diff options
author | Alvaro Herrera <alvherre@alvh.no-ip.org> | 2017-04-06 17:49:26 -0300 |
---|---|---|
committer | Alvaro Herrera <alvherre@alvh.no-ip.org> | 2017-04-06 17:51:53 -0300 |
commit | 7e534adcdc70866e7be74d626b0ed067c890a251 (patch) | |
tree | 609e0b65eb10538fdaa787fc29dee9655a6da710 /src/backend/utils/adt | |
parent | b2ff37d43cc81348fd8e9d9c5fcc9dfadf790763 (diff) | |
download | postgresql-7e534adcdc70866e7be74d626b0ed067c890a251.tar.gz postgresql-7e534adcdc70866e7be74d626b0ed067c890a251.zip |
Fix BRIN cost estimation
The original code was overly optimistic about the cost of scanning a
BRIN index, leading to BRIN indexes being selected when they'd be a
worse choice than some other index. This complete rewrite should be
more accurate.
Author: David Rowley, based on an earlier patch by Emre Hasegeli
Reviewed-by: Emre Hasegeli
Discussion: https://postgr.es/m/CAKJS1f9n-Wapop5Xz1dtGdpdqmzeGqQK4sV2MK-zZugfC14Xtw@mail.gmail.com
Diffstat (limited to 'src/backend/utils/adt')
-rw-r--r-- | src/backend/utils/adt/selfuncs.c | 197 |
1 files changed, 171 insertions, 26 deletions
diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c index f5cffcb2eac..b6893e22bfa 100644 --- a/src/backend/utils/adt/selfuncs.c +++ b/src/backend/utils/adt/selfuncs.c @@ -101,6 +101,7 @@ #include <float.h> #include <math.h> +#include "access/brin.h" #include "access/gin.h" #include "access/htup_details.h" #include "access/sysattr.h" @@ -7721,56 +7722,200 @@ brincostestimate(PlannerInfo *root, IndexPath *path, double loop_count, { IndexOptInfo *index = path->indexinfo; List *indexQuals = path->indexquals; - List *indexOrderBys = path->indexorderbys; double numPages = index->pages; - double numTuples = index->tuples; + RelOptInfo *baserel = index->rel; + RangeTblEntry *rte = planner_rt_fetch(baserel->relid, root); List *qinfos; Cost spc_seq_page_cost; Cost spc_random_page_cost; - double qual_op_cost; double qual_arg_cost; + double qualSelectivity; + BrinStatsData statsData; + double indexRanges; + double minimalRanges; + double estimatedRanges; + double selec; + Relation indexRel; + ListCell *l; + VariableStatData vardata; - /* Do preliminary analysis of indexquals */ - qinfos = deconstruct_indexquals(path); + Assert(rte->rtekind == RTE_RELATION); - /* fetch estimated page cost for tablespace containing index */ + /* fetch estimated page cost for the tablespace containing the index */ get_tablespace_page_costs(index->reltablespace, &spc_random_page_cost, &spc_seq_page_cost); /* - * BRIN indexes are always read in full; use that as startup cost. + * Obtain some data from the index itself. + */ + indexRel = index_open(index->indexoid, AccessShareLock); + brinGetStats(indexRel, &statsData); + index_close(indexRel, AccessShareLock); + + /* + * Compute index correlation * - * XXX maybe only include revmap pages here? + * Because we can use all index quals equally when scanning, we can use + * the largest correlation (in absolute value) among columns used by the + * query. Start at zero, the worst possible case. If we cannot find + * any correlation statistics, we will keep it as 0. + */ + *indexCorrelation = 0; + + qinfos = deconstruct_indexquals(path); + foreach(l, qinfos) + { + IndexQualInfo *qinfo = (IndexQualInfo *) lfirst(l); + AttrNumber attnum = index->indexkeys[qinfo->indexcol]; + + /* attempt to lookup stats in relation for this index column */ + if (attnum != 0) + { + /* Simple variable -- look to stats for the underlying table */ + if (get_relation_stats_hook && + (*get_relation_stats_hook) (root, rte, attnum, &vardata)) + { + /* + * The hook took control of acquiring a stats tuple. If it + * did supply a tuple, it'd better have supplied a freefunc. + */ + if (HeapTupleIsValid(vardata.statsTuple) && !vardata.freefunc) + elog(ERROR, + "no function provided to release variable stats with"); + } + else + { + vardata.statsTuple = + SearchSysCache3(STATRELATTINH, + ObjectIdGetDatum(rte->relid), + Int16GetDatum(attnum), + BoolGetDatum(false)); + vardata.freefunc = ReleaseSysCache; + } + } + else + { + /* + * Looks like we've found an expression column in the index. Let's + * see if there's any stats for it. + */ + + /* get the attnum from the 0-based