public inbox for [email protected]
help / color / mirror / Atom feed[PATCH v38 2/6] Introduce UniqueKey attributes on RelOptInfo struct.
9+ messages / 4 participants
[nested] [flat]
* [PATCH v38 2/6] Introduce UniqueKey attributes on RelOptInfo struct.
@ 2020-05-11 07:50 一挃 <[email protected]>
0 siblings, 0 replies; 9+ messages in thread
From: 一挃 @ 2020-05-11 07:50 UTC (permalink / raw)
UniqueKey is a set of exprs on RelOptInfo which represents the exprs
will be unique on the given RelOptInfo. You can see README.uniquekey
for more information.
---
src/backend/nodes/copyfuncs.c | 13 +
src/backend/nodes/list.c | 31 +
src/backend/nodes/makefuncs.c | 13 +
src/backend/nodes/outfuncs.c | 11 +
src/backend/nodes/readfuncs.c | 10 +
src/backend/optimizer/path/Makefile | 3 +-
src/backend/optimizer/path/README.uniquekey | 131 +++
src/backend/optimizer/path/allpaths.c | 10 +
src/backend/optimizer/path/joinpath.c | 9 +-
src/backend/optimizer/path/joinrels.c | 2 +
src/backend/optimizer/path/pathkeys.c | 3 +-
src/backend/optimizer/path/uniquekeys.c | 1135 +++++++++++++++++++
src/backend/optimizer/plan/planner.c | 13 +-
src/backend/optimizer/prep/prepunion.c | 2 +
src/backend/optimizer/util/appendinfo.c | 44 +
src/backend/optimizer/util/inherit.c | 18 +-
src/include/nodes/makefuncs.h | 3 +
src/include/nodes/nodes.h | 1 +
src/include/nodes/pathnodes.h | 29 +-
src/include/nodes/pg_list.h | 10 +
src/include/optimizer/appendinfo.h | 3 +
src/include/optimizer/optimizer.h | 2 +
src/include/optimizer/paths.h | 43 +
23 files changed, 1515 insertions(+), 24 deletions(-)
create mode 100644 src/backend/optimizer/path/README.uniquekey
create mode 100644 src/backend/optimizer/path/uniquekeys.c
diff --git a/src/backend/nodes/copyfuncs.c b/src/backend/nodes/copyfuncs.c
index da91cbd2b1..75c1c5e824 100644
--- a/src/backend/nodes/copyfuncs.c
+++ b/src/backend/nodes/copyfuncs.c
@@ -2296,6 +2296,16 @@ _copyPathKey(const PathKey *from)
return newnode;
}
+static UniqueKey *
+_copyUniqueKey(const UniqueKey *from)
+{
+ UniqueKey *newnode = makeNode(UniqueKey);
+
+ COPY_NODE_FIELD(exprs);
+ COPY_SCALAR_FIELD(multi_nullvals);
+
+ return newnode;
+}
/*
* _copyRestrictInfo
*/
@@ -5220,6 +5230,9 @@ copyObjectImpl(const void *from)
case T_PathKey:
retval = _copyPathKey(from);
break;
+ case T_UniqueKey:
+ retval = _copyUniqueKey(from);
+ break;
case T_RestrictInfo:
retval = _copyRestrictInfo(from);
break;
diff --git a/src/backend/nodes/list.c b/src/backend/nodes/list.c
index dbf6b30233..ca099495a1 100644
--- a/src/backend/nodes/list.c
+++ b/src/backend/nodes/list.c
@@ -702,6 +702,37 @@ list_member_oid(const List *list, Oid datum)
return false;
}
+/*
+ * return true iff every entry in "members" list is also present
+ * in the "target" list.
+ */
+bool
+list_is_subset(const List *members, const List *target)
+{
+ const ListCell *lc1, *lc2;
+
+ Assert(IsPointerList(members));
+ Assert(IsPointerList(target));
+ check_list_invariants(members);
+ check_list_invariants(target);
+
+ foreach(lc1, members)
+ {
+ bool found = false;
+ foreach(lc2, target)
+ {
+ if (equal(lfirst(lc1), lfirst(lc2)))
+ {
+ found = true;
+ break;
+ }
+ }
+ if (!found)
+ return false;
+ }
+ return true;
+}
+
/*
* Delete the n'th cell (counting from 0) in list.
*
diff --git a/src/backend/nodes/makefuncs.c b/src/backend/nodes/makefuncs.c
index 01c110cd2f..40415d0f5b 100644
--- a/src/backend/nodes/makefuncs.c
+++ b/src/backend/nodes/makefuncs.c
@@ -815,3 +815,16 @@ makeVacuumRelation(RangeVar *relation, Oid oid, List *va_cols)
v->va_cols = va_cols;
return v;
}
+
+
+/*
+ * makeUniqueKey
+ */
+UniqueKey*
+makeUniqueKey(List *exprs, bool multi_nullvals)
+{
+ UniqueKey * ukey = makeNode(UniqueKey);
+ ukey->exprs = exprs;
+ ukey->multi_nullvals = multi_nullvals;
+ return ukey;
+}
diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c
index 6493a03ff8..44154cde6a 100644
--- a/src/backend/nodes/outfuncs.c
+++ b/src/backend/nodes/outfuncs.c
@@ -2456,6 +2456,14 @@ _outPathKey(StringInfo str, const PathKey *node)
WRITE_BOOL_FIELD(pk_nulls_first);
}
+static void
+_outUniqueKey(StringInfo str, const UniqueKey *node)
+{
+ WRITE_NODE_TYPE("UNIQUEKEY");
+ WRITE_NODE_FIELD(exprs);
+ WRITE_BOOL_FIELD(multi_nullvals);
+}
+
static void
_outPathTarget(StringInfo str, const PathTarget *node)
{
@@ -4198,6 +4206,9 @@ outNode(StringInfo str, const void *obj)
case T_PathKey:
_outPathKey(str, obj);
break;
+ case T_UniqueKey:
+ _outUniqueKey(str, obj);
+ break;
case T_PathTarget:
_outPathTarget(str, obj);
break;
diff --git a/src/backend/nodes/readfuncs.c b/src/backend/nodes/readfuncs.c
index c5e136e9c3..b3e212bf1c 100644
--- a/src/backend/nodes/readfuncs.c
+++ b/src/backend/nodes/readfuncs.c
@@ -492,6 +492,14 @@ _readSetOperationStmt(void)
READ_DONE();
}
+static UniqueKey *
+_readUniqueKey(void)
+{
+ READ_LOCALS(UniqueKey);
+ READ_NODE_FIELD(exprs);
+ READ_BOOL_FIELD(multi_nullvals);
+ READ_DONE();
+}
/*
* Stuff from primnodes.h.
@@ -2717,6 +2725,8 @@ parseNodeString(void)
return_value = _readCommonTableExpr();
else if (MATCH("SETOPERATIONSTMT", 16))
return_value = _readSetOperationStmt();
+ else if (MATCH("UNIQUEKEY", 9))
+ return_value = _readUniqueKey();
else if (MATCH("ALIAS", 5))
return_value = _readAlias();
else if (MATCH("RANGEVAR", 8))
diff --git a/src/backend/optimizer/path/Makefile b/src/backend/optimizer/path/Makefile
index 1e199ff66f..7b9820c25f 100644
--- a/src/backend/optimizer/path/Makefile
+++ b/src/backend/optimizer/path/Makefile
@@ -21,6 +21,7 @@ OBJS = \
joinpath.o \
joinrels.o \
pathkeys.o \
- tidpath.o
+ tidpath.o \
+ uniquekeys.o
include $(top_srcdir)/src/backend/common.mk
diff --git a/src/backend/optimizer/path/README.uniquekey b/src/backend/optimizer/path/README.uniquekey
new file mode 100644
index 0000000000..5eac761995
--- /dev/null
+++ b/src/backend/optimizer/path/README.uniquekey
@@ -0,0 +1,131 @@
+1. What is UniqueKey?
+We can think UniqueKey is a set of exprs for a RelOptInfo, which we are insure
+that doesn't yields same result among all the rows. The simplest UniqueKey
+format is primary key.
+
+However we define the UnqiueKey as below.
+
+typedef struct UniqueKey
+{
+ NodeTag type;
+ List *exprs;
+ bool multi_nullvals;
+} UniqueKey;
+
+exprs is a list of exprs which is unique on current RelOptInfo. exprs = NIL
+is a special case of UniqueKey, which means there is only one row in that
+relation.it has a stronger semantic than others. like SELECT uk FROM t; uk is
+normal unique key and may have different values. SELECT colx FROM t WHERE uk =
+const. colx is unique AND we have only 1 value. This field can used for
+innerrel_is_unique. this logic is handled specially in add_uniquekey_for_onerow
+function.
+
+multi_nullvals: true means multi null values may exist in these exprs, so the
+uniqueness is not guaranteed in this case. This field is necessary for
+remove_useless_join & reduce_unique_semijoins where we don't mind these
+duplicated NULL values. It is set to true for 2 cases. One is a unique key
+from a unique index but the related column is nullable. The other one is for
+outer join. see populate_joinrel_uniquekeys for detail.
+
+
+The UniqueKey can be used at the following cases at least:
+1. remove_useless_joins.
+2. reduce_semianti_joins
+3. remove distinct node if distinct clause is unique.
+4. remove aggnode if group by clause is unique.
+5. Index Skip Scan (WIP)
+6. Aggregation Push Down without 2 phase aggregation if the join can't
+ duplicated the aggregated rows. (work in progress feature)
+
+2. How is it maintained?
+
+We have a set of populate_xxx_unqiuekeys functions to maintain the uniquekey on
+various cases. xxx includes baserel, joinrel, partitionedrel, distinctrel,
+groupedrel, unionrel. and we also need to convert the uniquekey from subquery
+to outer relation, which is what convert_subquery_uniquekeys does.
+
+1. The first part is about baserel. We handled 3 cases. suppose we have Unique
+Index on (a, b).
+
+1. SELECT a, b FROM t. UniqueKey (a, b)
+2. SELECT a FROM t WHERE b = 1; UniqueKey (a)
+3. SELECT .. FROM t WHERE a = 1 AND b = 1; UniqueKey (NIL). onerow case, every
+ column is Unique.
+
+2. The next part is joinrel, this part is most error-prone, we simplified the rules
+like below:
+1. If the relation's UniqueKey can't be duplicated after join, then is will be
+ still valid for the join rel. The function we used here is
+ innerrel_keeps_unique. The basic idea is innerrel.any_col = outer.uk.
+
+2. If the UnqiueKey can't keep valid via the rule 1, the combination of the
+ UniqueKey from both sides are valid for sure. We can prove this as: if the
+ unique exprs from rel1 is duplicated by rel2, the duplicated rows must
+ contains different unique exprs from rel2.
+
+More considerations about onerow:
+1. If relation with one row and it can't be duplicated, it is still possible
+ contains mulit_nullvas after outer join.
+2. If the either UniqueKey can be duplicated after join, the can get one row
+ only when both side is one row AND there is no outer join.
+3. Whenever the onerow UniqueKey is not a valid any more, we need to convert one
+ row UniqueKey to normal unique key since we don't store exprs for one-row
+ relation. get_exprs_from_uniquekeys will be used here.
+
+
+More considerations about multi_nullvals after join:
+1. If the original UnqiueKey has multi_nullvals, the final UniqueKey will have
+ mulit_nullvals in any case.
+2. If a unique key doesn't allow mulit_nullvals, after some outer join, it
+ allows some outer join.
+
+
+3. When we comes to subquery, we need to convert_subquery_unqiuekeys just like
+convert_subquery_pathkeys. Only the UniqueKey insides subquery is referenced as
+a Var in outer relation will be reused. The relationship between the outerrel.Var
+and subquery.exprs is built with outerel->subroot->processed_tlist.
+
+
+4. As for the SRF functions, it will break the uniqueness of uniquekey, However it
+is handled in adjust_paths_for_srfs, which happens after the query_planner. so
+we will maintain the UniqueKey until there and reset it to NIL at that
+places. This can't help on distinct/group by elimination cases but probably help
+in some other cases, like reduce_unqiue_semijoins/remove_useless_joins and it is
+semantic correctly.
+
+
+5. As for inherit table, we first main the UnqiueKey on childrel as well. But for
+partitioned table we need to maintain 2 different kinds of
+UnqiueKey. 1). UniqueKey on the parent relation 2). UniqueKey on child
+relation for partition wise query.
+
+Example:
+CREATE TABLE p (a int not null, b int not null) partition by list (a);
+CREATE TABLE p0 partition of p for values in (1);
+CREATE TABLE p1 partition of p for values in (2);
+
+create unique index p0_b on p0(b);
+create unique index p1_b on p1(b);
+
+Now b is only unique on partition level, so the distinct can't be removed on
+the following cases. SELECT DISTINCT b FROM p;
+
+Another example is SELECT DISTINCT a, b FROM p WHERE a = 1; Since only one
+partition is chosen, the UniqueKey on child relation is same as the UniqueKey on
+parent relation.
+
+Another usage of UniqueKey on partition level is it be helpful for
+partition-wise join.
+
+As for the UniqueKey on parent table level, it comes with 2 different ways,
+1). the UniqueKey is also derived in UniqueKey index, but the index must be same
+in all the related children relations and the unique index must contains
+Partition Key in it. Example:
+
+CREATE UNIQUE INDEX p_ab ON p(a, b); -- where a is the partition key.
+
+-- Query
+SELECT a, b FROM p; the (a, b) is a UniqueKey of p.
+
+2). If the parent relation has only one childrel, the UniqueKey on childrel is
+ the UniqueKey on parent as well.
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 37b4223adb..d52ad59f89 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -579,6 +579,12 @@ set_plain_rel_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
*/
check_index_predicates(root, rel);
+ /*
+ * Now that we've marked which partial indexes are suitable, we can now
+ * build the relation's unique keys.
+ */
+ populate_baserel_uniquekeys(root, rel, rel->indexlist);
+
/* Mark rel with estimated output rows, width, etc */
set_baserel_size_estimates(root, rel);
}
@@ -1297,6 +1303,8 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
/* Add paths to the append relation. */
add_paths_to_append_rel(root, rel, live_childrels);
+ if (IS_PARTITIONED_REL(rel))
+ populate_partitionedrel_uniquekeys(root, rel, live_childrels);
}
@@ -2304,6 +2312,8 @@ set_subquery_pathlist(PlannerInfo *root, RelOptInfo *rel,
pathkeys, required_outer));
}
+ convert_subquery_uniquekeys(root, rel, sub_final_rel);
+
/* If outer rel allows parallelism, do same for partial paths. */
if (rel->consider_parallel && bms_is_empty(required_outer))
{
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 57ce97fd53..697ac047fb 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -71,13 +71,6 @@ static void consider_parallel_mergejoin(PlannerInfo *root,
static void hash_inner_and_outer(PlannerInfo *root, RelOptInfo *joinrel,
RelOptInfo *outerrel, RelOptInfo *innerrel,
JoinType jointype, JoinPathExtraData *extra);
-static List *select_mergejoin_clauses(PlannerInfo *root,
- RelOptInfo *joinrel,
- RelOptInfo *outerrel,
- RelOptInfo *innerrel,
- List *restrictlist,
- JoinType jointype,
- bool *mergejoin_allowed);
static void generate_mergejoin_paths(PlannerInfo *root,
RelOptInfo *joinrel,
RelOptInfo *innerrel,
@@ -1927,7 +1920,7 @@ hash_inner_and_outer(PlannerInfo *root,
* if it is mergejoinable and involves vars from the two sub-relations
* currently of interest.
*/
-static List *
+List *
select_mergejoin_clauses(PlannerInfo *root,
RelOptInfo *joinrel,
RelOptInfo *outerrel,
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 0dbe2ac726..7271f044ec 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -924,6 +924,8 @@ populate_joinrel_with_paths(PlannerInfo *root, RelOptInfo *rel1,
/* Apply partitionwise join technique, if possible. */
try_partitionwise_join(root, rel1, rel2, joinrel, sjinfo, restrictlist);
+
+ populate_joinrel_uniquekeys(root, joinrel, rel1, rel2, restrictlist, sjinfo->jointype);
}
diff --git a/src/backend/optimizer/path/pathkeys.c b/src/backend/optimizer/path/pathkeys.c
index bd9a176d7d..139278829b 100644
--- a/src/backend/optimizer/path/pathkeys.c
+++ b/src/backend/optimizer/path/pathkeys.c
@@ -33,7 +33,6 @@ static bool pathkey_is_redundant(PathKey *new_pathkey, List *pathkeys);
static bool matches_boolean_partition_clause(RestrictInfo *rinfo,
RelOptInfo *partrel,
int partkeycol);
-static Var *find_var_for_subquery_tle(RelOptInfo *rel, TargetEntry *tle);
static bool right_merge_direction(PlannerInfo *root, PathKey *pathkey);
@@ -1035,7 +1034,7 @@ convert_subquery_pathkeys(PlannerInfo *root, RelOptInfo *rel,
* We need this to ensure that we don't return pathkeys describing values
* that are unavailable above the level of the subquery scan.
*/
-static Var *
+Var *
find_var_for_subquery_tle(RelOptInfo *rel, TargetEntry *tle)
{
ListCell *lc;
diff --git a/src/backend/optimizer/path/uniquekeys.c b/src/backend/optimizer/path/uniquekeys.c
new file mode 100644
index 0000000000..77ed2b2eff
--- /dev/null
+++ b/src/backend/optimizer/path/uniquekeys.c
@@ -0,0 +1,1135 @@
+/*-------------------------------------------------------------------------
+ *
+ * uniquekeys.c
+ * Utilities for matching and building unique keys
+ *
+ * Portions Copyright (c) 2020, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ * src/backend/optimizer/path/uniquekeys.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#include "nodes/makefuncs.h"
+#include "nodes/nodeFuncs.h"
+#include "optimizer/pathnode.h"
+#include "optimizer/paths.h"
+#include "optimizer/appendinfo.h"
+#include "optimizer/optimizer.h"
+#include "optimizer/tlist.h"
+#include "rewrite/rewriteManip.h"
+
+
+/*
+ * This struct is used to help populate_joinrel_uniquekeys.
+ *
+ * added_to_joinrel is true if a uniquekey (from outerrel or innerrel)
+ * has been added to joinrel.
+ * useful is true if the exprs of the uniquekey still appears in joinrel.
+ */
+typedef struct UniqueKeyContextData
+{
+ UniqueKey *uniquekey;
+ bool added_to_joinrel;
+ bool useful;
+} *UniqueKeyContext;
+
+static List *initililze_uniquecontext_for_joinrel(RelOptInfo *inputrel);
+static bool innerrel_keeps_unique(PlannerInfo *root,
+ RelOptInfo *outerrel,
+ RelOptInfo *innerrel,
+ List *restrictlist,
+ bool reverse);
+
+static List *get_exprs_from_uniqueindex(IndexOptInfo *unique_index,
+ List *const_exprs,
+ List *const_expr_opfamilies,
+ Bitmapset *used_varattrs,
+ bool *useful,
+ bool *multi_nullvals);
+static List *get_exprs_from_uniquekey(PlannerInfo *root,
+ RelOptInfo *joinrel,
+ RelOptInfo *rel1,
+ UniqueKey *ukey);
+static void add_uniquekey_for_onerow(RelOptInfo *rel);
+static bool add_combined_uniquekey(PlannerInfo *root,
+ RelOptInfo *joinrel,
+ RelOptInfo *outer_rel,
+ RelOptInfo *inner_rel,
+ UniqueKey *outer_ukey,
+ UniqueKey *inner_ukey,
+ JoinType jointype);
+
+/* Used for unique indexes checking for partitioned table */
+static bool index_constains_partkey(RelOptInfo *partrel, IndexOptInfo *ind);
+static IndexOptInfo *simple_copy_indexinfo_to_parent(PlannerInfo *root,
+ RelOptInfo *parentrel,
+ IndexOptInfo *from);
+static bool simple_indexinfo_equal(IndexOptInfo *ind1, IndexOptInfo *ind2);
+static void adjust_partition_unique_indexlist(PlannerInfo *root,
+ RelOptInfo *parentrel,
+ RelOptInfo *childrel,
+ List **global_unique_index);
+
+/* Helper function for grouped relation and distinct relation. */
+static void add_uniquekey_from_sortgroups(PlannerInfo *root,
+ RelOptInfo *rel,
+ List *sortgroups);
+
+/*
+ * populate_baserel_uniquekeys
+ * Populate 'baserel' uniquekeys list by looking at the rel's unique index
+ * and baserestrictinfo
+ */
+void
+populate_baserel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *baserel,
+ List *indexlist)
+{
+ ListCell *lc;
+ List *matched_uniq_indexes = NIL;
+
+ /* Attrs appears in rel->reltarget->exprs. */
+ Bitmapset *used_attrs = NULL;
+
+ List *const_exprs = NIL;
+ List *expr_opfamilies = NIL;
+
+ Assert(baserel->rtekind == RTE_RELATION);
+
+ foreach(lc, indexlist)
+ {
+ IndexOptInfo *ind = (IndexOptInfo *) lfirst(lc);
+ if (!ind->unique || !ind->immediate ||
+ (ind->indpred != NIL && !ind->predOK))
+ continue;
+ matched_uniq_indexes = lappend(matched_uniq_indexes, ind);
+ }
+
+ if (matched_uniq_indexes == NIL)
+ return;
+
+ /* Check which attrs is used in baserel->reltarget */
+ pull_varattnos((Node *)baserel->reltarget->exprs, baserel->relid, &used_attrs);
+
+ /* Check which attrno is used at a mergeable const filter */
+ foreach(lc, baserel->baserestrictinfo)
+ {
+ RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
+
+ if (rinfo->mergeopfamilies == NIL)
+ continue;
+
+ if (bms_is_empty(rinfo->left_relids))
+ {
+ const_exprs = lappend(const_exprs, get_rightop(rinfo->clause));
+ }
+ else if (bms_is_empty(rinfo->right_relids))
+ {
+ const_exprs = lappend(const_exprs, get_leftop(rinfo->clause));
+ }
+ else
+ continue;
+
+ expr_opfamilies = lappend(expr_opfamilies, rinfo->mergeopfamilies);
+ }
+
+ foreach(lc, matched_uniq_indexes)
+ {
+ bool multi_nullvals, useful;
+ List *exprs = get_exprs_from_uniqueindex(lfirst_node(IndexOptInfo, lc),
+ const_exprs,
+ expr_opfamilies,
+ used_attrs,
+ &useful,
+ &multi_nullvals);
+ if (useful)
+ {
+ if (exprs == NIL)
+ {
+ /* All the columns in Unique Index matched with a restrictinfo */
+ add_uniquekey_for_onerow(baserel);
+ return;
+ }
+ baserel->uniquekeys = lappend(baserel->uniquekeys,
+ makeUniqueKey(exprs, multi_nullvals));
+ }
+ }
+}
+
+
+/*
+ * populate_partitionedrel_uniquekeys
+ * The UniqueKey on partitionrel comes from 2 cases:
+ * 1). Only one partition is involved in this query, the unique key can be
+ * copied to parent rel from childrel.
+ * 2). There are some unique index which includes partition key and exists
+ * in all the related partitions.
+ * We never mind rule 2 if we hit rule 1.
+ */
+
+void
+populate_partitionedrel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *rel,
+ List *childrels)
+{
+ ListCell *lc;
+ List *global_uniq_indexlist = NIL;
+ RelOptInfo *childrel;
+ bool is_first = true;
+
+ Assert(IS_PARTITIONED_REL(rel));
+
+ if (childrels == NIL)
+ return;
+
+ /*
+ * If there is only one partition used in this query, the UniqueKey in childrel is
+ * still valid in parent level, but we need convert the format from child expr to
+ * parent expr.
+ */
+ if (list_length(childrels) == 1)
+ {
+ /* Check for Rule 1 */
+ RelOptInfo *childrel = linitial_node(RelOptInfo, childrels);
+ ListCell *lc;
+ Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
+ if (relation_is_onerow(childrel))
+ {
+ add_uniquekey_for_onerow(rel);
+ return;
+ }
+
+ foreach(lc, childrel->uniquekeys)
+ {
+ UniqueKey *ukey = lfirst_node(UniqueKey, lc);
+ AppendRelInfo *appinfo = find_appinfo_by_child(root, childrel->relid);
+ List *parent_exprs = NIL;
+ bool can_reuse = true;
+ ListCell *lc2;
+ foreach(lc2, ukey->exprs)
+ {
+ Var *var = (Var *)lfirst(lc2);
+ /*
+ * If the expr comes from a expression, it is hard to build the expression
+ * in parent so ignore that case for now.
+ */
+ if(!IsA(var, Var))
+ {
+ can_reuse = false;
+ break;
+ }
+ /* Convert it to parent var */
+ parent_exprs = lappend(parent_exprs, find_parent_var(appinfo, var));
+ }
+ if (can_reuse)
+ rel->uniquekeys = lappend(rel->uniquekeys,
+ makeUniqueKey(parent_exprs,
+ ukey->multi_nullvals));
+ }
+ }
+ else
+ {
+ /* Check for rule 2 */
+ childrel = linitial_node(RelOptInfo, childrels);
+ foreach(lc, childrel->indexlist)
+ {
+ IndexOptInfo *ind = lfirst(lc);
+ IndexOptInfo *modified_index;
+ if (!ind->unique || !ind->immediate ||
+ (ind->indpred != NIL && !ind->predOK))
+ continue;
+
+ /*
+ * During simple_copy_indexinfo_to_parent, we need to convert var from
+ * child var to parent var, index on expression is too complex to handle.
+ * so ignore it for now.
+ */
+ if (ind->indexprs != NIL)
+ continue;
+
+ modified_index = simple_copy_indexinfo_to_parent(root, rel, ind);
+ /*
+ * If the unique index doesn't contain partkey, then it is unique
+ * on this partition only, so it is useless for us.
+ */
+ if (!index_constains_partkey(rel, modified_index))
+ continue;
+
+ global_uniq_indexlist = lappend(global_uniq_indexlist, modified_index);
+ }
+
+ if (global_uniq_indexlist != NIL)
+ {
+ foreach(lc, childrels)
+ {
+ RelOptInfo *child = lfirst(lc);
+ if (is_first)
+ {
+ is_first = false;
+ continue;
+ }
+ adjust_partition_unique_indexlist(root, rel, child, &global_uniq_indexlist);
+ }
+ /* Now we have a list of unique index which are exactly same on all childrels,
+ * Set the UniqueKey just like it is non-partition table
+ */
+ populate_baserel_uniquekeys(root, rel, global_uniq_indexlist);
+ }
+ }
+}
+
+
+/*
+ * populate_distinctrel_uniquekeys
+ */
+void
+populate_distinctrel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *inputrel,
+ RelOptInfo *distinctrel)
+{
+ /* The unique key before the distinct is still valid. */
+ distinctrel->uniquekeys = list_copy(inputrel->uniquekeys);
+ add_uniquekey_from_sortgroups(root, distinctrel, root->parse->distinctClause);
+}
+
+/*
+ * populate_grouprel_uniquekeys
+ */
+void
+populate_grouprel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *grouprel,
+ RelOptInfo *inputrel)
+
+{
+ Query *parse = root->parse;
+ bool input_ukey_added = false;
+ ListCell *lc;
+
+ if (relation_is_onerow(inputrel))
+ {
+ add_uniquekey_for_onerow(grouprel);
+ return;
+ }
+ if (parse->groupingSets)
+ return;
+
+ /* A Normal group by without grouping set. */
+ if (parse->groupClause)
+ {
+ /*
+ * Current even the groupby clause is Unique already, but if query has aggref
+ * We have to create grouprel still. To keep the UnqiueKey short, we will check
+ * the UniqueKey of input_rel still valid, if so we reuse it.
+ */
+ foreach(lc, inputrel->uniquekeys)
+ {
+ UniqueKey *ukey = lfirst_node(UniqueKey, lc);
+ if (list_is_subset(ukey->exprs, grouprel->reltarget->exprs))
+ {
+ grouprel->uniquekeys = lappend(grouprel->uniquekeys,
+ ukey);
+ input_ukey_added = true;
+ }
+ }
+ if (!input_ukey_added)
+ /*
+ * group by clause must be a super-set of grouprel->reltarget->exprs except the
+ * aggregation expr, so if such exprs is unique already, no bother to generate
+ * new uniquekey for group by exprs.
+ */
+ add_uniquekey_from_sortgroups(root,
+ grouprel,
+ root->parse->groupClause);
+ }
+ else
+ /* It has aggregation but without a group by, so only one row returned */
+ add_uniquekey_for_onerow(grouprel);
+}
+
+/*
+ * simple_copy_uniquekeys
+ * Using a function for the one-line code makes us easy to check where we simply
+ * copied the uniquekey.
+ */
+void
+simple_copy_uniquekeys(RelOptInfo *oldrel,
+ RelOptInfo *newrel)
+{
+ newrel->uniquekeys = oldrel->uniquekeys;
+}
+
+/*
+ * populate_unionrel_uniquekeys
+ */
+void
+populate_unionrel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *unionrel)
+{
+ ListCell *lc;
+ List *exprs = NIL;
+
+ Assert(unionrel->uniquekeys == NIL);
+
+ foreach(lc, unionrel->reltarget->exprs)
+ {
+ exprs = lappend(exprs, lfirst(lc));
+ }
+
+ if (exprs == NIL)
+ /* SQL: select union select; is valid, we need to handle it here. */
+ add_uniquekey_for_onerow(unionrel);
+ else
+ unionrel->uniquekeys = lappend(unionrel->uniquekeys,
+ makeUniqueKey(exprs,false));
+
+}
+
+/*
+ * populate_joinrel_uniquekeys
+ *
+ * populate uniquekeys for joinrel. We will check each relation to see if its
+ * UniqueKey is still valid via innerrel_keeps_unique, if so, we add it to
+ * joinrel. The multi_nullvals field will be changed to true for some outer
+ * join cases and one-row UniqueKey needs to be converted to normal UniqueKey
+ * for the same case as well.
+ * For the uniquekey in either baserel which can't be unique after join, we still
+ * check to see if combination of UniqueKeys from both side is still useful for us.
+ * if yes, we add it to joinrel as well.
+ */
+void
+populate_joinrel_uniquekeys(PlannerInfo *root, RelOptInfo *joinrel,
+ RelOptInfo *outerrel, RelOptInfo *innerrel,
+ List *restrictlist, JoinType jointype)
+{
+ ListCell *lc, *lc2;
+ List *clause_list = NIL;
+ List *outerrel_ukey_ctx;
+ List *innerrel_ukey_ctx;
+ bool inner_onerow, outer_onerow;
+ bool mergejoin_allowed;
+
+ /* Care about the outerrel relation only for SEMI/ANTI join */
+ if (jointype == JOIN_SEMI || jointype == JOIN_ANTI)
+ {
+ foreach(lc, outerrel->uniquekeys)
+ {
+ UniqueKey *uniquekey = lfirst_node(UniqueKey, lc);
+ if (list_is_subset(uniquekey->exprs, joinrel->reltarget->exprs))
+ joinrel->uniquekeys = lappend(joinrel->uniquekeys, uniquekey);
+ }
+ return;
+ }
+
+ Assert(jointype == JOIN_LEFT || jointype == JOIN_FULL || jointype == JOIN_INNER);
+
+ /* Fast path */
+ if (innerrel->uniquekeys == NIL || outerrel->uniquekeys == NIL)
+ return;
+
+ inner_onerow = relation_is_onerow(innerrel);
+ outer_onerow = relation_is_onerow(outerrel);
+
+ outerrel_ukey_ctx = initililze_uniquecontext_for_joinrel(outerrel);
+ innerrel_ukey_ctx = initililze_uniquecontext_for_joinrel(innerrel);
+
+ clause_list = select_mergejoin_clauses(root, joinrel, outerrel, innerrel,
+ restrictlist, jointype,
+ &mergejoin_allowed);
+
+ if (innerrel_keeps_unique(root, innerrel, outerrel, clause_list, true /* reverse */))
+ {
+ bool outer_impact = jointype == JOIN_FULL;
+ foreach(lc, outerrel_ukey_ctx)
+ {
+ UniqueKeyContext ctx = (UniqueKeyContext)lfirst(lc);
+
+ if (!list_is_subset(ctx->uniquekey->exprs, joinrel->reltarget->exprs))
+ {
+ ctx->useful = false;
+ continue;
+ }
+
+ /* Outer relation has one row, and the unique key is not duplicated after join,
+ * the joinrel will still has one row unless the jointype == JOIN_FULL.
+ */
+ if (outer_onerow && !outer_impact)
+ {
+ add_uniquekey_for_onerow(joinrel);
+ return;
+ }
+ else if (outer_onerow)
+ {
+ /*
+ * The onerow outerrel becomes multi rows and multi_nullvals
+ * will be changed to true. We also need to set the exprs correctly since it
+ * can't be NIL any more.
+ */
+ ListCell *lc2;
+ foreach(lc2, get_exprs_from_uniquekey(root, joinrel, outerrel, NULL))
+ {
+ joinrel->uniquekeys = lappend(joinrel->uniquekeys,
+ makeUniqueKey(lfirst(lc2), true));
+ }
+ }
+ else
+ {
+ if (!ctx->uniquekey->multi_nullvals && outer_impact)
+ /* Change multi_nullvals to true due to the full join. */
+ joinrel->uniquekeys = lappend(joinrel->uniquekeys,
+ makeUniqueKey(ctx->uniquekey->exprs, true));
+ else
+ /* Just reuse it */
+ joinrel->uniquekeys = lappend(joinrel->uniquekeys,
+ ctx->uniquekey);
+ }
+ ctx->added_to_joinrel = true;
+ }
+ }
+
+ if (innerrel_keeps_unique(root, outerrel, innerrel, clause_list, false))
+ {
+ bool outer_impact = jointype == JOIN_FULL || jointype == JOIN_LEFT;;
+
+ foreach(lc, innerrel_ukey_ctx)
+ {
+ UniqueKeyContext ctx = (UniqueKeyContext)lfirst(lc);
+
+ if (!list_is_subset(ctx->uniquekey->exprs, joinrel->reltarget->exprs))
+ {
+ ctx->useful = false;
+ continue;
+ }
+
+ if (inner_onerow && !outer_impact)
+ {
+ add_uniquekey_for_onerow(joinrel);
+ return;
+ }
+ else if (inner_onerow)
+ {
+ ListCell *lc2;
+ foreach(lc2, get_exprs_from_uniquekey(root, joinrel, innerrel, NULL))
+ {
+ joinrel->uniquekeys = lappend(joinrel->uniquekeys,
+ makeUniqueKey(lfirst(lc2), true));
+ }
+ }
+ else
+ {
+ if (!ctx->uniquekey->multi_nullvals && outer_impact)
+ /* Need to change multi_nullvals to true due to the outer join. */
+ joinrel->uniquekeys = lappend(joinrel->uniquekeys,
+ makeUniqueKey(ctx->uniquekey->exprs,
+ true));
+ else
+ joinrel->uniquekeys = lappend(joinrel->uniquekeys,
+ ctx->uniquekey);
+
+ }
+ ctx->added_to_joinrel = true;
+ }
+ }
+
+ /*
+ * The combination of the UniqueKey from both sides is unique as well regardless
+ * of join type, but no bother to add it if its subset has been added to joinrel
+ * already or it is not useful for the joinrel.
+ */
+ foreach(lc, outerrel_ukey_ctx)
+ {
+ UniqueKeyContext ctx1 = (UniqueKeyContext) lfirst(lc);
+ if (ctx1->added_to_joinrel || !ctx1->useful)
+ continue;
+ foreach(lc2, innerrel_ukey_ctx)
+ {
+ UniqueKeyContext ctx2 = (UniqueKeyContext) lfirst(lc2);
+ if (ctx2->added_to_joinrel || !ctx2->useful)
+ continue;
+ if (add_combined_uniquekey(root, joinrel, outerrel, innerrel,
+ ctx1->uniquekey, ctx2->uniquekey,
+ jointype))
+ /* If we set a onerow UniqueKey to joinrel, we don't need other. */
+ return;
+ }
+ }
+}
+
+
+/*
+ * convert_subquery_uniquekeys
+ *
+ * Covert the UniqueKey in subquery to outer relation.
+ */
+void convert_subquery_uniquekeys(PlannerInfo *root,
+ RelOptInfo *currel,
+ RelOptInfo *sub_final_rel)
+{
+ ListCell *lc;
+
+ if (sub_final_rel->uniquekeys == NIL)
+ return;
+
+ if (relation_is_onerow(sub_final_rel))
+ {
+ add_uniquekey_for_onerow(currel);
+ return;
+ }
+
+ Assert(currel->subroot != NULL);
+
+ foreach(lc, sub_final_rel->uniquekeys)
+ {
+ UniqueKey *ukey = lfirst_node(UniqueKey, lc);
+ ListCell *lc;
+ List *exprs = NIL;
+ bool ukey_useful = true;
+
+ /* One row case is handled above */
+ Assert(ukey->exprs != NIL);
+ foreach(lc, ukey->exprs)
+ {
+ Var *var;
+ TargetEntry *tle = tlist_member(lfirst(lc),
+ currel->subroot->processed_tlist);
+ if (tle == NULL)
+ {
+ ukey_useful = false;
+ break;
+ }
+ var = find_var_for_subquery_tle(currel, tle);
+ if (var == NULL)
+ {
+ ukey_useful = false;
+ break;
+ }
+ exprs = lappend(exprs, var);
+ }
+
+ if (ukey_useful)
+ currel->uniquekeys = lappend(currel->uniquekeys,
+ makeUniqueKey(exprs,
+ ukey->multi_nullvals));
+
+ }
+}
+
+/*
+ * innerrel_keeps_unique
+ *
+ * Check if Unique key of the innerrel is valid after join. innerrel's UniqueKey
+ * will be still valid if innerrel's any-column mergeop outrerel's uniquekey
+ * exists in clause_list.
+ *
+ * Note: the clause_list must be a list of mergeable restrictinfo already.
+ */
+static bool
+innerrel_keeps_unique(PlannerInfo *root,
+ RelOptInfo *outerrel,
+ RelOptInfo *innerrel,
+ List *clause_list,
+ bool reverse)
+{
+ ListCell *lc, *lc2, *lc3;
+
+ if (outerrel->uniquekeys == NIL || innerrel->uniquekeys == NIL)
+ return false;
+
+ /* Check if there is outerrel's uniquekey in mergeable clause. */
+ foreach(lc, outerrel->uniquekeys)
+ {
+ List *outer_uq_exprs = lfirst_node(UniqueKey, lc)->exprs;
+ bool clauselist_matchs_all_exprs = true;
+ foreach(lc2, outer_uq_exprs)
+ {
+ Node *outer_uq_expr = lfirst(lc2);
+ bool find_uq_expr_in_clauselist = false;
+ foreach(lc3, clause_list)
+ {
+ RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc3);
+ Node *outer_expr;
+ if (reverse)
+ outer_expr = rinfo->outer_is_left ? get_rightop(rinfo->clause) : get_leftop(rinfo->clause);
+ else
+ outer_expr = rinfo->outer_is_left ? get_leftop(rinfo->clause) : get_rightop(rinfo->clause);
+ if (equal(outer_expr, outer_uq_expr))
+ {
+ find_uq_expr_in_clauselist = true;
+ break;
+ }
+ }
+ if (!find_uq_expr_in_clauselist)
+ {
+ /* No need to check the next exprs in the current uniquekey */
+ clauselist_matchs_all_exprs = false;
+ break;
+ }
+ }
+
+ if (clauselist_matchs_all_exprs)
+ return true;
+ }
+ return false;
+}
+
+
+/*
+ * relation_is_onerow
+ * Check if it is a one-row relation by checking UniqueKey.
+ */
+bool
+relation_is_onerow(RelOptInfo *rel)
+{
+ UniqueKey *ukey;
+ if (rel->uniquekeys == NIL)
+ return false;
+ ukey = linitial_node(UniqueKey, rel->uniquekeys);
+ return ukey->exprs == NIL && list_length(rel->uniquekeys) == 1;
+}
+
+/*
+ * relation_has_uniquekeys_for
+ * Returns true if we have proofs that 'rel' cannot return multiple rows with
+ * the same values in each of 'exprs'. Otherwise returns false.
+ */
+bool
+relation_has_uniquekeys_for(PlannerInfo *root, RelOptInfo *rel,
+ List *exprs, bool allow_multinulls)
+{
+ ListCell *lc;
+
+ /*
+ * For UniqueKey->onerow case, the uniquekey->exprs is empty as well
+ * so we can't rely on list_is_subset to handle this special cases
+ */
+ if (exprs == NIL)
+ return false;
+
+ foreach(lc, rel->uniquekeys)
+ {
+ UniqueKey *ukey = lfirst_node(UniqueKey, lc);
+ if (ukey->multi_nullvals && !allow_multinulls)
+ continue;
+ if (list_is_subset(ukey->exprs, exprs))
+ return true;
+ }
+ return false;
+}
+
+
+/*
+ * get_exprs_from_uniqueindex
+ *
+ * Return a list of exprs which is unique. set useful to false if this
+ * unique index is not useful for us.
+ */
+static List *
+get_exprs_from_uniqueindex(IndexOptInfo *unique_index,
+ List *const_exprs,
+ List *const_expr_opfamilies,
+ Bitmapset *used_varattrs,
+ bool *useful,
+ bool *multi_nullvals)
+{
+ List *exprs = NIL;
+ ListCell *indexpr_item;
+ int c = 0;
+
+ *useful = true;
+ *multi_nullvals = false;
+
+ indexpr_item = list_head(unique_index->indexprs);
+ for(c = 0; c < unique_index->ncolumns; c++)
+ {
+ int attr = unique_index->indexkeys[c];
+ Expr *expr;
+ bool matched_const = false;
+ ListCell *lc1, *lc2;
+
+ if(attr > 0)
+ {
+ expr = list_nth_node(TargetEntry, unique_index->indextlist, c)->expr;
+ }
+ else if (attr == 0)
+ {
+ /* Expression index */
+ expr = lfirst(indexpr_item);
+ indexpr_item = lnext(unique_index->indexprs, indexpr_item);
+ }
+ else /* attr < 0 */
+ {
+ /* Index on system column is not supported */
+ Assert(false);
+ }
+
+ /*
+ * Check index_col = Const case with regarding to opfamily checking
+ * If we can remove the index_col from the final UniqueKey->exprs.
+ */
+ forboth(lc1, const_exprs, lc2, const_expr_opfamilies)
+ {
+ if (list_member_oid((List *)lfirst(lc2), unique_index->opfamily[c])
+ && match_index_to_operand((Node *) lfirst(lc1), c, unique_index))
+ {
+ matched_const = true;
+ break;
+ }
+ }
+
+ if (matched_const)
+ continue;
+
+ /* Check if the indexed expr is used in rel */
+ if (attr > 0)
+ {
+ /*
+ * Normal Indexed column, if the col is not used, then the index is useless
+ * for uniquekey.
+ */
+ attr -= FirstLowInvalidHeapAttributeNumber;
+
+ if (!bms_is_member(attr, used_varattrs))
+ {
+ *useful = false;
+ break;
+ }
+ }
+ else if (!list_member(unique_index->rel->reltarget->exprs, expr))
+ {
+ /* Expression index but the expression is not used in rel */
+ *useful = false;
+ break;
+ }
+
+ /* check not null property. */
+ if (attr == 0)
+ {
+ /* We never know if a expression yields null or not */
+ *multi_nullvals = true;
+ }
+ else if (!bms_is_member(attr, unique_index->rel->notnullattrs)
+ && !bms_is_member(0 - FirstLowInvalidHeapAttributeNumber,
+ unique_index->rel->notnullattrs))
+ {
+ *multi_nullvals = true;
+ }
+
+ exprs = lappend(exprs, expr);
+ }
+ return exprs;
+}
+
+
+/*
+ * add_uniquekey_for_onerow
+ * If we are sure that the relation only returns one row, then all the columns
+ * are unique. However we don't need to create UniqueKey for every column, we
+ * just set exprs = NIL and overwrites all the other UniqueKey on this RelOptInfo
+ * since this one has strongest semantics.
+ */
+void
+add_uniquekey_for_onerow(RelOptInfo *rel)
+{
+ /*
+ * We overwrite the previous UniqueKey on purpose since this one has the
+ * strongest semantic.
+ */
+ rel->uniquekeys = list_make1(makeUniqueKey(NIL, false));
+}
+
+
+/*
+ * initililze_uniquecontext_for_joinrel
+ * Return a List of UniqueKeyContext for an inputrel
+ */
+static List *
+initililze_uniquecontext_for_joinrel(RelOptInfo *inputrel)
+{
+ List *res = NIL;
+ ListCell *lc;
+ foreach(lc, inputrel->uniquekeys)
+ {
+ UniqueKeyContext context;
+ context = palloc(sizeof(struct UniqueKeyContextData));
+ context->uniquekey = lfirst_node(UniqueKey, lc);
+ context->added_to_joinrel = false;
+ context->useful = true;
+ res = lappend(res, context);
+ }
+ return res;
+}
+
+
+/*
+ * get_exprs_from_uniquekey
+ * Unify the way of get List of exprs from a one-row UniqueKey or
+ * normal UniqueKey. for the onerow case, every expr in rel1 is a valid
+ * UniqueKey. Return a List of exprs.
+ *
+ * rel1: The relation which you want to get the exprs.
+ * ukey: The UniqueKey you want to get the exprs.
+ */
+static List *
+get_exprs_from_uniquekey(PlannerInfo *root, RelOptInfo *joinrel,
+ RelOptInfo *rel1, UniqueKey *ukey)
+{
+ ListCell *lc;
+ bool onerow = rel1 != NULL && relation_is_onerow(rel1);
+
+ List *res = NIL;
+ Assert(onerow || ukey);
+ if (onerow)
+ {
+ /* Only cares about the exprs still exist in joinrel */
+ foreach(lc, joinrel->reltarget->exprs)
+ {
+ Bitmapset *relids = pull_varnos(root, lfirst(lc));
+ if (bms_is_subset(relids, rel1->relids))
+ {
+ res = lappend(res, list_make1(lfirst(lc)));
+ }
+ }
+ }
+ else
+ {
+ res = list_make1(ukey->exprs);
+ }
+ return res;
+}
+
+/*
+ * Partitioned table Unique Keys.
+ * The partition table unique key is maintained as:
+ * 1. The index must be unique as usual.
+ * 2. The index must contains partition key.
+ * 3. The index must exist on all the child rel. see simple_indexinfo_equal for
+ * how we compare it.
+ */
+
+/*
+ * index_constains_partkey
+ * return true if the index contains the partiton key.
+ */
+static bool
+index_constains_partkey(RelOptInfo *partrel, IndexOptInfo *ind)
+{
+ ListCell *lc;
+ int i;
+ Assert(IS_PARTITIONED_REL(partrel));
+ Assert(partrel->part_scheme->partnatts > 0);
+
+ for(i = 0; i < partrel->part_scheme->partnatts; i++)
+ {
+ Node *part_expr = linitial(partrel->partexprs[i]);
+ bool found_in_index = false;
+ foreach(lc, ind->indextlist)
+ {
+ Expr *index_expr = lfirst_node(TargetEntry, lc)->expr;
+ if (equal(index_expr, part_expr))
+ {
+ found_in_index = true;
+ break;
+ }
+ }
+ if (!found_in_index)
+ return false;
+ }
+ return true;
+}
+
+/*
+ * simple_indexinfo_equal
+ *
+ * Used to check if the 2 index is same as each other. The index here
+ * is COPIED from childrel and did some tiny changes(see
+ * simple_copy_indexinfo_to_parent)
+ */
+static bool
+simple_indexinfo_equal(IndexOptInfo *ind1, IndexOptInfo *ind2)
+{
+ Size oid_cmp_len = sizeof(Oid) * ind1->ncolumns;
+
+ return ind1->ncolumns == ind2->ncolumns &&
+ ind1->unique == ind2->unique &&
+ memcmp(ind1->indexkeys, ind2->indexkeys, sizeof(int) * ind1->ncolumns) == 0 &&
+ memcmp(ind1->opfamily, ind2->opfamily, oid_cmp_len) == 0 &&
+ memcmp(ind1->opcintype, ind2->opcintype, oid_cmp_len) == 0 &&
+ memcmp(ind1->sortopfamily, ind2->sortopfamily, oid_cmp_len) == 0 &&
+ equal(get_tlist_exprs(ind1->indextlist, true),
+ get_tlist_exprs(ind2->indextlist, true));
+}
+
+
+/*
+ * The below macros are used for simple_copy_indexinfo_to_parent which is so
+ * customized that I don't want to put it to copyfuncs.c. So copy it here.
+ */
+#define COPY_POINTER_FIELD(fldname, sz) \
+ do { \
+ Size _size = (sz); \
+ newnode->fldname = palloc(_size); \
+ memcpy(newnode->fldname, from->fldname, _size); \
+ } while (0)
+
+#define COPY_NODE_FIELD(fldname) \
+ (newnode->fldname = copyObjectImpl(from->fldname))
+
+#define COPY_SCALAR_FIELD(fldname) \
+ (newnode->fldname = from->fldname)
+
+
+/*
+ * simple_copy_indexinfo_to_parent (from partition)
+ * Copy the IndexInfo from child relation to parent relation with some modification,
+ * which is used to test:
+ * 1. If the same index exists in all the childrels.
+ * 2. If the parentrel->reltarget/basicrestrict info matches this index.
+ */
+static IndexOptInfo *
+simple_copy_indexinfo_to_parent(PlannerInfo *root,
+ RelOptInfo *parentrel,
+ IndexOptInfo *from)
+{
+ IndexOptInfo *newnode = makeNode(IndexOptInfo);
+ AppendRelInfo *appinfo = find_appinfo_by_child(root, from->rel->relid);
+ ListCell *lc;
+ int idx = 0;
+
+ COPY_SCALAR_FIELD(ncolumns);
+ COPY_SCALAR_FIELD(nkeycolumns);
+ COPY_SCALAR_FIELD(unique);
+ COPY_SCALAR_FIELD(immediate);
+ /* We just need to know if it is NIL or not */
+ COPY_SCALAR_FIELD(indpred);
+ COPY_SCALAR_FIELD(predOK);
+ COPY_POINTER_FIELD(indexkeys, from->ncolumns * sizeof(int));
+ COPY_POINTER_FIELD(indexcollations, from->ncolumns * sizeof(Oid));
+ COPY_POINTER_FIELD(opfamily, from->ncolumns * sizeof(Oid));
+ COPY_POINTER_FIELD(opcintype, from->ncolumns * sizeof(Oid));
+ COPY_POINTER_FIELD(sortopfamily, from->ncolumns * sizeof(Oid));
+ COPY_NODE_FIELD(indextlist);
+
+ /* Convert index exprs on child expr to expr on parent */
+ foreach(lc, newnode->indextlist)
+ {
+ TargetEntry *tle = lfirst_node(TargetEntry, lc);
+ /* Index on expression is ignored */
+ Assert(IsA(tle->expr, Var));
+ tle->expr = (Expr *) find_parent_var(appinfo, (Var *) tle->expr);
+ newnode->indexkeys[idx] = castNode(Var, tle->expr)->varattno;
+ idx++;
+ }
+ newnode->rel = parentrel;
+ return newnode;
+}
+
+/*
+ * adjust_partition_unique_indexlist
+ *
+ * global_unique_indexes: At the beginning, it contains the copy & modified
+ * unique index from the first partition. And then check if each index in it still
+ * exists in the following partitions. If no, remove it. at last, it has an
+ * index list which exists in all the partitions.
+ */
+static void
+adjust_partition_unique_indexlist(PlannerInfo *root,
+ RelOptInfo *parentrel,
+ RelOptInfo *childrel,
+ List **global_unique_indexes)
+{
+ ListCell *lc, *lc2;
+ foreach(lc, *global_unique_indexes)
+ {
+ IndexOptInfo *g_ind = lfirst_node(IndexOptInfo, lc);
+ bool found_in_child = false;
+
+ foreach(lc2, childrel->indexlist)
+ {
+ IndexOptInfo *p_ind = lfirst_node(IndexOptInfo, lc2);
+ IndexOptInfo *p_ind_copy;
+ if (!p_ind->unique || !p_ind->immediate ||
+ (p_ind->indpred != NIL && !p_ind->predOK))
+ continue;
+ p_ind_copy = simple_copy_indexinfo_to_parent(root, parentrel, p_ind);
+ if (simple_indexinfo_equal(p_ind_copy, g_ind))
+ {
+ found_in_child = true;
+ break;
+ }
+ }
+ if (!found_in_child)
+ /* The index doesn't exist in childrel, remove it from global_unique_indexes */
+ *global_unique_indexes = foreach_delete_current(*global_unique_indexes, lc);
+ }
+}
+
+/* Helper function for groupres/distinctrel */
+static void
+add_uniquekey_from_sortgroups(PlannerInfo *root, RelOptInfo *rel, List *sortgroups)
+{
+ Query *parse = root->parse;
+ List *exprs;
+
+ /*
+ * XXX: If there are some vars which is not in current levelsup, the semantic is
+ * imprecise, should we avoid it or not? levelsup = 1 is just a demo, maybe we need to
+ * check every level other than 0, if so, looks we have to write another
+ * pull_var_walker.
+ */
+ List *upper_vars = pull_vars_of_level((Node*)sortgroups, 1);
+
+ if (upper_vars != NIL)
+ return;
+
+ exprs = get_sortgrouplist_exprs(sortgroups, parse->targetList);
+ rel->uniquekeys = lappend(rel->uniquekeys,
+ makeUniqueKey(exprs,
+ false /* sortgroupclause can't be multi_nullvals */));
+}
+
+
+/*
+ * add_combined_uniquekey
+ * The combination of both UniqueKeys is a valid UniqueKey for joinrel no matter
+ * the jointype.
