Build inherited extended stats on partitioned tables

Commit 859b3003de disabled building of extended stats for inheritance
trees, to prevent updating the same catalog row twice. While that
resolved the issue, it also means there are no extended stats for
declaratively partitioned tables, because there are no data in the
non-leaf relations.

That also means declaratively partitioned tables were not affected by
the issue 859b3003de addressed, which means this is a regression
affecting queries that calculate estimates for the whole inheritance
tree as a whole (which includes e.g. GROUP BY queries).

But because partitioned tables are empty, we can invert the condition
and build statistics only for the case with inheritance, without losing
anything. And we can consider them when calculating estimates.

It may be necessary to run ANALYZE on partitioned tables, to collect
proper statistics. For declarative partitioning there should no prior
statistics, and it might take time before autoanalyze is triggered. For
tables partitioned by inheritance the statistics may include data from
child relations (if built 859b3003de), contradicting the current code.

Report and patch by Justin Pryzby, minor fixes and cleanup by me.
Backpatch all the way back to PostgreSQL 10, where extended statistics
were introduced (same as 859b3003de).

Author: Justin Pryzby
Reported-by: Justin Pryzby
Backpatch-through: 10
Discussion: https://postgr.es/m/20210923212624.GI831%40telsasoft.com

4 files changed, 48 insertions, 19 deletions

diff --git a/src/backend/commands/analyze.c b/src/backend/commands/analyze.c
index cc9705d06bf..d447bc9b436 100644
--- a/src/backend/commands/analyze.c
+++ b/src/backend/commands/analyze.c
@@ -549,6 +549,7 @@ do_analyze_rel(Relation onerel, VacuumParams *params,
 	{
 		MemoryContext col_context,
 					old_context;
+		bool		build_ext_stats;
 
 		pgstat_progress_update_param(PROGRESS_ANALYZE_PHASE,
 									 PROGRESS_ANALYZE_PHASE_COMPUTE_STATS);
@@ -613,12 +614,27 @@ do_analyze_rel(Relation onerel, VacuumParams *params,
 		}
 
 		/*
+		 * Should we build extended statistics for this relation?
+		 *
+		 * The extended statistics catalog does not include an inheritance
+		 * flag, so we can't store statistics built both with and without
+		 * data from child relations. We can store just one set of statistics
+		 * per relation. For plain relations that's fine, but for inheritance
+		 * trees we have to pick whether to store statistics for just the
+		 * one relation or the whole tree. For plain inheritance we store
+		 * the (!inh) version, mostly for backwards compatibility reasons.
+		 * For partitioned tables that's pointless (the non-leaf tables are
+		 * always empty), so we store stats representing the whole tree.
+		 */
+		build_ext_stats = (onerel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE) ? inh : (!inh);
+
+		/*
 		 * Build extended statistics (if there are any).
 		 *
 		 * For now we only build extended statistics on individual relations,
 		 * not for relations representing inheritance trees.
 		 */
-		if (!inh)
+		if (build_ext_stats)
 			BuildRelationExtStatistics(onerel, totalrows, numrows, rows,
 									   attr_cnt, vacattrstats);
 	}
diff --git a/src/backend/statistics/dependencies.c b/src/backend/statistics/dependencies.c
index f1602cd3a26..bbc29b67117 100644
--- a/src/backend/statistics/dependencies.c
+++ b/src/backend/statistics/dependencies.c
@@ -1418,10 +1418,13 @@ dependencies_clauselist_selectivity(PlannerInfo *root,
 	int			unique_exprs_cnt;
 
 	/*
-	 * When dealing with inheritance trees, ignore extended stats (which were
-	 * built without data from child rels, and thus do not represent them).
+	 * When dealing with regular inheritance trees, ignore extended stats
+	 * (which were built without data from child rels, and thus do not
+	 * represent them). For partitioned tables data there's no data in the
+	 * non-leaf relations, so we build stats only for the inheritance tree.
+	 * So for partitioned tables we do consider extended stats.
 	 */
-	if (rte->inh)
+	if (rte->inh && rte->relkind != RELKIND_PARTITIONED_TABLE)
 		return 1.0;
 
 	/* check if there's any stats that might be useful for us. */
diff --git a/src/backend/statistics/extended_stats.c b/src/backend/statistics/extended_stats.c
index b23c15e9db4..5762621673b 100644
--- a/src/backend/statistics/extended_stats.c
+++ b/src/backend/statistics/extended_stats.c
@@ -1698,10 +1698,13 @@ statext_mcv_clauselist_selectivity(PlannerInfo *root, List *clauses, int varReli
 	RangeTblEntry *rte = planner_rt_fetch(rel->relid, root);
 
 	/*
-	 * When dealing with inheritance trees, ignore extended stats (which were
-	 * built without data from child rels, and thus do not represent them).
+	 * When dealing with regular inheritance trees, ignore extended stats
+	 * (which were built without data from child rels, and thus do not
+	 * represent them). For partitioned tables data there's no data in the
+	 * non-leaf relations, so we build stats only for the inheritance tree.
+	 * So for partitioned tables we do consider extended stats.
 	 */
-	if (rte->inh)
+	if (rte->inh && rte->relkind != RELKIND_PARTITIONED_TABLE)
 		return sel;
 
 	/* check if there's any stats that might be useful for us. */
diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c
index 9e39f9888e1..5d56748f0a7 100644
--- a/src/backend/utils/adt/selfuncs.c
+++ b/src/backend/utils/adt/selfuncs.c
@@ -3913,19 +3913,23 @@ estimate_multivariate_ndistinct(PlannerInfo *root, RelOptInfo *rel,
 	Oid			statOid = InvalidOid;
 	MVNDistinct *stats;
 	StatisticExtInfo *matched_info = NULL;
-	RangeTblEntry		*rte = planner_rt_fetch(rel->relid, root);
-
-	/*
-	 * When dealing with inheritance trees, ignore extended stats (which were
-	 * built without data from child rels, and thus do not represent them).
-	 */
-	if (rte->inh)
-		return false;
+	RangeTblEntry		*rte;
 
 	/* bail out immediately if the table has no extended statistics */
 	if (!rel->statlist)
 		return false;
 
+	/*
+	 * When dealing with regular inheritance trees, ignore extended stats
+	 * (which were built without data from child rels, and thus do not
+	 * represent them). For partitioned tables data there's no data in the
+	 * non-leaf relations, so we build stats only for the inheritance tree.
+	 * So for partitioned tables we do consider extended stats.
+	 */
+	rte = planner_rt_fetch(rel->relid, root);
+	if (rte->inh && rte->relkind != RELKIND_PARTITIONED_TABLE)
+		return false;
+
 	/* look for the ndistinct statistics object matching the most vars */
 	nmatches_vars = 0;			/* we require at least two matches */
 	nmatches_exprs = 0;
@@ -5242,11 +5246,14 @@ examine_variable(PlannerInfo *root, Node *node, int varRelid,
 				break;
 
 			/*
-			 * When dealing with inheritance trees, ignore extended stats (which
-			 * were built without data from child rels, and so do not represent
-			 * them).
+			 * When dealing with regular inheritance trees, ignore extended
+			 * stats (which were built without data from child rels, and thus
+			 * do not represent them). For partitioned tables data there's no
+			 * data in the non-leaf relations, so we build stats only for the
+			 * inheritance tree. So for partitioned tables we do consider
+			 * extended stats.
 			 */
-			if (rte->inh)
+			if (rte->inh && rte->relkind != RELKIND_PARTITIONED_TABLE)
 				break;
 
 			/* skip stats without per-expression stats */