| 1 | /*------------------------------------------------------------------------- |
| 2 | * |
| 3 | * ts_selfuncs.c |
| 4 | * Selectivity estimation functions for text search operators. |
| 5 | * |
| 6 | * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group |
| 7 | * |
| 8 | * |
| 9 | * IDENTIFICATION |
| 10 | * src/backend/tsearch/ts_selfuncs.c |
| 11 | * |
| 12 | *------------------------------------------------------------------------- |
| 13 | */ |
| 14 | #include "postgres.h" |
| 15 | |
| 16 | #include "access/htup_details.h" |
| 17 | #include "catalog/pg_statistic.h" |
| 18 | #include "catalog/pg_type.h" |
| 19 | #include "miscadmin.h" |
| 20 | #include "nodes/nodes.h" |
| 21 | #include "tsearch/ts_type.h" |
| 22 | #include "utils/builtins.h" |
| 23 | #include "utils/lsyscache.h" |
| 24 | #include "utils/selfuncs.h" |
| 25 | #include "utils/syscache.h" |
| 26 | |
| 27 | |
| 28 | /* |
| 29 | * The default text search selectivity is chosen to be small enough to |
| 30 | * encourage indexscans for typical table densities. See selfuncs.h and |
| 31 | * DEFAULT_EQ_SEL for details. |
| 32 | */ |
| 33 | #define DEFAULT_TS_MATCH_SEL 0.005 |
| 34 | |
| 35 | /* lookup table type for binary searching through MCELEMs */ |
| 36 | typedef struct |
| 37 | { |
| 38 | text *element; |
| 39 | float4 frequency; |
| 40 | } TextFreq; |
| 41 | |
| 42 | /* type of keys for bsearch'ing through an array of TextFreqs */ |
| 43 | typedef struct |
| 44 | { |
| 45 | char *lexeme; |
| 46 | int length; |
| 47 | } LexemeKey; |
| 48 | |
| 49 | static Selectivity tsquerysel(VariableStatData *vardata, Datum constval); |
| 50 | static Selectivity mcelem_tsquery_selec(TSQuery query, |
| 51 | Datum *mcelem, int nmcelem, |
| 52 | float4 *numbers, int nnumbers); |
| 53 | static Selectivity tsquery_opr_selec(QueryItem *item, char *operand, |
| 54 | TextFreq *lookup, int length, float4 minfreq); |
| 55 | static int compare_lexeme_textfreq(const void *e1, const void *e2); |
| 56 | |
| 57 | #define tsquery_opr_selec_no_stats(query) \ |
| 58 | tsquery_opr_selec(GETQUERY(query), GETOPERAND(query), NULL, 0, 0) |
| 59 | |
| 60 | |
| 61 | /* |
| 62 | * tsmatchsel -- Selectivity of "@@" |
| 63 | * |
| 64 | * restriction selectivity function for tsvector @@ tsquery and |
| 65 | * tsquery @@ tsvector |
| 66 | */ |
| 67 | Datum |
| 68 | tsmatchsel(PG_FUNCTION_ARGS) |
| 69 | { |
| 70 | PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0); |
| 71 | |
| 72 | #ifdef NOT_USED |
| 73 | Oid operator = PG_GETARG_OID(1); |
| 74 | #endif |
| 75 | List *args = (List *) PG_GETARG_POINTER(2); |
| 76 | int varRelid = PG_GETARG_INT32(3); |
| 77 | VariableStatData vardata; |
| 78 | Node *other; |
| 79 | bool varonleft; |
| 80 | Selectivity selec; |
| 81 | |
| 82 | /* |
| 83 | * If expression is not variable = something or something = variable, then |
| 84 | * punt and return a default estimate. |
| 85 | */ |
| 86 | if (!get_restriction_variable(root, args, varRelid, |
| 87 | &vardata, &other, &varonleft)) |
| 88 | PG_RETURN_FLOAT8(DEFAULT_TS_MATCH_SEL); |
| 89 | |
| 90 | /* |
| 91 | * Can't do anything useful if the something is not a constant, either. |
