Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * network_selfuncs.c
4 : * Functions for selectivity estimation of inet/cidr operators
5 : *
6 : * This module provides estimators for the subnet inclusion and overlap
7 : * operators. Estimates are based on null fraction, most common values,
8 : * and histogram of inet/cidr columns.
9 : *
10 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
11 : * Portions Copyright (c) 1994, Regents of the University of California
12 : *
13 : *
14 : * IDENTIFICATION
15 : * src/backend/utils/adt/network_selfuncs.c
16 : *
17 : *-------------------------------------------------------------------------
18 : */
19 : #include "postgres.h"
20 :
21 : #include <math.h>
22 :
23 : #include "access/htup_details.h"
24 : #include "catalog/pg_operator.h"
25 : #include "catalog/pg_statistic.h"
26 : #include "utils/fmgrprotos.h"
27 : #include "utils/inet.h"
28 : #include "utils/lsyscache.h"
29 : #include "utils/selfuncs.h"
30 :
31 :
32 : /* Default selectivity for the inet overlap operator */
33 : #define DEFAULT_OVERLAP_SEL 0.01
34 :
35 : /* Default selectivity for the various inclusion operators */
36 : #define DEFAULT_INCLUSION_SEL 0.005
37 :
38 : /* Default selectivity for specified operator */
39 : #define DEFAULT_SEL(operator) \
40 : ((operator) == OID_INET_OVERLAP_OP ? \
41 : DEFAULT_OVERLAP_SEL : DEFAULT_INCLUSION_SEL)
42 :
43 : /* Maximum number of items to consider in join selectivity calculations */
44 : #define MAX_CONSIDERED_ELEMS 1024
45 :
46 : static Selectivity networkjoinsel_inner(Oid operator,
47 : VariableStatData *vardata1, VariableStatData *vardata2);
48 : static Selectivity networkjoinsel_semi(Oid operator,
49 : VariableStatData *vardata1, VariableStatData *vardata2);
50 : static Selectivity mcv_population(float4 *mcv_numbers, int mcv_nvalues);
51 : static Selectivity inet_hist_value_sel(Datum *values, int nvalues,
52 : Datum constvalue, int opr_codenum);
53 : static Selectivity inet_mcv_join_sel(Datum *mcv1_values,
54 : float4 *mcv1_numbers, int mcv1_nvalues, Datum *mcv2_values,
55 : float4 *mcv2_numbers, int mcv2_nvalues, Oid operator);
56 : static Selectivity inet_mcv_hist_sel(Datum *mcv_values, float4 *mcv_numbers,
57 : int mcv_nvalues, Datum *hist_values, int hist_nvalues,
58 : int opr_codenum);
59 : static Selectivity inet_hist_inclusion_join_sel(Datum *hist1_values,
60 : int hist1_nvalues,
61 : Datum *hist2_values, int hist2_nvalues,
62 : int opr_codenum);
63 : static Selectivity inet_semi_join_sel(Datum lhs_value,
64 : bool mcv_exists, Datum *mcv_values, int mcv_nvalues,
65 : bool hist_exists, Datum *hist_values, int hist_nvalues,
66 : double hist_weight,
67 : FmgrInfo *proc, int opr_codenum);
68 : static int inet_opr_codenum(Oid operator);
69 : static int inet_inclusion_cmp(inet *left, inet *right, int opr_codenum);
70 : static int inet_masklen_inclusion_cmp(inet *left, inet *right,
71 : int opr_codenum);
72 : static int inet_hist_match_divider(inet *boundary, inet *query,
73 : int opr_codenum);
74 :
75 : /*
76 : * Selectivity estimation for the subnet inclusion/overlap operators
77 : */
78 : Datum
79 900 : networksel(PG_FUNCTION_ARGS)
80 : {
81 900 : PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
82 900 : Oid operator = PG_GETARG_OID(1);
83 900 : List *args = (List *) PG_GETARG_POINTER(2);
84 900 : int varRelid = PG_GETARG_INT32(3);
85 : VariableStatData vardata;
86 : Node *other;
87 : bool varonleft;
88 : Selectivity selec,
89 : mcv_selec,
90 : non_mcv_selec;
91 : Datum constvalue;
92 : Form_pg_statistic stats;
93 : AttStatsSlot hslot;
94 : double sumcommon,
95 : nullfrac;
96 : FmgrInfo proc;
97 :
98 : /*
99 : * If expression is not (variable op something) or (something op
100 : * variable), then punt and return a default estimate.
101 : */
102 900 : if (!get_restriction_variable(root, args, varRelid,
103 : &vardata, &other, &varonleft))
104 0 : PG_RETURN_FLOAT8(DEFAULT_SEL(operator));
105 :
106 : /*
107 : * Can't do anything useful if the something is not a constant, either.