index. */ + attnum = qinfo->indexcol + 1; + + if (get_index_stats_hook && + (*get_index_stats_hook) (root, index->indexoid, attnum, &vardata)) + { + /* + * The hook took control of acquiring a stats tuple. If it did + * supply a tuple, it'd better have supplied a freefunc. + */ + if (HeapTupleIsValid(vardata.statsTuple) && + !vardata.freefunc) + elog(ERROR, "no function provided to release variable stats with"); + } + else + { + vardata.statsTuple = SearchSysCache3(STATRELATTINH, + ObjectIdGetDatum(index->indexoid), + Int16GetDatum(attnum), + BoolGetDatum(false)); + vardata.freefunc = ReleaseSysCache; + } + } + + if (HeapTupleIsValid(vardata.statsTuple)) + { + float4 *numbers; + int nnumbers; + + if (get_attstatsslot(vardata.statsTuple, InvalidOid, 0, + STATISTIC_KIND_CORRELATION, + InvalidOid, + NULL, + NULL, NULL, + &numbers, &nnumbers)) + { + double varCorrelation = 0.0; + + if (nnumbers > 0) + varCorrelation = Abs(numbers[0]); + + if (varCorrelation > *indexCorrelation) + *indexCorrelation = varCorrelation; + + free_attstatsslot(InvalidOid, NULL, 0, numbers, nnumbers); + } + } + + ReleaseVariableStats(vardata); + } + + qualSelectivity = clauselist_selectivity(root, indexQuals, + baserel->relid, + JOIN_INNER, NULL, + baserel); + + /* work out the actual number of ranges in the index */ + indexRanges = Max(ceil((double) baserel->pages / statsData.pagesPerRange), + 1.0); + + /* + * Now calculate the minimum possible ranges we could match with if all of + * the rows were in the perfect order in the table's heap. */ - *indexStartupCost = spc_seq_page_cost * numPages * loop_count; + minimalRanges = ceil(indexRanges * qualSelectivity); /* - * To read a BRIN index there might be a bit of back and forth over - * regular pages, as revmap might point to them out of sequential order; - * calculate this as reading the whole index in random order. + * Now estimate the number of ranges that we'll touch by using the + * indexCorrelation from the stats. Careful not to divide by zero + * (note we're using the absolute value of the correlation). */ - *indexTotalCost = spc_random_page_cost * numPages * loop_count; + if (*indexCorrelation < 1.0e-10) + estimatedRanges = indexRanges; + else + estimatedRanges = Min(minimalRanges / *indexCorrelation, indexRanges); - *indexSelectivity = - clauselist_selectivity(root, indexQuals, - path->indexinfo->rel->relid, - JOIN_INNER, NULL, - path->indexinfo->rel); - *indexCorrelation = 1; + /* we expect to visit this portion of the table */ + selec = estimatedRanges / indexRanges; + + CLAMP_PROBABILITY(selec); + + *indexSelectivity = selec; /* - * Add on index qual eval costs, much as in genericcostestimate. + * Compute the index qual costs, much as in genericcostestimate, to add + * to the index costs. */ qual_arg_cost = other_operands_eval_cost(root, qinfos) + orderby_operands_eval_cost(root, path); - qual_op_cost = cpu_operator_cost * - (list_length(indexQuals) + list_length(indexOrderBys)); + /* + * Compute the startup cost as the cost to read the whole revmap + * sequentially, including the cost to execute the index quals. + */ + *indexStartupCost = + spc_seq_page_cost * statsData.revmapNumPages * loop_count; *indexStartupCost += qual_arg_cost; - *indexTotalCost += qual_arg_cost; - *indexTotalCost += (numTuples * *indexSelectivity) * (cpu_index_tuple_cost + qual_op_cost); - *indexPages = index->pages; - /* XXX what about pages_per_range? */ + /* + * To read a BRIN index there might be a bit of back and forth over + * regular pages, as revmap might point to them out of sequential order; + * calculate the total cost as reading the whole index in random order. + */ + *indexTotalCost = *indexStartupCost + + spc_random_page_cost * (numPages - statsData.revmapNumPages) * loop_count; + + /* + * Charge a small amount per range tuple which we expect to match to. This + * is meant to reflect the costs of manipulating the bitmap. The BRIN scan + * will set a bit for each page in the range when we find a matching + * range, so we must multiply the charge by the number of pages in the + * range. + */ + *indexTotalCost += 0.1 * cpu_operator_cost * estimatedRanges * + statsData.pagesPerRange; + + *indexPages = index->pages; } |