| 92 | */ |
| 93 | if (!IsA(other, Const)) |
| 94 | { |
| 95 | ReleaseVariableStats(vardata); |
| 96 | PG_RETURN_FLOAT8(DEFAULT_TS_MATCH_SEL); |
| 97 | } |
| 98 | |
| 99 | /* |
| 100 | * The "@@" operator is strict, so we can cope with NULL right away |
| 101 | */ |
| 102 | if (((Const *) other)->constisnull) |
| 103 | { |
| 104 | ReleaseVariableStats(vardata); |
| 105 | PG_RETURN_FLOAT8(0.0); |
| 106 | } |
| 107 | |
| 108 | /* |
| 109 | * OK, there's a Var and a Const we're dealing with here. We need the |
| 110 | * Const to be a TSQuery, else we can't do anything useful. We have to |
| 111 | * check this because the Var might be the TSQuery not the TSVector. |
| 112 | */ |
| 113 | if (((Const *) other)->consttype == TSQUERYOID) |
| 114 | { |
| 115 | /* tsvector @@ tsquery or the other way around */ |
| 116 | Assert(vardata.vartype == TSVECTOROID); |
| 117 | |
| 118 | selec = tsquerysel(&vardata, ((Const *) other)->constvalue); |
| 119 | } |
| 120 | else |
| 121 | { |
| 122 | /* If we can't see the query structure, must punt */ |
| 123 | selec = DEFAULT_TS_MATCH_SEL; |
| 124 | } |
| 125 | |
| 126 | ReleaseVariableStats(vardata); |
| 127 | |
| 128 | CLAMP_PROBABILITY(selec); |
| 129 | |
| 130 | PG_RETURN_FLOAT8((float8) selec); |
| 131 | } |
| 132 | |
| 133 | |
| 134 | /* |
| 135 | * tsmatchjoinsel -- join selectivity of "@@" |
| 136 | * |
| 137 | * join selectivity function for tsvector @@ tsquery and tsquery @@ tsvector |
| 138 | */ |
| 139 | Datum |
| 140 | tsmatchjoinsel(PG_FUNCTION_ARGS) |
| 141 | { |
| 142 | /* for the moment we just punt */ |
| 143 | PG_RETURN_FLOAT8(DEFAULT_TS_MATCH_SEL); |
| 144 | } |
| 145 | |
| 146 | |
| 147 | /* |
| 148 | * @@ selectivity for tsvector var vs tsquery constant |
| 149 | */ |
| 150 | static Selectivity |
| 151 | tsquerysel(VariableStatData *vardata, Datum constval) |
| 152 | { |
| 153 | Selectivity selec; |
| 154 | TSQuery query; |
| 155 | |
| 156 | /* The caller made sure the const is a TSQuery, so get it now */ |
| 157 | query = DatumGetTSQuery(constval); |
| 158 | |
| 159 | /* Empty query matches nothing */ |
| 160 | if (query->size == 0) |
| 161 | return (Selectivity) 0.0; |
| 162 | |
| 163 | if (HeapTupleIsValid(vardata->statsTuple)) |
| 164 | { |
| 165 | Form_pg_statistic stats; |
| 166 | AttStatsSlot sslot; |
| 167 | |
| 168 | stats = (Form_pg_statistic) GETSTRUCT(vardata->statsTuple); |
| 169 | |
| 170 | /* MCELEM will be an array of TEXT elements for a tsvector column */ |
| 171 | if (get_attstatsslot(&sslot, vardata->statsTuple, |
| 172 | STATISTIC_KIND_MCELEM, InvalidOid, |
| 173 | ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS)) |
| 174 | { |
| 175 | /* |
| 176 | * There is a most-common-elements slot for the tsvector Var, so |
| 177 | * use that. |
| 178 | */ |
| 179 | selec = mcelem_tsquery_selec(query, sslot.values, sslot.nvalues, |
| 180 | sslot.numbers, sslot.nnumbers); |
| 181 | free_attstatsslot(&sslot); |
| 182 | } |
| 183 | else |
| 184 | { |
| 185 | /* No most-common-elements info, so do without */ |
| 186 | selec = tsquery_opr_selec_no_stats(query); |
| 187 | } |