+ */
+bool
+add_combined_uniquekey(PlannerInfo *root,
+ RelOptInfo *joinrel,
+ RelOptInfo *outer_rel,
+ RelOptInfo *inner_rel,
+ UniqueKey *outer_ukey,
+ UniqueKey *inner_ukey,
+ JoinType jointype)
+{
+
+ ListCell *lc1, *lc2;
+
+ /* Either side has multi_nullvals or we have outer join,
+ * the combined UniqueKey has multi_nullvals */
+ bool multi_nullvals = outer_ukey->multi_nullvals ||
+ inner_ukey->multi_nullvals || IS_OUTER_JOIN(jointype);
+
+ /* The only case we can get onerow joinrel after join */
+ if (relation_is_onerow(outer_rel)
+ && relation_is_onerow(inner_rel)
+ && jointype == JOIN_INNER)
+ {
+ add_uniquekey_for_onerow(joinrel);
+ return true;
+ }
+
+ foreach(lc1, get_exprs_from_uniquekey(root, joinrel, outer_rel, outer_ukey))
+ {
+ foreach(lc2, get_exprs_from_uniquekey(root, joinrel, inner_rel, inner_ukey))
+ {
+ List *exprs = list_concat_copy(lfirst_node(List, lc1), lfirst_node(List, lc2));
+ joinrel->uniquekeys = lappend(joinrel->uniquekeys,
+ makeUniqueKey(exprs,
+ multi_nullvals));
+ }
+ }
+ return false;
+}
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index 424d25cbd5..8d8e493f5c 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -2387,6 +2387,8 @@ grouping_planner(PlannerInfo *root, bool inheritance_update,
add_path(final_rel, path);
}
+ simple_copy_uniquekeys(current_rel, final_rel);
+
/*
* Generate partial paths for final_rel, too, if outer query levels might
* be able to make use of them.
@@ -3899,6 +3901,8 @@ create_grouping_paths(PlannerInfo *root,
}
set_cheapest(grouped_rel);
+
+ populate_grouprel_uniquekeys(root, grouped_rel, input_rel);
return grouped_rel;
}
@@ -4621,7 +4625,7 @@ create_window_paths(PlannerInfo *root,
/* Now choose the best path(s) */
set_cheapest(window_rel);
-
+ simple_copy_uniquekeys(input_rel, window_rel);
return window_rel;
}
@@ -4935,7 +4939,7 @@ create_distinct_paths(PlannerInfo *root,
/* Now choose the best path(s) */
set_cheapest(distinct_rel);
-
+ populate_distinctrel_uniquekeys(root, input_rel, distinct_rel);
return distinct_rel;
}
@@ -5196,6 +5200,8 @@ create_ordered_paths(PlannerInfo *root,
*/
Assert(ordered_rel->pathlist != NIL);
+ simple_copy_uniquekeys(input_rel, ordered_rel);
+
return ordered_rel;
}
@@ -6073,6 +6079,9 @@ adjust_paths_for_srfs(PlannerInfo *root, RelOptInfo *rel,
if (list_length(targets) == 1)
return;
+ /* UniqueKey is not valid after handling the SRF. */
+ rel->uniquekeys = NIL;
+
/*
* Stack SRF-evaluation nodes atop each path for the rel.
*
diff --git a/src/backend/optimizer/prep/prepunion.c b/src/backend/optimizer/prep/prepunion.c
index becdcbb872..b7626545bf 100644
--- a/src/backend/optimizer/prep/prepunion.c
+++ b/src/backend/optimizer/prep/prepunion.c
@@ -689,6 +689,8 @@ generate_union_paths(SetOperationStmt *op, PlannerInfo *root,
/* Undo effects of possibly forcing tuple_fraction to 0 */
root->tuple_fraction = save_fraction;
+ /* Add the UniqueKeys */
+ populate_unionrel_uniquekeys(root, result_rel);
return result_rel;
}
diff --git a/src/backend/optimizer/util/appendinfo.c b/src/backend/optimizer/util/appendinfo.c
index 86922a273c..6817c9c787 100644
--- a/src/backend/optimizer/util/appendinfo.c
+++ b/src/backend/optimizer/util/appendinfo.c
@@ -746,3 +746,47 @@ find_appinfos_by_relids(PlannerInfo *root, Relids relids, int *nappinfos)
}
return appinfos;
}
+
+/*
+ * find_appinfo_by_child
+ *
+ */
+AppendRelInfo *
+find_appinfo_by_child(PlannerInfo *root, Index child_index)
+{
+ ListCell *lc;
+ foreach(lc, root->append_rel_list)
+ {
+ AppendRelInfo *appinfo = lfirst_node(AppendRelInfo, lc);
+ if (appinfo->child_relid == child_index)
+ return appinfo;
+ }
+ elog(ERROR, "parent relation cant be found");
+ return NULL;
+}
+
+/*
+ * find_parent_var
+ *
+ */
+Var *
+find_parent_var(AppendRelInfo *appinfo, Var *child_var)
+{
+ ListCell *lc;
+ Var *res = NULL;
+ Index attno = 1;
+ foreach(lc, appinfo->translated_vars)
+ {
+ Node *child_node = lfirst(lc);
+ if (equal(child_node, child_var))
+ {
+ res = copyObject(child_var);
+ res->varattno = attno;
+ res->varno = appinfo->parent_relid;
+ }
+ attno++;
+ }
+ if (res == NULL)
+ elog(ERROR, "parent var can't be found.");
+ return res;
+}
diff --git a/src/backend/optimizer/util/inherit.c b/src/backend/optimizer/util/inherit.c
index be1c9ddd96..3eec1f4d74 100644
--- a/src/backend/optimizer/util/inherit.c
+++ b/src/backend/optimizer/util/inherit.c
@@ -736,6 +736,7 @@ apply_child_basequals(PlannerInfo *root, RelOptInfo *parentrel,
{
Node *onecq = (Node *) lfirst(lc2);
bool pseudoconstant;
+ RestrictInfo *child_rinfo;
/* check for pseudoconstant (no Vars or volatile functions) */
pseudoconstant =
@@ -747,14 +748,15 @@ apply_child_basequals(PlannerInfo *root, RelOptInfo *parentrel,
root->hasPseudoConstantQuals = true;
}
/* reconstitute RestrictInfo with appropriate properties */
- childquals = lappend(childquals,
- make_restrictinfo(root,
- (Expr *) onecq,
- rinfo->is_pushed_down,
- rinfo->outerjoin_delayed,
- pseudoconstant,
- rinfo->security_level,
- NULL, NULL, NULL));
+ child_rinfo = make_restrictinfo(root,
+ (Expr *) onecq,
+ rinfo->is_pushed_down,
+ rinfo->outerjoin_delayed,
+ pseudoconstant,
+ rinfo->security_level,
+ NULL, NULL, NULL);
+ child_rinfo->mergeopfamilies = rinfo->mergeopfamilies;
+ childquals = lappend(childquals, child_rinfo);
/* track minimum security level among child quals */
cq_min_security = Min(cq_min_security, rinfo->security_level);
}
diff --git a/src/include/nodes/makefuncs.h b/src/include/nodes/makefuncs.h
index 48a7ebfe45..4fe1824eb0 100644
--- a/src/include/nodes/makefuncs.h
+++ b/src/include/nodes/makefuncs.h
@@ -16,6 +16,7 @@
#include "nodes/execnodes.h"
#include "nodes/parsenodes.h"
+#include "nodes/pathnodes.h"
extern A_Expr *makeA_Expr(A_Expr_Kind kind, List *name,
@@ -106,4 +107,6 @@ extern GroupingSet *makeGroupingSet(GroupingSetKind kind, List *content, int loc
extern VacuumRelation *makeVacuumRelation(RangeVar *relation, Oid oid, List *va_cols);
+extern UniqueKey* makeUniqueKey(List *exprs, bool multi_nullvals);
+
#endif /* MAKEFUNC_H */
diff --git a/src/include/nodes/nodes.h b/src/include/nodes/nodes.h
index e22df890ef..3320273ac1 100644
--- a/src/include/nodes/nodes.h
+++ b/src/include/nodes/nodes.h
@@ -265,6 +265,7 @@ typedef enum NodeTag
T_EquivalenceMember,
T_PathKey,
T_PathTarget,
+ T_UniqueKey,
T_RestrictInfo,
T_IndexClause,
T_PlaceHolderVar,
diff --git a/src/include/nodes/pathnodes.h b/src/include/nodes/pathnodes.h
index 0d61f04d27..80561b1003 100644
--- a/src/include/nodes/pathnodes.h
+++ b/src/include/nodes/pathnodes.h
@@ -739,6 +739,7 @@ typedef struct RelOptInfo
QualCost baserestrictcost; /* cost of evaluating the above */
Index baserestrict_min_security; /* min security_level found in
* baserestrictinfo */
+ List *uniquekeys; /* List of UniqueKey */
List *joininfo; /* RestrictInfo structures for join clauses
* involving this rel */
bool has_eclass_joins; /* T means joininfo is incomplete */
@@ -1059,6 +1060,28 @@ typedef struct PathKey
} PathKey;
+/*
+ * UniqueKey
+ *
+ * Represents the unique properties held by a RelOptInfo.
+ *
+ * exprs is a list of exprs which is unique on current RelOptInfo. exprs = NIL
+ * is a special case of UniqueKey, which means there is only 1 row in that
+ * relation.
+ * multi_nullvals: true means multi null values may exist in these exprs, so the
+ * uniqueness is not guaranteed in this case. This field is necessary for
+ * remove_useless_join & reduce_unique_semijoins where we don't mind these
+ * duplicated NULL values. It is set to true for 2 cases. One is a unique key
+ * from a unique index but the related column is nullable. The other one is for
+ * outer join. see populate_joinrel_uniquekeys for detail.
+ */
+typedef struct UniqueKey
+{
+ NodeTag type;
+ List *exprs;
+ bool multi_nullvals;
+} UniqueKey;
+
/*
* PathTarget
*
@@ -2496,7 +2519,7 @@ typedef enum
*
* flags indicating what kinds of grouping are possible.
* partial_costs_set is true if the agg_partial_costs and agg_final_costs
- * have been initialized.
+ * have been initialized.
* agg_partial_costs gives partial aggregation costs.
* agg_final_costs gives finalization costs.
* target_parallel_safe is true if target is parallel safe.
@@ -2526,8 +2549,8 @@ typedef struct
* limit_tuples is an estimated bound on the number of output tuples,
* or -1 if no LIMIT or couldn't estimate.
* count_est and offset_est are the estimated values of the LIMIT and OFFSET
- * expressions computed by preprocess_limit() (see comments for
- * preprocess_limit() for more information).
+ * expressions computed by preprocess_limit() (see comments for
+ * preprocess_limit() for more information).
*/
typedef struct
{
diff --git a/src/include/nodes/pg_list.h b/src/include/nodes/pg_list.h
index 404e03f132..85c65e910c 100644
--- a/src/include/nodes/pg_list.h
+++ b/src/include/nodes/pg_list.h
@@ -566,6 +566,16 @@ extern pg_nodiscard List *list_delete_first(List *list);
extern pg_nodiscard List *list_delete_last(List *list);
extern pg_nodiscard List *list_delete_nth_cell(List *list, int n);
extern pg_nodiscard List *list_delete_cell(List *list, ListCell *cell);
+extern bool list_is_subset(const List *members, const List *target);
+
+extern List *list_delete(List *list, void *datum);
+extern List *list_delete_ptr(List *list, void *datum);
+extern List *list_delete_int(List *list, int datum);
+extern List *list_delete_oid(List *list, Oid datum);
+extern List *list_delete_first(List *list);
+extern List *list_delete_last(List *list);
+extern List *list_delete_nth_cell(List *list, int n);
+extern List *list_delete_cell(List *list, ListCell *cell);
extern List *list_union(const List *list1, const List *list2);
extern List *list_union_ptr(const List *list1, const List *list2);
diff --git a/src/include/optimizer/appendinfo.h b/src/include/optimizer/appendinfo.h
index 4cbf8c26cc..43b8d05f69 100644
--- a/src/include/optimizer/appendinfo.h
+++ b/src/include/optimizer/appendinfo.h
@@ -32,4 +32,7 @@ extern Relids adjust_child_relids_multilevel(PlannerInfo *root, Relids relids,
extern AppendRelInfo **find_appinfos_by_relids(PlannerInfo *root,
Relids relids, int *nappinfos);
+extern AppendRelInfo *find_appinfo_by_child(PlannerInfo *root, Index child_index);
+extern Var *find_parent_var(AppendRelInfo *appinfo, Var *child_var);
+
#endif /* APPENDINFO_H */
diff --git a/src/include/optimizer/optimizer.h b/src/include/optimizer/optimizer.h
index d587952b7d..843aafa51e 100644
--- a/src/include/optimizer/optimizer.h
+++ b/src/include/optimizer/optimizer.h
@@ -23,6 +23,7 @@
#define OPTIMIZER_H
#include "nodes/parsenodes.h"
+#include "nodes/pathnodes.h"
/* Test if an expression node represents a SRF call. Beware multiple eval! */
#define IS_SRF_CALL(node) \
@@ -168,6 +169,7 @@ extern TargetEntry *get_sortgroupref_tle(Index sortref,
List *targetList);
extern TargetEntry *get_sortgroupclause_tle(SortGroupClause *sgClause,
List *targetList);
+extern Var *find_var_for_subquery_tle(RelOptInfo *rel, TargetEntry *tle);
extern Node *get_sortgroupclause_expr(SortGroupClause *sgClause,
List *targetList);
extern List *get_sortgrouplist_exprs(List *sgClauses,
diff --git a/src/include/optimizer/paths.h b/src/include/optimizer/paths.h
index 035d3e1206..1adf99c9ee 100644
--- a/src/include/optimizer/paths.h
+++ b/src/include/optimizer/paths.h
@@ -247,5 +247,48 @@ extern PathKey *make_canonical_pathkey(PlannerInfo *root,
int strategy, bool nulls_first);
extern void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
List *live_childrels);
+extern List *select_mergejoin_clauses(PlannerInfo *root,
+ RelOptInfo *joinrel,
+ RelOptInfo *outerrel,
+ RelOptInfo *innerrel,
+ List *restrictlist,
+ JoinType jointype,
+ bool *mergejoin_allowed);
+
+/*
+ * uniquekeys.c
+ * Utilities for matching and building unique keys
+ */
+extern void populate_baserel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *baserel,
+ List* unique_index_list);
+extern void populate_partitionedrel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *rel,
+ List *childrels);
+extern void populate_distinctrel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *inputrel,
+ RelOptInfo *distinctrel);
+extern void populate_grouprel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *grouprel,
+ RelOptInfo *inputrel);
+extern void populate_unionrel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *unionrel);
+extern void simple_copy_uniquekeys(RelOptInfo *oldrel,
+ RelOptInfo *newrel);
+extern void convert_subquery_uniquekeys(PlannerInfo *root,
+ RelOptInfo *currel,
+ RelOptInfo *sub_final_rel);
+extern void populate_joinrel_uniquekeys(PlannerInfo *root,
+ RelOptInfo *joinrel,
+ RelOptInfo *rel1,
+ RelOptInfo *rel2,
+ List *restrictlist,
+ JoinType jointype);
+
+extern bool relation_has_uniquekeys_for(PlannerInfo *root,
+ RelOptInfo *rel,
+ List *exprs,
+ bool allow_multinulls);
+extern bool relation_is_onerow(RelOptInfo *rel);
#endif /* PATHS_H */
--
2.26.2
--yh7cw345mbttadjz
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment; filename="v38-0003-Extend-UniqueKeys.patch"
^ permalink raw reply [nested|flat] 9+ messages in thread
* Re: Collecting statistics about contents of JSONB columns
@ 2022-01-23 00:24 Nikita Glukhov <[email protected]>
0 siblings, 1 reply; 9+ messages in thread
From: Nikita Glukhov @ 2022-01-23 00:24 UTC (permalink / raw)
To: Tomas Vondra <[email protected]>; +Cc: PostgreSQL Hackers <[email protected]>; Mahendra Thalor <[email protected]>; Oleg Bartunov <[email protected]>; Thomas Munro <[email protected]>
Hi!
I am glad that you found my very old patch interesting and started to
work on it. We failed to post it in 2016 mostly because we were not
satisfied with JSONB storage. Also we decided to wait for completion
of work on extended statistics as we thought that it could help us.
But in early 2017 we switched to SQL/JSON and forgot about this patch.
I think custom datatype is necessary for better performance. With a
plain JSONB we need to do a lot of work for extraction of path stats:
- iterate through MCV/Histogram JSONB arrays
- cast numeric values to float, string to text etc.
- build PG arrays from extracted datums
- form pg_statistic tuple.
With a custom data type we could store pg_statistic tuple unmodified
and use it without any overhead. But then we need modify a bit
VariableStatData and several functions to pass additional nullfrac
corrections.
Maybe simple record type (path text, data pg_statistic, ext_data jsonb)
would be enough.
Also there is an idea to store per-path separately in pg_statistic_ext
rows using expression like (jb #> '{foo,bar}') as stxexprs. This could
also help user to select which paths to analyze simply by using some
sort of CREATE STATISTICS. But it is really unclear how to:
* store pg_statistic_ext rows from typanalyze
* form path expressions for array elements (maybe add new jsonpath
operator)
* match various -> operator chains to stxexprs
* jsonpath-like languages need simple API for searching by stxexprs
Per-path statistics should only be collected for scalars. This can be
enabled by flag JsonAnalyzeContext.scalarsOnly. But there are is a
problem: computed MCVs and histograms will be broken and we will not be
able to use them for queries like (jb > const) in general case. Also
we will not be and use internally in scalarineqsel() and var_eq_const()
(see jsonSelectivity()). Instead, we will have to implement own
estimator functions for JSONB comparison operators that will correctly
use our hacked MCVs and histograms (and maybe not all cases will be
supported; for example, comparison to scalars only).
It's possible to replace objects and arrays with empty ones when
scalarsOnly is set to keep correct frequencies of non-scalars.
But there is an old bug in JSONB comparison: empty arrays are placed
before other values in the JSONB sort order, although they should go
after all scalars. So we need also to distinguish empty and non-empty
arrays here.
I tried to fix a major part of places marked as XXX and FIXME, the new
version of the patches is attached. There are a lot of changes, you
can see them in a step-by-step form in the corresponding branch
jsonb_stats_20220122 in our GitHub repo [1].
Below is the explanation of fixed XXXs:
Patch 0001
src/backend/commands/operatorcmds.c:
> XXX Maybe not the right name, because "estimator" implies we're
> calculating selectivity. But we're actually deriving statistics for
> an expression.
Renamed to "Derivator".
> XXX perhaps full "statistics" wording would be better
Renamed to "statistics".
src/backend/utils/adt/selfuncs.c:
> examine_operator_expression():
> XXX Not sure why this returns bool - we ignore the return value
> anyway. We might as well return the calculated vardata (or NULL).
Oprstat was changed to return void.
> XXX Not sure what to do about recursion - there can be another OpExpr
> in one of the arguments, and we should call this recursively from the
> oprstat procedure. But that's not possible, because it's marked as
> static.
Oprstat call chain: get_restriction_variable() => examine_variable() =>
examine_operator_expression().
The main thing here is that OpExpr with oprstat acts like ordinary Var.
> examine_variable():
> XXX Shouldn't this put more restrictions on the OpExpr? E.g. that
> one of the arguments has to be a Const or something?
This is simply a responsibility of oprstat.
Patch 0004
> XXX Not sure we want to add these functions to jsonb.h, which is the
> public API. Maybe it belongs rather to jsonb_typanalyze.c or
> elsewhere, closer to how it's used?
Maybe it needs to be moved the new file jsonb_utils.h. I think these
functions become very useful, when we start to build JSONBs with
predefined structure.
> pushJsonbValue():
> XXX I'm not quite sure why we actually do this? Why do we need to
> change how JsonbValue is converted to Jsonb for the statistics patch?
Scalars in JSONB are encoded as one-element preudo-array containers.
So when we are inserting binary JsonbValues, that was initialized
directly from JSONB datums, inside other non-empty JSONB in
pushJsonbValue(), all values except scalars are inserted as
expected but scalars become [scalar]. So we need to extract scalar
values in caller functions or in pushJsonbValue(). I think
autoextraction in pushJsonbValue() is better. This case simply was not
used before JSONB stats introduction.
Patch 0006
src/backend/utils/adt/jsonb_selfuncs.c
> jsonPathStatsGetSpecialStats()
> XXX This does not really extract any stats, it merely allocates the
> struct?
Renamed with "Alloc" suffix.
> jsonPathStatsGetArrayLengthStats()
> XXX Why doesn't this do jsonPathStatsGetTypeFreq check similar to
> what jsonPathStatsGetLengthStats does?
"length" stats were collected inside parent paths, but "array_length"
and "avg_array_length" stats were collected inside child array paths.
This resulted in inconsistencies in TypeFreq checks.
I have removed "length" stats, moved "array_length" and
"avg_array_length" to the parent path, and added separate
"object_length" stats. TypeFreq checks become consistent.
> XXX Seems to do essentially what jsonStatsFindPath, except that it
> also considers jsdata->prefix. Seems fairly easy to combine those
> into a single function.
I don't think that it would be better to combine those function into
one with considerPrefix flag, because jsonStatsFindPathStats() is a
some kind of low-level function which is called only in two places.
In all other places considerPrefix will be true. Also
jsonStatsFindPathStatsStr() is exported in jsonb_selfuncs.h for to give
external jsonpath-like query operators ability to use JSON statistics.
> jsonPathStatsCompare()
> XXX Seems a bit convoluted to first cast it to Datum, then Jsonb ...
> Datum const *pdatum = pv2;
> Jsonb *jsonb = DatumGetJsonbP(*pdatum);
The problem of simply using 'jsonb = *(Jsonb **) pv2' is that
DatumGetJsonbP() may deTOAST datums.
> XXX Not sure about this? Does empty path mean global stats?
Empty "path" is simply a sign of invalid stats data.
> jsonStatsConvertArray()
> FIXME Does this actually work on all 32/64-bit systems? What if typid
> is FLOAT8OID or something? Should look at TypeCache instead, probably.
Used get_typlenbyvalalign() instead of hardcoded values.
> jsonb_stats()
> XXX It might be useful to allow recursion, i.e.
> get_restriction_variable might derive statistics too.
> I don't think it does that now, right?
get_restriction_variable() already might derive statistics: it calls
examine_variable() on its both operands, and examine_variable() calls
examine_operator_expression().
> XXX Could we also get varonleft=false in useful cases?
All supported JSONB operators have signature jsonb OP arg.
If varonleft = false, then we need to derive stats for expression like
'{"foo": "bar"}' -> text_column
having stats for text_column. It is possible to implement too, but it
is not related to JSONB stats and needs to be done in the separate
patch.
> jsonSelectivityContains():
> XXX This really needs more comments explaining the logic.
I have refactored this function and added comments.
> jsonGetFloat4():
> XXX Not sure assert is a sufficient protection against different
> types of JSONB values to be passed in.
I have refactored this function by passing a default value for
non-numeric JSONB case.
> jsonPathAppendEntryWithLen()
> XXX Doesn't this need ecape_json too?
Comment is removed, because jsonPathAppendEntry() is called inside.
> jsonPathStatsGetTypeFreq()
> FIXME This is really hard to read/understand, with two nested ternary
> operators.
I have refactored this place.
> XXX Seems more like an error, no? Why ignore it?
It's not an error, it's request of non-numeric type. Length values
are always numeric, and frequence of non-numeric values is 0.
The possible case when it could happen is jsonpath '$.size() == "foo"'.
Estimator for operator == will check frequence of strings in .size()
and it will be 0.
> jsonPathStatsFormTuple()
> FIXME What does this mean?
There is no need to transform ordinary root path stats tuple, it can be
simply copied.
jsonb_typanalyze.c
> XXX We need entry+lenth because JSON path elements may contain null
> bytes, I guess?
'entry' can be not zero-terminated when it is pointing to JSONB keys,
so 'len' is necessary field. 'len' is also used for faster entry
comparison, to distinguish array entries ('len' == -1).
> XXX Sould be JsonPathEntryMatch as it deals with JsonPathEntry nodes
> not whole paths, no?
> XXX Again, maybe JsonPathEntryHash would be a better name?
Functions renamed using JsonPathEntry prefix, JsonPath typedef removed.
> JsonPathMatch()
> XXX Seems a bit silly to return int, when the return statement only
> really returns bool (because of how it compares paths). It's not really
> a comparator for sorting, for example.
This function is implementation of HashCompareFunc and it needs to
return int.
> jsonAnalyzeJson()
> XXX The name seems a bit weird, with the two json bits.
Renamed to jsonAnalyzeCollectPaths(). The two similar functions were
renamed too.
> jsonAnalyzeJson():
> XXX The mix of break/return statements in this block is really
> confusing.
I have refactored this place using only breaks.
> XXX not sure why we're doing this?
Manual recursion into containers by creating child iterator together
with skipNested=true flag is used to give jsonAnalyzeJsonValue()
ability to access jbvBinary containers.
> compute_json_stats()
> XXX Not sure what the first branch is doing (or supposed to)?
> XXX It's not immediately clear why this is (-1) and not simply
> NULL. It crashes, so presumably it's used to tweak the behavior,
> but it's not clear why/how, and it affects place that is pretty
> far away, and so not obvious. We should use some sort of flag
> with a descriptive name instead.
> XXX If I understand correctly, we simply collect all paths first,
> without accumulating any Values. And then in the next step we
> process each path independently, probably to save memory (we
> don't want to accumulate all values for all paths, with a lot
> of duplicities).
There are two variants of stats collection:
* single-pass - collect all values for all paths
* multi-pass - collect only values for a one path at each pass
The first variant can consume too much memory (jsonb iteration produces
a lot of garbage etc.), but works faster than second.
The first 'if (false)' is used for manual selection of one of this
variants. This selection should be controlled by some user-specified
option (maybe GUC), or the first variant can be simply removed.
jsonAnalyzeJson()'s parameter of type JsonPathAnlStats * determines
which paths we need to consider for the value collection:
* NULL - collect values for all paths
* -1 - do not collect any values
* stats - collect values only for a given path
The last variant really is unused because we already have another
function jsonAnalyzeJsonPath(), which is optimized for selective path
values collection (used object key accessor JSONB functions instead of
full JSONB iteration). I have replaced this strange parameter with
simple boolean flag.
> XXX Could the parameters be different on other platforms?
Used get_typlenbyvalalign(JSONBOID) instead of hardcoded values.
> jsonAnalyzePathValues()
> XXX Do we need to initialize all slots?
I have copied here the following comment from extended_stats.c:
/*
* The fields describing the stats->stavalues[n] element types default
* to the type of the data being analyzed, but the type-specific
* typanalyze function can change them if it wants to store something
* else.
*/
> XXX Not sure why we divide it by the number of json values?
We divide counts of lengths by the total number of json values to
compute correct nullfrac. I.e. not all input jsons have lengths,
length of scalar jsons is NULL.
[1]https://github.com/postgrespro/postgres/tree/jsonb_stats_20220122
--
Nikita Glukhov
Postgres Professional:http://www.postgrespro.com
The Russian Postgres Company
Attachments:
[text/x-patch] 0001-Add-pg_operator.oprstat-for-derived-operator-statist.patch (11.9K, ../../[email protected]/3-0001-Add-pg_operator.oprstat-for-derived-operator-statist.patch)
download | inline diff:
From f334b5a210dd57659da2e202f398052158a5291a Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Sat, 12 Nov 2016 18:59:43 +0300
Subject: [PATCH 1/6] Add pg_operator.oprstat for derived operator statistics
estimation
---
src/backend/catalog/pg_operator.c | 11 +++++
src/backend/commands/operatorcmds.c | 61 ++++++++++++++++++++++++++
src/backend/utils/adt/selfuncs.c | 38 ++++++++++++++++
src/backend/utils/cache/lsyscache.c | 24 ++++++++++
src/include/catalog/pg_operator.h | 4 ++
src/include/utils/lsyscache.h | 1 +
src/test/regress/expected/oidjoins.out | 1 +
7 files changed, 140 insertions(+)
diff --git a/src/backend/catalog/pg_operator.c b/src/backend/catalog/pg_operator.c
index 630bf3e56cc..9205e62c0eb 100644
--- a/src/backend/catalog/pg_operator.c
+++ b/src/backend/catalog/pg_operator.c
@@ -256,6 +256,7 @@ OperatorShellMake(const char *operatorName,
values[Anum_pg_operator_oprcode - 1] = ObjectIdGetDatum(InvalidOid);
values[Anum_pg_operator_oprrest - 1] = ObjectIdGetDatum(InvalidOid);
values[Anum_pg_operator_oprjoin - 1] = ObjectIdGetDatum(InvalidOid);
+ values[Anum_pg_operator_oprstat - 1] = ObjectIdGetDatum(InvalidOid);
/*
* create a new operator tuple
@@ -301,6 +302,7 @@ OperatorShellMake(const char *operatorName,
* negatorName X negator operator
* restrictionId X restriction selectivity procedure ID
* joinId X join selectivity procedure ID
+ * statsId X statistics derivation procedure ID
* canMerge merge join can be used with this operator
* canHash hash join can be used with this operator
*
@@ -333,6 +335,7 @@ OperatorCreate(const char *operatorName,
List *negatorName,
Oid restrictionId,
Oid joinId,
+ Oid statsId,
bool canMerge,
bool canHash)
{
@@ -505,6 +508,7 @@ OperatorCreate(const char *operatorName,
values[Anum_pg_operator_oprcode - 1] = ObjectIdGetDatum(procedureId);
values[Anum_pg_operator_oprrest - 1] = ObjectIdGetDatum(restrictionId);
values[Anum_pg_operator_oprjoin - 1] = ObjectIdGetDatum(joinId);
+ values[Anum_pg_operator_oprstat - 1] = ObjectIdGetDatum(statsId);
pg_operator_desc = table_open(OperatorRelationId, RowExclusiveLock);
@@ -855,6 +859,13 @@ makeOperatorDependencies(HeapTuple tuple,
add_exact_object_address(&referenced, addrs);
}
+ /* Dependency on statistics derivation function */
+ if (OidIsValid(oper->oprstat))
+ {
+ ObjectAddressSet(referenced, ProcedureRelationId, oper->oprstat);
+ add_exact_object_address(&referenced, addrs);
+ }
+
record_object_address_dependencies(&myself, addrs, DEPENDENCY_NORMAL);
free_object_addresses(addrs);
diff --git a/src/backend/commands/operatorcmds.c b/src/backend/commands/operatorcmds.c
index a5924d7d564..adf13e648a6 100644
--- a/src/backend/commands/operatorcmds.c
+++ b/src/backend/commands/operatorcmds.c
@@ -52,6 +52,7 @@
static Oid ValidateRestrictionEstimator(List *restrictionName);
static Oid ValidateJoinEstimator(List *joinName);
+static Oid ValidateStatisticsDerivator(List *joinName);
/*
* DefineOperator
@@ -78,10 +79,12 @@ DefineOperator(List *names, List *parameters)
List *commutatorName = NIL; /* optional commutator operator name */
List *negatorName = NIL; /* optional negator operator name */
List *restrictionName = NIL; /* optional restrict. sel. function */
+ List *statisticsName = NIL; /* optional stats estimat. procedure */
List *joinName = NIL; /* optional join sel. function */
Oid functionOid; /* functions converted to OID */
Oid restrictionOid;
Oid joinOid;
+ Oid statisticsOid;
Oid typeId[2]; /* to hold left and right arg */
int nargs;
ListCell *pl;
@@ -131,6 +134,8 @@ DefineOperator(List *names, List *parameters)
restrictionName = defGetQualifiedName(defel);
else if (strcmp(defel->defname, "join") == 0)
joinName = defGetQualifiedName(defel);
+ else if (strcmp(defel->defname, "statistics") == 0)
+ statisticsName = defGetQualifiedName(defel);
else if (strcmp(defel->defname, "hashes") == 0)
canHash = defGetBoolean(defel);
else if (strcmp(defel->defname, "merges") == 0)
@@ -246,6 +251,10 @@ DefineOperator(List *names, List *parameters)
joinOid = ValidateJoinEstimator(joinName);
else
joinOid = InvalidOid;
+ if (statisticsName)
+ statisticsOid = ValidateStatisticsDerivator(statisticsName);
+ else
+ statisticsOid = InvalidOid;
/*
* now have OperatorCreate do all the work..
@@ -260,6 +269,7 @@ DefineOperator(List *names, List *parameters)
negatorName, /* optional negator operator name */
restrictionOid, /* optional restrict. sel. function */
joinOid, /* optional join sel. function name */
+ statisticsOid, /* optional stats estimation procedure */
canMerge, /* operator merges */
canHash); /* operator hashes */
}
@@ -357,6 +367,40 @@ ValidateJoinEstimator(List *joinName)
return joinOid;
}
+/*
+ * Look up a statistics estimator function by name, and verify that it has the
+ * correct signature and we have the permissions to attach it to an operator.
+ */
+static Oid
+ValidateStatisticsDerivator(List *statName)
+{
+ Oid typeId[4];
+ Oid statOid;
+ AclResult aclresult;
+
+ typeId[0] = INTERNALOID; /* PlannerInfo */
+ typeId[1] = INTERNALOID; /* OpExpr */
+ typeId[2] = INT4OID; /* varRelid */
+ typeId[3] = INTERNALOID; /* VariableStatData */
+
+ statOid = LookupFuncName(statName, 4, typeId, false);
+
+ /* statistics estimators must return void */
+ if (get_func_rettype(statOid) != VOIDOID)
+ ereport(ERROR,
+ (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
+ errmsg("statistics estimator function %s must return type %s",
+ NameListToString(statName), "void")));
+
+ /* Require EXECUTE rights for the estimator */
+ aclresult = pg_proc_aclcheck(statOid, GetUserId(), ACL_EXECUTE);
+ if (aclresult != ACLCHECK_OK)
+ aclcheck_error(aclresult, OBJECT_FUNCTION,
+ NameListToString(statName));
+
+ return statOid;
+}
+
/*
* Guts of operator deletion.
*/
@@ -424,6 +468,9 @@ AlterOperator(AlterOperatorStmt *stmt)
List *joinName = NIL; /* optional join sel. function */
bool updateJoin = false;
Oid joinOid;
+ List *statName = NIL; /* optional statistics estimation procedure */
+ bool updateStat = false;
+ Oid statOid;
/* Look up the operator */
oprId = LookupOperWithArgs(stmt->opername, false);
@@ -454,6 +501,11 @@ AlterOperator(AlterOperatorStmt *stmt)
joinName = param;
updateJoin = true;
}
+ else if (pg_strcasecmp(defel->defname, "stats") == 0)
+ {
+ statName = param;
+ updateStat = true;
+ }
/*
* The rest of the options that CREATE accepts cannot be changed.
@@ -496,6 +548,10 @@ AlterOperator(AlterOperatorStmt *stmt)
joinOid = ValidateJoinEstimator(joinName);
else
joinOid = InvalidOid;
+ if (statName)
+ statOid = ValidateStatisticsDerivator(statName);
+ else
+ statOid = InvalidOid;
/* Perform additional checks, like OperatorCreate does */
if (!(OidIsValid(oprForm->oprleft) && OidIsValid(oprForm->oprright)))
@@ -536,6 +592,11 @@ AlterOperator(AlterOperatorStmt *stmt)
replaces[Anum_pg_operator_oprjoin - 1] = true;
values[Anum_pg_operator_oprjoin - 1] = joinOid;
}
+ if (updateStat)
+ {
+ replaces[Anum_pg_operator_oprstat - 1] = true;
+ values[Anum_pg_operator_oprstat - 1] = statOid;
+ }
tup = heap_modify_tuple(tup, RelationGetDescr(catalog),
values, nulls, replaces);
diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c
index 1fbb0b28c3b..d1dd049f1ae 100644
--- a/src/backend/utils/adt/selfuncs.c
+++ b/src/backend/utils/adt/selfuncs.c
@@ -4917,6 +4917,30 @@ ReleaseDummy(HeapTuple tuple)
pfree(tuple);
}
+/*
+ * examine_operator_expression
+ * Try to derive optimizer statistics for the operator expression using
+ * operator's oprstat function.
+ *
+ * There can be another OpExpr in one of the arguments, and it will be called
+ * recursively from the oprstat procedure through the following chain:
+ * get_restriction_variable() => examine_variable() =>
+ * examine_operator_expression().
+ */
+static void
+examine_operator_expression(PlannerInfo *root, OpExpr *opexpr, int varRelid,
+ VariableStatData *vardata)
+{
+ RegProcedure oprstat = get_oprstat(opexpr->opno);
+
+ if (OidIsValid(oprstat))
+ OidFunctionCall4(oprstat,
+ PointerGetDatum(root),
+ PointerGetDatum(opexpr),
+ Int32GetDatum(varRelid),
+ PointerGetDatum(vardata));
+}
+
/*
* examine_variable
* Try to look up statistical data about an expression.
@@ -5332,6 +5356,20 @@ examine_variable(PlannerInfo *root, Node *node, int varRelid,
pos++;
}
}
+
+ /*
+ * If there's no index or extended statistics matching the expression,
+ * try deriving the statistics from statistics on arguments of the
+ * operator expression (OpExpr). We do this last because it may be quite
+ * expensive, and it's unclear how accurate the statistics will be.
+ *
+ * More restrictions on the OpExpr (e.g. that one of the arguments
+ * has to be a Const or something) can be put by the operator itself
+ * in its oprstat function.
+ */
+ if (!vardata->statsTuple && IsA(basenode, OpExpr))
+ examine_operator_expression(root, (OpExpr *) basenode, varRelid,
+ vardata);
}
}
diff --git a/src/backend/utils/cache/lsyscache.c b/src/backend/utils/cache/lsyscache.c
index feef9998634..b5440b596cd 100644
--- a/src/backend/utils/cache/lsyscache.c
+++ b/src/backend/utils/cache/lsyscache.c
@@ -1567,6 +1567,30 @@ get_oprjoin(Oid opno)
return (RegProcedure) InvalidOid;
}
+/*
+ * get_oprstat
+ *
+ * Returns procedure id for estimating statistics for an operator.
+ */
+RegProcedure
+get_oprstat(Oid opno)
+{
+ HeapTuple tp;
+
+ tp = SearchSysCache1(OPEROID, ObjectIdGetDatum(opno));
+ if (HeapTupleIsValid(tp))
+ {
+ Form_pg_operator optup = (Form_pg_operator) GETSTRUCT(tp);
+ RegProcedure result;
+
+ result = optup->oprstat;
+ ReleaseSysCache(tp);
+ return result;
+ }
+ else
+ return (RegProcedure) InvalidOid;
+}
+
/* ---------- FUNCTION CACHE ---------- */
/*
diff --git a/src/include/catalog/pg_operator.h b/src/include/catalog/pg_operator.h
index 51263f550fe..ff1bb339f75 100644
--- a/src/include/catalog/pg_operator.h
+++ b/src/include/catalog/pg_operator.h
@@ -73,6 +73,9 @@ CATALOG(pg_operator,2617,OperatorRelationId)
/* OID of join estimator, or 0 */
regproc oprjoin BKI_DEFAULT(-) BKI_LOOKUP_OPT(pg_proc);
+
+ /* OID of statistics estimator, or 0 */
+ regproc oprstat BKI_DEFAULT(-) BKI_LOOKUP_OPT(pg_proc);
} FormData_pg_operator;
/* ----------------
@@ -95,6 +98,7 @@ extern ObjectAddress OperatorCreate(const char *operatorName,
List *negatorName,
Oid restrictionId,
Oid joinId,
+ Oid statisticsId,
bool canMerge,
bool canHash);
diff --git a/src/include/utils/lsyscache.h b/src/include/utils/lsyscache.h
index b8dd27d4a96..cc08fc50a50 100644
--- a/src/include/utils/lsyscache.h
+++ b/src/include/utils/lsyscache.h
@@ -118,6 +118,7 @@ extern Oid get_commutator(Oid opno);
extern Oid get_negator(Oid opno);
extern RegProcedure get_oprrest(Oid opno);
extern RegProcedure get_oprjoin(Oid opno);
+extern RegProcedure get_oprstat(Oid opno);
extern char *get_func_name(Oid funcid);
extern Oid get_func_namespace(Oid funcid);
extern Oid get_func_rettype(Oid funcid);
diff --git a/src/test/regress/expected/oidjoins.out b/src/test/regress/expected/oidjoins.out
index 215eb899be3..111ea99cdae 100644
--- a/src/test/regress/expected/oidjoins.out
+++ b/src/test/regress/expected/oidjoins.out
@@ -113,6 +113,7 @@ NOTICE: checking pg_operator {oprnegate} => pg_operator {oid}
NOTICE: checking pg_operator {oprcode} => pg_proc {oid}
NOTICE: checking pg_operator {oprrest} => pg_proc {oid}
NOTICE: checking pg_operator {oprjoin} => pg_proc {oid}
+NOTICE: checking pg_operator {oprstat} => pg_proc {oid}
NOTICE: checking pg_opfamily {opfmethod} => pg_am {oid}
NOTICE: checking pg_opfamily {opfnamespace} => pg_namespace {oid}
NOTICE: checking pg_opfamily {opfowner} => pg_authid {oid}
--
2.25.1
[text/x-patch] 0002-Add-stats_form_tuple.patch (3.9K, ../../[email protected]/4-0002-Add-stats_form_tuple.patch)
download | inline diff:
From 08e99152f37eeb73a879b640c97e528630581f3b Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Wed, 7 Dec 2016 16:12:55 +0300
Subject: [PATCH 2/6] Add stats_form_tuple()
---
src/backend/utils/adt/selfuncs.c | 55 ++++++++++++++++++++++++++++++++
src/include/utils/selfuncs.h | 22 +++++++++++++
2 files changed, 77 insertions(+)
diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c
index d1dd049f1ae..f534e862079 100644
--- a/src/backend/utils/adt/selfuncs.c
+++ b/src/backend/utils/adt/selfuncs.c
@@ -7954,3 +7954,58 @@ brincostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
*indexPages = index->pages;
}
+
+/*
+ * stats_form_tuple - Form pg_statistic tuple from StatsData.
+ *
+ * If 'data' parameter is NULL, form all-NULL tuple (nullfrac = 1.0).
+ */
+HeapTuple
+stats_form_tuple(StatsData *data)
+{
+ Relation rel;
+ HeapTuple tuple;
+ Datum values[Natts_pg_statistic];
+ bool nulls[Natts_pg_statistic];
+ int i;
+
+ for (i = 0; i < Natts_pg_statistic; ++i)
+ nulls[i] = false;
+
+ values[Anum_pg_statistic_starelid - 1] = ObjectIdGetDatum(InvalidOid);
+ values[Anum_pg_statistic_staattnum - 1] = Int16GetDatum(0);
+ values[Anum_pg_statistic_stainherit - 1] = BoolGetDatum(false);
+ values[Anum_pg_statistic_stanullfrac - 1] =
+ Float4GetDatum(data ? data->nullfrac : 1.0);
+ values[Anum_pg_statistic_stawidth - 1] =
+ Int32GetDatum(data ? data->width : 0);
+ values[Anum_pg_statistic_stadistinct - 1] =
+ Float4GetDatum(data ? data->distinct : 0);
+
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ StatsSlot *slot = data ? &data->slots[i] : NULL;
+
+ values[Anum_pg_statistic_stakind1 + i - 1] =
+ Int16GetDatum(slot ? slot->kind : 0);
+
+ values[Anum_pg_statistic_staop1 + i - 1] =
+ ObjectIdGetDatum(slot ? slot->opid : InvalidOid);
+
+ if (slot && DatumGetPointer(slot->numbers))
+ values[Anum_pg_statistic_stanumbers1 + i - 1] = slot->numbers;
+ else
+ nulls[Anum_pg_statistic_stanumbers1 + i - 1] = true;
+
+ if (slot && DatumGetPointer(slot->values))
+ values[Anum_pg_statistic_stavalues1 + i - 1] = slot->values;
+ else
+ nulls[Anum_pg_statistic_stavalues1 + i - 1] = true;
+ }
+
+ rel = table_open(StatisticRelationId, AccessShareLock);
+ tuple = heap_form_tuple(RelationGetDescr(rel), values, nulls);
+ table_close(rel, NoLock);
+
+ return tuple;
+}
diff --git a/src/include/utils/selfuncs.h b/src/include/utils/selfuncs.h
index 8f3d73edfb2..49fc3d717b1 100644
--- a/src/include/utils/selfuncs.h
+++ b/src/include/utils/selfuncs.h
@@ -16,6 +16,7 @@
#define SELFUNCS_H
#include "access/htup.h"
+#include "catalog/pg_statistic.h"
#include "fmgr.h"
#include "nodes/pathnodes.h"
@@ -133,6 +134,24 @@ typedef struct
double num_sa_scans; /* # indexscans from ScalarArrayOpExprs */
} GenericCosts;
+/* Single pg_statistic slot */
+typedef struct StatsSlot
+{
+ int16 kind; /* stakindN: statistic kind (STATISTIC_KIND_XXX) */
+ Oid opid; /* staopN: associated operator, if needed */
+ Datum numbers; /* stanumbersN: float4 array of numbers */
+ Datum values; /* stavaluesN: anyarray of values */
+} StatsSlot;
+
+/* Deformed pg_statistic tuple */
+typedef struct StatsData
+{
+ float4 nullfrac; /* stanullfrac: fraction of NULL values */
+ float4 distinct; /* stadistinct: number of distinct non-NULL values */
+ int32 width; /* stawidth: average width in bytes of non-NULL values */
+ StatsSlot slots[STATISTIC_NUM_SLOTS]; /* slots for different statistic types */
+} StatsData;
+
/* Hooks for plugins to get control when we ask for stats */
typedef bool (*get_relation_stats_hook_type) (PlannerInfo *root,
RangeTblEntry *rte,
@@ -231,6 +250,9 @@ extern void genericcostestimate(PlannerInfo *root, IndexPath *path,
double loop_count,
GenericCosts *costs);
+extern HeapTuple stats_form_tuple(StatsData *data);
+
+
/* Functions in array_selfuncs.c */
extern Selectivity scalararraysel_containment(PlannerInfo *root,
--
2.25.1
[text/x-patch] 0003-Add-symbolic-names-for-some-jsonb-operators.patch (7.1K, ../../[email protected]/5-0003-Add-symbolic-names-for-some-jsonb-operators.patch)
download | inline diff:
From 4735d6de9ed1f8f93d9592d9ffa1f662fcdafbe7 Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Sat, 12 Nov 2016 18:59:43 +0300
Subject: [PATCH 3/6] Add symbolic names for some jsonb operators
---
src/include/catalog/pg_operator.dat | 45 +++++++++++++++++------------
1 file changed, 26 insertions(+), 19 deletions(-)
diff --git a/src/include/catalog/pg_operator.dat b/src/include/catalog/pg_operator.dat
index bc5f8213f3a..8e0e65ad275 100644
--- a/src/include/catalog/pg_operator.dat
+++ b/src/include/catalog/pg_operator.dat
@@ -2067,7 +2067,7 @@
{ oid => '1751', descr => 'negate',
oprname => '-', oprkind => 'l', oprleft => '0', oprright => 'numeric',
oprresult => 'numeric', oprcode => 'numeric_uminus' },
-{ oid => '1752', descr => 'equal',
+{ oid => '1752', oid_symbol => 'NumericEqOperator', descr => 'equal',
oprname => '=', oprcanmerge => 't', oprcanhash => 't', oprleft => 'numeric',
oprright => 'numeric', oprresult => 'bool', oprcom => '=(numeric,numeric)',
oprnegate => '<>(numeric,numeric)', oprcode => 'numeric_eq',
@@ -2077,7 +2077,7 @@
oprresult => 'bool', oprcom => '<>(numeric,numeric)',
oprnegate => '=(numeric,numeric)', oprcode => 'numeric_ne',
oprrest => 'neqsel', oprjoin => 'neqjoinsel' },
-{ oid => '1754', descr => 'less than',
+{ oid => '1754', oid_symbol => 'NumericLtOperator', descr => 'less than',
oprname => '<', oprleft => 'numeric', oprright => 'numeric',
oprresult => 'bool', oprcom => '>(numeric,numeric)',
oprnegate => '>=(numeric,numeric)', oprcode => 'numeric_lt',
@@ -3172,70 +3172,77 @@
{ oid => '3967', descr => 'get value from json as text with path elements',
oprname => '#>>', oprleft => 'json', oprright => '_text', oprresult => 'text',
oprcode => 'json_extract_path_text' },
-{ oid => '3211', descr => 'get jsonb object field',
+{ oid => '3211', oid_symbol => 'JsonbObjectFieldOperator',
+ descr => 'get jsonb object field',
oprname => '->', oprleft => 'jsonb', oprright => 'text', oprresult => 'jsonb',
oprcode => 'jsonb_object_field' },
-{ oid => '3477', descr => 'get jsonb object field as text',
+{ oid => '3477', oid_symbol => 'JsonbObjectFieldTextOperator',
+ descr => 'get jsonb object field as text',
oprname => '->>', oprleft => 'jsonb', oprright => 'text', oprresult => 'text',
oprcode => 'jsonb_object_field_text' },
-{ oid => '3212', descr => 'get jsonb array element',
+{ oid => '3212', oid_symbol => 'JsonbArrayElementOperator',
+ descr => 'get jsonb array element',
oprname => '->', oprleft => 'jsonb', oprright => 'int4', oprresult => 'jsonb',
oprcode => 'jsonb_array_element' },
-{ oid => '3481', descr => 'get jsonb array element as text',
+{ oid => '3481', oid_symbol => 'JsonbArrayElementTextOperator',
+ descr => 'get jsonb array element as text',
oprname => '->>', oprleft => 'jsonb', oprright => 'int4', oprresult => 'text',
oprcode => 'jsonb_array_element_text' },
-{ oid => '3213', descr => 'get value from jsonb with path elements',
+{ oid => '3213', oid_symbol => 'JsonbExtractPathOperator',
+ descr => 'get value from jsonb with path elements',
oprname => '#>', oprleft => 'jsonb', oprright => '_text',
oprresult => 'jsonb', oprcode => 'jsonb_extract_path' },
-{ oid => '3206', descr => 'get value from jsonb as text with path elements',
+{ oid => '3206', oid_symbol => 'JsonbExtractPathTextOperator',
+ descr => 'get value from jsonb as text with path elements',
oprname => '#>>', oprleft => 'jsonb', oprright => '_text',
oprresult => 'text', oprcode => 'jsonb_extract_path_text' },
-{ oid => '3240', descr => 'equal',
+{ oid => '3240', oid_symbol => 'JsonbEqOperator', descr => 'equal',
oprname => '=', oprcanmerge => 't', oprcanhash => 't', oprleft => 'jsonb',
oprright => 'jsonb', oprresult => 'bool', oprcom => '=(jsonb,jsonb)',
oprnegate => '<>(jsonb,jsonb)', oprcode => 'jsonb_eq', oprrest => 'eqsel',
oprjoin => 'eqjoinsel' },
-{ oid => '3241', descr => 'not equal',
+{ oid => '3241', oid_symbol => 'JsonbNeOperator', descr => 'not equal',
oprname => '<>', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '<>(jsonb,jsonb)', oprnegate => '=(jsonb,jsonb)',
oprcode => 'jsonb_ne', oprrest => 'neqsel', oprjoin => 'neqjoinsel' },
-{ oid => '3242', descr => 'less than',
+{ oid => '3242', oid_symbol => 'JsonbLtOperator', descr => 'less than',
oprname => '<', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '>(jsonb,jsonb)', oprnegate => '>=(jsonb,jsonb)',
oprcode => 'jsonb_lt', oprrest => 'scalarltsel',
oprjoin => 'scalarltjoinsel' },
-{ oid => '3243', descr => 'greater than',
+{ oid => '3243', oid_symbol => 'JsonbGtOperator', descr => 'greater than',
oprname => '>', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '<(jsonb,jsonb)', oprnegate => '<=(jsonb,jsonb)',
oprcode => 'jsonb_gt', oprrest => 'scalargtsel',
oprjoin => 'scalargtjoinsel' },
-{ oid => '3244', descr => 'less than or equal',
+{ oid => '3244', oid_symbol => 'JsonbLeOperator', descr => 'less than or equal',
oprname => '<=', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '>=(jsonb,jsonb)', oprnegate => '>(jsonb,jsonb)',
oprcode => 'jsonb_le', oprrest => 'scalarlesel',
oprjoin => 'scalarlejoinsel' },
-{ oid => '3245', descr => 'greater than or equal',
+{ oid => '3245', oid_symbol => 'JsonbGeOperator',
+ descr => 'greater than or equal',
oprname => '>=', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '<=(jsonb,jsonb)', oprnegate => '<(jsonb,jsonb)',
oprcode => 'jsonb_ge', oprrest => 'scalargesel',
oprjoin => 'scalargejoinsel' },
-{ oid => '3246', descr => 'contains',
+{ oid => '3246', oid_symbol => 'JsonbContainsOperator', descr => 'contains',
oprname => '@>', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '<@(jsonb,jsonb)', oprcode => 'jsonb_contains',
oprrest => 'matchingsel', oprjoin => 'matchingjoinsel' },
-{ oid => '3247', descr => 'key exists',
+{ oid => '3247', oid_symbol => 'JsonbExistsOperator', descr => 'key exists',
oprname => '?', oprleft => 'jsonb', oprright => 'text', oprresult => 'bool',
oprcode => 'jsonb_exists', oprrest => 'matchingsel',
oprjoin => 'matchingjoinsel' },
-{ oid => '3248', descr => 'any key exists',
+{ oid => '3248', oid_symbol => 'JsonbExistsAnyOperator', descr => 'any key exists',
oprname => '?|', oprleft => 'jsonb', oprright => '_text', oprresult => 'bool',
oprcode => 'jsonb_exists_any', oprrest => 'matchingsel',
oprjoin => 'matchingjoinsel' },
-{ oid => '3249', descr => 'all keys exist',
+{ oid => '3249', oid_symbol => 'JsonbExistsAllOperator', descr => 'all keys exist',
oprname => '?&', oprleft => 'jsonb', oprright => '_text', oprresult => 'bool',
oprcode => 'jsonb_exists_all', oprrest => 'matchingsel',
oprjoin => 'matchingjoinsel' },
-{ oid => '3250', descr => 'is contained by',
+{ oid => '3250', oid_symbol => 'JsonbContainedOperator', descr => 'is contained by',
oprname => '<@', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '@>(jsonb,jsonb)', oprcode => 'jsonb_contained',
oprrest => 'matchingsel', oprjoin => 'matchingjoinsel' },
--
2.25.1
[text/x-patch] 0004-Add-helper-jsonb-functions-and-macros.patch (9.3K, ../../[email protected]/6-0004-Add-helper-jsonb-functions-and-macros.patch)
download | inline diff:
From 98fe4515c9494ed6f41e40c909b1458017049832 Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Fri, 16 Dec 2016 17:16:47 +0300
Subject: [PATCH 4/6] Add helper jsonb functions and macros
---
src/backend/utils/adt/jsonb_util.c | 27 +++++
src/backend/utils/adt/jsonfuncs.c | 10 +-
src/include/utils/jsonb.h | 165 ++++++++++++++++++++++++++++-
3 files changed, 195 insertions(+), 7 deletions(-)
diff --git a/src/backend/utils/adt/jsonb_util.c b/src/backend/utils/adt/jsonb_util.c
index 291fb722e2d..9c78bf3e903 100644
--- a/src/backend/utils/adt/jsonb_util.c
+++ b/src/backend/utils/adt/jsonb_util.c
@@ -388,6 +388,22 @@ findJsonbValueFromContainer(JsonbContainer *container, uint32 flags,
return NULL;
}
+/*
+ * findJsonbValueFromContainer() wrapper that sets up JsonbValue key string.
