 | doc_id_t doc_id_cmp, /*!< in: Doc ID to compare */ |
2650 | ibool read_only, /*!< in: TRUE if read the |
2651 | synced_doc_id only */ |
2652 | doc_id_t* doc_id) /*!< out: larger document id |
2653 | after comparing "doc_id_cmp" |
2654 | to the one stored in CONFIG |
2655 | table */ |
2656 | { |
2657 | trx_t* trx; |
2658 | pars_info_t* info; |
2659 | dberr_t error; |
2660 | fts_table_t fts_table; |
2661 | que_t* graph = NULL; |
2662 | fts_cache_t* cache = table->fts->cache; |
2663 | char table_name[MAX_FULL_NAME_LEN]; |
2664 | retry: |
2665 | ut_a(table->fts->doc_col != ULINT_UNDEFINED); |
2666 | |
2667 | fts_table.suffix = "CONFIG" ; |
2668 | fts_table.table_id = table->id; |
2669 | fts_table.type = FTS_COMMON_TABLE; |
2670 | fts_table.table = table; |
2671 | |
2672 | fts_table.parent = table->name.m_name; |
2673 | |
2674 | trx = trx_create(); |
2675 | if (srv_read_only_mode) { |
2676 | trx_start_internal_read_only(trx); |
2677 | } else { |
2678 | trx_start_internal(trx); |
2679 | } |
2680 | |
2681 | trx->op_info = "update the next FTS document id" ; |
2682 | |
2683 | info = pars_info_create(); |
2684 | |
2685 | pars_info_bind_function( |
2686 | info, "my_func" , fts_fetch_store_doc_id, doc_id); |
2687 | |
2688 | fts_get_table_name(&fts_table, table_name); |
2689 | pars_info_bind_id(info, true, "config_table" , table_name); |
2690 | |
2691 | graph = fts_parse_sql( |
2692 | &fts_table, info, |
2693 | "DECLARE FUNCTION my_func;\n" |
2694 | "DECLARE CURSOR c IS SELECT value FROM $config_table" |
2695 | " WHERE key = 'synced_doc_id' FOR UPDATE;\n" |
2696 | "BEGIN\n" |
2697 | "" |
2698 | "OPEN c;\n" |
2699 | "WHILE 1 = 1 LOOP\n" |
2700 | " FETCH c INTO my_func();\n" |
2701 | " IF c % NOTFOUND THEN\n" |
2702 | " EXIT;\n" |
2703 | " END IF;\n" |
2704 | "END LOOP;\n" |
2705 | "CLOSE c;" ); |
2706 | |
2707 | *doc_id = 0; |
2708 | |
2709 | error = fts_eval_sql(trx, graph); |
2710 | |
2711 | fts_que_graph_free_check_lock(&fts_table, NULL, graph); |
2712 | |
2713 | // FIXME: We need to retry deadlock errors |
2714 | if (error != DB_SUCCESS) { |
2715 | goto func_exit; |
2716 | } |
2717 | |
2718 | if (read_only) { |
2719 | goto func_exit; |
2720 | } |
2721 | |
2722 | if (doc_id_cmp == 0 && *doc_id) { |
2723 | cache->synced_doc_id = *doc_id - 1; |
2724 | } else { |
2725 | cache->synced_doc_id = ut_max(doc_id_cmp, *doc_id); |
2726 | } |
2727 | |
2728 | mutex_enter(&cache->doc_id_lock); |
2729 | /* For each sync operation, we will add next_doc_id by 1, |
2730 | so to mark a sync operation */ |
2731 | if (cache->next_doc_id < cache->synced_doc_id + 1) { |
2732 | cache->next_doc_id = cache->synced_doc_id + 1; |
2733 | } |
2734 | mutex_exit(&cache->doc_id_lock); |
2735 | |
2736 | if (doc_id_cmp > *doc_id) { |
2737 | error = fts_update_sync_doc_id( |
2738 | table, table->name.m_name, cache->synced_doc_id, trx); |
2739 | } |
2740 | |
2741 | *doc_id = cache->next_doc_id; |
2742 | |
2743 | func_exit: |
2744 | |
2745 | if (error == DB_SUCCESS) { |
2746 | fts_sql_commit(trx); |
2747 | } else { |
2748 | *doc_id = 0; |
2749 | |
2750 | ib::error() << "(" << ut_strerr(error) << ") while getting" |
2751 | " next doc id." ; |
2752 | fts_sql_rollback(trx); |
2753 | |
2754 | if (error == DB_DEADLOCK) { |
2755 | os_thread_sleep(FTS_DEADLOCK_RETRY_WAIT); |
2756 | goto retry; |
2757 | } |
2758 | } |
2759 | |
2760 | trx_free(trx); |
2761 | |
2762 | return(error); |
2763 | } |
2764 | |
2765 | /*********************************************************************//** |
2766 | Update the last document id. This function could create a new |
2767 | transaction to update the last document id. |
2768 | @return DB_SUCCESS if OK */ |
2769 | static |
2770 | dberr_t |
2771 | fts_update_sync_doc_id( |
2772 | /*===================*/ |
2773 | const dict_table_t* table, /*!< in: table */ |
2774 | const char* table_name, /*!< in: table name, or NULL */ |
2775 | doc_id_t doc_id, /*!< in: last document id */ |
2776 | trx_t* trx) /*!< in: update trx, or NULL */ |
2777 | { |
2778 | byte id[FTS_MAX_ID_LEN]; |
2779 | pars_info_t* info; |
2780 | fts_table_t fts_table; |
2781 | ulint id_len; |
2782 | que_t* graph = NULL; |
2783 | dberr_t error; |
2784 | ibool local_trx = FALSE; |
2785 | fts_cache_t* cache = table->fts->cache; |
2786 | char fts_name[MAX_FULL_NAME_LEN]; |
2787 | |
2788 | if (srv_read_only_mode) { |
2789 | return DB_READ_ONLY; |
2790 | } |
2791 | |
2792 | fts_table.suffix = "CONFIG" ; |
2793 | fts_table.table_id = table->id; |
2794 | fts_table.type = FTS_COMMON_TABLE; |
2795 | fts_table.table = table; |
2796 | if (table_name) { |
2797 | fts_table.parent = table_name; |
2798 | } else { |
2799 | fts_table.parent = table->name.m_name; |
2800 | } |
2801 | |
2802 | if (!trx) { |
2803 | trx = trx_create(); |
2804 | trx_start_internal(trx); |
2805 | |
2806 | trx->op_info = "setting last FTS document id" ; |
2807 | local_trx = TRUE; |
2808 | } |
2809 | |
2810 | info = pars_info_create(); |
2811 | |
2812 | id_len = (ulint) snprintf( |
2813 | (char*) id, sizeof(id), FTS_DOC_ID_FORMAT, doc_id + 1); |
2814 | |
2815 | pars_info_bind_varchar_literal(info, "doc_id" , id, id_len); |
2816 | |
2817 | fts_get_table_name(&fts_table, fts_name); |
2818 | pars_info_bind_id(info, true, "table_name" , fts_name); |
2819 | |
2820 | graph = fts_parse_sql( |
2821 | &fts_table, info, |
2822 | "BEGIN" |
2823 | " UPDATE $table_name SET value = :doc_id" |
2824 | " WHERE key = 'synced_doc_id';" ); |
2825 | |
2826 | error = fts_eval_sql(trx, graph); |
2827 | |
2828 | fts_que_graph_free_check_lock(&fts_table, NULL, graph); |
2829 | |
2830 | if (local_trx) { |
2831 | if (error == DB_SUCCESS) { |
2832 | fts_sql_commit(trx); |
2833 | cache->synced_doc_id = doc_id; |
2834 | } else { |
2835 | |
2836 | ib::error() << "(" << ut_strerr(error) << ") while" |
2837 | " updating last doc id." ; |
2838 | |
2839 | fts_sql_rollback(trx); |
2840 | } |
2841 | trx_free(trx); |
2842 | } |
2843 | |
2844 | return(error); |
2845 | } |
2846 | |
2847 | /*********************************************************************//** |
2848 | Create a new fts_doc_ids_t. |
2849 | @return new fts_doc_ids_t */ |
2850 | fts_doc_ids_t* |
2851 | fts_doc_ids_create(void) |
2852 | /*====================*/ |
2853 | { |
2854 | fts_doc_ids_t* fts_doc_ids; |
2855 | mem_heap_t* heap = mem_heap_create(512); |
2856 | |
2857 | fts_doc_ids = static_cast<fts_doc_ids_t*>( |
2858 | mem_heap_alloc(heap, sizeof(*fts_doc_ids))); |
2859 | |
2860 | fts_doc_ids->self_heap = ib_heap_allocator_create(heap); |
2861 | |
2862 | fts_doc_ids->doc_ids = static_cast<ib_vector_t*>(ib_vector_create( |
2863 | fts_doc_ids->self_heap, sizeof(fts_update_t), 32)); |
2864 | |
2865 | return(fts_doc_ids); |
2866 | } |
2867 | |
2868 | /*********************************************************************//** |
2869 | Free a fts_doc_ids_t. */ |
2870 | void |
2871 | fts_doc_ids_free( |
2872 | /*=============*/ |
2873 | fts_doc_ids_t* fts_doc_ids) |
2874 | { |
2875 | mem_heap_t* heap = static_cast<mem_heap_t*>( |
2876 | fts_doc_ids->self_heap->arg); |
2877 | |
2878 | memset(fts_doc_ids, 0, sizeof(*fts_doc_ids)); |
2879 | |
2880 | mem_heap_free(heap); |
2881 | } |
2882 | |
2883 | /*********************************************************************//** |
2884 | Do commit-phase steps necessary for the insertion of a new row. */ |
2885 | void |
2886 | fts_add( |
2887 | /*====*/ |
2888 | fts_trx_table_t*ftt, /*!< in: FTS trx table */ |
2889 | fts_trx_row_t* row) /*!< in: row */ |
2890 | { |
2891 | dict_table_t* table = ftt->table; |
2892 | doc_id_t doc_id = row->doc_id; |
2893 | |
2894 | ut_a(row->state == FTS_INSERT || row->state == FTS_MODIFY); |
2895 | |
2896 | fts_add_doc_by_id(ftt, doc_id, row->fts_indexes); |
2897 | |
2898 | mutex_enter(&table->fts->cache->deleted_lock); |
2899 | ++table->fts->cache->added; |
2900 | mutex_exit(&table->fts->cache->deleted_lock); |
2901 | |
2902 | if (!DICT_TF2_FLAG_IS_SET(table, DICT_TF2_FTS_HAS_DOC_ID) |
2903 | && doc_id >= table->fts->cache->next_doc_id) { |
2904 | table->fts->cache->next_doc_id = doc_id + 1; |
2905 | } |
2906 | } |
2907 | |
2908 | /*********************************************************************//** |
2909 | Do commit-phase steps necessary for the deletion of a row. |
2910 | @return DB_SUCCESS or error code */ |
2911 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
2912 | dberr_t |
2913 | fts_delete( |
2914 | /*=======*/ |
2915 | fts_trx_table_t*ftt, /*!< in: FTS trx table */ |
2916 | fts_trx_row_t* row) /*!< in: row */ |
2917 | { |
2918 | que_t* graph; |
2919 | fts_table_t fts_table; |
2920 | dberr_t error = DB_SUCCESS; |
2921 | doc_id_t write_doc_id; |
2922 | dict_table_t* table = ftt->table; |
2923 | doc_id_t doc_id = row->doc_id; |
2924 | trx_t* trx = ftt->fts_trx->trx; |
2925 | pars_info_t* info = pars_info_create(); |
2926 | fts_cache_t* cache = table->fts->cache; |
2927 | |
2928 | /* we do not index Documents whose Doc ID value is 0 */ |
2929 | if (doc_id == FTS_NULL_DOC_ID) { |
2930 | ut_ad(!DICT_TF2_FLAG_IS_SET(table, DICT_TF2_FTS_HAS_DOC_ID)); |
2931 | return(error); |
2932 | } |
2933 | |
2934 | ut_a(row->state == FTS_DELETE || row->state == FTS_MODIFY); |
2935 | |
2936 | FTS_INIT_FTS_TABLE(&fts_table, "DELETED" , FTS_COMMON_TABLE, table); |
2937 | |
2938 | /* Convert to "storage" byte order. */ |
2939 | fts_write_doc_id((byte*) &write_doc_id, doc_id); |
2940 | fts_bind_doc_id(info, "doc_id" , &write_doc_id); |
2941 | |
2942 | /* It is possible we update a record that has not yet been sync-ed |
2943 | into cache from last crash (delete Doc will not initialize the |
2944 | sync). Avoid any added counter accounting until the FTS cache |
2945 | is re-established and sync-ed */ |
2946 | if (table->fts->fts_status & ADDED_TABLE_SYNCED |
2947 | && doc_id > cache->synced_doc_id) { |
2948 | mutex_enter(&table->fts->cache->deleted_lock); |
2949 | |
2950 | /* The Doc ID could belong to those left in |
2951 | ADDED table from last crash. So need to check |
2952 | if it is less than first_doc_id when we initialize |
2953 | the Doc ID system after reboot */ |
2954 | if (doc_id >= table->fts->cache->first_doc_id |
2955 | && table->fts->cache->added > 0) { |
2956 | --table->fts->cache->added; |
2957 | } |
2958 | |
2959 | mutex_exit(&table->fts->cache->deleted_lock); |
2960 | |
2961 | /* Only if the row was really deleted. */ |
2962 | ut_a(row->state == FTS_DELETE || row->state == FTS_MODIFY); |
2963 | } |
2964 | |
2965 | /* Note the deleted document for OPTIMIZE to purge. */ |
2966 | if (error == DB_SUCCESS) { |
2967 | char table_name[MAX_FULL_NAME_LEN]; |
2968 | |
2969 | trx->op_info = "adding doc id to FTS DELETED" ; |
2970 | |
2971 | info->graph_owns_us = TRUE; |
2972 | |
2973 | fts_table.suffix = "DELETED" ; |
2974 | |
2975 | fts_get_table_name(&fts_table, table_name); |
2976 | pars_info_bind_id(info, true, "deleted" , table_name); |
2977 | |
2978 | graph = fts_parse_sql( |
2979 | &fts_table, |
2980 | info, |
2981 | "BEGIN INSERT INTO $deleted VALUES (:doc_id);" ); |
2982 | |
2983 | error = fts_eval_sql(trx, graph); |
2984 | |
2985 | fts_que_graph_free(graph); |
2986 | } else { |
2987 | pars_info_free(info); |
2988 | } |
2989 | |
2990 | /* Increment the total deleted count, this is used to calculate the |
2991 | number of documents indexed. */ |
2992 | if (error == DB_SUCCESS) { |
2993 | mutex_enter(&table->fts->cache->deleted_lock); |
2994 | |
2995 | ++table->fts->cache->deleted; |
2996 | |
2997 | mutex_exit(&table->fts->cache->deleted_lock); |
2998 | } |
2999 | |
3000 | return(error); |
3001 | } |
3002 | |
3003 | /*********************************************************************//** |
3004 | Do commit-phase steps necessary for the modification of a row. |
3005 | @return DB_SUCCESS or error code */ |
3006 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
3007 | dberr_t |
3008 | fts_modify( |
3009 | /*=======*/ |
3010 | fts_trx_table_t* ftt, /*!< in: FTS trx table */ |
3011 | fts_trx_row_t* row) /*!< in: row */ |
3012 | { |
3013 | dberr_t error; |
3014 | |
3015 | ut_a(row->state == FTS_MODIFY); |
3016 | |
3017 | error = fts_delete(ftt, row); |
3018 | |
3019 | if (error == DB_SUCCESS) { |
3020 | fts_add(ftt, row); |
3021 | } |
3022 | |
3023 | return(error); |
3024 | } |
3025 | |
3026 | /*********************************************************************//** |
3027 | The given transaction is about to be committed; do whatever is necessary |
3028 | from the FTS system's POV. |
3029 | @return DB_SUCCESS or error code */ |
3030 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
3031 | dberr_t |
3032 | fts_commit_table( |
3033 | /*=============*/ |
3034 | fts_trx_table_t* ftt) /*!< in: FTS table to commit*/ |
3035 | { |
3036 | if (srv_read_only_mode) { |
3037 | return DB_READ_ONLY; |
3038 | } |
3039 | |
3040 | const ib_rbt_node_t* node; |
3041 | ib_rbt_t* rows; |
3042 | dberr_t error = DB_SUCCESS; |
3043 | fts_cache_t* cache = ftt->table->fts->cache; |
3044 | trx_t* trx = trx_create(); |
3045 | |
3046 | trx_start_internal(trx); |
3047 | |
3048 | rows = ftt->rows; |
3049 | |
3050 | ftt->fts_trx->trx = trx; |
3051 | |
3052 | if (cache->get_docs == NULL) { |
3053 | rw_lock_x_lock(&cache->init_lock); |
3054 | if (cache->get_docs == NULL) { |
3055 | cache->get_docs = fts_get_docs_create(cache); |
3056 | } |
3057 | rw_lock_x_unlock(&cache->init_lock); |
3058 | } |
3059 | |
3060 | for (node = rbt_first(rows); |
3061 | node != NULL && error == DB_SUCCESS; |
3062 | node = rbt_next(rows, node)) { |
3063 | |
3064 | fts_trx_row_t* row = rbt_value(fts_trx_row_t, node); |
3065 | |
3066 | switch (row->state) { |
3067 | case FTS_INSERT: |
3068 | fts_add(ftt, row); |
3069 | break; |
3070 | |
3071 | case FTS_MODIFY: |
3072 | error = fts_modify(ftt, row); |
3073 | break; |
3074 | |
3075 | case FTS_DELETE: |
3076 | error = fts_delete(ftt, row); |
3077 | break; |
3078 | |
3079 | default: |
3080 | ut_error; |
3081 | } |
3082 | } |
3083 | |
3084 | fts_sql_commit(trx); |
3085 | |
3086 | trx_free(trx); |
3087 | |
3088 | return(error); |
3089 | } |
3090 | |
3091 | /*********************************************************************//** |
3092 | The given transaction is about to be committed; do whatever is necessary |
3093 | from the FTS system's POV. |
3094 | @return DB_SUCCESS or error code */ |
3095 | dberr_t |
3096 | fts_commit( |
3097 | /*=======*/ |
3098 | trx_t* trx) /*!< in: transaction */ |
3099 | { |
3100 | const ib_rbt_node_t* node; |
3101 | dberr_t error; |
3102 | ib_rbt_t* tables; |
3103 | fts_savepoint_t* savepoint; |
3104 | |
3105 | savepoint = static_cast<fts_savepoint_t*>( |
3106 | ib_vector_last(trx->fts_trx->savepoints)); |
3107 | tables = savepoint->tables; |
3108 | |
3109 | for (node = rbt_first(tables), error = DB_SUCCESS; |
3110 | node != NULL && error == DB_SUCCESS; |
3111 | node = rbt_next(tables, node)) { |
3112 | |
3113 | fts_trx_table_t** ftt; |
3114 | |
3115 | ftt = rbt_value(fts_trx_table_t*, node); |
3116 | |
3117 | error = fts_commit_table(*ftt); |
3118 | } |
3119 | |
3120 | return(error); |
3121 | } |
3122 | |
3123 | /*********************************************************************//** |
3124 | Initialize a document. */ |
3125 | void |
3126 | fts_doc_init( |
3127 | /*=========*/ |
3128 | fts_doc_t* doc) /*!< in: doc to initialize */ |
3129 | { |
3130 | mem_heap_t* heap = mem_heap_create(32); |
3131 | |
3132 | memset(doc, 0, sizeof(*doc)); |
3133 | |
3134 | doc->self_heap = ib_heap_allocator_create(heap); |
3135 | } |
3136 | |
3137 | /*********************************************************************//** |
3138 | Free document. */ |
3139 | void |
3140 | fts_doc_free( |
3141 | /*=========*/ |
3142 | fts_doc_t* doc) /*!< in: document */ |
3143 | { |
3144 | mem_heap_t* heap = static_cast<mem_heap_t*>(doc->self_heap->arg); |
3145 | |
3146 | if (doc->tokens) { |
3147 | rbt_free(doc->tokens); |
3148 | } |
3149 | |
3150 | ut_d(memset(doc, 0, sizeof(*doc))); |
3151 | |
3152 | mem_heap_free(heap); |
3153 | } |
3154 | |
3155 | /*********************************************************************//** |
3156 | Callback function for fetch that stores the text of an FTS document, |
3157 | converting each column to UTF-16. |
3158 | @return always FALSE */ |
3159 | ibool |
3160 | fts_query_expansion_fetch_doc( |
3161 | /*==========================*/ |
3162 | void* row, /*!< in: sel_node_t* */ |
3163 | void* user_arg) /*!< in: fts_doc_t* */ |
3164 | { |
3165 | que_node_t* exp; |
3166 | sel_node_t* node = static_cast<sel_node_t*>(row); |
3167 | fts_doc_t* result_doc = static_cast<fts_doc_t*>(user_arg); |
3168 | dfield_t* dfield; |
3169 | ulint len; |
3170 | ulint doc_len; |
3171 | fts_doc_t doc; |
3172 | CHARSET_INFO* doc_charset = NULL; |
3173 | ulint field_no = 0; |
3174 | |
3175 | len = 0; |
3176 | |
3177 | fts_doc_init(&doc); |
3178 | doc.found = TRUE; |
3179 | |
3180 | exp = node->select_list; |
3181 | doc_len = 0; |
3182 | |
3183 | doc_charset = result_doc->charset; |
3184 | |
3185 | /* Copy each indexed column content into doc->text.f_str */ |
3186 | while (exp) { |
3187 | dfield = que_node_get_val(exp); |
3188 | len = dfield_get_len(dfield); |
3189 | |
3190 | /* NULL column */ |
3191 | if (len == UNIV_SQL_NULL) { |
3192 | exp = que_node_get_next(exp); |
3193 | continue; |
3194 | } |
3195 | |
3196 | if (!doc_charset) { |
3197 | doc_charset = fts_get_charset(dfield->type.prtype); |
3198 | } |
3199 | |
3200 | doc.charset = doc_charset; |
3201 | |
3202 | if (dfield_is_ext(dfield)) { |
3203 | /* We ignore columns that are stored externally, this |
3204 | could result in too many words to search */ |
3205 | exp = que_node_get_next(exp); |
3206 | continue; |
3207 | } else { |
3208 | doc.text.f_n_char = 0; |
3209 | |
3210 | doc.text.f_str = static_cast<byte*>( |
3211 | dfield_get_data(dfield)); |
3212 | |
3213 | doc.text.f_len = len; |
3214 | } |
3215 | |
3216 | if (field_no == 0) { |
3217 | fts_tokenize_document(&doc, result_doc, |
3218 | result_doc->parser); |
3219 | } else { |
3220 | fts_tokenize_document_next(&doc, doc_len, result_doc, |
3221 | result_doc->parser); |
3222 | } |
3223 | |
3224 | exp = que_node_get_next(exp); |
3225 | |
3226 | doc_len += (exp) ? len + 1 : len; |
3227 | |
3228 | field_no++; |
3229 | } |
3230 | |
3231 | ut_ad(doc_charset); |
3232 | |
3233 | if (!result_doc->charset) { |
3234 | result_doc->charset = doc_charset; |
3235 | } |
3236 | |
3237 | fts_doc_free(&doc); |
3238 | |
3239 | return(FALSE); |
3240 | } |
3241 | |
3242 | /*********************************************************************//** |
3243 | fetch and tokenize the document. */ |
3244 | static |
3245 | void |
3246 | fts_fetch_doc_from_rec( |
3247 | /*===================*/ |
3248 | fts_get_doc_t* get_doc, /*!< in: FTS index's get_doc struct */ |
3249 | dict_index_t* clust_index, /*!< in: cluster index */ |
3250 | btr_pcur_t* pcur, /*!< in: cursor whose position |
3251 | has been stored */ |
3252 | ulint* offsets, /*!< in: offsets */ |
3253 | fts_doc_t* doc) /*!< out: fts doc to hold parsed |
3254 | documents */ |
3255 | { |
3256 | dict_index_t* index; |
3257 | dict_table_t* table; |
3258 | const rec_t* clust_rec; |
3259 | ulint num_field; |
3260 | const dict_field_t* ifield; |
3261 | const dict_col_t* col; |
3262 | ulint clust_pos; |
3263 | ulint i; |
3264 | ulint doc_len = 0; |
3265 | ulint processed_doc = 0; |
3266 | st_mysql_ftparser* parser; |
3267 | |
3268 | if (!get_doc) { |
3269 | return; |
3270 | } |
3271 | |
3272 | index = get_doc->index_cache->index; |
3273 | table = get_doc->index_cache->index->table; |
3274 | parser = get_doc->index_cache->index->parser; |
3275 | |
3276 | clust_rec = btr_pcur_get_rec(pcur); |
3277 | ut_ad(!page_rec_is_comp(clust_rec) |
3278 | || rec_get_status(clust_rec) == REC_STATUS_ORDINARY); |
3279 | |
3280 | num_field = dict_index_get_n_fields(index); |
3281 | |
3282 | for (i = 0; i < num_field; i++) { |
3283 | ifield = dict_index_get_nth_field(index, i); |
3284 | col = dict_field_get_col(ifield); |
3285 | clust_pos = dict_col_get_clust_pos(col, clust_index); |
3286 | |
3287 | if (!get_doc->index_cache->charset) { |
3288 | get_doc->index_cache->charset = fts_get_charset( |
3289 | ifield->col->prtype); |
3290 | } |
3291 | |
3292 | if (rec_offs_nth_extern(offsets, clust_pos)) { |
3293 | doc->text.f_str = |
3294 | btr_rec_copy_externally_stored_field( |
3295 | clust_rec, offsets, |
3296 | dict_table_page_size(table), |
3297 | clust_pos, &doc->text.f_len, |
