From 164535616040447207c36ea47c52b206f4d77dcf Mon Sep 17 00:00:00 2001 From: Ilia Evdokimov Date: Tue, 10 Dec 2024 16:25:29 +0300 Subject: [PATCH v2] Define macros for minimum rows of stats in ANALYZE This introduces two macros, STATS_MIN_ROWS and EXT_STATS_MIN_ROWS, to represent the default minimum number of rows sampled in ANALYZE. STATS_MIN_ROWS is used for single-column statistics, while EXT_STATS_MIN_ROWS is intended for extended statistics. Both macros replace the hardcoded value of 300, improving clarity. --- src/backend/commands/analyze.c | 31 +++++-------------- src/backend/statistics/extended_stats.c | 2 +- src/backend/tsearch/ts_typanalyze.c | 4 +-- src/backend/utils/adt/rangetypes_typanalyze.c | 7 ++--- .../statistics/extended_stats_internal.h | 11 +++++++ src/include/statistics/statistics.h | 23 ++++++++++++++ 6 files changed, 47 insertions(+), 31 deletions(-) diff --git a/src/backend/commands/analyze.c b/src/backend/commands/analyze.c index 9a56de2282..e358b0d828 100644 --- a/src/backend/commands/analyze.c +++ b/src/backend/commands/analyze.c @@ -1897,42 +1897,25 @@ std_typanalyze(VacAttrStats *stats) { /* Seems to be a scalar datatype */ stats->compute_stats = compute_scalar_stats; - /*-------------------- - * The following choice of minrows is based on the paper - * "Random sampling for histogram construction: how much is enough?" - * by Surajit Chaudhuri, Rajeev Motwani and Vivek Narasayya, in - * Proceedings of ACM SIGMOD International Conference on Management - * of Data, 1998, Pages 436-447. Their Corollary 1 to Theorem 5 - * says that for table size n, histogram size k, maximum relative - * error in bin size f, and error probability gamma, the minimum - * random sample size is - * r = 4 * k * ln(2*n/gamma) / f^2 - * Taking f = 0.5, gamma = 0.01, n = 10^6 rows, we obtain - * r = 305.82 * k - * Note that because of the log function, the dependence on n is - * quite weak; even at n = 10^12, a 300*k sample gives <= 0.66 - * bin size error with probability 0.99. So there's no real need to - * scale for n, which is a good thing because we don't necessarily - * know it at this point. - *-------------------- - */ - stats->minrows = 300 * stats->attstattarget; } else if (OidIsValid(eqopr)) { /* We can still recognize distinct values */ stats->compute_stats = compute_distinct_stats; - /* Might as well use the same minrows as above */ - stats->minrows = 300 * stats->attstattarget; } else { /* Can't do much but the trivial stuff */ stats->compute_stats = compute_trivial_stats; - /* Might as well use the same minrows as above */ - stats->minrows = 300 * stats->attstattarget; } + /* + * For the scalar types, STATS_MIN_ROWS is derived from research on + * optimal sample sizes for histograms. For other cases, + * we assume the same value. + */ + stats->minrows = (STATS_MIN_ROWS * stats->attstattarget); + return true; } diff --git a/src/backend/statistics/extended_stats.c b/src/backend/statistics/extended_stats.c index 99fdf208db..8cd024b5ce 100644 --- a/src/backend/statistics/extended_stats.c +++ b/src/backend/statistics/extended_stats.c @@ -320,7 +320,7 @@ ComputeExtStatisticsRows(Relation onerel, MemoryContextDelete(cxt); /* compute sample size based on the statistics target */ - return (300 * result); + return (EXT_STATS_MIN_ROWS * result); } /* diff --git a/src/backend/tsearch/ts_typanalyze.c b/src/backend/tsearch/ts_typanalyze.c index ccafe42729..befd90c9d5 100644 --- a/src/backend/tsearch/ts_typanalyze.c +++ b/src/backend/tsearch/ts_typanalyze.c @@ -17,6 +17,7 @@ #include "catalog/pg_operator.h" #include "commands/vacuum.h" #include "common/hashfn.h" +#include "statistics/statistics.h" #include "tsearch/ts_type.h" #include "utils/builtins.h" #include "varatt.h" @@ -64,8 +65,7 @@ ts_typanalyze(PG_FUNCTION_ARGS) stats->attstattarget = default_statistics_target; stats->compute_stats = compute_tsvector_stats; - /* see comment about the choice of minrows in commands/analyze.c */ - stats->minrows = 300 * stats->attstattarget; + stats->minrows = (STATS_MIN_ROWS * stats->attstattarget); PG_RETURN_BOOL(true); } diff --git a/src/backend/utils/adt/rangetypes_typanalyze.c b/src/backend/utils/adt/rangetypes_typanalyze.c index 3773f98115..7fcac5621c 100644 --- a/src/backend/utils/adt/rangetypes_typanalyze.c +++ b/src/backend/utils/adt/rangetypes_typanalyze.c @@ -26,6 +26,7 @@ #include "catalog/pg_operator.h" #include "commands/vacuum.h" +#include "statistics/statistics.h" #include "utils/float.h" #include "utils/fmgrprotos.h" #include "utils/lsyscache.h" @@ -56,8 +57,7 @@ range_typanalyze(PG_FUNCTION_ARGS) stats->compute_stats = compute_range_stats; stats->extra_data = typcache; - /* same as in std_typanalyze */ - stats->minrows = 300 * stats->attstattarget; + stats->minrows = (STATS_MIN_ROWS * stats->attstattarget); PG_RETURN_BOOL(true); } @@ -82,8 +82,7 @@ multirange_typanalyze(PG_FUNCTION_ARGS) stats->compute_stats = compute_range_stats; stats->extra_data = typcache; - /* same as in std_typanalyze */ - stats->minrows = 300 * stats->attstattarget; + stats->minrows = (STATS_MIN_ROWS * stats->attstattarget); PG_RETURN_BOOL(true); } diff --git a/src/include/statistics/extended_stats_internal.h b/src/include/statistics/extended_stats_internal.h index 8eed9b338d..69c7fcca02 100644 --- a/src/include/statistics/extended_stats_internal.h +++ b/src/include/statistics/extended_stats_internal.h @@ -17,6 +17,17 @@ #include "statistics/statistics.h" #include "utils/sortsupport.h" +/* Minimum of rows wanted for extended stats + * + * Based on research, a fixed size proportional to statistics_target + * (300 * statistics_target) is generally sufficient for accurate single-column + * histograms and MCV lists. Its suitability for extended statistics. + * + * XXX While this value works reasonably well for individual columns, its + * suitability for extended statistics is still an open question. + */ +#define EXT_STATS_MIN_ROWS 300 + typedef struct { Oid eqopr; /* '=' operator for datatype, if any */ diff --git a/src/include/statistics/statistics.h b/src/include/statistics/statistics.h index 7f2bf18716..f578f7fd1d 100644 --- a/src/include/statistics/statistics.h +++ b/src/include/statistics/statistics.h @@ -22,6 +22,29 @@ #define STATS_NDISTINCT_MAGIC 0xA352BFA4 /* struct identifier */ #define STATS_NDISTINCT_TYPE_BASIC 1 /* struct version */ +/*-------------------- + * Minimum of rows wanted for stats + * + * The following choice of minrows is based on the paper + * "Random sampling for histogram construction: how much is enough?" + * by Surajit Chaudhuri, Rajeev Motwani and Vivek Narasayya, in + * Proceedings of ACM SIGMOD International Conference on Management + * of Data, 1998, Pages 436-447. Their Corollary 1 to Theorem 5 + * says that for table size n, histogram size k, maximum relative + * error in bin size f, and error probability gamma, the minimum + * random sample size is + * r = 4 * k * ln(2*n/gamma) / f^2 + * Taking f = 0.5, gamma = 0.01, n = 10^6 rows, we obtain + * r = 305.82 * k + * Note that because of the log function, the dependence on n is + * quite weak; even at n = 10^12, a 300*k sample gives <= 0.66 + * bin size error with probability 0.99. So there's no real need to + * scale for n, which is a good thing because we don't necessarily + * know it at this point. + *-------------------- + */ +#define STATS_MIN_ROWS 300 + /* MVNDistinctItem represents a single combination of columns */ typedef struct MVNDistinctItem { -- 2.34.1