108 : */
109 900 : if (!IsA(other, Const))
110 : {
111 0 : ReleaseVariableStats(vardata);
112 0 : PG_RETURN_FLOAT8(DEFAULT_SEL(operator));
113 : }
114 :
115 : /* All of the operators handled here are strict. */
116 900 : if (((Const *) other)->constisnull)
117 : {
118 0 : ReleaseVariableStats(vardata);
119 0 : PG_RETURN_FLOAT8(0.0);
120 : }
121 900 : constvalue = ((Const *) other)->constvalue;
122 :
123 : /* Otherwise, we need stats in order to produce a non-default estimate. */
124 900 : if (!HeapTupleIsValid(vardata.statsTuple))
125 : {
126 900 : ReleaseVariableStats(vardata);
127 900 : PG_RETURN_FLOAT8(DEFAULT_SEL(operator));
128 : }
129 :
130 0 : stats = (Form_pg_statistic) GETSTRUCT(vardata.statsTuple);
131 0 : nullfrac = stats->stanullfrac;
132 :
133 : /*
134 : * If we have most-common-values info, add up the fractions of the MCV
135 : * entries that satisfy MCV OP CONST. These fractions contribute directly
136 : * to the result selectivity. Also add up the total fraction represented
137 : * by MCV entries.
138 : */
139 0 : fmgr_info(get_opcode(operator), &proc);
140 0 : mcv_selec = mcv_selectivity(&vardata, &proc, InvalidOid,
141 : constvalue, varonleft,
142 : &sumcommon);
143 :
144 : /*
145 : * If we have a histogram, use it to estimate the proportion of the
146 : * non-MCV population that satisfies the clause. If we don't, apply the
147 : * default selectivity to that population.
148 : */
149 0 : if (get_attstatsslot(&hslot, vardata.statsTuple,
150 : STATISTIC_KIND_HISTOGRAM, InvalidOid,
151 : ATTSTATSSLOT_VALUES))
152 : {
153 0 : int opr_codenum = inet_opr_codenum(operator);
154 :
155 : /* Commute if needed, so we can consider histogram to be on the left */
156 0 : if (!varonleft)
157 0 : opr_codenum = -opr_codenum;
158 0 : non_mcv_selec = inet_hist_value_sel(hslot.values, hslot.nvalues,
159 : constvalue, opr_codenum);
160 :
161 0 : free_attstatsslot(&hslot);
162 : }
163 : else
164 0 : non_mcv_selec = DEFAULT_SEL(operator);
165 :
166 : /* Combine selectivities for MCV and non-MCV populations */
167 0 : selec = mcv_selec + (1.0 - nullfrac - sumcommon) * non_mcv_selec;
168 :
169 : /* Result should be in range, but make sure... */
170 0 : CLAMP_PROBABILITY(selec);
171 :
172 0 : ReleaseVariableStats(vardata);
173 :
174 0 : PG_RETURN_FLOAT8(selec);
175 : }
176 :
177 : /*
178 : * Join selectivity estimation for the subnet inclusion/overlap operators
179 : *
180 : * This function has the same structure as eqjoinsel() in selfuncs.c.
181 : *
182 : * Throughout networkjoinsel and its subroutines, we have a performance issue
183 : * in that the amount of work to be done is O(N^2) in the length of the MCV
184 : * and histogram arrays. To keep the runtime from getting out of hand when
185 : * large statistics targets have been set, we arbitrarily limit the number of
186 : * values considered to 1024 (MAX_CONSIDERED_ELEMS). For the MCV arrays, this
187 : * is easy: just consider at most the first N elements. (Since the MCVs are
188 : * sorted by decreasing frequency, this correctly gets us the first N MCVs.)
189 : * For the histogram arrays, we decimate; that is consider only every k'th
190 : * element, where k is chosen so that no more than MAX_CONSIDERED_ELEMS
191 : * elements are considered. This should still give us a good random sample of
192 : * the non-MCV population. Decimation is done on-the-fly in the loops that
193 : * iterate over the histogram arrays.
194 : */
195 : Datum
196 0 : networkjoinsel(PG_FUNCTION_ARGS)
197 : {
198 0 : PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
199 0 : Oid operator = PG_GETARG_OID(1);
200 0 : List *args = (List *) PG_GETARG_POINTER(2);
201 : #ifdef NOT_USED
202 : JoinType jointype = (JoinType) PG_GETARG_INT16(3);
203 : #endif
204 0 : SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) PG_GETARG_POINTER(4);
205 : double selec;
206 : VariableStatData vardata1;
207 : VariableStatData vardata2;
208 : bool join_is_reversed;
209 :
210 0 : get_join_variables(root, args, sjinfo,
211 : &vardata1, &vardata2, &join_is_reversed);
212 :
213 0 : switch (sjinfo->jointype)
214 : {
215 0 : case JOIN_INNER:
216 : case JOIN_LEFT:
217 : case JOIN_FULL:
218 :
219 : /*
220 : * Selectivity for left/full join is not exactly the same as inner
221 : * join, but we neglect the difference, as eqjoinsel does.
