Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commite2eed78

Browse files
committed
Remove useless "rough estimate" path from mcelem_array_contained_selec.
The code in this function that tried to cope with a missing count histogramwas quite ineffective for anything except a perfectly flat distribution.Furthermore, since we were already punting for missing MCELEM slot, it'srather useless to sweat over missing DECHIST: there are no cases whereANALYZE will create the first but not the second. So just simplify thecode by punting rather than pretending we can do something useful.
1 parent4fb694a commite2eed78

File tree

1 file changed

+62
-76
lines changed

1 file changed

+62
-76
lines changed

‎src/backend/utils/adt/array_selfuncs.c‎

Lines changed: 62 additions & 76 deletions
Original file line numberDiff line numberDiff line change
@@ -242,8 +242,7 @@ scalararraysel_containment(PlannerInfo *root,
242242
}
243243

244244
/*
245-
* arraycontsel -- restriction selectivity for "arraycolumn @> const",
246-
* "arraycolumn && const" or "arraycolumn <@ const"
245+
* arraycontsel -- restriction selectivity for array @>, &&, <@ operators
247246
*/
248247
Datum
249248
arraycontsel(PG_FUNCTION_ARGS)
@@ -323,8 +322,7 @@ arraycontsel(PG_FUNCTION_ARGS)
323322
}
324323

325324
/*
326-
* arraycontjoinsel -- join selectivity for "arraycolumn @> const",
327-
* "arraycolumn && const" or "arraycolumn <@ const"
325+
* arraycontjoinsel -- join selectivity for array @>, &&, <@ operators
328326
*/
329327
Datum
330328
arraycontjoinsel(PG_FUNCTION_ARGS)
@@ -744,18 +742,18 @@ mcelem_array_contained_selec(Datum *mcelem, int nmcelem,
744742
if (numbers==NULL||nnumbers!=nmcelem+3)
745743
returnDEFAULT_CONTAIN_SEL;
746744

745+
/* Can't do much without a count histogram, either */
746+
if (hist==NULL||nhist<3)
747+
returnDEFAULT_CONTAIN_SEL;
748+
747749
/*
748750
* Grab some of the summary statistics that compute_array_stats() stores:
749751
* lowest frequency, frequency of null elements, and average distinct
750752
* element count.
751753
*/
752754
minfreq=numbers[nmcelem];
753755
nullelem_freq=numbers[nmcelem+2];
754-
755-
if (hist&&nhist>0)
756-
avg_count=hist[nhist-1];
757-
else
758-
avg_count=10.0f;/* default assumption */
756+
avg_count=hist[nhist-1];
759757

760758
/*
761759
* "rest" will be the sum of the frequencies of all elements not
@@ -853,83 +851,71 @@ mcelem_array_contained_selec(Datum *mcelem, int nmcelem,
853851
*/
854852
mult *=exp(-rest);
855853

856-
/* Check we have nonempty distinct element count histogram */
857-
if (hist&&nhist >=3)
858-
{
859-
/*----------
860-
* Using the distinct element count histogram requires
861-
*O(unique_nitems * (nmcelem + unique_nitems))
862-
* operations. Beyond a certain computational cost threshold, it's
863-
* reasonable to sacrifice accuracy for decreased planning time.
864-
* We limit the number of operations to EFFORT * nmcelem; since
865-
* nmcelem is limited by the column's statistics target, the work
866-
* done is user-controllable.
867-
*
868-
* If the number of operations would be too large, we can reduce it
869-
* without losing all accuracy by reducing unique_nitems and
870-
* considering only the most-common elements of the constant array.
871-
* To make the results exactly match what we would have gotten with
872-
* only those elements to start with, we'd have to remove any
873-
* discarded elements' frequencies from "mult", but since this is only
874-
* an approximation anyway, we don't bother with that. Therefore it's
875-
* sufficient to qsort elem_selec[] and take the largest elements.
876-
* (They will no longer match up with the elements of array_data[],
877-
* but we don't care.)
878-
*----------
879-
*/
854+
/*----------
855+
* Using the distinct element count histogram requires
856+
*O(unique_nitems * (nmcelem + unique_nitems))
857+
* operations. Beyond a certain computational cost threshold, it's
858+
* reasonable to sacrifice accuracy for decreased planning time. We limit
859+
* the number of operations to EFFORT * nmcelem; since nmcelem is limited
860+
* by the column's statistics target, the work done is user-controllable.
861+
*
862+
* If the number of operations would be too large, we can reduce it
863+
* without losing all accuracy by reducing unique_nitems and considering
864+
* only the most-common elements of the constant array. To make the
865+
* results exactly match what we would have gotten with only those
866+
* elements to start with, we'd have to remove any discarded elements'
867+
* frequencies from "mult", but since this is only an approximation
868+
* anyway, we don't bother with that. Therefore it's sufficient to qsort
869+
* elem_selec[] and take the largest elements. (They will no longer match
870+
* up with the elements of array_data[], but we don't care.)
871+
*----------
872+
*/
880873
#defineEFFORT 100
881874