+ */
+JsonbValue *
+findJsonbValueFromContainerLen(JsonbContainer *container, uint32 flags,
+ const char *key, uint32 keylen)
+{
+ JsonbValue k;
+
+ k.type = jbvString;
+ k.val.string.val = key;
+ k.val.string.len = keylen;
+
+ return findJsonbValueFromContainer(container, flags, &k);
+}
+
/*
* Find value by key in Jsonb object and fetch it into 'res', which is also
* returned.
@@ -602,6 +618,17 @@ pushJsonbValue(JsonbParseState **pstate, JsonbIteratorToken seq,
return pushJsonbValueScalar(pstate, seq, jbval);
}
+ /*
+ * XXX I'm not quite sure why we actually do this? Why do we need to change
+ * how JsonbValue is converted to Jsonb for the statistics patch?
+ */
+ /* push value from scalar container without its enclosing array */
+ if (*pstate && JsonbExtractScalar(jbval->val.binary.data, &v))
+ {
+ Assert(IsAJsonbScalar(&v));
+ return pushJsonbValueScalar(pstate, seq, &v);
+ }
+
/* unpack the binary and add each piece to the pstate */
it = JsonbIteratorInit(jbval->val.binary.data);
diff --git a/src/backend/utils/adt/jsonfuncs.c b/src/backend/utils/adt/jsonfuncs.c
index 0273f883d4f..ac03072839b 100644
--- a/src/backend/utils/adt/jsonfuncs.c
+++ b/src/backend/utils/adt/jsonfuncs.c
@@ -5371,7 +5371,8 @@ iterate_jsonb_values(Jsonb *jb, uint32 flags, void *state,
if (type == WJB_KEY)
{
if (flags & jtiKey)
- action(state, v.val.string.val, v.val.string.len);
+ action(state, unconstify(char *, v.val.string.val),
+ v.val.string.len);
continue;
}
@@ -5386,7 +5387,8 @@ iterate_jsonb_values(Jsonb *jb, uint32 flags, void *state,
{
case jbvString:
if (flags & jtiString)
- action(state, v.val.string.val, v.val.string.len);
+ action(state, unconstify(char *, v.val.string.val),
+ v.val.string.len);
break;
case jbvNumeric:
if (flags & jtiNumeric)
@@ -5508,7 +5510,9 @@ transform_jsonb_string_values(Jsonb *jsonb, void *action_state,
{
if ((type == WJB_VALUE || type == WJB_ELEM) && v.type == jbvString)
{
- out = transform_action(action_state, v.val.string.val, v.val.string.len);
+ out = transform_action(action_state,
+ unconstify(char *, v.val.string.val),
+ v.val.string.len);
v.val.string.val = VARDATA_ANY(out);
v.val.string.len = VARSIZE_ANY_EXHDR(out);
res = pushJsonbValue(&st, type, type < WJB_BEGIN_ARRAY ? &v : NULL);
diff --git a/src/include/utils/jsonb.h b/src/include/utils/jsonb.h
index 4cbe6edf218..e1ed712f26c 100644
--- a/src/include/utils/jsonb.h
+++ b/src/include/utils/jsonb.h
@@ -14,6 +14,7 @@
#include "lib/stringinfo.h"
#include "utils/array.h"
+#include "utils/builtins.h"
#include "utils/numeric.h"
/* Tokens used when sequentially processing a jsonb value */
@@ -229,8 +230,7 @@ typedef struct
#define JB_ROOT_IS_OBJECT(jbp_) ((*(uint32 *) VARDATA(jbp_) & JB_FOBJECT) != 0)
#define JB_ROOT_IS_ARRAY(jbp_) ((*(uint32 *) VARDATA(jbp_) & JB_FARRAY) != 0)
-
-enum jbvType
+typedef enum jbvType
{
/* Scalar types */
jbvNull = 0x0,
@@ -250,7 +250,7 @@ enum jbvType
* into JSON strings when outputted to json/jsonb.
*/
jbvDatetime = 0x20,
-};
+} JsonbValueType;
/*
* JsonbValue: In-memory representation of Jsonb. This is a convenient
@@ -269,7 +269,7 @@ struct JsonbValue
struct
{
int len;
- char *val; /* Not necessarily null-terminated */
+ const char *val; /* Not necessarily null-terminated */
} string; /* String primitive type */
struct
@@ -382,6 +382,10 @@ extern int compareJsonbContainers(JsonbContainer *a, JsonbContainer *b);
extern JsonbValue *findJsonbValueFromContainer(JsonbContainer *sheader,
uint32 flags,
JsonbValue *key);
+extern JsonbValue *findJsonbValueFromContainerLen(JsonbContainer *container,
+ uint32 flags,
+ const char *key,
+ uint32 keylen);
extern JsonbValue *getKeyJsonValueFromContainer(JsonbContainer *container,
const char *keyVal, int keyLen,
JsonbValue *res);
@@ -412,4 +416,157 @@ extern Datum jsonb_set_element(Jsonb *jb, Datum *path, int path_len,
JsonbValue *newval);
extern Datum jsonb_get_element(Jsonb *jb, Datum *path, int npath,
bool *isnull, bool as_text);
+
+/*
+ * XXX Not sure we want to add these functions to jsonb.h, which is the
+ * public API. Maybe it belongs rather to jsonb_typanalyze.c or elsewhere,
+ * closer to how it's used?
+ */
+
+/* helper inline functions for JsonbValue initialization */
+static inline JsonbValue *
+JsonValueInitObject(JsonbValue *val, int nPairs, int nPairsAllocated)
+{
+ val->type = jbvObject;
+ val->val.object.nPairs = nPairs;
+ val->val.object.pairs = nPairsAllocated ?
+ palloc(sizeof(JsonbPair) * nPairsAllocated) : NULL;
+
+ return val;
+}
+
+static inline JsonbValue *
+JsonValueInitArray(JsonbValue *val, int nElems, int nElemsAllocated,
+ bool rawScalar)
+{
+ val->type = jbvArray;
+ val->val.array.nElems = nElems;
+ val->val.array.elems = nElemsAllocated ?
+ palloc(sizeof(JsonbValue) * nElemsAllocated) : NULL;
+ val->val.array.rawScalar = rawScalar;
+
+ return val;
+}
+
+static inline JsonbValue *
+JsonValueInitBinary(JsonbValue *val, Jsonb *jb)
+{
+ val->type = jbvBinary;
+ val->val.binary.data = &(jb)->root;
+ val->val.binary.len = VARSIZE_ANY_EXHDR(jb);
+ return val;
+}
+
+
+static inline JsonbValue *
+JsonValueInitString(JsonbValue *jbv, const char *str)
+{
+ jbv->type = jbvString;
+ jbv->val.string.len = strlen(str);
+ jbv->val.string.val = memcpy(palloc(jbv->val.string.len + 1), str,
+ jbv->val.string.len + 1);
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitStringWithLen(JsonbValue *jbv, const char *str, int len)
+{
+ jbv->type = jbvString;
+ jbv->val.string.val = str;
+ jbv->val.string.len = len;
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitText(JsonbValue *jbv, text *txt)
+{
+ jbv->type = jbvString;
+ jbv->val.string.val = VARDATA_ANY(txt);
+ jbv->val.string.len = VARSIZE_ANY_EXHDR(txt);
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitNumeric(JsonbValue *jbv, Numeric num)
+{
+ jbv->type = jbvNumeric;
+ jbv->val.numeric = num;
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitInteger(JsonbValue *jbv, int64 i)
+{
+ jbv->type = jbvNumeric;
+ jbv->val.numeric = DatumGetNumeric(DirectFunctionCall1(
+ int8_numeric, Int64GetDatum(i)));
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitFloat(JsonbValue *jbv, float4 f)
+{
+ jbv->type = jbvNumeric;
+ jbv->val.numeric = DatumGetNumeric(DirectFunctionCall1(
+ float4_numeric, Float4GetDatum(f)));
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitDouble(JsonbValue *jbv, float8 f)
+{
+ jbv->type = jbvNumeric;
+ jbv->val.numeric = DatumGetNumeric(DirectFunctionCall1(
+ float8_numeric, Float8GetDatum(f)));
+ return jbv;
+}
+
+/* helper macros for jsonb building */
+#define pushJsonbKey(pstate, jbv, key) \
+ pushJsonbValue(pstate, WJB_KEY, JsonValueInitString(jbv, key))
+
+#define pushJsonbValueGeneric(Type, pstate, jbv, val) \
+ pushJsonbValue(pstate, WJB_VALUE, JsonValueInit##Type(jbv, val))
+
+#define pushJsonbElemGeneric(Type, pstate, jbv, val) \
+ pushJsonbValue(pstate, WJB_ELEM, JsonValueInit##Type(jbv, val))
+
+#define pushJsonbValueInteger(pstate, jbv, i) \
+ pushJsonbValueGeneric(Integer, pstate, jbv, i)
+
+#define pushJsonbValueFloat(pstate, jbv, f) \
+ pushJsonbValueGeneric(Float, pstate, jbv, f)
+
+#define pushJsonbElemFloat(pstate, jbv, f) \
+ pushJsonbElemGeneric(Float, pstate, jbv, f)
+
+#define pushJsonbElemString(pstate, jbv, txt) \
+ pushJsonbElemGeneric(String, pstate, jbv, txt)
+
+#define pushJsonbElemText(pstate, jbv, txt) \
+ pushJsonbElemGeneric(Text, pstate, jbv, txt)
+
+#define pushJsonbElemNumeric(pstate, jbv, num) \
+ pushJsonbElemGeneric(Numeric, pstate, jbv, num)
+
+#define pushJsonbElemInteger(pstate, jbv, num) \
+ pushJsonbElemGeneric(Integer, pstate, jbv, num)
+
+#define pushJsonbElemBinary(pstate, jbv, jbcont) \
+ pushJsonbElemGeneric(Binary, pstate, jbv, jbcont)
+
+#define pushJsonbKeyValueGeneric(Type, pstate, jbv, key, val) ( \
+ pushJsonbKey(pstate, jbv, key), \
+ pushJsonbValueGeneric(Type, pstate, jbv, val) \
+ )
+
+#define pushJsonbKeyValueString(pstate, jbv, key, val) \
+ pushJsonbKeyValueGeneric(String, pstate, jbv, key, val)
+
+#define pushJsonbKeyValueFloat(pstate, jbv, key, val) \
+ pushJsonbKeyValueGeneric(Float, pstate, jbv, key, val)
+
+#define pushJsonbKeyValueInteger(pstate, jbv, key, val) \
+ pushJsonbKeyValueGeneric(Integer, pstate, jbv, key, val)
+
#endif /* __JSONB_H__ */
--
2.25.1
[text/x-patch] 0005-Export-scalarineqsel.patch (1.5K, ../../[email protected]/7-0005-Export-scalarineqsel.patch)
download | inline diff:
From 45b82feed6e866bd337d036f68cdf46b6827d0b8 Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Sat, 12 Nov 2016 19:01:43 +0300
Subject: [PATCH 5/6] Export scalarineqsel()
---
src/backend/utils/adt/selfuncs.c | 2 +-
src/include/utils/selfuncs.h | 4 ++++
2 files changed, 5 insertions(+), 1 deletion(-)
diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c
index f534e862079..6ef31fdd79c 100644
--- a/src/backend/utils/adt/selfuncs.c
+++ b/src/backend/utils/adt/selfuncs.c
@@ -573,7 +573,7 @@ neqsel(PG_FUNCTION_ARGS)
* it will return an approximate estimate based on assuming that the constant
* value falls in the middle of the bin identified by binary search.
*/
-static double
+double
scalarineqsel(PlannerInfo *root, Oid operator, bool isgt, bool iseq,
Oid collation,
VariableStatData *vardata, Datum constval, Oid consttype)
diff --git a/src/include/utils/selfuncs.h b/src/include/utils/selfuncs.h
index 49fc3d717b1..88a95401c5f 100644
--- a/src/include/utils/selfuncs.h
+++ b/src/include/utils/selfuncs.h
@@ -249,6 +249,10 @@ extern List *add_predicate_to_index_quals(IndexOptInfo *index,
extern void genericcostestimate(PlannerInfo *root, IndexPath *path,
double loop_count,
GenericCosts *costs);
+extern double scalarineqsel(PlannerInfo *root, Oid operator, bool isgt,
+ bool iseq, Oid collation,
+ VariableStatData *vardata, Datum constval,
+ Oid consttype);
extern HeapTuple stats_form_tuple(StatsData *data);
--
2.25.1
[text/x-patch] 0006-Add-jsonb-statistics.patch (131.0K, ../../[email protected]/8-0006-Add-jsonb-statistics.patch)
download | inline diff:
From 485f4cc1c0e1d16cb809dedc3a2819780d1a2a23 Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Sat, 12 Nov 2016 19:19:33 +0300
Subject: [PATCH 6/6] Add jsonb statistics
---
src/backend/catalog/system_functions.sql | 36 +
src/backend/catalog/system_views.sql | 56 +
src/backend/utils/adt/Makefile | 2 +
src/backend/utils/adt/jsonb_selfuncs.c | 1582 +++++++++++++++++++++
src/backend/utils/adt/jsonb_typanalyze.c | 1360 ++++++++++++++++++
src/backend/utils/adt/jsonpath_exec.c | 2 +-
src/include/catalog/pg_operator.dat | 17 +-
src/include/catalog/pg_proc.dat | 11 +
src/include/catalog/pg_statistic.h | 2 +
src/include/catalog/pg_type.dat | 2 +-
src/include/utils/json_selfuncs.h | 113 ++
src/test/regress/expected/jsonb_stats.out | 713 ++++++++++
src/test/regress/expected/rules.out | 32 +
src/test/regress/parallel_schedule | 2 +-
src/test/regress/sql/jsonb_stats.sql | 249 ++++
15 files changed, 4168 insertions(+), 11 deletions(-)
create mode 100644 src/backend/utils/adt/jsonb_selfuncs.c
create mode 100644 src/backend/utils/adt/jsonb_typanalyze.c
create mode 100644 src/include/utils/json_selfuncs.h
create mode 100644 src/test/regress/expected/jsonb_stats.out
create mode 100644 src/test/regress/sql/jsonb_stats.sql
diff --git a/src/backend/catalog/system_functions.sql b/src/backend/catalog/system_functions.sql
index fd1421788e6..fdf236e4edc 100644
--- a/src/backend/catalog/system_functions.sql
+++ b/src/backend/catalog/system_functions.sql
@@ -594,6 +594,42 @@ LANGUAGE internal
STRICT IMMUTABLE PARALLEL SAFE
AS 'unicode_is_normalized';
+-- XXX is this function immutable / parallel safe?
+-- XXX do we actually need to cast to text and then to jsonb?
+CREATE FUNCTION pg_json_path_stats(tab regclass, path_index integer) RETURNS text
+AS $$
+ SELECT jsonb_pretty((
+ CASE
+ WHEN stakind1 = 8 THEN stavalues1
+ WHEN stakind2 = 8 THEN stavalues2
+ WHEN stakind3 = 8 THEN stavalues3
+ WHEN stakind4 = 8 THEN stavalues4
+ WHEN stakind5 = 8 THEN stavalues5
+ END::text::jsonb[])[$2])
+ FROM pg_statistic
+ WHERE starelid = $1
+$$ LANGUAGE 'sql';
+
+-- XXX is this function immutable / parallel safe?
+-- XXX do we actually need to cast to text and then to jsonb?
+CREATE FUNCTION pg_json_path_stats(tab regclass, path text) RETURNS text
+AS $$
+ SELECT jsonb_pretty(pathstats)
+ FROM (
+ SELECT unnest(
+ CASE
+ WHEN stakind1 = 8 THEN stavalues1
+ WHEN stakind2 = 8 THEN stavalues2
+ WHEN stakind3 = 8 THEN stavalues3
+ WHEN stakind4 = 8 THEN stavalues4
+ WHEN stakind5 = 8 THEN stavalues5
+ END::text::jsonb[]) pathstats
+ FROM pg_statistic
+ WHERE starelid = $1
+ ) paths
+ WHERE pathstats->>'path' = $2
+$$ LANGUAGE 'sql';
+
--
-- The default permissions for functions mean that anyone can execute them.
-- A number of functions shouldn't be executable by just anyone, but rather
diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sql
index 3cb69b1f87b..6fafa7aef31 100644
--- a/src/backend/catalog/system_views.sql
+++ b/src/backend/catalog/system_views.sql
@@ -364,6 +364,62 @@ CREATE VIEW pg_stats_ext_exprs WITH (security_barrier) AS
-- unprivileged users may read pg_statistic_ext but not pg_statistic_ext_data
REVOKE ALL ON pg_statistic_ext_data FROM public;
+-- XXX This probably needs to do the same checks as pg_stats, i.e.
+-- WHERE NOT attisdropped
+-- AND has_column_privilege(c.oid, a.attnum, 'select')
+-- AND (c.relrowsecurity = false OR NOT row_security_active(c.oid));
+CREATE VIEW pg_stats_json AS
+ SELECT
+ nspname AS schemaname,
+ relname AS tablename,
+ attname AS attname,
+
+ path->>'path' AS json_path,
+
+ stainherit AS inherited,
+
+ (path->'json'->>'nullfrac')::float4 AS null_frac,
+ (path->'json'->>'width')::float4 AS avg_width,
+ (path->'json'->>'distinct')::float4 AS n_distinct,
+
+ ARRAY(SELECT val FROM jsonb_array_elements(
+ path->'json'->'mcv'->'values') val)::anyarray
+ AS most_common_vals,
+
+ ARRAY(SELECT num::text::float4 FROM jsonb_array_elements(
+ path->'json'->'mcv'->'numbers') num)
+ AS most_common_freqs,
+
+ ARRAY(SELECT val FROM jsonb_array_elements(
+ path->'json'->'histogram'->'values') val)
+ AS histogram_bounds,
+
+ ARRAY(SELECT val::text::int FROM jsonb_array_elements(
+ path->'array_length'->'mcv'->'values') val)
+ AS most_common_array_lengths,
+
+ ARRAY(SELECT num::text::float4 FROM jsonb_array_elements(
+ path->'array_length'->'mcv'->'numbers') num)
+ AS most_common_array_length_freqs,
+
+ (path->'json'->>'correlation')::float4 AS correlation
+
+ FROM
+ pg_statistic s JOIN pg_class c ON (c.oid = s.starelid)
+ JOIN pg_attribute a ON (c.oid = attrelid AND attnum = s.staattnum)
+ LEFT JOIN pg_namespace n ON (n.oid = c.relnamespace),
+ LATERAL (
+ SELECT unnest((CASE
+ WHEN stakind1 = 8 THEN stavalues1
+ WHEN stakind2 = 8 THEN stavalues2
+ WHEN stakind3 = 8 THEN stavalues3
+ WHEN stakind4 = 8 THEN stavalues4
+ WHEN stakind5 = 8 THEN stavalues5
+ END ::text::jsonb[])[2:]) AS path
+ ) paths;
+
+-- no need to revoke any privileges, we've already revoked accss to pg_statistic
+
CREATE VIEW pg_publication_tables AS
SELECT
P.pubname AS pubname,
diff --git a/src/backend/utils/adt/Makefile b/src/backend/utils/adt/Makefile
index 41b486bceff..5e359ccf4fb 100644
--- a/src/backend/utils/adt/Makefile
+++ b/src/backend/utils/adt/Makefile
@@ -50,6 +50,8 @@ OBJS = \
jsonb.o \
jsonb_gin.o \
jsonb_op.o \
+ jsonb_selfuncs.o \
+ jsonb_typanalyze.o \
jsonb_util.o \
jsonfuncs.o \
jsonbsubs.o \
diff --git a/src/backend/utils/adt/jsonb_selfuncs.c b/src/backend/utils/adt/jsonb_selfuncs.c
new file mode 100644
index 00000000000..3b6d1406566
--- /dev/null
+++ b/src/backend/utils/adt/jsonb_selfuncs.c
@@ -0,0 +1,1582 @@
+/*-------------------------------------------------------------------------
+ *
+ * jsonb_selfuncs.c
+ * Functions for selectivity estimation of jsonb operators
+ *
+ * Copyright (c) 2016-2022, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ * src/backend/utils/adt/jsonb_selfuncs.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#include "postgres.h"
+
+#include <math.h>
+
+#include "fmgr.h"
+#include "access/heapam.h"
+#include "access/htup_details.h"
+#include "catalog/pg_operator.h"
+#include "catalog/pg_statistic.h"
+#include "catalog/pg_type.h"
+#include "common/string.h"
+#include "nodes/primnodes.h"
+#include "utils/builtins.h"
+#include "utils/json.h"
+#include "utils/jsonb.h"
+#include "utils/json_selfuncs.h"
+#include "utils/lsyscache.h"
+#include "utils/rel.h"
+#include "utils/selfuncs.h"
+
+#define DEFAULT_JSON_CONTAINS_SEL 0.001
+
+/*
+ * jsonGetField
+ * Given a JSONB document and a key, extract the JSONB value for the key.
+ */
+static inline Datum
+jsonGetField(Datum obj, const char *field)
+{
+ Jsonb *jb = DatumGetJsonbP(obj);
+ JsonbValue *jbv = findJsonbValueFromContainerLen(&jb->root, JB_FOBJECT,
+ field, strlen(field));
+ return jbv ? JsonbPGetDatum(JsonbValueToJsonb(jbv)) : PointerGetDatum(NULL);
+}
+
+/*
+ * jsonGetFloat4
+ * Given a JSONB value, interpret it as a float4 value.
+ *
+ * This expects the JSONB value to be a numeric, because that's how we store
+ * floats in JSONB, and we cast it to float4.
+ */
+static inline float4
+jsonGetFloat4(Datum jsonb, float4 default_val)
+{
+ Jsonb *jb;
+ JsonbValue jv;
+
+ if (!DatumGetPointer(jsonb))
+ return default_val;
+
+ jb = DatumGetJsonbP(jsonb);
+
+ if (!JsonbExtractScalar(&jb->root, &jv) || jv.type != jbvNumeric)
+ return default_val;
+
+ return DatumGetFloat4(DirectFunctionCall1(numeric_float4,
+ NumericGetDatum(jv.val.numeric)));
+}
+
+/*
+ * jsonStatsInit
+ * Given a pg_statistic tuple, expand STATISTIC_KIND_JSON into JsonStats.
+ */
+bool
+jsonStatsInit(JsonStats data, const VariableStatData *vardata)
+{
+ Jsonb *jb;
+ JsonbValue prefix;
+
+ if (!vardata->statsTuple)
+ return false;
+
+ data->statsTuple = vardata->statsTuple;
+ memset(&data->attslot, 0, sizeof(data->attslot));
+
+ /* Were there just NULL values in the column? No JSON stats, but still useful. */
+ if (((Form_pg_statistic) GETSTRUCT(data->statsTuple))->stanullfrac >= 1.0)
+ {
+ data->nullfrac = 1.0;
+ return true;
+ }
+
+ /* Do we have the JSON stats built in the pg_statistic? */
+ if (!get_attstatsslot(&data->attslot, data->statsTuple,
+ STATISTIC_KIND_JSON, InvalidOid,
+ ATTSTATSSLOT_NUMBERS | ATTSTATSSLOT_VALUES))
+ return false;
+
+ /*
+ * Valid JSON stats should have at least 2 elements in values:
+ * 0th - root path prefix
+ * 1st - root path stats
+ */
+ if (data->attslot.nvalues < 2)
+ {
+ free_attstatsslot(&data->attslot);
+ return false;
+ }
+
+ /* XXX If the ACL check was not OK, would we even get here? */
+ data->acl_ok = vardata->acl_ok;
+ data->rel = vardata->rel;
+ data->nullfrac =
+ data->attslot.nnumbers > 0 ? data->attslot.numbers[0] : 0.0;
+ data->pathdatums = data->attslot.values + 1;
+ data->npaths = data->attslot.nvalues - 1;
+
+ /* Extract root path prefix */
+ jb = DatumGetJsonbP(data->attslot.values[0]);
+ if (!JsonbExtractScalar(&jb->root, &prefix) || prefix.type != jbvString)
+ {
+ free_attstatsslot(&data->attslot);
+ return false;
+ }
+
+ data->prefix = prefix.val.string.val;
+ data->prefixlen = prefix.val.string.len;
+
+ /* Create path cache, initialze only two fields that acting as flags */
+ data->paths = palloc(sizeof(*data->paths) * data->npaths);
+
+ for (int i = 0; i < data->npaths; i++)
+ {
+ data->paths[i].data = NULL;
+ data->paths[i].path = NULL;
+ }
+
+ return true;
+}
+
+/*
+ * jsonStatsRelease
+ * Release resources (statistics slot) associated with the JsonStats value.
+ */
+void
+jsonStatsRelease(JsonStats data)
+{
+ free_attstatsslot(&data->attslot);
+}
+
+/*
+ * jsonPathStatsAllocSpecialStats
+ * Allocate a copy of JsonPathStats for accessing special (length etc.)
+ * stats for a given JSON path.
+ */
+static JsonPathStats
+jsonPathStatsAllocSpecialStats(JsonPathStats pstats, JsonPathStatsType type)
+{
+ JsonPathStats stats;
+
+ if (!pstats)
+ return NULL;
+
+ /* copy and replace stats type */
+ stats = palloc(sizeof(*stats));
+ *stats = *pstats;
+ stats->type = type;
+
+ return stats;
+}
+
+/*
+ * jsonPathStatsGetArrayLengthStats
+ * Extract statistics of array lengths for the path.
+ */
+JsonPathStats
+jsonPathStatsGetArrayLengthStats(JsonPathStats pstats)
+{
+ /*
+ * The array length statistics is relevant only for values that are arrays.
+ * So if we observed no such values, we know there can't be such
+ * statistics and so we simply return NULL.
+ */
+ if (jsonPathStatsGetTypeFreq(pstats, jbvArray, 0.0) <= 0.0)
+ return NULL;
+
+ return jsonPathStatsAllocSpecialStats(pstats, JsonPathStatsArrayLength);
+}
+
+/*
+ * jsonPathStatsGetObjectLengthStats
+ * Extract statistics of object length for the path.
+ */
+JsonPathStats
+jsonPathStatsGetObjectLengthStats(JsonPathStats pstats)
+{
+ /*
+ * The object length statistics is relevant only for values that are arrays.
+ * So if we observed no such values, we know there can't be such
+ * statistics and so we simply return NULL.
+ */
+ if (jsonPathStatsGetTypeFreq(pstats, jbvObject, 0.0) <= 0.0)
+ return NULL;
+
+ return jsonPathStatsAllocSpecialStats(pstats, JsonPathStatsObjectLength);
+}
+
+/*
+ * jsonPathStatsGetPath
+ * Try to use cached path name or extract it from per-path stats datum.
+ *
+ * Returns true on succces, false on error.
+ */
+static inline bool
+jsonPathStatsGetPath(JsonPathStats stats, Datum pathdatum,
+ const char **path, int *pathlen)
+{
+ *path = stats->path;
+
+ if (*path)
+ /* use cached path name */
+ *pathlen = stats->pathlen;
+ else
+ {
+ Jsonb *jsonb = DatumGetJsonbP(pathdatum);
+ JsonbValue pathkey;
+ JsonbValue *pathval;
+
+ /* extract path from the statistics represented as jsonb document */
+ JsonValueInitStringWithLen(&pathkey, "path", 4);
+ pathval = findJsonbValueFromContainer(&jsonb->root, JB_FOBJECT, &pathkey);
+
+ if (!pathval || pathval->type != jbvString)
+ return false; /* XXX invalid stats data, maybe throw error */
+
+ /* cache extracted path name */
+ *path = stats->path = pathval->val.string.val;
+ *pathlen = stats->pathlen = pathval->val.string.len;
+ }
+
+ return true;
+}
+
+/* Context for bsearch()ing paths */
+typedef struct JsonPathStatsSearchContext
+{
+ JsonStats stats;
+ const char *path;
+ int pathlen;
+} JsonPathStatsSearchContext;
+
+/*
+ * jsonPathStatsCompare
+ * Compare two JsonPathStats structs, so that we can sort them.
+ *
+ * We do this so that we can search for stats for a given path simply by
+ * bsearch().
+ *
+ * XXX We never build two structs for the same path, so we know the paths
+ * are different - one may be a prefix of the other, but then we sort the
+ * strings by length.
+ */
+static int
+jsonPathStatsCompare(const void *pv1, const void *pv2)
+{
+ JsonPathStatsSearchContext const *cxt = pv1;
+ Datum const *pathdatum = (Datum const *) pv2;
+ int index = pathdatum - cxt->stats->pathdatums;
+ JsonPathStats stats = &cxt->stats->paths[index];
+ const char *path;
+ int pathlen;
+ int res;
+
+ if (!jsonPathStatsGetPath(stats, *pathdatum, &path, &pathlen))
+ return 1; /* XXX invalid stats data */
+
+ /* compare the shared part first, then compare by length */
+ res = strncmp(cxt->path, path, Min(cxt->pathlen, pathlen));
+
+ return res ? res : cxt->pathlen - pathlen;
+}
+
+/*
+ * jsonStatsFindPath
+ * Find stats for a given path.
+ *
+ * The stats are sorted by path, so we can simply do bsearch().
+ * This is low-level function and jsdata->prefix is not considered, the caller
+ * should handle it by itself.
+ */
+static JsonPathStats
+jsonStatsFindPath(JsonStats jsdata, const char *path, int pathlen)
+{
+ JsonPathStatsSearchContext cxt;
+ JsonPathStats stats;
+ Datum *pdatum;
+ int index;
+
+ cxt.stats = jsdata;
+ cxt.path = path;
+ cxt.pathlen = pathlen;
+
+ pdatum = bsearch(&cxt, jsdata->pathdatums, jsdata->npaths,
+ sizeof(*jsdata->pathdatums), jsonPathStatsCompare);
+
+ if (!pdatum)
+ return NULL;
+
+ index = pdatum - jsdata->pathdatums;
+ stats = &jsdata->paths[index];
+
+ Assert(stats->path);
+ Assert(stats->pathlen == pathlen);
+
+ /* Init all fields if needed (stats->data == NULL means uninitialized) */
+ if (!stats->data)
+ {
+ stats->data = jsdata;
+ stats->datum = pdatum;
+ stats->type = JsonPathStatsValues;
+ }
+
+ return stats;
+}
+
+/*
+ * jsonStatsGetPathByStr
+ * Find stats for a given path string considering jsdata->prefix.
+ */
+JsonPathStats
+jsonStatsGetPathByStr(JsonStats jsdata, const char *subpath, int subpathlen)
+{
+ JsonPathStats stats;
+ char *path;
+ int pathlen;
+
+ if (jsdata->nullfrac >= 1.0)
+ return NULL;
+
+ pathlen = jsdata->prefixlen + subpathlen - 1;
+ path = palloc(pathlen);
+
+ memcpy(path, jsdata->prefix, jsdata->prefixlen);
+ memcpy(&path[jsdata->prefixlen], &subpath[1], subpathlen - 1);
+
+ stats = jsonStatsFindPath(jsdata, path, pathlen);
+
+ if (!stats)
+ pfree(path);
+
+ return stats;
+}
+
+/*
+ * jsonStatsGetRootPath
+ * Find JSON stats for root prefix path.
+ */
+static JsonPathStats
+jsonStatsGetRootPath(JsonStats jsdata)
+{
+ if (jsdata->nullfrac >= 1.0)
+ return NULL;
+
+ return jsonStatsFindPath(jsdata, jsdata->prefix, jsdata->prefixlen);
+}
+
+#define jsonStatsGetRootArrayPath(jsdata) \
+ jsonStatsGetPathByStr(jsdata, JSON_PATH_ROOT_ARRAY, JSON_PATH_ROOT_ARRAY_LEN)
+
+/*
+ * jsonPathAppendEntry
+ * Append entry (represented as simple string) to a path.
+ *
+ * NULL entry is treated as wildcard array accessor "[*]".
+ */
+void
+jsonPathAppendEntry(StringInfo path, const char *entry)
+{
+ if (entry)
+ {
+ appendStringInfoCharMacro(path, '.');
+ escape_json(path, entry);
+ }
+ else
+ appendStringInfoString(path, "[*]");
+}
+
+/*
+ * jsonPathAppendEntryWithLen
+ * Append string (represented as string + length) to a path.
+ */
+static void
+jsonPathAppendEntryWithLen(StringInfo path, const char *entry, int len)
+{
+ char *tmpentry = pnstrdup(entry, len);
+ jsonPathAppendEntry(path, tmpentry);
+ pfree(tmpentry);
+}
+
+/*
+ * jsonPathStatsGetSubpath
+ * Find JSON path stats for object key or array elements (if 'key' = NULL).
+ */
+JsonPathStats
+jsonPathStatsGetSubpath(JsonPathStats pstats, const char *key)
+{
+ JsonPathStats spstats;
+ StringInfoData str;
+
+ initStringInfo(&str);
+ appendBinaryStringInfo(&str, pstats->path, pstats->pathlen);
+ jsonPathAppendEntry(&str, key);
+
+ spstats = jsonStatsFindPath(pstats->data, str.data, str.len);
+ if (!spstats)
+ pfree(str.data);
+
+ return spstats;
+}
+
+/*
+ * jsonPathStatsGetArrayIndexSelectivity
+ * Given stats for a path, determine selectivity for an array index.
+ */
+Selectivity
+jsonPathStatsGetArrayIndexSelectivity(JsonPathStats pstats, int index)
+{
+ JsonPathStats lenstats = jsonPathStatsGetArrayLengthStats(pstats);
+ JsonbValue tmpjbv;
+ Jsonb *jb;
+
+ /*
+ * If we have no array length stats, assume all documents match.
+ *
+ * XXX Shouldn't this use a default smaller than 1.0? What do the selfuncs
+ * for regular arrays use?
+ */
+ if (!lenstats)
+ return 1.0;
+
+ jb = JsonbValueToJsonb(JsonValueInitInteger(&tmpjbv, index));
+
+ /* calculate fraction of elements smaller than the index */
+ return jsonSelectivity(lenstats, JsonbPGetDatum(jb), JsonbGtOperator);
+}
+
+/*
+ * jsonStatsGetPath
+ * Find JSON statistics for a given path.
+ *
+ * 'path' is an array of text datums of length 'pathlen' (can be zero).
+ */
+static JsonPathStats
+jsonStatsGetPath(JsonStats jsdata, Datum *path, int pathlen,
+ bool try_arrays_indexes, float4 *nullfrac)
+{
+ JsonPathStats pstats = jsonStatsGetRootPath(jsdata);
+ Selectivity sel = 1.0;
+
+ for (int i = 0; pstats && i < pathlen; i++)
+ {
+ char *key = TextDatumGetCString(path[i]);
+ char *tail;
+ int index;
+
+ if (!try_arrays_indexes)
+ {
+ /* Find object key stats */
+ pstats = jsonPathStatsGetSubpath(pstats, key);
+ pfree(key);
+ continue;
+ }
+
+ /* Try to interpret path entry as integer array index */
+ errno = 0;
+ index = strtoint(key, &tail, 10);
+
+ if (tail == key || *tail != '\0' || errno != 0)
+ {
+ /* Find object key stats */
+ pstats = jsonPathStatsGetSubpath(pstats, key);
+ }
+ else
+ {
+ /* Find array index stats */
+ /* FIXME consider object key "index" also */
+ JsonPathStats arrstats = jsonPathStatsGetSubpath(pstats, NULL);
+
+ if (arrstats)
+ {
+ float4 arrfreq = jsonPathStatsGetFreq(pstats, 0.0);
+
+ sel *= jsonPathStatsGetArrayIndexSelectivity(pstats, index);
+
+ if (arrfreq > 0.0)
+ sel /= arrfreq;
+ }
+
+ pstats = arrstats;
+ }
+
+ pfree(key);
+ }
+
+ *nullfrac = 1.0 - sel;
+
+ return pstats;
+}
+
+/*
+ * jsonPathStatsGetNextSubpathStats
+ * Iterate all collected subpaths of a given path.
+ *
+ * This function can be useful for estimation of selectivity of jsonpath
+ * '.*' and '.**' operators.
+ *
+ * The next found subpath is written into *pkeystats, which should be set to
+ * NULL before the first call.
+ *
+ * If keysOnly is true, emit only top-level object-key subpaths.
+ *
+ * Returns false on the end of iteration and true otherwise.
+ */
+bool
+jsonPathStatsGetNextSubpathStats(JsonPathStats stats, JsonPathStats *pkeystats,
+ bool keysOnly)
+{
+ JsonPathStats keystats = *pkeystats;
+ /* compute next index */
+ int index =
+ (keystats ? keystats->datum : stats->datum) - stats->data->pathdatums + 1;
+
+ if (stats->type != JsonPathStatsValues)
+ return false; /* length stats doe not have subpaths */
+
+ for (; index < stats->data->npaths; index++)
+ {
+ Datum *pathdatum = &stats->data->pathdatums[index];
+ const char *path;
+ int pathlen;
+
+ keystats = &stats->data->paths[index];
+
+ if (!jsonPathStatsGetPath(keystats, *pathdatum, &path, &pathlen))
+ break; /* invalid path stats */
+
+ /* Break, if subpath does not start from a desired prefix */
+ if (pathlen <= stats->pathlen ||
+ memcmp(path, stats->path, stats->pathlen))
+ break;
+
+ if (keysOnly)
+ {
+ const char *c = &path[stats->pathlen];
+
+ if (*c == '[')
+ {
+ Assert(c[1] == '*' && c[2] == ']');
+
+#if 0 /* TODO add separate flag for requesting top-level array accessors */
+ /* skip if it is not last key in the path */
+ if (pathlen > stats->pathlen + 3)
+#endif
+ continue; /* skip array accessors */
+ }
+ else if (*c == '.')
+ {
+ /* find end of '."key"' */
+ const char *pathend = path + pathlen;
+
+ if (++c >= pathend || *c != '"')
+ break; /* invalid path */
+
+ while (++c <= pathend && *c != '"')
+ if (*c == '\\') /* handle escaped chars */
+ c++;
+
+ if (c > pathend)
+ break; /* invalid path */
+
+ /* skip if it is not last key in the path */
+ if (c < pathend)
+ continue;
+ }
+ else
+ continue; /* invalid path */
+ }
+
+ /* Init path stats if needed */
+ if (!stats->data)
+ {
+ keystats->data = stats->data;
+ keystats->datum = pathdatum;
+ keystats->type = JsonPathStatsValues;
+ }
+
+ *pkeystats = keystats;
+
+ return true;
+ }
+
+ return false;
+}
+
+/*
+ * jsonStatsConvertArray
+ * Convert a JSONB array into an array of some regular data type.
+ *
+ * The "type" identifies what elements are in the input JSONB array, while
+ * typid determines the target type.
+ */
+static Datum
+jsonStatsConvertArray(Datum jsonbValueArray, JsonStatType type, Oid typid,
+ float4 multiplier)
+{
+ Datum *values;
+ Jsonb *jbvals;
+ JsonbValue jbv;
+ JsonbIterator *it;
+ JsonbIteratorToken r;
+ int nvalues;
+ int i;
+ int16 typlen;
+ bool typbyval;
+ char typalign;
+
+ if (!DatumGetPointer(jsonbValueArray))
+ return PointerGetDatum(NULL);
+
+ jbvals = DatumGetJsonbP(jsonbValueArray);
+
+ nvalues = JsonContainerSize(&jbvals->root);
+
+ values = palloc(sizeof(Datum) * nvalues);
+
+ for (i = 0, it = JsonbIteratorInit(&jbvals->root);
+ (r = JsonbIteratorNext(&it, &jbv, true)) != WJB_DONE;)
+ {
+ if (r == WJB_ELEM)
+ {
+ Datum value;
+
+ switch (type)
+ {
+ case JsonStatJsonb:
+ case JsonStatJsonbWithoutSubpaths:
+ value = JsonbPGetDatum(JsonbValueToJsonb(&jbv));
+ break;
+
+ case JsonStatText:
+ case JsonStatString:
+ Assert(jbv.type == jbvString);
+ value = PointerGetDatum(
+ cstring_to_text_with_len(jbv.val.string.val,
+ jbv.val.string.len));
+ break;
+
+ case JsonStatNumeric:
+ Assert(jbv.type == jbvNumeric);
+ value = NumericGetDatum(jbv.val.numeric);
+ break;
+
+ case JsonStatFloat4:
+ Assert(jbv.type == jbvNumeric);
+ value = DirectFunctionCall1(numeric_float4,
+ NumericGetDatum(jbv.val.numeric));
+ value = Float4GetDatum(DatumGetFloat4(value) * multiplier);
+ break;
+
+ default:
+ elog(ERROR, "invalid json stat type %d", type);
+ value = (Datum) 0;
+ break;
+ }
+
+ Assert(i < nvalues);
+ values[i++] = value;
+ }
+ }
+
+ Assert(i == nvalues);
+
+ get_typlenbyvalalign(typid, &typlen, &typbyval, &typalign);
+
+ return PointerGetDatum(
+ construct_array(values, nvalues, typid, typlen, typbyval, typalign));
+}
+
+/*
+ * jsonPathStatsExtractData
+ * Extract pg_statistics values from statistics for a single path.
+ *
+ * Extract ordinary MCV, Histogram, Correlation slots for a requested stats
+ * type. If requested stats for JSONB, include also transformed JSON slot for
+ * a path and possibly for its subpaths.
+ */
+static bool
+jsonPathStatsExtractData(JsonPathStats pstats, JsonStatType stattype,
+ float4 nullfrac, StatsData *statdata)
+{
+ Datum data;
+ Datum nullf;
+ Datum dist;
+ Datum width;
+ Datum mcv;
+ Datum hst;
+ Datum corr;
+ Oid type;
+ Oid eqop;
+ Oid ltop;
+ const char *key;
+ StatsSlot *slot = statdata->slots;
+
+ nullfrac = 1.0 - (1.0 - pstats->data->nullfrac) * (1.0 - nullfrac);
+
+ /*
+ * Depending on requested statistics type, select:
+ * - stavalues data type
+ * - corresponding eq/lt operators
+ * - JSONB field, containing stats slots for this statistics type
+ */
+ switch (stattype)
+ {
+ case JsonStatJsonb:
+ case JsonStatJsonbWithoutSubpaths:
+ key = pstats->type == JsonPathStatsArrayLength ? "array_length" :
+ pstats->type == JsonPathStatsObjectLength ? "object_length" :
+ "json";
+ type = JSONBOID;
+ eqop = JsonbEqOperator;
+ ltop = JsonbLtOperator;
+ break;
+ case JsonStatText:
+ key = "text";
+ type = TEXTOID;
+ eqop = TextEqualOperator;
+ ltop = TextLessOperator;
+ break;
+ case JsonStatString:
+ key = "string";
+ type = TEXTOID;
+ eqop = TextEqualOperator;
+ ltop = TextLessOperator;
+ break;
+ case JsonStatNumeric:
+ key = "numeric";
+ type = NUMERICOID;
+ eqop = NumericEqOperator;
+ ltop = NumericLtOperator;
+ break;
+ case JsonStatFloat4: /* special internal stats type */
+ default:
+ elog(ERROR, "invalid json statistic type %d", stattype);
+ break;
+ }
+
+ /* Extract object containing slots */
+ data = jsonGetField(*pstats->datum, key);
+
+ if (!DatumGetPointer(data))
+ return false;
+
+ nullf = jsonGetField(data, "nullfrac");
+ dist = jsonGetField(data, "distinct");
+ width = jsonGetField(data, "width");
+ mcv = jsonGetField(data, "mcv");
+ hst = jsonGetField(data, "histogram");
+ corr = jsonGetField(data, "correlation");
+
+ statdata->nullfrac = jsonGetFloat4(nullf, 0);
+ statdata->distinct = jsonGetFloat4(dist, 0);
+ statdata->width = (int32) jsonGetFloat4(width, 0);
+
+ statdata->nullfrac += (1.0 - statdata->nullfrac) * nullfrac;
+
+ /* Include MCV slot if exists */
+ if (DatumGetPointer(mcv))
+ {
+ slot->kind = STATISTIC_KIND_MCV;
+ slot->opid = eqop;
+ slot->numbers = jsonStatsConvertArray(jsonGetField(mcv, "numbers"),
+ JsonStatFloat4, FLOAT4OID,
+ 1.0 - nullfrac);
+ slot->values = jsonStatsConvertArray(jsonGetField(mcv, "values"),
+ stattype, type, 0);
+ slot++;
+ }
+
+ /* Include Histogram slot if exists */
+ if (DatumGetPointer(hst))
+ {
+ slot->kind = STATISTIC_KIND_HISTOGRAM;
+ slot->opid = ltop;
+ slot->numbers = jsonStatsConvertArray(jsonGetField(hst, "numbers"),
+ JsonStatFloat4, FLOAT4OID, 1.0);
+ slot->values = jsonStatsConvertArray(jsonGetField(hst, "values"),
+ stattype, type, 0);
+ slot++;
+ }
+
+ /* Include Correlation slot if exists */
+ if (DatumGetPointer(corr))
+ {
+ Datum correlation = Float4GetDatum(jsonGetFloat4(corr, 0));
+
+ slot->kind = STATISTIC_KIND_CORRELATION;
+ slot->opid = ltop;
+ slot->numbers = PointerGetDatum(construct_array(&correlation, 1,
+ FLOAT4OID, 4, true,
+ 'i'));
+ slot++;
+ }
+
+ /* Include JSON statistics for a given path and possibly for its subpaths */
+ if ((stattype == JsonStatJsonb ||
+ stattype == JsonStatJsonbWithoutSubpaths) &&
+ jsonAnalyzeBuildSubPathsData(pstats->data->pathdatums,
+ pstats->data->npaths,
+ pstats->datum - pstats->data->pathdatums,
+ pstats->path,
+ pstats->pathlen,
+ stattype == JsonStatJsonb,
+ nullfrac,
+ &slot->values,
+ &slot->numbers))
+ {
+ slot->kind = STATISTIC_KIND_JSON;
+ slot++;
+ }
+
+ return true;
+}
+
+static float4
+jsonPathStatsGetFloat(JsonPathStats pstats, const char *key, float4 defaultval)
+{
+ if (!pstats)
+ return defaultval;
+
+ return jsonGetFloat4(jsonGetField(*pstats->datum, key), defaultval);
+}
+
+float4
+jsonPathStatsGetFreq(JsonPathStats pstats, float4 defaultfreq)
+{
+ return jsonPathStatsGetFloat(pstats, "freq", defaultfreq);
+}
+
+float4
+jsonPathStatsGetAvgArraySize(JsonPathStats pstats)
+{
+ return jsonPathStatsGetFloat(pstats, "avg_array_length", 1.0);
+}
+
+/*
+ * jsonPathStatsGetTypeFreq
+ * Get frequency of different JSON object types for a given path.
+ *
+ * JSON documents don't have any particular schema, and the same path may point
+ * to values with different types in multiple documents. Consider for example
+ * two documents {"a" : "b"} and {"a" : 100} which have both a string and int
+ * for the same path. So we track the frequency of different JSON types for
+ * each path, so that we can consider this later.
+ */
+float4
+jsonPathStatsGetTypeFreq(JsonPathStats pstats, JsonbValueType type,
+ float4 defaultfreq)
+{
+ const char *key;
+
+ if (!pstats)
+ return defaultfreq;
+
+ /*
+ * When dealing with (object/array) length stats, we only really care about
+ * objects and arrays.
+ *
+ * Lengths are always numeric, so simply return 0 if requested frequency
+ * of non-numeric values.
+ */
+ if (pstats->type == JsonPathStatsArrayLength)
+ {
+ if (type != jbvNumeric)
+ return 0.0;
+
+ return jsonPathStatsGetFloat(pstats, "freq_array", defaultfreq);
+ }
+
+ if (pstats->type == JsonPathStatsObjectLength)
+ {
+ if (type != jbvNumeric)
+ return 0.0;
+
+ return jsonPathStatsGetFloat(pstats, "freq_object", defaultfreq);
+ }
+
+ /* Which JSON type are we interested in? Pick the right freq_type key. */
+ switch (type)
+ {
+ case jbvNull:
+ key = "freq_null";
+ break;
+ case jbvString:
+ key = "freq_string";
+ break;
+ case jbvNumeric:
+ key = "freq_numeric";
+ break;
+ case jbvBool:
+ key = "freq_boolean";
+ break;
+ case jbvObject:
+ key = "freq_object";
+ break;
+ case jbvArray:
+ key = "freq_array";
+ break;
+ default:
+ elog(ERROR, "Invalid jsonb value type: %d", type);
+ break;
+ }
+
+ return jsonPathStatsGetFloat(pstats, key, defaultfreq);
+}
+
+/*
+ * jsonPathStatsFormTuple
+ * For a pg_statistic tuple representing JSON statistics.
+ *
+ * XXX Maybe it's a bit expensive to first build StatsData and then transform it
+ * again while building the tuple. Could it be done in a single step? Would it be
+ * more efficient? Not sure how expensive it actually is, though.
+ */
+static HeapTuple
+jsonPathStatsFormTuple(JsonPathStats pstats, JsonStatType type, float4 nullfrac)
+{
+ StatsData statdata;
+
+ if (!pstats || !pstats->datum)
+ return NULL;
+
+ /*
+ * If it is the ordinary root path stats, there is no need to transform
+ * the tuple, it can be simply copied.
+ */
+ if (pstats->datum == &pstats->data->pathdatums[0] &&
+ pstats->type == JsonPathStatsValues)
+ return heap_copytuple(pstats->data->statsTuple);
+
+ MemSet(&statdata, 0, sizeof(statdata));
+
+ if (!jsonPathStatsExtractData(pstats, type, nullfrac, &statdata))
+ return NULL;
+
+ return stats_form_tuple(&statdata);
+}
+
+/*
+ * jsonStatsGetPathTuple
+ * Extract JSON statistics for a text[] path and form pg_statistics tuple.
+ */
+static HeapTuple
+jsonStatsGetPathTuple(JsonStats jsdata, JsonStatType type,
+ Datum *path, int pathlen, bool try_arrays_indexes)
+{
+ float4 nullfrac;
+ JsonPathStats pstats = jsonStatsGetPath(jsdata, path, pathlen,
+ try_arrays_indexes, &nullfrac);
+
+ return jsonPathStatsFormTuple(pstats, type, nullfrac);
+}
+
+/*
+ * jsonStatsGetArrayIndexStatsTuple
+ * Extract JSON statistics for a array index and form pg_statistics tuple.
+ */
+static HeapTuple
+jsonStatsGetArrayIndexStatsTuple(JsonStats jsdata, JsonStatType type, int32 index)
+{
+ /* Extract statistics for root array elements */
+ JsonPathStats arrstats = jsonStatsGetRootArrayPath(jsdata);
+ JsonPathStats rootstats;
+ Selectivity index_sel;
+
+ if (!arrstats)
+ return NULL;
+
+ /* Compute relative selectivity of 'EXISTS($[index])' */
+ rootstats = jsonStatsGetRootPath(jsdata);
+ index_sel = jsonPathStatsGetArrayIndexSelectivity(rootstats, index);
+ index_sel /= jsonPathStatsGetFreq(arrstats, 0.0);
+
+ /* Form pg_statistics tuple, taking into account array index selectivity */
+ return jsonPathStatsFormTuple(arrstats, type, 1.0 - index_sel);
+}
+
+/*
+ * jsonStatsGetPathFreq
+ * Return frequency of a path (fraction of documents containing it).