222 : */
223 0 : selec = networkjoinsel_inner(operator, &vardata1, &vardata2);
224 0 : break;
225 0 : case JOIN_SEMI:
226 : case JOIN_ANTI:
227 : /* Here, it's important that we pass the outer var on the left. */
228 0 : if (!join_is_reversed)
229 0 : selec = networkjoinsel_semi(operator, &vardata1, &vardata2);
230 : else
231 0 : selec = networkjoinsel_semi(get_commutator(operator),
232 : &vardata2, &vardata1);
233 0 : break;
234 0 : default:
235 : /* other values not expected here */
236 0 : elog(ERROR, "unrecognized join type: %d",
237 : (int) sjinfo->jointype);
238 : selec = 0; /* keep compiler quiet */
239 : break;
240 : }
241 :
242 0 : ReleaseVariableStats(vardata1);
243 0 : ReleaseVariableStats(vardata2);
244 :
245 0 : CLAMP_PROBABILITY(selec);
246 :
247 0 : PG_RETURN_FLOAT8((float8) selec);
248 : }
249 :
250 : /*
251 : * Inner join selectivity estimation for subnet inclusion/overlap operators
252 : *
253 : * Calculates MCV vs MCV, MCV vs histogram and histogram vs histogram
254 : * selectivity for join using the subnet inclusion operators. Unlike the
255 : * join selectivity function for the equality operator, eqjoinsel_inner(),
256 : * one to one matching of the values is not enough. Network inclusion
257 : * operators are likely to match many to many, so we must check all pairs.
258 : * (Note: it might be possible to exploit understanding of the histogram's
259 : * btree ordering to reduce the work needed, but we don't currently try.)
260 : * Also, MCV vs histogram selectivity is not neglected as in eqjoinsel_inner().
261 : */
262 : static Selectivity
263 0 : networkjoinsel_inner(Oid operator,
264 : VariableStatData *vardata1, VariableStatData *vardata2)
265 : {
266 : Form_pg_statistic stats;
267 0 : double nullfrac1 = 0.0,
268 0 : nullfrac2 = 0.0;
269 0 : Selectivity selec = 0.0,
270 0 : sumcommon1 = 0.0,
271 0 : sumcommon2 = 0.0;
272 0 : bool mcv1_exists = false,
273 0 : mcv2_exists = false,
274 0 : hist1_exists = false,
275 0 : hist2_exists = false;
276 : int opr_codenum;
277 0 : int mcv1_length = 0,
278 0 : mcv2_length = 0;
279 : AttStatsSlot mcv1_slot;
280 : AttStatsSlot mcv2_slot;
281 : AttStatsSlot hist1_slot;
282 : AttStatsSlot hist2_slot;
283 :
284 0 : if (HeapTupleIsValid(vardata1->statsTuple))
285 : {
286 0 : stats = (Form_pg_statistic) GETSTRUCT(vardata1->statsTuple);
287 0 : nullfrac1 = stats->stanullfrac;
288 :
289 0 : mcv1_exists = get_attstatsslot(&mcv1_slot, vardata1->statsTuple,
290 : STATISTIC_KIND_MCV, InvalidOid,
291 : ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS);
292 0 : hist1_exists = get_attstatsslot(&hist1_slot, vardata1->statsTuple,
293 : STATISTIC_KIND_HISTOGRAM, InvalidOid,
294 : ATTSTATSSLOT_VALUES);
295 : /* Arbitrarily limit number of MCVs considered */
296 0 : mcv1_length = Min(mcv1_slot.nvalues, MAX_CONSIDERED_ELEMS);
297 0 : if (mcv1_exists)
298 0 : sumcommon1 = mcv_population(mcv1_slot.numbers, mcv1_length);
299 : }
300 : else
301 : {
302 0 : memset(&mcv1_slot, 0, sizeof(mcv1_slot));
303 0 : memset(&hist1_slot, 0, sizeof(hist1_slot));
304 : }
305 :
306 0 : if (HeapTupleIsValid(vardata2->statsTuple))
307 : {
308 0 : stats = (Form_pg_statistic) GETSTRUCT(vardata2->statsTuple);
309 0 : nullfrac2 = stats->stanullfrac;
310 :
311 0 : mcv2_exists = get_attstatsslot(&mcv2_slot, vardata2->statsTuple,
312 : STATISTIC_KIND_MCV, InvalidOid,
313 : ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS);
314 0 : hist2_exists = get_attstatsslot(&hist2_slot, vardata2->statsTuple,
315 : STATISTIC_KIND_HISTOGRAM, InvalidOid,
316 : ATTSTATSSLOT_VALUES);
317 : /* Arbitrarily limit number of MCVs considered */
318 0 : mcv2_length = Min(mcv2_slot.nvalues, MAX_CONSIDERED_ELEMS);
319 0 : if (mcv2_exists)
320 0 : sumcommon2 = mcv_population(mcv2_slot.numbers, mcv2_length);
321 : }
322 : else
323 : {
324 0 : memset(&mcv2_slot, 0, sizeof(mcv2_slot));
325 0 : memset(&hist2_slot, 0, sizeof(hist2_slot));
326 : }
327 :
328 0 : opr_codenum = inet_opr_codenum(operator);
329 :
330 : /*
331 : * Calculate selectivity for MCV vs MCV matches.
332 : */
333 0 : if (mcv1_exists && mcv2_exists)
334 0 : selec += inet_mcv_join_sel(mcv1_slot.values, mcv1_slot.numbers,
335 : mcv1_length,
336 : mcv2_slot.values, mcv2_slot.numbers,
337 : mcv2_length,
338 : operator);
339 :
340 : /*
341 : * Add in selectivities for MCV vs histogram matches, scaling according to
342 : * the fractions of the populations represented by the histograms. Note
343 : * that the second case needs to commute the operator.