3298 | static_cast<mem_heap_t*>( |
3299 | doc->self_heap->arg)); |
3300 | } else { |
3301 | doc->text.f_str = (byte*) rec_get_nth_field( |
3302 | clust_rec, offsets, clust_pos, |
3303 | &doc->text.f_len); |
3304 | } |
3305 | |
3306 | doc->found = TRUE; |
3307 | doc->charset = get_doc->index_cache->charset; |
3308 | |
3309 | /* Null Field */ |
3310 | if (doc->text.f_len == UNIV_SQL_NULL || doc->text.f_len == 0) { |
3311 | continue; |
3312 | } |
3313 | |
3314 | if (processed_doc == 0) { |
3315 | fts_tokenize_document(doc, NULL, parser); |
3316 | } else { |
3317 | fts_tokenize_document_next(doc, doc_len, NULL, parser); |
3318 | } |
3319 | |
3320 | processed_doc++; |
3321 | doc_len += doc->text.f_len + 1; |
3322 | } |
3323 | } |
3324 | |
3325 | /** Fetch the data from tuple and tokenize the document. |
3326 | @param[in] get_doc FTS index's get_doc struct |
3327 | @param[in] tuple tuple should be arranged in table schema order |
3328 | @param[out] doc fts doc to hold parsed documents. */ |
3329 | static |
3330 | void |
3331 | fts_fetch_doc_from_tuple( |
3332 | fts_get_doc_t* get_doc, |
3333 | const dtuple_t* tuple, |
3334 | fts_doc_t* doc) |
3335 | { |
3336 | dict_index_t* index; |
3337 | st_mysql_ftparser* parser; |
3338 | ulint doc_len = 0; |
3339 | ulint processed_doc = 0; |
3340 | ulint num_field; |
3341 | |
3342 | if (get_doc == NULL) { |
3343 | return; |
3344 | } |
3345 | |
3346 | index = get_doc->index_cache->index; |
3347 | parser = get_doc->index_cache->index->parser; |
3348 | num_field = dict_index_get_n_fields(index); |
3349 | |
3350 | for (ulint i = 0; i < num_field; i++) { |
3351 | const dict_field_t* ifield; |
3352 | const dict_col_t* col; |
3353 | ulint pos; |
3354 | dfield_t* field; |
3355 | |
3356 | ifield = dict_index_get_nth_field(index, i); |
3357 | col = dict_field_get_col(ifield); |
3358 | pos = dict_col_get_no(col); |
3359 | field = dtuple_get_nth_field(tuple, pos); |
3360 | |
3361 | if (!get_doc->index_cache->charset) { |
3362 | get_doc->index_cache->charset = fts_get_charset( |
3363 | ifield->col->prtype); |
3364 | } |
3365 | |
3366 | ut_ad(!dfield_is_ext(field)); |
3367 | |
3368 | doc->text.f_str = (byte*) dfield_get_data(field); |
3369 | doc->text.f_len = dfield_get_len(field); |
3370 | doc->found = TRUE; |
3371 | doc->charset = get_doc->index_cache->charset; |
3372 | |
3373 | /* field data is NULL. */ |
3374 | if (doc->text.f_len == UNIV_SQL_NULL || doc->text.f_len == 0) { |
3375 | continue; |
3376 | } |
3377 | |
3378 | if (processed_doc == 0) { |
3379 | fts_tokenize_document(doc, NULL, parser); |
3380 | } else { |
3381 | fts_tokenize_document_next(doc, doc_len, NULL, parser); |
3382 | } |
3383 | |
3384 | processed_doc++; |
3385 | doc_len += doc->text.f_len + 1; |
3386 | } |
3387 | } |
3388 | |
3389 | /** Fetch the document from tuple, tokenize the text data and |
3390 | insert the text data into fts auxiliary table and |
3391 | its cache. Moreover this tuple fields doesn't contain any information |
3392 | about externally stored field. This tuple contains data directly |
3393 | converted from mysql. |
3394 | @param[in] ftt FTS transaction table |
3395 | @param[in] doc_id doc id |
3396 | @param[in] tuple tuple from where data can be retrieved |
3397 | and tuple should be arranged in table |
3398 | schema order. */ |
3399 | void |
3400 | fts_add_doc_from_tuple( |
3401 | fts_trx_table_t*ftt, |
3402 | doc_id_t doc_id, |
3403 | const dtuple_t* tuple) |
3404 | { |
3405 | mtr_t mtr; |
3406 | fts_cache_t* cache = ftt->table->fts->cache; |
3407 | |
3408 | ut_ad(cache->get_docs); |
3409 | |
3410 | if (!(ftt->table->fts->fts_status & ADDED_TABLE_SYNCED)) { |
3411 | fts_init_index(ftt->table, FALSE); |
3412 | } |
3413 | |
3414 | mtr_start(&mtr); |
3415 | |
3416 | ulint num_idx = ib_vector_size(cache->get_docs); |
3417 | |
3418 | for (ulint i = 0; i < num_idx; ++i) { |
3419 | fts_doc_t doc; |
3420 | dict_table_t* table; |
3421 | fts_get_doc_t* get_doc; |
3422 | |
3423 | get_doc = static_cast<fts_get_doc_t*>( |
3424 | ib_vector_get(cache->get_docs, i)); |
3425 | table = get_doc->index_cache->index->table; |
3426 | |
3427 | fts_doc_init(&doc); |
3428 | fts_fetch_doc_from_tuple( |
3429 | get_doc, tuple, &doc); |
3430 | |
3431 | if (doc.found) { |
3432 | mtr_commit(&mtr); |
3433 | rw_lock_x_lock(&table->fts->cache->lock); |
3434 | |
3435 | if (table->fts->cache->stopword_info.status |
3436 | & STOPWORD_NOT_INIT) { |
3437 | fts_load_stopword(table, NULL, NULL, |
3438 | NULL, TRUE, TRUE); |
3439 | } |
3440 | |
3441 | fts_cache_add_doc( |
3442 | table->fts->cache, |
3443 | get_doc->index_cache, |
3444 | doc_id, doc.tokens); |
3445 | |
3446 | rw_lock_x_unlock(&table->fts->cache->lock); |
3447 | |
3448 | if (cache->total_size > fts_max_cache_size / 5 |
3449 | || fts_need_sync) { |
3450 | fts_sync(cache->sync, true, false, false); |
3451 | } |
3452 | |
3453 | mtr_start(&mtr); |
3454 | |
3455 | } |
3456 | |
3457 | fts_doc_free(&doc); |
3458 | } |
3459 | |
3460 | mtr_commit(&mtr); |
3461 | } |
3462 | |
3463 | /*********************************************************************//** |
3464 | This function fetches the document inserted during the committing |
3465 | transaction, and tokenize the inserted text data and insert into |
3466 | FTS auxiliary table and its cache. |
3467 | @return TRUE if successful */ |
3468 | static |
3469 | ulint |
3470 | fts_add_doc_by_id( |
3471 | /*==============*/ |
3472 | fts_trx_table_t*ftt, /*!< in: FTS trx table */ |
3473 | doc_id_t doc_id, /*!< in: doc id */ |
3474 | ib_vector_t* fts_indexes MY_ATTRIBUTE((unused))) |
3475 | /*!< in: affected fts indexes */ |
3476 | { |
3477 | mtr_t mtr; |
3478 | mem_heap_t* heap; |
3479 | btr_pcur_t pcur; |
3480 | dict_table_t* table; |
3481 | dtuple_t* tuple; |
3482 | dfield_t* dfield; |
3483 | fts_get_doc_t* get_doc; |
3484 | doc_id_t temp_doc_id; |
3485 | dict_index_t* clust_index; |
3486 | dict_index_t* fts_id_index; |
3487 | ibool is_id_cluster; |
3488 | fts_cache_t* cache = ftt->table->fts->cache; |
3489 | |
3490 | ut_ad(cache->get_docs); |
3491 | |
3492 | /* If Doc ID has been supplied by the user, then the table |
3493 | might not yet be sync-ed */ |
3494 | |
3495 | if (!(ftt->table->fts->fts_status & ADDED_TABLE_SYNCED)) { |
3496 | fts_init_index(ftt->table, FALSE); |
3497 | } |
3498 | |
3499 | /* Get the first FTS index's get_doc */ |
3500 | get_doc = static_cast<fts_get_doc_t*>( |
3501 | ib_vector_get(cache->get_docs, 0)); |
3502 | ut_ad(get_doc); |
3503 | |
3504 | table = get_doc->index_cache->index->table; |
3505 | |
3506 | heap = mem_heap_create(512); |
3507 | |
3508 | clust_index = dict_table_get_first_index(table); |
3509 | fts_id_index = table->fts_doc_id_index; |
3510 | |
3511 | /* Check whether the index on FTS_DOC_ID is cluster index */ |
3512 | is_id_cluster = (clust_index == fts_id_index); |
3513 | |
3514 | mtr_start(&mtr); |
3515 | btr_pcur_init(&pcur); |
3516 | |
3517 | /* Search based on Doc ID. Here, we'll need to consider the case |
3518 | when there is no primary index on Doc ID */ |
3519 | tuple = dtuple_create(heap, 1); |
3520 | dfield = dtuple_get_nth_field(tuple, 0); |
3521 | dfield->type.mtype = DATA_INT; |
3522 | dfield->type.prtype = DATA_NOT_NULL | DATA_UNSIGNED | DATA_BINARY_TYPE; |
3523 | |
3524 | mach_write_to_8((byte*) &temp_doc_id, doc_id); |
3525 | dfield_set_data(dfield, &temp_doc_id, sizeof(temp_doc_id)); |
3526 | |
3527 | btr_pcur_open_with_no_init( |
3528 | fts_id_index, tuple, PAGE_CUR_LE, BTR_SEARCH_LEAF, |
3529 | &pcur, 0, &mtr); |
3530 | |
3531 | /* If we have a match, add the data to doc structure */ |
3532 | if (btr_pcur_get_low_match(&pcur) == 1) { |
3533 | const rec_t* rec; |
3534 | btr_pcur_t* doc_pcur; |
3535 | const rec_t* clust_rec; |
3536 | btr_pcur_t clust_pcur; |
3537 | ulint* offsets = NULL; |
3538 | ulint num_idx = ib_vector_size(cache->get_docs); |
3539 | |
3540 | rec = btr_pcur_get_rec(&pcur); |
3541 | |
3542 | /* Doc could be deleted */ |
3543 | if (page_rec_is_infimum(rec) |
3544 | || rec_get_deleted_flag(rec, dict_table_is_comp(table))) { |
3545 | |
3546 | goto func_exit; |
3547 | } |
3548 | |
3549 | if (is_id_cluster) { |
3550 | clust_rec = rec; |
3551 | doc_pcur = &pcur; |
3552 | } else { |
3553 | dtuple_t* clust_ref; |
3554 | ulint n_fields; |
3555 | |
3556 | btr_pcur_init(&clust_pcur); |
3557 | n_fields = dict_index_get_n_unique(clust_index); |
3558 | |
3559 | clust_ref = dtuple_create(heap, n_fields); |
3560 | dict_index_copy_types(clust_ref, clust_index, n_fields); |
3561 | |
3562 | row_build_row_ref_in_tuple( |
3563 | clust_ref, rec, fts_id_index, NULL); |
3564 | |
3565 | btr_pcur_open_with_no_init( |
3566 | clust_index, clust_ref, PAGE_CUR_LE, |
3567 | BTR_SEARCH_LEAF, &clust_pcur, 0, &mtr); |
3568 | |
3569 | doc_pcur = &clust_pcur; |
3570 | clust_rec = btr_pcur_get_rec(&clust_pcur); |
3571 | |
3572 | } |
3573 | |
3574 | offsets = rec_get_offsets(clust_rec, clust_index, NULL, true, |
3575 | ULINT_UNDEFINED, &heap); |
3576 | |
3577 | for (ulint i = 0; i < num_idx; ++i) { |
3578 | fts_doc_t doc; |
3579 | dict_table_t* table; |
3580 | fts_get_doc_t* get_doc; |
3581 | |
3582 | get_doc = static_cast<fts_get_doc_t*>( |
3583 | ib_vector_get(cache->get_docs, i)); |
3584 | |
3585 | table = get_doc->index_cache->index->table; |
3586 | |
3587 | fts_doc_init(&doc); |
3588 | |
3589 | fts_fetch_doc_from_rec( |
3590 | get_doc, clust_index, doc_pcur, offsets, &doc); |
3591 | |
3592 | if (doc.found) { |
3593 | ibool success MY_ATTRIBUTE((unused)); |
3594 | |
3595 | btr_pcur_store_position(doc_pcur, &mtr); |
3596 | mtr_commit(&mtr); |
3597 | |
3598 | rw_lock_x_lock(&table->fts->cache->lock); |
3599 | |
3600 | if (table->fts->cache->stopword_info.status |
3601 | & STOPWORD_NOT_INIT) { |
3602 | fts_load_stopword(table, NULL, NULL, |
3603 | NULL, TRUE, TRUE); |
3604 | } |
3605 | |
3606 | fts_cache_add_doc( |
3607 | table->fts->cache, |
3608 | get_doc->index_cache, |
3609 | doc_id, doc.tokens); |
3610 | |
3611 | bool need_sync = false; |
3612 | if ((cache->total_size > fts_max_cache_size / 10 |
3613 | || fts_need_sync) |
3614 | && !cache->sync->in_progress) { |
3615 | need_sync = true; |
3616 | } |
3617 | |
3618 | rw_lock_x_unlock(&table->fts->cache->lock); |
3619 | |
3620 | DBUG_EXECUTE_IF( |
3621 | "fts_instrument_sync" , |
3622 | fts_optimize_request_sync_table(table); |
3623 | os_event_wait(cache->sync->event); |
3624 | ); |
3625 | |
3626 | DBUG_EXECUTE_IF( |
3627 | "fts_instrument_sync_debug" , |
3628 | fts_sync(cache->sync, true, true, false); |
3629 | ); |
3630 | |
3631 | DEBUG_SYNC_C("fts_instrument_sync_request" ); |
3632 | DBUG_EXECUTE_IF( |
3633 | "fts_instrument_sync_request" , |
3634 | fts_optimize_request_sync_table(table); |
3635 | ); |
3636 | |
3637 | if (need_sync) { |
3638 | fts_optimize_request_sync_table(table); |
3639 | } |
3640 | |
3641 | mtr_start(&mtr); |
3642 | |
3643 | if (i < num_idx - 1) { |
3644 | |
3645 | success = btr_pcur_restore_position( |
3646 | BTR_SEARCH_LEAF, doc_pcur, |
3647 | &mtr); |
3648 | |
3649 | ut_ad(success); |
3650 | } |
3651 | } |
3652 | |
3653 | fts_doc_free(&doc); |
3654 | } |
3655 | |
3656 | if (!is_id_cluster) { |
3657 | btr_pcur_close(doc_pcur); |
3658 | } |
3659 | } |
3660 | func_exit: |
3661 | mtr_commit(&mtr); |
3662 | |
3663 | btr_pcur_close(&pcur); |
3664 | |
3665 | mem_heap_free(heap); |
3666 | return(TRUE); |
3667 | } |
3668 | |
3669 | |
3670 | /*********************************************************************//** |
3671 | Callback function to read a single ulint column. |
3672 | return always returns TRUE */ |
3673 | static |
3674 | ibool |
3675 | fts_read_ulint( |
3676 | /*===========*/ |
3677 | void* row, /*!< in: sel_node_t* */ |
3678 | void* user_arg) /*!< in: pointer to ulint */ |
3679 | { |
3680 | sel_node_t* sel_node = static_cast<sel_node_t*>(row); |
3681 | ulint* value = static_cast<ulint*>(user_arg); |
3682 | que_node_t* exp = sel_node->select_list; |
3683 | dfield_t* dfield = que_node_get_val(exp); |
3684 | void* data = dfield_get_data(dfield); |
3685 | |
3686 | *value = static_cast<ulint>(mach_read_from_4( |
3687 | static_cast<const byte*>(data))); |
3688 | |
3689 | return(TRUE); |
3690 | } |
3691 | |
3692 | /*********************************************************************//** |
3693 | Get maximum Doc ID in a table if index "FTS_DOC_ID_INDEX" exists |
3694 | @return max Doc ID or 0 if index "FTS_DOC_ID_INDEX" does not exist */ |
3695 | doc_id_t |
3696 | fts_get_max_doc_id( |
3697 | /*===============*/ |
3698 | dict_table_t* table) /*!< in: user table */ |
3699 | { |
3700 | dict_index_t* index; |
3701 | dict_field_t* dfield MY_ATTRIBUTE((unused)) = NULL; |
3702 | doc_id_t doc_id = 0; |
3703 | mtr_t mtr; |
3704 | btr_pcur_t pcur; |
3705 | |
3706 | index = table->fts_doc_id_index; |
3707 | |
3708 | if (!index) { |
3709 | return(0); |
3710 | } |
3711 | |
3712 | ut_ad(!index->is_instant()); |
3713 | |
3714 | dfield = dict_index_get_nth_field(index, 0); |
3715 | |
3716 | #if 0 /* This can fail when renaming a column to FTS_DOC_ID_COL_NAME. */ |
3717 | ut_ad(innobase_strcasecmp(FTS_DOC_ID_COL_NAME, dfield->name) == 0); |
3718 | #endif |
3719 | |
3720 | mtr_start(&mtr); |
3721 | |
3722 | /* fetch the largest indexes value */ |
3723 | btr_pcur_open_at_index_side( |
3724 | false, index, BTR_SEARCH_LEAF, &pcur, true, 0, &mtr); |
3725 | |
3726 | if (!page_is_empty(btr_pcur_get_page(&pcur))) { |
3727 | const rec_t* rec = NULL; |
3728 | ulint offsets_[REC_OFFS_NORMAL_SIZE]; |
3729 | ulint* offsets = offsets_; |
3730 | mem_heap_t* heap = NULL; |
3731 | ulint len; |
3732 | const void* data; |
3733 | |
3734 | rec_offs_init(offsets_); |
3735 | |
3736 | do { |
3737 | rec = btr_pcur_get_rec(&pcur); |
3738 | |
3739 | if (page_rec_is_user_rec(rec)) { |
3740 | break; |
3741 | } |
3742 | } while (btr_pcur_move_to_prev(&pcur, &mtr)); |
3743 | |
3744 | if (!rec) { |
3745 | goto func_exit; |
3746 | } |
3747 | |
3748 | ut_ad(!rec_is_default_row(rec, index)); |
3749 | offsets = rec_get_offsets( |
3750 | rec, index, offsets, true, ULINT_UNDEFINED, &heap); |
3751 | |
3752 | data = rec_get_nth_field(rec, offsets, 0, &len); |
3753 | |
3754 | doc_id = static_cast<doc_id_t>(fts_read_doc_id( |
3755 | static_cast<const byte*>(data))); |
3756 | } |
3757 | |
3758 | func_exit: |
3759 | btr_pcur_close(&pcur); |
3760 | mtr_commit(&mtr); |
3761 | return(doc_id); |
3762 | } |
3763 | |
3764 | /*********************************************************************//** |
3765 | Fetch document with the given document id. |
3766 | @return DB_SUCCESS if OK else error */ |
3767 | dberr_t |
3768 | fts_doc_fetch_by_doc_id( |
3769 | /*====================*/ |
3770 | fts_get_doc_t* get_doc, /*!< in: state */ |
3771 | doc_id_t doc_id, /*!< in: id of document to |
3772 | fetch */ |
3773 | dict_index_t* index_to_use, /*!< in: caller supplied FTS index, |
3774 | or NULL */ |
3775 | ulint option, /*!< in: search option, if it is |
3776 | greater than doc_id or equal */ |
3777 | fts_sql_callback |
3778 | callback, /*!< in: callback to read */ |
3779 | void* arg) /*!< in: callback arg */ |
3780 | { |
3781 | pars_info_t* info; |
3782 | dberr_t error; |
3783 | const char* select_str; |
3784 | doc_id_t write_doc_id; |
3785 | dict_index_t* index; |
3786 | trx_t* trx = trx_create(); |
3787 | que_t* graph; |
3788 | |
3789 | trx->op_info = "fetching indexed FTS document" ; |
3790 | |
3791 | /* The FTS index can be supplied by caller directly with |
3792 | "index_to_use", otherwise, get it from "get_doc" */ |
3793 | index = (index_to_use) ? index_to_use : get_doc->index_cache->index; |
3794 | |
3795 | if (get_doc && get_doc->get_document_graph) { |
3796 | info = get_doc->get_document_graph->info; |
3797 | } else { |
3798 | info = pars_info_create(); |
3799 | } |
3800 | |
3801 | /* Convert to "storage" byte order. */ |
3802 | fts_write_doc_id((byte*) &write_doc_id, doc_id); |
3803 | fts_bind_doc_id(info, "doc_id" , &write_doc_id); |
3804 | pars_info_bind_function(info, "my_func" , callback, arg); |
3805 | |
3806 | select_str = fts_get_select_columns_str(index, info, info->heap); |
3807 | pars_info_bind_id(info, TRUE, "table_name" , index->table->name.m_name); |
3808 | |
3809 | if (!get_doc || !get_doc->get_document_graph) { |
3810 | if (option == FTS_FETCH_DOC_BY_ID_EQUAL) { |
3811 | graph = fts_parse_sql( |
3812 | NULL, |
3813 | info, |
3814 | mem_heap_printf(info->heap, |
3815 | "DECLARE FUNCTION my_func;\n" |
3816 | "DECLARE CURSOR c IS" |
3817 | " SELECT %s FROM $table_name" |
3818 | " WHERE %s = :doc_id;\n" |
3819 | "BEGIN\n" |
3820 | "" |
3821 | "OPEN c;\n" |
3822 | "WHILE 1 = 1 LOOP\n" |
3823 | " FETCH c INTO my_func();\n" |
3824 | " IF c %% NOTFOUND THEN\n" |
3825 | " EXIT;\n" |
3826 | " END IF;\n" |
3827 | "END LOOP;\n" |
3828 | "CLOSE c;" , |
3829 | select_str, FTS_DOC_ID_COL_NAME)); |
3830 | } else { |
3831 | ut_ad(option == FTS_FETCH_DOC_BY_ID_LARGE); |
3832 | |
3833 | /* This is used for crash recovery of table with |
3834 | hidden DOC ID or FTS indexes. We will scan the table |
3835 | to re-processing user table rows whose DOC ID or |
3836 | FTS indexed documents have not been sync-ed to disc |
3837 | during recent crash. |
3838 | In the case that all fulltext indexes are dropped |
3839 | for a table, we will keep the "hidden" FTS_DOC_ID |
3840 | column, and this scan is to retreive the largest |
3841 | DOC ID being used in the table to determine the |
3842 | appropriate next DOC ID. |
3843 | In the case of there exists fulltext index(es), this |
3844 | operation will re-tokenize any docs that have not |
3845 | been sync-ed to the disk, and re-prime the FTS |
3846 | cached */ |
3847 | graph = fts_parse_sql( |
3848 | NULL, |
3849 | info, |
3850 | mem_heap_printf(info->heap, |
3851 | "DECLARE FUNCTION my_func;\n" |
3852 | "DECLARE CURSOR c IS" |
3853 | " SELECT %s, %s FROM $table_name" |
3854 | " WHERE %s > :doc_id;\n" |
3855 | "BEGIN\n" |
3856 | "" |
3857 | "OPEN c;\n" |
3858 | "WHILE 1 = 1 LOOP\n" |
3859 | " FETCH c INTO my_func();\n" |
3860 | " IF c %% NOTFOUND THEN\n" |
3861 | " EXIT;\n" |
3862 | " END IF;\n" |
3863 | "END LOOP;\n" |
3864 | "CLOSE c;" , |
3865 | FTS_DOC_ID_COL_NAME, |
3866 | select_str, FTS_DOC_ID_COL_NAME)); |
3867 | } |
3868 | if (get_doc) { |
3869 | get_doc->get_document_graph = graph; |
3870 | } |
3871 | } else { |
3872 | graph = get_doc->get_document_graph; |
3873 | } |
3874 | |
3875 | error = fts_eval_sql(trx, graph); |
3876 | fts_sql_commit(trx); |
3877 | trx_free(trx); |
3878 | |
3879 | if (!get_doc) { |
3880 | fts_que_graph_free(graph); |
3881 | } |
3882 | |
3883 | return(error); |
3884 | } |
3885 | |
3886 | /*********************************************************************//** |
3887 | Write out a single word's data as new entry/entries in the INDEX table. |
3888 | @return DB_SUCCESS if all OK. */ |
3889 | dberr_t |
3890 | fts_write_node( |
3891 | /*===========*/ |
3892 | trx_t* trx, /*!< in: transaction */ |
3893 | que_t** graph, /*!< in: query graph */ |
3894 | fts_table_t* fts_table, /*!< in: aux table */ |
3895 | fts_string_t* word, /*!< in: word in UTF-8 */ |
3896 | fts_node_t* node) /*!< in: node columns */ |
3897 | { |
3898 | pars_info_t* info; |
3899 | dberr_t error; |
3900 | ib_uint32_t doc_count; |
3901 | ib_time_t start_time; |
3902 | doc_id_t last_doc_id; |
3903 | doc_id_t first_doc_id; |
3904 | char table_name[MAX_FULL_NAME_LEN]; |
3905 | |
3906 | ut_a(node->ilist != NULL); |
3907 | |
3908 | if (*graph) { |
3909 | info = (*graph)->info; |
3910 | } else { |
3911 | info = pars_info_create(); |
3912 | |
3913 | fts_get_table_name(fts_table, table_name); |
3914 | pars_info_bind_id(info, true, "index_table_name" , table_name); |
3915 | } |
3916 | |
3917 | pars_info_bind_varchar_literal(info, "token" , word->f_str, word->f_len); |
3918 | |
3919 | /* Convert to "storage" byte order. */ |
3920 | fts_write_doc_id((byte*) &first_doc_id, node->first_doc_id); |
3921 | fts_bind_doc_id(info, "first_doc_id" , &first_doc_id); |
3922 | |
3923 | /* Convert to "storage" byte order. */ |
3924 | fts_write_doc_id((byte*) &last_doc_id, node->last_doc_id); |
3925 | fts_bind_doc_id(info, "last_doc_id" , &last_doc_id); |
3926 | |
3927 | ut_a(node->last_doc_id >= node->first_doc_id); |
3928 | |
3929 | /* Convert to "storage" byte order. */ |
3930 | mach_write_to_4((byte*) &doc_count, node->doc_count); |
3931 | pars_info_bind_int4_literal( |
3932 | info, "doc_count" , (const ib_uint32_t*) &doc_count); |
3933 | |