| 188 | |
| 189 | /* |
| 190 | * MCE stats count only non-null rows, so adjust for null rows. |
| 191 | */ |
| 192 | selec *= (1.0 - stats->stanullfrac); |
| 193 | } |
| 194 | else |
| 195 | { |
| 196 | /* No stats at all, so do without */ |
| 197 | selec = tsquery_opr_selec_no_stats(query); |
| 198 | /* we assume no nulls here, so no stanullfrac correction */ |
| 199 | } |
| 200 | |
| 201 | return selec; |
| 202 | } |
| 203 | |
| 204 | /* |
| 205 | * Extract data from the pg_statistic arrays into useful format. |
| 206 | */ |
| 207 | static Selectivity |
| 208 | mcelem_tsquery_selec(TSQuery query, Datum *mcelem, int nmcelem, |
| 209 | float4 *numbers, int nnumbers) |
| 210 | { |
| 211 | float4 minfreq; |
| 212 | TextFreq *lookup; |
| 213 | Selectivity selec; |
| 214 | int i; |
| 215 | |
| 216 | /* |
| 217 | * There should be two more Numbers than Values, because the last two |
| 218 | * cells are taken for minimal and maximal frequency. Punt if not. |
| 219 | * |
| 220 | * (Note: the MCELEM statistics slot definition allows for a third extra |
| 221 | * number containing the frequency of nulls, but we're not expecting that |
| 222 | * to appear for a tsvector column.) |
| 223 | */ |
| 224 | if (nnumbers != nmcelem + 2) |
| 225 | return tsquery_opr_selec_no_stats(query); |
| 226 | |
| 227 | /* |
| 228 | * Transpose the data into a single array so we can use bsearch(). |
| 229 | */ |
| 230 | lookup = (TextFreq *) palloc(sizeof(TextFreq) * nmcelem); |
| 231 | for (i = 0; i < nmcelem; i++) |
| 232 | { |
| 233 | /* |
| 234 | * The text Datums came from an array, so it cannot be compressed or |
| 235 | * stored out-of-line -- it's safe to use VARSIZE_ANY*. |
| 236 | */ |
| 237 | Assert(!VARATT_IS_COMPRESSED(mcelem[i]) && !VARATT_IS_EXTERNAL(mcelem[i])); |
| 238 | lookup[i].element = (text *) DatumGetPointer(mcelem[i]); |
| 239 | lookup[i].frequency = numbers[i]; |
| 240 | } |
| 241 | |
| 242 | /* |
| 243 | * Grab the lowest frequency. compute_tsvector_stats() stored it for us in |
| 244 | * the one before the last cell of the Numbers array. See ts_typanalyze.c |
| 245 | */ |
| 246 | minfreq = numbers[nnumbers - 2]; |
| 247 | |
| 248 | selec = tsquery_opr_selec(GETQUERY(query), GETOPERAND(query), lookup, |
| 249 | nmcelem, minfreq); |
| 250 | |
| 251 | pfree(lookup); |
| 252 | |
| 253 | return selec; |
| 254 | } |
| 255 | |
| 256 | /* |
| 257 | * Traverse the tsquery in preorder, calculating selectivity as: |
| 258 | * |
| 259 | * selec(left_oper) * selec(right_oper) in AND & PHRASE nodes, |
| 260 | * |
| 261 | * selec(left_oper) + selec(right_oper) - |
| 262 | * selec(left_oper) * selec(right_oper) in OR nodes, |
| 263 | * |
| 264 | * 1 - select(oper) in NOT nodes |
| 265 | * |
| 266 | * histogram-based estimation in prefix VAL nodes |
| 267 | * |
| 268 | * freq[val] in exact VAL nodes, if the value is in MCELEM |
| 269 | * min(freq[MCELEM]) / 2 in VAL nodes, if it is not |
| 270 | * |
| 271 | * The MCELEM array is already sorted (see ts_typanalyze.c), so we can use |
| 272 | * binary search for determining freq[MCELEM]. |
| 273 | * |