344 : */
345 0 : if (mcv1_exists && hist2_exists)
346 0 : selec += (1.0 - nullfrac2 - sumcommon2) *
347 0 : inet_mcv_hist_sel(mcv1_slot.values, mcv1_slot.numbers, mcv1_length,
348 : hist2_slot.values, hist2_slot.nvalues,
349 : opr_codenum);
350 0 : if (mcv2_exists && hist1_exists)
351 0 : selec += (1.0 - nullfrac1 - sumcommon1) *
352 0 : inet_mcv_hist_sel(mcv2_slot.values, mcv2_slot.numbers, mcv2_length,
353 : hist1_slot.values, hist1_slot.nvalues,
354 : -opr_codenum);
355 :
356 : /*
357 : * Add in selectivity for histogram vs histogram matches, again scaling
358 : * appropriately.
359 : */
360 0 : if (hist1_exists && hist2_exists)
361 0 : selec += (1.0 - nullfrac1 - sumcommon1) *
362 0 : (1.0 - nullfrac2 - sumcommon2) *
363 0 : inet_hist_inclusion_join_sel(hist1_slot.values, hist1_slot.nvalues,
364 : hist2_slot.values, hist2_slot.nvalues,
365 : opr_codenum);
366 :
367 : /*
368 : * If useful statistics are not available then use the default estimate.
369 : * We can apply null fractions if known, though.
370 : */
371 0 : if ((!mcv1_exists && !hist1_exists) || (!mcv2_exists && !hist2_exists))
372 0 : selec = (1.0 - nullfrac1) * (1.0 - nullfrac2) * DEFAULT_SEL(operator);
373 :
374 : /* Release stats. */
375 0 : free_attstatsslot(&mcv1_slot);
376 0 : free_attstatsslot(&mcv2_slot);
377 0 : free_attstatsslot(&hist1_slot);
378 0 : free_attstatsslot(&hist2_slot);
379 :
380 0 : return selec;
381 : }
382 :
383 : /*
384 : * Semi join selectivity estimation for subnet inclusion/overlap operators
385 : *
386 : * Calculates MCV vs MCV, MCV vs histogram, histogram vs MCV, and histogram vs
387 : * histogram selectivity for semi/anti join cases.
388 : */
389 : static Selectivity
390 0 : networkjoinsel_semi(Oid operator,
391 : VariableStatData *vardata1, VariableStatData *vardata2)
392 : {
393 : Form_pg_statistic stats;
394 0 : Selectivity selec = 0.0,
395 0 : sumcommon1 = 0.0,
396 0 : sumcommon2 = 0.0;
397 0 : double nullfrac1 = 0.0,
398 0 : nullfrac2 = 0.0,
399 0 : hist2_weight = 0.0;
400 0 : bool mcv1_exists = false,
401 0 : mcv2_exists = false,
402 0 : hist1_exists = false,
403 0 : hist2_exists = false;
404 : int opr_codenum;
405 : FmgrInfo proc;
406 : int i,
407 0 : mcv1_length = 0,
408 0 : mcv2_length = 0;
409 : AttStatsSlot mcv1_slot;
410 : AttStatsSlot mcv2_slot;
411 : AttStatsSlot hist1_slot;
412 : AttStatsSlot hist2_slot;
413 :
414 0 : if (HeapTupleIsValid(vardata1->statsTuple))
415 : {
416 0 : stats = (Form_pg_statistic) GETSTRUCT(vardata1->statsTuple);
417 0 : nullfrac1 = stats->stanullfrac;
418 :
419 0 : mcv1_exists = get_attstatsslot(&mcv1_slot, vardata1->statsTuple,
420 : STATISTIC_KIND_MCV, InvalidOid,
421 : ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS);
422 0 : hist1_exists = get_attstatsslot(&hist1_slot, vardata1->statsTuple,
423 : STATISTIC_KIND_HISTOGRAM, InvalidOid,
424 : ATTSTATSSLOT_VALUES);
425 : /* Arbitrarily limit number of MCVs considered */
426 0 : mcv1_length = Min(mcv1_slot.nvalues, MAX_CONSIDERED_ELEMS);
427 0 : if (mcv1_exists)
428 0 : sumcommon1 = mcv_population(mcv1_slot.numbers, mcv1_length);
429 : }
430 : else
431 : {
432 0 : memset(&mcv1_slot, 0, sizeof(mcv1_slot));
433 0 : memset(&hist1_slot, 0, sizeof(hist1_slot));
434 : }
435 :
436 0 : if (HeapTupleIsValid(vardata2->statsTuple))
437 : {
438 0 : stats = (Form_pg_statistic) GETSTRUCT(vardata2->statsTuple);
439 0 : nullfrac2 = stats->stanullfrac;
440 :
441 0 : mcv2_exists = get_attstatsslot(&mcv2_slot, vardata2->statsTuple,
442 : STATISTIC_KIND_MCV, InvalidOid,
443 : ATTSTATSSLOT_VALUES | ATTSTATSSLOT_NUMBERS);
444 0 : hist2_exists = get_attstatsslot(&hist2_slot, vardata2->statsTuple,
445 : STATISTIC_KIND_HISTOGRAM, InvalidOid,
446 : ATTSTATSSLOT_VALUES);
447 : /* Arbitrarily limit number of MCVs considered */
448 0 : mcv2_length = Min(mcv2_slot.nvalues, MAX_CONSIDERED_ELEMS);
449 0 : if (mcv2_exists)
450 0 : sumcommon2 = mcv_population(mcv2_slot.numbers, mcv2_length);
451 : }
452 : else
453 : {
454 0 : memset(&mcv2_slot, 0, sizeof(mcv2_slot));
455 0 : memset(&hist2_slot, 0, sizeof(hist2_slot));
456 : }
457 :
458 0 : opr_codenum = inet_opr_codenum(operator);
459 0 : fmgr_info(get_opcode(operator), &proc);
460 :
461 : /* Estimate number of input rows represented by RHS histogram. */
462 0 : if (hist2_exists && vardata2->rel)
463 0 : hist2_weight = (1.0 - nullfrac2 - sumcommon2) * vardata2->rel->rows;
464 :
465 : /*
466 : * Consider each element of the LHS MCV list, matching it to whatever RHS
467 : * stats we have. Scale according to the known frequency of the MCV.
468 : */
469 0 : if (mcv1_exists && (mcv2_exists || hist2_exists))
470 : {
471 0 : for (i = 0; i < mcv1_length; i++)
472 : {
473 0 : selec += mcv1_slot.numbers[i] *
474 0 : inet_semi_join_sel(mcv1_slot.values[i],
475 : mcv2_exists, mcv2_slot.values, mcv2_length,
476 : hist2_exists,
477 : hist2_slot.values, hist2_slot.nvalues,
478 : hist2_weight,
479 : &proc, opr_codenum);
480 : }
481 : }
482 :
483 : /*
484 : * Consider each element of the LHS histogram, except for the first and
485 : * last elements, which we exclude on the grounds that they're outliers
486 : * and thus not very representative. Scale on the assumption that each
487 : * such histogram element represents an equal share of the LHS histogram
488 : * population (which is a bit bogus, because the members of its bucket may
489 : * not all act the same with respect to the join clause, but it's hard to
490 : * do better).
491 : *
492 : * If there are too many histogram elements, decimate to limit runtime.
493 : */
494 0 : if (hist1_exists && hist1_slot.nvalues > 2 && (mcv2_exists || hist2_exists))
495 : {
496 0 : double hist_selec_sum = 0.0;
497 : int k,
498 : n;
499 :
500 0 : k = (hist1_slot.nvalues - 3) / MAX_CONSIDERED_ELEMS + 1;
501 :
502 0 : n = 0;
503 0 : for (i = 1; i < hist1_slot.nvalues - 1; i += k)
504 : {
505 0 : hist_selec_sum +=
506 0 : inet_semi_join_sel(hist1_slot.values[i],
507 : mcv2_exists, mcv2_slot.values, mcv2_length,
508 : hist2_exists,
509 : hist2_slot.values, hist2_slot.nvalues,
510 : hist2_weight,
511 : &proc, opr_codenum);
512 0 : n++;
513 : }
514 :
515 0 : selec += (1.0 - nullfrac1 - sumcommon1) * hist_selec_sum / n;
516 : }
517 :
518 : /*
519 : * If useful statistics are not available then use the default estimate.
520 : * We can apply null fractions if known, though.
521 : */
522 0 : if ((!mcv1_exists && !hist1_exists) || (!mcv2_exists && !hist2_exists))
523 0 : selec = (1.0 - nullfrac1) * (1.0 - nullfrac2) * DEFAULT_SEL(operator);
524 :
525 : /* Release stats. */
526 0 : free_attstatsslot(&mcv1_slot);
527 0 : free_attstatsslot(&mcv2_slot);
528 0 : free_attstatsslot(&hist1_slot);
529 0 : free_attstatsslot(&hist2_slot);
530 :
531 0 : return selec;
532 : }
533 :
534 : /*
535 : * Compute the fraction of a relation's population that is represented
536 : * by the MCV list.
537 : */
538 : static Selectivity
539 0 : mcv_population(float4 *mcv_numbers, int mcv_nvalues)
540 : {
541 0 : Selectivity sumcommon = 0.0;
542 : int i;
543 :
544 0 : for (i = 0; i < mcv_nvalues; i++)
545 : {
546 0 : sumcommon += mcv_numbers[i];
547 : }
548 :
549 0 : return sumcommon;
550 : }
551 :
552 : /*
553 : * Inet histogram vs single value selectivity estimation
554 : *
555 : * Estimate the fraction of the histogram population that satisfies
556 : * "value OPR CONST". (The result needs to be scaled to reflect the
557 : * proportion of the total population represented by the histogram.)
558 : *
559 : * The histogram is originally for the inet btree comparison operators.
560 : * Only the common bits of the network part and the length of the network part
561 : * (masklen) are interesting for the subnet inclusion operators. Fortunately,
562 : * btree comparison treats the network part as the major sort key. Even so,
563 : * the length of the network part would not really be significant in the
564 : * histogram. This would lead to big mistakes for data sets with uneven
565 : * masklen distribution. To reduce this problem, comparisons with the left
566 : * and the right sides of the buckets are used together.
567 : *
568 : * Histogram bucket matches are calculated in two forms. If the constant
569 : * matches both bucket endpoints the bucket is considered as fully matched.
570 : * The second form is to match the bucket partially; we recognize this when
571 : * the constant matches just one endpoint, or the two endpoints fall on
572 : * opposite sides of the constant. (Note that when the constant matches an
573 : * interior histogram element, it gets credit for partial matches to the
574 : * buckets on both sides, while a match to a histogram endpoint gets credit
575 : * for only one partial match. This is desirable.)
576 : *
577 : * The divider in the partial bucket match is imagined as the distance
578 : * between the decisive bits and the common bits of the addresses. It will
579 : * be used as a power of two as it is the natural scale for the IP network
580 : * inclusion. This partial bucket match divider calculation is an empirical
581 : * formula and subject to change with more experiment.
582 : *
583 : * For a partial match, we try to calculate dividers for both of the
584 : * boundaries. If the address family of a boundary value does not match the
585 : * constant or comparison of the length of the network parts is not correct
586 : * for the operator, the divider for that boundary will not be taken into
587 : * account. If both of the dividers are valid, the greater one will be used
588 : * to minimize the mistake in buckets that have disparate masklens. This
589 : * calculation is unfair when dividers can be calculated for both of the
590 : * boundaries but they are far from each other; but it is not a common
591 : * situation as the boundaries are expected to share most of their significant
592 : * bits of their masklens. The mistake would be greater, if we would use the
593 : * minimum instead of the maximum, and we don't know a sensible way to combine
594 : * them.
595 : *
596 : * For partial match in buckets that have different address families on the
597 : * left and right sides, only the boundary with the same address family is
598 : * taken into consideration. This can cause more mistakes for these buckets
599 : * if the masklens of their boundaries are also disparate. But this can only
600 : * happen in one bucket, since only two address families exist. It seems a
601 : * better option than not considering these buckets at all.
602 : */
603 : static Selectivity
604 0 : inet_hist_value_sel(Datum *values, int nvalues, Datum constvalue,
605 : int opr_codenum)
606 : {
607 0 : Selectivity match = 0.0;
608 : inet *query,
609 : *left,
610 : *right;
611 : int i,
612 : k,
613 : n;
614 : int left_order,
615 : right_order,
616 : left_divider,
617 : right_divider;
618 :
619 : /* guard against zero-divide below */
620 0 : if (nvalues <= 1)
621 0 : return 0.0;
622 :
623 : /* if there are too many histogram elements, decimate to limit runtime */
624 0 : k = (nvalues - 2) / MAX_CONSIDERED_ELEMS + 1;
625 :
626 0 : query = DatumGetInetPP(constvalue);
627 :
628 : /* "left" is the left boundary value of the current bucket ... */
629 0 : left = DatumGetInetPP(values[0]);
630 0 : left_order = inet_inclusion_cmp(left, query, opr_codenum);
631 :
632 0 : n = 0;
633 0 : for (i = k; i < nvalues; i += k)
634 : {
635 : /* ... and "right" is the right boundary value */
636 0 : right = DatumGetInetPP(values[i]);
637 0 : right_order = inet_inclusion_cmp(right, query, opr_codenum);
638 :
639 0 : if (left_order == 0 && right_order == 0)
640 : {
641 : /* The whole bucket matches, since both endpoints do. */
642 0 : match += 1.0;
643 : }
644 0 : else if ((left_order <= 0 && right_order >= 0) ||
645 0 : (left_order >= 0 && right_order <= 0))
646 : {
647 : /* Partial bucket match. */
648 0 : left_divider = inet_hist_match_divider(left, query, opr_codenum);
649 0 : right_divider = inet_hist_match_divider(right, query, opr_codenum);
650 :
651 0 : if (left_divider >= 0 || right_divider >= 0)
652 0 : match += 1.0 / pow(2.0, Max(left_divider, right_divider));
653 : }
654 :
655 : /* Shift the variables. */
656 0 : left = right;
657 0 : left_order = right_order;
658 :
659 : /* Count the number of buckets considered. */
660 0 : n++;
661 : }
662 :
663 0 : return match / n;
664 : }
665 :
666 : /*
667 : * Inet MCV vs MCV join selectivity estimation
668 : *
669 : * We simply add up the fractions of the populations that satisfy the clause.
670 : * The result is exact and does not need to be scaled further.
671 : */
672 : static Selectivity
673 0 : inet_mcv_join_sel(Datum *mcv1_values, float4 *mcv1_numbers, int mcv1_nvalues,
674 : Datum *mcv2_values, float4 *mcv2_numbers, int mcv2_nvalues,
675 : Oid operator)
676 : {
677 0 : Selectivity selec = 0.0;
678 : FmgrInfo proc;
679 : int i,
680 : j;
681 :
682 0 : fmgr_info(get_opcode(operator), &proc);
683 :
684 0 : for (i = 0; i < mcv1_nvalues; i++)
685 : {
686 0 : for (j = 0; j < mcv2_nvalues; j++)
687 0 : if (DatumGetBool(FunctionCall2(&proc,
688 : mcv1_values[i],
689 : mcv2_values[j])))
690 0 : selec += mcv1_numbers[i] * mcv2_numbers[j];
691 : }
692 0 : return selec;
693 : }
694 :
695 : /*
696 : * Inet MCV vs histogram join selectivity estimation
697 : *
698 : * For each MCV on the lefthand side, estimate the fraction of the righthand's
699 : * histogram population that satisfies the join clause, and add those up,
700 : * scaling by the MCV's frequency. The result still needs to be scaled
701 : * according to the fraction of the righthand's population represented by
702 : * the histogram.
703 : */
704 : static Selectivity
705 0 : inet_mcv_hist_sel(Datum *mcv_values, float4 *mcv_numbers, int mcv_nvalues,
706 : Datum *hist_values, int hist_nvalues,
707 : int opr_codenum)
708 : {
709 0 : Selectivity selec = 0.0;
710 : int i;
711 :
712 : /*
713 : * We'll call inet_hist_value_selec with the histogram on the left, so we
714 : * must commute the operator.
715 : */
716 0 : opr_codenum = -opr_codenum;
717 :
718 0 : for (i = 0; i < mcv_nvalues; i++)
719 : {
720 0 : selec += mcv_numbers[i] *
721 0 : inet_hist_value_sel(hist_values, hist_nvalues, mcv_values[i],
722 : opr_codenum);
723 : }
724 0 : return selec;
725 : }
726 :
727 : /*
728 : * Inet histogram vs histogram join selectivity estimation
729 : *
730 : * Here, we take all values listed in the second histogram (except for the
731 : * first and last elements, which are excluded on the grounds of possibly
732 : * not being very representative) and treat them as a uniform sample of
733 : * the non-MCV population for that relation. For each one, we apply
734 : * inet_hist_value_selec to see what fraction of the first histogram
735 : * it matches.
736 : *
737 : * We could alternatively do this the other way around using the operator's
738 : * commutator. XXX would it be worthwhile to do it both ways and take the
739 : * average? That would at least avoid non-commutative estimation results.
740 : */
741 : static Selectivity
742 0 : inet_hist_inclusion_join_sel(Datum *hist1_values, int hist1_nvalues,
743 : Datum *hist2_values, int hist2_nvalues,
744 : int opr_codenum)
745 : {
746 0 : double match = 0.0;
747 : int i,
748 : k,
749 : n;
750 :
751 0 : if (hist2_nvalues <= 2)
752 0 : return 0.0; /* no interior histogram elements */
753 :
754 : /* if there are too many histogram elements, decimate to limit runtime */
755 0 : k = (hist2_nvalues - 3) / MAX_CONSIDERED_ELEMS + 1;
756 :
757 0 : n = 0;
758 0 : for (i = 1; i < hist2_nvalues - 1; i += k)
759 : {
760 0 : match += inet_hist_value_sel(hist1_values, hist1_nvalues,
761 0 : hist2_values[i], opr_codenum);
762 0 : n++;
763 : }
764 :
765 0 : return match / n;
766 : }
767 :
768 : /*
769 : * Inet semi join selectivity estimation for one value
770 : *
771 : * The function calculates the probability that there is at least one row
772 : * in the RHS table that satisfies the "lhs_value op column" condition.
773 : * It is used in semi join estimation to check a sample from the left hand
774 : * side table.
775 : *
776 : * The MCV and histogram from the right hand side table should be provided as
777 : * arguments with the lhs_value from the left hand side table for the join.
778 : * hist_weight is the total number of rows represented by the histogram.
779 : * For example, if the table has 1000 rows, and 10% of the rows are in the MCV
780 : * list, and another 10% are NULLs, hist_weight would be 800.
781 : *
782 : * First, the lhs_value will be matched to the most common values. If it
783 : * matches any of them, 1.0 will be returned, because then there is surely
784 : * a match.
785 : *
786 : * Otherwise, the histogram will be used to estimate the number of rows in
787 : * the second table that match the condition. If the estimate is greater
788 : * than 1.0, 1.0 will be returned, because it means there is a greater chance
789 : * that the lhs_value will match more than one row in the table. If it is
790 : * between 0.0 and 1.0, it will be returned as the probability.
791 : */
792 : static Selectivity
793 0 : inet_semi_join_sel(Datum lhs_value,
794 : bool mcv_exists, Datum *mcv_values, int mcv_nvalues,
795 : bool hist_exists, Datum *hist_values, int hist_nvalues,
796 : double hist_weight,
797 : FmgrInfo *proc, int opr_codenum)
798 : {
799 0 : if (mcv_exists)
800 : {
801 : int i;
802 :
803 0 : for (i = 0; i < mcv_nvalues; i++)
804 : {
805 0 : if (DatumGetBool(FunctionCall2(proc,
806 : lhs_value,
807 : mcv_values[i])))
808 0 : return 1.0;
809 : }
810 : }
811 :
812 0 : if (hist_exists && hist_weight > 0)
813 : {
814 : Selectivity hist_selec;
815 :
816 : /* Commute operator, since we're passing lhs_value on the right */
817 0 : hist_selec = inet_hist_value_sel(hist_values, hist_nvalues,
818 : lhs_value, -opr_codenum);
819 :
820 0 : if (hist_selec > 0)
821 0 : return Min(1.0, hist_weight * hist_selec);
822 : }
823 :
824 0 : return 0.0;
825 : }
826 :
827 : /*
828 : * Assign useful code numbers for the subnet inclusion/overlap operators
829 : *
830 : * Only inet_masklen_inclusion_cmp() and inet_hist_match_divider() depend
831 : * on the exact codes assigned here; but many other places in this file
832 : * know that they can negate a code to obtain the code for the commutator
833 : * operator.
834 : */
835 : static int
836 0 : inet_opr_codenum(Oid operator)
837 : {
838 0 : switch (operator)
839 : {
840 0 : case OID_INET_SUP_OP:
841 0 : return -2;
842 0 : case OID_INET_SUPEQ_OP:
843 0 : return -1;
844 0 : case OID_INET_OVERLAP_OP:
845 0 : return 0;
846 0 : case OID_INET_SUBEQ_OP:
847 0 : return 1;
848 0 : case OID_INET_SUB_OP:
849 0 : return 2;
850 0 : default:
851 0 : elog(ERROR, "unrecognized operator %u for inet selectivity",
852 : operator);
853 : }
854 : return 0; /* unreached, but keep compiler quiet */
855 : }
856 :
857 : /*
858 : * Comparison function for the subnet inclusion/overlap operators
859 : *
860 : * If the comparison is okay for the specified inclusion operator, the return
861 : * value will be 0. Otherwise the return value will be less than or greater
862 : * than 0 as appropriate for the operator.
863 : *
864 : * Comparison is compatible with the basic comparison function for the inet
865 : * type. See network_cmp_internal() in network.c for the original. Basic
866 : * comparison operators are implemented with the network_cmp_internal()
867 : * function. It is possible to implement the subnet inclusion operators with
868 : * this function.
869 : *
870 : * Comparison is first on the common bits of the network part, then on the
871 : * length of the network part (masklen) as in the network_cmp_internal()
872 : * function. Only the first part is in this function. The second part is
873 : * separated to another function for reusability. The difference between the
874 : * second part and the original network_cmp_internal() is that the inclusion
875 : * operator is considered while comparing the lengths of the network parts.
876 : * See the inet_masklen_inclusion_cmp() function below.
877 : */
878 : static int
879 0 : inet_inclusion_cmp(inet *left, inet *right, int opr_codenum)
880 : {
881 0 : if (ip_family(left) == ip_family(right))
882 : {
883 : int order;
884 :
885 0 : order = bitncmp(ip_addr(left), ip_addr(right),
886 0 : Min(ip_bits(left), ip_bits(right)));
887 0 : if (order != 0)
888 0 : return order;
889 :
890 0 : return inet_masklen_inclusion_cmp(left, right, opr_codenum);
891 : }
892 :
893 0 : return ip_family(left) - ip_family(right);
894 : }
895 :
896 : /*
897 : * Masklen comparison function for the subnet inclusion/overlap operators
898 : *
899 : * Compares the lengths of the network parts of the inputs. If the comparison
900 : * is okay for the specified inclusion operator, the return value will be 0.
901 : * Otherwise the return value will be less than or greater than 0 as
902 : * appropriate for the operator.
903 : */
904 : static int
905 0 : inet_masklen_inclusion_cmp(inet *left, inet *right, int opr_codenum)
906 : {
907 : int order;
908 :
909 0 : order = (int) ip_bits(left) - (int) ip_bits(right);
910 :
911 : /*
912 : * Return 0 if the operator would accept this combination of masklens.
913 : * Note that opr_codenum zero (overlaps) will accept all cases.
914 : */
915 0 : if ((order > 0 && opr_codenum >= 0) ||
916 0 : (order == 0 && opr_codenum >= -1 && opr_codenum <= 1) ||
917 0 : (order < 0 && opr_codenum <= 0))
918 0 : return 0;
919 :
920 : /*
921 : * Otherwise, return a negative value for sup/supeq (notionally, the RHS
922 : * needs to have a larger masklen than it has, which would make it sort
923 : * later), or a positive value for sub/subeq (vice versa).
924 : */
925 0 : return opr_codenum;
926 : }
927 :
928 : /*
929 : * Inet histogram partial match divider calculation
930 : *
931 : * First the families and the lengths of the network parts are compared using
932 : * the subnet inclusion operator. If those are acceptable for the operator,
933 : * the divider will be calculated using the masklens and the common bits of
934 : * the addresses. -1 will be returned if it cannot be calculated.
935 : *
936 : * See commentary for inet_hist_value_sel() for some rationale for this.
937 : */
938 : static int
939 0 : inet_hist_match_divider(inet *boundary, inet *query, int opr_codenum)
940 : {
941 0 : if (ip_family(boundary) == ip_family(query) &&
942 0 : inet_masklen_inclusion_cmp(boundary, query, opr_codenum) == 0)
943 : {
944 : int min_bits,
945 : decisive_bits;
946 :
947 0 : min_bits = Min(ip_bits(boundary), ip_bits(query));
948 :
949 : /*
950 : * Set decisive_bits to the masklen of the one that should contain the
951 : * other according to the operator.
952 : */
953 0 : if (opr_codenum < 0)
954 0 : decisive_bits = ip_bits(boundary);
955 0 : else if (opr_codenum > 0)
956 0 : decisive_bits = ip_bits(query);
957 : else
958 0 : decisive_bits = min_bits;
959 :
960 : /*
961 : * Now return the number of non-common decisive bits. (This will be
962 : * zero if the boundary and query in fact match, else positive.)
963 : */
964 0 : if (min_bits > 0)
965 0 : return decisive_bits - bitncommon(ip_addr(boundary),
966 0 : ip_addr(query),
967 : min_bits);
968 0 : return decisive_bits;
969 : }
970 :
971 0 : return -1;
972 : }
|