882-
if ((nmcelem+unique_nitems)>0&&
883-
unique_nitems>EFFORT*nmcelem / (nmcelem+unique_nitems))
884-
{
885-
/*
886-
* Use the quadratic formula to solve for largest allowable N;
887-
* we have A = 1, B = nmcelem, C = - EFFORT * nmcelem.
888-
*/
889-
doubleb= (double)nmcelem;
890-
intn;
891-
892-
n= (int) ((sqrt(b*b+4*EFFORT*b)-b) /2);
893-
894-
/* Sort, then take just the first n elements */
895-
qsort(elem_selec,unique_nitems,sizeof(float),
896-
float_compare_desc);
897-
unique_nitems=n;
898-
}
899-
875+
if ((nmcelem+unique_nitems)>0&&
876+
unique_nitems>EFFORT*nmcelem / (nmcelem+unique_nitems))
877+
{
900878
/*
901-
* Calculate probabilities of each distinct element count for both
902-
* mcelems and constant elements. At this point, assume independent
903-
* element occurrence.
879+
* Use the quadratic formula to solve for largest allowable N. We
880+
* have A = 1, B = nmcelem, C = - EFFORT * nmcelem.
904881
*/
905-
dist=calc_distr(elem_selec,unique_nitems,unique_nitems,0.0f);
906-
mcelem_dist=calc_distr(numbers,nmcelem,unique_nitems,rest);
882+
doubleb=(double)nmcelem;
883+
intn;
907884

908-
/* ignore hist[nhist-1], which is the avg not a histogram member */
909-
hist_part=calc_hist(hist,nhist-1,unique_nitems);
885+
n= (int) ((sqrt(b*b+4*EFFORT*b)-b) /2);
910886

911-
selec=0.0f;
912-
for (i=0;i <=unique_nitems;i++)
913-
{
914-
/*
915-
* mult * dist[i] / mcelem_dist[i] gives us probability of qual
916-
* matching from assumption of independent element occurrence with
917-
* the condition that distinct element count = i.
918-
*/
919-
if (mcelem_dist[i]>0)
920-
selec+=hist_part[i]*mult*dist[i] /mcelem_dist[i];
921-
}
922-
923-
pfree(dist);
924-
pfree(mcelem_dist);
925-
pfree(hist_part);
887+
/* Sort, then take just the first n elements */
888+
qsort(elem_selec,unique_nitems,sizeof(float),
889+
float_compare_desc);
890+
unique_nitems=n;
926891
}
927-
else
892+
893+
/*
894+
* Calculate probabilities of each distinct element count for both
895+
* mcelems and constant elements. At this point, assume independent
896+
* element occurrence.
897+
*/
898+
dist=calc_distr(elem_selec,unique_nitems,unique_nitems,0.0f);
899+
mcelem_dist=calc_distr(numbers,nmcelem,unique_nitems,rest);
900+
901+
/* ignore hist[nhist-1], which is the average not a histogram member */
902+
hist_part=calc_hist(hist,nhist-1,unique_nitems);
903+
904+
selec=0.0f;
905+
for (i=0;i <=unique_nitems;i++)
928906
{
929-
/* We don't have histogram. Use a rough estimate. */
930-
selec=mult;
907+
/*
908+
* mult * dist[i] / mcelem_dist[i] gives us probability of qual
909+
* matching from assumption of independent element occurrence with
910+
* the condition that distinct element count = i.
911+
*/
912+
if (mcelem_dist[i]>0)
913+
selec+=hist_part[i]*mult*dist[i] /mcelem_dist[i];
931914
}
932915

916+
pfree(dist);
917+
pfree(mcelem_dist);
918+
pfree(hist_part);
933919
pfree(elem_selec);
934920

935921
/* Take into account occurrence of NULL element. */

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp