database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-07-17 06:41:09 +03:00
Code Activity

Measure the number of all-visible pages for use in index-only scan costing.

Browse Source 
Add a column pg_class.relallvisible to remember the number of pages that
were all-visible according to the visibility map as of the last VACUUM
(or ANALYZE, or some other operations that update pg_class.relpages).
Use relallvisible/relpages, instead of an arbitrary constant, to estimate
how many heap page fetches can be avoided during an index-only scan.

This is pretty primitive and will no doubt see refinements once we've
acquired more field experience with the index-only scan mechanism, but
it's way better than using a constant.

Note: I had to adjust an underspecified query in the window.sql regression
test, because it was changing answers when the plan changed to use an
index-only scan.  Some of the adjacent tests perhaps should be adjusted
as well, but I didn't do that here.
This commit is contained in:
Tom Lane
2011-10-14 17:23:01 -04:00
parent dea95c7a7b
commit e6858e6657
22 changed files with 246 additions and 72 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
src/backend/optimizer/path/costsize.c
View File

@ -120,9 +120,6 @@ bool enable_material = true;
bool enable_mergejoin = true;
bool enable_hashjoin = true;
/* Possibly this should become a GUC too */
static double visibility_fraction = 0.9;
typedef struct
{
PlannerInfo *root;
@ -324,9 +321,10 @@ cost_index(IndexPath *path, PlannerInfo *root,
*
* If it's an index-only scan, then we will not need to fetch any heap
* pages for which the visibility map shows all tuples are visible.
* Unfortunately, we have no stats as to how much of the heap is
* all-visible, and that's likely to be a rather unstable number anyway.
* We use an arbitrary constant visibility_fraction to estimate this.
* Hence, reduce the estimated number of heap fetches accordingly.
* We use the measured fraction of the entire heap that is all-visible,
* which might not be particularly relevant to the subset of the heap
* that this query will fetch; but it's not clear how to do better.
*----------
*/
if (outer_rel != NULL && outer_rel->rows > 1)
@ -347,7 +345,7 @@ cost_index(IndexPath *path, PlannerInfo *root,
root);
if (indexonly)
pages_fetched = ceil(pages_fetched * visibility_fraction);
pages_fetched = ceil(pages_fetched * (1.0 - baserel->allvisfrac));
max_IO_cost = (pages_fetched * spc_random_page_cost) / num_scans;
@ -369,7 +367,7 @@ cost_index(IndexPath *path, PlannerInfo *root,
root);
if (indexonly)
pages_fetched = ceil(pages_fetched * visibility_fraction);
pages_fetched = ceil(pages_fetched * (1.0 - baserel->allvisfrac));
min_IO_cost = (pages_fetched * spc_random_page_cost) / num_scans;
}
@ -385,7 +383,7 @@ cost_index(IndexPath *path, PlannerInfo *root,
root);
if (indexonly)
pages_fetched = ceil(pages_fetched * visibility_fraction);
pages_fetched = ceil(pages_fetched * (1.0 - baserel->allvisfrac));
/* max_IO_cost is for the perfectly uncorrelated case (csquared=0) */
max_IO_cost = pages_fetched * spc_random_page_cost;
@ -394,7 +392,7 @@ cost_index(IndexPath *path, PlannerInfo *root,
pages_fetched = ceil(indexSelectivity * (double) baserel->pages);
if (indexonly)
pages_fetched = ceil(pages_fetched * visibility_fraction);
pages_fetched = ceil(pages_fetched * (1.0 - baserel->allvisfrac));
min_IO_cost = spc_random_page_cost;
if (pages_fetched > 1)

30
src/backend/optimizer/util/plancat.c
View File

@ -116,7 +116,7 @@ get_relation_info(PlannerInfo *root, Oid relationObjectId, bool inhparent,
*/
if (!inhparent)
estimate_rel_size(relation, rel->attr_widths - rel->min_attr,
&rel->pages, &rel->tuples);
&rel->pages, &rel->tuples, &rel->allvisfrac);
/*
* Make list of indexes. Ignore indexes on system catalogs if told to.
@ -339,8 +339,10 @@ get_relation_info(PlannerInfo *root, Oid relationObjectId, bool inhparent,
}
else
{
double allvisfrac; /* dummy */
estimate_rel_size(indexRelation, NULL,
&info->pages, &info->tuples);
&info->pages, &info->tuples, &allvisfrac);
if (info->tuples > rel->tuples)
info->tuples = rel->tuples;
}
@ -369,17 +371,21 @@ get_relation_info(PlannerInfo *root, Oid relationObjectId, bool inhparent,
/*
* estimate_rel_size - estimate # pages and # tuples in a table or index
*
* We also estimate the fraction of the pages that are marked all-visible in
* the visibility map, for use in estimation of index-only scans.
*
* If attr_widths isn't NULL, it points to the zero-index entry of the
* relation's attr_widths[] cache; we fill this in if we have need to compute
* the attribute widths for estimation purposes.
*/
void
estimate_rel_size(Relation rel, int32 *attr_widths,
BlockNumber *pages, double *tuples)
BlockNumber *pages, double *tuples, double *allvisfrac)
{
BlockNumber curpages;
BlockNumber relpages;
double reltuples;
BlockNumber relallvisible;
double density;
switch (rel->rd_rel->relkind)
@ -432,11 +438,13 @@ estimate_rel_size(Relation rel, int32 *attr_widths,
if (curpages == 0)
{
*tuples = 0;
*allvisfrac = 0;
break;
}
/* coerce values in pg_class to more desirable types */
relpages = (BlockNumber) rel->rd_rel->relpages;
reltuples = (double) rel->rd_rel->reltuples;
relallvisible = (BlockNumber) rel->rd_rel->relallvisible;
/*
* If it's an index, discount the metapage while estimating the
@ -480,21 +488,37 @@ estimate_rel_size(Relation rel, int32 *attr_widths,
density = (BLCKSZ - SizeOfPageHeaderData) / tuple_width;
}
*tuples = rint(density * (double) curpages);
/*
* We use relallvisible as-is, rather than scaling it up like we
* do for the pages and tuples counts, on the theory that any
* pages added since the last VACUUM are most likely not marked
* all-visible. But costsize.c wants it converted to a fraction.
*/
if (relallvisible == 0 || curpages <= 0)
*allvisfrac = 0;
else if ((double) relallvisible >= curpages)
*allvisfrac = 1;
else
*allvisfrac = (double) relallvisible / curpages;
break;
case RELKIND_SEQUENCE:
/* Sequences always have a known size */
*pages = 1;
*tuples = 1;
*allvisfrac = 0;
break;
case RELKIND_FOREIGN_TABLE:
/* Just use whatever's in pg_class */
*pages = rel->rd_rel->relpages;
*tuples = rel->rd_rel->reltuples;
*allvisfrac = 0;
break;
default:
/* else it has no disk storage; probably shouldn't get here? */
*pages = 0;
*tuples = 0;
*allvisfrac = 0;
break;
}
}

2
src/backend/optimizer/util/relnode.c
View File

@ -109,6 +109,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptKind reloptkind)
rel->indexlist = NIL;
rel->pages = 0;
rel->tuples = 0;
rel->allvisfrac = 0;
rel->subplan = NULL;
rel->subroot = NULL;
rel->baserestrictinfo = NIL;
@ -362,6 +363,7 @@ build_join_rel(PlannerInfo *root,
joinrel->indexlist = NIL;
joinrel->pages = 0;
joinrel->tuples = 0;
joinrel->allvisfrac = 0;
joinrel->subplan = NULL;
joinrel->subroot = NULL;
joinrel->baserestrictinfo = NIL;