3934 | /* Set copy_name to FALSE since it's a static. */ |
3935 | pars_info_bind_literal( |
3936 | info, "ilist" , node->ilist, node->ilist_size, |
3937 | DATA_BLOB, DATA_BINARY_TYPE); |
3938 | |
3939 | if (!*graph) { |
3940 | |
3941 | *graph = fts_parse_sql( |
3942 | fts_table, |
3943 | info, |
3944 | "BEGIN\n" |
3945 | "INSERT INTO $index_table_name VALUES" |
3946 | " (:token, :first_doc_id," |
3947 | " :last_doc_id, :doc_count, :ilist);" ); |
3948 | } |
3949 | |
3950 | start_time = ut_time(); |
3951 | error = fts_eval_sql(trx, *graph); |
3952 | elapsed_time += ut_time() - start_time; |
3953 | ++n_nodes; |
3954 | |
3955 | return(error); |
3956 | } |
3957 | |
3958 | /*********************************************************************//** |
3959 | Add rows to the DELETED_CACHE table. |
3960 | @return DB_SUCCESS if all went well else error code*/ |
3961 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
3962 | dberr_t |
3963 | fts_sync_add_deleted_cache( |
3964 | /*=======================*/ |
3965 | fts_sync_t* sync, /*!< in: sync state */ |
3966 | ib_vector_t* doc_ids) /*!< in: doc ids to add */ |
3967 | { |
3968 | ulint i; |
3969 | pars_info_t* info; |
3970 | que_t* graph; |
3971 | fts_table_t fts_table; |
3972 | char table_name[MAX_FULL_NAME_LEN]; |
3973 | doc_id_t dummy = 0; |
3974 | dberr_t error = DB_SUCCESS; |
3975 | ulint n_elems = ib_vector_size(doc_ids); |
3976 | |
3977 | ut_a(ib_vector_size(doc_ids) > 0); |
3978 | |
3979 | ib_vector_sort(doc_ids, fts_update_doc_id_cmp); |
3980 | |
3981 | info = pars_info_create(); |
3982 | |
3983 | fts_bind_doc_id(info, "doc_id" , &dummy); |
3984 | |
3985 | FTS_INIT_FTS_TABLE( |
3986 | &fts_table, "DELETED_CACHE" , FTS_COMMON_TABLE, sync->table); |
3987 | |
3988 | fts_get_table_name(&fts_table, table_name); |
3989 | pars_info_bind_id(info, true, "table_name" , table_name); |
3990 | |
3991 | graph = fts_parse_sql( |
3992 | &fts_table, |
3993 | info, |
3994 | "BEGIN INSERT INTO $table_name VALUES (:doc_id);" ); |
3995 | |
3996 | for (i = 0; i < n_elems && error == DB_SUCCESS; ++i) { |
3997 | fts_update_t* update; |
3998 | doc_id_t write_doc_id; |
3999 | |
4000 | update = static_cast<fts_update_t*>(ib_vector_get(doc_ids, i)); |
4001 | |
4002 | /* Convert to "storage" byte order. */ |
4003 | fts_write_doc_id((byte*) &write_doc_id, update->doc_id); |
4004 | fts_bind_doc_id(info, "doc_id" , &write_doc_id); |
4005 | |
4006 | error = fts_eval_sql(sync->trx, graph); |
4007 | } |
4008 | |
4009 | fts_que_graph_free(graph); |
4010 | |
4011 | return(error); |
4012 | } |
4013 | |
4014 | /** Write the words and ilist to disk. |
4015 | @param[in,out] trx transaction |
4016 | @param[in] index_cache index cache |
4017 | @param[in] unlock_cache whether unlock cache when write node |
4018 | @return DB_SUCCESS if all went well else error code */ |
4019 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
4020 | dberr_t |
4021 | fts_sync_write_words( |
4022 | trx_t* trx, |
4023 | fts_index_cache_t* index_cache, |
4024 | bool unlock_cache) |
4025 | { |
4026 | fts_table_t fts_table; |
4027 | ulint n_nodes = 0; |
4028 | ulint n_words = 0; |
4029 | const ib_rbt_node_t* rbt_node; |
4030 | dberr_t error = DB_SUCCESS; |
4031 | ibool print_error = FALSE; |
4032 | dict_table_t* table = index_cache->index->table; |
4033 | |
4034 | FTS_INIT_INDEX_TABLE( |
4035 | &fts_table, NULL, FTS_INDEX_TABLE, index_cache->index); |
4036 | |
4037 | n_words = rbt_size(index_cache->words); |
4038 | |
4039 | /* We iterate over the entire tree, even if there is an error, |
4040 | since we want to free the memory used during caching. */ |
4041 | for (rbt_node = rbt_first(index_cache->words); |
4042 | rbt_node; |
4043 | rbt_node = rbt_next(index_cache->words, rbt_node)) { |
4044 | |
4045 | ulint i; |
4046 | ulint selected; |
4047 | fts_tokenizer_word_t* word; |
4048 | |
4049 | word = rbt_value(fts_tokenizer_word_t, rbt_node); |
4050 | |
4051 | selected = fts_select_index( |
4052 | index_cache->charset, word->text.f_str, |
4053 | word->text.f_len); |
4054 | |
4055 | fts_table.suffix = fts_get_suffix(selected); |
4056 | |
4057 | /* We iterate over all the nodes even if there was an error */ |
4058 | for (i = 0; i < ib_vector_size(word->nodes); ++i) { |
4059 | |
4060 | fts_node_t* fts_node = static_cast<fts_node_t*>( |
4061 | ib_vector_get(word->nodes, i)); |
4062 | |
4063 | if (fts_node->synced) { |
4064 | continue; |
4065 | } else { |
4066 | fts_node->synced = true; |
4067 | } |
4068 | |
4069 | /*FIXME: we need to handle the error properly. */ |
4070 | if (error == DB_SUCCESS) { |
4071 | if (unlock_cache) { |
4072 | rw_lock_x_unlock( |
4073 | &table->fts->cache->lock); |
4074 | } |
4075 | |
4076 | error = fts_write_node( |
4077 | trx, |
4078 | &index_cache->ins_graph[selected], |
4079 | &fts_table, &word->text, fts_node); |
4080 | |
4081 | DEBUG_SYNC_C("fts_write_node" ); |
4082 | DBUG_EXECUTE_IF("fts_write_node_crash" , |
4083 | DBUG_SUICIDE();); |
4084 | |
4085 | DBUG_EXECUTE_IF("fts_instrument_sync_sleep" , |
4086 | os_thread_sleep(1000000); |
4087 | ); |
4088 | |
4089 | if (unlock_cache) { |
4090 | rw_lock_x_lock( |
4091 | &table->fts->cache->lock); |
4092 | } |
4093 | } |
4094 | } |
4095 | |
4096 | n_nodes += ib_vector_size(word->nodes); |
4097 | |
4098 | if (error != DB_SUCCESS && !print_error) { |
4099 | ib::error() << "(" << ut_strerr(error) << ") writing" |
4100 | " word node to FTS auxiliary index table." ; |
4101 | print_error = TRUE; |
4102 | } |
4103 | } |
4104 | |
4105 | if (fts_enable_diag_print) { |
4106 | printf("Avg number of nodes: %lf\n" , |
4107 | (double) n_nodes / (double) (n_words > 1 ? n_words : 1)); |
4108 | } |
4109 | |
4110 | return(error); |
4111 | } |
4112 | |
4113 | /*********************************************************************//** |
4114 | Begin Sync, create transaction, acquire locks, etc. */ |
4115 | static |
4116 | void |
4117 | fts_sync_begin( |
4118 | /*===========*/ |
4119 | fts_sync_t* sync) /*!< in: sync state */ |
4120 | { |
4121 | fts_cache_t* cache = sync->table->fts->cache; |
4122 | |
4123 | n_nodes = 0; |
4124 | elapsed_time = 0; |
4125 | |
4126 | sync->start_time = ut_time(); |
4127 | |
4128 | sync->trx = trx_create(); |
4129 | trx_start_internal(sync->trx); |
4130 | |
4131 | if (fts_enable_diag_print) { |
4132 | ib::info() << "FTS SYNC for table " << sync->table->name |
4133 | << ", deleted count: " |
4134 | << ib_vector_size(cache->deleted_doc_ids) |
4135 | << " size: " << cache->total_size << " bytes" ; |
4136 | } |
4137 | } |
4138 | |
4139 | /*********************************************************************//** |
4140 | Run SYNC on the table, i.e., write out data from the index specific |
4141 | cache to the FTS aux INDEX table and FTS aux doc id stats table. |
4142 | @return DB_SUCCESS if all OK */ |
4143 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
4144 | dberr_t |
4145 | fts_sync_index( |
4146 | /*===========*/ |
4147 | fts_sync_t* sync, /*!< in: sync state */ |
4148 | fts_index_cache_t* index_cache) /*!< in: index cache */ |
4149 | { |
4150 | trx_t* trx = sync->trx; |
4151 | |
4152 | trx->op_info = "doing SYNC index" ; |
4153 | |
4154 | if (fts_enable_diag_print) { |
4155 | ib::info() << "SYNC words: " << rbt_size(index_cache->words); |
4156 | } |
4157 | |
4158 | ut_ad(rbt_validate(index_cache->words)); |
4159 | |
4160 | return(fts_sync_write_words(trx, index_cache, sync->unlock_cache)); |
4161 | } |
4162 | |
4163 | /** Check if index cache has been synced completely |
4164 | @param[in,out] index_cache index cache |
4165 | @return true if index is synced, otherwise false. */ |
4166 | static |
4167 | bool |
4168 | fts_sync_index_check( |
4169 | fts_index_cache_t* index_cache) |
4170 | { |
4171 | const ib_rbt_node_t* rbt_node; |
4172 | |
4173 | for (rbt_node = rbt_first(index_cache->words); |
4174 | rbt_node != NULL; |
4175 | rbt_node = rbt_next(index_cache->words, rbt_node)) { |
4176 | |
4177 | fts_tokenizer_word_t* word; |
4178 | word = rbt_value(fts_tokenizer_word_t, rbt_node); |
4179 | |
4180 | fts_node_t* fts_node; |
4181 | fts_node = static_cast<fts_node_t*>(ib_vector_last(word->nodes)); |
4182 | |
4183 | if (!fts_node->synced) { |
4184 | return(false); |
4185 | } |
4186 | } |
4187 | |
4188 | return(true); |
4189 | } |
4190 | |
4191 | /** Reset synced flag in index cache when rollback |
4192 | @param[in,out] index_cache index cache */ |
4193 | static |
4194 | void |
4195 | fts_sync_index_reset( |
4196 | fts_index_cache_t* index_cache) |
4197 | { |
4198 | const ib_rbt_node_t* rbt_node; |
4199 | |
4200 | for (rbt_node = rbt_first(index_cache->words); |
4201 | rbt_node != NULL; |
4202 | rbt_node = rbt_next(index_cache->words, rbt_node)) { |
4203 | |
4204 | fts_tokenizer_word_t* word; |
4205 | word = rbt_value(fts_tokenizer_word_t, rbt_node); |
4206 | |
4207 | fts_node_t* fts_node; |
4208 | fts_node = static_cast<fts_node_t*>(ib_vector_last(word->nodes)); |
4209 | |
4210 | fts_node->synced = false; |
4211 | } |
4212 | } |
4213 | |
4214 | /** Commit the SYNC, change state of processed doc ids etc. |
4215 | @param[in,out] sync sync state |
4216 | @return DB_SUCCESS if all OK */ |
4217 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
4218 | dberr_t |
4219 | fts_sync_commit( |
4220 | fts_sync_t* sync) |
4221 | { |
4222 | dberr_t error; |
4223 | trx_t* trx = sync->trx; |
4224 | fts_cache_t* cache = sync->table->fts->cache; |
4225 | doc_id_t last_doc_id; |
4226 | |
4227 | trx->op_info = "doing SYNC commit" ; |
4228 | |
4229 | /* After each Sync, update the CONFIG table about the max doc id |
4230 | we just sync-ed to index table */ |
4231 | error = fts_cmp_set_sync_doc_id(sync->table, sync->max_doc_id, FALSE, |
4232 | &last_doc_id); |
4233 | |
4234 | /* Get the list of deleted documents that are either in the |
4235 | cache or were headed there but were deleted before the add |
4236 | thread got to them. */ |
4237 | |
4238 | if (error == DB_SUCCESS && ib_vector_size(cache->deleted_doc_ids) > 0) { |
4239 | |
4240 | error = fts_sync_add_deleted_cache( |
4241 | sync, cache->deleted_doc_ids); |
4242 | } |
4243 | |
4244 | /* We need to do this within the deleted lock since fts_delete() can |
4245 | attempt to add a deleted doc id to the cache deleted id array. */ |
4246 | fts_cache_clear(cache); |
4247 | DEBUG_SYNC_C("fts_deleted_doc_ids_clear" ); |
4248 | fts_cache_init(cache); |
4249 | rw_lock_x_unlock(&cache->lock); |
4250 | |
4251 | if (error == DB_SUCCESS) { |
4252 | |
4253 | fts_sql_commit(trx); |
4254 | |
4255 | } else if (error != DB_SUCCESS) { |
4256 | |
4257 | fts_sql_rollback(trx); |
4258 | |
4259 | ib::error() << "(" << ut_strerr(error) << ") during SYNC." ; |
4260 | } |
4261 | |
4262 | if (fts_enable_diag_print && elapsed_time) { |
4263 | ib::info() << "SYNC for table " << sync->table->name |
4264 | << ": SYNC time: " |
4265 | << (ut_time() - sync->start_time) |
4266 | << " secs: elapsed " |
4267 | << (double) n_nodes / elapsed_time |
4268 | << " ins/sec" ; |
4269 | } |
4270 | |
4271 | /* Avoid assertion in trx_free(). */ |
4272 | trx->dict_operation_lock_mode = 0; |
4273 | trx_free(trx); |
4274 | |
4275 | return(error); |
4276 | } |
4277 | |
4278 | /** Rollback a sync operation |
4279 | @param[in,out] sync sync state */ |
4280 | static |
4281 | void |
4282 | fts_sync_rollback( |
4283 | fts_sync_t* sync) |
4284 | { |
4285 | trx_t* trx = sync->trx; |
4286 | fts_cache_t* cache = sync->table->fts->cache; |
4287 | |
4288 | for (ulint i = 0; i < ib_vector_size(cache->indexes); ++i) { |
4289 | ulint j; |
4290 | fts_index_cache_t* index_cache; |
4291 | |
4292 | index_cache = static_cast<fts_index_cache_t*>( |
4293 | ib_vector_get(cache->indexes, i)); |
4294 | |
4295 | /* Reset synced flag so nodes will not be skipped |
4296 | in the next sync, see fts_sync_write_words(). */ |
4297 | fts_sync_index_reset(index_cache); |
4298 | |
4299 | for (j = 0; fts_index_selector[j].value; ++j) { |
4300 | |
4301 | if (index_cache->ins_graph[j] != NULL) { |
4302 | |
4303 | fts_que_graph_free_check_lock( |
4304 | NULL, index_cache, |
4305 | index_cache->ins_graph[j]); |
4306 | |
4307 | index_cache->ins_graph[j] = NULL; |
4308 | } |
4309 | |
4310 | if (index_cache->sel_graph[j] != NULL) { |
4311 | |
4312 | fts_que_graph_free_check_lock( |
4313 | NULL, index_cache, |
4314 | index_cache->sel_graph[j]); |
4315 | |
4316 | index_cache->sel_graph[j] = NULL; |
4317 | } |
4318 | } |
4319 | } |
4320 | |
4321 | rw_lock_x_unlock(&cache->lock); |
4322 | |
4323 | fts_sql_rollback(trx); |
4324 | |
4325 | /* Avoid assertion in trx_free(). */ |
4326 | trx->dict_operation_lock_mode = 0; |
4327 | trx_free(trx); |
4328 | } |
4329 | |
4330 | /** Run SYNC on the table, i.e., write out data from the cache to the |
4331 | FTS auxiliary INDEX table and clear the cache at the end. |
4332 | @param[in,out] sync sync state |
4333 | @param[in] unlock_cache whether unlock cache lock when write node |
4334 | @param[in] wait whether wait when a sync is in progress |
4335 | @param[in] has_dict whether has dict operation lock |
4336 | @return DB_SUCCESS if all OK */ |
4337 | static |
4338 | dberr_t |
4339 | fts_sync( |
4340 | fts_sync_t* sync, |
4341 | bool unlock_cache, |
4342 | bool wait, |
4343 | bool has_dict) |
4344 | { |
4345 | if (srv_read_only_mode) { |
4346 | return DB_READ_ONLY; |
4347 | } |
4348 | |
4349 | ulint i; |
4350 | dberr_t error = DB_SUCCESS; |
4351 | fts_cache_t* cache = sync->table->fts->cache; |
4352 | |
4353 | rw_lock_x_lock(&cache->lock); |
4354 | |
4355 | /* Check if cache is being synced. |
4356 | Note: we release cache lock in fts_sync_write_words() to |
4357 | avoid long wait for the lock by other threads. */ |
4358 | while (sync->in_progress) { |
4359 | rw_lock_x_unlock(&cache->lock); |
4360 | |
4361 | if (wait) { |
4362 | os_event_wait(sync->event); |
4363 | } else { |
4364 | return(DB_SUCCESS); |
4365 | } |
4366 | |
4367 | rw_lock_x_lock(&cache->lock); |
4368 | } |
4369 | |
4370 | sync->unlock_cache = unlock_cache; |
4371 | sync->in_progress = true; |
4372 | |
4373 | DEBUG_SYNC_C("fts_sync_begin" ); |
4374 | fts_sync_begin(sync); |
4375 | |
4376 | /* When sync in background, we hold dict operation lock |
4377 | to prevent DDL like DROP INDEX, etc. */ |
4378 | if (has_dict) { |
4379 | sync->trx->dict_operation_lock_mode = RW_S_LATCH; |
4380 | } |
4381 | |
4382 | begin_sync: |
4383 | if (cache->total_size > fts_max_cache_size) { |
4384 | /* Avoid the case: sync never finish when |
4385 | insert/update keeps comming. */ |
4386 | ut_ad(sync->unlock_cache); |
4387 | sync->unlock_cache = false; |
4388 | } |
4389 | |
4390 | for (i = 0; i < ib_vector_size(cache->indexes); ++i) { |
4391 | fts_index_cache_t* index_cache; |
4392 | |
4393 | index_cache = static_cast<fts_index_cache_t*>( |
4394 | ib_vector_get(cache->indexes, i)); |
4395 | |
4396 | if (index_cache->index->to_be_dropped |
4397 | || index_cache->index->table->to_be_dropped) { |
4398 | continue; |
4399 | } |
4400 | |
4401 | index_cache->index->index_fts_syncing = true; |
4402 | DBUG_EXECUTE_IF("fts_instrument_sync_sleep_drop_waits" , |
4403 | os_thread_sleep(10000000); |
4404 | ); |
4405 | |
4406 | error = fts_sync_index(sync, index_cache); |
4407 | |
4408 | if (error != DB_SUCCESS && !sync->interrupted) { |
4409 | |
4410 | goto end_sync; |
4411 | } |
4412 | } |
4413 | |
4414 | DBUG_EXECUTE_IF("fts_instrument_sync_interrupted" , |
4415 | sync->interrupted = true; |
4416 | error = DB_INTERRUPTED; |
4417 | goto end_sync; |
4418 | ); |
4419 | |
4420 | /* Make sure all the caches are synced. */ |
4421 | for (i = 0; i < ib_vector_size(cache->indexes); ++i) { |
4422 | fts_index_cache_t* index_cache; |
4423 | |
4424 | index_cache = static_cast<fts_index_cache_t*>( |
4425 | ib_vector_get(cache->indexes, i)); |
4426 | |
4427 | if (index_cache->index->to_be_dropped |
4428 | || index_cache->index->table->to_be_dropped |
4429 | || fts_sync_index_check(index_cache)) { |
4430 | continue; |
4431 | } |
4432 | |
4433 | goto begin_sync; |
4434 | } |
4435 | |
4436 | end_sync: |
4437 | if (error == DB_SUCCESS && !sync->interrupted) { |
4438 | error = fts_sync_commit(sync); |
4439 | } else { |
4440 | fts_sync_rollback(sync); |
4441 | } |
4442 | |
4443 | rw_lock_x_lock(&cache->lock); |
4444 | /* Clear fts syncing flags of any indexes incase sync is |
4445 | interrupeted */ |
4446 | for (i = 0; i < ib_vector_size(cache->indexes); ++i) { |
4447 | static_cast<fts_index_cache_t*>( |
4448 | ib_vector_get(cache->indexes, i)) |
4449 | ->index->index_fts_syncing = false; |
4450 | } |
4451 | |
4452 | sync->interrupted = false; |
4453 | sync->in_progress = false; |
4454 | os_event_set(sync->event); |
4455 | rw_lock_x_unlock(&cache->lock); |
4456 | |
4457 | /* We need to check whether an optimize is required, for that |
4458 | we make copies of the two variables that control the trigger. These |
4459 | variables can change behind our back and we don't want to hold the |
4460 | lock for longer than is needed. */ |
4461 | mutex_enter(&cache->deleted_lock); |
4462 | |
4463 | cache->added = 0; |
4464 | cache->deleted = 0; |
4465 | |
4466 | mutex_exit(&cache->deleted_lock); |
4467 | |
4468 | return(error); |
4469 | } |
4470 | |
4471 | /** Run SYNC on the table, i.e., write out data from the cache to the |
4472 | FTS auxiliary INDEX table and clear the cache at the end. |
4473 | @param[in,out] table fts table |
4474 | @param[in] unlock_cache whether unlock cache when write node |
4475 | @param[in] wait whether wait for existing sync to finish |
4476 | @param[in] has_dict whether has dict operation lock |
4477 | @return DB_SUCCESS on success, error code on failure. */ |
4478 | dberr_t |
4479 | fts_sync_table( |
4480 | dict_table_t* table, |
4481 | bool unlock_cache, |
4482 | bool wait, |
4483 | bool has_dict) |
4484 | { |
4485 | dberr_t err = DB_SUCCESS; |
4486 | |
4487 | ut_ad(table->fts); |
4488 | |
4489 | if (table->space && table->fts->cache |
4490 | && !dict_table_is_corrupted(table)) { |
4491 | err = fts_sync(table->fts->cache->sync, |
4492 | unlock_cache, wait, has_dict); |
4493 | } |
4494 | |
4495 | return(err); |
4496 | } |
4497 | |
4498 | /** Check if a fts token is a stopword or less than fts_min_token_size |
4499 | or greater than fts_max_token_size. |
4500 | @param[in] token token string |
4501 | @param[in] stopwords stopwords rb tree |
4502 | @param[in] cs token charset |
4503 | @retval true if it is not stopword and length in range |
4504 | @retval false if it is stopword or lenght not in range */ |
4505 | bool |
4506 | fts_check_token( |
4507 | const fts_string_t* token, |
4508 | const ib_rbt_t* stopwords, |
4509 | const CHARSET_INFO* cs) |
4510 | { |
4511 | ut_ad(cs != NULL || stopwords == NULL); |
4512 | |
4513 | ib_rbt_bound_t parent; |
4514 | |
4515 | return(token->f_n_char >= fts_min_token_size |
4516 | && token->f_n_char <= fts_max_token_size |
4517 | && (stopwords == NULL |
4518 | || rbt_search(stopwords, &parent, token) != 0)); |
4519 | } |
4520 | |
4521 | /** Add the token and its start position to the token's list of positions. |
4522 | @param[in,out] result_doc result doc rb tree |
4523 | @param[in] str token string |
4524 | @param[in] position token position */ |
4525 | static |
4526 | void |
4527 | fts_add_token( |
4528 | fts_doc_t* result_doc, |
4529 | fts_string_t str, |
4530 | ulint position) |
4531 | { |
4532 | /* Ignore string whose character number is less than |
4533 | "fts_min_token_size" or more than "fts_max_token_size" */ |
4534 | |
4535 | if (fts_check_token(&str, NULL, result_doc->charset)) { |
4536 | |
4537 | mem_heap_t* heap; |
4538 | fts_string_t t_str; |
4539 | fts_token_t* token; |
4540 | ib_rbt_bound_t parent; |
4541 | ulint newlen; |
4542 | |
4543 | heap = static_cast<mem_heap_t*>(result_doc->self_heap->arg); |
4544 | |
4545 | t_str.f_n_char = str.f_n_char; |
4546 | |
4547 | t_str.f_len = str.f_len * result_doc->charset->casedn_multiply + 1; |
4548 | |
4549 | t_str.f_str = static_cast<byte*>( |
4550 | mem_heap_alloc(heap, t_str.f_len)); |
4551 | |
4552 | /* For binary collations, a case sensitive search is |
4553 | performed. Hence don't convert to lower case. */ |
4554 | if (my_binary_compare(result_doc->charset)) { |
4555 | memcpy(t_str.f_str, str.f_str, str.f_len); |
4556 | t_str.f_str[str.f_len]= 0; |
4557 | newlen= str.f_len; |
4558 | } else { |
4559 | newlen = innobase_fts_casedn_str( |
4560 | result_doc->charset, (char*) str.f_str, str.f_len, |
4561 | (char*) t_str.f_str, t_str.f_len); |
4562 | } |
4563 | |
4564 | t_str.f_len = newlen; |
4565 | t_str.f_str[newlen] = 0; |
4566 | |
4567 | /* Add the word to the document statistics. If the word |
4568 | hasn't been seen before we create a new entry for it. */ |
4569 | if (rbt_search(result_doc->tokens, &parent, &t_str) != 0) { |
4570 | fts_token_t new_token; |
4571 | |
4572 | new_token.text.f_len = newlen; |
4573 | new_token.text.f_str = t_str.f_str; |
4574 | new_token.text.f_n_char = t_str.f_n_char; |
4575 | |
4576 | new_token.positions = ib_vector_create( |
4577 | result_doc->self_heap, sizeof(ulint), 32); |
4578 | |
4579 | parent.last = rbt_add_node( |
4580 | result_doc->tokens, &parent, &new_token); |
4581 | |
4582 | ut_ad(rbt_validate(result_doc->tokens)); |
4583 | } |
4584 | |
4585 | token = rbt_value(fts_token_t, parent.last); |
4586 | ib_vector_push(token->positions, &position); |
4587 | } |
4588 | } |
4589 | |
4590 | /******************************************************************** |
4591 | Process next token from document starting at the given position, i.e., add |
4592 | the token's start position to the token's list of positions. |
4593 | @return number of characters handled in this call */ |
4594 | static |
4595 | ulint |
4596 | fts_process_token( |
4597 | /*==============*/ |
4598 | fts_doc_t* doc, /* in/out: document to |
4599 | tokenize */ |
4600 | fts_doc_t* result, /* out: if provided, save |
4601 | result here */ |
4602 | ulint start_pos, /*!< in: start position in text */ |
4603 | ulint add_pos) /*!< in: add this position to all |
4604 | tokens from this tokenization */ |
4605 | { |
4606 | ulint ret; |
4607 | fts_string_t str; |
4608 | ulint position; |
4609 | fts_doc_t* result_doc; |
4610 | byte buf[FTS_MAX_WORD_LEN + 1]; |
4611 | |
4612 | str.f_str = buf; |
4613 | |
4614 | /* Determine where to save the result. */ |
4615 | result_doc = (result != NULL) ? result : doc; |
4616 | |
4617 | /* The length of a string in characters is set here only. */ |
4618 | |
4619 | ret = innobase_mysql_fts_get_token( |
4620 | doc->charset, doc->text.f_str + start_pos, |
4621 | doc->text.f_str + doc->text.f_len, &str); |
4622 | |
4623 | position = start_pos + ret - str.f_len + add_pos; |
4624 | |
4625 | fts_add_token(result_doc, str, position); |
4626 | |
4627 | return(ret); |
4628 | } |
4629 | |
4630 | /*************************************************************//** |
4631 | Get token char size by charset |
4632 | @return token size */ |
4633 | ulint |
4634 | fts_get_token_size( |
4635 | /*===============*/ |
4636 | const CHARSET_INFO* cs, /*!< in: Character set */ |
4637 | const char* token, /*!< in: token */ |
4638 | ulint len) /*!< in: token length */ |
4639 | { |
4640 | char* start; |
4641 | char* end; |
4642 | ulint size = 0; |
4643 | |
4644 | /* const_cast is for reinterpret_cast below, or it will fail. */ |
4645 | start = const_cast<char*>(token); |
4646 | end = start + len; |
4647 | while (start < end) { |
4648 | int ctype; |
4649 | int mbl; |
4650 | |
4651 | mbl = cs->cset->ctype( |
4652 | cs, &ctype, |
4653 | reinterpret_cast<uchar*>(start), |
4654 | reinterpret_cast<uchar*>(end)); |
4655 | |
4656 | size++; |
4657 | |
4658 | start += mbl > 0 ? mbl : (mbl < 0 ? -mbl : 1); |
4659 | } |
4660 | |
4661 | return(size); |
4662 | } |
4663 | |
4664 | /*************************************************************//** |
4665 | FTS plugin parser 'myql_parser' callback function for document tokenize. |
4666 | Refer to 'st_mysql_ftparser_param' for more detail. |
4667 | @return always returns 0 */ |
4668 | int |
4669 | fts_tokenize_document_internal( |
4670 | /*===========================*/ |
4671 | MYSQL_FTPARSER_PARAM* param, /*!< in: parser parameter */ |
4672 | const char* doc,/*!< in/out: document */ |
4673 | int len) /*!< in: document length */ |
4674 | { |
4675 | fts_string_t str; |
4676 | byte buf[FTS_MAX_WORD_LEN + 1]; |
4677 | /* JAN: TODO: MySQL 5.7 |
4678 | MYSQL_FTPARSER_BOOLEAN_INFO bool_info = |
4679 | { FT_TOKEN_WORD, 0, 0, 0, 0, 0, ' ', 0 }; |
4680 | */ |
4681 | MYSQL_FTPARSER_BOOLEAN_INFO bool_info = |
4682 | { FT_TOKEN_WORD, 0, 0, 0, 0, ' ', 0}; |
4683 | |
4684 | ut_ad(len >= 0); |
4685 | |
4686 | str.f_str = buf; |
4687 | |
4688 | for (ulint i = 0, inc = 0; i < static_cast<ulint>(len); i += inc) { |
4689 | inc = innobase_mysql_fts_get_token( |
4690 | const_cast<CHARSET_INFO*>(param->cs), |
4691 | (uchar*)(doc) + i, |
4692 | (uchar*)(doc) + len, |
4693 | &str); |
4694 | |
4695 | if (str.f_len > 0) { |
4696 | /* JAN: TODO: MySQL 5.7 |
4697 | bool_info.position = |
4698 | static_cast<int>(i + inc - str.f_len); |
4699 | ut_ad(bool_info.position >= 0); |
4700 | */ |
4701 | |
4702 | /* Stop when add word fails */ |
4703 | if (param->mysql_add_word( |
4704 | param, |
4705 | reinterpret_cast<char*>(str.f_str), |
4706 | static_cast<int>(str.f_len), |
4707 | &bool_info)) { |
4708 | break; |
4709 | } |
4710 | } |
4711 | } |
4712 | |
4713 | return(0); |
4714 | } |
4715 | |
4716 | /******************************************************************//** |
4717 | FTS plugin parser 'myql_add_word' callback function for document tokenize. |
4718 | Refer to 'st_mysql_ftparser_param' for more detail. |
4719 | @return always returns 0 */ |
4720 | static |
4721 | int |
4722 | fts_tokenize_add_word_for_parser( |
4723 | /*=============================*/ |
4724 | MYSQL_FTPARSER_PARAM* param, /* in: parser paramter */ |
4725 | const char* word, /* in: token word */ |
4726 | int word_len, /* in: word len */ |
4727 | MYSQL_FTPARSER_BOOLEAN_INFO*) |
4728 | { |
4729 | fts_string_t str; |
4730 | fts_tokenize_param_t* fts_param; |
4731 | fts_doc_t* result_doc; |
4732 | ulint position; |
4733 | |
4734 | fts_param = static_cast<fts_tokenize_param_t*>(param->mysql_ftparam); |
4735 | result_doc = fts_param->result_doc; |
4736 | ut_ad(result_doc != NULL); |
4737 | |
4738 | str.f_str = (byte*)(word); |
4739 | str.f_len = ulint(word_len); |
4740 | str.f_n_char = fts_get_token_size( |
4741 | const_cast<CHARSET_INFO*>(param->cs), word, str.f_len); |
4742 | |
4743 | /* JAN: TODO: MySQL 5.7 FTS |
4744 | ut_ad(boolean_info->position >= 0); |
4745 | position = boolean_info->position + fts_param->add_pos; |
4746 | */ |
4747 | position = fts_param->add_pos; |
4748 | |
4749 | fts_add_token(result_doc, str, position); |
4750 | |
4751 | return(0); |
4752 | } |
4753 | |
4754 | /******************************************************************//** |
4755 | Parse a document using an external / user supplied parser */ |
4756 | static |
4757 | void |
4758 | fts_tokenize_by_parser( |
4759 | /*===================*/ |
4760 | fts_doc_t* doc, /* in/out: document to tokenize */ |
4761 | st_mysql_ftparser* parser, /* in: plugin fts parser */ |
4762 | fts_tokenize_param_t* fts_param) /* in: fts tokenize param */ |
4763 | { |
4764 | MYSQL_FTPARSER_PARAM param; |
4765 | |
4766 | ut_a(parser); |
4767 | |
4768 | /* Set paramters for param */ |
4769 | param.mysql_parse = fts_tokenize_document_internal; |
4770 | param.mysql_add_word = fts_tokenize_add_word_for_parser; |
4771 | param.mysql_ftparam = fts_param; |
4772 | param.cs = doc->charset; |
4773 | param.doc = reinterpret_cast<char*>(doc->text.f_str); |
4774 | param.length = static_cast<int>(doc->text.f_len); |
4775 | param.mode= MYSQL_FTPARSER_SIMPLE_MODE; |
4776 | |
4777 | PARSER_INIT(parser, ¶m); |
4778 | parser->parse(¶m); |
4779 | PARSER_DEINIT(parser, ¶m); |
4780 | } |
4781 | |
4782 | /** Tokenize a document. |
4783 | @param[in,out] doc document to tokenize |
4784 | @param[out] result tokenization result |
4785 | @param[in] parser pluggable parser */ |
4786 | static |
4787 | void |
4788 | fts_tokenize_document( |
4789 | fts_doc_t* doc, |
4790 | fts_doc_t* result, |
4791 | st_mysql_ftparser* parser) |
4792 | { |
4793 | ut_a(!doc->tokens); |
4794 | ut_a(doc->charset); |
4795 | |
4796 | doc->tokens = rbt_create_arg_cmp( |
4797 | sizeof(fts_token_t), innobase_fts_text_cmp, (void*) doc->charset); |
4798 | |
4799 | if (parser != NULL) { |
4800 | fts_tokenize_param_t fts_param; |
4801 | |
4802 | fts_param.result_doc = (result != NULL) ? result : doc; |
4803 | fts_param.add_pos = 0; |
4804 | |
4805 | fts_tokenize_by_parser(doc, parser, &fts_param); |
4806 | } else { |
4807 | ulint inc; |
4808 | |
4809 | for (ulint i = 0; i < doc->text.f_len; i += inc) { |
4810 | inc = fts_process_token(doc, result, i, 0); |
4811 | ut_a(inc > 0); |
4812 | } |
4813 | } |
4814 | } |
4815 | |
4816 | /** Continue to tokenize a document. |
4817 | @param[in,out] doc document to tokenize |
4818 | @param[in] add_pos add this position to all tokens from this tokenization |
4819 | @param[out] result tokenization result |
4820 | @param[in] parser pluggable parser */ |
4821 | static |
4822 | void |
4823 | fts_tokenize_document_next( |
4824 | fts_doc_t* doc, |
4825 | ulint add_pos, |
4826 | fts_doc_t* result, |
4827 | st_mysql_ftparser* parser) |
4828 | { |
4829 | ut_a(doc->tokens); |
4830 | |
4831 | if (parser) { |
4832 | fts_tokenize_param_t fts_param; |
4833 | |
4834 | fts_param.result_doc = (result != NULL) ? result : doc; |
4835 | fts_param.add_pos = add_pos; |
4836 | |
4837 | fts_tokenize_by_parser(doc, parser, &fts_param); |
4838 | } else { |
4839 | ulint inc; |
4840 | |
4841 | for (ulint i = 0; i < doc->text.f_len; i += inc) { |
4842 | inc = fts_process_token(doc, result, i, add_pos); |
4843 | ut_a(inc > 0); |
4844 | } |
4845 | } |
4846 | } |
4847 | |
4848 | /** Create the vector of fts_get_doc_t instances. |
4849 | @param[in,out] cache fts cache |
4850 | @return vector of fts_get_doc_t instances */ |
4851 | static |
4852 | ib_vector_t* |
4853 | fts_get_docs_create( |
4854 | fts_cache_t* cache) |
4855 | { |
4856 | ib_vector_t* get_docs; |
4857 | |
4858 | ut_ad(rw_lock_own(&cache->init_lock, RW_LOCK_X)); |
4859 | |
4860 | /* We need one instance of fts_get_doc_t per index. */ |
4861 | get_docs = ib_vector_create(cache->self_heap, sizeof(fts_get_doc_t), 4); |
4862 | |
4863 | /* Create the get_doc instance, we need one of these |
4864 | per FTS index. */ |
4865 | for (ulint i = 0; i < ib_vector_size(cache->indexes); ++i) { |
4866 | |
4867 | dict_index_t** index; |
4868 | fts_get_doc_t* get_doc; |
4869 | |
4870 | index = static_cast<dict_index_t**>( |
4871 | ib_vector_get(cache->indexes, i)); |
4872 | |
4873 | get_doc = static_cast<fts_get_doc_t*>( |
4874 | ib_vector_push(get_docs, NULL)); |
4875 | |
4876 | memset(get_doc, 0x0, sizeof(*get_doc)); |
4877 | |
4878 | get_doc->index_cache = fts_get_index_cache(cache, *index); |
4879 | get_doc->cache = cache; |
4880 | |
4881 | /* Must find the index cache. */ |
4882 | ut_a(get_doc->index_cache != NULL); |
4883 | } |
4884 | |
4885 | return(get_docs); |
4886 | } |
4887 | |
4888 | /******************************************************************** |
4889 | Release any resources held by the fts_get_doc_t instances. */ |
4890 | static |
4891 | void |
4892 | fts_get_docs_clear( |
4893 | /*===============*/ |
4894 | ib_vector_t* get_docs) /*!< in: Doc retrieval vector */ |
4895 | { |
4896 | ulint i; |
4897 | |
4898 | /* Release the get doc graphs if any. */ |
4899 | for (i = 0; i < ib_vector_size(get_docs); ++i) { |
4900 | |
4901 | fts_get_doc_t* get_doc = static_cast<fts_get_doc_t*>( |
4902 | ib_vector_get(get_docs, i)); |
4903 | |
4904 | if (get_doc->get_document_graph != NULL) { |
4905 | |
4906 | ut_a(get_doc->index_cache); |
4907 | |
4908 | fts_que_graph_free(get_doc->get_document_graph); |
4909 | get_doc->get_document_graph = NULL; |
4910 | } |
4911 | } |
4912 | } |
4913 | |
4914 | /*********************************************************************//** |
4915 | Get the initial Doc ID by consulting the CONFIG table |
4916 | @return initial Doc ID */ |
4917 | doc_id_t |
4918 | fts_init_doc_id( |
4919 | /*============*/ |
4920 | const dict_table_t* table) /*!< in: table */ |
4921 | { |
4922 | doc_id_t max_doc_id = 0; |
4923 | |
4924 | rw_lock_x_lock(&table->fts->cache->lock); |
4925 | |
4926 | /* Return if the table is already initialized for DOC ID */ |
4927 | if (table->fts->cache->first_doc_id != FTS_NULL_DOC_ID) { |
4928 | rw_lock_x_unlock(&table->fts->cache->lock); |
4929 | return(0); |
4930 | } |
4931 | |
4932 | DEBUG_SYNC_C("fts_initialize_doc_id" ); |
4933 | |
4934 | /* Then compare this value with the ID value stored in the CONFIG |
4935 | table. The larger one will be our new initial Doc ID */ |
4936 | fts_cmp_set_sync_doc_id(table, 0, FALSE, &max_doc_id); |
4937 | |
4938 | /* If DICT_TF2_FTS_ADD_DOC_ID is set, we are in the process of |
4939 | creating index (and add doc id column. No need to recovery |
4940 | documents */ |
4941 | if (!DICT_TF2_FLAG_IS_SET(table, DICT_TF2_FTS_ADD_DOC_ID)) { |
4942 | fts_init_index((dict_table_t*) table, TRUE); |
4943 | } |
4944 | |
4945 | table->fts->fts_status |= ADDED_TABLE_SYNCED; |
4946 | |
4947 | table->fts->cache->first_doc_id = max_doc_id; |
4948 | |
4949 | rw_lock_x_unlock(&table->fts->cache->lock); |
4950 | |
4951 | ut_ad(max_doc_id > 0); |
4952 | |
4953 | return(max_doc_id); |
4954 | } |
4955 | |
4956 | #ifdef FTS_MULT_INDEX |
4957 | /*********************************************************************//** |
4958 | Check if the index is in the affected set. |
4959 | @return TRUE if index is updated */ |
4960 | static |
4961 | ibool |
4962 | fts_is_index_updated( |
4963 | /*=================*/ |
4964 | const ib_vector_t* fts_indexes, /*!< in: affected FTS indexes */ |
4965 | const fts_get_doc_t* get_doc) /*!< in: info for reading |
4966 | document */ |
4967 | { |
4968 | ulint i; |
4969 | dict_index_t* index = get_doc->index_cache->index; |
4970 | |
4971 | for (i = 0; i < ib_vector_size(fts_indexes); ++i) { |
4972 | const dict_index_t* updated_fts_index; |
4973 | |
4974 | updated_fts_index = static_cast<const dict_index_t*>( |
4975 | ib_vector_getp_const(fts_indexes, i)); |
4976 | |
4977 | ut_a(updated_fts_index != NULL); |
4978 | |
4979 | if (updated_fts_index == index) { |
4980 | return(TRUE); |
4981 | } |
4982 | } |
4983 | |
4984 | return(FALSE); |
4985 | } |
4986 | #endif |
4987 | |
4988 | /*********************************************************************//** |
4989 | Fetch COUNT(*) from specified table. |
4990 | @return the number of rows in the table */ |
4991 | ulint |
4992 | fts_get_rows_count( |
4993 | /*===============*/ |
4994 | fts_table_t* fts_table) /*!< in: fts table to read */ |
4995 | { |
4996 | trx_t* trx; |
4997 | pars_info_t* info; |
4998 | que_t* graph; |
4999 | dberr_t error; |
5000 | ulint count = 0; |
5001 | char table_name[MAX_FULL_NAME_LEN]; |
5002 | |
5003 | trx = trx_create(); |
5004 | trx->op_info = "fetching FT table rows count" ; |
5005 | |
5006 | info = pars_info_create(); |
5007 | |
5008 | pars_info_bind_function(info, "my_func" , fts_read_ulint, &count); |
5009 | |
5010 | fts_get_table_name(fts_table, table_name); |
5011 | pars_info_bind_id(info, true, "table_name" , table_name); |
5012 | |
5013 | graph = fts_parse_sql( |
5014 | fts_table, |
5015 | info, |
5016 | "DECLARE FUNCTION my_func;\n" |
5017 | "DECLARE CURSOR c IS" |
5018 | " SELECT COUNT(*)" |
5019 | " FROM $table_name;\n" |
5020 | "BEGIN\n" |
5021 | "\n" |
5022 | "OPEN c;\n" |
5023 | "WHILE 1 = 1 LOOP\n" |
5024 | " FETCH c INTO my_func();\n" |
5025 | " IF c % NOTFOUND THEN\n" |
5026 | " EXIT;\n" |
5027 | " END IF;\n" |
5028 | "END LOOP;\n" |
5029 | "CLOSE c;" ); |
5030 | |
5031 | for (;;) { |
5032 | error = fts_eval_sql(trx, graph); |
5033 | |
5034 | if (error == DB_SUCCESS) { |
5035 | fts_sql_commit(trx); |
5036 | |
5037 | break; /* Exit the loop. */ |
5038 | } else { |
5039 | fts_sql_rollback(trx); |
5040 | |
5041 | if (error == DB_LOCK_WAIT_TIMEOUT) { |
5042 | ib::warn() << "lock wait timeout reading" |
5043 | " FTS table. Retrying!" ; |
5044 | |
5045 | trx->error_state = DB_SUCCESS; |
5046 | } else { |
5047 | ib::error() << "(" << ut_strerr(error) |
5048 | << ") while reading FTS table." ; |
5049 | |
5050 | break; /* Exit the loop. */ |
5051 | } |
5052 | } |
5053 | } |
5054 | |
5055 | fts_que_graph_free(graph); |
5056 | |
5057 | trx_free(trx); |
5058 | |
5059 | return(count); |
5060 | } |
5061 | |
5062 | #ifdef FTS_CACHE_SIZE_DEBUG |
5063 | /*********************************************************************//** |
5064 | Read the max cache size parameter from the config table. */ |
5065 | static |
5066 | void |
5067 | fts_update_max_cache_size( |
5068 | /*======================*/ |
5069 | fts_sync_t* sync) /*!< in: sync state */ |
5070 | { |
5071 | trx_t* trx; |
5072 | fts_table_t fts_table; |
5073 | |
5074 | trx = trx_create(); |
5075 | |
5076 | FTS_INIT_FTS_TABLE(&fts_table, "CONFIG" , FTS_COMMON_TABLE, sync->table); |
5077 | |
5078 | /* The size returned is in bytes. */ |
5079 | sync->max_cache_size = fts_get_max_cache_size(trx, &fts_table); |
5080 | |
5081 | fts_sql_commit(trx); |
5082 | |
5083 | trx_free(trx); |
5084 | } |
5085 | #endif /* FTS_CACHE_SIZE_DEBUG */ |
5086 | |
5087 | /*********************************************************************//** |
5088 | Free the modified rows of a table. */ |
5089 | UNIV_INLINE |
5090 | void |
5091 | fts_trx_table_rows_free( |
5092 | /*====================*/ |
5093 | ib_rbt_t* rows) /*!< in: rbt of rows to free */ |
5094 | { |
5095 | const ib_rbt_node_t* node; |
5096 | |
5097 | for (node = rbt_first(rows); node; node = rbt_first(rows)) { |
5098 | fts_trx_row_t* row; |
5099 | |
5100 | row = rbt_value(fts_trx_row_t, node); |
5101 | |
5102 | if (row->fts_indexes != NULL) { |
5103 | /* This vector shouldn't be using the |
5104 | heap allocator. */ |
5105 | ut_a(row->fts_indexes->allocator->arg == NULL); |
5106 | |
5107 | ib_vector_free(row->fts_indexes); |
5108 | row->fts_indexes = NULL; |
5109 | } |
5110 | |
5111 | ut_free(rbt_remove_node(rows, node)); |
5112 | } |
5113 | |
5114 | ut_a(rbt_empty(rows)); |
5115 | rbt_free(rows); |
5116 | } |
5117 | |
5118 | /*********************************************************************//** |
5119 | Free an FTS savepoint instance. */ |
5120 | UNIV_INLINE |
5121 | void |
5122 | fts_savepoint_free( |
5123 | /*===============*/ |
5124 | fts_savepoint_t* savepoint) /*!< in: savepoint instance */ |
5125 | { |
5126 | const ib_rbt_node_t* node; |
5127 | ib_rbt_t* tables = savepoint->tables; |
5128 | |
5129 | /* Nothing to free! */ |
5130 | if (tables == NULL) { |
5131 | return; |
5132 | } |
5133 | |
5134 | for (node = rbt_first(tables); node; node = rbt_first(tables)) { |
5135 | fts_trx_table_t* ftt; |
5136 | fts_trx_table_t** fttp; |
5137 | |
5138 | fttp = rbt_value(fts_trx_table_t*, node); |
5139 | ftt = *fttp; |
5140 | |
5141 | /* This can be NULL if a savepoint was released. */ |
5142 | if (ftt->rows != NULL) { |
5143 | fts_trx_table_rows_free(ftt->rows); |
5144 | ftt->rows = NULL; |
5145 | } |
5146 | |
5147 | /* This can be NULL if a savepoint was released. */ |
5148 | if (ftt->added_doc_ids != NULL) { |
5149 | fts_doc_ids_free(ftt->added_doc_ids); |
5150 | ftt->added_doc_ids = NULL; |
5151 | } |
5152 | |
5153 | /* The default savepoint name must be NULL. */ |
5154 | if (ftt->docs_added_graph) { |
5155 | fts_que_graph_free(ftt->docs_added_graph); |
5156 | } |
5157 | |
5158 | /* NOTE: We are responsible for free'ing the node */ |
5159 | ut_free(rbt_remove_node(tables, node)); |
5160 | } |
5161 | |
5162 | ut_a(rbt_empty(tables)); |
5163 | rbt_free(tables); |
5164 | savepoint->tables = NULL; |
5165 | } |
5166 | |
5167 | /*********************************************************************//** |
5168 | Free an FTS trx. */ |
5169 | void |
5170 | fts_trx_free( |
5171 | /*=========*/ |
5172 | fts_trx_t* fts_trx) /* in, own: FTS trx */ |
5173 | { |
5174 | ulint i; |
5175 | |
5176 | for (i = 0; i < ib_vector_size(fts_trx->savepoints); ++i) { |
5177 | fts_savepoint_t* savepoint; |
5178 | |
5179 | savepoint = static_cast<fts_savepoint_t*>( |
5180 | ib_vector_get(fts_trx->savepoints, i)); |
5181 | |
5182 | /* The default savepoint name must be NULL. */ |
5183 | if (i == 0) { |
5184 | ut_a(savepoint->name == NULL); |
5185 | } |
5186 | |
5187 | fts_savepoint_free(savepoint); |
5188 | } |
5189 | |
5190 | for (i = 0; i < ib_vector_size(fts_trx->last_stmt); ++i) { |
5191 | fts_savepoint_t* savepoint; |
5192 | |
5193 | savepoint = static_cast<fts_savepoint_t*>( |
5194 | ib_vector_get(fts_trx->last_stmt, i)); |
5195 | |
5196 | /* The default savepoint name must be NULL. */ |
5197 | if (i == 0) { |
5198 | ut_a(savepoint->name == NULL); |
5199 | } |