+ */
+static float4
+jsonStatsGetPathFreq(JsonStats jsdata, Datum *path, int pathlen,
+ bool try_array_indexes)
+{
+ float4 nullfrac;
+ JsonPathStats pstats = jsonStatsGetPath(jsdata, path, pathlen,
+ try_array_indexes, &nullfrac);
+ float4 freq = (1.0 - nullfrac) * jsonPathStatsGetFreq(pstats, 0.0);
+
+ CLAMP_PROBABILITY(freq);
+ return freq;
+}
+
+/*
+ * jsonbStatsVarOpConst
+ * Prepare optimizer statistics for a given operator, from JSON stats.
+ *
+ * This handles only OpExpr expressions, with variable and a constant. We get
+ * the constant as is, and the variable is represented by statistics fetched
+ * by get_restriction_variable().
+ *
+ * opid - OID of the operator (input parameter)
+ * resdata - pointer to calculated statistics for result of operator
+ * vardata - statistics for the restriction variable
+ * cnst - constant from the operator expression
+ *
+ * Returns true when useful optimizer statistics have been calculated.
+ */
+static bool
+jsonbStatsVarOpConst(Oid opid, VariableStatData *resdata,
+ const VariableStatData *vardata, Const *cnst)
+{
+ JsonStatData jsdata;
+ JsonStatType statype = JsonStatJsonb;
+
+ if (!jsonStatsInit(&jsdata, vardata))
+ return false;
+
+ switch (opid)
+ {
+ case JsonbObjectFieldTextOperator:
+ statype = JsonStatText;
+ /* FALLTHROUGH */
+ case JsonbObjectFieldOperator:
+ {
+ if (cnst->consttype != TEXTOID)
+ {
+ jsonStatsRelease(&jsdata);
+ return false;
+ }
+
+ resdata->statsTuple = jsonStatsGetPathTuple(&jsdata, statype,
+ &cnst->constvalue, 1,
+ false);
+ break;
+ }
+
+ case JsonbArrayElementTextOperator:
+ statype = JsonStatText;
+ /* FALLTHROUGH */
+ case JsonbArrayElementOperator:
+ {
+ if (cnst->consttype != INT4OID)
+ {
+ jsonStatsRelease(&jsdata);
+ return false;
+ }
+
+ resdata->statsTuple =
+ jsonStatsGetArrayIndexStatsTuple(&jsdata, statype,
+ DatumGetInt32(cnst->constvalue));
+ break;
+ }
+
+ case JsonbExtractPathTextOperator:
+ statype = JsonStatText;
+ /* FALLTHROUGH */
+ case JsonbExtractPathOperator:
+ {
+ Datum *path;
+ bool *nulls;
+ int pathlen;
+ bool have_nulls = false;
+
+ if (cnst->consttype != TEXTARRAYOID)
+ {
+ jsonStatsRelease(&jsdata);
+ return false;
+ }
+
+ deconstruct_array(DatumGetArrayTypeP(cnst->constvalue), TEXTOID,
+ -1, false, 'i', &path, &nulls, &pathlen);
+
+ for (int i = 0; i < pathlen; i++)
+ {
+ if (nulls[i])
+ {
+ have_nulls = true;
+ break;
+ }
+ }
+
+ if (!have_nulls)
+ resdata->statsTuple = jsonStatsGetPathTuple(&jsdata, statype,
+ path, pathlen,
+ true);
+
+ pfree(path);
+ pfree(nulls);
+ break;
+ }
+
+ default:
+ jsonStatsRelease(&jsdata);
+ return false;
+ }
+
+ if (!resdata->statsTuple)
+ resdata->statsTuple = stats_form_tuple(NULL); /* form all-NULL tuple */
+
+ resdata->acl_ok = vardata->acl_ok;
+ resdata->freefunc = heap_freetuple;
+ Assert(resdata->rel == vardata->rel);
+ Assert(resdata->atttype ==
+ (statype == JsonStatJsonb ? JSONBOID :
+ statype == JsonStatText ? TEXTOID :
+ /* statype == JsonStatFreq */ BOOLOID));
+
+ jsonStatsRelease(&jsdata);
+ return true;
+}
+
+/*
+ * jsonb_stats
+ * Statistics estimation procedure for JSONB data type.
+ *
+ * This only supports OpExpr expressions, with (Var op Const) shape.
+ *
+ * Var really can be a chain of OpExprs with derived statistics
+ * (jsonb_column -> 'key1' -> key2'), because get_restriction_variable()
+ * already handles this case.
+ */
+Datum
+jsonb_stats(PG_FUNCTION_ARGS)
+{
+ PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
+ OpExpr *opexpr = (OpExpr *) PG_GETARG_POINTER(1);
+ int varRelid = PG_GETARG_INT32(2);
+ VariableStatData *resdata = (VariableStatData *) PG_GETARG_POINTER(3);
+ VariableStatData vardata;
+ Node *constexpr;
+ bool varonleft;
+
+ /* should only be called for OpExpr expressions */
+ Assert(IsA(opexpr, OpExpr));
+
+ /* Is the expression simple enough? (Var op Const) or similar? */
+ if (!get_restriction_variable(root, opexpr->args, varRelid,
+ &vardata, &constexpr, &varonleft))
+ PG_RETURN_VOID();
+
+ /* XXX Could we also get varonleft=false in useful cases? */
+ if (IsA(constexpr, Const) && varonleft)
+ jsonbStatsVarOpConst(opexpr->opno, resdata, &vardata,
+ (Const *) constexpr);
+
+ ReleaseVariableStats(vardata);
+
+ PG_RETURN_VOID();
+}
+
+/*
+ * jsonSelectivity
+ * Use JSON statistics to estimate selectivity for (in)equalities.
+ *
+ * The statistics is represented as (arrays of) JSON values etc. so we
+ * need to pass the right operators to the functions.
+ */
+Selectivity
+jsonSelectivity(JsonPathStats stats, Datum scalar, Oid operator)
+{
+ VariableStatData vardata;
+ Selectivity sel;
+
+ if (!stats)
+ return 0.0;
+
+ vardata.atttype = JSONBOID;
+ vardata.atttypmod = -1;
+ vardata.isunique = false;
+ vardata.rel = stats->data->rel;
+ vardata.var = NULL;
+ vardata.vartype = JSONBOID;
+ vardata.acl_ok = stats->data->acl_ok;
+ vardata.statsTuple = jsonPathStatsFormTuple(stats,
+ JsonStatJsonbWithoutSubpaths, 0.0);
+
+ if (operator == JsonbEqOperator)
+ sel = var_eq_const(&vardata, operator, InvalidOid, scalar, false, true, false);
+ else
+ sel = scalarineqsel(NULL, operator,
+ /* is it greater or greater-or-equal? */
+ operator == JsonbGtOperator ||
+ operator == JsonbGeOperator,
+ /* is it equality? */
+ operator == JsonbLeOperator ||
+ operator == JsonbGeOperator,
+ InvalidOid,
+ &vardata, scalar, JSONBOID);
+
+ if (vardata.statsTuple)
+ heap_freetuple(vardata.statsTuple);
+
+ return sel;
+}
+
+/*
+ * jsonAccumulateSubPathSelectivity
+ * Transform absolute subpath selectivity into relative and accumulate it
+ * into parent path simply by multiplication of relative selectivities.
+ */
+static void
+jsonAccumulateSubPathSelectivity(Selectivity subpath_abs_sel,
+ Selectivity path_freq,
+ Selectivity *path_relative_sel,
+ JsonPathStats array_path_stats)
+{
+ Selectivity sel = subpath_abs_sel / path_freq; /* relative selectivity */
+
+ /* XXX Try to take into account array length */
+ if (array_path_stats)
+ sel = 1.0 - pow(1.0 - sel,
+ jsonPathStatsGetAvgArraySize(array_path_stats));
+
+ /* Accumulate selectivity of subpath into parent path */
+ *path_relative_sel *= sel;
+}
+
+/*
+ * jsonSelectivityContains
+ * Estimate selectivity for containment operator on JSON.
+ *
+ * Iterate through query jsonb elements, build paths to its leaf elements,
+ * calculate selectivies of 'path == scalar' in leaves, multiply relative
+ * selectivities of subpaths at each path level, propagate computed
+ * selectivities to the root.
+ */
+static Selectivity
+jsonSelectivityContains(JsonStats stats, Jsonb *jb)
+{
+ JsonbValue v;
+ JsonbIterator *it;
+ JsonbIteratorToken r;
+ StringInfoData pathstr; /* path string */
+ struct Path /* path stack entry */
+ {
+ struct Path *parent; /* parent entry */
+ int len; /* associated length of pathstr */
+ Selectivity freq; /* absolute frequence of path */
+ Selectivity sel; /* relative selectivity of subpaths */
+ JsonPathStats stats; /* statistics for the path */
+ bool is_array_accesor; /* is it '[*]' ? */
+ } root, /* root path entry */
+ *path = &root; /* path entry stack */
+ Selectivity sel; /* resulting selectivity */
+ Selectivity scalarSel; /* selectivity of 'jsonb == scalar' */
+
+ /* Initialize root path string */
+ initStringInfo(&pathstr);
+ appendBinaryStringInfo(&pathstr, stats->prefix, stats->prefixlen);
+
+ /* Initialize root path entry */
+ root.parent = NULL;
+ root.len = pathstr.len;
+ root.stats = jsonStatsFindPath(stats, pathstr.data, pathstr.len);
+ root.freq = jsonPathStatsGetFreq(root.stats, 0.0);
+ root.sel = 1.0;
+ root.is_array_accesor = pathstr.data[pathstr.len - 1] == ']';
+
+ /* Return 0, if NULL fraction is 1. */
+ if (root.freq <= 0.0)
+ return 0.0;
+
+ /*
+ * Selectivity of query 'jsonb @> scalar' consists of selectivities of
+ * 'jsonb == scalar' and 'jsonb[*] == scalar'. Selectivity of
+ * 'jsonb[*] == scalar' will be computed in root.sel, but for
+ * 'jsonb == scalar' we need additional computation.
+ */
+ if (JsonContainerIsScalar(&jb->root))
+ scalarSel = jsonSelectivity(root.stats, JsonbPGetDatum(jb),
+ JsonbEqOperator);
+ else
+ scalarSel = 0.0;
+
+ it = JsonbIteratorInit(&jb->root);
+
+ while ((r = JsonbIteratorNext(&it, &v, false)) != WJB_DONE)
+ {
+ switch (r)
+ {
+ case WJB_BEGIN_OBJECT:
+ {
+ struct Path *p;
+ Selectivity freq =
+ jsonPathStatsGetTypeFreq(path->stats, jbvObject, 0.0);
+
+ /* If there are no objects, selectivity is 0. */
+ if (freq <= 0.0)
+ return 0.0;
+
+ /*
+ * Push path entry for object keys, actual key names are
+ * appended later in WJB_KEY case.
+ */
+ p = palloc(sizeof(*p));
+ p->len = pathstr.len;
+ p->parent = path;
+ p->stats = NULL;
+ p->freq = freq;
+ p->sel = 1.0;
+ p->is_array_accesor = false;
+ path = p;
+ break;
+ }
+
+ case WJB_BEGIN_ARRAY:
+ {
+ struct Path *p;
+ JsonPathStats pstats;
+ Selectivity freq;
+
+ /*
+ * First, find stats for the parent path if needed, it will be
+ * used in jsonAccumulateSubPathSelectivity().
+ */
+ if (!path->stats)
+ path->stats = jsonStatsFindPath(stats, pathstr.data,
+ pathstr.len);
+
+ /* Appeend path string entry for array elements, get stats. */
+ jsonPathAppendEntry(&pathstr, NULL);
+ pstats = jsonStatsFindPath(stats, pathstr.data, pathstr.len);
+ freq = jsonPathStatsGetFreq(pstats, 0.0);
+
+ /* If there are no arrays, return 0 or scalar selectivity */
+ if (freq <= 0.0)
+ return scalarSel;
+
+ /* Push path entry for array elements. */
+ p = palloc(sizeof(*p));
+ p->len = pathstr.len;
+ p->parent = path;
+ p->stats = pstats;
+ p->freq = freq;
+ p->sel = 1.0;
+ p->is_array_accesor = true;
+ path = p;
+ break;
+ }
+
+ case WJB_END_OBJECT:
+ case WJB_END_ARRAY:
+ {
+ struct Path *p = path;
+ /* Absoulte selectivity of the path with its all subpaths */
+ Selectivity abs_sel = p->sel * p->freq;
+
+ /* Pop last path entry */
+ path = path->parent;
+ pfree(p);
+ pathstr.len = path->len;
+ pathstr.data[pathstr.len] = '\0';
+
+ /* Accumulate selectivity into parent path */
+ jsonAccumulateSubPathSelectivity(abs_sel, path->freq,
+ &path->sel,
+ path->is_array_accesor ?
+ path->parent->stats : NULL);
+ break;
+ }
+
+ case WJB_KEY:
+ {
+ /* Remove previous key in the path string */
+ pathstr.len = path->parent->len;
+ pathstr.data[pathstr.len] = '\0';
+
+ /* Append current key to path string */
+ jsonPathAppendEntryWithLen(&pathstr, v.val.string.val,
+ v.val.string.len);
+ path->len = pathstr.len;
+ break;
+ }
+
+ case WJB_VALUE:
+ case WJB_ELEM:
+ {
+ /*
+ * Extract statistics for a path. Array elements share the
+ * same statistics that was extracted in WJB_BEGIN_ARRAY.
+ */
+ JsonPathStats pstats = r == WJB_ELEM ? path->stats :
+ jsonStatsFindPath(stats, pathstr.data, pathstr.len);
+ Selectivity abs_sel; /* Absolute selectivity of 'path == scalar' */
+
+ if (pstats)
+ {
+ /* Make scalar jsonb datum and compute selectivity */
+ Datum scalar = JsonbPGetDatum(JsonbValueToJsonb(&v));
+
+ abs_sel = jsonSelectivity(pstats, scalar, JsonbEqOperator);
+ }
+ else
+ abs_sel = 0.0;
+
+ /* Accumulate selectivity into parent path */
+ jsonAccumulateSubPathSelectivity(abs_sel, path->freq,
+ &path->sel,
+ path->is_array_accesor ?
+ path->parent->stats : NULL);
+ break;
+ }
+
+ default:
+ break;
+ }
+ }
+
+ /* Compute absolute selectivity for root, including raw scalar case. */
+ sel = root.sel * root.freq + scalarSel;
+ CLAMP_PROBABILITY(sel);
+ return sel;
+}
+
+/*
+ * jsonSelectivityExists
+ * Estimate selectivity for JSON "exists" operator.
+ */
+static Selectivity
+jsonSelectivityExists(JsonStats stats, Datum key)
+{
+ JsonPathStats rootstats;
+ JsonPathStats arrstats;
+ JsonbValue jbvkey;
+ Datum jbkey;
+ Selectivity keysel;
+ Selectivity scalarsel;
+ Selectivity arraysel;
+ Selectivity sel;
+
+ JsonValueInitStringWithLen(&jbvkey,
+ VARDATA_ANY(key), VARSIZE_ANY_EXHDR(key));
+
+ jbkey = JsonbPGetDatum(JsonbValueToJsonb(&jbvkey));
+
+ keysel = jsonStatsGetPathFreq(stats, &key, 1, false);
+
+ rootstats = jsonStatsGetRootPath(stats);
+ scalarsel = jsonSelectivity(rootstats, jbkey, JsonbEqOperator);
+
+ arrstats = jsonStatsGetRootArrayPath(stats);
+ arraysel = jsonSelectivity(arrstats, jbkey, JsonbEqOperator);
+ arraysel = 1.0 - pow(1.0 - arraysel,
+ jsonPathStatsGetAvgArraySize(rootstats));
+
+ sel = keysel + scalarsel + arraysel;
+ CLAMP_PROBABILITY(sel);
+ return sel;
+}
+
+static Selectivity
+jsonb_sel_internal(JsonStats stats, Oid operator, Const *cnst, bool varonleft)
+{
+ switch (operator)
+ {
+ case JsonbExistsOperator:
+ if (!varonleft || cnst->consttype != TEXTOID)
+ break;
+
+ return jsonSelectivityExists(stats, cnst->constvalue);
+
+ case JsonbExistsAnyOperator:
+ case JsonbExistsAllOperator:
+ {
+ Datum *keys;
+ bool *nulls;
+ Selectivity freq = 1.0;
+ int nkeys;
+ int i;
+ bool all = operator == JsonbExistsAllOperator;
+
+ if (!varonleft || cnst->consttype != TEXTARRAYOID)
+ break;
+
+ deconstruct_array(DatumGetArrayTypeP(cnst->constvalue), TEXTOID,
+ -1, false, 'i', &keys, &nulls, &nkeys);
+
+ for (i = 0; i < nkeys; i++)
+ if (!nulls[i])
+ {
+ Selectivity pathfreq = jsonSelectivityExists(stats,
+ keys[i]);
+ freq *= all ? pathfreq : (1.0 - pathfreq);
+ }
+
+ pfree(keys);
+ pfree(nulls);
+
+ if (!all)
+ freq = 1.0 - freq;
+
+ return freq;
+ }
+
+ case JsonbContainedOperator:
+ if (varonleft || cnst->consttype != JSONBOID)
+ break;
+
+ return jsonSelectivityContains(stats,
+ DatumGetJsonbP(cnst->constvalue));
+
+ case JsonbContainsOperator:
+ if (!varonleft || cnst->consttype != JSONBOID)
+ break;
+
+ return jsonSelectivityContains(stats,
+ DatumGetJsonbP(cnst->constvalue));
+
+ default:
+ break;
+ }
+
+ return DEFAULT_JSON_CONTAINS_SEL;
+}
+
+/*
+ * jsonb_sel
+ * The main procedure estimating selectivity for all JSONB operators.
+ */
+Datum
+jsonb_sel(PG_FUNCTION_ARGS)
+{
+ PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
+ Oid operator = PG_GETARG_OID(1);
+ List *args = (List *) PG_GETARG_POINTER(2);
+ int varRelid = PG_GETARG_INT32(3);
+ double sel = DEFAULT_JSON_CONTAINS_SEL;
+ Node *other;
+ bool varonleft;
+ VariableStatData vardata;
+
+ if (get_restriction_variable(root, args, varRelid,
+ &vardata, &other, &varonleft))
+ {
+ if (IsA(other, Const))
+ {
+ Const *cnst = (Const *) other;
+
+ if (cnst->constisnull)
+ sel = 0.0;
+ else
+ {
+ JsonStatData stats;
+
+ if (jsonStatsInit(&stats, &vardata))
+ {
+ sel = jsonb_sel_internal(&stats, operator, cnst, varonleft);
+ jsonStatsRelease(&stats);
+ }
+ }
+ }
+
+ ReleaseVariableStats(vardata);
+ }
+
+ PG_RETURN_FLOAT8((float8) sel);
+}
diff --git a/src/backend/utils/adt/jsonb_typanalyze.c b/src/backend/utils/adt/jsonb_typanalyze.c
new file mode 100644
index 00000000000..65551f6ff93
--- /dev/null
+++ b/src/backend/utils/adt/jsonb_typanalyze.c
@@ -0,0 +1,1360 @@
+/*-------------------------------------------------------------------------
+ *
+ * jsonb_typanalyze.c
+ * Functions for gathering statistics from jsonb columns
+ *
+ * Copyright (c) 2016-2022, PostgreSQL Global Development Group
+ *
+ * Functions in this module are used to analyze contents of JSONB columns
+ * and build optimizer statistics. In principle we extract paths from all
+ * sampled documents and calculate the usual statistics (MCV, histogram)
+ * for each path - in principle each path is treated as a column.
+ *
+ * Because we're not enforcing any JSON schema, the documents may differ
+ * a lot - the documents may contain large number of different keys, the
+ * types of values may be entirely different, etc. This makes it more
+ * challenging than building stats for regular columns. For example not
+ * only do we need to decide which values to keep in the MCV, but also
+ * which paths to keep (in case the documents are so variable we can't
+ * keep all paths).
+ *
+ * The statistics is stored in pg_statistic, in a slot with a new stakind
+ * value (STATISTIC_KIND_JSON). The statistics is serialized as an array
+ * of JSONB values, eash element storing statistics for one path.
+ *
+ * For each path, we store the following keys:
+ *
+ * - path - path this stats is for, serialized as jsonpath
+ * - freq - frequency of documents containing this path
+ * - json - the regular per-column stats (MCV, histogram, ...)
+ * - freq_null - frequency of JSON null values
+ * - freq_array - frequency of JSON array values
+ * - freq_object - frequency of JSON object values
+ * - freq_string - frequency of JSON string values
+ * - freq_numeric - frequency of JSON numeric values
+ *
+ * This is stored in the stavalues array.
+ *
+ * The first element of stavalues is a path prefix. It is used for avoiding
+ * path transformations when the derived statistics for the chains of ->
+ * operators is computed.
+ *
+ * The per-column stats (stored in the "json" key) have additional internal
+ * structure, to allow storing multiple stakind types (histogram, mcv). See
+ * jsonAnalyzeMakeScalarStats for details.
+ *
+ *
+ * XXX It's a bit weird the "regular" stats are stored in the "json" key,
+ * while the JSON stats (frequencies of different JSON types) are right
+ * at the top level.
+ *
+ *
+ * IDENTIFICATION
+ * src/backend/utils/adt/jsonb_typanalyze.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#include "postgres.h"
+
+#include "fmgr.h"
+#include "access/hash.h"
+#include "access/detoast.h"
+#include "catalog/pg_attribute.h"
+#include "catalog/pg_operator.h"
+#include "catalog/pg_type.h"
+#include "commands/vacuum.h"
+#include "utils/builtins.h"
+#include "utils/guc.h"
+#include "utils/hsearch.h"
+#include "utils/json.h"
+#include "utils/jsonb.h"
+#include "utils/json_selfuncs.h"
+#include "utils/lsyscache.h"
+#include "utils/memutils.h"
+
+typedef struct JsonPathEntry JsonPathEntry;
+
+/*
+ * Element of a path in the JSON document (i.e. not jsonpath). Elements
+ * are linked together to build longer paths.
+ *
+ * 'entry' can be not zero-terminated when it is pointing to JSONB keys, so
+ * 'len' is necessary. 'len' is also used for faster entry comparison, to
+ * distinguish array entries ('len' == -1).
+ */
+typedef struct JsonPathEntry
+{
+ JsonPathEntry *parent;
+ const char *entry; /* element of the path as a string */
+ int len; /* length of entry string (may be 0 or -1) */
+ uint32 hash; /* hash of the whole path (with parent) */
+} JsonPathEntry;
+
+#define JsonPathEntryIsArray(entry) ((entry)->len == -1)
+
+/*
+ * An array containing a dynamic number of values extracted from JSON documents.
+ * All values should have the same data type:
+ * jsonb - ordinary path stats, values of different JSON types
+ * int32 - array/object length stats
+ * text - separate stats fro strings
+ * numeric - separate stats fro numbers
+ */
+typedef struct JsonValues
+{
+ Datum *buf;
+ int count;
+ int allocated;
+} JsonValues;
+
+/*
+ * Scalar statistics built for an array of values, extracted from a JSON
+ * document (for one particular path).
+ */
+typedef struct JsonScalarStats
+{
+ JsonValues values;
+ VacAttrStats stats;
+} JsonScalarStats;
+
+/*
+ * Statistics calculated for a set of values.
+ *
+ *
+ * XXX This seems rather complicated and needs simplification. We're not
+ * really using all the various JsonScalarStats bits, there's a lot of
+ * duplication (e.g. each JsonScalarStats contains it's own array, which
+ * has a copy of data from the one in "jsons").
+ */
+typedef struct JsonValueStats
+{
+ JsonScalarStats jsons; /* stats for all JSON types together */
+
+#ifdef JSON_ANALYZE_SCALARS /* XXX */
+ JsonScalarStats strings; /* stats for JSON strings */
+ JsonScalarStats numerics; /* stats for JSON numerics */
+#endif
+
+ JsonScalarStats arrlens; /* stats of array lengths */
+ JsonScalarStats objlens; /* stats of object lengths */
+
+ int nnulls; /* number of JSON null values */
+ int ntrue; /* number of JSON true values */
+ int nfalse; /* number of JSON false values */
+ int nobjects; /* number of JSON objects */
+ int narrays; /* number of JSON arrays */
+ int nstrings; /* number of JSON strings */
+ int nnumerics; /* number of JSON numerics */
+
+ int64 narrelems; /* total number of array elements
+ * (for avg. array length) */
+} JsonValueStats;
+
+/* Main structure for analyzed JSON path */
+typedef struct JsonPathAnlStats
+{
+ JsonPathEntry path; /* path entry chain, used for hashing */
+ JsonValueStats vstats; /* collected values and raw computed stats */
+ Jsonb *stats; /* stats converted into jsonb form */
+ char *pathstr; /* full path string */
+ double freq; /* frequence of the path */
+ int depth; /* nesting level, i.e. path length */
+} JsonPathAnlStats;
+
+/* various bits needed while analyzing JSON */
+typedef struct JsonAnalyzeContext
+{
+ VacAttrStats *stats;
+ MemoryContext mcxt;
+ AnalyzeAttrFetchFunc fetchfunc;
+ HTAB *pathshash;
+ JsonPathAnlStats *root;
+ double totalrows;
+ double total_width;
+ int samplerows;
+ int target;
+ int null_cnt;
+ int analyzed_cnt;
+ int maxdepth;
+ bool scalarsOnly;
+} JsonAnalyzeContext;
+
+/*
+ * JsonPathMatch
+ * Determine when two JSON paths (list of JsonPathEntry) match.
+ *
+ * Returned int instead of bool, because it is an implementation of
+ * HashCompareFunc.
+ */
+static int
+JsonPathEntryMatch(const void *key1, const void *key2, Size keysize)
+{
+ const JsonPathEntry *path1 = key1;
+ const JsonPathEntry *path2 = key2;
+
+ return path1->parent != path2->parent ||
+ path1->len != path2->len ||
+ (path1->len > 0 &&
+ strncmp(path1->entry, path2->entry, path1->len));
+}
+
+/*
+ * JsonPathHash
+ * Calculate hash of the path entry.
+ *
+ * Parent hash should be already calculated.
+ */
+static uint32
+JsonPathEntryHash(const void *key, Size keysize)
+{
+ const JsonPathEntry *path = key;
+ uint32 hash = path->parent ? path->parent->hash : 0;
+
+ hash = (hash << 1) | (hash >> 31);
+ hash ^= path->len < 0 ? 0 :
+ DatumGetUInt32(hash_any((const unsigned char *) path->entry, path->len));
+
+ return hash;
+}
+
+/*
+ * jsonAnalyzeAddPath
+ * Add an entry for a JSON path to the working list of statistics.
+ *
+ * Returns a pointer to JsonPathAnlStats (which might have already existed
+ * if the path was in earlier document), which can then be populated or
+ * updated.
+ */
+static inline JsonPathAnlStats *
+jsonAnalyzeAddPath(JsonAnalyzeContext *ctx, JsonPathEntry *parent,
+ const char *entry, int len)
+{
+ JsonPathEntry path;
+ JsonPathAnlStats *stats;
+ bool found;
+
+ /* Init path entry */
+ path.parent = parent;
+ path.entry = entry;
+ path.len = len;
+ path.hash = JsonPathEntryHash(&path, 0);
+
+ /* See if we already saw this path earlier. */
+ stats = hash_search_with_hash_value(ctx->pathshash, &path, path.hash,
+ HASH_ENTER, &found);
+
+ /*
+ * Nope, it's the first time we see this path, so initialize all the
+ * fields (path string, counters, ...).
+ */
+ if (!found)
+ {
+ JsonPathAnlStats *parent = (JsonPathAnlStats *) stats->path.parent;
+ JsonPathEntry *path = &stats->path;
+ const char *ppath = parent->pathstr;
+ StringInfoData si;
+ MemoryContext oldcxt = MemoryContextSwitchTo(ctx->mcxt);
+
+ /* NULL entries are treated as wildcard array accessors "[*]" */
+ if (path->entry)
+ /* Copy path entry name into the right MemoryContext */
+ path->entry = pnstrdup(path->entry, path->len);
+
+ /* Initialze full path string */
+ initStringInfo(&si);
+ appendStringInfoString(&si, ppath);
+ jsonPathAppendEntry(&si, path->entry);
+ stats->pathstr = si.data;
+
+ MemoryContextSwitchTo(oldcxt);
+
+ /* initialize the stats counter for this path entry */
+ memset(&stats->vstats, 0, sizeof(JsonValueStats));
+ stats->stats = NULL;
+ stats->freq = 0.0;
+ stats->depth = parent->depth + 1;
+
+ /* update maximal depth */
+ if (ctx->maxdepth < stats->depth)
+ ctx->maxdepth = stats->depth;
+ }
+
+ return stats;
+}
+
+/*
+ * JsonValuesAppend
+ * Add a JSON value to the dynamic array (enlarge it if needed).
+ *
+ * XXX This is likely one of the problems - the documents may be pretty
+ * large, with a lot of different values for each path. At that point
+ * it's problematic to keep all of that in memory at once. So maybe we
+ * need to introduce some sort of compaction (e.g. we could try
+ * deduplicating the values), limit on size of the array or something.
+ */
+static inline void
+JsonValuesAppend(JsonValues *values, Datum value, int initialSize)
+{
+ if (values->count >= values->allocated)
+ {
+ if (values->allocated)
+ {
+ values->allocated = values->allocated * 2;
+ values->buf = repalloc(values->buf,
+ sizeof(values->buf[0]) * values->allocated);
+ }
+ else
+ {
+ values->allocated = initialSize;
+ values->buf = palloc(sizeof(values->buf[0]) * values->allocated);
+ }
+ }
+
+ values->buf[values->count++] = value;
+}
+
+/*
+ * jsonAnalyzeJsonValue
+ * Process a value extracted from the document (for a given path).
+ */
+static inline void
+jsonAnalyzeJsonValue(JsonAnalyzeContext *ctx, JsonValueStats *vstats,
+ JsonbValue *jv)
+{
+ JsonbValue *jbv;
+ JsonbValue jbvtmp;
+ Datum value;
+
+ /* XXX if analyzing only scalar values, make containers empty */
+ if (ctx->scalarsOnly && jv->type == jbvBinary)
+ {
+ if (JsonContainerIsObject(jv->val.binary.data))
+ jbv = JsonValueInitObject(&jbvtmp, 0, 0);
+ else
+ {
+ Assert(JsonContainerIsArray(jv->val.binary.data));
+ jbv = JsonValueInitArray(&jbvtmp, 0, 0, false);
+ }
+ }
+ else
+ jbv = jv;
+
+ /* always add it to the "global" JSON stats, shared by all types */
+ JsonValuesAppend(&vstats->jsons.values,
+ JsonbPGetDatum(JsonbValueToJsonb(jbv)),
+ ctx->target);
+
+ /* also update the type-specific counters */
+ switch (jv->type)
+ {
+ case jbvNull:
+ vstats->nnulls++;
+ break;
+
+ case jbvBool:
+ if (jv->val.boolean)
+ vstats->ntrue++;
+ else
+ vstats->nfalse++;
+ break;
+
+ case jbvString:
+ vstats->nstrings++;
+#ifdef JSON_ANALYZE_SCALARS
+ value = PointerGetDatum(
+ cstring_to_text_with_len(jv->val.string.val,
+ jv->val.string.len));
+ JsonValuesAppend(&vstats->strings.values, value, ctx->target);
+#endif
+ break;
+
+ case jbvNumeric:
+ vstats->nnumerics++;
+#ifdef JSON_ANALYZE_SCALARS
+ value = PointerGetDatum(jv->val.numeric);
+ JsonValuesAppend(&vstats->numerics.values, value, ctx->target);
+#endif
+ break;
+
+ case jbvBinary:
+ if (JsonContainerIsObject(jv->val.binary.data))
+ {
+ uint32 size = JsonContainerSize(jv->val.binary.data);
+
+ value = DatumGetInt32(size);
+ vstats->nobjects++;
+ JsonValuesAppend(&vstats->objlens.values, value, ctx->target);
+ }
+ else if (JsonContainerIsArray(jv->val.binary.data))
+ {
+ uint32 size = JsonContainerSize(jv->val.binary.data);
+
+ value = DatumGetInt32(size);
+ vstats->narrays++;
+ JsonValuesAppend(&vstats->arrlens.values, value, ctx->target);
+ vstats->narrelems += size;
+ }
+ break;
+
+ default:
+ elog(ERROR, "invalid scalar json value type %d", jv->type);
+ break;
+ }
+}
+
+/*
+ * jsonAnalyzeCollectPaths
+ * Parse the JSON document and collect all paths and their values.
+ */
+static void
+jsonAnalyzeCollectPaths(JsonAnalyzeContext *ctx, Jsonb *jb, void *param)
+{
+ JsonbValue jv;
+ JsonbIterator *it;
+ JsonbIteratorToken tok;
+ JsonPathAnlStats *stats = ctx->root;
+ bool collect_values = (bool)(intptr_t) param;
+ bool scalar = false;
+
+ if (collect_values && !JB_ROOT_IS_SCALAR(jb))
+ jsonAnalyzeJsonValue(ctx, &stats->vstats, JsonValueInitBinary(&jv, jb));
+
+ it = JsonbIteratorInit(&jb->root);
+
+ while ((tok = JsonbIteratorNext(&it, &jv, true)) != WJB_DONE)
+ {
+ switch (tok)
+ {
+ case WJB_BEGIN_OBJECT:
+ /*
+ * Read next token to see if the object is empty or not.
+ * If not, make stats for the first key. Subsequent WJB_KEYs
+ * and WJB_END_OBJECT will expect that stats will be pointing
+ * to the key of current object.
+ */
+ tok = JsonbIteratorNext(&it, &jv, true);
+
+ if (tok == WJB_END_OBJECT)
+ /* Empty object, simply skip stats initialization. */
+ break;
+
+ if (tok != WJB_KEY)
+ elog(ERROR, "unexpected jsonb iterator token: %d", tok);
+
+ stats = jsonAnalyzeAddPath(ctx, &stats->path,
+ jv.val.string.val,
+ jv.val.string.len);
+ break;
+
+ case WJB_BEGIN_ARRAY:
+ /* Make stats for non-scalar array and use it for all elements */
+ if (!(scalar = jv.val.array.rawScalar))
+ stats = jsonAnalyzeAddPath(ctx, &stats->path, NULL, -1);
+ break;
+
+ case WJB_END_ARRAY:
+ if (scalar)
+ break;
+ /* FALLTHROUGH */
+ case WJB_END_OBJECT:
+ /* Reset to parent stats */
+ stats = (JsonPathAnlStats *) stats->path.parent;
+ break;
+
+ case WJB_KEY:
+ /*
+ * Stats should point to the previous key of current object,
+ * use its parent path as a base path.
+ */
+ stats = jsonAnalyzeAddPath(ctx, stats->path.parent,
+ jv.val.string.val,
+ jv.val.string.len);
+ break;
+
+ case WJB_VALUE:
+ case WJB_ELEM:
+ if (collect_values)
+ jsonAnalyzeJsonValue(ctx, &stats->vstats, &jv);
+
+ /*
+ * Manually recurse into container by creating child iterator.
+ * We use skipNested=true to give jsonAnalyzeJsonValue()
+ * ability to access jbvBinary containers.
+ */
+ if (jv.type == jbvBinary)
+ {
+ JsonbIterator *it2 = JsonbIteratorInit(jv.val.binary.data);
+
+ it2->parent = it;
+ it = it2;
+ }
+ break;
+
+ default:
+ break;
+ }
+ }
+}
+
+/*
+ * jsonAnalyzeCollectSubpath
+ * Recursively extract trailing part of a path and collect its values.
+ */
+static void
+jsonAnalyzeCollectSubpath(JsonAnalyzeContext *ctx, JsonPathAnlStats *pstats,
+ JsonbValue *jbv, JsonPathEntry **entries,
+ int start_entry)
+{
+ JsonbValue scalar;
+ int i;
+
+ for (i = start_entry; i < pstats->depth; i++)
+ {
+ JsonPathEntry *entry = entries[i];
+ JsonbContainer *jbc = jbv->val.binary.data;
+ JsonbValueType type = jbv->type;
+
+ if (i > start_entry)
+ pfree(jbv);
+
+ if (type != jbvBinary)
+ return;
+
+ if (JsonPathEntryIsArray(entry))
+ {
+ JsonbIterator *it;
+ JsonbIteratorToken r;
+ JsonbValue elem;
+
+ if (!JsonContainerIsArray(jbc) || JsonContainerIsScalar(jbc))
+ return;
+
+ it = JsonbIteratorInit(jbc);
+
+ while ((r = JsonbIteratorNext(&it, &elem, true)) != WJB_DONE)
+ {
+ if (r == WJB_ELEM)
+ jsonAnalyzeCollectSubpath(ctx, pstats, &elem, entries, i + 1);
+ }
+
+ return;
+ }
+ else
+ {
+ if (!JsonContainerIsObject(jbc))
+ return;
+
+ jbv = findJsonbValueFromContainerLen(jbc, JB_FOBJECT,
+ entry->entry, entry->len);
+
+ if (!jbv)
+ return;
+ }
+ }
+
+ if (i == start_entry &&
+ jbv->type == jbvBinary &&
+ JsonbExtractScalar(jbv->val.binary.data, &scalar))
+ jbv = &scalar;
+
+ jsonAnalyzeJsonValue(ctx, &pstats->vstats, jbv);
+
+ if (i > start_entry)
+ pfree(jbv);
+}
+
+/*
+ * jsonAnalyzeCollectPath
+ * Extract a single path from JSON documents and collect its values.
+ */
+static void
+jsonAnalyzeCollectPath(JsonAnalyzeContext *ctx, Jsonb *jb, void *param)
+{
+ JsonPathAnlStats *pstats = (JsonPathAnlStats *) param;
+ JsonbValue jbvtmp;
+ JsonbValue *jbv = JsonValueInitBinary(&jbvtmp, jb);
+ JsonPathEntry *path;
+ JsonPathEntry **entries;
+ int i;
+
+ entries = palloc(sizeof(*entries) * pstats->depth);
+
+ /* Build entry array in direct order */
+ for (path = &pstats->path, i = pstats->depth - 1;
+ path->parent && i >= 0;
+ path = path->parent, i--)
+ entries[i] = path;
+
+ jsonAnalyzeCollectSubpath(ctx, pstats, jbv, entries, 0);
+
+ pfree(entries);
+}
+
+static Datum
+jsonAnalyzePathFetch(VacAttrStatsP stats, int rownum, bool *isnull)
+{
+ *isnull = false;
+ return stats->exprvals[rownum];
+}
+
+/*
+ * jsonAnalyzePathValues
+ * Calculate per-column statistics for values for a single path.
+ *
+ * We have already accumulated all the values for the path, so we simply
+ * call the typanalyze function for the proper data type, and then
+ * compute_stats (which points to compute_scalar_stats or so).
+ */
+static void
+jsonAnalyzePathValues(JsonAnalyzeContext *ctx, JsonScalarStats *sstats,
+ Oid typid, double freq)
+{
+ JsonValues *values = &sstats->values;
+ VacAttrStats *stats = &sstats->stats;
+ FormData_pg_attribute attr;
+ FormData_pg_type type;
+ int i;
+
+ if (!sstats->values.count)
+ return;
+
+ get_typlenbyvalalign(typid, &type.typlen, &type.typbyval, &type.typalign);
+
+ attr.attstattarget = ctx->target;
+
+ stats->attr = &attr;
+ stats->attrtypid = typid;
+ stats->attrtypmod = -1;
+ stats->attrtype = &type;
+ stats->anl_context = ctx->stats->anl_context;
+
+ stats->exprvals = values->buf;
+
+ /*
+ * The fields describing the stats->stavalues[n] element types default to
+ * the type of the data being analyzed, but the type-specific typanalyze
+ * function can change them if it wants to store something else.
+ */
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ stats->statypid[i] = stats->attrtypid;
+ stats->statyplen[i] = stats->attrtype->typlen;
+ stats->statypbyval[i] = stats->attrtype->typbyval;
+ stats->statypalign[i] = stats->attrtype->typalign;
+ }
+
+ std_typanalyze(stats);
+
+ stats->compute_stats(stats, jsonAnalyzePathFetch,
+ values->count,
+ ctx->totalrows / ctx->samplerows * values->count);
+
+ /*
+ * We've only kept the non-null values, so compute_stats will always
+ * leave this as 1.0. But we have enough info to calculate the correct
+ * value.
+ */
+ stats->stanullfrac = (float4)(1.0 - freq);
+
+ /*
+ * Similarly, we need to correct the MCV frequencies, becuse those are
+ * also calculated only from the non-null values. All we need to do is
+ * simply multiply that with the non-NULL frequency.
+ */
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ if (stats->stakind[i] == STATISTIC_KIND_MCV)
+ {
+ int j;
+ for (j = 0; j < stats->numnumbers[i]; j++)
+ stats->stanumbers[i][j] *= freq;
+ }
+ }
+}
+
+/*
+ * jsonAnalyzeMakeScalarStats
+ * Serialize scalar stats into a JSON representation.
+ *
+ * We simply produce a JSON document with a list of predefined keys:
+ *
+ * - nullfrac
+ * - distinct
+ * - width
+ * - correlation
+ * - mcv or histogram
+ *
+ * For the mcv / histogram, we store a nested values / numbers.
+ */
+static JsonbValue *
+jsonAnalyzeMakeScalarStats(JsonbParseState **ps, const char *name,
+ const VacAttrStats *stats)
+{
+ JsonbValue val;
+ int i;
+ int j;
+
+ pushJsonbKey(ps, &val, name);
+
+ pushJsonbValue(ps, WJB_BEGIN_OBJECT, NULL);
+
+ pushJsonbKeyValueFloat(ps, &val, "nullfrac", stats->stanullfrac);
+ pushJsonbKeyValueFloat(ps, &val, "distinct", stats->stadistinct);
+ pushJsonbKeyValueInteger(ps, &val, "width", stats->stawidth);
+
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ if (!stats->stakind[i])
+ break;
+
+ switch (stats->stakind[i])
+ {
+ case STATISTIC_KIND_MCV:
+ pushJsonbKey(ps, &val, "mcv");
+ break;
+
+ case STATISTIC_KIND_HISTOGRAM:
+ pushJsonbKey(ps, &val, "histogram");
+ break;
+
+ case STATISTIC_KIND_CORRELATION:
+ pushJsonbKeyValueFloat(ps, &val, "correlation",
+ stats->stanumbers[i][0]);
+ continue;
+
+ default:
+ elog(ERROR, "unexpected stakind %d", stats->stakind[i]);
+ break;
+ }
+
+ pushJsonbValue(ps, WJB_BEGIN_OBJECT, NULL);
+
+ if (stats->numvalues[i] > 0)
+ {
+ pushJsonbKey(ps, &val, "values");
+ pushJsonbValue(ps, WJB_BEGIN_ARRAY, NULL);
+ for (j = 0; j < stats->numvalues[i]; j++)
+ {
+ Datum v = stats->stavalues[i][j];
+ if (stats->attrtypid == JSONBOID)
+ pushJsonbElemBinary(ps, &val, DatumGetJsonbP(v));
+ else if (stats->attrtypid == TEXTOID)
+ pushJsonbElemText(ps, &val, DatumGetTextP(v));
+ else if (stats->attrtypid == NUMERICOID)
+ pushJsonbElemNumeric(ps, &val, DatumGetNumeric(v));
+ else if (stats->attrtypid == INT4OID)
+ pushJsonbElemInteger(ps, &val, DatumGetInt32(v));
+ else
+ elog(ERROR, "unexpected stat value type %d",
+ stats->attrtypid);
+ }
+ pushJsonbValue(ps, WJB_END_ARRAY, NULL);
+ }
+
+ if (stats->numnumbers[i] > 0)
+ {
+ pushJsonbKey(ps, &val, "numbers");
+ pushJsonbValue(ps, WJB_BEGIN_ARRAY, NULL);
+ for (j = 0; j < stats->numnumbers[i]; j++)
+ pushJsonbElemFloat(ps, &val, stats->stanumbers[i][j]);
+ pushJsonbValue(ps, WJB_END_ARRAY, NULL);
+ }
+
+ pushJsonbValue(ps, WJB_END_OBJECT, NULL);
+ }
+
+ return pushJsonbValue(ps, WJB_END_OBJECT, NULL);
+}
+
+/*
+ * jsonAnalyzeBuildPathStats
+ * Serialize statistics for a particular json path.
+ *
+ * This includes both the per-column stats (stored in "json" key) and the
+ * JSON specific stats (like frequencies of different object types).
+ */
+static Jsonb *
+jsonAnalyzeBuildPathStats(JsonPathAnlStats *pstats)
+{
+ const JsonValueStats *vstats = &pstats->vstats;
+ float4 freq = pstats->freq;
+ bool fullstats = true; /* pstats->path.parent != NULL */
+ JsonbValue val;
+ JsonbValue *jbv;
+ JsonbParseState *ps = NULL;
+
+ pushJsonbValue(&ps, WJB_BEGIN_OBJECT, NULL);
+
+ pushJsonbKeyValueString(&ps, &val, "path", pstats->pathstr);
+
+ pushJsonbKeyValueFloat(&ps, &val, "freq", freq);
+
+ pushJsonbKeyValueFloat(&ps, &val, "freq_null",
+ freq * vstats->nnulls /
+ vstats->jsons.values.count);
+
+ pushJsonbKeyValueFloat(&ps, &val, "freq_boolean",
+ freq * (vstats->nfalse + vstats->ntrue) /
+ vstats->jsons.values.count);
+
+ pushJsonbKeyValueFloat(&ps, &val, "freq_string",
+ freq * vstats->nstrings /
+ vstats->jsons.values.count);
+
+ pushJsonbKeyValueFloat(&ps, &val, "freq_numeric",
+ freq * vstats->nnumerics /
+ vstats->jsons.values.count);
+
+ pushJsonbKeyValueFloat(&ps, &val, "freq_array",
+ freq * vstats->narrays /
+ vstats->jsons.values.count);
+
+ pushJsonbKeyValueFloat(&ps, &val, "freq_object",
+ freq * vstats->nobjects /
+ vstats->jsons.values.count);
+
+ /*
+ * We keep array length stats here for queries like jsonpath '$.size() > 5'.
+ * Object lengths stats can be useful for other query lanuages.
+ */
+ if (vstats->arrlens.values.count)
+ jsonAnalyzeMakeScalarStats(&ps, "array_length", &vstats->arrlens.stats);
+
+ if (vstats->objlens.values.count)
+ jsonAnalyzeMakeScalarStats(&ps, "object_length", &vstats->objlens.stats);
+
+ if (vstats->narrays)
+ pushJsonbKeyValueFloat(&ps, &val, "avg_array_length",
+ (float4) vstats->narrelems / vstats->narrays);
+
+ if (fullstats)
+ {
+#ifdef JSON_ANALYZE_SCALARS
+ jsonAnalyzeMakeScalarStats(&ps, "string", &vstats->strings.stats);
+ jsonAnalyzeMakeScalarStats(&ps, "numeric", &vstats->numerics.stats);
+#endif
+ jsonAnalyzeMakeScalarStats(&ps, "json", &vstats->jsons.stats);
+ }
+
+ jbv = pushJsonbValue(&ps, WJB_END_OBJECT, NULL);
+
+ return JsonbValueToJsonb(jbv);
+}
+
+/*
+ * jsonAnalyzeCalcPathFreq
+ * Calculate path frequency, i.e. how many documents contain this path.
+ */
+static void
+jsonAnalyzeCalcPathFreq(JsonAnalyzeContext *ctx, JsonPathAnlStats *pstats)
+{
+ JsonPathAnlStats *parent = (JsonPathAnlStats *) pstats->path.parent;
+
+ if (parent)
+ {
+ int count = JsonPathEntryIsArray(&pstats->path) ?
+ parent->vstats.narrays : pstats->vstats.jsons.values.count;
+
+ pstats->freq = parent->freq * count / parent->vstats.jsons.values.count;
+
+ CLAMP_PROBABILITY(pstats->freq);
+ }
+ else
+ pstats->freq = (double) ctx->analyzed_cnt / ctx->samplerows;
+}
+
+/*
+ * jsonAnalyzePath
+ * Build statistics for values accumulated for this path.
+ *
+ * We're done with accumulating values for this path, so calculate the
+ * statistics for the various arrays.
+ *
+ * XXX I wonder if we could introduce some simple heuristict on which
+ * paths to keep, similarly to what we do for MCV lists. For example a
+ * path that occurred just once is not very interesting, so we could
+ * decide to ignore it and not build the stats. Although that won't
+ * save much, because there'll be very few values accumulated.
+ */
+static void
+jsonAnalyzePath(JsonAnalyzeContext *ctx, JsonPathAnlStats *pstats)
+{
+ MemoryContext oldcxt;
+ JsonValueStats *vstats = &pstats->vstats;
+
+ jsonAnalyzeCalcPathFreq(ctx, pstats);
+
+ /* values combining all object types */
+ jsonAnalyzePathValues(ctx, &vstats->jsons, JSONBOID, pstats->freq);
+
+ /*
+ * Lengths and array lengths. We divide counts by the total number of json
+ * values to compute correct nullfrac (i.e. not all jsons have lengths).
+ */
+ jsonAnalyzePathValues(ctx, &vstats->arrlens, INT4OID,
+ pstats->freq * vstats->arrlens.values.count /
+ vstats->jsons.values.count);
+ jsonAnalyzePathValues(ctx, &vstats->objlens, INT4OID,
+ pstats->freq * vstats->objlens.values.count /
+ vstats->jsons.values.count);
+
+#ifdef JSON_ANALYZE_SCALARS
+ /* stats for values of string/numeric types only */
+ jsonAnalyzePathValues(ctx, &vstats->strings, TEXTOID, pstats->freq);
+ jsonAnalyzePathValues(ctx, &vstats->numerics, NUMERICOID, pstats->freq);
+#endif
+
+ oldcxt = MemoryContextSwitchTo(ctx->stats->anl_context);
+ pstats->stats = jsonAnalyzeBuildPathStats(pstats);
+ MemoryContextSwitchTo(oldcxt);
+}
+
+/*
+ * JsonPathStatsCompare
+ * Compare two path stats (by path string).
+ *
+ * We store the stats sorted by path string, and this is the comparator.
+ */
+static int
+JsonPathStatsCompare(const void *pv1, const void *pv2)
+{
+ return strcmp((*((const JsonPathAnlStats **) pv1))->pathstr,
+ (*((const JsonPathAnlStats **) pv2))->pathstr);
+}
+
+/*
+ * jsonAnalyzeSortPaths
+ * Reads all stats stored in the hash table and sorts them.
+ */
+static JsonPathAnlStats **
+jsonAnalyzeSortPaths(JsonAnalyzeContext *ctx, int *p_npaths)
+{
+ HASH_SEQ_STATUS hseq;
+ JsonPathAnlStats *path;
+ JsonPathAnlStats **paths;
+ int npaths;
+
+ npaths = hash_get_num_entries(ctx->pathshash) + 1;
+ paths = MemoryContextAlloc(ctx->mcxt, sizeof(*paths) * npaths);
+
+ paths[0] = ctx->root;
+
+ hash_seq_init(&hseq, ctx->pathshash);
+
+ for (int i = 1; (path = hash_seq_search(&hseq)) != NULL; i++)
+ paths[i] = path;
+
+ pg_qsort(paths, npaths, sizeof(*paths), JsonPathStatsCompare);
+
+ *p_npaths = npaths;
+ return paths;
+}
+
+/*
+ * jsonAnalyzeBuildPathStatsArray
+ * Build jsonb datum array for path stats, that will be used as stavalues.
+ *
+ * The first element is a path prefix.
+ */
+static Datum *
+jsonAnalyzeBuildPathStatsArray(JsonPathAnlStats **paths, int npaths, int *nvals,
+ const char *prefix, int prefixlen)
+{
+ Datum *values = palloc(sizeof(Datum) * (npaths + 1));
+ JsonbValue *jbvprefix = palloc(sizeof(JsonbValue));
+ int i;
+
+ JsonValueInitStringWithLen(jbvprefix,
+ memcpy(palloc(prefixlen), prefix, prefixlen),
+ prefixlen);
+
+ values[0] = JsonbPGetDatum(JsonbValueToJsonb(jbvprefix));
+
+ for (i = 0; i < npaths; i++)
+ values[i + 1] = JsonbPGetDatum(paths[i]->stats);
+
+ *nvals = npaths + 1;
+
+ return values;
+}
+
+/*
+ * jsonAnalyzeMakeStats
+ * Build stavalues jsonb array for the root path prefix.
+ */
+static Datum *
+jsonAnalyzeMakeStats(JsonAnalyzeContext *ctx, JsonPathAnlStats **paths,
+ int npaths, int *numvalues)
+{
+ Datum *values;
+ MemoryContext oldcxt = MemoryContextSwitchTo(ctx->stats->anl_context);
+
+ values = jsonAnalyzeBuildPathStatsArray(paths, npaths, numvalues,
+ JSON_PATH_ROOT, JSON_PATH_ROOT_LEN);
+
+ MemoryContextSwitchTo(oldcxt);
+
+ return values;
+}
+
+/*
+ * jsonAnalyzeBuildSubPathsData
+ * Build statvalues and stanumbers arrays for the subset of paths starting
+ * from a given prefix.
+ *
+ * pathsDatums[index] should point to the desired path.
+ */
+bool
+jsonAnalyzeBuildSubPathsData(Datum *pathsDatums, int npaths, int index,
+ const char *path, int pathlen,
+ bool includeSubpaths, float4 nullfrac,
+ Datum *pvals, Datum *pnums)
+{
+ JsonPathAnlStats **pvalues = palloc(sizeof(*pvalues) * npaths);
+ Datum *values;
+ Datum numbers[1];
+ JsonbValue pathkey;
+ int nsubpaths = 0;
+ int nvalues;
+ int i;
+
+ JsonValueInitStringWithLen(&pathkey, "path", 4);
+
+ for (i = index; i < npaths; i++)
+ {
+ /* Extract path name */
+ Jsonb *jb = DatumGetJsonbP(pathsDatums[i]);
+ JsonbValue *jbv = findJsonbValueFromContainer(&jb->root, JB_FOBJECT,
+ &pathkey);
+
+ /* Check if path name starts with a given prefix */
+ if (!jbv || jbv->type != jbvString ||
+ jbv->val.string.len < pathlen ||
+ memcmp(jbv->val.string.val, path, pathlen))
+ break;
+
+ pfree(jbv);
+
+ /* Collect matching path */
+ pvalues[nsubpaths] = palloc(sizeof(**pvalues));
+ pvalues[nsubpaths]->stats = jb;
+
+ nsubpaths++;
+
+ /*
+ * The path should go before its subpaths, so if subpaths are not
+ * needed the loop is broken after the first matching path.
+ */
+ if (!includeSubpaths)
+ break;
+ }
+
+ if (!nsubpaths)
+ {
+ pfree(pvalues);
+ return false;
+ }
+
+ /* Construct new array from the selected paths */
+ values = jsonAnalyzeBuildPathStatsArray(pvalues, nsubpaths, &nvalues,
+ path, pathlen);
+ *pvals = PointerGetDatum(construct_array(values, nvalues, JSONBOID, -1,
+ false, 'i'));
+
+ pfree(pvalues);
+ pfree(values);
+
+ numbers[0] = Float4GetDatum(nullfrac);
+ *pnums = PointerGetDatum(construct_array(numbers, 1, FLOAT4OID, 4,
+ true /*FLOAT4PASSBYVAL*/, 'i'));
+
+ return true;
+}
+
+/*
+ * jsonAnalyzeInit
+ * Initialize the analyze context so that we can start adding paths.
+ */
+static void
+jsonAnalyzeInit(JsonAnalyzeContext *ctx, VacAttrStats *stats,
+ AnalyzeAttrFetchFunc fetchfunc,
+ int samplerows, double totalrows)
+{
+ HASHCTL hash_ctl;
+
+ memset(ctx, 0, sizeof(*ctx));
+
+ ctx->stats = stats;
+ ctx->fetchfunc = fetchfunc;
+ ctx->mcxt = CurrentMemoryContext;
+ ctx->samplerows = samplerows;
+ ctx->totalrows = totalrows;
+ ctx->target = stats->attr->attstattarget;
+ ctx->scalarsOnly = false;
+
+ MemSet(&hash_ctl, 0, sizeof(hash_ctl));
+ hash_ctl.keysize = sizeof(JsonPathEntry);
+ hash_ctl.entrysize = sizeof(JsonPathAnlStats);
+ hash_ctl.hash = JsonPathEntryHash;
+ hash_ctl.match = JsonPathEntryMatch;
+ hash_ctl.hcxt = ctx->mcxt;
+
+ ctx->pathshash = hash_create("JSON analyze path table", 100, &hash_ctl,
+ HASH_ELEM | HASH_FUNCTION | HASH_COMPARE | HASH_CONTEXT);
+
+ ctx->root = MemoryContextAllocZero(ctx->mcxt, sizeof(JsonPathAnlStats));
+ ctx->root->pathstr = JSON_PATH_ROOT;
+}
+
+/*
+ * jsonAnalyzePass
+ * One analysis pass over the JSON column.
+ *
+ * Performs one analysis pass on the JSON documents, and passes them to the
+ * custom analyzefunc.
+ */
+static void
+jsonAnalyzePass(JsonAnalyzeContext *ctx,
+ void (*analyzefunc)(JsonAnalyzeContext *, Jsonb *, void *),
+ void *analyzearg)
+{
+ int row_num;
+
+ MemoryContext tmpcxt = AllocSetContextCreate(CurrentMemoryContext,
+ "Json Analyze Pass Context",
+ ALLOCSET_DEFAULT_MINSIZE,
+ ALLOCSET_DEFAULT_INITSIZE,
+ ALLOCSET_DEFAULT_MAXSIZE);
+
+ MemoryContext oldcxt = MemoryContextSwitchTo(tmpcxt);
+
+ ctx->null_cnt = 0;
+ ctx->analyzed_cnt = 0;
+ ctx->total_width = 0;
+
+ /* Loop over the arrays. */
+ for (row_num = 0; row_num < ctx->samplerows; row_num++)
+ {
+ Datum value;
+ Jsonb *jb;
+ Size width;
+ bool isnull;
+
+ vacuum_delay_point();
+
+ value = ctx->fetchfunc(ctx->stats, row_num, &isnull);
+
+ if (isnull)
+ {
+ /* json is null, just count that */
+ ctx->null_cnt++;
+ continue;
+ }
+
+ width = toast_raw_datum_size(value);
+
+ ctx->total_width += VARSIZE_ANY(DatumGetPointer(value)); /* FIXME raw width? */
+
+ /* Skip too-large values. */
+#define JSON_WIDTH_THRESHOLD (100 * 1024)
+
+ if (width > JSON_WIDTH_THRESHOLD)
+ continue;
+
+ ctx->analyzed_cnt++;
+
+ jb = DatumGetJsonbP(value);
+
+ MemoryContextSwitchTo(oldcxt);
+
+ analyzefunc(ctx, jb, analyzearg);
+
+ oldcxt = MemoryContextSwitchTo(tmpcxt);
+ MemoryContextReset(tmpcxt);
+ }
+
+ MemoryContextSwitchTo(oldcxt);
+}
+
+/*
+ * compute_json_stats() -- compute statistics for a json column
+ */
+static void
+compute_json_stats(VacAttrStats *stats, AnalyzeAttrFetchFunc fetchfunc,
+ int samplerows, double totalrows)
+{
+ JsonAnalyzeContext ctx;
+ JsonPathAnlStats **paths;
+ int npaths;
+ bool sigle_pass = false; /* FIXME make GUC or simply remove */
+
+ jsonAnalyzeInit(&ctx, stats, fetchfunc, samplerows, totalrows);
+
+ /*
+ * Collect and analyze JSON path values in single or multiple passes.
+ * Sigle-pass collection is faster but consumes much more memory than
+ * collecting and analyzing by the one path at pass.
+ */
+ if (sigle_pass)
+ {
+ /* Collect all values of all paths */
+ jsonAnalyzePass(&ctx, jsonAnalyzeCollectPaths, (void *)(intptr_t) true);
+
+ /*
+ * Now that we're done with processing the documents, we sort the paths
+ * we extracted and calculate stats for each of them.
+ *
+ * XXX I wonder if we could do this in two phases, to maybe not collect
+ * (or even accumulate) values for paths that are not interesting.
+ */
+ paths = jsonAnalyzeSortPaths(&ctx, &npaths);
+
+ for (int i = 0; i < npaths; i++)
+ jsonAnalyzePath(&ctx, paths[i]);
+ }
+ else
+ {
+ MemoryContext oldcxt;
+ MemoryContext tmpcxt = AllocSetContextCreate(CurrentMemoryContext,
+ "Json Analyze Tmp Context",
+ ALLOCSET_DEFAULT_MINSIZE,
+ ALLOCSET_DEFAULT_INITSIZE,
+ ALLOCSET_DEFAULT_MAXSIZE);
+
+ elog(DEBUG1, "analyzing %s attribute \"%s\"",
+ stats->attrtypid == JSONBOID ? "jsonb" : "json",
+ NameStr(stats->attr->attname));
+
+ elog(DEBUG1, "collecting json paths");
+
+ oldcxt = MemoryContextSwitchTo(tmpcxt);
+
+ /* Collect all paths first without accumulating any Values, sort them */
+ jsonAnalyzePass(&ctx, jsonAnalyzeCollectPaths, (void *)(intptr_t) false);
+ paths = jsonAnalyzeSortPaths(&ctx, &npaths);
+
+ /*
+ * Next, process each path independently to save memory (we don't want
+ * to accumulate all values for all paths, with a lot of duplicities).
+ */
+ MemoryContextReset(tmpcxt);
+
+ for (int i = 0; i < npaths; i++)
+ {
+ JsonPathAnlStats *path = paths[i];
+
+ elog(DEBUG1, "analyzing json path (%d/%d) %s",
+ i + 1, npaths, path->pathstr);
+
+ jsonAnalyzePass(&ctx, jsonAnalyzeCollectPath, path);
+ jsonAnalyzePath(&ctx, path);
+
+ MemoryContextReset(tmpcxt);
+ }
+
+ MemoryContextSwitchTo(oldcxt);
+
+ MemoryContextDelete(tmpcxt);
+ }
+
+ /* We can only compute real stats if we found some non-null values. */
+ if (ctx.null_cnt >= samplerows)
+ {
+ /* We found only nulls; assume the column is entirely null */
+ stats->stats_valid = true;
+ stats->stanullfrac = 1.0;
+ stats->stawidth = 0; /* "unknown" */
+ stats->stadistinct = 0.0; /* "unknown" */
+ }
+ else if (!ctx.analyzed_cnt)
+ {
+ int nonnull_cnt = samplerows - ctx.null_cnt;
+
+ /* We found some non-null values, but they were all too wide */
+ stats->stats_valid = true;
+ /* Do the simple null-frac and width stats */
+ stats->stanullfrac = (double) ctx.null_cnt / (double) samplerows;
+ stats->stawidth = ctx.total_width / (double) nonnull_cnt;
+ /* Assume all too-wide values are distinct, so it's a unique column */
+ stats->stadistinct = -1.0 * (1.0 - stats->stanullfrac);
+ }
+ else
+ {
+ VacAttrStats *jsstats = &ctx.root->vstats.jsons.stats;
+ int i;
+ int empty_slot = -1;
+
+ stats->stats_valid = true;
+
+ stats->stanullfrac = jsstats->stanullfrac;
+ stats->stawidth = jsstats->stawidth;
+ stats->stadistinct = jsstats->stadistinct;
+
+ /*
+ * We need to store the statistics the statistics slots. We simply
+ * store the regular stats in the first slots, and then we put the
+ * JSON stats into the first empty slot.
+ */
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ /* once we hit an empty slot, we're done */
+ if (!jsstats->staop[i])
+ {
+ empty_slot = i; /* remember the empty slot */
+ break;
+ }
+
+ stats->stakind[i] = jsstats->stakind[i];
+ stats->staop[i] = jsstats->staop[i];
+ stats->stanumbers[i] = jsstats->stanumbers[i];
+ stats->stavalues[i] = jsstats->stavalues[i];
+ stats->statypid[i] = jsstats->statypid[i];
+ stats->statyplen[i] = jsstats->statyplen[i];
+ stats->statypbyval[i] = jsstats->statypbyval[i];
+ stats->statypalign[i] = jsstats->statypalign[i];
+ stats->numnumbers[i] = jsstats->numnumbers[i];
+ stats->numvalues[i] = jsstats->numvalues[i];
+ }
+
+ Assert((empty_slot >= 0) && (empty_slot < STATISTIC_NUM_SLOTS));
+
+ stats->stakind[empty_slot] = STATISTIC_KIND_JSON;
+ stats->staop[empty_slot] = InvalidOid;
+ stats->numnumbers[empty_slot] = 1;
+ stats->stanumbers[empty_slot] = MemoryContextAlloc(stats->anl_context,
+ sizeof(float4));
+ stats->stanumbers[empty_slot][0] = 0.0; /* nullfrac */
+ stats->stavalues[empty_slot] =
+ jsonAnalyzeMakeStats(&ctx, paths, npaths,
+ &stats->numvalues[empty_slot]);
+
+ /* We are storing jsonb values */
+ stats->statypid[empty_slot] = JSONBOID;
+ get_typlenbyvalalign(stats->statypid[empty_slot],
+ &stats->statyplen[empty_slot],
+ &stats->statypbyval[empty_slot],
+ &stats->statypalign[empty_slot]);
+ }
+}
+
+/*
+ * json_typanalyze -- typanalyze function for jsonb
+ */
+Datum
+jsonb_typanalyze(PG_FUNCTION_ARGS)
+{
+ VacAttrStats *stats = (VacAttrStats *) PG_GETARG_POINTER(0);
+ Form_pg_attribute attr = stats->attr;
+
+ /* If the attstattarget column is negative, use the default value */
+ /* NB: it is okay to scribble on stats->attr since it's a copy */
+ if (attr->attstattarget < 0)
+ attr->attstattarget = default_statistics_target;
+
+ stats->compute_stats = compute_json_stats;
+ /* see comment about the choice of minrows in commands/analyze.c */
+ stats->minrows = 300 * attr->attstattarget;
+
+ PG_RETURN_BOOL(true);
+}
diff --git a/src/backend/utils/adt/jsonpath_exec.c b/src/backend/utils/adt/jsonpath_exec.c
index eff3734b6ab..94ae6e1f385 100644
--- a/src/backend/utils/adt/jsonpath_exec.c
+++ b/src/backend/utils/adt/jsonpath_exec.c
@@ -1723,7 +1723,7 @@ executeLikeRegex(JsonPathItem *jsp, JsonbValue *str, JsonbValue *rarg,
cxt->cflags = jspConvertRegexFlags(jsp->content.like_regex.flags);
}
- if (RE_compile_and_execute(cxt->regex, str->val.string.val,
+ if (RE_compile_and_execute(cxt->regex, unconstify(char *, str->val.string.val),
str->val.string.len,
cxt->cflags, DEFAULT_COLLATION_OID, 0, NULL))
return jpbTrue;
diff --git a/src/include/catalog/pg_operator.dat b/src/include/catalog/pg_operator.dat
index 8e0e65ad275..9805bb15038 100644
--- a/src/include/catalog/pg_operator.dat
+++ b/src/include/catalog/pg_operator.dat
@@ -3175,7 +3175,7 @@
{ oid => '3211', oid_symbol => 'JsonbObjectFieldOperator',
descr => 'get jsonb object field',
oprname => '->', oprleft => 'jsonb', oprright => 'text', oprresult => 'jsonb',
- oprcode => 'jsonb_object_field' },
+ oprcode => 'jsonb_object_field', oprstat => 'jsonb_stats' },
{ oid => '3477', oid_symbol => 'JsonbObjectFieldTextOperator',
descr => 'get jsonb object field as text',
oprname => '->>', oprleft => 'jsonb', oprright => 'text', oprresult => 'text',
@@ -3183,7 +3183,7 @@
{ oid => '3212', oid_symbol => 'JsonbArrayElementOperator',
descr => 'get jsonb array element',
oprname => '->', oprleft => 'jsonb', oprright => 'int4', oprresult => 'jsonb',
- oprcode => 'jsonb_array_element' },
+ oprcode => 'jsonb_array_element', oprstat => 'jsonb_stats' },
{ oid => '3481', oid_symbol => 'JsonbArrayElementTextOperator',
descr => 'get jsonb array element as text',
oprname => '->>', oprleft => 'jsonb', oprright => 'int4', oprresult => 'text',
@@ -3191,7 +3191,8 @@
{ oid => '3213', oid_symbol => 'JsonbExtractPathOperator',
descr => 'get value from jsonb with path elements',
oprname => '#>', oprleft => 'jsonb', oprright => '_text',
- oprresult => 'jsonb', oprcode => 'jsonb_extract_path' },
+ oprresult => 'jsonb', oprcode => 'jsonb_extract_path',
+ oprstat => 'jsonb_stats' },
{ oid => '3206', oid_symbol => 'JsonbExtractPathTextOperator',
descr => 'get value from jsonb as text with path elements',
oprname => '#>>', oprleft => 'jsonb', oprright => '_text',
@@ -3229,23 +3230,23 @@
{ oid => '3246', oid_symbol => 'JsonbContainsOperator', descr => 'contains',
oprname => '@>', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '<@(jsonb,jsonb)', oprcode => 'jsonb_contains',
- oprrest => 'matchingsel', oprjoin => 'matchingjoinsel' },
+ oprrest => 'jsonb_sel', oprjoin => 'matchingjoinsel' },
{ oid => '3247', oid_symbol => 'JsonbExistsOperator', descr => 'key exists',
oprname => '?', oprleft => 'jsonb', oprright => 'text', oprresult => 'bool',
- oprcode => 'jsonb_exists', oprrest => 'matchingsel',
+ oprcode => 'jsonb_exists', oprrest => 'jsonb_sel',
oprjoin => 'matchingjoinsel' },
{ oid => '3248', oid_symbol => 'JsonbExistsAnyOperator', descr => 'any key exists',
oprname => '?|', oprleft => 'jsonb', oprright => '_text', oprresult => 'bool',
- oprcode => 'jsonb_exists_any', oprrest => 'matchingsel',
+ oprcode => 'jsonb_exists_any', oprrest => 'jsonb_sel',
oprjoin => 'matchingjoinsel' },
{ oid => '3249', oid_symbol => 'JsonbExistsAllOperator', descr => 'all keys exist',
oprname => '?&', oprleft => 'jsonb', oprright => '_text', oprresult => 'bool',
- oprcode => 'jsonb_exists_all', oprrest => 'matchingsel',
+ oprcode => 'jsonb_exists_all', oprrest => 'jsonb_sel',
oprjoin => 'matchingjoinsel' },
{ oid => '3250', oid_symbol => 'JsonbContainedOperator', descr => 'is contained by',
oprname => '<@', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '@>(jsonb,jsonb)', oprcode => 'jsonb_contained',
- oprrest => 'matchingsel', oprjoin => 'matchingjoinsel' },
+ oprrest => 'jsonb_sel', oprjoin => 'matchingjoinsel' },
{ oid => '3284', descr => 'concatenate',
oprname => '||', oprleft => 'jsonb', oprright => 'jsonb',
oprresult => 'jsonb', oprcode => 'jsonb_concat' },
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index 0859dc81cac..2dc3f134ea5 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -11731,4 +11731,15 @@
prorettype => 'bytea', proargtypes => 'pg_brin_minmax_multi_summary',
prosrc => 'brin_minmax_multi_summary_send' },
+# jsonb statistics
+{ oid => '8526', descr => 'jsonb typanalyze',
+ proname => 'jsonb_typanalyze', provolatile => 's', prorettype => 'bool',
+ proargtypes => 'internal', prosrc => 'jsonb_typanalyze' },
+{ oid => '8527', descr => 'jsonb selectivity estimation',
+ proname => 'jsonb_sel', provolatile => 's', prorettype => 'float8',
+ proargtypes => 'internal oid internal int4', prosrc => 'jsonb_sel' },
+{ oid => '8528', descr => 'jsonb statsistics estimation',
+ proname => 'jsonb_stats', provolatile => 's', prorettype => 'void',
+ proargtypes => 'internal internal int4 internal', prosrc => 'jsonb_stats' },
+
]
diff --git a/src/include/catalog/pg_statistic.h b/src/include/catalog/pg_statistic.h
index cdf74481398..c4f53ebd1ba 100644
--- a/src/include/catalog/pg_statistic.h
+++ b/src/include/catalog/pg_statistic.h
@@ -277,6 +277,8 @@ DECLARE_FOREIGN_KEY((starelid, staattnum), pg_attribute, (attrelid, attnum));
*/
#define STATISTIC_KIND_BOUNDS_HISTOGRAM 7
+#define STATISTIC_KIND_JSON 8
+
#endif /* EXPOSE_TO_CLIENT_CODE */
#endif /* PG_STATISTIC_H */
diff --git a/src/include/catalog/pg_type.dat b/src/include/catalog/pg_type.dat
index df458794635..b867db42f45 100644
--- a/src/include/catalog/pg_type.dat
+++ b/src/include/catalog/pg_type.dat
@@ -445,7 +445,7 @@
typname => 'jsonb', typlen => '-1', typbyval => 'f', typcategory => 'U',
typsubscript => 'jsonb_subscript_handler', typinput => 'jsonb_in',
typoutput => 'jsonb_out', typreceive => 'jsonb_recv', typsend => 'jsonb_send',
- typalign => 'i', typstorage => 'x' },
+ typanalyze => 'jsonb_typanalyze', typalign => 'i', typstorage => 'x' },
{ oid => '4072', array_type_oid => '4073', descr => 'JSON path',
typname => 'jsonpath', typlen => '-1', typbyval => 'f', typcategory => 'U',
typinput => 'jsonpath_in', typoutput => 'jsonpath_out',
diff --git a/src/include/utils/json_selfuncs.h b/src/include/utils/json_selfuncs.h
new file mode 100644
index 00000000000..9a36567ae65
--- /dev/null
+++ b/src/include/utils/json_selfuncs.h
@@ -0,0 +1,113 @@
+/*-------------------------------------------------------------------------
+ *
+ * json_selfuncs.h
+ * JSON cost estimation functions.
+ *
+ *
+ * Portions Copyright (c) 2016-2022, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ * src/include/utils/json_selfuncs.h
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#ifndef JSON_SELFUNCS_H_
+#define JSON_SELFUNCS_H 1
+
+#include "postgres.h"
+#include "access/htup.h"
+#include "utils/jsonb.h"
+#include "utils/lsyscache.h"
+#include "utils/selfuncs.h"
+
+#define JSON_PATH_ROOT "$"
+#define JSON_PATH_ROOT_LEN 1
+
+#define JSON_PATH_ROOT_ARRAY "$[*]"
+#define JSON_PATH_ROOT_ARRAY_LEN 4
+
+typedef enum
+{
+ JsonPathStatsValues,
+ JsonPathStatsArrayLength,
+ JsonPathStatsObjectLength
+} JsonPathStatsType;
+
+typedef struct JsonStatData JsonStatData, *JsonStats;
+
+/* Per-path JSON stats */
+typedef struct JsonPathStatsData
+{
+ JsonStats data; /* pointer to per-column control structure */
+ Datum *datum; /* pointer to JSONB datum with stats data */
+ const char *path; /* path string, points directly to JSONB data */
+ int pathlen; /* path length */
+ JsonPathStatsType type; /* type of stats (values, lengths etc.) */
+} JsonPathStatsData, *JsonPathStats;
+
+/* Per-column JSON stats */
+struct JsonStatData
+{
+ HeapTuple statsTuple; /* original pg_statistic tuple */
+ AttStatsSlot attslot; /* data extracted from STATISTIC_KIND_JSON
+ * slot of statsTuple */
+ RelOptInfo *rel; /* Relation, or NULL if not identifiable */
+ Datum *pathdatums; /* path JSONB datums */
+ JsonPathStatsData *paths; /* cached paths */
+ int npaths; /* number of paths */
+ float4 nullfrac; /* NULL fraction */
+ const char *prefix; /* global path prefix which needs to be used
+ * for searching in pathdatums */
+ int prefixlen; /* path prefix length */
+ bool acl_ok; /* ACL check is Ok */
+};
+
+typedef enum JsonStatType
+{
+ JsonStatJsonb,
+ JsonStatJsonbWithoutSubpaths,
+ JsonStatText,
+ JsonStatFloat4,
+ JsonStatString,
+ JsonStatNumeric,
+ JsonStatFreq,
+} JsonStatType;
+
+extern bool jsonStatsInit(JsonStats stats, const VariableStatData *vardata);
+extern void jsonStatsRelease(JsonStats data);
+
+extern JsonPathStats jsonStatsGetPathByStr(JsonStats stats,
+ const char *path, int pathlen);
+
+extern JsonPathStats jsonPathStatsGetSubpath(JsonPathStats stats,
+ const char *subpath);
+
+extern bool jsonPathStatsGetNextSubpathStats(JsonPathStats stats,
+ JsonPathStats *keystats,
+ bool keysOnly);
+
+extern JsonPathStats jsonPathStatsGetArrayLengthStats(JsonPathStats pstats);
+extern JsonPathStats jsonPathStatsGetObjectLengthStats(JsonPathStats pstats);
+
+extern float4 jsonPathStatsGetFreq(JsonPathStats pstats, float4 defaultfreq);
+
+extern float4 jsonPathStatsGetTypeFreq(JsonPathStats pstats,
+ JsonbValueType type, float4 defaultfreq);
+
+extern float4 jsonPathStatsGetAvgArraySize(JsonPathStats pstats);
+
+extern Selectivity jsonPathStatsGetArrayIndexSelectivity(JsonPathStats pstats,
+ int index);
+
+extern Selectivity jsonSelectivity(JsonPathStats stats, Datum scalar, Oid oper);
+
+extern void jsonPathAppendEntry(StringInfo path, const char *entry);
+
+extern bool jsonAnalyzeBuildSubPathsData(Datum *pathsDatums,
+ int npaths, int index,
+ const char *path, int pathlen,
+ bool includeSubpaths, float4 nullfrac,
+ Datum *pvals, Datum *pnums);
+
+#endif /* JSON_SELFUNCS_H */
diff --git a/src/test/regress/expected/jsonb_stats.out b/src/test/regress/expected/jsonb_stats.out
new file mode 100644
index 00000000000..c7b1e644099
--- /dev/null
+++ b/src/test/regress/expected/jsonb_stats.out
@@ -0,0 +1,713 @@
+CREATE OR REPLACE FUNCTION explain_jsonb(sql_query text)
+RETURNS TABLE(explain_line json) AS
+$$
+BEGIN
+ RETURN QUERY EXECUTE 'EXPLAIN (ANALYZE, FORMAT json) ' || sql_query;
+END;
+$$ LANGUAGE plpgsql;
+CREATE OR REPLACE FUNCTION get_plan_and_actual_rows(sql_query text)
+RETURNS TABLE(plan integer, actual integer) AS
+$$
+ SELECT
+ (plan->>'Plan Rows')::integer plan,
+ (plan->>'Actual Rows')::integer actual
+ FROM (
+ SELECT explain_jsonb(sql_query) #> '{0,Plan,Plans,0}'
+ ) p(plan)
+$$ LANGUAGE sql;
+CREATE OR REPLACE FUNCTION check_estimate(sql_query text, accuracy real)
+RETURNS boolean AS
+$$
+ SELECT plan BETWEEN actual / (1 + accuracy) AND (actual + 1) * (1 + accuracy)
+ FROM (SELECT * FROM get_plan_and_actual_rows(sql_query)) x
+$$ LANGUAGE sql;
+CREATE OR REPLACE FUNCTION check_estimate2(sql_query text, accuracy real)
+RETURNS TABLE(min integer, max integer) AS
+$$
+ SELECT (actual * (1 - accuracy))::integer, ((actual + 1) * (1 + accuracy))::integer
+ FROM (SELECT * FROM get_plan_and_actual_rows(sql_query)) x
+$$ LANGUAGE sql;
+CREATE TABLE jsonb_stats_test(js jsonb);
+INSERT INTO jsonb_stats_test SELECT NULL FROM generate_series(1, 1000);
+INSERT INTO jsonb_stats_test SELECT 'null' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT 'true' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT 'false' FROM generate_series(1, 500);
+INSERT INTO jsonb_stats_test SELECT '12345' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT (1000 * (i % 10))::text::jsonb FROM generate_series(1, 400) i;
+INSERT INTO jsonb_stats_test SELECT i::text::jsonb FROM generate_series(1, 500) i;
+INSERT INTO jsonb_stats_test SELECT '"foo"' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT format('"bar%s"', i % 10)::jsonb FROM generate_series(1, 400) i;
+INSERT INTO jsonb_stats_test SELECT format('"baz%s"', i)::jsonb FROM generate_series(1, 500) i;
+INSERT INTO jsonb_stats_test SELECT '{}' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', 'bar') FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', 'baz' || (i % 10)) FROM generate_series(1, 300) i;
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', i % 10) FROM generate_series(1, 200) i;
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('"foo \"bar"', i % 10) FROM generate_series(1, 200) i;
+INSERT INTO jsonb_stats_test SELECT '[]' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT '["foo"]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[12345]' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT '[["foo"]]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[{"key": "foo"}]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, "foo"]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, 12345]' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT '[null, ["foo"]]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, {"key": "foo"}]' FROM generate_series(1, 200);
+-- Build random variable-length integer arrays
+SELECT setseed(0.0);
+ setseed
+---------
+
+(1 row)
+
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array())
+FROM generate_series(1, 1000);
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int))
+FROM generate_series(1, 4000);
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int,
+ floor(random() * 10)::int))
+FROM generate_series(1, 3000);
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int,
+ floor(random() * 10)::int,
+ floor(random() * 10)::int))
+FROM generate_series(1, 2000);
+ANALYZE jsonb_stats_test;
+CREATE OR REPLACE FUNCTION check_jsonb_stats_test_estimate(sql_condition text, accuracy real)
+RETURNS boolean AS
+$$
+ SELECT check_estimate('SELECT count(*) FROM jsonb_stats_test WHERE ' || sql_condition, accuracy)
+$$ LANGUAGE sql;
+DROP FUNCTION IF EXISTS check_jsonb_stats_test_estimate2(text, real);
+NOTICE: function check_jsonb_stats_test_estimate2(text,pg_catalog.float4) does not exist, skipping
+CREATE OR REPLACE FUNCTION check_jsonb_stats_test_estimate2(sql_condition text, accuracy real)
+RETURNS TABLE(plan integer, actual integer) AS
+$$
+ SELECT get_plan_and_actual_rows('SELECT count(*) FROM jsonb_stats_test WHERE ' || sql_condition)
+$$ LANGUAGE sql;
+-- Check NULL estimate
+SELECT check_jsonb_stats_test_estimate($$js IS NULL$$, 0.03);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key}' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{1000000}' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key1' -> 'bad_key2' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key1,bad_key2}' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key1' -> 1 IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key1,1}' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 -> 'foo' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{1000000,foo}' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' = '123'$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 = '123'$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check null eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = 'null'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> 'null'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check boolean eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = 'true'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> 'true'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js = 'false'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> 'false'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check numeric eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '12345'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js#>'{}' = '12345'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js = '3000'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js = '1234'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '6000'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check numeric range estimate
+SELECT check_jsonb_stats_test_estimate($$js < '0'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js < '100'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js < '1000'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js < '3456'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js < '10000'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js < '100000'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js > '100' AND js < '600'$$, 0.5);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js > '6800' AND js < '12000'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check string eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '"foo"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js = '"bar7"'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js = '"baz1234"'$$, 10);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '"bar4"'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check string range estimate
+SELECT check_jsonb_stats_test_estimate($$js > '"foo"'$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js > '"bar"'$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js > '"baz"'$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check object eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '{}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js > '{}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check object key eq estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"bar"'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"baz"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"baz5"'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{foo}' = '"bar"'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check object key range estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' >= '"baz2"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' < '"baz9"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' >= '"baz2"' AND js -> 'foo' < '"baz9"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check array eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '[]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js >= '[]' AND js < '{}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check variable-length array element eq estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 = '1'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 = '6'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 = '8'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 = '1'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check variable-length array element range estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 < '7'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 < '7'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 < '7'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 < '7'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check variable-length array containment estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[]'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1]'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[100]'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2]'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 100]'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2, 100]'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2, 3]'$$, 5);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '1'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '100'$$, 10);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 @> '1'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 @> '1'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 @> '1'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 @> '1'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 @> '[1]'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": []}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1]}'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [100]}'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2]}'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 100]}'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2, 100]}'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2, 3]}'$$, 3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- check misc containment
+SELECT check_jsonb_stats_test_estimate($$js @> '"foo"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '12345'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[12345]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '["foo"]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[["foo", "bar"]]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[["foo"]]'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[{"key": "foo"}]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[null]'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check object key null estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' IS NULL$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' IS NOT NULL$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> '"foo \"bar"' IS NOT NULL$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NOT NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check object key existence
+SELECT check_jsonb_stats_test_estimate($$js ? 'bad_key'$$, 10);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ? 'foo'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ? 'array'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ?| '{foo,bad_key}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ?| '{foo,array}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ?& '{foo,bad_key}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ?& '{foo,bar}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out
index d652f7b5fb4..9b633baa179 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -2537,6 +2537,38 @@ pg_stats_ext_exprs| SELECT cn.nspname AS schemaname,
LEFT JOIN pg_namespace sn ON ((sn.oid = s.stxnamespace)))
JOIN LATERAL ( SELECT unnest(pg_get_statisticsobjdef_expressions(s.oid)) AS expr,
unnest(sd.stxdexpr) AS a) stat ON ((stat.expr IS NOT NULL)));
+pg_stats_json| SELECT n.nspname AS schemaname,
+ c.relname AS tablename,
+ a.attname,
+ (paths.path ->> 'path'::text) AS json_path,
+ s.stainherit AS inherited,
+ (((paths.path -> 'json'::text) ->> 'nullfrac'::text))::real AS null_frac,
+ (((paths.path -> 'json'::text) ->> 'width'::text))::real AS avg_width,
+ (((paths.path -> 'json'::text) ->> 'distinct'::text))::real AS n_distinct,
+ ARRAY( SELECT val.value AS val
+ FROM jsonb_array_elements((((paths.path -> 'json'::text) -> 'mcv'::text) -> 'values'::text)) val(value)) AS most_common_vals,
+ ARRAY( SELECT ((num.value)::text)::real AS num
+ FROM jsonb_array_elements((((paths.path -> 'json'::text) -> 'mcv'::text) -> 'numbers'::text)) num(value)) AS most_common_freqs,
+ ARRAY( SELECT val.value AS val
+ FROM jsonb_array_elements((((paths.path -> 'json'::text) -> 'histogram'::text) -> 'values'::text)) val(value)) AS histogram_bounds,
+ ARRAY( SELECT ((val.value)::text)::integer AS val
+ FROM jsonb_array_elements((((paths.path -> 'array_length'::text) -> 'mcv'::text) -> 'values'::text)) val(value)) AS most_common_array_lengths,
+ ARRAY( SELECT ((num.value)::text)::real AS num
+ FROM jsonb_array_elements((((paths.path -> 'array_length'::text) -> 'mcv'::text) -> 'numbers'::text)) num(value)) AS most_common_array_length_freqs,
+ (((paths.path -> 'json'::text) ->> 'correlation'::text))::real AS correlation
+ FROM (((pg_statistic s
+ JOIN pg_class c ON ((c.oid = s.starelid)))
+ JOIN pg_attribute a ON (((c.oid = a.attrelid) AND (a.attnum = s.staattnum))))
+ LEFT JOIN pg_namespace n ON ((n.oid = c.relnamespace))),
+ LATERAL ( SELECT unnest((((
+ CASE
+ WHEN (s.stakind1 = 8) THEN s.stavalues1
+ WHEN (s.stakind2 = 8) THEN s.stavalues2
+ WHEN (s.stakind3 = 8) THEN s.stavalues3
+ WHEN (s.stakind4 = 8) THEN s.stavalues4
+ WHEN (s.stakind5 = 8) THEN s.stavalues5
+ ELSE NULL::anyarray
+ END)::text)::jsonb[])[2:]) AS path) paths;
pg_tables| SELECT n.nspname AS schemaname,
c.relname AS tablename,
pg_get_userbyid(c.relowner) AS tableowner,
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index 5b0c73d7e37..d108cc62107 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -112,7 +112,7 @@ test: select_views portals_p2 foreign_key cluster dependency guc bitmapops combo
# ----------
# Another group of parallel tests (JSON related)
# ----------
-test: json jsonb json_encoding jsonpath jsonpath_encoding jsonb_jsonpath
+test: json jsonb json_encoding jsonpath jsonpath_encoding jsonb_jsonpath jsonb_stats
# ----------
# Another group of parallel tests
diff --git a/src/test/regress/sql/jsonb_stats.sql b/src/test/regress/sql/jsonb_stats.sql
new file mode 100644
index 00000000000..fac71d09ded
--- /dev/null
+++ b/src/test/regress/sql/jsonb_stats.sql
@@ -0,0 +1,249 @@
+CREATE OR REPLACE FUNCTION explain_jsonb(sql_query text)
+RETURNS TABLE(explain_line json) AS
+$$
+BEGIN
+ RETURN QUERY EXECUTE 'EXPLAIN (ANALYZE, FORMAT json) ' || sql_query;
+END;
+$$ LANGUAGE plpgsql;
+
+CREATE OR REPLACE FUNCTION get_plan_and_actual_rows(sql_query text)
+RETURNS TABLE(plan integer, actual integer) AS
+$$
+ SELECT
+ (plan->>'Plan Rows')::integer plan,
+ (plan->>'Actual Rows')::integer actual
+ FROM (
+ SELECT explain_jsonb(sql_query) #> '{0,Plan,Plans,0}'
+ ) p(plan)
+$$ LANGUAGE sql;
+
+CREATE OR REPLACE FUNCTION check_estimate(sql_query text, accuracy real)
+RETURNS boolean AS
+$$
+ SELECT plan BETWEEN actual / (1 + accuracy) AND (actual + 1) * (1 + accuracy)
+ FROM (SELECT * FROM get_plan_and_actual_rows(sql_query)) x
+$$ LANGUAGE sql;
+
+CREATE OR REPLACE FUNCTION check_estimate2(sql_query text, accuracy real)
+RETURNS TABLE(min integer, max integer) AS
+$$
+ SELECT (actual * (1 - accuracy))::integer, ((actual + 1) * (1 + accuracy))::integer
+ FROM (SELECT * FROM get_plan_and_actual_rows(sql_query)) x
+$$ LANGUAGE sql;
+
+CREATE TABLE jsonb_stats_test(js jsonb);
+
+INSERT INTO jsonb_stats_test SELECT NULL FROM generate_series(1, 1000);
+
+INSERT INTO jsonb_stats_test SELECT 'null' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT 'true' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT 'false' FROM generate_series(1, 500);
+
+INSERT INTO jsonb_stats_test SELECT '12345' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT (1000 * (i % 10))::text::jsonb FROM generate_series(1, 400) i;
+INSERT INTO jsonb_stats_test SELECT i::text::jsonb FROM generate_series(1, 500) i;
+
+INSERT INTO jsonb_stats_test SELECT '"foo"' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT format('"bar%s"', i % 10)::jsonb FROM generate_series(1, 400) i;
+INSERT INTO jsonb_stats_test SELECT format('"baz%s"', i)::jsonb FROM generate_series(1, 500) i;
+
+INSERT INTO jsonb_stats_test SELECT '{}' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', 'bar') FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', 'baz' || (i % 10)) FROM generate_series(1, 300) i;
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', i % 10) FROM generate_series(1, 200) i;
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('"foo \"bar"', i % 10) FROM generate_series(1, 200) i;
+
+INSERT INTO jsonb_stats_test SELECT '[]' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT '["foo"]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[12345]' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT '[["foo"]]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[{"key": "foo"}]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, "foo"]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, 12345]' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT '[null, ["foo"]]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, {"key": "foo"}]' FROM generate_series(1, 200);
+
+-- Build random variable-length integer arrays
+SELECT setseed(0.0);
+
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array())
+FROM generate_series(1, 1000);
+
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int))
+FROM generate_series(1, 4000);
+
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int,
+ floor(random() * 10)::int))
+FROM generate_series(1, 3000);
+
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int,
+ floor(random() * 10)::int,
+ floor(random() * 10)::int))
+FROM generate_series(1, 2000);
+
+
+ANALYZE jsonb_stats_test;
+
+CREATE OR REPLACE FUNCTION check_jsonb_stats_test_estimate(sql_condition text, accuracy real)
+RETURNS boolean AS
+$$
+ SELECT check_estimate('SELECT count(*) FROM jsonb_stats_test WHERE ' || sql_condition, accuracy)
+$$ LANGUAGE sql;
+
+DROP FUNCTION IF EXISTS check_jsonb_stats_test_estimate2(text, real);
+
+CREATE OR REPLACE FUNCTION check_jsonb_stats_test_estimate2(sql_condition text, accuracy real)
+RETURNS TABLE(plan integer, actual integer) AS
+$$
+ SELECT get_plan_and_actual_rows('SELECT count(*) FROM jsonb_stats_test WHERE ' || sql_condition)
+$$ LANGUAGE sql;
+
+-- Check NULL estimate
+SELECT check_jsonb_stats_test_estimate($$js IS NULL$$, 0.03);
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key}' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js #> '{1000000}' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key1' -> 'bad_key2' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key1,bad_key2}' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key1' -> 1 IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key1,1}' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 -> 'foo' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js #> '{1000000,foo}' IS NULL$$, 0.01);
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' = '123'$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 = '123'$$, 0.01);
+
+-- Check null eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = 'null'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> 'null'$$, 0.1);
+
+-- Check boolean eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = 'true'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> 'true'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js = 'false'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> 'false'$$, 0.1);
+
+-- Check numeric eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '12345'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js#>'{}' = '12345'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js = '3000'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js = '1234'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js @> '6000'$$, 0.2);
+
+-- Check numeric range estimate
+SELECT check_jsonb_stats_test_estimate($$js < '0'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js < '100'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js < '1000'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js < '3456'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js < '10000'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js < '100000'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js > '100' AND js < '600'$$, 0.5);
+SELECT check_jsonb_stats_test_estimate($$js > '6800' AND js < '12000'$$, 0.1);
+
+-- Check string eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '"foo"'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js = '"bar7"'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js = '"baz1234"'$$, 10);
+SELECT check_jsonb_stats_test_estimate($$js @> '"bar4"'$$, 0.3);
+
+-- Check string range estimate
+SELECT check_jsonb_stats_test_estimate($$js > '"foo"'$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js > '"bar"'$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js > '"baz"'$$, 0.01);
+
+-- Check object eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '{}'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js > '{}'$$, 0.1);
+
+-- Check object key eq estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"bar"'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"baz"'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"baz5"'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js #> '{foo}' = '"bar"'$$, 0.2);
+
+-- Check object key range estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' >= '"baz2"'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' < '"baz9"'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' >= '"baz2"' AND js -> 'foo' < '"baz9"'$$, 0.1);
+
+-- Check array eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '[]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js >= '[]' AND js < '{}'$$, 0.1);
+
+-- Check variable-length array element eq estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 = '1'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 = '6'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 = '8'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 = '1'$$, 0.2);
+
+-- Check variable-length array element range estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 < '7'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 < '7'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 < '7'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 < '7'$$, 0.1);
+
+-- Check variable-length array containment estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[]'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1]'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[100]'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2]'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 100]'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2, 100]'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2, 3]'$$, 5);
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '1'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '100'$$, 10);
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 @> '1'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 @> '1'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 @> '1'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 @> '1'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 @> '[1]'$$, 0.3);
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": []}'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1]}'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [100]}'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2]}'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 100]}'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2, 100]}'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2, 3]}'$$, 3);
+
+-- check misc containment
+SELECT check_jsonb_stats_test_estimate($$js @> '"foo"'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '12345'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '[]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '[12345]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '["foo"]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '[["foo", "bar"]]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '[["foo"]]'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js @> '[{"key": "foo"}]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '[null]'$$, 0.3);
+
+-- Check object key null estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' IS NULL$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' IS NOT NULL$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> '"foo \"bar"' IS NOT NULL$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NOT NULL$$, 0.01);
+
+-- Check object key existence
+SELECT check_jsonb_stats_test_estimate($$js ? 'bad_key'$$, 10);
+SELECT check_jsonb_stats_test_estimate($$js ? 'foo'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js ? 'array'$$, 0.1);
+
+SELECT check_jsonb_stats_test_estimate($$js ?| '{foo,bad_key}'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js ?| '{foo,array}'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js ?& '{foo,bad_key}'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js ?& '{foo,bar}'$$, 0.1);
--
2.25.1
^ permalink raw reply [nested|flat] 9+ messages in thread
* Re: Collecting statistics about contents of JSONB columns
@ 2022-01-24 22:19 Tomas Vondra <[email protected]>
parent: Nikita Glukhov <[email protected]>
0 siblings, 1 reply; 9+ messages in thread
From: Tomas Vondra @ 2022-01-24 22:19 UTC (permalink / raw)
To: Nikita Glukhov <[email protected]>; +Cc: PostgreSQL Hackers <[email protected]>; Mahendra Thalor <[email protected]>; Oleg Bartunov <[email protected]>; Thomas Munro <[email protected]>
On 1/23/22 01:24, Nikita Glukhov wrote:
> Hi!
>
> I am glad that you found my very old patch interesting and started to
> work on it. We failed to post it in 2016 mostly because we were not
> satisfied with JSONB storage. Also we decided to wait for completion
> of work on extended statistics as we thought that it could help us.
> But in early 2017 we switched to SQL/JSON and forgot about this patch.
>
Understood. Let's see how feasible this idea is and if we can move this
forward.
>
> I think custom datatype is necessary for better performance. With a
> plain JSONB we need to do a lot of work for extraction of path stats:
> - iterate through MCV/Histogram JSONB arrays
> - cast numeric values to float, string to text etc.
> - build PG arrays from extracted datums
> - form pg_statistic tuple.
>
> With a custom data type we could store pg_statistic tuple unmodified
> and use it without any overhead. But then we need modify a bit
> VariableStatData and several functions to pass additional nullfrac
> corrections.
>
I'm not against evaluating/exploring alternative storage formats, but my
feeling is the real impact on performance will be fairly low. At least I
haven't seen this as very expensive while profiling the patch so far. Of
course, I may be wrong, and it may be more significant in some cases.
> Maybe simple record type (path text, data pg_statistic, ext_data jsonb)
> would be enough.
>
Maybe, but then you still need to store a bunch of those, right? So
either an array (likely toasted) or 1:M table. I'm not sure it's goiing
to be much cheaper than JSONB.
I'd suggest we focus on what we need to store first, which seems like
tha primary question, and worry about the exact storage format then.
> Also there is an idea to store per-path separately in pg_statistic_ext
> rows using expression like (jb #> '{foo,bar}') as stxexprs. This could
> also help user to select which paths to analyze simply by using some
> sort of CREATE STATISTICS. But it is really unclear how to:
> * store pg_statistic_ext rows from typanalyze
> * form path expressions for array elements (maybe add new jsonpath
> operator)
> * match various -> operator chains to stxexprs
> * jsonpath-like languages need simple API for searching by stxexprs
>
Sure, you can do statistics on expressions, right? Of course, if that
expression produces JSONB value, that's not very useful at the moment.
Maybe we could have two typanalyze functions - one for regular analyze,
one for extended statistics?
That being said, I'm not sure extended stats are a good match for this.
My feeling was we should collect these stats for all JSONB columns,
which is why I argued for putting that in pg_statistic.
>
>
> Per-path statistics should only be collected for scalars. This can be
> enabled by flag JsonAnalyzeContext.scalarsOnly. But there are is a
> problem: computed MCVs and histograms will be broken and we will not be
> able to use them for queries like (jb > const) in general case. Also
> we will not be and use internally in scalarineqsel() and var_eq_const()
> (see jsonSelectivity()). Instead, we will have to implement own
> estimator functions for JSONB comparison operators that will correctly
> use our hacked MCVs and histograms (and maybe not all cases will be
> supported; for example, comparison to scalars only).
>
Yeah, but maybe that's an acceptable trade-off? I mean, if we can
improve estimates for most clauses, and there's a some number of clauses
that are estimated just like without stats, that's still an improvement,
right?
> It's possible to replace objects and arrays with empty ones when
> scalarsOnly is set to keep correct frequencies of non-scalars.
> But there is an old bug in JSONB comparison: empty arrays are placed
> before other values in the JSONB sort order, although they should go
> after all scalars. So we need also to distinguish empty and non-empty
> arrays here.
>
Hmmm ...
>
>
> I tried to fix a major part of places marked as XXX and FIXME, the new
> version of the patches is attached. There are a lot of changes, you
> can see them in a step-by-step form in the corresponding branch
> jsonb_stats_20220122 in our GitHub repo [1].
>
Thanks! I'll go through the changes soon.
regards
--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
^ permalink raw reply [nested|flat] 9+ messages in thread
* Re: Collecting statistics about contents of JSONB columns
@ 2022-01-25 16:56 Mahendra Singh Thalor <[email protected]>
parent: Tomas Vondra <[email protected]>
0 siblings, 1 reply; 9+ messages in thread
From: Mahendra Singh Thalor @ 2022-01-25 16:56 UTC (permalink / raw)
To: Tomas Vondra <[email protected]>; +Cc: Nikita Glukhov <[email protected]>; PostgreSQL Hackers <[email protected]>; Mahendra Thalor <[email protected]>; Oleg Bartunov <[email protected]>; Thomas Munro <[email protected]>
On Tue, 25 Jan 2022 at 03:50, Tomas Vondra <[email protected]>
wrote:
>
> On 1/23/22 01:24, Nikita Glukhov wrote:
> > Hi!
> >
> > I am glad that you found my very old patch interesting and started to
> > work on it. We failed to post it in 2016 mostly because we were not
> > satisfied with JSONB storage. Also we decided to wait for completion
> > of work on extended statistics as we thought that it could help us.
> > But in early 2017 we switched to SQL/JSON and forgot about this patch.
> >
>
> Understood. Let's see how feasible this idea is and if we can move this
> forward.
>
> >
> > I think custom datatype is necessary for better performance. With a
> > plain JSONB we need to do a lot of work for extraction of path stats:
> > - iterate through MCV/Histogram JSONB arrays
> > - cast numeric values to float, string to text etc.
> > - build PG arrays from extracted datums
> > - form pg_statistic tuple.
> >
> > With a custom data type we could store pg_statistic tuple unmodified
> > and use it without any overhead. But then we need modify a bit
> > VariableStatData and several functions to pass additional nullfrac
> > corrections.
> >
>
> I'm not against evaluating/exploring alternative storage formats, but my
> feeling is the real impact on performance will be fairly low. At least I
> haven't seen this as very expensive while profiling the patch so far. Of
> course, I may be wrong, and it may be more significant in some cases.
>
> > Maybe simple record type (path text, data pg_statistic, ext_data jsonb)
> > would be enough.
> >
> Maybe, but then you still need to store a bunch of those, right? So
> either an array (likely toasted) or 1:M table. I'm not sure it's goiing
> to be much cheaper than JSONB.
>
> I'd suggest we focus on what we need to store first, which seems like
> tha primary question, and worry about the exact storage format then.
>
>
> > Also there is an idea to store per-path separately in pg_statistic_ext
> > rows using expression like (jb #> '{foo,bar}') as stxexprs. This could
> > also help user to select which paths to analyze simply by using some
> > sort of CREATE STATISTICS. But it is really unclear how to:
> > * store pg_statistic_ext rows from typanalyze
> > * form path expressions for array elements (maybe add new jsonpath
> > operator)
> > * match various -> operator chains to stxexprs
> > * jsonpath-like languages need simple API for searching by stxexprs
> >
>
> Sure, you can do statistics on expressions, right? Of course, if that
> expression produces JSONB value, that's not very useful at the moment.
> Maybe we could have two typanalyze functions - one for regular analyze,
> one for extended statistics?
>
> That being said, I'm not sure extended stats are a good match for this.
> My feeling was we should collect these stats for all JSONB columns,
> which is why I argued for putting that in pg_statistic.
>
> >
> >
> > Per-path statistics should only be collected for scalars. This can be
> > enabled by flag JsonAnalyzeContext.scalarsOnly. But there are is a
> > problem: computed MCVs and histograms will be broken and we will not be
> > able to use them for queries like (jb > const) in general case. Also
> > we will not be and use internally in scalarineqsel() and var_eq_const()
> > (see jsonSelectivity()). Instead, we will have to implement own
> > estimator functions for JSONB comparison operators that will correctly
> > use our hacked MCVs and histograms (and maybe not all cases will be
> > supported; for example, comparison to scalars only).
> >
>
> Yeah, but maybe that's an acceptable trade-off? I mean, if we can
> improve estimates for most clauses, and there's a some number of clauses
> that are estimated just like without stats, that's still an improvement,
> right?
>
> > It's possible to replace objects and arrays with empty ones when
> > scalarsOnly is set to keep correct frequencies of non-scalars.
> > But there is an old bug in JSONB comparison: empty arrays are placed
> > before other values in the JSONB sort order, although they should go
> > after all scalars. So we need also to distinguish empty and non-empty
> > arrays here.
> >
>
> Hmmm ...
> >
> >
> > I tried to fix a major part of places marked as XXX and FIXME, the new
> > version of the patches is attached. There are a lot of changes, you
> > can see them in a step-by-step form in the corresponding branch
> > jsonb_stats_20220122 in our GitHub repo [1].
> >
>
> Thanks! I'll go through the changes soon.
>
>
Thanks, Nikita and Tomas for these patches.
For the last few days, I was trying to understand these patches, and based
on Tomas's suggestion, I was doing some performance tests.
With the attached .SQL file, I can see that analyze is taking more time
with these patches.
*Setup: *
autovacuum=off
rest all are default settings.
Insert attached file with and without the patch to compare the time taken
by analyze.
*With json patches:*
postgres=# analyze test ;
ANALYZE
Time: *28464.062 ms (00:28.464)*
postgres=#
postgres=# SELECT pg_size_pretty(
pg_total_relation_size('pg_catalog.pg_statistic') );
pg_size_pretty
----------------
328 kB
(1 row)
--
*Without json patches:*
postgres=# analyze test ;
ANALYZE
*Time: 120.864* ms
postgres=# SELECT pg_size_pretty(
pg_total_relation_size('pg_catalog.pg_statistic') );
pg_size_pretty
----------------
272 kB
I haven't found the root cause of this but I feel that this time is due to
a loop of all the paths.
In my test data, there is a total of 951 different-2 paths. While doing
analysis, first we check all the sample rows(30000) and we collect all the
different-2 paths (here 951), and after that for every single path, we loop
over all the sample rows again to collect stats for a particular path. I
feel that these loops might be taking time.
I will run perf and will try to find out the root cause of this.
Apart from this performance issue, I haven't found any crashes or issues.