| 274 | * If we don't have stats for the tsvector, we still use this logic, |
| 275 | * except we use default estimates for VAL nodes. This case is signaled |
| 276 | * by lookup == NULL. |
| 277 | */ |
| 278 | static Selectivity |
| 279 | tsquery_opr_selec(QueryItem *item, char *operand, |
| 280 | TextFreq *lookup, int length, float4 minfreq) |
| 281 | { |
| 282 | Selectivity selec; |
| 283 | |
| 284 | /* since this function recurses, it could be driven to stack overflow */ |
| 285 | check_stack_depth(); |
| 286 | |
| 287 | if (item->type == QI_VAL) |
| 288 | { |
| 289 | QueryOperand *oper = (QueryOperand *) item; |
| 290 | LexemeKey key; |
| 291 | |
| 292 | /* |
| 293 | * Prepare the key for bsearch(). |
| 294 | */ |
| 295 | key.lexeme = operand + oper->distance; |
| 296 | key.length = oper->length; |
| 297 | |
| 298 | if (oper->prefix) |
| 299 | { |
| 300 | /* Prefix match, ie the query item is lexeme:* */ |
| 301 | Selectivity matched, |
| 302 | allmces; |
| 303 | int i, |
| 304 | n_matched; |
| 305 | |
| 306 | /* |
| 307 | * Our strategy is to scan through the MCELEM list and combine the |
| 308 | * frequencies of the ones that match the prefix. We then |
| 309 | * extrapolate the fraction of matching MCELEMs to the remaining |
| 310 | * rows, assuming that the MCELEMs are representative of the whole |
| 311 | * lexeme population in this respect. (Compare |
| 312 | * histogram_selectivity().) Note that these are most common |
| 313 | * elements not most common values, so they're not mutually |
| 314 | * exclusive. We treat occurrences as independent events. |
| 315 | * |
| 316 | * This is only a good plan if we have a pretty fair number of |
| 317 | * MCELEMs available; we set the threshold at 100. If no stats or |
| 318 | * insufficient stats, arbitrarily use DEFAULT_TS_MATCH_SEL*4. |
| 319 | */ |
| 320 | if (lookup == NULL || length < 100) |
| 321 | return (Selectivity) (DEFAULT_TS_MATCH_SEL * 4); |
| 322 | |
| 323 | matched = allmces = 0; |
| 324 | n_matched = 0; |
| 325 | for (i = 0; i < length; i++) |
| 326 | { |
| 327 | TextFreq *t = lookup + i; |
| 328 | int tlen = VARSIZE_ANY_EXHDR(t->element); |
| 329 | |
| 330 | if (tlen >= key.length && |
| 331 | strncmp(key.lexeme, VARDATA_ANY(t->element), |
| 332 | key.length) == 0) |
| 333 | { |
| 334 | matched += t->frequency - matched * t->frequency; |
| 335 | n_matched++; |
| 336 | } |
| 337 | allmces += t->frequency - allmces * t->frequency; |
| 338 | } |
| 339 | |
| 340 | /* Clamp to ensure sanity in the face of roundoff error */ |
| 341 | CLAMP_PROBABILITY(matched); |
| 342 | CLAMP_PROBABILITY(allmces); |
| 343 | |
| 344 | selec = matched + (1.0 - allmces) * ((double) n_matched / length); |
| 345 | |
| 346 | /* |
| 347 | * In any case, never believe that a prefix match has selectivity |
| 348 | * less than we would assign for a non-MCELEM lexeme. This |
| 349 | * preserves the property that "word:*" should be estimated to |
| 350 | * match at least as many rows as "word" would be. |
| 351 | */ |
| 352 | selec = Max(Min(DEFAULT_TS_MATCH_SEL, minfreq / 2), selec); |
| 353 | } |
| 354 | else |
| 355 | { |
| 356 | /* Regular exact lexeme match */ |
| 357 | TextFreq *searchres; |
| 358 | |
| 359 | /* If no stats for the variable, use DEFAULT_TS_MATCH_SEL */ |
| 360 | if (lookup == NULL) |
| 361 | return (Selectivity) DEFAULT_TS_MATCH_SEL; |
| 362 | |
| 363 | searchres = (TextFreq *) bsearch(&key, lookup, length, |
| 364 | sizeof(TextFreq), |
| 365 | compare_lexeme_textfreq); |
| 366 | |
| 367 | if (searchres) |
| 368 | { |
| 369 | /* |
| 370 | * The element is in MCELEM. Return precise selectivity (or |
| 371 | * at least as precise as ANALYZE could find out). |
| 372 | */ |
| 373 | selec = searchres->frequency; |
| 374 | } |
| 375 | else |
| 376 | { |
| 377 | /* |
| 378 | * The element is not in MCELEM. Punt, but assume that the |
| 379 | * selectivity cannot be more than minfreq / 2. |
| 380 | */ |
| 381 | selec = Min(DEFAULT_TS_MATCH_SEL, minfreq / 2); |
| 382 | } |
| 383 | } |
| 384 | } |
| 385 | else |
| 386 | { |
| 387 | /* Current TSQuery node is an operator */ |
| 388 | Selectivity s1, |
| 389 | s2; |
| 390 | |
| 391 | switch (item->qoperator.oper) |
| 392 | { |
| 393 | case OP_NOT: |
| 394 | selec = 1.0 - tsquery_opr_selec(item + 1, operand, |
| 395 | lookup, length, minfreq); |
| 396 | break; |
| 397 | |
| 398 | case OP_PHRASE: |
| 399 | case OP_AND: |
| 400 | s1 = tsquery_opr_selec(item + 1, operand, |
| 401 | lookup, length, minfreq); |
| 402 | s2 = tsquery_opr_selec(item + item->qoperator.left, operand, |
| 403 | lookup, length, minfreq); |
| 404 | selec = s1 * s2; |
| 405 | break; |
| 406 | |
| 407 | case OP_OR: |
| 408 | s1 = tsquery_opr_selec(item + 1, operand, |
| 409 | lookup, length, minfreq); |
| 410 | s2 = tsquery_opr_selec(item + item->qoperator.left, operand, |
| 411 | lookup, length, minfreq); |
| 412 | selec = s1 + s2 - s1 * s2; |
| 413 | break; |
| 414 | |
| 415 | default: |
| 416 | elog(ERROR, "unrecognized operator: %d" , item->qoperator.oper); |
| 417 | selec = 0; /* keep compiler quiet */ |
| 418 | break; |
| 419 | } |
| 420 | } |
| 421 | |
| 422 | /* Clamp intermediate results to stay sane despite roundoff error */ |
| 423 | CLAMP_PROBABILITY(selec); |
| 424 | |
| 425 | return selec; |
| 426 | } |
| 427 | |
| 428 | /* |
| 429 | * bsearch() comparator for a lexeme (non-NULL terminated string with length) |
| 430 | * and a TextFreq. Use length, then byte-for-byte comparison, because that's |
| 431 | * how ANALYZE code sorted data before storing it in a statistic tuple. |
| 432 | * See ts_typanalyze.c for details. |
| 433 | */ |
| 434 | static int |
| 435 | compare_lexeme_textfreq(const void *e1, const void *e2) |
| 436 | { |
| 437 | const LexemeKey *key = (const LexemeKey *) e1; |
| 438 | const TextFreq *t = (const TextFreq *) e2; |
| 439 | int len1, |
| 440 | len2; |
| 441 | |
| 442 | len1 = key->length; |
| 443 | len2 = VARSIZE_ANY_EXHDR(t->element); |
| 444 | |
| 445 | /* Compare lengths first, possibly avoiding a strncmp call */ |
| 446 | if (len1 > len2) |
| 447 | return 1; |
| 448 | else if (len1 < len2) |
| 449 | return -1; |
| 450 | |
| 451 | /* Fall back on byte-for-byte comparison */ |
| 452 | return strncmp(key->lexeme, VARDATA_ANY(t->element), len1); |
| 453 | } |
| 454 | |