5200 | |
5201 | fts_savepoint_free(savepoint); |
5202 | } |
5203 | |
5204 | if (fts_trx->heap) { |
5205 | mem_heap_free(fts_trx->heap); |
5206 | } |
5207 | } |
5208 | |
5209 | /*********************************************************************//** |
5210 | Extract the doc id from the FTS hidden column. |
5211 | @return doc id that was extracted from rec */ |
5212 | doc_id_t |
5213 | fts_get_doc_id_from_row( |
5214 | /*====================*/ |
5215 | dict_table_t* table, /*!< in: table */ |
5216 | dtuple_t* row) /*!< in: row whose FTS doc id we |
5217 | want to extract.*/ |
5218 | { |
5219 | dfield_t* field; |
5220 | doc_id_t doc_id = 0; |
5221 | |
5222 | ut_a(table->fts->doc_col != ULINT_UNDEFINED); |
5223 | |
5224 | field = dtuple_get_nth_field(row, table->fts->doc_col); |
5225 | |
5226 | ut_a(dfield_get_len(field) == sizeof(doc_id)); |
5227 | ut_a(dfield_get_type(field)->mtype == DATA_INT); |
5228 | |
5229 | doc_id = fts_read_doc_id( |
5230 | static_cast<const byte*>(dfield_get_data(field))); |
5231 | |
5232 | return(doc_id); |
5233 | } |
5234 | |
5235 | /** Extract the doc id from the record that belongs to index. |
5236 | @param[in] table table |
5237 | @param[in] rec record contains FTS_DOC_ID |
5238 | @param[in] index index of rec |
5239 | @param[in] heap heap memory |
5240 | @return doc id that was extracted from rec */ |
5241 | doc_id_t |
5242 | fts_get_doc_id_from_rec( |
5243 | dict_table_t* table, |
5244 | const rec_t* rec, |
5245 | const dict_index_t* index, |
5246 | mem_heap_t* heap) |
5247 | { |
5248 | ulint len; |
5249 | const byte* data; |
5250 | ulint col_no; |
5251 | doc_id_t doc_id = 0; |
5252 | ulint offsets_[REC_OFFS_NORMAL_SIZE]; |
5253 | ulint* offsets = offsets_; |
5254 | mem_heap_t* my_heap = heap; |
5255 | |
5256 | ut_a(table->fts->doc_col != ULINT_UNDEFINED); |
5257 | |
5258 | rec_offs_init(offsets_); |
5259 | |
5260 | offsets = rec_get_offsets( |
5261 | rec, index, offsets, true, ULINT_UNDEFINED, &my_heap); |
5262 | |
5263 | col_no = dict_col_get_index_pos( |
5264 | &table->cols[table->fts->doc_col], index); |
5265 | |
5266 | ut_ad(col_no != ULINT_UNDEFINED); |
5267 | |
5268 | data = rec_get_nth_field(rec, offsets, col_no, &len); |
5269 | |
5270 | ut_a(len == 8); |
5271 | ut_ad(8 == sizeof(doc_id)); |
5272 | doc_id = static_cast<doc_id_t>(mach_read_from_8(data)); |
5273 | |
5274 | if (my_heap && !heap) { |
5275 | mem_heap_free(my_heap); |
5276 | } |
5277 | |
5278 | return(doc_id); |
5279 | } |
5280 | |
5281 | /*********************************************************************//** |
5282 | Search the index specific cache for a particular FTS index. |
5283 | @return the index specific cache else NULL */ |
5284 | fts_index_cache_t* |
5285 | fts_find_index_cache( |
5286 | /*=================*/ |
5287 | const fts_cache_t* cache, /*!< in: cache to search */ |
5288 | const dict_index_t* index) /*!< in: index to search for */ |
5289 | { |
5290 | /* We cast away the const because our internal function, takes |
5291 | non-const cache arg and returns a non-const pointer. */ |
5292 | return(static_cast<fts_index_cache_t*>( |
5293 | fts_get_index_cache((fts_cache_t*) cache, index))); |
5294 | } |
5295 | |
5296 | /*********************************************************************//** |
5297 | Search cache for word. |
5298 | @return the word node vector if found else NULL */ |
5299 | const ib_vector_t* |
5300 | fts_cache_find_word( |
5301 | /*================*/ |
5302 | const fts_index_cache_t*index_cache, /*!< in: cache to search */ |
5303 | const fts_string_t* text) /*!< in: word to search for */ |
5304 | { |
5305 | ib_rbt_bound_t parent; |
5306 | const ib_vector_t* nodes = NULL; |
5307 | #ifdef UNIV_DEBUG |
5308 | dict_table_t* table = index_cache->index->table; |
5309 | fts_cache_t* cache = table->fts->cache; |
5310 | |
5311 | ut_ad(rw_lock_own(&cache->lock, RW_LOCK_X)); |
5312 | #endif /* UNIV_DEBUG */ |
5313 | |
5314 | /* Lookup the word in the rb tree */ |
5315 | if (rbt_search(index_cache->words, &parent, text) == 0) { |
5316 | const fts_tokenizer_word_t* word; |
5317 | |
5318 | word = rbt_value(fts_tokenizer_word_t, parent.last); |
5319 | |
5320 | nodes = word->nodes; |
5321 | } |
5322 | |
5323 | return(nodes); |
5324 | } |
5325 | |
5326 | /*********************************************************************//** |
5327 | Append deleted doc ids to vector. */ |
5328 | void |
5329 | fts_cache_append_deleted_doc_ids( |
5330 | /*=============================*/ |
5331 | const fts_cache_t* cache, /*!< in: cache to use */ |
5332 | ib_vector_t* vector) /*!< in: append to this vector */ |
5333 | { |
5334 | mutex_enter(const_cast<ib_mutex_t*>(&cache->deleted_lock)); |
5335 | |
5336 | if (cache->deleted_doc_ids == NULL) { |
5337 | mutex_exit((ib_mutex_t*) &cache->deleted_lock); |
5338 | return; |
5339 | } |
5340 | |
5341 | |
5342 | for (ulint i = 0; i < ib_vector_size(cache->deleted_doc_ids); ++i) { |
5343 | fts_update_t* update; |
5344 | |
5345 | update = static_cast<fts_update_t*>( |
5346 | ib_vector_get(cache->deleted_doc_ids, i)); |
5347 | |
5348 | ib_vector_push(vector, &update->doc_id); |
5349 | } |
5350 | |
5351 | mutex_exit((ib_mutex_t*) &cache->deleted_lock); |
5352 | } |
5353 | |
5354 | /*********************************************************************//** |
5355 | Wait for the background thread to start. We poll to detect change |
5356 | of state, which is acceptable, since the wait should happen only |
5357 | once during startup. |
5358 | @return true if the thread started else FALSE (i.e timed out) */ |
5359 | ibool |
5360 | fts_wait_for_background_thread_to_start( |
5361 | /*====================================*/ |
5362 | dict_table_t* table, /*!< in: table to which the thread |
5363 | is attached */ |
5364 | ulint max_wait) /*!< in: time in microseconds, if |
5365 | set to 0 then it disables |
5366 | timeout checking */ |
5367 | { |
5368 | ulint count = 0; |
5369 | ibool done = FALSE; |
5370 | |
5371 | ut_a(max_wait == 0 || max_wait >= FTS_MAX_BACKGROUND_THREAD_WAIT); |
5372 | |
5373 | for (;;) { |
5374 | fts_t* fts = table->fts; |
5375 | |
5376 | mutex_enter(&fts->bg_threads_mutex); |
5377 | |
5378 | if (fts->fts_status & BG_THREAD_READY) { |
5379 | |
5380 | done = TRUE; |
5381 | } |
5382 | |
5383 | mutex_exit(&fts->bg_threads_mutex); |
5384 | |
5385 | if (!done) { |
5386 | os_thread_sleep(FTS_MAX_BACKGROUND_THREAD_WAIT); |
5387 | |
5388 | if (max_wait > 0) { |
5389 | |
5390 | max_wait -= FTS_MAX_BACKGROUND_THREAD_WAIT; |
5391 | |
5392 | /* We ignore the residual value. */ |
5393 | if (max_wait < FTS_MAX_BACKGROUND_THREAD_WAIT) { |
5394 | break; |
5395 | } |
5396 | } |
5397 | |
5398 | ++count; |
5399 | } else { |
5400 | break; |
5401 | } |
5402 | |
5403 | if (count >= FTS_BACKGROUND_THREAD_WAIT_COUNT) { |
5404 | ib::error() << "The background thread for the FTS" |
5405 | " table " << table->name |
5406 | << " refuses to start" ; |
5407 | |
5408 | count = 0; |
5409 | } |
5410 | } |
5411 | |
5412 | return(done); |
5413 | } |
5414 | |
5415 | /*********************************************************************//** |
5416 | Add the FTS document id hidden column. */ |
5417 | void |
5418 | fts_add_doc_id_column( |
5419 | /*==================*/ |
5420 | dict_table_t* table, /*!< in/out: Table with FTS index */ |
5421 | mem_heap_t* heap) /*!< in: temporary memory heap, or NULL */ |
5422 | { |
5423 | dict_mem_table_add_col( |
5424 | table, heap, |
5425 | FTS_DOC_ID_COL_NAME, |
5426 | DATA_INT, |
5427 | dtype_form_prtype( |
5428 | DATA_NOT_NULL | DATA_UNSIGNED |
5429 | | DATA_BINARY_TYPE | DATA_FTS_DOC_ID, 0), |
5430 | sizeof(doc_id_t)); |
5431 | DICT_TF2_FLAG_SET(table, DICT_TF2_FTS_HAS_DOC_ID); |
5432 | } |
5433 | |
5434 | /** Add new fts doc id to the update vector. |
5435 | @param[in] table the table that contains the FTS index. |
5436 | @param[in,out] ufield the fts doc id field in the update vector. |
5437 | No new memory is allocated for this in this |
5438 | function. |
5439 | @param[in,out] next_doc_id the fts doc id that has been added to the |
5440 | update vector. If 0, a new fts doc id is |
5441 | automatically generated. The memory provided |
5442 | for this argument will be used by the update |
5443 | vector. Ensure that the life time of this |
5444 | memory matches that of the update vector. |
5445 | @return the fts doc id used in the update vector */ |
5446 | doc_id_t |
5447 | fts_update_doc_id( |
5448 | dict_table_t* table, |
5449 | upd_field_t* ufield, |
5450 | doc_id_t* next_doc_id) |
5451 | { |
5452 | doc_id_t doc_id; |
5453 | dberr_t error = DB_SUCCESS; |
5454 | |
5455 | if (*next_doc_id) { |
5456 | doc_id = *next_doc_id; |
5457 | } else { |
5458 | /* Get the new document id that will be added. */ |
5459 | error = fts_get_next_doc_id(table, &doc_id); |
5460 | } |
5461 | |
5462 | if (error == DB_SUCCESS) { |
5463 | dict_index_t* clust_index; |
5464 | dict_col_t* col = dict_table_get_nth_col( |
5465 | table, table->fts->doc_col); |
5466 | |
5467 | ufield->exp = NULL; |
5468 | |
5469 | ufield->new_val.len = sizeof(doc_id); |
5470 | |
5471 | clust_index = dict_table_get_first_index(table); |
5472 | |
5473 | ufield->field_no = dict_col_get_clust_pos(col, clust_index); |
5474 | dict_col_copy_type(col, dfield_get_type(&ufield->new_val)); |
5475 | |
5476 | /* It is possible we update record that has |
5477 | not yet be sync-ed from last crash. */ |
5478 | |
5479 | /* Convert to storage byte order. */ |
5480 | ut_a(doc_id != FTS_NULL_DOC_ID); |
5481 | fts_write_doc_id((byte*) next_doc_id, doc_id); |
5482 | |
5483 | ufield->new_val.data = next_doc_id; |
5484 | ufield->new_val.ext = 0; |
5485 | } |
5486 | |
5487 | return(doc_id); |
5488 | } |
5489 | |
5490 | /** fts_t constructor. |
5491 | @param[in] table table with FTS indexes |
5492 | @param[in,out] heap memory heap where 'this' is stored */ |
5493 | fts_t::fts_t( |
5494 | const dict_table_t* table, |
5495 | mem_heap_t* heap) |
5496 | : |
5497 | bg_threads(0), |
5498 | fts_status(0), |
5499 | add_wq(NULL), |
5500 | cache(NULL), |
5501 | doc_col(ULINT_UNDEFINED), |
5502 | fts_heap(heap) |
5503 | { |
5504 | ut_a(table->fts == NULL); |
5505 | |
5506 | mutex_create(LATCH_ID_FTS_BG_THREADS, &bg_threads_mutex); |
5507 | |
5508 | ib_alloc_t* heap_alloc = ib_heap_allocator_create(fts_heap); |
5509 | |
5510 | indexes = ib_vector_create(heap_alloc, sizeof(dict_index_t*), 4); |
5511 | |
5512 | dict_table_get_all_fts_indexes(table, indexes); |
5513 | } |
5514 | |
5515 | /** fts_t destructor. */ |
5516 | fts_t::~fts_t() |
5517 | { |
5518 | mutex_free(&bg_threads_mutex); |
5519 | |
5520 | ut_ad(add_wq == NULL); |
5521 | |
5522 | if (cache != NULL) { |
5523 | fts_cache_clear(cache); |
5524 | fts_cache_destroy(cache); |
5525 | cache = NULL; |
5526 | } |
5527 | |
5528 | /* There is no need to call ib_vector_free() on this->indexes |
5529 | because it is stored in this->fts_heap. */ |
5530 | } |
5531 | |
5532 | /*********************************************************************//** |
5533 | Create an instance of fts_t. |
5534 | @return instance of fts_t */ |
5535 | fts_t* |
5536 | fts_create( |
5537 | /*=======*/ |
5538 | dict_table_t* table) /*!< in/out: table with FTS indexes */ |
5539 | { |
5540 | fts_t* fts; |
5541 | mem_heap_t* heap; |
5542 | |
5543 | heap = mem_heap_create(512); |
5544 | |
5545 | fts = static_cast<fts_t*>(mem_heap_alloc(heap, sizeof(*fts))); |
5546 | |
5547 | new(fts) fts_t(table, heap); |
5548 | |
5549 | return(fts); |
5550 | } |
5551 | |
5552 | /*********************************************************************//** |
5553 | Free the FTS resources. */ |
5554 | void |
5555 | fts_free( |
5556 | /*=====*/ |
5557 | dict_table_t* table) /*!< in/out: table with FTS indexes */ |
5558 | { |
5559 | fts_t* fts = table->fts; |
5560 | |
5561 | fts->~fts_t(); |
5562 | |
5563 | mem_heap_free(fts->fts_heap); |
5564 | |
5565 | table->fts = NULL; |
5566 | } |
5567 | |
5568 | #if 0 // TODO: Enable this in WL#6608 |
5569 | /*********************************************************************//** |
5570 | Signal FTS threads to initiate shutdown. */ |
5571 | void |
5572 | fts_start_shutdown( |
5573 | /*===============*/ |
5574 | dict_table_t* table, /*!< in: table with FTS indexes */ |
5575 | fts_t* fts) /*!< in: fts instance that needs |
5576 | to be informed about shutdown */ |
5577 | { |
5578 | mutex_enter(&fts->bg_threads_mutex); |
5579 | |
5580 | fts->fts_status |= BG_THREAD_STOP; |
5581 | |
5582 | mutex_exit(&fts->bg_threads_mutex); |
5583 | |
5584 | } |
5585 | |
5586 | /*********************************************************************//** |
5587 | Wait for FTS threads to shutdown. */ |
5588 | void |
5589 | fts_shutdown( |
5590 | /*=========*/ |
5591 | dict_table_t* table, /*!< in: table with FTS indexes */ |
5592 | fts_t* fts) /*!< in: fts instance to shutdown */ |
5593 | { |
5594 | mutex_enter(&fts->bg_threads_mutex); |
5595 | |
5596 | ut_a(fts->fts_status & BG_THREAD_STOP); |
5597 | |
5598 | dict_table_wait_for_bg_threads_to_exit(table, 20000); |
5599 | |
5600 | mutex_exit(&fts->bg_threads_mutex); |
5601 | } |
5602 | #endif |
5603 | |
5604 | /*********************************************************************//** |
5605 | Take a FTS savepoint. */ |
5606 | UNIV_INLINE |
5607 | void |
5608 | fts_savepoint_copy( |
5609 | /*===============*/ |
5610 | const fts_savepoint_t* src, /*!< in: source savepoint */ |
5611 | fts_savepoint_t* dst) /*!< out: destination savepoint */ |
5612 | { |
5613 | const ib_rbt_node_t* node; |
5614 | const ib_rbt_t* tables; |
5615 | |
5616 | tables = src->tables; |
5617 | |
5618 | for (node = rbt_first(tables); node; node = rbt_next(tables, node)) { |
5619 | |
5620 | fts_trx_table_t* ftt_dst; |
5621 | const fts_trx_table_t** ftt_src; |
5622 | |
5623 | ftt_src = rbt_value(const fts_trx_table_t*, node); |
5624 | |
5625 | ftt_dst = fts_trx_table_clone(*ftt_src); |
5626 | |
5627 | rbt_insert(dst->tables, &ftt_dst, &ftt_dst); |
5628 | } |
5629 | } |
5630 | |
5631 | /*********************************************************************//** |
5632 | Take a FTS savepoint. */ |
5633 | void |
5634 | fts_savepoint_take( |
5635 | /*===============*/ |
5636 | fts_trx_t* fts_trx, /*!< in: fts transaction */ |
5637 | const char* name) /*!< in: savepoint name */ |
5638 | { |
5639 | mem_heap_t* heap; |
5640 | fts_savepoint_t* savepoint; |
5641 | fts_savepoint_t* last_savepoint; |
5642 | |
5643 | ut_a(name != NULL); |
5644 | |
5645 | heap = fts_trx->heap; |
5646 | |
5647 | /* The implied savepoint must exist. */ |
5648 | ut_a(ib_vector_size(fts_trx->savepoints) > 0); |
5649 | |
5650 | last_savepoint = static_cast<fts_savepoint_t*>( |
5651 | ib_vector_last(fts_trx->savepoints)); |
5652 | savepoint = fts_savepoint_create(fts_trx->savepoints, name, heap); |
5653 | |
5654 | if (last_savepoint->tables != NULL) { |
5655 | fts_savepoint_copy(last_savepoint, savepoint); |
5656 | } |
5657 | } |
5658 | |
5659 | /*********************************************************************//** |
5660 | Lookup a savepoint instance by name. |
5661 | @return ULINT_UNDEFINED if not found */ |
5662 | UNIV_INLINE |
5663 | ulint |
5664 | fts_savepoint_lookup( |
5665 | /*==================*/ |
5666 | ib_vector_t* savepoints, /*!< in: savepoints */ |
5667 | const char* name) /*!< in: savepoint name */ |
5668 | { |
5669 | ulint i; |
5670 | |
5671 | ut_a(ib_vector_size(savepoints) > 0); |
5672 | |
5673 | for (i = 1; i < ib_vector_size(savepoints); ++i) { |
5674 | fts_savepoint_t* savepoint; |
5675 | |
5676 | savepoint = static_cast<fts_savepoint_t*>( |
5677 | ib_vector_get(savepoints, i)); |
5678 | |
5679 | if (strcmp(name, savepoint->name) == 0) { |
5680 | return(i); |
5681 | } |
5682 | } |
5683 | |
5684 | return(ULINT_UNDEFINED); |
5685 | } |
5686 | |
5687 | /*********************************************************************//** |
5688 | Release the savepoint data identified by name. All savepoints created |
5689 | after the named savepoint are kept. |
5690 | @return DB_SUCCESS or error code */ |
5691 | void |
5692 | fts_savepoint_release( |
5693 | /*==================*/ |
5694 | trx_t* trx, /*!< in: transaction */ |
5695 | const char* name) /*!< in: savepoint name */ |
5696 | { |
5697 | ut_a(name != NULL); |
5698 | |
5699 | ib_vector_t* savepoints = trx->fts_trx->savepoints; |
5700 | |
5701 | ut_a(ib_vector_size(savepoints) > 0); |
5702 | |
5703 | ulint i = fts_savepoint_lookup(savepoints, name); |
5704 | if (i != ULINT_UNDEFINED) { |
5705 | ut_a(i >= 1); |
5706 | |
5707 | fts_savepoint_t* savepoint; |
5708 | savepoint = static_cast<fts_savepoint_t*>( |
5709 | ib_vector_get(savepoints, i)); |
5710 | |
5711 | if (i == ib_vector_size(savepoints) - 1) { |
5712 | /* If the savepoint is the last, we save its |
5713 | tables to the previous savepoint. */ |
5714 | fts_savepoint_t* prev_savepoint; |
5715 | prev_savepoint = static_cast<fts_savepoint_t*>( |
5716 | ib_vector_get(savepoints, i - 1)); |
5717 | |
5718 | ib_rbt_t* tables = savepoint->tables; |
5719 | savepoint->tables = prev_savepoint->tables; |
5720 | prev_savepoint->tables = tables; |
5721 | } |
5722 | |
5723 | fts_savepoint_free(savepoint); |
5724 | ib_vector_remove(savepoints, *(void**)savepoint); |
5725 | |
5726 | /* Make sure we don't delete the implied savepoint. */ |
5727 | ut_a(ib_vector_size(savepoints) > 0); |
5728 | } |
5729 | } |
5730 | |
5731 | /**********************************************************************//** |
5732 | Refresh last statement savepoint. */ |
5733 | void |
5734 | fts_savepoint_laststmt_refresh( |
5735 | /*===========================*/ |
5736 | trx_t* trx) /*!< in: transaction */ |
5737 | { |
5738 | |
5739 | fts_trx_t* fts_trx; |
5740 | fts_savepoint_t* savepoint; |
5741 | |
5742 | fts_trx = trx->fts_trx; |
5743 | |
5744 | savepoint = static_cast<fts_savepoint_t*>( |
5745 | ib_vector_pop(fts_trx->last_stmt)); |
5746 | fts_savepoint_free(savepoint); |
5747 | |
5748 | ut_ad(ib_vector_is_empty(fts_trx->last_stmt)); |
5749 | savepoint = fts_savepoint_create(fts_trx->last_stmt, NULL, NULL); |
5750 | } |
5751 | |
5752 | /******************************************************************** |
5753 | Undo the Doc ID add/delete operations in last stmt */ |
5754 | static |
5755 | void |
5756 | fts_undo_last_stmt( |
5757 | /*===============*/ |
5758 | fts_trx_table_t* s_ftt, /*!< in: Transaction FTS table */ |
5759 | fts_trx_table_t* l_ftt) /*!< in: last stmt FTS table */ |
5760 | { |
5761 | ib_rbt_t* s_rows; |
5762 | ib_rbt_t* l_rows; |
5763 | const ib_rbt_node_t* node; |
5764 | |
5765 | l_rows = l_ftt->rows; |
5766 | s_rows = s_ftt->rows; |
5767 | |
5768 | for (node = rbt_first(l_rows); |
5769 | node; |
5770 | node = rbt_next(l_rows, node)) { |
5771 | fts_trx_row_t* l_row = rbt_value(fts_trx_row_t, node); |
5772 | ib_rbt_bound_t parent; |
5773 | |
5774 | rbt_search(s_rows, &parent, &(l_row->doc_id)); |
5775 | |
5776 | if (parent.result == 0) { |
5777 | fts_trx_row_t* s_row = rbt_value( |
5778 | fts_trx_row_t, parent.last); |
5779 | |
5780 | switch (l_row->state) { |
5781 | case FTS_INSERT: |
5782 | ut_free(rbt_remove_node(s_rows, parent.last)); |
5783 | break; |
5784 | |
5785 | case FTS_DELETE: |
5786 | if (s_row->state == FTS_NOTHING) { |
5787 | s_row->state = FTS_INSERT; |
5788 | } else if (s_row->state == FTS_DELETE) { |
5789 | ut_free(rbt_remove_node( |
5790 | s_rows, parent.last)); |
5791 | } |
5792 | break; |
5793 | |
5794 | /* FIXME: Check if FTS_MODIFY need to be addressed */ |
5795 | case FTS_MODIFY: |
5796 | case FTS_NOTHING: |
5797 | break; |
5798 | default: |
5799 | ut_error; |
5800 | } |
5801 | } |
5802 | } |
5803 | } |
5804 | |
5805 | /**********************************************************************//** |
5806 | Rollback to savepoint indentified by name. |
5807 | @return DB_SUCCESS or error code */ |
5808 | void |
5809 | fts_savepoint_rollback_last_stmt( |
5810 | /*=============================*/ |
5811 | trx_t* trx) /*!< in: transaction */ |
5812 | { |
5813 | ib_vector_t* savepoints; |
5814 | fts_savepoint_t* savepoint; |
5815 | fts_savepoint_t* last_stmt; |
5816 | fts_trx_t* fts_trx; |
5817 | ib_rbt_bound_t parent; |
5818 | const ib_rbt_node_t* node; |
5819 | ib_rbt_t* l_tables; |
5820 | ib_rbt_t* s_tables; |
5821 | |
5822 | fts_trx = trx->fts_trx; |
5823 | savepoints = fts_trx->savepoints; |
5824 | |
5825 | savepoint = static_cast<fts_savepoint_t*>(ib_vector_last(savepoints)); |
5826 | last_stmt = static_cast<fts_savepoint_t*>( |
5827 | ib_vector_last(fts_trx->last_stmt)); |
5828 | |
5829 | l_tables = last_stmt->tables; |
5830 | s_tables = savepoint->tables; |
5831 | |
5832 | for (node = rbt_first(l_tables); |
5833 | node; |
5834 | node = rbt_next(l_tables, node)) { |
5835 | |
5836 | fts_trx_table_t** l_ftt; |
5837 | |
5838 | l_ftt = rbt_value(fts_trx_table_t*, node); |
5839 | |
5840 | rbt_search_cmp( |
5841 | s_tables, &parent, &(*l_ftt)->table->id, |
5842 | fts_trx_table_id_cmp, NULL); |
5843 | |
5844 | if (parent.result == 0) { |
5845 | fts_trx_table_t** s_ftt; |
5846 | |
5847 | s_ftt = rbt_value(fts_trx_table_t*, parent.last); |
5848 | |
5849 | fts_undo_last_stmt(*s_ftt, *l_ftt); |
5850 | } |
5851 | } |
5852 | } |
5853 | |
5854 | /**********************************************************************//** |
5855 | Rollback to savepoint indentified by name. |
5856 | @return DB_SUCCESS or error code */ |
5857 | void |
5858 | fts_savepoint_rollback( |
5859 | /*===================*/ |
5860 | trx_t* trx, /*!< in: transaction */ |
5861 | const char* name) /*!< in: savepoint name */ |
5862 | { |
5863 | ulint i; |
5864 | ib_vector_t* savepoints; |
5865 | |
5866 | ut_a(name != NULL); |
5867 | |
5868 | savepoints = trx->fts_trx->savepoints; |
5869 | |
5870 | /* We pop all savepoints from the the top of the stack up to |
5871 | and including the instance that was found. */ |
5872 | i = fts_savepoint_lookup(savepoints, name); |
5873 | |
5874 | if (i != ULINT_UNDEFINED) { |
5875 | fts_savepoint_t* savepoint; |
5876 | |
5877 | ut_a(i > 0); |
5878 | |
5879 | while (ib_vector_size(savepoints) > i) { |
5880 | fts_savepoint_t* savepoint; |
5881 | |
5882 | savepoint = static_cast<fts_savepoint_t*>( |
5883 | ib_vector_pop(savepoints)); |
5884 | |
5885 | if (savepoint->name != NULL) { |
5886 | /* Since name was allocated on the heap, the |
5887 | memory will be released when the transaction |
5888 | completes. */ |
5889 | savepoint->name = NULL; |
5890 | |
5891 | fts_savepoint_free(savepoint); |
5892 | } |
5893 | } |
5894 | |
5895 | /* Pop all a elements from the top of the stack that may |
5896 | have been released. We have to be careful that we don't |
5897 | delete the implied savepoint. */ |
5898 | |
5899 | for (savepoint = static_cast<fts_savepoint_t*>( |
5900 | ib_vector_last(savepoints)); |
5901 | ib_vector_size(savepoints) > 1 |
5902 | && savepoint->name == NULL; |
5903 | savepoint = static_cast<fts_savepoint_t*>( |
5904 | ib_vector_last(savepoints))) { |
5905 | |
5906 | ib_vector_pop(savepoints); |
5907 | } |
5908 | |
5909 | /* Make sure we don't delete the implied savepoint. */ |
5910 | ut_a(ib_vector_size(savepoints) > 0); |
5911 | |
5912 | /* Restore the savepoint. */ |
5913 | fts_savepoint_take(trx->fts_trx, name); |
5914 | } |
5915 | } |
5916 | |
5917 | /** Check if a table is an FTS auxiliary table name. |
5918 | @param[out] table FTS table info |
5919 | @param[in] name Table name |
5920 | @param[in] len Length of table name |
5921 | @return true if the name matches an auxiliary table name pattern */ |
5922 | static |
5923 | bool |
5924 | fts_is_aux_table_name( |
5925 | fts_aux_table_t* table, |
5926 | const char* name, |
5927 | ulint len) |
5928 | { |
5929 | const char* ptr; |
5930 | char* end; |
5931 | char my_name[MAX_FULL_NAME_LEN + 1]; |
5932 | |
5933 | ut_ad(len <= MAX_FULL_NAME_LEN); |
5934 | ut_memcpy(my_name, name, len); |
5935 | my_name[len] = 0; |
5936 | end = my_name + len; |
5937 | |
5938 | ptr = static_cast<const char*>(memchr(my_name, '/', len)); |
5939 | |
5940 | if (ptr != NULL) { |
5941 | /* We will start the match after the '/' */ |
5942 | ++ptr; |
5943 | len = ulint(end - ptr); |
5944 | } |
5945 | |
5946 | /* All auxiliary tables are prefixed with "FTS_" and the name |
5947 | length will be at the very least greater than 20 bytes. */ |
5948 | if (ptr != NULL && len > 20 && strncmp(ptr, "FTS_" , 4) == 0) { |
5949 | ulint i; |
5950 | |
5951 | /* Skip the prefix. */ |
5952 | ptr += 4; |
5953 | len -= 4; |
5954 | |
5955 | /* Try and read the table id. */ |
5956 | if (!fts_read_object_id(&table->parent_id, ptr)) { |
5957 | return(false); |
5958 | } |
5959 | |
5960 | /* Skip the table id. */ |
5961 | ptr = static_cast<const char*>(memchr(ptr, '_', len)); |
5962 | |
5963 | if (ptr == NULL) { |
5964 | return(false); |
5965 | } |
5966 | |
5967 | /* Skip the underscore. */ |
5968 | ++ptr; |
5969 | ut_a(end > ptr); |
5970 | len = ulint(end - ptr); |
5971 | |
5972 | /* First search the common table suffix array. */ |
5973 | for (i = 0; fts_common_tables[i] != NULL; ++i) { |
5974 | |
5975 | if (strncmp(ptr, fts_common_tables[i], len) == 0) { |
5976 | return(true); |
5977 | } |
5978 | } |
5979 | |
5980 | /* Could be obsolete common tables. */ |
5981 | if (strncmp(ptr, "ADDED" , len) == 0 |
5982 | || strncmp(ptr, "STOPWORDS" , len) == 0) { |
5983 | return(true); |
5984 | } |
5985 | |
5986 | /* Try and read the index id. */ |
5987 | if (!fts_read_object_id(&table->index_id, ptr)) { |
5988 | return(false); |
5989 | } |
5990 | |
5991 | /* Skip the table id. */ |
5992 | ptr = static_cast<const char*>(memchr(ptr, '_', len)); |
5993 | |
5994 | if (ptr == NULL) { |
5995 | return(false); |
5996 | } |
5997 | |
5998 | /* Skip the underscore. */ |
5999 | ++ptr; |
6000 | ut_a(end > ptr); |
6001 | len = ulint(end - ptr); |
6002 | |
6003 | /* Search the FT index specific array. */ |
6004 | for (i = 0; i < FTS_NUM_AUX_INDEX; ++i) { |
6005 | |
6006 | if (strncmp(ptr, fts_get_suffix(i), len) == 0) { |
6007 | return(true); |
6008 | } |
6009 | } |
6010 | |
6011 | /* Other FT index specific table(s). */ |
6012 | if (strncmp(ptr, "DOC_ID" , len) == 0) { |
6013 | return(true); |
6014 | } |
6015 | } |
6016 | |
6017 | return(false); |
6018 | } |
6019 | |
6020 | /**********************************************************************//** |
6021 | Callback function to read a single table ID column. |
6022 | @return Always return TRUE */ |
6023 | static |
6024 | ibool |
6025 | fts_read_tables( |
6026 | /*============*/ |
6027 | void* row, /*!< in: sel_node_t* */ |
6028 | void* user_arg) /*!< in: pointer to ib_vector_t */ |
6029 | { |
6030 | int i; |
6031 | fts_aux_table_t*table; |
6032 | mem_heap_t* heap; |
6033 | ibool done = FALSE; |
6034 | ib_vector_t* tables = static_cast<ib_vector_t*>(user_arg); |
6035 | sel_node_t* sel_node = static_cast<sel_node_t*>(row); |
6036 | que_node_t* exp = sel_node->select_list; |
6037 | |
6038 | /* Must be a heap allocated vector. */ |
6039 | ut_a(tables->allocator->arg != NULL); |
6040 | |
6041 | /* We will use this heap for allocating strings. */ |
6042 | heap = static_cast<mem_heap_t*>(tables->allocator->arg); |
6043 | table = static_cast<fts_aux_table_t*>(ib_vector_push(tables, NULL)); |
6044 | |
6045 | memset(table, 0x0, sizeof(*table)); |
6046 | |
6047 | /* Iterate over the columns and read the values. */ |
6048 | for (i = 0; exp && !done; exp = que_node_get_next(exp), ++i) { |
6049 | |
6050 | dfield_t* dfield = que_node_get_val(exp); |
6051 | void* data = dfield_get_data(dfield); |
6052 | ulint len = dfield_get_len(dfield); |
6053 | |
6054 | ut_a(len != UNIV_SQL_NULL); |
6055 | |
6056 | /* Note: The column numbers below must match the SELECT */ |
6057 | switch (i) { |
6058 | case 0: /* NAME */ |
6059 | |
6060 | if (!fts_is_aux_table_name( |
6061 | table, static_cast<const char*>(data), len)) { |
6062 | ib_vector_pop(tables); |
6063 | done = TRUE; |
6064 | break; |
6065 | } |
6066 | |
6067 | table->name = static_cast<char*>( |
6068 | mem_heap_alloc(heap, len + 1)); |
6069 | memcpy(table->name, data, len); |
6070 | table->name[len] = 0; |
6071 | break; |
6072 | |
6073 | case 1: /* ID */ |
6074 | ut_a(len == 8); |
6075 | table->id = mach_read_from_8( |
6076 | static_cast<const byte*>(data)); |
6077 | break; |
6078 | |
6079 | default: |
6080 | ut_error; |
6081 | } |
6082 | } |
6083 | |
6084 | return(TRUE); |
6085 | } |
6086 | |
6087 | /******************************************************************//** |
6088 | Callback that sets a hex formatted FTS table's flags2 in |
6089 | SYS_TABLES. The flags is stored in MIX_LEN column. |
6090 | @return FALSE if all OK */ |
6091 | static |
6092 | ibool |
6093 | fts_set_hex_format( |
6094 | /*===============*/ |
6095 | void* row, /*!< in: sel_node_t* */ |
6096 | void* user_arg) /*!< in: bool set/unset flag */ |
6097 | { |
6098 | sel_node_t* node = static_cast<sel_node_t*>(row); |
6099 | dfield_t* dfield = que_node_get_val(node->select_list); |
6100 | |
6101 | ut_ad(dtype_get_mtype(dfield_get_type(dfield)) == DATA_INT); |
6102 | ut_ad(dfield_get_len(dfield) == sizeof(ib_uint32_t)); |
6103 | /* There should be at most one matching record. So the value |
6104 | must be the default value. */ |
6105 | ut_ad(mach_read_from_4(static_cast<byte*>(user_arg)) |
6106 | == ULINT32_UNDEFINED); |
6107 | |
6108 | ulint flags2 = mach_read_from_4( |
6109 | static_cast<byte*>(dfield_get_data(dfield))); |
6110 | |
6111 | flags2 |= DICT_TF2_FTS_AUX_HEX_NAME; |
6112 | |
6113 | mach_write_to_4(static_cast<byte*>(user_arg), flags2); |
6114 | |
6115 | return(FALSE); |
6116 | } |
6117 | |
6118 | /*****************************************************************//** |
6119 | Update the DICT_TF2_FTS_AUX_HEX_NAME flag in SYS_TABLES. |
6120 | @return DB_SUCCESS or error code. */ |
6121 | static |
6122 | dberr_t |
6123 | fts_update_hex_format_flag( |
6124 | /*=======================*/ |
6125 | trx_t* trx, /*!< in/out: transaction that |
6126 | covers the update */ |
6127 | table_id_t table_id, /*!< in: Table for which we want |
6128 | to set the root table->flags2 */ |
6129 | bool dict_locked) /*!< in: set to true if the |
6130 | caller already owns the |
6131 | dict_sys_t::mutex. */ |
6132 | { |
6133 | pars_info_t* info; |
6134 | ib_uint32_t flags2; |
6135 | |
6136 | static const char sql[] = |
6137 | "PROCEDURE UPDATE_HEX_FORMAT_FLAG() IS\n" |
6138 | "DECLARE FUNCTION my_func;\n" |
6139 | "DECLARE CURSOR c IS\n" |
6140 | " SELECT MIX_LEN" |
6141 | " FROM SYS_TABLES" |
6142 | " WHERE ID = :table_id FOR UPDATE;" |
6143 | "\n" |
6144 | "BEGIN\n" |
6145 | "OPEN c;\n" |
6146 | "WHILE 1 = 1 LOOP\n" |
6147 | " FETCH c INTO my_func();\n" |
6148 | " IF c % NOTFOUND THEN\n" |
6149 | " EXIT;\n" |
6150 | " END IF;\n" |
6151 | "END LOOP;\n" |
6152 | "UPDATE SYS_TABLES" |
6153 | " SET MIX_LEN = :flags2" |
6154 | " WHERE ID = :table_id;\n" |
6155 | "CLOSE c;\n" |
6156 | "END;\n" ; |
6157 | |
6158 | flags2 = ULINT32_UNDEFINED; |
6159 | |
6160 | info = pars_info_create(); |
6161 | |
6162 | pars_info_add_ull_literal(info, "table_id" , table_id); |
6163 | pars_info_bind_int4_literal(info, "flags2" , &flags2); |
6164 | |
6165 | pars_info_bind_function( |
6166 | info, "my_func" , fts_set_hex_format, &flags2); |
6167 | |
6168 | if (trx_get_dict_operation(trx) == TRX_DICT_OP_NONE) { |
6169 | trx_set_dict_operation(trx, TRX_DICT_OP_INDEX); |
6170 | } |
6171 | |
6172 | dberr_t err = que_eval_sql(info, sql, !dict_locked, trx); |
6173 | |
6174 | ut_a(flags2 != ULINT32_UNDEFINED); |
6175 | |
6176 | return(err); |
6177 | } |
6178 | |
6179 | /*********************************************************************//** |
6180 | Rename an aux table to HEX format. It's called when "%016llu" is used |
6181 | to format an object id in table name, which only happens in Windows. */ |
6182 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
6183 | dberr_t |
6184 | fts_rename_one_aux_table_to_hex_format( |
6185 | /*===================================*/ |
6186 | trx_t* trx, /*!< in: transaction */ |
6187 | const fts_aux_table_t* aux_table, /*!< in: table info */ |
6188 | const dict_table_t* parent_table) /*!< in: parent table name */ |
6189 | { |
6190 | const char* ptr; |
6191 | fts_table_t fts_table; |
6192 | char new_name[MAX_FULL_NAME_LEN]; |
6193 | dberr_t error; |
6194 | |
6195 | ptr = strchr(aux_table->name, '/'); |
6196 | ut_a(ptr != NULL); |
6197 | ++ptr; |
6198 | /* Skip "FTS_", table id and underscore */ |
6199 | for (ulint i = 0; i < 2; ++i) { |
6200 | ptr = strchr(ptr, '_'); |
6201 | ut_a(ptr != NULL); |
6202 | ++ptr; |
6203 | } |
6204 | |
6205 | fts_table.suffix = NULL; |
6206 | if (aux_table->index_id == 0) { |
6207 | fts_table.type = FTS_COMMON_TABLE; |
6208 | |
6209 | for (ulint i = 0; fts_common_tables[i] != NULL; ++i) { |
6210 | if (strcmp(ptr, fts_common_tables[i]) == 0) { |
6211 | fts_table.suffix = fts_common_tables[i]; |
6212 | break; |
6213 | } |
6214 | } |
6215 | } else { |
6216 | fts_table.type = FTS_INDEX_TABLE; |
6217 | |
6218 | /* Skip index id and underscore */ |
6219 | ptr = strchr(ptr, '_'); |
6220 | ut_a(ptr != NULL); |
6221 | ++ptr; |
6222 | |
6223 | for (ulint i = 0; fts_index_selector[i].value; ++i) { |
6224 | if (strcmp(ptr, fts_get_suffix(i)) == 0) { |
6225 | fts_table.suffix = fts_get_suffix(i); |
6226 | break; |
6227 | } |
6228 | } |
6229 | } |
6230 | |
6231 | ut_a(fts_table.suffix != NULL); |
6232 | |
6233 | fts_table.parent = parent_table->name.m_name; |
6234 | fts_table.table_id = aux_table->parent_id; |
6235 | fts_table.index_id = aux_table->index_id; |
6236 | fts_table.table = parent_table; |
6237 | |
6238 | fts_get_table_name(&fts_table, new_name); |
6239 | ut_ad(strcmp(new_name, aux_table->name) != 0); |
6240 | |
6241 | if (trx_get_dict_operation(trx) == TRX_DICT_OP_NONE) { |
6242 | trx_set_dict_operation(trx, TRX_DICT_OP_INDEX); |
6243 | } |
6244 | |
6245 | error = row_rename_table_for_mysql(aux_table->name, new_name, trx, |
6246 | FALSE); |
6247 | |
6248 | if (error != DB_SUCCESS) { |
6249 | ib::warn() << "Failed to rename aux table '" |
6250 | << aux_table->name << "' to new format '" |
6251 | << new_name << "'." ; |
6252 | } else { |
6253 | ib::info() << "Renamed aux table '" << aux_table->name |
6254 | << "' to '" << new_name << "'." ; |
6255 | } |
6256 | |
6257 | return(error); |
6258 | } |
6259 | |
6260 | /**********************************************************************//** |
6261 | Rename all aux tables of a parent table to HEX format. Also set aux tables' |
6262 | flags2 and parent table's flags2 with DICT_TF2_FTS_AUX_HEX_NAME. |
6263 | It's called when "%016llu" is used to format an object id in table name, |
6264 | which only happens in Windows. |
6265 | Note the ids in tables are correct but the names are old ambiguous ones. |
6266 | |
6267 | This function should make sure that either all the parent table and aux tables |
6268 | are set DICT_TF2_FTS_AUX_HEX_NAME with flags2 or none of them are set */ |
6269 | static MY_ATTRIBUTE((nonnull, warn_unused_result)) |
6270 | dberr_t |
6271 | fts_rename_aux_tables_to_hex_format_low( |
6272 | /*====================================*/ |
6273 | trx_t* trx, /*!< in: transaction */ |
6274 | dict_table_t* parent_table, /*!< in: parent table */ |
6275 | ib_vector_t* tables) /*!< in: aux tables to rename. */ |
6276 | { |
6277 | dberr_t error; |
6278 | ulint count; |
6279 | |
6280 | ut_ad(!DICT_TF2_FLAG_IS_SET(parent_table, DICT_TF2_FTS_AUX_HEX_NAME)); |
6281 | ut_ad(!ib_vector_is_empty(tables)); |
6282 | |
6283 | error = fts_update_hex_format_flag(trx, parent_table->id, true); |
6284 | |
6285 | if (error != DB_SUCCESS) { |
6286 | ib::warn() << "Setting parent table " << parent_table->name |
6287 | << " to hex format failed." ; |
6288 | fts_sql_rollback(trx); |
6289 | return(error); |
6290 | } |
6291 | |
6292 | DICT_TF2_FLAG_SET(parent_table, DICT_TF2_FTS_AUX_HEX_NAME); |
6293 | |
6294 | for (count = 0; count < ib_vector_size(tables); ++count) { |
6295 | dict_table_t* table; |
6296 | fts_aux_table_t* aux_table; |
6297 | |
6298 | aux_table = static_cast<fts_aux_table_t*>( |
6299 | ib_vector_get(tables, count)); |
6300 | |
6301 | table = dict_table_open_on_id(aux_table->id, TRUE, |
6302 | DICT_TABLE_OP_NORMAL); |
6303 | |
6304 | ut_ad(table != NULL); |
6305 | ut_ad(!DICT_TF2_FLAG_IS_SET(table, DICT_TF2_FTS_AUX_HEX_NAME)); |
6306 | |
6307 | /* Set HEX_NAME flag here to make sure we can get correct |
6308 | new table name in following function */ |
6309 | DICT_TF2_FLAG_SET(table, DICT_TF2_FTS_AUX_HEX_NAME); |
6310 | error = fts_rename_one_aux_table_to_hex_format(trx, |
6311 | aux_table, parent_table); |
6312 | /* We will rollback the trx if the error != DB_SUCCESS, |
6313 | so setting the flag here is the same with setting it in |
6314 | row_rename_table_for_mysql */ |
6315 | DBUG_EXECUTE_IF("rename_aux_table_fail" , error = DB_ERROR;); |
6316 | |
6317 | if (error != DB_SUCCESS) { |
6318 | dict_table_close(table, TRUE, FALSE); |
6319 | |
6320 | ib::warn() << "Failed to rename one aux table " |
6321 | << aux_table->name << ". Will revert" |
6322 | " all successful rename operations." ; |
6323 | |
6324 | fts_sql_rollback(trx); |
6325 | break; |
6326 | } |
6327 | |
6328 | error = fts_update_hex_format_flag(trx, aux_table->id, true); |
6329 | dict_table_close(table, TRUE, FALSE); |
6330 | |
6331 | if (error != DB_SUCCESS) { |
6332 | ib::warn() << "Setting aux table " << aux_table->name |
6333 | << " to hex format failed." ; |
6334 | |
6335 | fts_sql_rollback(trx); |
6336 | break; |
6337 | } |
6338 | } |
6339 | |
6340 | if (error != DB_SUCCESS) { |
6341 | ut_ad(count != ib_vector_size(tables)); |
6342 | |
6343 | /* If rename fails, thr trx would be rolled back, we can't |
6344 | use it any more, we'll start a new background trx to do |
6345 | the reverting. */ |
6346 | |
6347 | ut_ad(!trx_is_started(trx)); |
6348 | |
6349 | bool not_rename = false; |
6350 | |
6351 | /* Try to revert those succesful rename operations |
6352 | in order to revert the ibd file rename. */ |
6353 | for (ulint i = 0; i <= count; ++i) { |
6354 | dict_table_t* table; |
6355 | fts_aux_table_t* aux_table; |
6356 | trx_t* trx_bg; |
6357 | dberr_t err; |
6358 | |
6359 | aux_table = static_cast<fts_aux_table_t*>( |
6360 | ib_vector_get(tables, i)); |
6361 | |
6362 | table = dict_table_open_on_id(aux_table->id, TRUE, |
6363 | DICT_TABLE_OP_NORMAL); |
6364 | ut_ad(table != NULL); |
6365 | |
6366 | if (not_rename) { |
6367 | DICT_TF2_FLAG_UNSET(table, |
6368 | DICT_TF2_FTS_AUX_HEX_NAME); |
6369 | } |
6370 | |
6371 | if (!DICT_TF2_FLAG_IS_SET(table, |
6372 | DICT_TF2_FTS_AUX_HEX_NAME)) { |
6373 | dict_table_close(table, TRUE, FALSE); |
6374 | continue; |
6375 | } |
6376 | |
6377 | trx_bg = trx_create(); |
6378 | trx_bg->op_info = "Revert half done rename" ; |
6379 | trx_bg->dict_operation_lock_mode = RW_X_LATCH; |
6380 | trx_start_for_ddl(trx_bg, TRX_DICT_OP_TABLE); |
6381 | |
6382 | DICT_TF2_FLAG_UNSET(table, DICT_TF2_FTS_AUX_HEX_NAME); |
6383 | err = row_rename_table_for_mysql(table->name.m_name, |
6384 | aux_table->name, |
6385 | trx_bg, FALSE); |
6386 | |
6387 | trx_bg->dict_operation_lock_mode = 0; |
6388 | dict_table_close(table, TRUE, FALSE); |
6389 | |
6390 | if (err != DB_SUCCESS) { |
6391 | ib::warn() << "Failed to revert table " |
6392 | << table->name << ". Please revert" |
6393 | " manually." ; |
6394 | fts_sql_rollback(trx_bg); |
6395 | trx_free(trx_bg); |
6396 | /* Continue to clear aux tables' flags2 */ |
6397 | not_rename = true; |
6398 | continue; |
6399 | } |
6400 | |
6401 | fts_sql_commit(trx_bg); |
6402 | trx_free(trx_bg); |
6403 | } |
6404 | |
6405 | DICT_TF2_FLAG_UNSET(parent_table, DICT_TF2_FTS_AUX_HEX_NAME); |
6406 | } |
6407 | |
6408 | return(error); |
6409 | } |
6410 | |
6411 | /**********************************************************************//** |
6412 | Convert an id, which is actually a decimal number but was regard as a HEX |
6413 | from a string, to its real value. */ |
6414 | static |
6415 | ib_id_t |
6416 | fts_fake_hex_to_dec( |
6417 | /*================*/ |
6418 | ib_id_t id) /*!< in: number to convert */ |
6419 | { |
6420 | ib_id_t dec_id = 0; |
6421 | char tmp_id[FTS_AUX_MIN_TABLE_ID_LENGTH]; |
6422 | |
6423 | #ifdef UNIV_DEBUG |
6424 | int ret = |
6425 | #endif /* UNIV_DEBUG */ |
6426 | sprintf(tmp_id, UINT64PFx, id); |
6427 | ut_ad(ret == 16); |
6428 | #ifdef UNIV_DEBUG |
6429 | ret = |
6430 | #endif /* UNIV_DEBUG */ |
6431 | sscanf(tmp_id, "%016" UINT64scan, &dec_id); |
6432 | ut_ad(ret == 1); |
6433 | |
6434 | return dec_id; |
6435 | } |
6436 | |
6437 | /*********************************************************************//** |
6438 | Compare two fts_aux_table_t parent_ids. |
6439 | @return < 0 if n1 < n2, 0 if n1 == n2, > 0 if n1 > n2 */ |
6440 | UNIV_INLINE |
6441 | int |
6442 | fts_check_aux_table_parent_id_cmp( |
6443 | /*==============================*/ |
6444 | const void* p1, /*!< in: id1 */ |