Thanks and Regards
Mahendra Singh Thalor
EnterpriseDB: http://www.enterprisedb.com
Attachments:
[application/octet-stream] test_jsonb_1.sql (21.6K, ../../CAKYtNArmMg=cU6SogL6u-b21Gx=rMoR9G1ah6kLANRw7LhHukA@mail.gmail.com/3-test_jsonb_1.sql)
download
^ permalink raw reply [nested|flat] 9+ messages in thread
* Re: Collecting statistics about contents of JSONB columns
@ 2022-02-04 02:47 Tomas Vondra <[email protected]>
parent: Mahendra Singh Thalor <[email protected]>
0 siblings, 2 replies; 9+ messages in thread
From: Tomas Vondra @ 2022-02-04 02:47 UTC (permalink / raw)
To: Mahendra Singh Thalor <[email protected]>; +Cc: Nikita Glukhov <[email protected]>; PostgreSQL Hackers <[email protected]>; Mahendra Thalor <[email protected]>; Oleg Bartunov <[email protected]>; Thomas Munro <[email protected]>
On 1/25/22 17:56, Mahendra Singh Thalor wrote:
>
> ...
>
> For the last few days, I was trying to understand these patches, and
> based on Tomas's suggestion, I was doing some performance tests.
>
> With the attached .SQL file, I can see that analyze is taking more time
> with these patches.
>
> *Setup: *
> autovacuum=off
> rest all are default settings.
>
> Insert attached file with and without the patch to compare the time
> taken by analyze.
>
> *With json patches:*
> postgres=# analyze test ;
> ANALYZE
> Time: *28464.062 ms (00:28.464)*
> postgres=#
> postgres=# SELECT pg_size_pretty(
> pg_total_relation_size('pg_catalog.pg_statistic') );
> pg_size_pretty
> ----------------
> 328 kB
> (1 row)
> --
>
> *Without json patches:*
> postgres=# analyze test ;
> ANALYZE
> *Time: 120.864* ms
> postgres=# SELECT pg_size_pretty(
> pg_total_relation_size('pg_catalog.pg_statistic') );
> pg_size_pretty
> ----------------
> 272 kB
>
> I haven't found the root cause of this but I feel that this time is due
> to a loop of all the paths.
> In my test data, there is a total of 951 different-2 paths. While doing
> analysis, first we check all the sample rows(30000) and we collect all
> the different-2 paths (here 951), and after that for every single path,
> we loop over all the sample rows again to collect stats for a particular
> path. I feel that these loops might be taking time.
>
> I will run perf and will try to find out the root cause of this.
>
Thanks, I've been doing some performance tests too, and you're right it
takes quite a bit of time. I wanted to compare how the timing changes
with complexity of the JSON documents (nesting, number of keys, ...) so
I wrote a simple python script to generate random JSON documents with
different parameters - see the attached json-generate.py script.
It's a bit crude, but it generates synthetic documents with a chosen
number of levels, keys per level, distinct key values, etc. The
generated documents are loaded directly into a "json_test" database,
into a table "test_table" with a single jsonb column called "v".
Tweaking this to connect to a different database, or just dump the
generated documents to a file, should be trivial.
The attached bash script runs the data generator for a couple of
combinations, and them measures how long it takes to analyze the table,
how large the statistics are (in a rather crude way), etc.
The results look like this (the last two columns are analyze duration in
milliseconds, for master and with the patch):
levels keys unique keys paths master patched
----------------------------------------------------------
1 1 1 2 153 122
1 1 1000 1001 134 1590
1 8 8 9 157 367
1 8 1000 1001 155 1838
1 64 64 65 189 2298
1 64 1000 1001 46 9322
2 1 1 3 237 197
2 1 1000 30580 152 46468
2 8 8 73 245 1780
So yeah, it's significantly slower - in most cases not as much as you
observed, but an order of magnitude slower than master. For size of the
statistics, it's similar:
levels keys unique keys paths table size master patched
------------------------------------------------------------------
1 1 1 2 1843200 16360 24325
1 1 1000 1001 1843200 16819 1425400
1 8 8 9 4710400 28948 88837
1 8 1000 1001 6504448 42694 3915802
1 64 64 65 30154752 209713 689842
1 64 1000 1001 49086464 1093 7755214
2 1 1 3 2572288 24883 48727
2 1 1000 30580 2572288 11422 26396365
2 8 8 73 23068672 164785 862258
This is measured by by dumping pg_statistic for the column, so in
database it might be compressed etc. It's larger, but that's somewhat
expected because we simply store more detailed stats. The size grows
with the number of paths extracted - which is expected, of course.
If you noticed why this doesn't show data for additional combinations
(e.g. 2 levels 8 keys and 1000 distinct key values), then that's the bad
news - that takes ages (multiple minutes) and then it gets killed by OOM
killer because it eats gigabytes of memory.
I agree the slowness is largely due to extracting all paths and then
processing them one by one - which means we have to loop over the tuples
over and over. In this case there's about 850k distinct paths extracted,
so we do ~850k loops over 30k tuples. That's gotta take time.
I don't know what exactly to do about this, but I already mentioned we
may need to pick a subset of paths to keep, similarly to how we pick
items for MCV. I mean, if we only saw a path once or twice, it's
unlikely to be interesting enough to build stats for it. I haven't
tried, but I'd bet most of the 850k paths might be ignored like this.
The other thing we might do is making it the loops more efficient. For
example, we might track which documents contain each path (by a small
bitmap or something), so that in the loop we can skip rows that don't
contain the path we're currently processing. Or something like that.
Of course, this can't eliminate all the overhead - we've building more
stats and that has a cost. In the "common" case of stable "fixed" schema
with the same paths in all documents we'll still need to do loop for
each of them. So it's bound to be slower than master.
Which probably means it's a bad idea to do this for all JSONB columns,
because in many cases the extra stats are not needed so the extra
analyze time would be a waste. So I guess we'll need some way to enable
this only for selected columns ... I argued against the idea to
implement this as extended statistics in the first message, but it's a
reasonably nice way to do such stuff (expression stats are a precedent).
> Apart from this performance issue, I haven't found any crashes or issues.
>
Well, I haven't seen any crashes either, but as I mentioned for complex
documents (2 levels, many distinct keys) the ANALYZE starts consuming a
lot of memory and may get killed by OOM. For example if you generate
documents like this
./json-generate.py 30000 2 8 1000 6 1000
and then run ANALYZE, that'll take ages and it very quickly gets into a
situation like this (generated from gdb by calling MemoryContextStats on
TopMemoryContext):
-------------------------------------------------------------------------
TopMemoryContext: 80776 total in 6 blocks; 13992 free (18 chunks); 66784
used
...
TopPortalContext: 8192 total in 1 blocks; 7656 free (0 chunks); 536 used
PortalContext: 1024 total in 1 blocks; 488 free (0 chunks); 536
used: <unnamed>
Analyze: 472726496 total in 150 blocks; 3725776 free (4 chunks);
469000720 used
Analyze Column: 921177696 total in 120 blocks; 5123256 free
(238 chunks); 916054440 used
Json Analyze Tmp Context: 8192 total in 1 blocks; 5720 free
(1 chunks); 2472 used
Json Analyze Pass Context: 8192 total in 1 blocks; 7928
free (0 chunks); 264 used
JSON analyze path table: 1639706040 total in 25084 blocks;
1513640 free (33 chunks); 1638192400 used
Vacuum: 8192 total in 1 blocks; 7448 free (0 chunks); 744 used
...
Grand total: 3035316184 bytes in 25542 blocks; 10971120 free (352
chunks); 3024345064 used
-------------------------------------------------------------------------
Yes, that's backend 3GB of memory, out of which 1.6GB is in "analyze
path table" context, 400MB in "analyze" and 900MB in "analyze column"
contexts. I mean, that seems a bit excessive. And it grows over time, so
after a while my laptop gives up and kills the backend.
I'm not sure if it's a memory leak (which would be fixable), or it's due
to keeping stats for all the extracted paths. I mean, in this particular
case we have 850k paths - even if stats are just 1kB per path, that's
850MB. This requires more investigation.
regards
--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
Attachments:
[text/x-python] json-generate.py (1.8K, ../../[email protected]/2-json-generate.py)
download | inline:
#!/usr/bin/python
# python json-generate.py NUM_OF_DOCUMENT NEST_LEVELS KEYS_PER_LEVEL UNIQUE_KEYS KEY_LENGTH UNIQUE_VALUES
import psycopg2
import hashlib
import random
import sys
from psycopg2.extras import Json
# number of document to generate
ndocuments = int(sys.argv[1])
# number of levels of nesting
nlevels = int(sys.argv[2])
# number of distinct paths
nkeys_unique = int(sys.argv[3])
# number of paths at each level (so total npaths_per_level ** nlevels)
nkeys_per_level = int(sys.argv[4])
# string length
keys_length = int(sys.argv[5])
# unique values (in all paths combined)
nvalues_unique = int(sys.argv[6])
# pre-generate random paths
paths = [hashlib.md5(str(i).encode()).hexdigest()[0:keys_length] for i in range(0,nkeys_unique)]
# pick paths randomly from the pre-generated array
def generate_paths(npaths):
return random.choices(paths, k=npaths)
# generate random value (simply an integer in a given range)
def generate_value():
return random.randint(0, nvalues_unique)
# generate document with nlevels, recursively
def generate_document(npaths, nlevels):
# this is the deepest level, so generate random
if nlevels == 0:
return generate_value()
# not the deepest level, so generate a dictionary with paths/values
doc = {}
paths = generate_paths(npaths)
# for each path, generate a sub-document (or a value)
for p in paths:
doc.update({p : generate_document(npaths, nlevels-1)})
return doc
if __name__ == '__main__':
conn = psycopg2.connect('host=localhost dbname=json_test')
cur = conn.cursor()
cur.execute('begin')
for i in range(0, ndocuments):
# batches
if i > 0 and i % 1000 == 0:
cur.execute('commit')
cur.execute('begin')
cur.execute('insert into test_table (v) values (%(json)s)', {'json' : Json(generate_document(nkeys_per_level, nlevels))})
if i % 1000 != 0:
cur.execute('commit')
[application/x-shellscript] json-test.sh (1.2K, ../../[email protected]/3-json-test.sh)
download
[image/svg+xml] json-stats-perf.svg (34.7K, ../../[email protected]/4-json-stats-perf.svg)
download | view image
^ permalink raw reply [nested|flat] 9+ messages in thread
* Re: Collecting statistics about contents of JSONB columns
@ 2022-02-04 03:00 Tomas Vondra <[email protected]>
parent: Tomas Vondra <[email protected]>
1 sibling, 0 replies; 9+ messages in thread
From: Tomas Vondra @ 2022-02-04 03:00 UTC (permalink / raw)
To: Mahendra Singh Thalor <[email protected]>; +Cc: Nikita Glukhov <[email protected]>; PostgreSQL Hackers <[email protected]>; Mahendra Thalor <[email protected]>; Oleg Bartunov <[email protected]>; Thomas Munro <[email protected]>
On 2/4/22 03:47, Tomas Vondra wrote:
> ./json-generate.py 30000 2 8 1000 6 1000
Sorry, this should be (different order of parameters):
./json-generate.py 30000 2 1000 8 6 1000
regards
--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
^ permalink raw reply [nested|flat] 9+ messages in thread
* Re: Collecting statistics about contents of JSONB columns
@ 2022-03-10 22:58 Nikita Glukhov <[email protected]>
parent: Tomas Vondra <[email protected]>
1 sibling, 1 reply; 9+ messages in thread
From: Nikita Glukhov @ 2022-03-10 22:58 UTC (permalink / raw)
To: Tomas Vondra <[email protected]>; Mahendra Singh Thalor <[email protected]>; +Cc: PostgreSQL Hackers <[email protected]>; Mahendra Thalor <[email protected]>; Oleg Bartunov <[email protected]>; Thomas Munro <[email protected]>
On 04.02.2022 05:47, Tomas Vondra wrote:
> On 1/25/22 17:56, Mahendra Singh Thalor wrote:
> >
>> ...
>>
>> For the last few days, I was trying to understand these patches, and
>> based on Tomas's suggestion, I was doing some performance tests.
>>
>> With the attached .SQL file, I can see that analyze is taking more
>> time with these patches.
>>
>> I haven't found the root cause of this but I feel that this time is
>> due to a loop of all the paths.
>> In my test data, there is a total of 951 different-2 paths. While
>> doing analysis, first we check all the sample rows(30000) and we
>> collect all the different-2 paths (here 951), and after that for
>> every single path, we loop over all the sample rows again to collect
>> stats for a particular path. I feel that these loops might be taking
>> time.
>>
> Thanks, I've been doing some performance tests too, and you're right
> it takes quite a bit of time.
That is absolutely not surprising, I have warned about poor performance
in cases with a large number of paths.
> I agree the slowness is largely due to extracting all paths and then
> processing them one by one - which means we have to loop over the
> tuples over and over. In this case there's about 850k distinct paths
> extracted, so we do ~850k loops over 30k tuples. That's gotta take time.
>
> I don't know what exactly to do about this, but I already mentioned we
> may need to pick a subset of paths to keep, similarly to how we pick
> items for MCV. I mean, if we only saw a path once or twice, it's
> unlikely to be interesting enough to build stats for it. I haven't
> tried, but I'd bet most of the 850k paths might be ignored like this.
>
> The other thing we might do is making it the loops more efficient. For
> example, we might track which documents contain each path (by a small
> bitmap or something), so that in the loop we can skip rows that don't
> contain the path we're currently processing. Or something like that.
>
>> Apart from this performance issue, I haven't found any crashes or
>> issues.
>>
>
> Well, I haven't seen any crashes either, but as I mentioned for
> complex documents (2 levels, many distinct keys) the ANALYZE starts
> consuming a lot of memory and may get killed by OOM. For example if
> you generate documents like this
>
> ./json-generate.py 30000 2 8 1000 6 1000
>
> and then run ANALYZE, that'll take ages and it very quickly gets into
> a situation like this (generated from gdb by calling
> MemoryContextStats on TopMemoryContext): and then run ANALYZE, that'll
> take ages and it very quickly gets into a situation like this
> (generated from gdb by calling MemoryContextStats on TopMemoryContext):
>
> -------------------------------------------------------------------------
> TopMemoryContext: 80776 total in 6 blocks; 13992 free (18 chunks);
> 66784 used
> ...
> TopPortalContext: 8192 total in 1 blocks; 7656 free (0 chunks); 536
> used
> PortalContext: 1024 total in 1 blocks; 488 free (0 chunks); 536
> used: <unnamed>
> Analyze: 472726496 total in 150 blocks; 3725776 free (4 chunks);
> 469000720 used
> Analyze Column: 921177696 total in 120 blocks; 5123256 free
> (238 chunks); 916054440 used
> Json Analyze Tmp Context: 8192 total in 1 blocks; 5720 free
> (1 chunks); 2472 used
> Json Analyze Pass Context: 8192 total in 1 blocks; 7928
> free (0 chunks); 264 used
> JSON analyze path table: 1639706040 total in 25084 blocks;
> 1513640 free (33 chunks); 1638192400 used
> Vacuum: 8192 total in 1 blocks; 7448 free (0 chunks); 744 used
> ...
> Grand total: 3035316184 bytes in 25542 blocks; 10971120 free (352
> chunks); 3024345064 used
> -------------------------------------------------------------------------
>
>
> Yes, that's backend 3GB of memory, out of which 1.6GB is in "analyze
> path table" context, 400MB in "analyze" and 900MB in "analyze column"
> contexts. I mean, that seems a bit excessive. And it grows over time,
> so after a while my laptop gives up and kills the backend.
>
> I'm not sure if it's a memory leak (which would be fixable), or it's
> due to keeping stats for all the extracted paths. I mean, in this
> particular case we have 850k paths - even if stats are just 1kB per
> path, that's 850MB. This requires more investigation.
Thank you for the tests and investigation.
I have tried to reduce memory consumption and speed up row scanning:
1. "JSON analyze path table" context contained ~1KB JsonPathAnlStats
structure per JSON path in the global hash table. I have moved
JsonPathAnlStats to the stack of compute_json_stats(), and hash
table now consumes ~70 bytes per path.
2. I have fixed copying of resulting JSONB stats into context, which
reduced the size of "Analyze Column" context.
3. I have optimized consumption of single-pass algorithm by storing
only value lists in the non-temporary context. That helped to
execute "2 64 64" test case in 30 seconds. Single-pass is a
bit faster in non-TOASTed cases, and much faster in TOASTed.
But it consumes much more memory and still goes to OOM in the
cases with more than ~100k paths.
4. I have implemented per-path document lists/bitmaps, which really
speed up the case "2 8 1000". List is converted into bitmap when
it becomes larger than bitmap.
5. Also I have fixed some bugs.
All these changes you can find commit form in our GitHub repository
on the branch jsonb_stats_20220310 [1].
Updated results of the test:
levels keys uniq keys paths master multi-pass single-pass
ms MB ms MB
-------------------------------------------------------------------
1 1 1 2 153 122 10 82 14
1 1 1000 1001 134 105 11 78 38
1 8 8 9 157 384 19 328 32
1 8 1000 1001 155 454 23 402 72
1 64 64 65 129 2889 45 2386 155
1 64 1000 1001 158 3990 94 1447 177
2 1 1 3 237 147 10 91 16
2 1 1000 30577 152 264 32 394 234
2 8 8 72 245 1943 37 1692 139
2 8 1000 852333 152 9175 678 OOM
2 64 64 4161 1784 ~1 hour 53 30018 1750
2 64 1000 1001001 4715 ~4 hours 1600 OOM
The two last multi-pass results are too slow, because JSONBs becomes
TOASTed. For measuring master in these tests, I disabled
WIDTH_THRESHOLD check which skipped TOASTed values > 1KB.
Next, I am going to try to disable all-paths collection and implement
collection of most common paths (and/or hashed paths maybe).
[1]https://github.com/postgrespro/postgres/tree/jsonb_stats_20220310
--
Nikita Glukhov
Postgres Professional:http://www.postgrespro.com
The Russian Postgres Company
Attachments:
[text/x-patch] 0001-Add-pg_operator.oprstat-for-derived-operator-statist-20220310.patch (11.9K, ../../[email protected]/3-0001-Add-pg_operator.oprstat-for-derived-operator-statist-20220310.patch)
download | inline diff:
From 300b43180b287424b595c237b452d55f4d582bc5 Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Sat, 12 Nov 2016 18:59:43 +0300
Subject: [PATCH 1/6] Add pg_operator.oprstat for derived operator statistics
estimation
---
src/backend/catalog/pg_operator.c | 11 +++++
src/backend/commands/operatorcmds.c | 61 ++++++++++++++++++++++++++
src/backend/utils/adt/selfuncs.c | 38 ++++++++++++++++
src/backend/utils/cache/lsyscache.c | 24 ++++++++++
src/include/catalog/pg_operator.h | 4 ++
src/include/utils/lsyscache.h | 1 +
src/test/regress/expected/oidjoins.out | 1 +
7 files changed, 140 insertions(+)
diff --git a/src/backend/catalog/pg_operator.c b/src/backend/catalog/pg_operator.c
index 630bf3e56cc..9205e62c0eb 100644
--- a/src/backend/catalog/pg_operator.c
+++ b/src/backend/catalog/pg_operator.c
@@ -256,6 +256,7 @@ OperatorShellMake(const char *operatorName,
values[Anum_pg_operator_oprcode - 1] = ObjectIdGetDatum(InvalidOid);
values[Anum_pg_operator_oprrest - 1] = ObjectIdGetDatum(InvalidOid);
values[Anum_pg_operator_oprjoin - 1] = ObjectIdGetDatum(InvalidOid);
+ values[Anum_pg_operator_oprstat - 1] = ObjectIdGetDatum(InvalidOid);
/*
* create a new operator tuple
@@ -301,6 +302,7 @@ OperatorShellMake(const char *operatorName,
* negatorName X negator operator
* restrictionId X restriction selectivity procedure ID
* joinId X join selectivity procedure ID
+ * statsId X statistics derivation procedure ID
* canMerge merge join can be used with this operator
* canHash hash join can be used with this operator
*
@@ -333,6 +335,7 @@ OperatorCreate(const char *operatorName,
List *negatorName,
Oid restrictionId,
Oid joinId,
+ Oid statsId,
bool canMerge,
bool canHash)
{
@@ -505,6 +508,7 @@ OperatorCreate(const char *operatorName,
values[Anum_pg_operator_oprcode - 1] = ObjectIdGetDatum(procedureId);
values[Anum_pg_operator_oprrest - 1] = ObjectIdGetDatum(restrictionId);
values[Anum_pg_operator_oprjoin - 1] = ObjectIdGetDatum(joinId);
+ values[Anum_pg_operator_oprstat - 1] = ObjectIdGetDatum(statsId);
pg_operator_desc = table_open(OperatorRelationId, RowExclusiveLock);
@@ -855,6 +859,13 @@ makeOperatorDependencies(HeapTuple tuple,
add_exact_object_address(&referenced, addrs);
}
+ /* Dependency on statistics derivation function */
+ if (OidIsValid(oper->oprstat))
+ {
+ ObjectAddressSet(referenced, ProcedureRelationId, oper->oprstat);
+ add_exact_object_address(&referenced, addrs);
+ }
+
record_object_address_dependencies(&myself, addrs, DEPENDENCY_NORMAL);
free_object_addresses(addrs);
diff --git a/src/backend/commands/operatorcmds.c b/src/backend/commands/operatorcmds.c
index a5924d7d564..adf13e648a6 100644
--- a/src/backend/commands/operatorcmds.c
+++ b/src/backend/commands/operatorcmds.c
@@ -52,6 +52,7 @@
static Oid ValidateRestrictionEstimator(List *restrictionName);
static Oid ValidateJoinEstimator(List *joinName);
+static Oid ValidateStatisticsDerivator(List *joinName);
/*
* DefineOperator
@@ -78,10 +79,12 @@ DefineOperator(List *names, List *parameters)
List *commutatorName = NIL; /* optional commutator operator name */
List *negatorName = NIL; /* optional negator operator name */
List *restrictionName = NIL; /* optional restrict. sel. function */
+ List *statisticsName = NIL; /* optional stats estimat. procedure */
List *joinName = NIL; /* optional join sel. function */
Oid functionOid; /* functions converted to OID */
Oid restrictionOid;
Oid joinOid;
+ Oid statisticsOid;
Oid typeId[2]; /* to hold left and right arg */
int nargs;
ListCell *pl;
@@ -131,6 +134,8 @@ DefineOperator(List *names, List *parameters)
restrictionName = defGetQualifiedName(defel);
else if (strcmp(defel->defname, "join") == 0)
joinName = defGetQualifiedName(defel);
+ else if (strcmp(defel->defname, "statistics") == 0)
+ statisticsName = defGetQualifiedName(defel);
else if (strcmp(defel->defname, "hashes") == 0)
canHash = defGetBoolean(defel);
else if (strcmp(defel->defname, "merges") == 0)
@@ -246,6 +251,10 @@ DefineOperator(List *names, List *parameters)
joinOid = ValidateJoinEstimator(joinName);
else
joinOid = InvalidOid;
+ if (statisticsName)
+ statisticsOid = ValidateStatisticsDerivator(statisticsName);
+ else
+ statisticsOid = InvalidOid;
/*
* now have OperatorCreate do all the work..
@@ -260,6 +269,7 @@ DefineOperator(List *names, List *parameters)
negatorName, /* optional negator operator name */
restrictionOid, /* optional restrict. sel. function */
joinOid, /* optional join sel. function name */
+ statisticsOid, /* optional stats estimation procedure */
canMerge, /* operator merges */
canHash); /* operator hashes */
}
@@ -357,6 +367,40 @@ ValidateJoinEstimator(List *joinName)
return joinOid;
}
+/*
+ * Look up a statistics estimator function by name, and verify that it has the
+ * correct signature and we have the permissions to attach it to an operator.
+ */
+static Oid
+ValidateStatisticsDerivator(List *statName)
+{
+ Oid typeId[4];
+ Oid statOid;
+ AclResult aclresult;
+
+ typeId[0] = INTERNALOID; /* PlannerInfo */
+ typeId[1] = INTERNALOID; /* OpExpr */
+ typeId[2] = INT4OID; /* varRelid */
+ typeId[3] = INTERNALOID; /* VariableStatData */
+
+ statOid = LookupFuncName(statName, 4, typeId, false);
+
+ /* statistics estimators must return void */
+ if (get_func_rettype(statOid) != VOIDOID)
+ ereport(ERROR,
+ (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
+ errmsg("statistics estimator function %s must return type %s",
+ NameListToString(statName), "void")));
+
+ /* Require EXECUTE rights for the estimator */
+ aclresult = pg_proc_aclcheck(statOid, GetUserId(), ACL_EXECUTE);
+ if (aclresult != ACLCHECK_OK)
+ aclcheck_error(aclresult, OBJECT_FUNCTION,
+ NameListToString(statName));
+
+ return statOid;
+}
+
/*
* Guts of operator deletion.
*/
@@ -424,6 +468,9 @@ AlterOperator(AlterOperatorStmt *stmt)
List *joinName = NIL; /* optional join sel. function */
bool updateJoin = false;
Oid joinOid;
+ List *statName = NIL; /* optional statistics estimation procedure */
+ bool updateStat = false;
+ Oid statOid;
/* Look up the operator */
oprId = LookupOperWithArgs(stmt->opername, false);
@@ -454,6 +501,11 @@ AlterOperator(AlterOperatorStmt *stmt)
joinName = param;
updateJoin = true;
}
+ else if (pg_strcasecmp(defel->defname, "stats") == 0)
+ {
+ statName = param;
+ updateStat = true;
+ }
/*
* The rest of the options that CREATE accepts cannot be changed.
@@ -496,6 +548,10 @@ AlterOperator(AlterOperatorStmt *stmt)
joinOid = ValidateJoinEstimator(joinName);
else
joinOid = InvalidOid;
+ if (statName)
+ statOid = ValidateStatisticsDerivator(statName);
+ else
+ statOid = InvalidOid;
/* Perform additional checks, like OperatorCreate does */
if (!(OidIsValid(oprForm->oprleft) && OidIsValid(oprForm->oprright)))
@@ -536,6 +592,11 @@ AlterOperator(AlterOperatorStmt *stmt)
replaces[Anum_pg_operator_oprjoin - 1] = true;
values[Anum_pg_operator_oprjoin - 1] = joinOid;
}
+ if (updateStat)
+ {
+ replaces[Anum_pg_operator_oprstat - 1] = true;
+ values[Anum_pg_operator_oprstat - 1] = statOid;
+ }
tup = heap_modify_tuple(tup, RelationGetDescr(catalog),
values, nulls, replaces);
diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c
index 1fbb0b28c3b..d1dd049f1ae 100644
--- a/src/backend/utils/adt/selfuncs.c
+++ b/src/backend/utils/adt/selfuncs.c
@@ -4917,6 +4917,30 @@ ReleaseDummy(HeapTuple tuple)
pfree(tuple);
}
+/*
+ * examine_operator_expression
+ * Try to derive optimizer statistics for the operator expression using
+ * operator's oprstat function.
+ *
+ * There can be another OpExpr in one of the arguments, and it will be called
+ * recursively from the oprstat procedure through the following chain:
+ * get_restriction_variable() => examine_variable() =>
+ * examine_operator_expression().
+ */
+static void
+examine_operator_expression(PlannerInfo *root, OpExpr *opexpr, int varRelid,
+ VariableStatData *vardata)
+{
+ RegProcedure oprstat = get_oprstat(opexpr->opno);
+
+ if (OidIsValid(oprstat))
+ OidFunctionCall4(oprstat,
+ PointerGetDatum(root),
+ PointerGetDatum(opexpr),
+ Int32GetDatum(varRelid),
+ PointerGetDatum(vardata));
+}
+
/*
* examine_variable
* Try to look up statistical data about an expression.
@@ -5332,6 +5356,20 @@ examine_variable(PlannerInfo *root, Node *node, int varRelid,
pos++;
}
}
+
+ /*
+ * If there's no index or extended statistics matching the expression,
+ * try deriving the statistics from statistics on arguments of the
+ * operator expression (OpExpr). We do this last because it may be quite
+ * expensive, and it's unclear how accurate the statistics will be.
+ *
+ * More restrictions on the OpExpr (e.g. that one of the arguments
+ * has to be a Const or something) can be put by the operator itself
+ * in its oprstat function.
+ */
+ if (!vardata->statsTuple && IsA(basenode, OpExpr))
+ examine_operator_expression(root, (OpExpr *) basenode, varRelid,
+ vardata);
}
}
diff --git a/src/backend/utils/cache/lsyscache.c b/src/backend/utils/cache/lsyscache.c
index feef9998634..b5440b596cd 100644
--- a/src/backend/utils/cache/lsyscache.c
+++ b/src/backend/utils/cache/lsyscache.c
@@ -1567,6 +1567,30 @@ get_oprjoin(Oid opno)
return (RegProcedure) InvalidOid;
}
+/*
+ * get_oprstat
+ *
+ * Returns procedure id for estimating statistics for an operator.
+ */
+RegProcedure
+get_oprstat(Oid opno)
+{
+ HeapTuple tp;
+
+ tp = SearchSysCache1(OPEROID, ObjectIdGetDatum(opno));
+ if (HeapTupleIsValid(tp))
+ {
+ Form_pg_operator optup = (Form_pg_operator) GETSTRUCT(tp);
+ RegProcedure result;
+
+ result = optup->oprstat;
+ ReleaseSysCache(tp);
+ return result;
+ }
+ else
+ return (RegProcedure) InvalidOid;
+}
+
/* ---------- FUNCTION CACHE ---------- */
/*
diff --git a/src/include/catalog/pg_operator.h b/src/include/catalog/pg_operator.h
index 51263f550fe..ff1bb339f75 100644
--- a/src/include/catalog/pg_operator.h
+++ b/src/include/catalog/pg_operator.h
@@ -73,6 +73,9 @@ CATALOG(pg_operator,2617,OperatorRelationId)
/* OID of join estimator, or 0 */
regproc oprjoin BKI_DEFAULT(-) BKI_LOOKUP_OPT(pg_proc);
+
+ /* OID of statistics estimator, or 0 */
+ regproc oprstat BKI_DEFAULT(-) BKI_LOOKUP_OPT(pg_proc);
} FormData_pg_operator;
/* ----------------
@@ -95,6 +98,7 @@ extern ObjectAddress OperatorCreate(const char *operatorName,
List *negatorName,
Oid restrictionId,
Oid joinId,
+ Oid statisticsId,
bool canMerge,
bool canHash);
diff --git a/src/include/utils/lsyscache.h b/src/include/utils/lsyscache.h
index b8dd27d4a96..cc08fc50a50 100644
--- a/src/include/utils/lsyscache.h
+++ b/src/include/utils/lsyscache.h
@@ -118,6 +118,7 @@ extern Oid get_commutator(Oid opno);
extern Oid get_negator(Oid opno);
extern RegProcedure get_oprrest(Oid opno);
extern RegProcedure get_oprjoin(Oid opno);
+extern RegProcedure get_oprstat(Oid opno);
extern char *get_func_name(Oid funcid);
extern Oid get_func_namespace(Oid funcid);
extern Oid get_func_rettype(Oid funcid);
diff --git a/src/test/regress/expected/oidjoins.out b/src/test/regress/expected/oidjoins.out
index 215eb899be3..111ea99cdae 100644
--- a/src/test/regress/expected/oidjoins.out
+++ b/src/test/regress/expected/oidjoins.out
@@ -113,6 +113,7 @@ NOTICE: checking pg_operator {oprnegate} => pg_operator {oid}
NOTICE: checking pg_operator {oprcode} => pg_proc {oid}
NOTICE: checking pg_operator {oprrest} => pg_proc {oid}
NOTICE: checking pg_operator {oprjoin} => pg_proc {oid}
+NOTICE: checking pg_operator {oprstat} => pg_proc {oid}
NOTICE: checking pg_opfamily {opfmethod} => pg_am {oid}
NOTICE: checking pg_opfamily {opfnamespace} => pg_namespace {oid}
NOTICE: checking pg_opfamily {opfowner} => pg_authid {oid}
--
2.25.1
[text/x-patch] 0002-Add-stats_form_tuple-20220310.patch (3.9K, ../../[email protected]/4-0002-Add-stats_form_tuple-20220310.patch)
download | inline diff:
From 7f4bd696f8162137762826478f450074df5b7939 Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Wed, 7 Dec 2016 16:12:55 +0300
Subject: [PATCH 2/6] Add stats_form_tuple()
---
src/backend/utils/adt/selfuncs.c | 55 ++++++++++++++++++++++++++++++++
src/include/utils/selfuncs.h | 22 +++++++++++++
2 files changed, 77 insertions(+)
diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c
index d1dd049f1ae..f534e862079 100644
--- a/src/backend/utils/adt/selfuncs.c
+++ b/src/backend/utils/adt/selfuncs.c
@@ -7954,3 +7954,58 @@ brincostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
*indexPages = index->pages;
}
+
+/*
+ * stats_form_tuple - Form pg_statistic tuple from StatsData.
+ *
+ * If 'data' parameter is NULL, form all-NULL tuple (nullfrac = 1.0).
+ */
+HeapTuple
+stats_form_tuple(StatsData *data)
+{
+ Relation rel;
+ HeapTuple tuple;
+ Datum values[Natts_pg_statistic];
+ bool nulls[Natts_pg_statistic];
+ int i;
+
+ for (i = 0; i < Natts_pg_statistic; ++i)
+ nulls[i] = false;
+
+ values[Anum_pg_statistic_starelid - 1] = ObjectIdGetDatum(InvalidOid);
+ values[Anum_pg_statistic_staattnum - 1] = Int16GetDatum(0);
+ values[Anum_pg_statistic_stainherit - 1] = BoolGetDatum(false);
+ values[Anum_pg_statistic_stanullfrac - 1] =
+ Float4GetDatum(data ? data->nullfrac : 1.0);
+ values[Anum_pg_statistic_stawidth - 1] =
+ Int32GetDatum(data ? data->width : 0);
+ values[Anum_pg_statistic_stadistinct - 1] =
+ Float4GetDatum(data ? data->distinct : 0);
+
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ StatsSlot *slot = data ? &data->slots[i] : NULL;
+
+ values[Anum_pg_statistic_stakind1 + i - 1] =
+ Int16GetDatum(slot ? slot->kind : 0);
+
+ values[Anum_pg_statistic_staop1 + i - 1] =
+ ObjectIdGetDatum(slot ? slot->opid : InvalidOid);
+
+ if (slot && DatumGetPointer(slot->numbers))
+ values[Anum_pg_statistic_stanumbers1 + i - 1] = slot->numbers;
+ else
+ nulls[Anum_pg_statistic_stanumbers1 + i - 1] = true;
+
+ if (slot && DatumGetPointer(slot->values))
+ values[Anum_pg_statistic_stavalues1 + i - 1] = slot->values;
+ else
+ nulls[Anum_pg_statistic_stavalues1 + i - 1] = true;
+ }
+
+ rel = table_open(StatisticRelationId, AccessShareLock);
+ tuple = heap_form_tuple(RelationGetDescr(rel), values, nulls);
+ table_close(rel, NoLock);
+
+ return tuple;
+}
diff --git a/src/include/utils/selfuncs.h b/src/include/utils/selfuncs.h
index 8f3d73edfb2..49fc3d717b1 100644
--- a/src/include/utils/selfuncs.h
+++ b/src/include/utils/selfuncs.h
@@ -16,6 +16,7 @@
#define SELFUNCS_H
#include "access/htup.h"
+#include "catalog/pg_statistic.h"
#include "fmgr.h"
#include "nodes/pathnodes.h"
@@ -133,6 +134,24 @@ typedef struct
double num_sa_scans; /* # indexscans from ScalarArrayOpExprs */
} GenericCosts;
+/* Single pg_statistic slot */
+typedef struct StatsSlot
+{
+ int16 kind; /* stakindN: statistic kind (STATISTIC_KIND_XXX) */
+ Oid opid; /* staopN: associated operator, if needed */
+ Datum numbers; /* stanumbersN: float4 array of numbers */
+ Datum values; /* stavaluesN: anyarray of values */
+} StatsSlot;
+
+/* Deformed pg_statistic tuple */
+typedef struct StatsData
+{
+ float4 nullfrac; /* stanullfrac: fraction of NULL values */
+ float4 distinct; /* stadistinct: number of distinct non-NULL values */
+ int32 width; /* stawidth: average width in bytes of non-NULL values */
+ StatsSlot slots[STATISTIC_NUM_SLOTS]; /* slots for different statistic types */
+} StatsData;
+
/* Hooks for plugins to get control when we ask for stats */
typedef bool (*get_relation_stats_hook_type) (PlannerInfo *root,
RangeTblEntry *rte,
@@ -231,6 +250,9 @@ extern void genericcostestimate(PlannerInfo *root, IndexPath *path,
double loop_count,
GenericCosts *costs);
+extern HeapTuple stats_form_tuple(StatsData *data);
+
+
/* Functions in array_selfuncs.c */
extern Selectivity scalararraysel_containment(PlannerInfo *root,
--
2.25.1
[text/x-patch] 0003-Add-symbolic-names-for-some-jsonb-operators-20220310.patch (7.1K, ../../[email protected]/5-0003-Add-symbolic-names-for-some-jsonb-operators-20220310.patch)
download | inline diff:
From 5fb7ee5ba041d8fc74a04cf40e9f3b1bd324610f Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Sat, 12 Nov 2016 18:59:43 +0300
Subject: [PATCH 3/6] Add symbolic names for some jsonb operators
---
src/include/catalog/pg_operator.dat | 45 +++++++++++++++++------------
1 file changed, 26 insertions(+), 19 deletions(-)
diff --git a/src/include/catalog/pg_operator.dat b/src/include/catalog/pg_operator.dat
index bc5f8213f3a..8e0e65ad275 100644
--- a/src/include/catalog/pg_operator.dat
+++ b/src/include/catalog/pg_operator.dat
@@ -2067,7 +2067,7 @@
{ oid => '1751', descr => 'negate',
oprname => '-', oprkind => 'l', oprleft => '0', oprright => 'numeric',
oprresult => 'numeric', oprcode => 'numeric_uminus' },
-{ oid => '1752', descr => 'equal',
+{ oid => '1752', oid_symbol => 'NumericEqOperator', descr => 'equal',
oprname => '=', oprcanmerge => 't', oprcanhash => 't', oprleft => 'numeric',
oprright => 'numeric', oprresult => 'bool', oprcom => '=(numeric,numeric)',
oprnegate => '<>(numeric,numeric)', oprcode => 'numeric_eq',
@@ -2077,7 +2077,7 @@
oprresult => 'bool', oprcom => '<>(numeric,numeric)',
oprnegate => '=(numeric,numeric)', oprcode => 'numeric_ne',
oprrest => 'neqsel', oprjoin => 'neqjoinsel' },
-{ oid => '1754', descr => 'less than',
+{ oid => '1754', oid_symbol => 'NumericLtOperator', descr => 'less than',
oprname => '<', oprleft => 'numeric', oprright => 'numeric',
oprresult => 'bool', oprcom => '>(numeric,numeric)',
oprnegate => '>=(numeric,numeric)', oprcode => 'numeric_lt',
@@ -3172,70 +3172,77 @@
{ oid => '3967', descr => 'get value from json as text with path elements',
oprname => '#>>', oprleft => 'json', oprright => '_text', oprresult => 'text',
oprcode => 'json_extract_path_text' },
-{ oid => '3211', descr => 'get jsonb object field',
+{ oid => '3211', oid_symbol => 'JsonbObjectFieldOperator',
+ descr => 'get jsonb object field',
oprname => '->', oprleft => 'jsonb', oprright => 'text', oprresult => 'jsonb',
oprcode => 'jsonb_object_field' },
-{ oid => '3477', descr => 'get jsonb object field as text',
+{ oid => '3477', oid_symbol => 'JsonbObjectFieldTextOperator',
+ descr => 'get jsonb object field as text',
oprname => '->>', oprleft => 'jsonb', oprright => 'text', oprresult => 'text',
oprcode => 'jsonb_object_field_text' },
-{ oid => '3212', descr => 'get jsonb array element',
+{ oid => '3212', oid_symbol => 'JsonbArrayElementOperator',
+ descr => 'get jsonb array element',
oprname => '->', oprleft => 'jsonb', oprright => 'int4', oprresult => 'jsonb',
oprcode => 'jsonb_array_element' },
-{ oid => '3481', descr => 'get jsonb array element as text',
+{ oid => '3481', oid_symbol => 'JsonbArrayElementTextOperator',
+ descr => 'get jsonb array element as text',
oprname => '->>', oprleft => 'jsonb', oprright => 'int4', oprresult => 'text',
oprcode => 'jsonb_array_element_text' },
-{ oid => '3213', descr => 'get value from jsonb with path elements',
+{ oid => '3213', oid_symbol => 'JsonbExtractPathOperator',
+ descr => 'get value from jsonb with path elements',
oprname => '#>', oprleft => 'jsonb', oprright => '_text',
oprresult => 'jsonb', oprcode => 'jsonb_extract_path' },
-{ oid => '3206', descr => 'get value from jsonb as text with path elements',
+{ oid => '3206', oid_symbol => 'JsonbExtractPathTextOperator',
+ descr => 'get value from jsonb as text with path elements',
oprname => '#>>', oprleft => 'jsonb', oprright => '_text',
oprresult => 'text', oprcode => 'jsonb_extract_path_text' },
-{ oid => '3240', descr => 'equal',
+{ oid => '3240', oid_symbol => 'JsonbEqOperator', descr => 'equal',
oprname => '=', oprcanmerge => 't', oprcanhash => 't', oprleft => 'jsonb',
oprright => 'jsonb', oprresult => 'bool', oprcom => '=(jsonb,jsonb)',
oprnegate => '<>(jsonb,jsonb)', oprcode => 'jsonb_eq', oprrest => 'eqsel',
oprjoin => 'eqjoinsel' },
-{ oid => '3241', descr => 'not equal',
+{ oid => '3241', oid_symbol => 'JsonbNeOperator', descr => 'not equal',
oprname => '<>', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '<>(jsonb,jsonb)', oprnegate => '=(jsonb,jsonb)',
oprcode => 'jsonb_ne', oprrest => 'neqsel', oprjoin => 'neqjoinsel' },
-{ oid => '3242', descr => 'less than',
+{ oid => '3242', oid_symbol => 'JsonbLtOperator', descr => 'less than',
oprname => '<', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '>(jsonb,jsonb)', oprnegate => '>=(jsonb,jsonb)',
oprcode => 'jsonb_lt', oprrest => 'scalarltsel',
oprjoin => 'scalarltjoinsel' },
-{ oid => '3243', descr => 'greater than',
+{ oid => '3243', oid_symbol => 'JsonbGtOperator', descr => 'greater than',
oprname => '>', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '<(jsonb,jsonb)', oprnegate => '<=(jsonb,jsonb)',
oprcode => 'jsonb_gt', oprrest => 'scalargtsel',
oprjoin => 'scalargtjoinsel' },
-{ oid => '3244', descr => 'less than or equal',
+{ oid => '3244', oid_symbol => 'JsonbLeOperator', descr => 'less than or equal',
oprname => '<=', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '>=(jsonb,jsonb)', oprnegate => '>(jsonb,jsonb)',
oprcode => 'jsonb_le', oprrest => 'scalarlesel',
oprjoin => 'scalarlejoinsel' },
-{ oid => '3245', descr => 'greater than or equal',
+{ oid => '3245', oid_symbol => 'JsonbGeOperator',
+ descr => 'greater than or equal',
oprname => '>=', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '<=(jsonb,jsonb)', oprnegate => '<(jsonb,jsonb)',
oprcode => 'jsonb_ge', oprrest => 'scalargesel',
oprjoin => 'scalargejoinsel' },
-{ oid => '3246', descr => 'contains',
+{ oid => '3246', oid_symbol => 'JsonbContainsOperator', descr => 'contains',
oprname => '@>', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '<@(jsonb,jsonb)', oprcode => 'jsonb_contains',
oprrest => 'matchingsel', oprjoin => 'matchingjoinsel' },
-{ oid => '3247', descr => 'key exists',
+{ oid => '3247', oid_symbol => 'JsonbExistsOperator', descr => 'key exists',
oprname => '?', oprleft => 'jsonb', oprright => 'text', oprresult => 'bool',
oprcode => 'jsonb_exists', oprrest => 'matchingsel',
oprjoin => 'matchingjoinsel' },
-{ oid => '3248', descr => 'any key exists',
+{ oid => '3248', oid_symbol => 'JsonbExistsAnyOperator', descr => 'any key exists',
oprname => '?|', oprleft => 'jsonb', oprright => '_text', oprresult => 'bool',
oprcode => 'jsonb_exists_any', oprrest => 'matchingsel',
oprjoin => 'matchingjoinsel' },
-{ oid => '3249', descr => 'all keys exist',
+{ oid => '3249', oid_symbol => 'JsonbExistsAllOperator', descr => 'all keys exist',
oprname => '?&', oprleft => 'jsonb', oprright => '_text', oprresult => 'bool',
oprcode => 'jsonb_exists_all', oprrest => 'matchingsel',
oprjoin => 'matchingjoinsel' },
-{ oid => '3250', descr => 'is contained by',
+{ oid => '3250', oid_symbol => 'JsonbContainedOperator', descr => 'is contained by',
oprname => '<@', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '@>(jsonb,jsonb)', oprcode => 'jsonb_contained',
oprrest => 'matchingsel', oprjoin => 'matchingjoinsel' },
--
2.25.1
[text/x-patch] 0004-Add-helper-jsonb-functions-and-macros-20220310.patch (9.3K, ../../[email protected]/6-0004-Add-helper-jsonb-functions-and-macros-20220310.patch)
download | inline diff:
From 6813b8ec7e8c45e04924d845c0f984d51ffc5f24 Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Fri, 16 Dec 2016 17:16:47 +0300
Subject: [PATCH 4/6] Add helper jsonb functions and macros
---
src/backend/utils/adt/jsonb_util.c | 27 +++++
src/backend/utils/adt/jsonfuncs.c | 10 +-
src/include/utils/jsonb.h | 165 ++++++++++++++++++++++++++++-
3 files changed, 195 insertions(+), 7 deletions(-)
diff --git a/src/backend/utils/adt/jsonb_util.c b/src/backend/utils/adt/jsonb_util.c
index 60442758b32..2715173fa37 100644
--- a/src/backend/utils/adt/jsonb_util.c
+++ b/src/backend/utils/adt/jsonb_util.c
@@ -389,6 +389,22 @@ findJsonbValueFromContainer(JsonbContainer *container, uint32 flags,
return NULL;
}
+/*
+ * findJsonbValueFromContainer() wrapper that sets up JsonbValue key string.
+ */
+JsonbValue *
+findJsonbValueFromContainerLen(JsonbContainer *container, uint32 flags,
+ const char *key, uint32 keylen)
+{
+ JsonbValue k;
+
+ k.type = jbvString;
+ k.val.string.val = key;
+ k.val.string.len = keylen;
+
+ return findJsonbValueFromContainer(container, flags, &k);
+}
+
/*
* Find value by key in Jsonb object and fetch it into 'res', which is also
* returned.
@@ -603,6 +619,17 @@ pushJsonbValue(JsonbParseState **pstate, JsonbIteratorToken seq,
return pushJsonbValueScalar(pstate, seq, jbval);
}
+ /*
+ * XXX I'm not quite sure why we actually do this? Why do we need to change
+ * how JsonbValue is converted to Jsonb for the statistics patch?
+ */
+ /* push value from scalar container without its enclosing array */
+ if (*pstate && JsonbExtractScalar(jbval->val.binary.data, &v))
+ {
+ Assert(IsAJsonbScalar(&v));
+ return pushJsonbValueScalar(pstate, seq, &v);
+ }
+
/* unpack the binary and add each piece to the pstate */
it = JsonbIteratorInit(jbval->val.binary.data);
diff --git a/src/backend/utils/adt/jsonfuncs.c b/src/backend/utils/adt/jsonfuncs.c
index 29664aa6e40..5e7a2f27483 100644
--- a/src/backend/utils/adt/jsonfuncs.c
+++ b/src/backend/utils/adt/jsonfuncs.c
@@ -5261,7 +5261,8 @@ iterate_jsonb_values(Jsonb *jb, uint32 flags, void *state,
if (type == WJB_KEY)
{
if (flags & jtiKey)
- action(state, v.val.string.val, v.val.string.len);
+ action(state, unconstify(char *, v.val.string.val),
+ v.val.string.len);
continue;
}
@@ -5276,7 +5277,8 @@ iterate_jsonb_values(Jsonb *jb, uint32 flags, void *state,
{
case jbvString:
if (flags & jtiString)
- action(state, v.val.string.val, v.val.string.len);
+ action(state, unconstify(char *, v.val.string.val),
+ v.val.string.len);
break;
case jbvNumeric:
if (flags & jtiNumeric)
@@ -5398,7 +5400,9 @@ transform_jsonb_string_values(Jsonb *jsonb, void *action_state,
{
if ((type == WJB_VALUE || type == WJB_ELEM) && v.type == jbvString)
{
- out = transform_action(action_state, v.val.string.val, v.val.string.len);
+ out = transform_action(action_state,
+ unconstify(char *, v.val.string.val),
+ v.val.string.len);
v.val.string.val = VARDATA_ANY(out);
v.val.string.len = VARSIZE_ANY_EXHDR(out);
res = pushJsonbValue(&st, type, type < WJB_BEGIN_ARRAY ? &v : NULL);
diff --git a/src/include/utils/jsonb.h b/src/include/utils/jsonb.h
index 4cbe6edf218..e1ed712f26c 100644
--- a/src/include/utils/jsonb.h
+++ b/src/include/utils/jsonb.h
@@ -14,6 +14,7 @@
#include "lib/stringinfo.h"
#include "utils/array.h"
+#include "utils/builtins.h"
#include "utils/numeric.h"
/* Tokens used when sequentially processing a jsonb value */
@@ -229,8 +230,7 @@ typedef struct
#define JB_ROOT_IS_OBJECT(jbp_) ((*(uint32 *) VARDATA(jbp_) & JB_FOBJECT) != 0)
#define JB_ROOT_IS_ARRAY(jbp_) ((*(uint32 *) VARDATA(jbp_) & JB_FARRAY) != 0)
-
-enum jbvType
+typedef enum jbvType
{
/* Scalar types */
jbvNull = 0x0,
@@ -250,7 +250,7 @@ enum jbvType
* into JSON strings when outputted to json/jsonb.
*/
jbvDatetime = 0x20,
-};
+} JsonbValueType;
/*
* JsonbValue: In-memory representation of Jsonb. This is a convenient
@@ -269,7 +269,7 @@ struct JsonbValue
struct
{
int len;
- char *val; /* Not necessarily null-terminated */
+ const char *val; /* Not necessarily null-terminated */
} string; /* String primitive type */
struct
@@ -382,6 +382,10 @@ extern int compareJsonbContainers(JsonbContainer *a, JsonbContainer *b);
extern JsonbValue *findJsonbValueFromContainer(JsonbContainer *sheader,
uint32 flags,
JsonbValue *key);
+extern JsonbValue *findJsonbValueFromContainerLen(JsonbContainer *container,
+ uint32 flags,
+ const char *key,
+ uint32 keylen);
extern JsonbValue *getKeyJsonValueFromContainer(JsonbContainer *container,
const char *keyVal, int keyLen,
JsonbValue *res);
@@ -412,4 +416,157 @@ extern Datum jsonb_set_element(Jsonb *jb, Datum *path, int path_len,
JsonbValue *newval);
extern Datum jsonb_get_element(Jsonb *jb, Datum *path, int npath,
bool *isnull, bool as_text);
+
+/*
+ * XXX Not sure we want to add these functions to jsonb.h, which is the
+ * public API. Maybe it belongs rather to jsonb_typanalyze.c or elsewhere,
+ * closer to how it's used?
+ */
+
+/* helper inline functions for JsonbValue initialization */
+static inline JsonbValue *
+JsonValueInitObject(JsonbValue *val, int nPairs, int nPairsAllocated)
+{
+ val->type = jbvObject;
+ val->val.object.nPairs = nPairs;
+ val->val.object.pairs = nPairsAllocated ?
+ palloc(sizeof(JsonbPair) * nPairsAllocated) : NULL;
+
+ return val;
+}
+
+static inline JsonbValue *
+JsonValueInitArray(JsonbValue *val, int nElems, int nElemsAllocated,
+ bool rawScalar)
+{
+ val->type = jbvArray;
+ val->val.array.nElems = nElems;
+ val->val.array.elems = nElemsAllocated ?
