Thanks for looking! On Wed, Jul 8, 2026 at 10:17 AM Michael Paquier <[email protected]> wrote: > > On Tue, Jul 07, 2026 at 04:13:30PM +0800, Ewan Young wrote: > > The attached patch validates the bounds histogram at import time and > > rejects an ill-formed one with a WARNING, matching the treatment of the > > other inconsistent inputs. Because the histogram element type is a range > > for both range- and multirange-typed columns, the single check covers > > both. > > I have been looking at that, and while the consequences of buggy > inputs are minor when loaded back, I don't really mind putting more > defenses to inform about that at the front of the restore functions. > > Now, your patch is entirely blind about extended statistics; these can > also load histogram bounds. The function you are introducing to > filter the inputs provided could be reused in this secondary case, > just by moving to stat_utils.c. Done in v2. I moved the check to stat_utils.c as stats_check_bounds_histogram() and now call it from both pg_restore_attribute_stats() and pg_restore_extended_stats(). In the extended path I reject an ill-formed histogram the same way the sibling stakinds there already do — WARNING plus goto pg_statistic_error.
> > I am not really convinced that any of this stuff would be worth a > backpatch, even if it's non-invasive. We don't have guards for > buggy infinite values, even if using empty or unordered bounds feels > kind of a stupid thing to do if one is a table owner. Agreed — v2 targets master only, no back-branch patches. > -- > Michael -- Regards, Ewan Young
From 1b06614bfdf5d91703cfa3ef166691feb0b5dda4 Mon Sep 17 00:00:00 2001 From: Ewan Young <[email protected]> Date: Tue, 7 Jul 2026 23:40:30 +0800 Subject: [PATCH v2] Reject ill-formed range bounds histograms in stats import functions pg_restore_attribute_stats() and pg_restore_extended_stats() (and their pg_set_*_stats() counterparts) stored a STATISTIC_KIND_BOUNDS_HISTOGRAM verbatim, without checking the invariant that ANALYZE's compute_range_stats() guarantees: the histogram must not contain any empty range, and its lower and upper bounds must each appear in ascending order. The range and multirange selectivity estimators rely on this. An unsorted histogram makes calc_length_hist_frac() fail an assertion (or compute a nonsensical, possibly negative, selectivity on a non-assert build), and an empty range trips a "shouldn't happen" elog(ERROR), when the planner later reads the imported statistics. For instance, after importing an unsorted bounds histogram, a query such as "WHERE r <@ ..." crashes the backend during planning. Validate the bounds histogram at import time and reject an ill-formed one with a WARNING, like other inconsistent inputs. The check lives in stat_utils.c so that both the per-attribute and the extended-statistics import paths, which can both load a bounds histogram, can share it. It covers both range- and multirange-typed columns, whose bounds histograms are both arrays of ranges. This is the same class of problem as commit f6e4ec0a705, which rejected oversized MCV lists in pg_restore_extended_stats(). --- src/backend/statistics/attribute_stats.c | 3 +- src/backend/statistics/extended_stats_funcs.c | 2 +- src/backend/statistics/stat_utils.c | 88 +++++++++++++++++++ src/include/statistics/stat_utils.h | 2 + src/test/regress/expected/stats_import.out | 56 ++++++++++++ src/test/regress/sql/stats_import.sql | 34 +++++++ 6 files changed, 183 insertions(+), 2 deletions(-) diff --git a/src/backend/statistics/attribute_stats.c b/src/backend/statistics/attribute_stats.c index 1cc4d657231..cb0375374b2 100644 --- a/src/backend/statistics/attribute_stats.c +++ b/src/backend/statistics/attribute_stats.c @@ -488,7 +488,8 @@ attribute_statistics_update(FunctionCallInfo fcinfo) atttypid, atttypmod, &converted); - if (converted) + if (converted && + stats_check_bounds_histogram(stavalues)) { statatt_set_slot(values, nulls, replaces, STATISTIC_KIND_BOUNDS_HISTOGRAM, diff --git a/src/backend/statistics/extended_stats_funcs.c b/src/backend/statistics/extended_stats_funcs.c index a5dce8a2206..f66e2360df4 100644 --- a/src/backend/statistics/extended_stats_funcs.c +++ b/src/backend/statistics/extended_stats_funcs.c @@ -1491,7 +1491,7 @@ import_pg_statistic(Relation pgsd, JsonbContainer *cont, extexprargname[RANGE_BOUNDS_HISTOGRAM_ELEM], &val_ok); - if (val_ok) + if (val_ok && stats_check_bounds_histogram(stavalues)) statatt_set_slot(values, nulls, replaces, STATISTIC_KIND_BOUNDS_HISTOGRAM, InvalidOid, InvalidOid, diff --git a/src/backend/statistics/stat_utils.c b/src/backend/statistics/stat_utils.c index 0b190e88237..3369ac2f170 100644 --- a/src/backend/statistics/stat_utils.c +++ b/src/backend/statistics/stat_utils.c @@ -33,6 +33,7 @@ #include "utils/array.h" #include "utils/builtins.h" #include "utils/lsyscache.h" +#include "utils/rangetypes.h" #include "utils/rel.h" #include "utils/syscache.h" #include "utils/typcache.h" @@ -742,3 +743,90 @@ statatt_init_empty_tuple(Oid reloid, int16 attnum, bool inherited, nulls[Anum_pg_statistic_stacoll1 + slotnum - 1] = false; } } + +/* + * Check that an imported bounds histogram (STATISTIC_KIND_BOUNDS_HISTOGRAM) is + * well-formed the way ANALYZE builds it in compute_range_stats(): it must not + * contain any empty range, and both its lower and its upper bounds must appear + * in non-decreasing order. The range and multirange selectivity estimators + * rely on this: they split the histogram into separate lower- and upper-bound + * histograms and assume each is sorted (e.g. calc_length_hist_frac() Asserts + * that successive bound distances do not decrease, and the estimator throws an + * internal error if it finds an empty range). Importing an unsorted or + * empty-containing histogram would therefore crash the backend, or produce + * nonsensical estimates on non-assert builds, when the histogram is later read + * by the planner. Reject such input with a WARNING, like other inconsistent + * inputs. + * + * This is shared between the per-attribute (pg_restore_attribute_stats()) and + * the extended-statistics (pg_restore_extended_stats()) import paths, both of + * which can load a bounds histogram. + * + * For both range- and multirange-typed columns the histogram is an array of + * ranges, so we take the range type from the array's element type. + */ +bool +stats_check_bounds_histogram(Datum arrayval) +{ + ArrayType *arr = DatumGetArrayTypeP(arrayval); + Oid rngtypid = ARR_ELEMTYPE(arr); + TypeCacheEntry *typcache; + int16 elmlen; + bool elmbyval; + char elmalign; + Datum *elems; + bool *nulls; + int nelems; + RangeBound prev_lower = {0}; + RangeBound prev_upper = {0}; + + typcache = lookup_type_cache(rngtypid, TYPECACHE_RANGE_INFO); + + /* + * The element type should always be a range type here, but be defensive: + * if it isn't, the bounds histogram is never consulted by the range + * estimator, so there is nothing to verify. + */ + if (typcache->rngelemtype == NULL) + return true; + + get_typlenbyvalalign(rngtypid, &elmlen, &elmbyval, &elmalign); + deconstruct_array(arr, rngtypid, elmlen, elmbyval, elmalign, + &elems, &nulls, &nelems); + + for (int i = 0; i < nelems; i++) + { + RangeBound lower, + upper; + bool empty; + + /* NULL elements are already rejected by statatt_build_stavalues() */ + range_deserialize(typcache, DatumGetRangeTypeP(elems[i]), + &lower, &upper, &empty); + + if (empty) + { + ereport(WARNING, + (errcode(ERRCODE_INVALID_PARAMETER_VALUE), + errmsg("\"%s\" must not contain empty ranges", + "range_bounds_histogram"))); + return false; + } + + if (i > 0 && + (range_cmp_bounds(typcache, &lower, &prev_lower) < 0 || + range_cmp_bounds(typcache, &upper, &prev_upper) < 0)) + { + ereport(WARNING, + (errcode(ERRCODE_INVALID_PARAMETER_VALUE), + errmsg("\"%s\" must have its lower and upper bounds sorted in ascending order", + "range_bounds_histogram"))); + return false; + } + + prev_lower = lower; + prev_upper = upper; + } + + return true; +} diff --git a/src/include/statistics/stat_utils.h b/src/include/statistics/stat_utils.h index 74da7790579..9f617a0f42b 100644 --- a/src/include/statistics/stat_utils.h +++ b/src/include/statistics/stat_utils.h @@ -58,4 +58,6 @@ extern Datum statatt_build_stavalues(const char *staname, FmgrInfo *array_in, Da extern bool statatt_get_elem_type(Oid atttypid, char atttyptype, Oid *elemtypid, Oid *elem_eq_opr); +extern bool stats_check_bounds_histogram(Datum arrayval); + #endif /* STATS_UTILS_H */ diff --git a/src/test/regress/expected/stats_import.out b/src/test/regress/expected/stats_import.out index bfdff77afa7..d4e5247b22a 100644 --- a/src/test/regress/expected/stats_import.out +++ b/src/test/regress/expected/stats_import.out @@ -1187,6 +1187,34 @@ AND attname = 'arange'; stats_import | test | arange | f | 0.29 | 0 | 0 | | | | | | | | {399,499,Infinity} | 0.5 | {"[-1,1)","[0,4)","[1,4)","[1,100)"} (1 row) +-- warn: range bounds histogram whose bounds are not sorted ascending +SELECT