6445 | const void* p2) /*!< in: id2 */ |
6446 | { |
6447 | const fts_aux_table_t* fa1 = static_cast<const fts_aux_table_t*>(p1); |
6448 | const fts_aux_table_t* fa2 = static_cast<const fts_aux_table_t*>(p2); |
6449 | |
6450 | return static_cast<int>(fa1->parent_id - fa2->parent_id); |
6451 | } |
6452 | |
6453 | /** Mark all the fts index associated with the parent table as corrupted. |
6454 | @param[in] trx transaction |
6455 | @param[in, out] parent_table fts index associated with this parent table |
6456 | will be marked as corrupted. */ |
6457 | static |
6458 | void |
6459 | fts_parent_all_index_set_corrupt( |
6460 | trx_t* trx, |
6461 | dict_table_t* parent_table) |
6462 | { |
6463 | fts_t* fts = parent_table->fts; |
6464 | |
6465 | if (trx_get_dict_operation(trx) == TRX_DICT_OP_NONE) { |
6466 | trx_set_dict_operation(trx, TRX_DICT_OP_INDEX); |
6467 | } |
6468 | |
6469 | for (ulint j = 0; j < ib_vector_size(fts->indexes); j++) { |
6470 | dict_index_t* index = static_cast<dict_index_t*>( |
6471 | ib_vector_getp_const(fts->indexes, j)); |
6472 | dict_set_corrupted(index, |
6473 | trx, "DROP ORPHANED TABLE" ); |
6474 | } |
6475 | } |
6476 | |
6477 | /** Mark the fts index which index id matches the id as corrupted. |
6478 | @param[in] trx transaction |
6479 | @param[in] id index id to search |
6480 | @param[in, out] parent_table parent table to check with all |
6481 | the index. */ |
6482 | static |
6483 | void |
6484 | fts_set_index_corrupt( |
6485 | trx_t* trx, |
6486 | index_id_t id, |
6487 | dict_table_t* table) |
6488 | { |
6489 | fts_t* fts = table->fts; |
6490 | |
6491 | if (trx_get_dict_operation(trx) == TRX_DICT_OP_NONE) { |
6492 | trx_set_dict_operation(trx, TRX_DICT_OP_INDEX); |
6493 | } |
6494 | |
6495 | for (ulint j = 0; j < ib_vector_size(fts->indexes); j++) { |
6496 | dict_index_t* index = static_cast<dict_index_t*>( |
6497 | ib_vector_getp_const(fts->indexes, j)); |
6498 | if (index->id == id) { |
6499 | dict_set_corrupted(index, trx, |
6500 | "DROP ORPHANED TABLE" ); |
6501 | break; |
6502 | } |
6503 | } |
6504 | } |
6505 | |
6506 | /** Check the index for the aux table is corrupted. |
6507 | @param[in] aux_table auxiliary table |
6508 | @retval nonzero if index is corrupted, zero for valid index */ |
6509 | static |
6510 | ulint |
6511 | fts_check_corrupt_index( |
6512 | fts_aux_table_t* aux_table) |
6513 | { |
6514 | dict_table_t* table; |
6515 | dict_index_t* index; |
6516 | table = dict_table_open_on_id( |
6517 | aux_table->parent_id, TRUE, DICT_TABLE_OP_NORMAL); |
6518 | |
6519 | if (table == NULL) { |
6520 | return(0); |
6521 | } |
6522 | |
6523 | for (index = UT_LIST_GET_FIRST(table->indexes); |
6524 | index; |
6525 | index = UT_LIST_GET_NEXT(indexes, index)) { |
6526 | if (index->id == aux_table->index_id) { |
6527 | ut_ad(index->type & DICT_FTS); |
6528 | dict_table_close(table, true, false); |
6529 | return index->is_corrupted(); |
6530 | } |
6531 | } |
6532 | |
6533 | dict_table_close(table, true, false); |
6534 | return(0); |
6535 | } |
6536 | |
6537 | /* Get parent table name if it's a fts aux table |
6538 | @param[in] aux_table_name aux table name |
6539 | @param[in] aux_table_len aux table length |
6540 | @return parent table name, or NULL */ |
6541 | char* |
6542 | fts_get_parent_table_name( |
6543 | const char* aux_table_name, |
6544 | ulint aux_table_len) |
6545 | { |
6546 | fts_aux_table_t aux_table; |
6547 | char* parent_table_name = NULL; |
6548 | |
6549 | if (fts_is_aux_table_name(&aux_table, aux_table_name, aux_table_len)) { |
6550 | dict_table_t* parent_table; |
6551 | |
6552 | parent_table = dict_table_open_on_id( |
6553 | aux_table.parent_id, TRUE, DICT_TABLE_OP_NORMAL); |
6554 | |
6555 | if (parent_table != NULL) { |
6556 | parent_table_name = mem_strdupl( |
6557 | parent_table->name.m_name, |
6558 | strlen(parent_table->name.m_name)); |
6559 | |
6560 | dict_table_close(parent_table, TRUE, FALSE); |
6561 | } |
6562 | } |
6563 | |
6564 | return(parent_table_name); |
6565 | } |
6566 | |
6567 | /** Check the validity of the parent table. |
6568 | @param[in] aux_table auxiliary table |
6569 | @return true if it is a valid table or false if it is not */ |
6570 | static |
6571 | bool |
6572 | fts_valid_parent_table( |
6573 | const fts_aux_table_t* aux_table) |
6574 | { |
6575 | dict_table_t* parent_table; |
6576 | bool valid = false; |
6577 | |
6578 | parent_table = dict_table_open_on_id( |
6579 | aux_table->parent_id, TRUE, DICT_TABLE_OP_NORMAL); |
6580 | |
6581 | if (parent_table != NULL && parent_table->fts != NULL) { |
6582 | if (aux_table->index_id == 0) { |
6583 | valid = true; |
6584 | } else { |
6585 | index_id_t id = aux_table->index_id; |
6586 | dict_index_t* index; |
6587 | |
6588 | /* Search for the FT index in the table's list. */ |
6589 | for (index = UT_LIST_GET_FIRST(parent_table->indexes); |
6590 | index; |
6591 | index = UT_LIST_GET_NEXT(indexes, index)) { |
6592 | if (index->id == id) { |
6593 | valid = true; |
6594 | break; |
6595 | } |
6596 | |
6597 | } |
6598 | } |
6599 | } |
6600 | |
6601 | if (parent_table) { |
6602 | dict_table_close(parent_table, TRUE, FALSE); |
6603 | } |
6604 | |
6605 | return(valid); |
6606 | } |
6607 | |
6608 | /** Try to rename all aux tables of the specified parent table. |
6609 | @param[in] aux_tables aux_tables to be renamed |
6610 | @param[in] parent_table parent table of all aux |
6611 | tables stored in tables. */ |
6612 | static |
6613 | void |
6614 | fts_rename_aux_tables_to_hex_format( |
6615 | ib_vector_t* aux_tables, |
6616 | dict_table_t* parent_table) |
6617 | { |
6618 | dberr_t err; |
6619 | trx_t* trx_rename = trx_create(); |
6620 | trx_rename->op_info = "Rename aux tables to hex format" ; |
6621 | trx_rename->dict_operation_lock_mode = RW_X_LATCH; |
6622 | trx_start_for_ddl(trx_rename, TRX_DICT_OP_TABLE); |
6623 | |
6624 | err = fts_rename_aux_tables_to_hex_format_low(trx_rename, |
6625 | parent_table, aux_tables); |
6626 | |
6627 | trx_rename->dict_operation_lock_mode = 0; |
6628 | |
6629 | if (err != DB_SUCCESS) { |
6630 | |
6631 | ib::warn() << "Rollback operations on all aux tables of " |
6632 | "table " << parent_table->name << ". All the fts index " |
6633 | "associated with the table are marked as corrupted. " |
6634 | "Please rebuild the index again." ; |
6635 | |
6636 | /* Corrupting the fts index related to parent table. */ |
6637 | trx_t* trx_corrupt; |
6638 | trx_corrupt = trx_create(); |
6639 | trx_corrupt->dict_operation_lock_mode = RW_X_LATCH; |
6640 | trx_start_for_ddl(trx_corrupt, TRX_DICT_OP_TABLE); |
6641 | fts_parent_all_index_set_corrupt(trx_corrupt, parent_table); |
6642 | trx_corrupt->dict_operation_lock_mode = 0; |
6643 | fts_sql_commit(trx_corrupt); |
6644 | trx_free(trx_corrupt); |
6645 | } else { |
6646 | fts_sql_commit(trx_rename); |
6647 | } |
6648 | |
6649 | trx_free(trx_rename); |
6650 | ib_vector_reset(aux_tables); |
6651 | } |
6652 | |
6653 | /** Set the hex format flag for the parent table. |
6654 | @param[in, out] parent_table parent table |
6655 | @param[in] trx transaction */ |
6656 | static |
6657 | void |
6658 | fts_set_parent_hex_format_flag( |
6659 | dict_table_t* parent_table, |
6660 | trx_t* trx) |
6661 | { |
6662 | if (!DICT_TF2_FLAG_IS_SET(parent_table, |
6663 | DICT_TF2_FTS_AUX_HEX_NAME)) { |
6664 | DBUG_EXECUTE_IF("parent_table_flag_fail" , DBUG_SUICIDE();); |
6665 | |
6666 | dberr_t err = fts_update_hex_format_flag( |
6667 | trx, parent_table->id, true); |
6668 | |
6669 | if (err != DB_SUCCESS) { |
6670 | ib::fatal() << "Setting parent table " |
6671 | << parent_table->name |
6672 | << "to hex format failed. Please try " |
6673 | << "to restart the server again, if it " |
6674 | << "doesn't work, the system tables " |
6675 | << "might be corrupted." ; |
6676 | } else { |
6677 | DICT_TF2_FLAG_SET( |
6678 | parent_table, DICT_TF2_FTS_AUX_HEX_NAME); |
6679 | } |
6680 | } |
6681 | } |
6682 | |
6683 | /** Drop the obsolete auxilary table. |
6684 | @param[in] tables tables to be dropped. */ |
6685 | static |
6686 | void |
6687 | fts_drop_obsolete_aux_table_from_vector( |
6688 | ib_vector_t* tables) |
6689 | { |
6690 | dberr_t err; |
6691 | |
6692 | for (ulint count = 0; count < ib_vector_size(tables); |
6693 | ++count) { |
6694 | |
6695 | fts_aux_table_t* aux_drop_table; |
6696 | aux_drop_table = static_cast<fts_aux_table_t*>( |
6697 | ib_vector_get(tables, count)); |
6698 | trx_t* trx_drop = trx_create(); |
6699 | trx_drop->op_info = "Drop obsolete aux tables" ; |
6700 | trx_drop->dict_operation_lock_mode = RW_X_LATCH; |
6701 | trx_start_for_ddl(trx_drop, TRX_DICT_OP_TABLE); |
6702 | |
6703 | err = row_drop_table_for_mysql( |
6704 | aux_drop_table->name, trx_drop, false, true); |
6705 | |
6706 | trx_drop->dict_operation_lock_mode = 0; |
6707 | |
6708 | if (err != DB_SUCCESS) { |
6709 | /* We don't need to worry about the |
6710 | failure, since server would try to |
6711 | drop it on next restart, even if |
6712 | the table was broken. */ |
6713 | ib::warn() << "Failed to drop obsolete aux table " |
6714 | << aux_drop_table->name << ", which is " |
6715 | << "harmless. will try to drop it on next " |
6716 | << "restart." ; |
6717 | |
6718 | fts_sql_rollback(trx_drop); |
6719 | } else { |
6720 | ib::info() << "Dropped obsolete aux" |
6721 | " table '" << aux_drop_table->name |
6722 | << "'." ; |
6723 | |
6724 | fts_sql_commit(trx_drop); |
6725 | } |
6726 | |
6727 | trx_free(trx_drop); |
6728 | } |
6729 | } |
6730 | |
6731 | /** Drop all the auxiliary table present in the vector. |
6732 | @param[in] trx transaction |
6733 | @param[in] tables tables to be dropped */ |
6734 | static |
6735 | void |
6736 | fts_drop_aux_table_from_vector( |
6737 | trx_t* trx, |
6738 | ib_vector_t* tables) |
6739 | { |
6740 | for (ulint count = 0; count < ib_vector_size(tables); |
6741 | ++count) { |
6742 | fts_aux_table_t* aux_drop_table; |
6743 | aux_drop_table = static_cast<fts_aux_table_t*>( |
6744 | ib_vector_get(tables, count)); |
6745 | |
6746 | /* Check for the validity of the parent table */ |
6747 | if (!fts_valid_parent_table(aux_drop_table)) { |
6748 | |
6749 | ib::warn() << "Parent table of FTS auxiliary table " |
6750 | << aux_drop_table->name << " not found." ; |
6751 | |
6752 | dberr_t err = fts_drop_table(trx, aux_drop_table->name); |
6753 | if (err == DB_FAIL) { |
6754 | |
6755 | char* path = fil_make_filepath( |
6756 | NULL, aux_drop_table->name, IBD, false); |
6757 | |
6758 | if (path != NULL) { |
6759 | os_file_delete_if_exists( |
6760 | innodb_data_file_key, |
6761 | path , NULL); |
6762 | ut_free(path); |
6763 | } |
6764 | } |
6765 | } |
6766 | } |
6767 | } |
6768 | |
6769 | /**********************************************************************//** |
6770 | Check and drop all orphaned FTS auxiliary tables, those that don't have |
6771 | a parent table or FTS index defined on them. |
6772 | @return DB_SUCCESS or error code */ |
6773 | static MY_ATTRIBUTE((nonnull)) |
6774 | void |
6775 | fts_check_and_drop_orphaned_tables( |
6776 | /*===============================*/ |
6777 | trx_t* trx, /*!< in: transaction */ |
6778 | ib_vector_t* tables) /*!< in: tables to check */ |
6779 | { |
6780 | mem_heap_t* heap; |
6781 | ib_vector_t* aux_tables_to_rename; |
6782 | ib_vector_t* invalid_aux_tables; |
6783 | ib_vector_t* valid_aux_tables; |
6784 | ib_vector_t* drop_aux_tables; |
6785 | ib_vector_t* obsolete_aux_tables; |
6786 | ib_alloc_t* heap_alloc; |
6787 | |
6788 | heap = mem_heap_create(1024); |
6789 | heap_alloc = ib_heap_allocator_create(heap); |
6790 | |
6791 | /* We store all aux tables belonging to the same parent table here, |
6792 | and rename all these tables in a batch mode. */ |
6793 | aux_tables_to_rename = ib_vector_create(heap_alloc, |
6794 | sizeof(fts_aux_table_t), 128); |
6795 | |
6796 | /* We store all fake auxiliary table and orphaned table here. */ |
6797 | invalid_aux_tables = ib_vector_create(heap_alloc, |
6798 | sizeof(fts_aux_table_t), 128); |
6799 | |
6800 | /* We store all valid aux tables. We use this to filter the |
6801 | fake auxiliary table from invalid auxiliary tables. */ |
6802 | valid_aux_tables = ib_vector_create(heap_alloc, |
6803 | sizeof(fts_aux_table_t), 128); |
6804 | |
6805 | /* We store all auxiliary tables to be dropped. */ |
6806 | drop_aux_tables = ib_vector_create(heap_alloc, |
6807 | sizeof(fts_aux_table_t), 128); |
6808 | |
6809 | /* We store all obsolete auxiliary tables to be dropped. */ |
6810 | obsolete_aux_tables = ib_vector_create(heap_alloc, |
6811 | sizeof(fts_aux_table_t), 128); |
6812 | |
6813 | /* Sort by parent_id first, in case rename will fail */ |
6814 | ib_vector_sort(tables, fts_check_aux_table_parent_id_cmp); |
6815 | |
6816 | for (ulint i = 0; i < ib_vector_size(tables); ++i) { |
6817 | dict_table_t* parent_table; |
6818 | fts_aux_table_t* aux_table; |
6819 | bool drop = false; |
6820 | dict_table_t* table; |
6821 | fts_aux_table_t* next_aux_table = NULL; |
6822 | ib_id_t orig_parent_id = 0; |
6823 | ib_id_t orig_index_id = 0; |
6824 | bool rename = false; |
6825 | |
6826 | aux_table = static_cast<fts_aux_table_t*>( |
6827 | ib_vector_get(tables, i)); |
6828 | |
6829 | table = dict_table_open_on_id( |
6830 | aux_table->id, TRUE, DICT_TABLE_OP_NORMAL); |
6831 | orig_parent_id = aux_table->parent_id; |
6832 | orig_index_id = aux_table->index_id; |
6833 | |
6834 | if (table == NULL |
6835 | || strcmp(table->name.m_name, aux_table->name)) { |
6836 | |
6837 | bool fake_aux = false; |
6838 | |
6839 | if (table != NULL) { |
6840 | dict_table_close(table, TRUE, FALSE); |
6841 | } |
6842 | |
6843 | if (i + 1 < ib_vector_size(tables)) { |
6844 | next_aux_table = static_cast<fts_aux_table_t*>( |
6845 | ib_vector_get(tables, i + 1)); |
6846 | } |
6847 | |
6848 | /* To know whether aux table is fake fts or |
6849 | orphan fts table. */ |
6850 | for (ulint count = 0; |
6851 | count < ib_vector_size(valid_aux_tables); |
6852 | count++) { |
6853 | fts_aux_table_t* valid_aux; |
6854 | valid_aux = static_cast<fts_aux_table_t*>( |
6855 | ib_vector_get(valid_aux_tables, count)); |
6856 | if (strcmp(valid_aux->name, |
6857 | aux_table->name) == 0) { |
6858 | fake_aux = true; |
6859 | break; |
6860 | } |
6861 | } |
6862 | |
6863 | /* All aux tables of parent table, whose id is |
6864 | last_parent_id, have been checked, try to rename |
6865 | them if necessary. */ |
6866 | if ((next_aux_table == NULL |
6867 | || orig_parent_id != next_aux_table->parent_id) |
6868 | && (!ib_vector_is_empty(aux_tables_to_rename))) { |
6869 | |
6870 | ib_id_t parent_id = fts_fake_hex_to_dec( |
6871 | aux_table->parent_id); |
6872 | |
6873 | parent_table = dict_table_open_on_id( |
6874 | parent_id, TRUE, |
6875 | DICT_TABLE_OP_NORMAL); |
6876 | |
6877 | fts_rename_aux_tables_to_hex_format( |
6878 | aux_tables_to_rename, parent_table); |
6879 | |
6880 | dict_table_close(parent_table, TRUE, |
6881 | FALSE); |
6882 | } |
6883 | |
6884 | /* If the aux table is fake aux table. Skip it. */ |
6885 | if (!fake_aux) { |
6886 | ib_vector_push(invalid_aux_tables, aux_table); |
6887 | } |
6888 | |
6889 | continue; |
6890 | } else if (!DICT_TF2_FLAG_IS_SET(table, |
6891 | DICT_TF2_FTS_AUX_HEX_NAME)) { |
6892 | |
6893 | aux_table->parent_id = fts_fake_hex_to_dec( |
6894 | aux_table->parent_id); |
6895 | |
6896 | if (aux_table->index_id != 0) { |
6897 | aux_table->index_id = fts_fake_hex_to_dec( |
6898 | aux_table->index_id); |
6899 | } |
6900 | |
6901 | ut_ad(aux_table->id > aux_table->parent_id); |
6902 | |
6903 | /* Check whether parent table id and index id |
6904 | are stored as decimal format. */ |
6905 | if (fts_valid_parent_table(aux_table)) { |
6906 | |
6907 | parent_table = dict_table_open_on_id( |
6908 | aux_table->parent_id, true, |
6909 | DICT_TABLE_OP_NORMAL); |
6910 | |
6911 | ut_ad(parent_table != NULL); |
6912 | ut_ad(parent_table->fts != NULL); |
6913 | |
6914 | if (!DICT_TF2_FLAG_IS_SET( |
6915 | parent_table, |
6916 | DICT_TF2_FTS_AUX_HEX_NAME)) { |
6917 | rename = true; |
6918 | } |
6919 | |
6920 | dict_table_close(parent_table, TRUE, FALSE); |
6921 | } |
6922 | |
6923 | if (!rename) { |
6924 | /* Reassign the original value of |
6925 | aux table if it is not in decimal format */ |
6926 | aux_table->parent_id = orig_parent_id; |
6927 | aux_table->index_id = orig_index_id; |
6928 | } |
6929 | } |
6930 | |
6931 | if (table != NULL) { |
6932 | dict_table_close(table, TRUE, FALSE); |
6933 | } |
6934 | |
6935 | if (!rename) { |
6936 | /* Check the validity of the parent table. */ |
6937 | if (!fts_valid_parent_table(aux_table)) { |
6938 | drop = true; |
6939 | } |
6940 | } |
6941 | |
6942 | /* Filter out the fake aux table by comparing with the |
6943 | current valid auxiliary table name. */ |
6944 | for (ulint count = 0; |
6945 | count < ib_vector_size(invalid_aux_tables); count++) { |
6946 | fts_aux_table_t* invalid_aux; |
6947 | invalid_aux = static_cast<fts_aux_table_t*>( |
6948 | ib_vector_get(invalid_aux_tables, count)); |
6949 | if (strcmp(invalid_aux->name, aux_table->name) == 0) { |
6950 | ib_vector_remove( |
6951 | invalid_aux_tables, |
6952 | *reinterpret_cast<void**>(invalid_aux)); |
6953 | break; |
6954 | } |
6955 | } |
6956 | |
6957 | ib_vector_push(valid_aux_tables, aux_table); |
6958 | |
6959 | /* If the index associated with aux table is corrupted, |
6960 | skip it. */ |
6961 | if (fts_check_corrupt_index(aux_table) > 0) { |
6962 | |
6963 | if (i + 1 < ib_vector_size(tables)) { |
6964 | next_aux_table = static_cast<fts_aux_table_t*>( |
6965 | ib_vector_get(tables, i + 1)); |
6966 | } |
6967 | |
6968 | if (next_aux_table == NULL |
6969 | || orig_parent_id != next_aux_table->parent_id) { |
6970 | |
6971 | parent_table = dict_table_open_on_id( |
6972 | aux_table->parent_id, TRUE, |
6973 | DICT_TABLE_OP_NORMAL); |
6974 | |
6975 | if (!ib_vector_is_empty(aux_tables_to_rename)) { |
6976 | fts_rename_aux_tables_to_hex_format( |
6977 | aux_tables_to_rename, parent_table); |
6978 | } else { |
6979 | fts_set_parent_hex_format_flag( |
6980 | parent_table, trx); |
6981 | } |
6982 | |
6983 | dict_table_close(parent_table, TRUE, FALSE); |
6984 | } |
6985 | |
6986 | continue; |
6987 | } |
6988 | |
6989 | parent_table = dict_table_open_on_id( |
6990 | aux_table->parent_id, TRUE, DICT_TABLE_OP_NORMAL); |
6991 | |
6992 | if (drop) { |
6993 | ib_vector_push(drop_aux_tables, aux_table); |
6994 | } else { |
6995 | if (FTS_IS_OBSOLETE_AUX_TABLE(aux_table->name)) { |
6996 | ib_vector_push(obsolete_aux_tables, aux_table); |
6997 | continue; |
6998 | } |
6999 | } |
7000 | |
7001 | /* If the aux table is in decimal format, we should |
7002 | rename it, so push it to aux_tables_to_rename */ |
7003 | if (!drop && rename) { |
7004 | bool rename_table = true; |
7005 | for (ulint count = 0; |
7006 | count < ib_vector_size(aux_tables_to_rename); |
7007 | count++) { |
7008 | fts_aux_table_t* rename_aux = |
7009 | static_cast<fts_aux_table_t*>( |
7010 | ib_vector_get(aux_tables_to_rename, |
7011 | count)); |
7012 | if (strcmp(rename_aux->name, |
7013 | aux_table->name) == 0) { |
7014 | rename_table = false; |
7015 | break; |
7016 | } |
7017 | } |
7018 | |
7019 | if (rename_table) { |
7020 | ib_vector_push(aux_tables_to_rename, |
7021 | aux_table); |
7022 | } |
7023 | } |
7024 | |
7025 | if (i + 1 < ib_vector_size(tables)) { |
7026 | next_aux_table = static_cast<fts_aux_table_t*>( |
7027 | ib_vector_get(tables, i + 1)); |
7028 | } |
7029 | |
7030 | if ((next_aux_table == NULL |
7031 | || orig_parent_id != next_aux_table->parent_id) |
7032 | && !ib_vector_is_empty(aux_tables_to_rename)) { |
7033 | |
7034 | ut_ad(rename); |
7035 | ut_ad(!DICT_TF2_FLAG_IS_SET( |
7036 | parent_table, DICT_TF2_FTS_AUX_HEX_NAME)); |
7037 | |
7038 | fts_rename_aux_tables_to_hex_format( |
7039 | aux_tables_to_rename,parent_table); |
7040 | } |
7041 | |
7042 | /* The IDs are already in correct hex format. */ |
7043 | if (!drop && !rename) { |
7044 | dict_table_t* table; |
7045 | |
7046 | table = dict_table_open_on_id( |
7047 | aux_table->id, TRUE, DICT_TABLE_OP_NORMAL); |
7048 | |
7049 | if (table != NULL |
7050 | && strcmp(table->name.m_name, aux_table->name)) { |
7051 | dict_table_close(table, TRUE, FALSE); |
7052 | table = NULL; |
7053 | } |
7054 | |
7055 | if (table != NULL |
7056 | && !DICT_TF2_FLAG_IS_SET( |
7057 | table, |
7058 | DICT_TF2_FTS_AUX_HEX_NAME)) { |
7059 | |
7060 | DBUG_EXECUTE_IF("aux_table_flag_fail" , |
7061 | ib::warn() << "Setting aux table " |
7062 | << table->name << " to hex " |
7063 | "format failed." ; |
7064 | fts_set_index_corrupt( |
7065 | trx, aux_table->index_id, |
7066 | parent_table); |
7067 | goto table_exit;); |
7068 | |
7069 | dberr_t err = fts_update_hex_format_flag( |
7070 | trx, table->id, true); |
7071 | |
7072 | if (err != DB_SUCCESS) { |
7073 | ib::warn() << "Setting aux table " |
7074 | << table->name << " to hex " |
7075 | "format failed." ; |
7076 | |
7077 | fts_set_index_corrupt( |
7078 | trx, aux_table->index_id, |
7079 | parent_table); |
7080 | } else { |
7081 | DICT_TF2_FLAG_SET(table, |
7082 | DICT_TF2_FTS_AUX_HEX_NAME); |
7083 | } |
7084 | } |
7085 | #ifndef DBUG_OFF |
7086 | table_exit: |
7087 | #endif /* !DBUG_OFF */ |
7088 | |
7089 | if (table != NULL) { |
7090 | dict_table_close(table, TRUE, FALSE); |
7091 | } |
7092 | |
7093 | ut_ad(parent_table != NULL); |
7094 | |
7095 | fts_set_parent_hex_format_flag( |
7096 | parent_table, trx); |
7097 | } |
7098 | |
7099 | if (parent_table != NULL) { |
7100 | dict_table_close(parent_table, TRUE, FALSE); |
7101 | } |
7102 | } |
7103 | |
7104 | fts_drop_aux_table_from_vector(trx, invalid_aux_tables); |
7105 | fts_drop_aux_table_from_vector(trx, drop_aux_tables); |
7106 | fts_sql_commit(trx); |
7107 | |
7108 | fts_drop_obsolete_aux_table_from_vector(obsolete_aux_tables); |
7109 | |
7110 | /* Free the memory allocated at the beginning */ |
7111 | if (heap != NULL) { |
7112 | mem_heap_free(heap); |
7113 | } |
7114 | } |
7115 | |
7116 | /**********************************************************************//** |
7117 | Drop all orphaned FTS auxiliary tables, those that don't have a parent |
7118 | table or FTS index defined on them. */ |
7119 | void |
7120 | fts_drop_orphaned_tables(void) |
7121 | /*==========================*/ |
7122 | { |
7123 | trx_t* trx; |
7124 | pars_info_t* info; |
7125 | mem_heap_t* heap; |
7126 | que_t* graph; |
7127 | ib_vector_t* tables; |
7128 | ib_alloc_t* heap_alloc; |
7129 | |
7130 | heap = mem_heap_create(1024); |
7131 | heap_alloc = ib_heap_allocator_create(heap); |
7132 | |
7133 | /* We store the table ids of all the FTS indexes that were found. */ |
7134 | tables = ib_vector_create(heap_alloc, sizeof(fts_aux_table_t), 128); |
7135 | |
7136 | /* Get the list of all known .ibd files and check for orphaned |
7137 | FTS auxiliary files in that list. We need to remove them because |
7138 | users can't map them back to table names and this will create |
7139 | unnecessary clutter. */ |
7140 | |
7141 | mutex_enter(&fil_system.mutex); |
7142 | |
7143 | for (fil_space_t* space = UT_LIST_GET_FIRST(fil_system.space_list); |
7144 | space != NULL; |
7145 | space = UT_LIST_GET_NEXT(space_list, space)) { |
7146 | |
7147 | if (space->purpose != FIL_TYPE_TABLESPACE) { |
7148 | continue; |
7149 | } |
7150 | |
7151 | fts_aux_table_t fts_aux_table; |
7152 | memset(&fts_aux_table, 0x0, sizeof fts_aux_table); |
7153 | |
7154 | size_t len = strlen(space->name); |
7155 | |
7156 | if (!fts_is_aux_table_name(&fts_aux_table, space->name, len)) { |
7157 | continue; |
7158 | } |
7159 | |
7160 | fts_aux_table.id = space->id; |
7161 | fts_aux_table.name = mem_heap_strdupl(heap, space->name, len); |
7162 | ib_vector_push(tables, &fts_aux_table); |
7163 | } |
7164 | |
7165 | mutex_exit(&fil_system.mutex); |
7166 | |
7167 | trx = trx_create(); |
7168 | trx->op_info = "dropping orphaned FTS tables" ; |
7169 | row_mysql_lock_data_dictionary(trx); |
7170 | |
7171 | info = pars_info_create(); |
7172 | |
7173 | pars_info_bind_function(info, "my_func" , fts_read_tables, tables); |
7174 | |
7175 | graph = fts_parse_sql_no_dict_lock( |
7176 | info, |
7177 | "DECLARE FUNCTION my_func;\n" |
7178 | "DECLARE CURSOR c IS" |
7179 | " SELECT NAME, ID" |
7180 | " FROM SYS_TABLES;\n" |
7181 | "BEGIN\n" |
7182 | "\n" |
7183 | "OPEN c;\n" |
7184 | "WHILE 1 = 1 LOOP\n" |
7185 | " FETCH c INTO my_func();\n" |
7186 | " IF c % NOTFOUND THEN\n" |
7187 | " EXIT;\n" |
7188 | " END IF;\n" |
7189 | "END LOOP;\n" |
7190 | "CLOSE c;" ); |
7191 | |
7192 | for (;;) { |
7193 | dberr_t error = fts_eval_sql(trx, graph); |
7194 | |
7195 | if (error == DB_SUCCESS) { |
7196 | fts_check_and_drop_orphaned_tables(trx, tables); |
7197 | break; /* Exit the loop. */ |
7198 | } else { |
7199 | ib_vector_reset(tables); |
7200 | |
7201 | fts_sql_rollback(trx); |
7202 | |
7203 | if (error == DB_LOCK_WAIT_TIMEOUT) { |
7204 | ib::warn() << "lock wait timeout reading" |
7205 | " SYS_TABLES. Retrying!" ; |
7206 | |
7207 | trx->error_state = DB_SUCCESS; |
7208 | } else { |
7209 | ib::error() << "(" << ut_strerr(error) |
7210 | << ") while reading SYS_TABLES." ; |
7211 | |
7212 | break; /* Exit the loop. */ |
7213 | } |
7214 | } |
7215 | } |
7216 | |
7217 | que_graph_free(graph); |
7218 | |
7219 | row_mysql_unlock_data_dictionary(trx); |
7220 | |
7221 | trx_free(trx); |
7222 | |
7223 | if (heap != NULL) { |
7224 | mem_heap_free(heap); |
7225 | } |
7226 | } |
7227 | |
7228 | /**********************************************************************//** |
7229 | Check whether user supplied stopword table is of the right format. |
7230 | Caller is responsible to hold dictionary locks. |
7231 | @return the stopword column charset if qualifies */ |
7232 | CHARSET_INFO* |
7233 | fts_valid_stopword_table( |
7234 | /*=====================*/ |
7235 | const char* stopword_table_name) /*!< in: Stopword table |
7236 | name */ |
7237 | { |
7238 | dict_table_t* table; |
7239 | dict_col_t* col = NULL; |
7240 | |
7241 | if (!stopword_table_name) { |
7242 | return(NULL); |
7243 | } |
7244 | |
7245 | table = dict_table_get_low(stopword_table_name); |
7246 | |
7247 | if (!table) { |
7248 | ib::error() << "User stopword table " << stopword_table_name |
7249 | << " does not exist." ; |
7250 | |
7251 | return(NULL); |
7252 | } else { |
7253 | const char* col_name; |
7254 | |
7255 | col_name = dict_table_get_col_name(table, 0); |
7256 | |
7257 | if (ut_strcmp(col_name, "value" )) { |
7258 | ib::error() << "Invalid column name for stopword" |
7259 | " table " << stopword_table_name << ". Its" |
7260 | " first column must be named as 'value'." ; |
7261 | |
7262 | return(NULL); |
7263 | } |
7264 | |
7265 | col = dict_table_get_nth_col(table, 0); |
7266 | |
7267 | if (col->mtype != DATA_VARCHAR |
7268 | && col->mtype != DATA_VARMYSQL) { |
7269 | ib::error() << "Invalid column type for stopword" |
7270 | " table " << stopword_table_name << ". Its" |
7271 | " first column must be of varchar type" ; |
7272 | |
7273 | return(NULL); |
7274 | } |
7275 | } |
7276 | |
7277 | ut_ad(col); |
7278 | |
7279 | return(fts_get_charset(col->prtype)); |
7280 | } |
7281 | |
7282 | /**********************************************************************//** |
7283 | This function loads the stopword into the FTS cache. It also |
7284 | records/fetches stopword configuration to/from FTS configure |
7285 | table, depending on whether we are creating or reloading the |
7286 | FTS. |
7287 | @return TRUE if load operation is successful */ |
7288 | ibool |
7289 | fts_load_stopword( |
7290 | /*==============*/ |
7291 | const dict_table_t* |
7292 | table, /*!< in: Table with FTS */ |
7293 | trx_t* trx, /*!< in: Transactions */ |
7294 | const char* global_stopword_table, /*!< in: Global stopword table |
7295 | name */ |
7296 | const char* session_stopword_table, /*!< in: Session stopword table |
7297 | name */ |
7298 | ibool stopword_is_on, /*!< in: Whether stopword |
7299 | option is turned on/off */ |
7300 | ibool reload) /*!< in: Whether it is |
7301 | for reloading FTS table */ |
7302 | { |
7303 | fts_table_t fts_table; |
7304 | fts_string_t str; |
7305 | dberr_t error = DB_SUCCESS; |
7306 | ulint use_stopword; |
7307 | fts_cache_t* cache; |
7308 | const char* stopword_to_use = NULL; |
7309 | ibool new_trx = FALSE; |
7310 | byte str_buffer[MAX_FULL_NAME_LEN + 1]; |
7311 | |
7312 | FTS_INIT_FTS_TABLE(&fts_table, "CONFIG" , FTS_COMMON_TABLE, table); |
7313 | |
7314 | cache = table->fts->cache; |
7315 | |
7316 | if (!reload && !(cache->stopword_info.status |
7317 | & STOPWORD_NOT_INIT)) { |
7318 | return(TRUE); |
7319 | } |
7320 | |
7321 | if (!trx) { |
7322 | trx = trx_create(); |
7323 | if (srv_read_only_mode) { |
7324 | trx_start_internal_read_only(trx); |
7325 | } else { |
7326 | trx_start_internal(trx); |
7327 | } |
7328 | trx->op_info = "upload FTS stopword" ; |
7329 | new_trx = TRUE; |
7330 | } |
7331 | |
7332 | /* First check whether stopword filtering is turned off */ |
7333 | if (reload) { |
7334 | error = fts_config_get_ulint( |
7335 | trx, &fts_table, FTS_USE_STOPWORD, &use_stopword); |
7336 | } else { |
7337 | use_stopword = (ulint) stopword_is_on; |
7338 | |
7339 | error = fts_config_set_ulint( |
7340 | trx, &fts_table, FTS_USE_STOPWORD, use_stopword); |
7341 | } |
7342 | |
7343 | if (error != DB_SUCCESS) { |
7344 | goto cleanup; |
7345 | } |
7346 | |
7347 | /* If stopword is turned off, no need to continue to load the |
7348 | stopword into cache, but still need to do initialization */ |
7349 | if (!use_stopword) { |
7350 | cache->stopword_info.status = STOPWORD_OFF; |
7351 | goto cleanup; |
7352 | } |
7353 | |
7354 | if (reload) { |
7355 | /* Fetch the stopword table name from FTS config |
7356 | table */ |
7357 | str.f_n_char = 0; |
7358 | str.f_str = str_buffer; |
7359 | str.f_len = sizeof(str_buffer) - 1; |
7360 | |
7361 | error = fts_config_get_value( |
7362 | trx, &fts_table, FTS_STOPWORD_TABLE_NAME, &str); |
7363 | |
7364 | if (error != DB_SUCCESS) { |
7365 | goto cleanup; |
7366 | } |
7367 | |
7368 | if (strlen((char*) str.f_str) > 0) { |
7369 | stopword_to_use = (const char*) str.f_str; |
7370 | } |
7371 | } else { |
7372 | stopword_to_use = (session_stopword_table) |
7373 | ? session_stopword_table : global_stopword_table; |
7374 | } |
7375 | |
7376 | if (stopword_to_use |
7377 | && fts_load_user_stopword(table->fts, stopword_to_use, |
7378 | &cache->stopword_info)) { |
7379 | /* Save the stopword table name to the configure |
7380 | table */ |
7381 | if (!reload) { |
7382 | str.f_n_char = 0; |
7383 | str.f_str = (byte*) stopword_to_use; |
7384 | str.f_len = ut_strlen(stopword_to_use); |
7385 | |
7386 | error = fts_config_set_value( |
7387 | trx, &fts_table, FTS_STOPWORD_TABLE_NAME, &str); |
7388 | } |
7389 | } else { |
7390 | /* Load system default stopword list */ |
7391 | fts_load_default_stopword(&cache->stopword_info); |
7392 | } |
7393 | |
7394 | cleanup: |
7395 | if (new_trx) { |
7396 | if (error == DB_SUCCESS) { |
7397 | fts_sql_commit(trx); |
7398 | } else { |
7399 | fts_sql_rollback(trx); |
7400 | } |
7401 | |
7402 | trx_free(trx); |
7403 | } |
7404 | |
7405 | if (!cache->stopword_info.cached_stopword) { |
7406 | cache->stopword_info.cached_stopword = rbt_create_arg_cmp( |
7407 | sizeof(fts_tokenizer_word_t), innobase_fts_text_cmp, |
7408 | &my_charset_latin1); |
7409 | } |
7410 | |
7411 | return(error == DB_SUCCESS); |
7412 | } |
7413 | |
7414 | /**********************************************************************//** |
7415 | Callback function when we initialize the FTS at the start up |
7416 | time. It recovers the maximum Doc IDs presented in the current table. |
7417 | @return: always returns TRUE */ |
7418 | static |
7419 | ibool |
7420 | fts_init_get_doc_id( |
7421 | /*================*/ |
7422 | void* row, /*!< in: sel_node_t* */ |
7423 | void* user_arg) /*!< in: fts cache */ |
7424 | { |
7425 | doc_id_t doc_id = FTS_NULL_DOC_ID; |
7426 | sel_node_t* node = static_cast<sel_node_t*>(row); |
7427 | que_node_t* exp = node->select_list; |
7428 | fts_cache_t* cache = static_cast<fts_cache_t*>(user_arg); |
7429 | |
7430 | ut_ad(ib_vector_is_empty(cache->get_docs)); |
7431 | |
7432 | /* Copy each indexed column content into doc->text.f_str */ |
7433 | if (exp) { |
7434 | dfield_t* dfield = que_node_get_val(exp); |
7435 | dtype_t* type = dfield_get_type(dfield); |
7436 | void* data = dfield_get_data(dfield); |
7437 | |
7438 | ut_a(dtype_get_mtype(type) == DATA_INT); |
7439 | |
7440 | doc_id = static_cast<doc_id_t>(mach_read_from_8( |
7441 | static_cast<const byte*>(data))); |
7442 | |
7443 | if (doc_id >= cache->next_doc_id) { |
7444 | cache->next_doc_id = doc_id + 1; |
7445 | } |
7446 | } |
7447 | |
7448 | return(TRUE); |
7449 | } |
7450 | |
7451 | /**********************************************************************//** |
7452 | Callback function when we initialize the FTS at the start up |
7453 | time. It recovers Doc IDs that have not sync-ed to the auxiliary |
7454 | table, and require to bring them back into FTS index. |
7455 | @return: always returns TRUE */ |
7456 | static |
7457 | ibool |
7458 | fts_init_recover_doc( |
7459 | /*=================*/ |
7460 | void* row, /*!< in: sel_node_t* */ |
7461 | void* user_arg) /*!< in: fts cache */ |
7462 | { |
7463 | |
7464 | fts_doc_t doc; |
7465 | ulint doc_len = 0; |
7466 | ulint field_no = 0; |
7467 | fts_get_doc_t* get_doc = static_cast<fts_get_doc_t*>(user_arg); |
7468 | doc_id_t doc_id = FTS_NULL_DOC_ID; |
7469 | sel_node_t* node = static_cast<sel_node_t*>(row); |
7470 | que_node_t* exp = node->select_list; |
7471 | fts_cache_t* cache = get_doc->cache; |
7472 | st_mysql_ftparser* parser = get_doc->index_cache->index->parser; |
7473 | |
7474 | fts_doc_init(&doc); |
7475 | doc.found = TRUE; |
7476 | |
7477 | ut_ad(cache); |
7478 | |
7479 | /* Copy each indexed column content into doc->text.f_str */ |
7480 | while (exp) { |
7481 | dfield_t* dfield = que_node_get_val(exp); |
7482 | ulint len = dfield_get_len(dfield); |
7483 | |
7484 | if (field_no == 0) { |
7485 | dtype_t* type = dfield_get_type(dfield); |
7486 | void* data = dfield_get_data(dfield); |
7487 | |
7488 | ut_a(dtype_get_mtype(type) == DATA_INT); |
7489 | |
7490 | doc_id = static_cast<doc_id_t>(mach_read_from_8( |
7491 | static_cast<const byte*>(data))); |
7492 | |
7493 | field_no++; |
7494 | exp = que_node_get_next(exp); |
7495 | continue; |
7496 | } |
7497 | |
7498 | if (len == UNIV_SQL_NULL) { |
7499 | exp = que_node_get_next(exp); |
7500 | continue; |
7501 | } |
7502 | |
7503 | ut_ad(get_doc); |
7504 | |
7505 | if (!get_doc->index_cache->charset) { |
7506 | get_doc->index_cache->charset = fts_get_charset( |
7507 | dfield->type.prtype); |
7508 | } |
7509 | |
7510 | doc.charset = get_doc->index_cache->charset; |
7511 | |
7512 | if (dfield_is_ext(dfield)) { |
7513 | dict_table_t* table = cache->sync->table; |
7514 | |
7515 | doc.text.f_str = btr_copy_externally_stored_field( |
7516 | &doc.text.f_len, |
7517 | static_cast<byte*>(dfield_get_data(dfield)), |
7518 | dict_table_page_size(table), len, |
7519 | static_cast<mem_heap_t*>(doc.self_heap->arg)); |
7520 | } else { |
7521 | doc.text.f_str = static_cast<byte*>( |
7522 | dfield_get_data(dfield)); |
7523 | |
7524 | doc.text.f_len = len; |
7525 | } |
7526 | |
7527 | if (field_no == 1) { |
7528 | fts_tokenize_document(&doc, NULL, parser); |
7529 | } else { |
7530 | fts_tokenize_document_next(&doc, doc_len, NULL, parser); |
7531 | } |
7532 | |
7533 | exp = que_node_get_next(exp); |
7534 | |
7535 | doc_len += (exp) ? len + 1 : len; |
7536 | |
7537 | field_no++; |
7538 | } |
7539 | |
7540 | fts_cache_add_doc(cache, get_doc->index_cache, doc_id, doc.tokens); |
7541 | |
7542 | fts_doc_free(&doc); |
7543 | |
7544 | cache->added++; |
7545 | |
7546 | if (doc_id >= cache->next_doc_id) { |
7547 | cache->next_doc_id = doc_id + 1; |
7548 | } |
7549 | |
7550 | return(TRUE); |
7551 | } |
7552 | |
7553 | /**********************************************************************//** |
7554 | This function brings FTS index in sync when FTS index is first |
7555 | used. There are documents that have not yet sync-ed to auxiliary |
7556 | tables from last server abnormally shutdown, we will need to bring |
7557 | such document into FTS cache before any further operations |
7558 | @return TRUE if all OK */ |
7559 | ibool |
7560 | fts_init_index( |
7561 | /*===========*/ |
7562 | dict_table_t* table, /*!< in: Table with FTS */ |
7563 | ibool has_cache_lock) /*!< in: Whether we already have |
7564 | cache lock */ |
7565 | { |
7566 | dict_index_t* index; |
7567 | doc_id_t start_doc; |
7568 | fts_get_doc_t* get_doc = NULL; |
7569 | fts_cache_t* cache = table->fts->cache; |
7570 | bool need_init = false; |
7571 | |
7572 | ut_ad(!mutex_own(&dict_sys->mutex)); |
7573 | |
7574 | /* First check cache->get_docs is initialized */ |
7575 | if (!has_cache_lock) { |
7576 | rw_lock_x_lock(&cache->lock); |
7577 | } |
7578 | |
7579 | rw_lock_x_lock(&cache->init_lock); |
7580 | if (cache->get_docs == NULL) { |
7581 | cache->get_docs = fts_get_docs_create(cache); |
7582 | } |
7583 | rw_lock_x_unlock(&cache->init_lock); |
7584 | |
7585 | if (table->fts->fts_status & ADDED_TABLE_SYNCED) { |
7586 | goto func_exit; |
7587 | } |
7588 | |
7589 | need_init = true; |
7590 | |
7591 | start_doc = cache->synced_doc_id; |
7592 | |
7593 | if (!start_doc) { |
7594 | fts_cmp_set_sync_doc_id(table, 0, TRUE, &start_doc); |
7595 | cache->synced_doc_id = start_doc; |
7596 | } |
7597 | |
7598 | /* No FTS index, this is the case when previous FTS index |
7599 | dropped, and we re-initialize the Doc ID system for subsequent |
7600 | insertion */ |
7601 | if (ib_vector_is_empty(cache->get_docs)) { |
7602 | index = table->fts_doc_id_index; |
7603 | |
7604 | ut_a(index); |
7605 | |
7606 | fts_doc_fetch_by_doc_id(NULL, start_doc, index, |
7607 | FTS_FETCH_DOC_BY_ID_LARGE, |
7608 | fts_init_get_doc_id, cache); |
7609 | } else { |
7610 | if (table->fts->cache->stopword_info.status |
7611 | & STOPWORD_NOT_INIT) { |
7612 | fts_load_stopword(table, NULL, NULL, NULL, TRUE, TRUE); |
7613 | } |
7614 | |
7615 | for (ulint i = 0; i < ib_vector_size(cache->get_docs); ++i) { |
7616 | get_doc = static_cast<fts_get_doc_t*>( |
7617 | ib_vector_get(cache->get_docs, i)); |
7618 | |
7619 | index = get_doc->index_cache->index; |
7620 | |
7621 | fts_doc_fetch_by_doc_id(NULL, start_doc, index, |
7622 | FTS_FETCH_DOC_BY_ID_LARGE, |
7623 | fts_init_recover_doc, get_doc); |
7624 | } |
7625 | } |
7626 | |
7627 | table->fts->fts_status |= ADDED_TABLE_SYNCED; |
7628 | |
7629 | fts_get_docs_clear(cache->get_docs); |
7630 | |
7631 | func_exit: |
7632 | if (!has_cache_lock) { |
7633 | rw_lock_x_unlock(&cache->lock); |
7634 | } |
7635 | |
7636 | if (need_init) { |
7637 | mutex_enter(&dict_sys->mutex); |
7638 | /* Register the table with the optimize thread. */ |
7639 | fts_optimize_add_table(table); |
7640 | mutex_exit(&dict_sys->mutex); |
7641 | } |
7642 | |
7643 | return(TRUE); |
7644 | } |
7645 | |
7646 | /** Check if the all the auxillary tables associated with FTS index are in |
7647 | consistent state. For now consistency is check only by ensuring |
7648 | index->page_no != FIL_NULL |
7649 | @param[out] base_table table has host fts index |
7650 | @param[in,out] trx trx handler */ |
7651 | void |
7652 | fts_check_corrupt( |
7653 | dict_table_t* base_table, |
7654 | trx_t* trx) |
7655 | { |
7656 | bool sane = true; |
7657 | fts_table_t fts_table; |
7658 | |
7659 | /* Iterate over the common table and check for their sanity. */ |
7660 | FTS_INIT_FTS_TABLE(&fts_table, NULL, FTS_COMMON_TABLE, base_table); |
7661 | |
7662 | for (ulint i = 0; fts_common_tables[i] != NULL && sane; ++i) { |
7663 | |
7664 | char table_name[MAX_FULL_NAME_LEN]; |
7665 | |
7666 | fts_table.suffix = fts_common_tables[i]; |
7667 | fts_get_table_name(&fts_table, table_name); |
7668 | |
7669 | dict_table_t* aux_table = dict_table_open_on_name( |
7670 | table_name, true, FALSE, DICT_ERR_IGNORE_NONE); |
7671 | |
7672 | if (aux_table == NULL) { |
7673 | dict_set_corrupted( |
7674 | dict_table_get_first_index(base_table), |
7675 | trx, "FTS_SANITY_CHECK" ); |
7676 | ut_ad(base_table->corrupted == TRUE); |
7677 | sane = false; |
7678 | continue; |
7679 | } |
7680 | |
7681 | for (dict_index_t* aux_table_index = |
7682 | UT_LIST_GET_FIRST(aux_table->indexes); |
7683 | aux_table_index != NULL; |
7684 | aux_table_index = |
7685 | UT_LIST_GET_NEXT(indexes, aux_table_index)) { |
7686 | |
7687 | /* Check if auxillary table needed for FTS is sane. */ |
7688 | if (aux_table_index->page == FIL_NULL) { |
7689 | dict_set_corrupted( |
7690 | dict_table_get_first_index(base_table), |
7691 | trx, "FTS_SANITY_CHECK" ); |
7692 | ut_ad(base_table->corrupted == TRUE); |
7693 | sane = false; |
7694 | } |
7695 | } |
7696 | |
7697 | dict_table_close(aux_table, FALSE, FALSE); |
7698 | } |
7699 | } |
7700 | |