+ palloc(sizeof(JsonbValue) * nElemsAllocated) : NULL;
+ val->val.array.rawScalar = rawScalar;
+
+ return val;
+}
+
+static inline JsonbValue *
+JsonValueInitBinary(JsonbValue *val, Jsonb *jb)
+{
+ val->type = jbvBinary;
+ val->val.binary.data = &(jb)->root;
+ val->val.binary.len = VARSIZE_ANY_EXHDR(jb);
+ return val;
+}
+
+
+static inline JsonbValue *
+JsonValueInitString(JsonbValue *jbv, const char *str)
+{
+ jbv->type = jbvString;
+ jbv->val.string.len = strlen(str);
+ jbv->val.string.val = memcpy(palloc(jbv->val.string.len + 1), str,
+ jbv->val.string.len + 1);
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitStringWithLen(JsonbValue *jbv, const char *str, int len)
+{
+ jbv->type = jbvString;
+ jbv->val.string.val = str;
+ jbv->val.string.len = len;
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitText(JsonbValue *jbv, text *txt)
+{
+ jbv->type = jbvString;
+ jbv->val.string.val = VARDATA_ANY(txt);
+ jbv->val.string.len = VARSIZE_ANY_EXHDR(txt);
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitNumeric(JsonbValue *jbv, Numeric num)
+{
+ jbv->type = jbvNumeric;
+ jbv->val.numeric = num;
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitInteger(JsonbValue *jbv, int64 i)
+{
+ jbv->type = jbvNumeric;
+ jbv->val.numeric = DatumGetNumeric(DirectFunctionCall1(
+ int8_numeric, Int64GetDatum(i)));
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitFloat(JsonbValue *jbv, float4 f)
+{
+ jbv->type = jbvNumeric;
+ jbv->val.numeric = DatumGetNumeric(DirectFunctionCall1(
+ float4_numeric, Float4GetDatum(f)));
+ return jbv;
+}
+
+static inline JsonbValue *
+JsonValueInitDouble(JsonbValue *jbv, float8 f)
+{
+ jbv->type = jbvNumeric;
+ jbv->val.numeric = DatumGetNumeric(DirectFunctionCall1(
+ float8_numeric, Float8GetDatum(f)));
+ return jbv;
+}
+
+/* helper macros for jsonb building */
+#define pushJsonbKey(pstate, jbv, key) \
+ pushJsonbValue(pstate, WJB_KEY, JsonValueInitString(jbv, key))
+
+#define pushJsonbValueGeneric(Type, pstate, jbv, val) \
+ pushJsonbValue(pstate, WJB_VALUE, JsonValueInit##Type(jbv, val))
+
+#define pushJsonbElemGeneric(Type, pstate, jbv, val) \
+ pushJsonbValue(pstate, WJB_ELEM, JsonValueInit##Type(jbv, val))
+
+#define pushJsonbValueInteger(pstate, jbv, i) \
+ pushJsonbValueGeneric(Integer, pstate, jbv, i)
+
+#define pushJsonbValueFloat(pstate, jbv, f) \
+ pushJsonbValueGeneric(Float, pstate, jbv, f)
+
+#define pushJsonbElemFloat(pstate, jbv, f) \
+ pushJsonbElemGeneric(Float, pstate, jbv, f)
+
+#define pushJsonbElemString(pstate, jbv, txt) \
+ pushJsonbElemGeneric(String, pstate, jbv, txt)
+
+#define pushJsonbElemText(pstate, jbv, txt) \
+ pushJsonbElemGeneric(Text, pstate, jbv, txt)
+
+#define pushJsonbElemNumeric(pstate, jbv, num) \
+ pushJsonbElemGeneric(Numeric, pstate, jbv, num)
+
+#define pushJsonbElemInteger(pstate, jbv, num) \
+ pushJsonbElemGeneric(Integer, pstate, jbv, num)
+
+#define pushJsonbElemBinary(pstate, jbv, jbcont) \
+ pushJsonbElemGeneric(Binary, pstate, jbv, jbcont)
+
+#define pushJsonbKeyValueGeneric(Type, pstate, jbv, key, val) ( \
+ pushJsonbKey(pstate, jbv, key), \
+ pushJsonbValueGeneric(Type, pstate, jbv, val) \
+ )
+
+#define pushJsonbKeyValueString(pstate, jbv, key, val) \
+ pushJsonbKeyValueGeneric(String, pstate, jbv, key, val)
+
+#define pushJsonbKeyValueFloat(pstate, jbv, key, val) \
+ pushJsonbKeyValueGeneric(Float, pstate, jbv, key, val)
+
+#define pushJsonbKeyValueInteger(pstate, jbv, key, val) \
+ pushJsonbKeyValueGeneric(Integer, pstate, jbv, key, val)
+
#endif /* __JSONB_H__ */
--
2.25.1
[text/x-patch] 0005-Export-scalarineqsel-20220310.patch (1.5K, ../../[email protected]/7-0005-Export-scalarineqsel-20220310.patch)
download | inline diff:
From f74f8951d35eaa872a2add0bf73d340c5a14e3bc Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Sat, 12 Nov 2016 19:01:43 +0300
Subject: [PATCH 5/6] Export scalarineqsel()
---
src/backend/utils/adt/selfuncs.c | 2 +-
src/include/utils/selfuncs.h | 4 ++++
2 files changed, 5 insertions(+), 1 deletion(-)
diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c
index f534e862079..6ef31fdd79c 100644
--- a/src/backend/utils/adt/selfuncs.c
+++ b/src/backend/utils/adt/selfuncs.c
@@ -573,7 +573,7 @@ neqsel(PG_FUNCTION_ARGS)
* it will return an approximate estimate based on assuming that the constant
* value falls in the middle of the bin identified by binary search.
*/
-static double
+double
scalarineqsel(PlannerInfo *root, Oid operator, bool isgt, bool iseq,
Oid collation,
VariableStatData *vardata, Datum constval, Oid consttype)
diff --git a/src/include/utils/selfuncs.h b/src/include/utils/selfuncs.h
index 49fc3d717b1..88a95401c5f 100644
--- a/src/include/utils/selfuncs.h
+++ b/src/include/utils/selfuncs.h
@@ -249,6 +249,10 @@ extern List *add_predicate_to_index_quals(IndexOptInfo *index,
extern void genericcostestimate(PlannerInfo *root, IndexPath *path,
double loop_count,
GenericCosts *costs);
+extern double scalarineqsel(PlannerInfo *root, Oid operator, bool isgt,
+ bool iseq, Oid collation,
+ VariableStatData *vardata, Datum constval,
+ Oid consttype);
extern HeapTuple stats_form_tuple(StatsData *data);
--
2.25.1
[text/x-patch] 0006-Add-jsonb-statistics-20220310.patch (137.0K, ../../[email protected]/8-0006-Add-jsonb-statistics-20220310.patch)
download | inline diff:
From 768c3cc2e47a8a85098af4aad38c58156c00b374 Mon Sep 17 00:00:00 2001
From: Nikita Glukhov <[email protected]>
Date: Fri, 11 Mar 2022 01:06:50 +0300
Subject: [PATCH 6/6] Add jsonb statistics
---
src/backend/catalog/system_functions.sql | 36 +
src/backend/catalog/system_views.sql | 56 +
src/backend/utils/adt/Makefile | 2 +
src/backend/utils/adt/jsonb_selfuncs.c | 1582 ++++++++++++++++++++
src/backend/utils/adt/jsonb_typanalyze.c | 1627 +++++++++++++++++++++
src/backend/utils/adt/jsonpath_exec.c | 2 +-
src/include/catalog/pg_operator.dat | 17 +-
src/include/catalog/pg_proc.dat | 11 +
src/include/catalog/pg_statistic.h | 2 +
src/include/catalog/pg_type.dat | 2 +-
src/include/utils/json_selfuncs.h | 113 ++
src/test/regress/expected/jsonb_stats.out | 713 +++++++++
src/test/regress/expected/rules.out | 32 +
src/test/regress/parallel_schedule | 2 +-
src/test/regress/sql/jsonb_stats.sql | 249 ++++
15 files changed, 4435 insertions(+), 11 deletions(-)
create mode 100644 src/backend/utils/adt/jsonb_selfuncs.c
create mode 100644 src/backend/utils/adt/jsonb_typanalyze.c
create mode 100644 src/include/utils/json_selfuncs.h
create mode 100644 src/test/regress/expected/jsonb_stats.out
create mode 100644 src/test/regress/sql/jsonb_stats.sql
diff --git a/src/backend/catalog/system_functions.sql b/src/backend/catalog/system_functions.sql
index 81bac6f5812..0b9f68e88ff 100644
--- a/src/backend/catalog/system_functions.sql
+++ b/src/backend/catalog/system_functions.sql
@@ -594,6 +594,42 @@ LANGUAGE internal
STRICT IMMUTABLE PARALLEL SAFE
AS 'unicode_is_normalized';
+-- XXX is this function immutable / parallel safe?
+-- XXX do we actually need to cast to text and then to jsonb?
+CREATE FUNCTION pg_json_path_stats(tab regclass, path_index integer) RETURNS text
+AS $$
+ SELECT jsonb_pretty((
+ CASE
+ WHEN stakind1 = 8 THEN stavalues1
+ WHEN stakind2 = 8 THEN stavalues2
+ WHEN stakind3 = 8 THEN stavalues3
+ WHEN stakind4 = 8 THEN stavalues4
+ WHEN stakind5 = 8 THEN stavalues5
+ END::text::jsonb[])[$2])
+ FROM pg_statistic
+ WHERE starelid = $1
+$$ LANGUAGE 'sql';
+
+-- XXX is this function immutable / parallel safe?
+-- XXX do we actually need to cast to text and then to jsonb?
+CREATE FUNCTION pg_json_path_stats(tab regclass, path text) RETURNS text
+AS $$
+ SELECT jsonb_pretty(pathstats)
+ FROM (
+ SELECT unnest(
+ CASE
+ WHEN stakind1 = 8 THEN stavalues1
+ WHEN stakind2 = 8 THEN stavalues2
+ WHEN stakind3 = 8 THEN stavalues3
+ WHEN stakind4 = 8 THEN stavalues4
+ WHEN stakind5 = 8 THEN stavalues5
+ END::text::jsonb[]) pathstats
+ FROM pg_statistic
+ WHERE starelid = $1
+ ) paths
+ WHERE pathstats->>'path' = $2
+$$ LANGUAGE 'sql';
+
--
-- The default permissions for functions mean that anyone can execute them.
-- A number of functions shouldn't be executable by just anyone, but rather
diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sql
index 40b7bca5a96..9d32eeb4a3a 100644
--- a/src/backend/catalog/system_views.sql
+++ b/src/backend/catalog/system_views.sql
@@ -364,6 +364,62 @@ CREATE VIEW pg_stats_ext_exprs WITH (security_barrier) AS
-- unprivileged users may read pg_statistic_ext but not pg_statistic_ext_data
REVOKE ALL ON pg_statistic_ext_data FROM public;
+-- XXX This probably needs to do the same checks as pg_stats, i.e.
+-- WHERE NOT attisdropped
+-- AND has_column_privilege(c.oid, a.attnum, 'select')
+-- AND (c.relrowsecurity = false OR NOT row_security_active(c.oid));
+CREATE VIEW pg_stats_json AS
+ SELECT
+ nspname AS schemaname,
+ relname AS tablename,
+ attname AS attname,
+
+ path->>'path' AS json_path,
+
+ stainherit AS inherited,
+
+ (path->'json'->>'nullfrac')::float4 AS null_frac,
+ (path->'json'->>'width')::float4 AS avg_width,
+ (path->'json'->>'distinct')::float4 AS n_distinct,
+
+ ARRAY(SELECT val FROM jsonb_array_elements(
+ path->'json'->'mcv'->'values') val)::anyarray
+ AS most_common_vals,
+
+ ARRAY(SELECT num::text::float4 FROM jsonb_array_elements(
+ path->'json'->'mcv'->'numbers') num)
+ AS most_common_freqs,
+
+ ARRAY(SELECT val FROM jsonb_array_elements(
+ path->'json'->'histogram'->'values') val)
+ AS histogram_bounds,
+
+ ARRAY(SELECT val::text::int FROM jsonb_array_elements(
+ path->'array_length'->'mcv'->'values') val)
+ AS most_common_array_lengths,
+
+ ARRAY(SELECT num::text::float4 FROM jsonb_array_elements(
+ path->'array_length'->'mcv'->'numbers') num)
+ AS most_common_array_length_freqs,
+
+ (path->'json'->>'correlation')::float4 AS correlation
+
+ FROM
+ pg_statistic s JOIN pg_class c ON (c.oid = s.starelid)
+ JOIN pg_attribute a ON (c.oid = attrelid AND attnum = s.staattnum)
+ LEFT JOIN pg_namespace n ON (n.oid = c.relnamespace),
+ LATERAL (
+ SELECT unnest((CASE
+ WHEN stakind1 = 8 THEN stavalues1
+ WHEN stakind2 = 8 THEN stavalues2
+ WHEN stakind3 = 8 THEN stavalues3
+ WHEN stakind4 = 8 THEN stavalues4
+ WHEN stakind5 = 8 THEN stavalues5
+ END ::text::jsonb[])[2:]) AS path
+ ) paths;
+
+-- no need to revoke any privileges, we've already revoked accss to pg_statistic
+
CREATE VIEW pg_publication_tables AS
SELECT
P.pubname AS pubname,
diff --git a/src/backend/utils/adt/Makefile b/src/backend/utils/adt/Makefile
index 41b486bceff..5e359ccf4fb 100644
--- a/src/backend/utils/adt/Makefile
+++ b/src/backend/utils/adt/Makefile
@@ -50,6 +50,8 @@ OBJS = \
jsonb.o \
jsonb_gin.o \
jsonb_op.o \
+ jsonb_selfuncs.o \
+ jsonb_typanalyze.o \
jsonb_util.o \
jsonfuncs.o \
jsonbsubs.o \
diff --git a/src/backend/utils/adt/jsonb_selfuncs.c b/src/backend/utils/adt/jsonb_selfuncs.c
new file mode 100644
index 00000000000..f5520f88a1d
--- /dev/null
+++ b/src/backend/utils/adt/jsonb_selfuncs.c
@@ -0,0 +1,1582 @@
+/*-------------------------------------------------------------------------
+ *
+ * jsonb_selfuncs.c
+ * Functions for selectivity estimation of jsonb operators
+ *
+ * Copyright (c) 2016-2022, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ * src/backend/utils/adt/jsonb_selfuncs.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#include "postgres.h"
+
+#include <math.h>
+
+#include "fmgr.h"
+#include "access/heapam.h"
+#include "access/htup_details.h"
+#include "catalog/pg_operator.h"
+#include "catalog/pg_statistic.h"
+#include "catalog/pg_type.h"
+#include "common/string.h"
+#include "nodes/primnodes.h"
+#include "utils/builtins.h"
+#include "utils/json.h"
+#include "utils/jsonb.h"
+#include "utils/json_selfuncs.h"
+#include "utils/lsyscache.h"
+#include "utils/rel.h"
+#include "utils/selfuncs.h"
+
+#define DEFAULT_JSON_CONTAINS_SEL 0.001
+
+/*
+ * jsonGetField
+ * Given a JSONB document and a key, extract the JSONB value for the key.
+ */
+static inline Datum
+jsonGetField(Datum obj, const char *field)
+{
+ Jsonb *jb = DatumGetJsonbP(obj);
+ JsonbValue *jbv = findJsonbValueFromContainerLen(&jb->root, JB_FOBJECT,
+ field, strlen(field));
+ return jbv ? JsonbPGetDatum(JsonbValueToJsonb(jbv)) : PointerGetDatum(NULL);
+}
+
+/*
+ * jsonGetFloat4
+ * Given a JSONB value, interpret it as a float4 value.
+ *
+ * This expects the JSONB value to be a numeric, because that's how we store
+ * floats in JSONB, and we cast it to float4.
+ */
+static inline float4
+jsonGetFloat4(Datum jsonb, float4 default_val)
+{
+ Jsonb *jb;
+ JsonbValue jv;
+
+ if (!DatumGetPointer(jsonb))
+ return default_val;
+
+ jb = DatumGetJsonbP(jsonb);
+
+ if (!JsonbExtractScalar(&jb->root, &jv) || jv.type != jbvNumeric)
+ return default_val;
+
+ return DatumGetFloat4(DirectFunctionCall1(numeric_float4,
+ NumericGetDatum(jv.val.numeric)));
+}
+
+/*
+ * jsonStatsInit
+ * Given a pg_statistic tuple, expand STATISTIC_KIND_JSON into JsonStats.
+ */
+bool
+jsonStatsInit(JsonStats data, const VariableStatData *vardata)
+{
+ Jsonb *jb;
+ JsonbValue prefix;
+
+ if (!vardata->statsTuple)
+ return false;
+
+ data->statsTuple = vardata->statsTuple;
+ memset(&data->attslot, 0, sizeof(data->attslot));
+
+ /* Were there just NULL values in the column? No JSON stats, but still useful. */
+ if (((Form_pg_statistic) GETSTRUCT(data->statsTuple))->stanullfrac >= 1.0)
+ {
+ data->nullfrac = 1.0;
+ return true;
+ }
+
+ /* Do we have the JSON stats built in the pg_statistic? */
+ if (!get_attstatsslot(&data->attslot, data->statsTuple,
+ STATISTIC_KIND_JSON, InvalidOid,
+ ATTSTATSSLOT_NUMBERS | ATTSTATSSLOT_VALUES))
+ return false;
+
+ /*
+ * Valid JSON stats should have at least 2 elements in values:
+ * 0th - root path prefix
+ * 1st - root path stats
+ */
+ if (data->attslot.nvalues < 2)
+ {
+ free_attstatsslot(&data->attslot);
+ return false;
+ }
+
+ /* XXX If the ACL check was not OK, would we even get here? */
+ data->acl_ok = vardata->acl_ok;
+ data->rel = vardata->rel;
+ data->nullfrac =
+ data->attslot.nnumbers > 0 ? data->attslot.numbers[0] : 0.0;
+ data->pathdatums = data->attslot.values + 1;
+ data->npaths = data->attslot.nvalues - 1;
+
+ /* Extract root path prefix */
+ jb = DatumGetJsonbP(data->attslot.values[0]);
+ if (!JsonbExtractScalar(&jb->root, &prefix) || prefix.type != jbvString)
+ {
+ free_attstatsslot(&data->attslot);
+ return false;
+ }
+
+ data->prefix = prefix.val.string.val;
+ data->prefixlen = prefix.val.string.len;
+
+ /* Create path cache, initialze only two fields that acting as flags */
+ data->paths = palloc(sizeof(*data->paths) * data->npaths);
+
+ for (int i = 0; i < data->npaths; i++)
+ {
+ data->paths[i].data = NULL;
+ data->paths[i].path = NULL;
+ }
+
+ return true;
+}
+
+/*
+ * jsonStatsRelease
+ * Release resources (statistics slot) associated with the JsonStats value.
+ */
+void
+jsonStatsRelease(JsonStats data)
+{
+ free_attstatsslot(&data->attslot);
+}
+
+/*
+ * jsonPathStatsAllocSpecialStats
+ * Allocate a copy of JsonPathStats for accessing special (length etc.)
+ * stats for a given JSON path.
+ */
+static JsonPathStats
+jsonPathStatsAllocSpecialStats(JsonPathStats pstats, JsonPathStatsType type)
+{
+ JsonPathStats stats;
+
+ if (!pstats)
+ return NULL;
+
+ /* copy and replace stats type */
+ stats = palloc(sizeof(*stats));
+ *stats = *pstats;
+ stats->type = type;
+
+ return stats;
+}
+
+/*
+ * jsonPathStatsGetArrayLengthStats
+ * Extract statistics of array lengths for the path.
+ */
+JsonPathStats
+jsonPathStatsGetArrayLengthStats(JsonPathStats pstats)
+{
+ /*
+ * The array length statistics is relevant only for values that are arrays.
+ * So if we observed no such values, we know there can't be such
+ * statistics and so we simply return NULL.
+ */
+ if (jsonPathStatsGetTypeFreq(pstats, jbvArray, 0.0) <= 0.0)
+ return NULL;
+
+ return jsonPathStatsAllocSpecialStats(pstats, JsonPathStatsArrayLength);
+}
+
+/*
+ * jsonPathStatsGetObjectLengthStats
+ * Extract statistics of object length for the path.
+ */
+JsonPathStats
+jsonPathStatsGetObjectLengthStats(JsonPathStats pstats)
+{
+ /*
+ * The object length statistics is relevant only for values that are arrays.
+ * So if we observed no such values, we know there can't be such
+ * statistics and so we simply return NULL.
+ */
+ if (jsonPathStatsGetTypeFreq(pstats, jbvObject, 0.0) <= 0.0)
+ return NULL;
+
+ return jsonPathStatsAllocSpecialStats(pstats, JsonPathStatsObjectLength);
+}
+
+/*
+ * jsonPathStatsGetPath
+ * Try to use cached path name or extract it from per-path stats datum.
+ *
+ * Returns true on succces, false on error.
+ */
+static inline bool
+jsonPathStatsGetPath(JsonPathStats stats, Datum pathdatum,
+ const char **path, int *pathlen)
+{
+ *path = stats->path;
+
+ if (*path)
+ /* use cached path name */
+ *pathlen = stats->pathlen;
+ else
+ {
+ Jsonb *jsonb = DatumGetJsonbP(pathdatum);
+ JsonbValue pathkey;
+ JsonbValue *pathval;
+
+ /* extract path from the statistics represented as jsonb document */
+ JsonValueInitStringWithLen(&pathkey, "path", 4);
+ pathval = findJsonbValueFromContainer(&jsonb->root, JB_FOBJECT, &pathkey);
+
+ if (!pathval || pathval->type != jbvString)
+ return false; /* XXX invalid stats data, maybe throw error */
+
+ /* cache extracted path name */
+ *path = stats->path = pathval->val.string.val;
+ *pathlen = stats->pathlen = pathval->val.string.len;
+ }
+
+ return true;
+}
+
+/* Context for bsearch()ing paths */
+typedef struct JsonPathStatsSearchContext
+{
+ JsonStats stats;
+ const char *path;
+ int pathlen;
+} JsonPathStatsSearchContext;
+
+/*
+ * jsonPathStatsCompare
+ * Compare two JsonPathStats structs, so that we can sort them.
+ *
+ * We do this so that we can search for stats for a given path simply by
+ * bsearch().
+ *
+ * XXX We never build two structs for the same path, so we know the paths
+ * are different - one may be a prefix of the other, but then we sort the
+ * strings by length.
+ */
+static int
+jsonPathStatsCompare(const void *pv1, const void *pv2)
+{
+ JsonPathStatsSearchContext const *cxt = pv1;
+ Datum const *pathdatum = (Datum const *) pv2;
+ int index = pathdatum - cxt->stats->pathdatums;
+ JsonPathStats stats = &cxt->stats->paths[index];
+ const char *path;
+ int pathlen;
+ int res;
+
+ if (!jsonPathStatsGetPath(stats, *pathdatum, &path, &pathlen))
+ return 1; /* XXX invalid stats data */
+
+ /* compare the shared part first, then compare by length */
+ res = strncmp(cxt->path, path, Min(cxt->pathlen, pathlen));
+
+ return res ? res : cxt->pathlen - pathlen;
+}
+
+/*
+ * jsonStatsFindPath
+ * Find stats for a given path.
+ *
+ * The stats are sorted by path, so we can simply do bsearch().
+ * This is low-level function and jsdata->prefix is not considered, the caller
+ * should handle it by itself.
+ */
+static JsonPathStats
+jsonStatsFindPath(JsonStats jsdata, const char *path, int pathlen)
+{
+ JsonPathStatsSearchContext cxt;
+ JsonPathStats stats;
+ Datum *pdatum;
+ int index;
+
+ cxt.stats = jsdata;
+ cxt.path = path;
+ cxt.pathlen = pathlen;
+
+ pdatum = bsearch(&cxt, jsdata->pathdatums, jsdata->npaths,
+ sizeof(*jsdata->pathdatums), jsonPathStatsCompare);
+
+ if (!pdatum)
+ return NULL;
+
+ index = pdatum - jsdata->pathdatums;
+ stats = &jsdata->paths[index];
+
+ Assert(stats->path);
+ Assert(stats->pathlen == pathlen);
+
+ /* Init all fields if needed (stats->data == NULL means uninitialized) */
+ if (!stats->data)
+ {
+ stats->data = jsdata;
+ stats->datum = pdatum;
+ stats->type = JsonPathStatsValues;
+ }
+
+ return stats;
+}
+
+/*
+ * jsonStatsGetPathByStr
+ * Find stats for a given path string considering jsdata->prefix.
+ */
+JsonPathStats
+jsonStatsGetPathByStr(JsonStats jsdata, const char *subpath, int subpathlen)
+{
+ JsonPathStats stats;
+ char *path;
+ int pathlen;
+
+ if (jsdata->nullfrac >= 1.0)
+ return NULL;
+
+ pathlen = jsdata->prefixlen + subpathlen - 1;
+ path = palloc(pathlen);
+
+ memcpy(path, jsdata->prefix, jsdata->prefixlen);
+ memcpy(&path[jsdata->prefixlen], &subpath[1], subpathlen - 1);
+
+ stats = jsonStatsFindPath(jsdata, path, pathlen);
+
+ if (!stats)
+ pfree(path);
+
+ return stats;
+}
+
+/*
+ * jsonStatsGetRootPath
+ * Find JSON stats for root prefix path.
+ */
+static JsonPathStats
+jsonStatsGetRootPath(JsonStats jsdata)
+{
+ if (jsdata->nullfrac >= 1.0)
+ return NULL;
+
+ return jsonStatsFindPath(jsdata, jsdata->prefix, jsdata->prefixlen);
+}
+
+#define jsonStatsGetRootArrayPath(jsdata) \
+ jsonStatsGetPathByStr(jsdata, JSON_PATH_ROOT_ARRAY, JSON_PATH_ROOT_ARRAY_LEN)
+
+/*
+ * jsonPathAppendEntry
+ * Append entry (represented as simple string) to a path.
+ *
+ * NULL entry is treated as wildcard array accessor "[*]".
+ */
+void
+jsonPathAppendEntry(StringInfo path, const char *entry)
+{
+ if (entry)
+ {
+ appendStringInfoCharMacro(path, '.');
+ escape_json(path, entry);
+ }
+ else
+ appendStringInfoString(path, "[*]");
+}
+
+/*
+ * jsonPathAppendEntryWithLen
+ * Append string (represented as string + length) to a path.
+ */
+static void
+jsonPathAppendEntryWithLen(StringInfo path, const char *entry, int len)
+{
+ char *tmpentry = pnstrdup(entry, len);
+ jsonPathAppendEntry(path, tmpentry);
+ pfree(tmpentry);
+}
+
+/*
+ * jsonPathStatsGetSubpath
+ * Find JSON path stats for object key or array elements (if 'key' = NULL).
+ */
+JsonPathStats
+jsonPathStatsGetSubpath(JsonPathStats pstats, const char *key)
+{
+ JsonPathStats spstats;
+ StringInfoData str;
+
+ initStringInfo(&str);
+ appendBinaryStringInfo(&str, pstats->path, pstats->pathlen);
+ jsonPathAppendEntry(&str, key);
+
+ spstats = jsonStatsFindPath(pstats->data, str.data, str.len);
+ if (!spstats)
+ pfree(str.data);
+
+ return spstats;
+}
+
+/*
+ * jsonPathStatsGetArrayIndexSelectivity
+ * Given stats for a path, determine selectivity for an array index.
+ */
+Selectivity
+jsonPathStatsGetArrayIndexSelectivity(JsonPathStats pstats, int index)
+{
+ JsonPathStats lenstats = jsonPathStatsGetArrayLengthStats(pstats);
+ JsonbValue tmpjbv;
+ Jsonb *jb;
+
+ /*
+ * If we have no array length stats, assume all documents match.
+ *
+ * XXX Shouldn't this use a default smaller than 1.0? What do the selfuncs
+ * for regular arrays use?
+ */
+ if (!lenstats)
+ return 1.0;
+
+ jb = JsonbValueToJsonb(JsonValueInitInteger(&tmpjbv, index));
+
+ /* calculate fraction of elements smaller than the index */
+ return jsonSelectivity(lenstats, JsonbPGetDatum(jb), JsonbGtOperator);
+}
+
+/*
+ * jsonStatsGetPath
+ * Find JSON statistics for a given path.
+ *
+ * 'path' is an array of text datums of length 'pathlen' (can be zero).
+ */
+static JsonPathStats
+jsonStatsGetPath(JsonStats jsdata, Datum *path, int pathlen,
+ bool try_arrays_indexes, float4 *nullfrac)
+{
+ JsonPathStats pstats = jsonStatsGetRootPath(jsdata);
+ Selectivity sel = 1.0;
+
+ for (int i = 0; pstats && i < pathlen; i++)
+ {
+ char *key = TextDatumGetCString(path[i]);
+ char *tail;
+ int index;
+
+ if (!try_arrays_indexes)
+ {
+ /* Find object key stats */
+ pstats = jsonPathStatsGetSubpath(pstats, key);
+ pfree(key);
+ continue;
+ }
+
+ /* Try to interpret path entry as integer array index */
+ errno = 0;
+ index = strtoint(key, &tail, 10);
+
+ if (tail == key || *tail != '\0' || errno != 0)
+ {
+ /* Find object key stats */
+ pstats = jsonPathStatsGetSubpath(pstats, key);
+ }
+ else
+ {
+ /* Find array index stats */
+ /* FIXME consider object key "index" also */
+ JsonPathStats arrstats = jsonPathStatsGetSubpath(pstats, NULL);
+
+ if (arrstats)
+ {
+ float4 arrfreq = jsonPathStatsGetFreq(pstats, 0.0);
+
+ sel *= jsonPathStatsGetArrayIndexSelectivity(pstats, index);
+
+ if (arrfreq > 0.0)
+ sel /= arrfreq;
+ }
+
+ pstats = arrstats;
+ }
+
+ pfree(key);
+ }
+
+ *nullfrac = 1.0 - sel;
+
+ return pstats;
+}
+
+/*
+ * jsonPathStatsGetNextSubpathStats
+ * Iterate all collected subpaths of a given path.
+ *
+ * This function can be useful for estimation of selectivity of jsonpath
+ * '.*' and '.**' operators.
+ *
+ * The next found subpath is written into *pkeystats, which should be set to
+ * NULL before the first call.
+ *
+ * If keysOnly is true, emit only top-level object-key subpaths.
+ *
+ * Returns false on the end of iteration and true otherwise.
+ */
+bool
+jsonPathStatsGetNextSubpathStats(JsonPathStats stats, JsonPathStats *pkeystats,
+ bool keysOnly)
+{
+ JsonPathStats keystats = *pkeystats;
+ /* compute next index */
+ int index =
+ (keystats ? keystats->datum : stats->datum) - stats->data->pathdatums + 1;
+
+ if (stats->type != JsonPathStatsValues)
+ return false; /* length stats doe not have subpaths */
+
+ for (; index < stats->data->npaths; index++)
+ {
+ Datum *pathdatum = &stats->data->pathdatums[index];
+ const char *path;
+ int pathlen;
+
+ keystats = &stats->data->paths[index];
+
+ if (!jsonPathStatsGetPath(keystats, *pathdatum, &path, &pathlen))
+ break; /* invalid path stats */
+
+ /* Break, if subpath does not start from a desired prefix */
+ if (pathlen <= stats->pathlen ||
+ memcmp(path, stats->path, stats->pathlen))
+ break;
+
+ if (keysOnly)
+ {
+ const char *c = &path[stats->pathlen];
+
+ if (*c == '[')
+ {
+ Assert(c[1] == '*' && c[2] == ']');
+
+#if 0 /* TODO add separate flag for requesting top-level array accessors */
+ /* skip if it is not last key in the path */
+ if (pathlen > stats->pathlen + 3)
+#endif
+ continue; /* skip array accessors */
+ }
+ else if (*c == '.')
+ {
+ /* find end of '."key"' */
+ const char *pathend = path + pathlen - 1;
+
+ if (++c >= pathend || *c != '"')
+ break; /* invalid path */
+
+ while (++c <= pathend && *c != '"')
+ if (*c == '\\') /* handle escaped chars */
+ c++;
+
+ if (c > pathend)
+ break; /* invalid path */
+
+ /* skip if it is not last key in the path */
+ if (c < pathend)
+ continue;
+ }
+ else
+ continue; /* invalid path */
+ }
+
+ /* Init path stats if needed */
+ if (!keystats->data)
+ {
+ keystats->data = stats->data;
+ keystats->datum = pathdatum;
+ keystats->type = JsonPathStatsValues;
+ }
+
+ *pkeystats = keystats;
+
+ return true;
+ }
+
+ return false;
+}
+
+/*
+ * jsonStatsConvertArray
+ * Convert a JSONB array into an array of some regular data type.
+ *
+ * The "type" identifies what elements are in the input JSONB array, while
+ * typid determines the target type.
+ */
+static Datum
+jsonStatsConvertArray(Datum jsonbValueArray, JsonStatType type, Oid typid,
+ float4 multiplier)
+{
+ Datum *values;
+ Jsonb *jbvals;
+ JsonbValue jbv;
+ JsonbIterator *it;
+ JsonbIteratorToken r;
+ int nvalues;
+ int i;
+ int16 typlen;
+ bool typbyval;
+ char typalign;
+
+ if (!DatumGetPointer(jsonbValueArray))
+ return PointerGetDatum(NULL);
+
+ jbvals = DatumGetJsonbP(jsonbValueArray);
+
+ nvalues = JsonContainerSize(&jbvals->root);
+
+ values = palloc(sizeof(Datum) * nvalues);
+
+ for (i = 0, it = JsonbIteratorInit(&jbvals->root);
+ (r = JsonbIteratorNext(&it, &jbv, true)) != WJB_DONE;)
+ {
+ if (r == WJB_ELEM)
+ {
+ Datum value;
+
+ switch (type)
+ {
+ case JsonStatJsonb:
+ case JsonStatJsonbWithoutSubpaths:
+ value = JsonbPGetDatum(JsonbValueToJsonb(&jbv));
+ break;
+
+ case JsonStatText:
+ case JsonStatString:
+ Assert(jbv.type == jbvString);
+ value = PointerGetDatum(
+ cstring_to_text_with_len(jbv.val.string.val,
+ jbv.val.string.len));
+ break;
+
+ case JsonStatNumeric:
+ Assert(jbv.type == jbvNumeric);
+ value = NumericGetDatum(jbv.val.numeric);
+ break;
+
+ case JsonStatFloat4:
+ Assert(jbv.type == jbvNumeric);
+ value = DirectFunctionCall1(numeric_float4,
+ NumericGetDatum(jbv.val.numeric));
+ value = Float4GetDatum(DatumGetFloat4(value) * multiplier);
+ break;
+
+ default:
+ elog(ERROR, "invalid json stat type %d", type);
+ value = (Datum) 0;
+ break;
+ }
+
+ Assert(i < nvalues);
+ values[i++] = value;
+ }
+ }
+
+ Assert(i == nvalues);
+
+ get_typlenbyvalalign(typid, &typlen, &typbyval, &typalign);
+
+ return PointerGetDatum(
+ construct_array(values, nvalues, typid, typlen, typbyval, typalign));
+}
+
+/*
+ * jsonPathStatsExtractData
+ * Extract pg_statistics values from statistics for a single path.
+ *
+ * Extract ordinary MCV, Histogram, Correlation slots for a requested stats
+ * type. If requested stats for JSONB, include also transformed JSON slot for
+ * a path and possibly for its subpaths.
+ */
+static bool
+jsonPathStatsExtractData(JsonPathStats pstats, JsonStatType stattype,
+ float4 nullfrac, StatsData *statdata)
+{
+ Datum data;
+ Datum nullf;
+ Datum dist;
+ Datum width;
+ Datum mcv;
+ Datum hst;
+ Datum corr;
+ Oid type;
+ Oid eqop;
+ Oid ltop;
+ const char *key;
+ StatsSlot *slot = statdata->slots;
+
+ nullfrac = 1.0 - (1.0 - pstats->data->nullfrac) * (1.0 - nullfrac);
+
+ /*
+ * Depending on requested statistics type, select:
+ * - stavalues data type
+ * - corresponding eq/lt operators
+ * - JSONB field, containing stats slots for this statistics type
+ */
+ switch (stattype)
+ {
+ case JsonStatJsonb:
+ case JsonStatJsonbWithoutSubpaths:
+ key = pstats->type == JsonPathStatsArrayLength ? "array_length" :
+ pstats->type == JsonPathStatsObjectLength ? "object_length" :
+ "json";
+ type = JSONBOID;
+ eqop = JsonbEqOperator;
+ ltop = JsonbLtOperator;
+ break;
+ case JsonStatText:
+ key = "text";
+ type = TEXTOID;
+ eqop = TextEqualOperator;
+ ltop = TextLessOperator;
+ break;
+ case JsonStatString:
+ key = "string";
+ type = TEXTOID;
+ eqop = TextEqualOperator;
+ ltop = TextLessOperator;
+ break;
+ case JsonStatNumeric:
+ key = "numeric";
+ type = NUMERICOID;
+ eqop = NumericEqOperator;
+ ltop = NumericLtOperator;
+ break;
+ case JsonStatFloat4: /* special internal stats type */
+ default:
+ elog(ERROR, "invalid json statistic type %d", stattype);
+ break;
+ }
+
+ /* Extract object containing slots */
+ data = jsonGetField(*pstats->datum, key);
+
+ if (!DatumGetPointer(data))
+ return false;
+
+ nullf = jsonGetField(data, "nullfrac");
+ dist = jsonGetField(data, "distinct");
+ width = jsonGetField(data, "width");
+ mcv = jsonGetField(data, "mcv");
+ hst = jsonGetField(data, "histogram");
+ corr = jsonGetField(data, "correlation");
+
+ statdata->nullfrac = jsonGetFloat4(nullf, 0);
+ statdata->distinct = jsonGetFloat4(dist, 0);
+ statdata->width = (int32) jsonGetFloat4(width, 0);
+
+ statdata->nullfrac += (1.0 - statdata->nullfrac) * nullfrac;
+
+ /* Include MCV slot if exists */
+ if (DatumGetPointer(mcv))
+ {
+ slot->kind = STATISTIC_KIND_MCV;
+ slot->opid = eqop;
+ slot->numbers = jsonStatsConvertArray(jsonGetField(mcv, "numbers"),
+ JsonStatFloat4, FLOAT4OID,
+ 1.0 - nullfrac);
+ slot->values = jsonStatsConvertArray(jsonGetField(mcv, "values"),
+ stattype, type, 0);
+ slot++;
+ }
+
+ /* Include Histogram slot if exists */
+ if (DatumGetPointer(hst))
+ {
+ slot->kind = STATISTIC_KIND_HISTOGRAM;
+ slot->opid = ltop;
+ slot->numbers = jsonStatsConvertArray(jsonGetField(hst, "numbers"),
+ JsonStatFloat4, FLOAT4OID, 1.0);
+ slot->values = jsonStatsConvertArray(jsonGetField(hst, "values"),
+ stattype, type, 0);
+ slot++;
+ }
+
+ /* Include Correlation slot if exists */
+ if (DatumGetPointer(corr))
+ {
+ Datum correlation = Float4GetDatum(jsonGetFloat4(corr, 0));
+
+ slot->kind = STATISTIC_KIND_CORRELATION;
+ slot->opid = ltop;
+ slot->numbers = PointerGetDatum(construct_array(&correlation, 1,
+ FLOAT4OID, 4, true,
+ 'i'));
+ slot++;
+ }
+
+ /* Include JSON statistics for a given path and possibly for its subpaths */
+ if ((stattype == JsonStatJsonb ||
+ stattype == JsonStatJsonbWithoutSubpaths) &&
+ jsonAnalyzeBuildSubPathsData(pstats->data->pathdatums,
+ pstats->data->npaths,
+ pstats->datum - pstats->data->pathdatums,
+ pstats->path,
+ pstats->pathlen,
+ stattype == JsonStatJsonb,
+ nullfrac,
+ &slot->values,
+ &slot->numbers))
+ {
+ slot->kind = STATISTIC_KIND_JSON;
+ slot++;
+ }
+
+ return true;
+}
+
+static float4
+jsonPathStatsGetFloat(JsonPathStats pstats, const char *key, float4 defaultval)
+{
+ if (!pstats)
+ return defaultval;
+
+ return jsonGetFloat4(jsonGetField(*pstats->datum, key), defaultval);
+}
+
+float4
+jsonPathStatsGetFreq(JsonPathStats pstats, float4 defaultfreq)
+{
+ return jsonPathStatsGetFloat(pstats, "freq", defaultfreq);
+}
+
+float4
+jsonPathStatsGetAvgArraySize(JsonPathStats pstats)
+{
+ return jsonPathStatsGetFloat(pstats, "avg_array_length", 1.0);
+}
+
+/*
+ * jsonPathStatsGetTypeFreq
+ * Get frequency of different JSON object types for a given path.
+ *
+ * JSON documents don't have any particular schema, and the same path may point
+ * to values with different types in multiple documents. Consider for example
+ * two documents {"a" : "b"} and {"a" : 100} which have both a string and int
+ * for the same path. So we track the frequency of different JSON types for
+ * each path, so that we can consider this later.
+ */
+float4
+jsonPathStatsGetTypeFreq(JsonPathStats pstats, JsonbValueType type,
+ float4 defaultfreq)
+{
+ const char *key;
+
+ if (!pstats)
+ return defaultfreq;
+
+ /*
+ * When dealing with (object/array) length stats, we only really care about
+ * objects and arrays.
+ *
+ * Lengths are always numeric, so simply return 0 if requested frequency
+ * of non-numeric values.
+ */
+ if (pstats->type == JsonPathStatsArrayLength)
+ {
+ if (type != jbvNumeric)
+ return 0.0;
+
+ return jsonPathStatsGetFloat(pstats, "freq_array", defaultfreq);
+ }
+
+ if (pstats->type == JsonPathStatsObjectLength)
+ {
+ if (type != jbvNumeric)
+ return 0.0;
+
+ return jsonPathStatsGetFloat(pstats, "freq_object", defaultfreq);
+ }
+
+ /* Which JSON type are we interested in? Pick the right freq_type key. */
+ switch (type)
+ {
+ case jbvNull:
+ key = "freq_null";
+ break;
+ case jbvString:
+ key = "freq_string";
+ break;
+ case jbvNumeric:
+ key = "freq_numeric";
+ break;
+ case jbvBool:
+ key = "freq_boolean";
+ break;
+ case jbvObject:
+ key = "freq_object";
+ break;
+ case jbvArray:
+ key = "freq_array";
+ break;
+ default:
+ elog(ERROR, "Invalid jsonb value type: %d", type);
+ break;
+ }
+
+ return jsonPathStatsGetFloat(pstats, key, defaultfreq);
+}
+
+/*
+ * jsonPathStatsFormTuple
+ * For a pg_statistic tuple representing JSON statistics.
+ *
+ * XXX Maybe it's a bit expensive to first build StatsData and then transform it
+ * again while building the tuple. Could it be done in a single step? Would it be
+ * more efficient? Not sure how expensive it actually is, though.
+ */
+static HeapTuple
+jsonPathStatsFormTuple(JsonPathStats pstats, JsonStatType type, float4 nullfrac)
+{
+ StatsData statdata;
+
+ if (!pstats || !pstats->datum)
+ return NULL;
+
+ /*
+ * If it is the ordinary root path stats, there is no need to transform
+ * the tuple, it can be simply copied.
+ */
+ if (pstats->datum == &pstats->data->pathdatums[0] &&
+ pstats->type == JsonPathStatsValues)
+ return heap_copytuple(pstats->data->statsTuple);
+
+ MemSet(&statdata, 0, sizeof(statdata));
+
+ if (!jsonPathStatsExtractData(pstats, type, nullfrac, &statdata))
+ return NULL;
+
+ return stats_form_tuple(&statdata);
+}
+
+/*
+ * jsonStatsGetPathTuple
+ * Extract JSON statistics for a text[] path and form pg_statistics tuple.
+ */
+static HeapTuple
+jsonStatsGetPathTuple(JsonStats jsdata, JsonStatType type,
+ Datum *path, int pathlen, bool try_arrays_indexes)
+{
+ float4 nullfrac;
+ JsonPathStats pstats = jsonStatsGetPath(jsdata, path, pathlen,
+ try_arrays_indexes, &nullfrac);
+
+ return jsonPathStatsFormTuple(pstats, type, nullfrac);
+}
+
+/*
+ * jsonStatsGetArrayIndexStatsTuple
+ * Extract JSON statistics for a array index and form pg_statistics tuple.
+ */
+static HeapTuple
+jsonStatsGetArrayIndexStatsTuple(JsonStats jsdata, JsonStatType type, int32 index)
+{
+ /* Extract statistics for root array elements */
+ JsonPathStats arrstats = jsonStatsGetRootArrayPath(jsdata);
+ JsonPathStats rootstats;
+ Selectivity index_sel;
+
+ if (!arrstats)
+ return NULL;
+
+ /* Compute relative selectivity of 'EXISTS($[index])' */
+ rootstats = jsonStatsGetRootPath(jsdata);
+ index_sel = jsonPathStatsGetArrayIndexSelectivity(rootstats, index);
+ index_sel /= jsonPathStatsGetFreq(arrstats, 0.0);
+
+ /* Form pg_statistics tuple, taking into account array index selectivity */
+ return jsonPathStatsFormTuple(arrstats, type, 1.0 - index_sel);
+}
+
+/*
+ * jsonStatsGetPathFreq
+ * Return frequency of a path (fraction of documents containing it).
+ */
+static float4
+jsonStatsGetPathFreq(JsonStats jsdata, Datum *path, int pathlen,
+ bool try_array_indexes)
+{
+ float4 nullfrac;
+ JsonPathStats pstats = jsonStatsGetPath(jsdata, path, pathlen,
+ try_array_indexes, &nullfrac);
+ float4 freq = (1.0 - nullfrac) * jsonPathStatsGetFreq(pstats, 0.0);
+
+ CLAMP_PROBABILITY(freq);
+ return freq;
+}
+
+/*
+ * jsonbStatsVarOpConst
+ * Prepare optimizer statistics for a given operator, from JSON stats.
+ *
+ * This handles only OpExpr expressions, with variable and a constant. We get
+ * the constant as is, and the variable is represented by statistics fetched
+ * by get_restriction_variable().
+ *
+ * opid - OID of the operator (input parameter)
+ * resdata - pointer to calculated statistics for result of operator
+ * vardata - statistics for the restriction variable
+ * cnst - constant from the operator expression
+ *
+ * Returns true when useful optimizer statistics have been calculated.
+ */
+static bool
+jsonbStatsVarOpConst(Oid opid, VariableStatData *resdata,
+ const VariableStatData *vardata, Const *cnst)
+{
+ JsonStatData jsdata;
+ JsonStatType statype = JsonStatJsonb;
+
+ if (!jsonStatsInit(&jsdata, vardata))
+ return false;
+
+ switch (opid)
+ {
+ case JsonbObjectFieldTextOperator:
+ statype = JsonStatText;
+ /* FALLTHROUGH */
+ case JsonbObjectFieldOperator:
+ {
+ if (cnst->consttype != TEXTOID)
+ {
+ jsonStatsRelease(&jsdata);
+ return false;
+ }
+
+ resdata->statsTuple = jsonStatsGetPathTuple(&jsdata, statype,
+ &cnst->constvalue, 1,
+ false);
+ break;
+ }
+
+ case JsonbArrayElementTextOperator:
+ statype = JsonStatText;
+ /* FALLTHROUGH */
+ case JsonbArrayElementOperator:
+ {
+ if (cnst->consttype != INT4OID)
+ {
+ jsonStatsRelease(&jsdata);
+ return false;
+ }
+
+ resdata->statsTuple =
+ jsonStatsGetArrayIndexStatsTuple(&jsdata, statype,
+ DatumGetInt32(cnst->constvalue));
+ break;
+ }
+
+ case JsonbExtractPathTextOperator:
+ statype = JsonStatText;
+ /* FALLTHROUGH */
+ case JsonbExtractPathOperator:
+ {
+ Datum *path;
+ bool *nulls;
+ int pathlen;
+ bool have_nulls = false;
+
+ if (cnst->consttype != TEXTARRAYOID)
+ {
+ jsonStatsRelease(&jsdata);
+ return false;
+ }
+
+ deconstruct_array(DatumGetArrayTypeP(cnst->constvalue), TEXTOID,
+ -1, false, 'i', &path, &nulls, &pathlen);
+
+ for (int i = 0; i < pathlen; i++)
+ {
+ if (nulls[i])
+ {
+ have_nulls = true;
+ break;
+ }
+ }
+
+ if (!have_nulls)
+ resdata->statsTuple = jsonStatsGetPathTuple(&jsdata, statype,
+ path, pathlen,
+ true);
+
+ pfree(path);
+ pfree(nulls);
+ break;
+ }
+
+ default:
+ jsonStatsRelease(&jsdata);
+ return false;
+ }
+
+ if (!resdata->statsTuple)
+ resdata->statsTuple = stats_form_tuple(NULL); /* form all-NULL tuple */
+
+ resdata->acl_ok = vardata->acl_ok;
+ resdata->freefunc = heap_freetuple;
+ Assert(resdata->rel == vardata->rel);
+ Assert(resdata->atttype ==
+ (statype == JsonStatJsonb ? JSONBOID :
+ statype == JsonStatText ? TEXTOID :
+ /* statype == JsonStatFreq */ BOOLOID));
+
+ jsonStatsRelease(&jsdata);
+ return true;
+}
+
+/*
+ * jsonb_stats
+ * Statistics estimation procedure for JSONB data type.
+ *
+ * This only supports OpExpr expressions, with (Var op Const) shape.
+ *
+ * Var really can be a chain of OpExprs with derived statistics
+ * (jsonb_column -> 'key1' -> key2'), because get_restriction_variable()
+ * already handles this case.
+ */
+Datum
+jsonb_stats(PG_FUNCTION_ARGS)
+{
+ PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
+ OpExpr *opexpr = (OpExpr *) PG_GETARG_POINTER(1);
+ int varRelid = PG_GETARG_INT32(2);
+ VariableStatData *resdata = (VariableStatData *) PG_GETARG_POINTER(3);
+ VariableStatData vardata;
+ Node *constexpr;
+ bool varonleft;
+
+ /* should only be called for OpExpr expressions */
+ Assert(IsA(opexpr, OpExpr));
+
+ /* Is the expression simple enough? (Var op Const) or similar? */
+ if (!get_restriction_variable(root, opexpr->args, varRelid,
+ &vardata, &constexpr, &varonleft))
+ PG_RETURN_VOID();
+
+ /* XXX Could we also get varonleft=false in useful cases? */
+ if (IsA(constexpr, Const) && varonleft)
+ jsonbStatsVarOpConst(opexpr->opno, resdata, &vardata,
+ (Const *) constexpr);
+
+ ReleaseVariableStats(vardata);
+
+ PG_RETURN_VOID();
+}
+
+/*
+ * jsonSelectivity
+ * Use JSON statistics to estimate selectivity for (in)equalities.
+ *
+ * The statistics is represented as (arrays of) JSON values etc. so we
+ * need to pass the right operators to the functions.
+ */
+Selectivity
+jsonSelectivity(JsonPathStats stats, Datum scalar, Oid operator)
+{
+ VariableStatData vardata;
+ Selectivity sel;
+
+ if (!stats)
+ return 0.0;
+
+ vardata.atttype = JSONBOID;
+ vardata.atttypmod = -1;
+ vardata.isunique = false;
+ vardata.rel = stats->data->rel;
+ vardata.var = NULL;
+ vardata.vartype = JSONBOID;
+ vardata.acl_ok = stats->data->acl_ok;
+ vardata.statsTuple = jsonPathStatsFormTuple(stats,
+ JsonStatJsonbWithoutSubpaths, 0.0);
+
+ if (operator == JsonbEqOperator)
+ sel = var_eq_const(&vardata, operator, InvalidOid, scalar, false, true, false);
+ else
+ sel = scalarineqsel(NULL, operator,
+ /* is it greater or greater-or-equal? */
+ operator == JsonbGtOperator ||
+ operator == JsonbGeOperator,
+ /* is it equality? */
+ operator == JsonbLeOperator ||
+ operator == JsonbGeOperator,
+ InvalidOid,
+ &vardata, scalar, JSONBOID);
+
+ if (vardata.statsTuple)
+ heap_freetuple(vardata.statsTuple);
+
+ return sel;
+}
+
+/*
+ * jsonAccumulateSubPathSelectivity
+ * Transform absolute subpath selectivity into relative and accumulate it
+ * into parent path simply by multiplication of relative selectivities.
+ */
+static void
+jsonAccumulateSubPathSelectivity(Selectivity subpath_abs_sel,
+ Selectivity path_freq,
+ Selectivity *path_relative_sel,
+ JsonPathStats array_path_stats)
+{
+ Selectivity sel = subpath_abs_sel / path_freq; /* relative selectivity */
+
+ /* XXX Try to take into account array length */
+ if (array_path_stats)
+ sel = 1.0 - pow(1.0 - sel,
+ jsonPathStatsGetAvgArraySize(array_path_stats));
+
+ /* Accumulate selectivity of subpath into parent path */
+ *path_relative_sel *= sel;
+}
+
+/*
+ * jsonSelectivityContains
+ * Estimate selectivity for containment operator on JSON.
+ *
+ * Iterate through query jsonb elements, build paths to its leaf elements,
+ * calculate selectivies of 'path == scalar' in leaves, multiply relative
+ * selectivities of subpaths at each path level, propagate computed
+ * selectivities to the root.
+ */
+static Selectivity
+jsonSelectivityContains(JsonStats stats, Jsonb *jb)
+{
+ JsonbValue v;
+ JsonbIterator *it;
+ JsonbIteratorToken r;
+ StringInfoData pathstr; /* path string */
+ struct Path /* path stack entry */
+ {
+ struct Path *parent; /* parent entry */
+ int len; /* associated length of pathstr */
+ Selectivity freq; /* absolute frequence of path */
+ Selectivity sel; /* relative selectivity of subpaths */
+ JsonPathStats stats; /* statistics for the path */
+ bool is_array_accesor; /* is it '[*]' ? */
+ } root, /* root path entry */
+ *path = &root; /* path entry stack */
+ Selectivity sel; /* resulting selectivity */
+ Selectivity scalarSel; /* selectivity of 'jsonb == scalar' */
+
+ /* Initialize root path string */
+ initStringInfo(&pathstr);
+ appendBinaryStringInfo(&pathstr, stats->prefix, stats->prefixlen);
+
+ /* Initialize root path entry */
+ root.parent = NULL;
+ root.len = pathstr.len;
+ root.stats = jsonStatsFindPath(stats, pathstr.data, pathstr.len);
+ root.freq = jsonPathStatsGetFreq(root.stats, 0.0);
+ root.sel = 1.0;
+ root.is_array_accesor = pathstr.data[pathstr.len - 1] == ']';
+
+ /* Return 0, if NULL fraction is 1. */
+ if (root.freq <= 0.0)
+ return 0.0;
+
+ /*
+ * Selectivity of query 'jsonb @> scalar' consists of selectivities of
+ * 'jsonb == scalar' and 'jsonb[*] == scalar'. Selectivity of
+ * 'jsonb[*] == scalar' will be computed in root.sel, but for
+ * 'jsonb == scalar' we need additional computation.