pg_catalog.pg_restore_attribute_stats( + 'schemaname', 'stats_import', + 'relname', 'test', + 'attname', 'arange', + 'inherited', false::boolean, + 'range_bounds_histogram', '{"[50,60)","[1,2)","[90,100)","[5,6)"}'::text + ); +WARNING: "range_bounds_histogram" must have its lower and upper bounds sorted in ascending order + pg_restore_attribute_stats +---------------------------- + f +(1 row) + +-- warn: range bounds histogram containing an empty range +SELECT pg_catalog.pg_restore_attribute_stats( + 'schemaname', 'stats_import', + 'relname', 'test', + 'attname', 'arange', + 'inherited', false::boolean, + 'range_bounds_histogram', '{empty,"[1,2)","[3,4)"}'::text + ); +WARNING: "range_bounds_histogram" must not contain empty ranges + pg_restore_attribute_stats +---------------------------- + f +(1 row) + -- warn: cannot set most_common_elems for range type, rest ok SELECT pg_catalog.pg_restore_attribute_stats( 'schemaname', 'stats_import', @@ -2358,6 +2386,34 @@ HINT: "range_length_histogram", "range_empty_frac", and "range_bounds_histogram f (1 row) +-- warn: range bounds histogram whose bounds are not sorted ascending +SELECT pg_catalog.pg_restore_extended_stats( + 'schemaname', 'stats_import', + 'relname', 'test_mr', + 'statistics_schemaname', 'stats_import', + 'statistics_name', 'test_mr_stat', + 'inherited', false, + 'exprs', '[{"range_length_histogram": "{10179,10189,10199}", "range_empty_frac": "0", "range_bounds_histogram": "{\"[50,60)\",\"[1,2)\",\"[90,100)\"}"}]'::jsonb); +WARNING: "range_bounds_histogram" must have its lower and upper bounds sorted in ascending order + pg_restore_extended_stats +--------------------------- + f +(1 row) + +-- warn: range bounds histogram containing an empty range +SELECT pg_catalog.pg_restore_extended_stats( + 'schemaname', 'stats_import', + 'relname', 'test_mr', + 'statistics_schemaname', 'stats_import', + 'statistics_name', 'test_mr_stat', + 'inherited', false, + 'exprs', '[{"range_length_histogram": "{10179,10189,10199}", "range_empty_frac": "0", "range_bounds_histogram": "{empty,\"[1,2)\",\"[3,4)\"}"}]'::jsonb); +WARNING: "range_bounds_histogram" must not contain empty ranges + pg_restore_extended_stats +--------------------------- + f +(1 row) + -- ok: multirange stats SELECT pg_catalog.pg_restore_extended_stats( 'schemaname', 'stats_import', diff --git a/src/test/regress/sql/stats_import.sql b/src/test/regress/sql/stats_import.sql index 58140315efb..e54f04ad638 100644 --- a/src/test/regress/sql/stats_import.sql +++ b/src/test/regress/sql/stats_import.sql @@ -885,6 +885,24 @@ AND tablename = 'test' AND inherited = false AND attname = 'arange'; +-- warn: range bounds histogram whose bounds are not sorted ascending +SELECT pg_catalog.pg_restore_attribute_stats( + 'schemaname', 'stats_import', + 'relname', 'test', + 'attname', 'arange', + 'inherited', false::boolean, + 'range_bounds_histogram', '{"[50,60)","[1,2)","[90,100)","[5,6)"}'::text + ); + +-- warn: range bounds histogram containing an empty range +SELECT pg_catalog.pg_restore_attribute_stats( + 'schemaname', 'stats_import', + 'relname', 'test', + 'attname', 'arange', + 'inherited', false::boolean, + 'range_bounds_histogram', '{empty,"[1,2)","[3,4)"}'::text + ); + -- warn: cannot set most_common_elems for range type, rest ok SELECT pg_catalog.pg_restore_attribute_stats( 'schemaname', 'stats_import', @@ -1696,6 +1714,22 @@ SELECT pg_catalog.pg_restore_extended_stats( 'statistics_name', 'test_mr_stat', 'inherited', false, 'exprs', '[{"range_bounds_histogram": "{\"[1,10200)\"}", "range_length_histogram": "{10179}"}]'::jsonb); +-- warn: range bounds histogram whose bounds are not sorted ascending +SELECT pg_catalog.pg_restore_extended_stats( + 'schemaname', 'stats_import', + 'relname', 'test_mr', + 'statistics_schemaname', 'stats_import', + 'statistics_name', 'test_mr_stat', + 'inherited', false, + 'exprs', '[{"range_length_histogram": "{10179,10189,10199}", "range_empty_frac": "0", "range_bounds_histogram": "{\"[50,60)\",\"[1,2)\",\"[90,100)\"}"}]'::jsonb); +-- warn: range bounds histogram containing an empty range +SELECT pg_catalog.pg_restore_extended_stats( + 'schemaname', 'stats_import', + 'relname', 'test_mr', + 'statistics_schemaname', 'stats_import', + 'statistics_name', 'test_mr_stat', + 'inherited', false, + 'exprs', '[{"range_length_histogram": "{10179,10189,10199}", "range_empty_frac": "0", "range_bounds_histogram": "{empty,\"[1,2)\",\"[3,4)\"}"}]'::jsonb); -- ok: multirange stats SELECT pg_catalog.pg_restore_extended_stats( -- 2.47.3