+ */
+ if (JsonContainerIsScalar(&jb->root))
+ scalarSel = jsonSelectivity(root.stats, JsonbPGetDatum(jb),
+ JsonbEqOperator);
+ else
+ scalarSel = 0.0;
+
+ it = JsonbIteratorInit(&jb->root);
+
+ while ((r = JsonbIteratorNext(&it, &v, false)) != WJB_DONE)
+ {
+ switch (r)
+ {
+ case WJB_BEGIN_OBJECT:
+ {
+ struct Path *p;
+ Selectivity freq =
+ jsonPathStatsGetTypeFreq(path->stats, jbvObject, 0.0);
+
+ /* If there are no objects, selectivity is 0. */
+ if (freq <= 0.0)
+ return 0.0;
+
+ /*
+ * Push path entry for object keys, actual key names are
+ * appended later in WJB_KEY case.
+ */
+ p = palloc(sizeof(*p));
+ p->len = pathstr.len;
+ p->parent = path;
+ p->stats = NULL;
+ p->freq = freq;
+ p->sel = 1.0;
+ p->is_array_accesor = false;
+ path = p;
+ break;
+ }
+
+ case WJB_BEGIN_ARRAY:
+ {
+ struct Path *p;
+ JsonPathStats pstats;
+ Selectivity freq;
+
+ /*
+ * First, find stats for the parent path if needed, it will be
+ * used in jsonAccumulateSubPathSelectivity().
+ */
+ if (!path->stats)
+ path->stats = jsonStatsFindPath(stats, pathstr.data,
+ pathstr.len);
+
+ /* Appeend path string entry for array elements, get stats. */
+ jsonPathAppendEntry(&pathstr, NULL);
+ pstats = jsonStatsFindPath(stats, pathstr.data, pathstr.len);
+ freq = jsonPathStatsGetFreq(pstats, 0.0);
+
+ /* If there are no arrays, return 0 or scalar selectivity */
+ if (freq <= 0.0)
+ return scalarSel;
+
+ /* Push path entry for array elements. */
+ p = palloc(sizeof(*p));
+ p->len = pathstr.len;
+ p->parent = path;
+ p->stats = pstats;
+ p->freq = freq;
+ p->sel = 1.0;
+ p->is_array_accesor = true;
+ path = p;
+ break;
+ }
+
+ case WJB_END_OBJECT:
+ case WJB_END_ARRAY:
+ {
+ struct Path *p = path;
+ /* Absoulte selectivity of the path with its all subpaths */
+ Selectivity abs_sel = p->sel * p->freq;
+
+ /* Pop last path entry */
+ path = path->parent;
+ pfree(p);
+ pathstr.len = path->len;
+ pathstr.data[pathstr.len] = '\0';
+
+ /* Accumulate selectivity into parent path */
+ jsonAccumulateSubPathSelectivity(abs_sel, path->freq,
+ &path->sel,
+ path->is_array_accesor ?
+ path->parent->stats : NULL);
+ break;
+ }
+
+ case WJB_KEY:
+ {
+ /* Remove previous key in the path string */
+ pathstr.len = path->parent->len;
+ pathstr.data[pathstr.len] = '\0';
+
+ /* Append current key to path string */
+ jsonPathAppendEntryWithLen(&pathstr, v.val.string.val,
+ v.val.string.len);
+ path->len = pathstr.len;
+ break;
+ }
+
+ case WJB_VALUE:
+ case WJB_ELEM:
+ {
+ /*
+ * Extract statistics for a path. Array elements share the
+ * same statistics that was extracted in WJB_BEGIN_ARRAY.
+ */
+ JsonPathStats pstats = r == WJB_ELEM ? path->stats :
+ jsonStatsFindPath(stats, pathstr.data, pathstr.len);
+ Selectivity abs_sel; /* Absolute selectivity of 'path == scalar' */
+
+ if (pstats)
+ {
+ /* Make scalar jsonb datum and compute selectivity */
+ Datum scalar = JsonbPGetDatum(JsonbValueToJsonb(&v));
+
+ abs_sel = jsonSelectivity(pstats, scalar, JsonbEqOperator);
+ }
+ else
+ abs_sel = 0.0;
+
+ /* Accumulate selectivity into parent path */
+ jsonAccumulateSubPathSelectivity(abs_sel, path->freq,
+ &path->sel,
+ path->is_array_accesor ?
+ path->parent->stats : NULL);
+ break;
+ }
+
+ default:
+ break;
+ }
+ }
+
+ /* Compute absolute selectivity for root, including raw scalar case. */
+ sel = root.sel * root.freq + scalarSel;
+ CLAMP_PROBABILITY(sel);
+ return sel;
+}
+
+/*
+ * jsonSelectivityExists
+ * Estimate selectivity for JSON "exists" operator.
+ */
+static Selectivity
+jsonSelectivityExists(JsonStats stats, Datum key)
+{
+ JsonPathStats rootstats;
+ JsonPathStats arrstats;
+ JsonbValue jbvkey;
+ Datum jbkey;
+ Selectivity keysel;
+ Selectivity scalarsel;
+ Selectivity arraysel;
+ Selectivity sel;
+
+ JsonValueInitStringWithLen(&jbvkey,
+ VARDATA_ANY(key), VARSIZE_ANY_EXHDR(key));
+
+ jbkey = JsonbPGetDatum(JsonbValueToJsonb(&jbvkey));
+
+ keysel = jsonStatsGetPathFreq(stats, &key, 1, false);
+
+ rootstats = jsonStatsGetRootPath(stats);
+ scalarsel = jsonSelectivity(rootstats, jbkey, JsonbEqOperator);
+
+ arrstats = jsonStatsGetRootArrayPath(stats);
+ arraysel = jsonSelectivity(arrstats, jbkey, JsonbEqOperator);
+ arraysel = 1.0 - pow(1.0 - arraysel,
+ jsonPathStatsGetAvgArraySize(rootstats));
+
+ sel = keysel + scalarsel + arraysel;
+ CLAMP_PROBABILITY(sel);
+ return sel;
+}
+
+static Selectivity
+jsonb_sel_internal(JsonStats stats, Oid operator, Const *cnst, bool varonleft)
+{
+ switch (operator)
+ {
+ case JsonbExistsOperator:
+ if (!varonleft || cnst->consttype != TEXTOID)
+ break;
+
+ return jsonSelectivityExists(stats, cnst->constvalue);
+
+ case JsonbExistsAnyOperator:
+ case JsonbExistsAllOperator:
+ {
+ Datum *keys;
+ bool *nulls;
+ Selectivity freq = 1.0;
+ int nkeys;
+ int i;
+ bool all = operator == JsonbExistsAllOperator;
+
+ if (!varonleft || cnst->consttype != TEXTARRAYOID)
+ break;
+
+ deconstruct_array(DatumGetArrayTypeP(cnst->constvalue), TEXTOID,
+ -1, false, 'i', &keys, &nulls, &nkeys);
+
+ for (i = 0; i < nkeys; i++)
+ if (!nulls[i])
+ {
+ Selectivity pathfreq = jsonSelectivityExists(stats,
+ keys[i]);
+ freq *= all ? pathfreq : (1.0 - pathfreq);
+ }
+
+ pfree(keys);
+ pfree(nulls);
+
+ if (!all)
+ freq = 1.0 - freq;
+
+ return freq;
+ }
+
+ case JsonbContainedOperator:
+ if (varonleft || cnst->consttype != JSONBOID)
+ break;
+
+ return jsonSelectivityContains(stats,
+ DatumGetJsonbP(cnst->constvalue));
+
+ case JsonbContainsOperator:
+ if (!varonleft || cnst->consttype != JSONBOID)
+ break;
+
+ return jsonSelectivityContains(stats,
+ DatumGetJsonbP(cnst->constvalue));
+
+ default:
+ break;
+ }
+
+ return DEFAULT_JSON_CONTAINS_SEL;
+}
+
+/*
+ * jsonb_sel
+ * The main procedure estimating selectivity for all JSONB operators.
+ */
+Datum
+jsonb_sel(PG_FUNCTION_ARGS)
+{
+ PlannerInfo *root = (PlannerInfo *) PG_GETARG_POINTER(0);
+ Oid operator = PG_GETARG_OID(1);
+ List *args = (List *) PG_GETARG_POINTER(2);
+ int varRelid = PG_GETARG_INT32(3);
+ double sel = DEFAULT_JSON_CONTAINS_SEL;
+ Node *other;
+ bool varonleft;
+ VariableStatData vardata;
+
+ if (get_restriction_variable(root, args, varRelid,
+ &vardata, &other, &varonleft))
+ {
+ if (IsA(other, Const))
+ {
+ Const *cnst = (Const *) other;
+
+ if (cnst->constisnull)
+ sel = 0.0;
+ else
+ {
+ JsonStatData stats;
+
+ if (jsonStatsInit(&stats, &vardata))
+ {
+ sel = jsonb_sel_internal(&stats, operator, cnst, varonleft);
+ jsonStatsRelease(&stats);
+ }
+ }
+ }
+
+ ReleaseVariableStats(vardata);
+ }
+
+ PG_RETURN_FLOAT8((float8) sel);
+}
diff --git a/src/backend/utils/adt/jsonb_typanalyze.c b/src/backend/utils/adt/jsonb_typanalyze.c
new file mode 100644
index 00000000000..7882db23a87
--- /dev/null
+++ b/src/backend/utils/adt/jsonb_typanalyze.c
@@ -0,0 +1,1627 @@
+/*-------------------------------------------------------------------------
+ *
+ * jsonb_typanalyze.c
+ * Functions for gathering statistics from jsonb columns
+ *
+ * Copyright (c) 2016-2022, PostgreSQL Global Development Group
+ *
+ * Functions in this module are used to analyze contents of JSONB columns
+ * and build optimizer statistics. In principle we extract paths from all
+ * sampled documents and calculate the usual statistics (MCV, histogram)
+ * for each path - in principle each path is treated as a column.
+ *
+ * Because we're not enforcing any JSON schema, the documents may differ
+ * a lot - the documents may contain large number of different keys, the
+ * types of values may be entirely different, etc. This makes it more
+ * challenging than building stats for regular columns. For example not
+ * only do we need to decide which values to keep in the MCV, but also
+ * which paths to keep (in case the documents are so variable we can't
+ * keep all paths).
+ *
+ * The statistics is stored in pg_statistic, in a slot with a new stakind
+ * value (STATISTIC_KIND_JSON). The statistics is serialized as an array
+ * of JSONB values, eash element storing statistics for one path.
+ *
+ * For each path, we store the following keys:
+ *
+ * - path - path this stats is for, serialized as jsonpath
+ * - freq - frequency of documents containing this path
+ * - json - the regular per-column stats (MCV, histogram, ...)
+ * - freq_null - frequency of JSON null values
+ * - freq_array - frequency of JSON array values
+ * - freq_object - frequency of JSON object values
+ * - freq_string - frequency of JSON string values
+ * - freq_numeric - frequency of JSON numeric values
+ *
+ * This is stored in the stavalues array.
+ *
+ * The first element of stavalues is a path prefix. It is used for avoiding
+ * path transformations when the derived statistics for the chains of ->
+ * operators is computed.
+ *
+ * The per-column stats (stored in the "json" key) have additional internal
+ * structure, to allow storing multiple stakind types (histogram, mcv). See
+ * jsonAnalyzeMakeScalarStats for details.
+ *
+ *
+ * XXX It's a bit weird the "regular" stats are stored in the "json" key,
+ * while the JSON stats (frequencies of different JSON types) are right
+ * at the top level.
+ *
+ *
+ * IDENTIFICATION
+ * src/backend/utils/adt/jsonb_typanalyze.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#include "postgres.h"
+
+#include "fmgr.h"
+#include "access/hash.h"
+#include "access/detoast.h"
+#include "catalog/pg_attribute.h"
+#include "catalog/pg_operator.h"
+#include "catalog/pg_type.h"
+#include "commands/vacuum.h"
+#include "utils/builtins.h"
+#include "utils/guc.h"
+#include "utils/hsearch.h"
+#include "utils/json.h"
+#include "utils/jsonb.h"
+#include "utils/json_selfuncs.h"
+#include "utils/lsyscache.h"
+#include "utils/memutils.h"
+
+typedef struct JsonPathEntry JsonPathEntry;
+
+/*
+ * Element of a path in the JSON document (i.e. not jsonpath). Elements
+ * are linked together to build longer paths.
+ *
+ * 'entry' can be not zero-terminated when it is pointing to JSONB keys, so
+ * 'len' is necessary. 'len' is also used for faster entry comparison, to
+ * distinguish array entries ('len' == -1).
+ */
+typedef struct JsonPathEntry
+{
+ JsonPathEntry *parent;
+ const char *entry; /* element of the path as a string */
+ int len; /* length of entry string (may be 0 or -1) */
+ uint32 hash; /* hash of the whole path (with parent) */
+ char *pathstr; /* full path string */
+ int depth; /* nesting level, i.e. path length */
+} JsonPathEntry;
+
+#define JsonPathEntryIsArray(entry) ((entry)->len == -1)
+
+/*
+ * An array containing a dynamic number of values extracted from JSON documents.
+ * All values should have the same data type:
+ * jsonb - ordinary path stats, values of different JSON types
+ * int32 - array/object length stats
+ * text - separate stats fro strings
+ * numeric - separate stats fro numbers
+ */
+typedef struct JsonValues
+{
+ Datum *buf;
+ int count;
+ int allocated;
+} JsonValues;
+
+/*
+ * Scalar statistics built for an array of values, extracted from a JSON
+ * document (for one particular path).
+ */
+typedef struct JsonScalarStats
+{
+ JsonValues values;
+ VacAttrStats stats;
+} JsonScalarStats;
+
+/*
+ * Statistics calculated for a set of values.
+ *
+ *
+ * XXX This seems rather complicated and needs simplification. We're not
+ * really using all the various JsonScalarStats bits, there's a lot of
+ * duplication (e.g. each JsonScalarStats contains it's own array, which
+ * has a copy of data from the one in "jsons").
+ */
+typedef struct JsonValueStats
+{
+ JsonScalarStats jsons; /* stats for all JSON types together */
+
+#ifdef JSON_ANALYZE_SCALARS /* XXX */
+ JsonScalarStats strings; /* stats for JSON strings */
+ JsonScalarStats numerics; /* stats for JSON numerics */
+#endif
+
+ JsonScalarStats arrlens; /* stats of array lengths */
+ JsonScalarStats objlens; /* stats of object lengths */
+
+ int nnulls; /* number of JSON null values */
+ int ntrue; /* number of JSON true values */
+ int nfalse; /* number of JSON false values */
+ int nobjects; /* number of JSON objects */
+ int narrays; /* number of JSON arrays */
+ int nstrings; /* number of JSON strings */
+ int nnumerics; /* number of JSON numerics */
+
+ int64 narrelems; /* total number of array elements
+ * (for avg. array length) */
+} JsonValueStats;
+
+typedef struct JsonPathDocBitmap
+{
+ bool is_list;
+ int size;
+ int allocated;
+ union
+ {
+ int32 *list;
+ uint8 *bitmap;
+ } data;
+} JsonPathDocBitmap;
+
+/* JSON path and list of documents containing it */
+typedef struct JsonPathAnlDocs
+{
+ JsonPathEntry path;
+ JsonPathDocBitmap bitmap;
+} JsonPathAnlDocs;
+
+/* Main structure for analyzed JSON path */
+typedef struct JsonPathAnlStats
+{
+ JsonPathEntry path;
+ double freq; /* frequence of the path */
+ JsonValueStats vstats; /* collected values and raw computed stats */
+ Jsonb *stats; /* stats converted into jsonb form */
+} JsonPathAnlStats;
+
+/* Some parent path stats counters that used for frequency calculations */
+typedef struct JsonPathParentStats
+{
+ double freq;
+ int count;
+ int narrays;
+} JsonPathParentStats;
+
+/* various bits needed while analyzing JSON */
+typedef struct JsonAnalyzeContext
+{
+ VacAttrStats *stats;
+ MemoryContext mcxt;
+ AnalyzeAttrFetchFunc fetchfunc;
+ HTAB *pathshash;
+ JsonPathAnlStats *root;
+ double totalrows;
+ double total_width;
+ int samplerows;
+ int current_rownum;
+ int target;
+ int null_cnt;
+ int analyzed_cnt;
+ int maxdepth;
+ bool scalarsOnly;
+ bool single_pass;
+} JsonAnalyzeContext;
+
+/*
+ * JsonPathMatch
+ * Determine when two JSON paths (list of JsonPathEntry) match.
+ *
+ * Returned int instead of bool, because it is an implementation of
+ * HashCompareFunc.
+ */
+static int
+JsonPathEntryMatch(const void *key1, const void *key2, Size keysize)
+{
+ const JsonPathEntry *path1 = key1;
+ const JsonPathEntry *path2 = key2;
+
+ return path1->parent != path2->parent ||
+ path1->len != path2->len ||
+ (path1->len > 0 &&
+ strncmp(path1->entry, path2->entry, path1->len));
+}
+
+/*
+ * JsonPathHash
+ * Calculate hash of the path entry.
+ *
+ * Parent hash should be already calculated.
+ */
+static uint32
+JsonPathEntryHash(const void *key, Size keysize)
+{
+ const JsonPathEntry *path = key;
+ uint32 hash = path->parent ? path->parent->hash : 0;
+
+ hash = (hash << 1) | (hash >> 31);
+ hash ^= path->len < 0 ? 0 :
+ DatumGetUInt32(hash_any((const unsigned char *) path->entry, path->len));
+
+ return hash;
+}
+
+static void
+jsonStatsBitmapInit(JsonPathDocBitmap *bitmap)
+{
+ memset(bitmap, 0, sizeof(*bitmap));
+ bitmap->is_list = true;
+}
+
+static void
+jsonStatsBitmapAdd(JsonAnalyzeContext *cxt, JsonPathDocBitmap *bitmap, int doc)
+{
+ /* Use more compact list representation if not too many bits set */
+ if (bitmap->is_list)
+ {
+ int *list = bitmap->data.list;
+
+#if 1 /* Enable list representation */
+ if (bitmap->size > 0 && list[bitmap->size - 1] == doc)
+ return;
+
+ if (bitmap->size < cxt->samplerows / sizeof(list[0]) / 8)
+ {
+ if (bitmap->size >= bitmap->allocated)
+ {
+ MemoryContext oldcxt = MemoryContextSwitchTo(cxt->mcxt);
+
+ if (bitmap->allocated)
+ {
+ bitmap->allocated *= 2;
+ list = repalloc(list, sizeof(list[0]) * bitmap->allocated);
+ }
+ else
+ {
+ bitmap->allocated = 8;
+ list = palloc(sizeof(list[0]) * bitmap->allocated);
+ }
+
+ bitmap->data.list = list;
+
+ MemoryContextSwitchTo(oldcxt);
+ }
+
+ list[bitmap->size++] = doc;
+ return;
+ }
+#endif
+ /* convert list to bitmap */
+ bitmap->allocated = (cxt->samplerows + 7) / 8;
+ bitmap->data.bitmap = MemoryContextAllocZero(cxt->mcxt, bitmap->allocated);
+ bitmap->is_list = false;
+
+ if (list)
+ {
+ for (int i = 0; i < bitmap->size; i++)
+ {
+ int d = list[i];
+
+ bitmap->data.bitmap[d / 8] |= (1 << (d % 8));
+ }
+
+ pfree(list);
+ }
+ }
+
+ /* set bit in bitmap */
+ if (doc < cxt->samplerows &&
+ !(bitmap->data.bitmap[doc / 8] & (1 << (doc % 8))))
+ {
+ bitmap->data.bitmap[doc / 8] |= (1 << (doc % 8));
+ bitmap->size++;
+ }
+}
+
+static bool
+jsonStatsBitmapNext(JsonPathDocBitmap *bitmap, int *pbit)
+{
+ uint8 *bmp = bitmap->data.bitmap;
+ uint8 *pb;
+ uint8 *pb_end = &bmp[bitmap->allocated];
+ int bit = *pbit;
+
+ Assert(!bitmap->is_list);
+
+ if (bit < 0)
+ {
+ pb = bmp;
+ bit = 0;
+ }
+ else
+ {
+ ++bit;
+ pb = &bmp[bit / 8];
+ bit %= 8;
+ }
+
+ for (; pb < pb_end; pb++, bit = 0)
+ {
+ uint8 b;
+
+ /* Skip zero bytes */
+ if (!bit)
+ {
+ while (!*pb)
+ {
+ if (++pb >= pb_end)
+ return false;
+ }
+ }
+
+ b = *pb;
+
+ /* Skip zero bits */
+ while (bit < 8 && !(b & (1 << bit)))
+ bit++;
+
+ if (bit >= 8)
+ continue; /* Non-zero bit not found, go to next byte */
+
+ /* Output next non-zero bit */
+ *pbit = (pb - bmp) * 8 + bit;
+ return true;
+ }
+
+ return false;
+}
+
+static void
+jsonStatsAnlInit(JsonPathAnlStats *stats)
+{
+ /* initialize the stats counter for this path entry */
+ memset(&stats->vstats, 0, sizeof(JsonValueStats));
+ stats->stats = NULL;
+ stats->freq = 0.0;
+}
+
+/*
+ * jsonAnalyzeAddPath
+ * Add an entry for a JSON path to the working list of statistics.
+ *
+ * Returns a pointer to JsonPathAnlStats (which might have already existed
+ * if the path was in earlier document), which can then be populated or
+ * updated.
+ */
+static inline JsonPathEntry *
+jsonAnalyzeAddPath(JsonAnalyzeContext *ctx, JsonPathEntry *parent,
+ const char *entry, int len)
+{
+ JsonPathEntry path;
+ JsonPathEntry *stats;
+ bool found;
+
+ /* Init path entry */
+ path.parent = parent;
+ path.entry = entry;
+ path.len = len;
+ path.hash = JsonPathEntryHash(&path, 0);
+
+ /* See if we already saw this path earlier. */
+ stats = hash_search_with_hash_value(ctx->pathshash, &path, path.hash,
+ HASH_ENTER, &found);
+
+ /*
+ * Nope, it's the first time we see this path, so initialize all the
+ * fields (path string, counters, ...).
+ */
+ if (!found)
+ {
+ JsonPathEntry *parent = stats->parent;
+ const char *ppath = parent->pathstr;
+ StringInfoData si;
+ MemoryContext oldcxt;
+
+ oldcxt = MemoryContextSwitchTo(ctx->mcxt);
+
+ /* NULL entries are treated as wildcard array accessors "[*]" */
+ if (stats->entry)
+ /* Copy path entry name into the right MemoryContext */
+ stats->entry = pnstrdup(stats->entry, stats->len);
+
+ MemoryContextSwitchTo(oldcxt);
+
+ /* Initialze full path string */
+ initStringInfo(&si);
+ appendStringInfoString(&si, ppath);
+ jsonPathAppendEntry(&si, stats->entry);
+
+ MemoryContextSwitchTo(ctx->mcxt);
+ stats->pathstr = pstrdup(si.data);
+ MemoryContextSwitchTo(oldcxt);
+
+ pfree(si.data);
+
+ if (ctx->single_pass)
+ jsonStatsAnlInit((JsonPathAnlStats *) stats);
+ else
+ jsonStatsBitmapInit(&((JsonPathAnlDocs *) stats)->bitmap);
+
+ stats->depth = parent->depth + 1;
+
+ /* update maximal depth */
+ if (ctx->maxdepth < stats->depth)
+ ctx->maxdepth = stats->depth;
+ }
+
+ return stats;
+}
+
+/*
+ * JsonValuesAppend
+ * Add a JSON value to the dynamic array (enlarge it if needed).
+ *
+ * XXX This is likely one of the problems - the documents may be pretty
+ * large, with a lot of different values for each path. At that point
+ * it's problematic to keep all of that in memory at once. So maybe we
+ * need to introduce some sort of compaction (e.g. we could try
+ * deduplicating the values), limit on size of the array or something.
+ */
+static inline void
+JsonValuesAppend(JsonValues *values, Datum value, int initialSize)
+{
+ if (values->count >= values->allocated)
+ {
+ if (values->allocated)
+ {
+ values->allocated = values->allocated * 2;
+ values->buf = repalloc(values->buf,
+ sizeof(values->buf[0]) * values->allocated);
+ }
+ else
+ {
+ values->allocated = initialSize;
+ values->buf = palloc(sizeof(values->buf[0]) * values->allocated);
+ }
+ }
+
+ values->buf[values->count++] = value;
+}
+
+/*
+ * jsonAnalyzeJsonValue
+ * Process a value extracted from the document (for a given path).
+ */
+static inline void
+jsonAnalyzeJsonValue(JsonAnalyzeContext *ctx, JsonValueStats *vstats,
+ JsonbValue *jv)
+{
+ JsonbValue *jbv;
+ JsonbValue jbvtmp;
+ Jsonb *jb;
+ Datum value;
+ MemoryContext oldcxt = NULL;
+
+ /* XXX if analyzing only scalar values, make containers empty */
+ if (ctx->scalarsOnly && jv->type == jbvBinary)
+ {
+ if (JsonContainerIsObject(jv->val.binary.data))
+ jbv = JsonValueInitObject(&jbvtmp, 0, 0);
+ else
+ {
+ Assert(JsonContainerIsArray(jv->val.binary.data));
+ jbv = JsonValueInitArray(&jbvtmp, 0, 0, false);
+ }
+ }
+ else
+ jbv = jv;
+
+ jb = JsonbValueToJsonb(jbv);
+
+ if (ctx->single_pass)
+ {
+ oldcxt = MemoryContextSwitchTo(ctx->stats->anl_context);
+ jb = memcpy(palloc(VARSIZE(jb)), jb, VARSIZE(jb));
+ }
+
+ /* always add it to the "global" JSON stats, shared by all types */
+ JsonValuesAppend(&vstats->jsons.values,
+ JsonbPGetDatum(jb),
+ ctx->target);
+
+ /* also update the type-specific counters */
+ switch (jv->type)
+ {
+ case jbvNull:
+ vstats->nnulls++;
+ break;
+
+ case jbvBool:
+ if (jv->val.boolean)
+ vstats->ntrue++;
+ else
+ vstats->nfalse++;
+ break;
+
+ case jbvString:
+ vstats->nstrings++;
+#ifdef JSON_ANALYZE_SCALARS
+ value = PointerGetDatum(
+ cstring_to_text_with_len(jv->val.string.val,
+ jv->val.string.len));
+ JsonValuesAppend(&vstats->strings.values, value, ctx->target);
+#endif
+ break;
+
+ case jbvNumeric:
+ vstats->nnumerics++;
+#ifdef JSON_ANALYZE_SCALARS
+ value = PointerGetDatum(jv->val.numeric);
+ JsonValuesAppend(&vstats->numerics.values, value, ctx->target);
+#endif
+ break;
+
+ case jbvBinary:
+ if (JsonContainerIsObject(jv->val.binary.data))
+ {
+ uint32 size = JsonContainerSize(jv->val.binary.data);
+
+ value = DatumGetInt32(size);
+ vstats->nobjects++;
+ JsonValuesAppend(&vstats->objlens.values, value, ctx->target);
+ }
+ else if (JsonContainerIsArray(jv->val.binary.data))
+ {
+ uint32 size = JsonContainerSize(jv->val.binary.data);
+
+ value = DatumGetInt32(size);
+ vstats->narrays++;
+ JsonValuesAppend(&vstats->arrlens.values, value, ctx->target);
+ vstats->narrelems += size;
+ }
+ break;
+
+ default:
+ elog(ERROR, "invalid scalar json value type %d", jv->type);
+ break;
+ }
+
+ if (ctx->single_pass)
+ MemoryContextSwitchTo(oldcxt);
+}
+
+/*
+ * jsonAnalyzeCollectPaths
+ * Parse the JSON document and collect all paths and their values.
+ */
+static void
+jsonAnalyzeCollectPaths(JsonAnalyzeContext *ctx, Jsonb *jb, void *param)
+{
+ JsonbValue jv;
+ JsonbIterator *it;
+ JsonbIteratorToken tok;
+ JsonPathEntry *stats = &ctx->root->path;
+ int doc = ctx->current_rownum;
+ bool collect_values = (bool)(intptr_t) param;
+ bool scalar = false;
+
+ if (collect_values && !JB_ROOT_IS_SCALAR(jb))
+ jsonAnalyzeJsonValue(ctx, &((JsonPathAnlStats *) stats)->vstats,
+ JsonValueInitBinary(&jv, jb));
+
+ it = JsonbIteratorInit(&jb->root);
+
+ while ((tok = JsonbIteratorNext(&it, &jv, true)) != WJB_DONE)
+ {
+ switch (tok)
+ {
+ case WJB_BEGIN_OBJECT:
+ /*
+ * Read next token to see if the object is empty or not.
+ * If not, make stats for the first key. Subsequent WJB_KEYs
+ * and WJB_END_OBJECT will expect that stats will be pointing
+ * to the key of current object.
+ */
+ tok = JsonbIteratorNext(&it, &jv, true);
+
+ if (tok == WJB_END_OBJECT)
+ /* Empty object, simply skip stats initialization. */
+ break;
+
+ if (tok != WJB_KEY)
+ elog(ERROR, "unexpected jsonb iterator token: %d", tok);
+
+ stats = jsonAnalyzeAddPath(ctx, stats,
+ jv.val.string.val,
+ jv.val.string.len);
+ break;
+
+ case WJB_BEGIN_ARRAY:
+ /* Make stats for non-scalar array and use it for all elements */
+ if (!(scalar = jv.val.array.rawScalar))
+ stats = jsonAnalyzeAddPath(ctx, stats, NULL, -1);
+ break;
+
+ case WJB_END_ARRAY:
+ if (scalar)
+ break;
+ /* FALLTHROUGH */
+ case WJB_END_OBJECT:
+ /* Reset to parent stats */
+ stats = stats->parent;
+ break;
+
+ case WJB_KEY:
+ /*
+ * Stats should point to the previous key of current object,
+ * use its parent path as a base path.
+ */
+ stats = jsonAnalyzeAddPath(ctx, stats->parent,
+ jv.val.string.val,
+ jv.val.string.len);
+ break;
+
+ case WJB_VALUE:
+ case WJB_ELEM:
+ if (collect_values)
+ jsonAnalyzeJsonValue(ctx,
+ &((JsonPathAnlStats *) stats)->vstats,
+ &jv);
+ else if (stats != &ctx->root->path)
+ jsonStatsBitmapAdd(ctx,
+ &((JsonPathAnlDocs *) stats)->bitmap,
+ doc);
+
+ /*
+ * Manually recurse into container by creating child iterator.
+ * We use skipNested=true to give jsonAnalyzeJsonValue()
+ * ability to access jbvBinary containers.
+ */
+ if (jv.type == jbvBinary)
+ {
+ JsonbIterator *it2 = JsonbIteratorInit(jv.val.binary.data);
+
+ it2->parent = it;
+ it = it2;
+ }
+ break;
+
+ default:
+ break;
+ }
+ }
+}
+
+/*
+ * jsonAnalyzeCollectSubpath
+ * Recursively extract trailing part of a path and collect its values.
+ */
+static void
+jsonAnalyzeCollectSubpath(JsonAnalyzeContext *ctx, JsonPathAnlStats *pstats,
+ JsonbValue *jbv, JsonPathEntry **entries,
+ int start_entry)
+{
+ JsonbValue scalar;
+ int i;
+
+ for (i = start_entry; i < pstats->path.depth; i++)
+ {
+ JsonPathEntry *entry = entries[i];
+ JsonbContainer *jbc = jbv->val.binary.data;
+ JsonbValueType type = jbv->type;
+
+ if (i > start_entry)
+ pfree(jbv);
+
+ if (type != jbvBinary)
+ return;
+
+ if (JsonPathEntryIsArray(entry))
+ {
+ JsonbIterator *it;
+ JsonbIteratorToken r;
+ JsonbValue elem;
+
+ if (!JsonContainerIsArray(jbc) || JsonContainerIsScalar(jbc))
+ return;
+
+ it = JsonbIteratorInit(jbc);
+
+ while ((r = JsonbIteratorNext(&it, &elem, true)) != WJB_DONE)
+ {
+ if (r == WJB_ELEM)
+ jsonAnalyzeCollectSubpath(ctx, pstats, &elem, entries, i + 1);
+ }
+
+ return;
+ }
+ else
+ {
+ if (!JsonContainerIsObject(jbc))
+ return;
+
+ jbv = findJsonbValueFromContainerLen(jbc, JB_FOBJECT,
+ entry->entry, entry->len);
+
+ if (!jbv)
+ return;
+ }
+ }
+
+ if (i == start_entry &&
+ jbv->type == jbvBinary &&
+ JsonbExtractScalar(jbv->val.binary.data, &scalar))
+ jbv = &scalar;
+
+ jsonAnalyzeJsonValue(ctx, &pstats->vstats, jbv);
+
+ if (i > start_entry)
+ pfree(jbv);
+}
+
+/*
+ * jsonAnalyzeCollectPath
+ * Extract a single path from JSON documents and collect its values.
+ */
+static void
+jsonAnalyzeCollectPath(JsonAnalyzeContext *ctx, Jsonb *jb, void *param)
+{
+ JsonPathAnlStats *pstats = (JsonPathAnlStats *) param;
+ JsonbValue jbvtmp;
+ JsonbValue *jbv = JsonValueInitBinary(&jbvtmp, jb);
+ JsonPathEntry *path;
+ JsonPathEntry **entries;
+ int i;
+
+ entries = palloc(sizeof(*entries) * pstats->path.depth);
+
+ /* Build entry array in direct order */
+ for (path = &pstats->path, i = pstats->path.depth - 1;
+ path->parent && i >= 0;
+ path = path->parent, i--)
+ entries[i] = path;
+
+ jsonAnalyzeCollectSubpath(ctx, pstats, jbv, entries, 0);
+
+ pfree(entries);
+}
+
+static Datum
+jsonAnalyzePathFetch(VacAttrStatsP stats, int rownum, bool *isnull)
+{
+ *isnull = false;
+ return stats->exprvals[rownum];
+}
+
+/*
+ * jsonAnalyzePathValues
+ * Calculate per-column statistics for values for a single path.
+ *
+ * We have already accumulated all the values for the path, so we simply
+ * call the typanalyze function for the proper data type, and then
+ * compute_stats (which points to compute_scalar_stats or so).
+ */
+static void
+jsonAnalyzePathValues(JsonAnalyzeContext *ctx, JsonScalarStats *sstats,
+ Oid typid, double freq, bool use_anl_context)
+{
+ JsonValues *values = &sstats->values;
+ VacAttrStats *stats = &sstats->stats;
+ FormData_pg_attribute attr;
+ FormData_pg_type type;
+ int i;
+
+ if (!sstats->values.count)
+ return;
+
+ get_typlenbyvalalign(typid, &type.typlen, &type.typbyval, &type.typalign);
+
+ attr.attstattarget = ctx->target;
+
+ stats->attr = &attr;
+ stats->attrtypid = typid;
+ stats->attrtypmod = -1;
+ stats->attrtype = &type;
+ stats->anl_context = use_anl_context ? ctx->stats->anl_context : CurrentMemoryContext;
+
+ stats->exprvals = values->buf;
+
+ /*
+ * The fields describing the stats->stavalues[n] element types default to
+ * the type of the data being analyzed, but the type-specific typanalyze
+ * function can change them if it wants to store something else.
+ */
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ stats->statypid[i] = stats->attrtypid;
+ stats->statyplen[i] = stats->attrtype->typlen;
+ stats->statypbyval[i] = stats->attrtype->typbyval;
+ stats->statypalign[i] = stats->attrtype->typalign;
+ }
+
+ std_typanalyze(stats);
+
+ stats->compute_stats(stats, jsonAnalyzePathFetch,
+ values->count,
+ ctx->totalrows / ctx->samplerows * values->count);
+
+ /*
+ * We've only kept the non-null values, so compute_stats will always
+ * leave this as 1.0. But we have enough info to calculate the correct
+ * value.
+ */
+ stats->stanullfrac = (float4)(1.0 - freq);
+
+ /*
+ * Similarly, we need to correct the MCV frequencies, becuse those are
+ * also calculated only from the non-null values. All we need to do is
+ * simply multiply that with the non-NULL frequency.
+ */
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ if (stats->stakind[i] == STATISTIC_KIND_MCV)
+ {
+ int j;
+ for (j = 0; j < stats->numnumbers[i]; j++)
+ stats->stanumbers[i][j] *= freq;
+ }
+ }
+}
+
+/*
+ * jsonAnalyzeMakeScalarStats
+ * Serialize scalar stats into a JSON representation.
+ *
+ * We simply produce a JSON document with a list of predefined keys:
+ *
+ * - nullfrac
+ * - distinct
+ * - width
+ * - correlation
+ * - mcv or histogram
+ *
+ * For the mcv / histogram, we store a nested values / numbers.
+ */
+static JsonbValue *
+jsonAnalyzeMakeScalarStats(JsonbParseState **ps, const char *name,
+ const VacAttrStats *stats)
+{
+ JsonbValue val;
+ int i;
+ int j;
+
+ pushJsonbKey(ps, &val, name);
+
+ pushJsonbValue(ps, WJB_BEGIN_OBJECT, NULL);
+
+ pushJsonbKeyValueFloat(ps, &val, "nullfrac", stats->stanullfrac);
+ pushJsonbKeyValueFloat(ps, &val, "distinct", stats->stadistinct);
+ pushJsonbKeyValueInteger(ps, &val, "width", stats->stawidth);
+
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ if (!stats->stakind[i])
+ break;
+
+ switch (stats->stakind[i])
+ {
+ case STATISTIC_KIND_MCV:
+ pushJsonbKey(ps, &val, "mcv");
+ break;
+
+ case STATISTIC_KIND_HISTOGRAM:
+ pushJsonbKey(ps, &val, "histogram");
+ break;
+
+ case STATISTIC_KIND_CORRELATION:
+ pushJsonbKeyValueFloat(ps, &val, "correlation",
+ stats->stanumbers[i][0]);
+ continue;
+
+ default:
+ elog(ERROR, "unexpected stakind %d", stats->stakind[i]);
+ break;
+ }
+
+ pushJsonbValue(ps, WJB_BEGIN_OBJECT, NULL);
+
+ if (stats->numvalues[i] > 0)
+ {
+ pushJsonbKey(ps, &val, "values");
+ pushJsonbValue(ps, WJB_BEGIN_ARRAY, NULL);
+ for (j = 0; j < stats->numvalues[i]; j++)
+ {
+ Datum v = stats->stavalues[i][j];
+ if (stats->attrtypid == JSONBOID)
+ pushJsonbElemBinary(ps, &val, DatumGetJsonbP(v));
+ else if (stats->attrtypid == TEXTOID)
+ pushJsonbElemText(ps, &val, DatumGetTextP(v));
+ else if (stats->attrtypid == NUMERICOID)
+ pushJsonbElemNumeric(ps, &val, DatumGetNumeric(v));
+ else if (stats->attrtypid == INT4OID)
+ pushJsonbElemInteger(ps, &val, DatumGetInt32(v));
+ else
+ elog(ERROR, "unexpected stat value type %d",
+ stats->attrtypid);
+ }
+ pushJsonbValue(ps, WJB_END_ARRAY, NULL);
+ }
+
+ if (stats->numnumbers[i] > 0)
+ {
+ pushJsonbKey(ps, &val, "numbers");
+ pushJsonbValue(ps, WJB_BEGIN_ARRAY, NULL);
+ for (j = 0; j < stats->numnumbers[i]; j++)
+ pushJsonbElemFloat(ps, &val, stats->stanumbers[i][j]);
+ pushJsonbValue(ps, WJB_END_ARRAY, NULL);
+ }
+
+ pushJsonbValue(ps, WJB_END_OBJECT, NULL);
+ }
+
+ return pushJsonbValue(ps, WJB_END_OBJECT, NULL);
+}
+
+static void
+pushJsonbKeyValueFloatNonZero(JsonbParseState **ps, JsonbValue *jbv,
+ const char *field, double val)
+{
+ if (val != 0.0)
+ pushJsonbKeyValueFloat(ps, jbv, field, val);
+}
+
+/*
+ * jsonAnalyzeBuildPathStats
+ * Serialize statistics for a particular json path.
+ *
+ * This includes both the per-column stats (stored in "json" key) and the
+ * JSON specific stats (like frequencies of different object types).
+ */
+static Jsonb *
+jsonAnalyzeBuildPathStats(JsonPathAnlStats *pstats)
+{
+ const JsonValueStats *vstats = &pstats->vstats;
+ float4 freq = pstats->freq;
+ bool fullstats = true; /* pstats->path.parent != NULL */
+ JsonbValue val;
+ JsonbValue *jbv;
+ JsonbParseState *ps = NULL;
+
+ pushJsonbValue(&ps, WJB_BEGIN_OBJECT, NULL);
+
+ pushJsonbKeyValueString(&ps, &val, "path", pstats->path.pathstr);
+
+ pushJsonbKeyValueFloat(&ps, &val, "freq", freq);
+
+ pushJsonbKeyValueFloatNonZero(&ps, &val, "freq_null",
+ freq * vstats->nnulls /
+ vstats->jsons.values.count);
+
+ pushJsonbKeyValueFloatNonZero(&ps, &val, "freq_boolean",
+ freq * (vstats->nfalse + vstats->ntrue) /
+ vstats->jsons.values.count);
+
+ pushJsonbKeyValueFloatNonZero(&ps, &val, "freq_string",
+ freq * vstats->nstrings /
+ vstats->jsons.values.count);
+
+ pushJsonbKeyValueFloatNonZero(&ps, &val, "freq_numeric",
+ freq * vstats->nnumerics /
+ vstats->jsons.values.count);
+
+ pushJsonbKeyValueFloatNonZero(&ps, &val, "freq_array",
+ freq * vstats->narrays /
+ vstats->jsons.values.count);
+
+ pushJsonbKeyValueFloatNonZero(&ps, &val, "freq_object",
+ freq * vstats->nobjects /
+ vstats->jsons.values.count);
+
+ /*
+ * We keep array length stats here for queries like jsonpath '$.size() > 5'.
+ * Object lengths stats can be useful for other query lanuages.
+ */
+ if (vstats->arrlens.values.count)
+ jsonAnalyzeMakeScalarStats(&ps, "array_length", &vstats->arrlens.stats);
+
+ if (vstats->objlens.values.count)
+ jsonAnalyzeMakeScalarStats(&ps, "object_length", &vstats->objlens.stats);
+
+ if (vstats->narrays)
+ pushJsonbKeyValueFloat(&ps, &val, "avg_array_length",
+ (float4) vstats->narrelems / vstats->narrays);
+
+ if (fullstats)
+ {
+#ifdef JSON_ANALYZE_SCALARS
+ jsonAnalyzeMakeScalarStats(&ps, "string", &vstats->strings.stats);
+ jsonAnalyzeMakeScalarStats(&ps, "numeric", &vstats->numerics.stats);
+#endif
+ jsonAnalyzeMakeScalarStats(&ps, "json", &vstats->jsons.stats);
+ }
+
+ jbv = pushJsonbValue(&ps, WJB_END_OBJECT, NULL);
+
+ return JsonbValueToJsonb(jbv);
+}
+
+/*
+ * jsonAnalyzeCalcPathFreq
+ * Calculate path frequency, i.e. how many documents contain this path.
+ */
+static void
+jsonAnalyzeCalcPathFreq(JsonAnalyzeContext *ctx, JsonPathAnlStats *pstats,
+ JsonPathParentStats *parent)
+{
+ if (pstats->path.parent)
+ {
+ int count = JsonPathEntryIsArray(&pstats->path) ?
+ parent->narrays : pstats->vstats.jsons.values.count;
+
+ pstats->freq = parent->freq * count / parent->count;
+
+ CLAMP_PROBABILITY(pstats->freq);
+ }
+ else
+ pstats->freq = (double) ctx->analyzed_cnt / ctx->samplerows;
+}
+
+/*
+ * jsonAnalyzePath
+ * Build statistics for values accumulated for this path.
+ *
+ * We're done with accumulating values for this path, so calculate the
+ * statistics for the various arrays.
+ *
+ * XXX I wonder if we could introduce some simple heuristict on which
+ * paths to keep, similarly to what we do for MCV lists. For example a
+ * path that occurred just once is not very interesting, so we could
+ * decide to ignore it and not build the stats. Although that won't
+ * save much, because there'll be very few values accumulated.
+ */
+static Jsonb *
+jsonAnalyzePath(JsonAnalyzeContext *ctx, JsonPathAnlStats *pstats,
+ JsonPathParentStats *parent_stats)
+{
+ JsonValueStats *vstats = &pstats->vstats;
+ Jsonb *stats;
+
+ jsonAnalyzeCalcPathFreq(ctx, pstats, parent_stats);
+
+ /* values combining all object types */
+ jsonAnalyzePathValues(ctx, &vstats->jsons, JSONBOID, pstats->freq,
+ /* store root stats in analyze context */
+ !parent_stats);
+
+ /*
+ * Lengths and array lengths. We divide counts by the total number of json
+ * values to compute correct nullfrac (i.e. not all jsons have lengths).
+ */
+ jsonAnalyzePathValues(ctx, &vstats->arrlens, INT4OID,
+ pstats->freq * vstats->arrlens.values.count /
+ vstats->jsons.values.count, false);
+ jsonAnalyzePathValues(ctx, &vstats->objlens, INT4OID,
+ pstats->freq * vstats->objlens.values.count /
+ vstats->jsons.values.count, false);
+
+#ifdef JSON_ANALYZE_SCALARS
+ /* stats for values of string/numeric types only */
+ jsonAnalyzePathValues(ctx, &vstats->strings, TEXTOID, pstats->freq, false);
+ jsonAnalyzePathValues(ctx, &vstats->numerics, NUMERICOID, pstats->freq, false);
+#endif
+
+ /* Build jsonb with path stats */
+ stats = jsonAnalyzeBuildPathStats(pstats);
+
+ /* Copy stats to non-temporary context */
+ return memcpy(MemoryContextAlloc(ctx->stats->anl_context, VARSIZE(stats)),
+ stats, VARSIZE(stats));
+}
+
+/*
+ * JsonPathStatsCompare
+ * Compare two path stats (by path string).
+ *
+ * We store the stats sorted by path string, and this is the comparator.
+ */
+static int
+JsonPathStatsCompare(const void *pv1, const void *pv2)
+{
+ return strcmp((*((const JsonPathEntry **) pv1))->pathstr,
+ (*((const JsonPathEntry **) pv2))->pathstr);
+}
+
+/*
+ * jsonAnalyzeSortPaths
+ * Reads all stats stored in the hash table and sorts them.
+ */
+static JsonPathEntry **
+jsonAnalyzeSortPaths(JsonAnalyzeContext *ctx, int *p_npaths)
+{
+ HASH_SEQ_STATUS hseq;
+ JsonPathEntry *path;
+ JsonPathEntry **paths;
+ int npaths;
+
+ npaths = hash_get_num_entries(ctx->pathshash) + 1;
+ paths = MemoryContextAlloc(ctx->mcxt, sizeof(*paths) * npaths);
+
+ paths[0] = &ctx->root->path;
+
+ hash_seq_init(&hseq, ctx->pathshash);
+
+ for (int i = 1; (path = hash_seq_search(&hseq)) != NULL; i++)
+ paths[i] = path;
+
+ pg_qsort(paths, npaths, sizeof(*paths), JsonPathStatsCompare);
+
+ *p_npaths = npaths;
+ return paths;
+}
+
+/*
+ * jsonAnalyzeBuildPathStatsArray
+ * Build jsonb datum array for path stats, that will be used as stavalues.
+ *
+ * The first element is a path prefix.
+ */
+static Datum *
+jsonAnalyzeBuildPathStatsArray(Jsonb **pstats, int npaths, int *nvals,
+ const char *prefix, int prefixlen)
+{
+ Datum *values = palloc(sizeof(Datum) * (npaths + 1));
+ JsonbValue *jbvprefix = palloc(sizeof(JsonbValue));
+ int i;
+
+ JsonValueInitStringWithLen(jbvprefix,
+ memcpy(palloc(prefixlen), prefix, prefixlen),
+ prefixlen);
+
+ values[0] = JsonbPGetDatum(JsonbValueToJsonb(jbvprefix));
+
+ for (i = 0; i < npaths; i++)
+ values[i + 1] = JsonbPGetDatum(pstats[i]);
+
+ *nvals = npaths + 1;
+
+ return values;
+}
+
+/*
+ * jsonAnalyzeMakeStats
+ * Build stavalues jsonb array for the root path prefix.
+ */
+static Datum *
+jsonAnalyzeMakeStats(JsonAnalyzeContext *ctx, Jsonb **paths,
+ int npaths, int *numvalues)
+{
+ Datum *values;
+ MemoryContext oldcxt = MemoryContextSwitchTo(ctx->stats->anl_context);
+
+ values = jsonAnalyzeBuildPathStatsArray(paths, npaths, numvalues,
+ JSON_PATH_ROOT, JSON_PATH_ROOT_LEN);
+
+ MemoryContextSwitchTo(oldcxt);
+
+ return values;
+}
+
+/*
+ * jsonAnalyzeBuildSubPathsData
+ * Build statvalues and stanumbers arrays for the subset of paths starting
+ * from a given prefix.
+ *
+ * pathsDatums[index] should point to the desired path.
+ */
+bool
+jsonAnalyzeBuildSubPathsData(Datum *pathsDatums, int npaths, int index,
+ const char *path, int pathlen,
+ bool includeSubpaths, float4 nullfrac,
+ Datum *pvals, Datum *pnums)
+{
+ Jsonb **pvalues = palloc(sizeof(*pvalues) * npaths);
+ Datum *values;
+ Datum numbers[1];
+ JsonbValue pathkey;
+ int nsubpaths = 0;
+ int nvalues;
+ int i;
+
+ JsonValueInitStringWithLen(&pathkey, "path", 4);
+
+ for (i = index; i < npaths; i++)
+ {
+ /* Extract path name */
+ Jsonb *jb = DatumGetJsonbP(pathsDatums[i]);
+ JsonbValue *jbv = findJsonbValueFromContainer(&jb->root, JB_FOBJECT,
+ &pathkey);
+
+ /* Check if path name starts with a given prefix */
+ if (!jbv || jbv->type != jbvString ||
+ jbv->val.string.len < pathlen ||
+ memcmp(jbv->val.string.val, path, pathlen))
+ break;
+
+ pfree(jbv);
+
+ /* Collect matching path */
+ pvalues[nsubpaths] = jb;
+
+ nsubpaths++;
+
+ /*
+ * The path should go before its subpaths, so if subpaths are not
+ * needed the loop is broken after the first matching path.
+ */
+ if (!includeSubpaths)
+ break;
+ }
+
+ if (!nsubpaths)
+ {
+ pfree(pvalues);
+ return false;
+ }
+
+ /* Construct new array from the selected paths */
+ values = jsonAnalyzeBuildPathStatsArray(pvalues, nsubpaths, &nvalues,
+ path, pathlen);
+ *pvals = PointerGetDatum(construct_array(values, nvalues, JSONBOID, -1,
+ false, 'i'));
+
+ pfree(pvalues);
+ pfree(values);
+
+ numbers[0] = Float4GetDatum(nullfrac);
+ *pnums = PointerGetDatum(construct_array(numbers, 1, FLOAT4OID, 4,
+ true /*FLOAT4PASSBYVAL*/, 'i'));
+
+ return true;
+}
+
+/*
+ * jsonAnalyzeInit
+ * Initialize the analyze context so that we can start adding paths.
+ */
+static void
+jsonAnalyzeInit(JsonAnalyzeContext *ctx, VacAttrStats *stats,
+ AnalyzeAttrFetchFunc fetchfunc,
+ int samplerows, double totalrows, bool single_pass)
+{
+ HASHCTL hash_ctl;
+
+ memset(ctx, 0, sizeof(*ctx));
+
+ ctx->stats = stats;
+ ctx->fetchfunc = fetchfunc;
+ ctx->mcxt = CurrentMemoryContext;
+ ctx->samplerows = samplerows;
+ ctx->totalrows = totalrows;
+ ctx->target = stats->attr->attstattarget;
+ ctx->scalarsOnly = false;
+ ctx->single_pass = single_pass;
+
+ MemSet(&hash_ctl, 0, sizeof(hash_ctl));
+ hash_ctl.keysize = sizeof(JsonPathEntry);
+ hash_ctl.entrysize = ctx->single_pass ? sizeof(JsonPathAnlStats) : sizeof(JsonPathAnlDocs);
+ hash_ctl.hash = JsonPathEntryHash;
+ hash_ctl.match = JsonPathEntryMatch;
+ hash_ctl.hcxt = ctx->mcxt;
+
+ ctx->pathshash = hash_create("JSON analyze path table", 100, &hash_ctl,
+ HASH_ELEM | HASH_FUNCTION | HASH_COMPARE | HASH_CONTEXT);
+
+ ctx->root = MemoryContextAllocZero(ctx->mcxt, sizeof(JsonPathAnlStats));
+ ctx->root->path.pathstr = JSON_PATH_ROOT;
+}
+
+/*
+ * jsonAnalyzePass
+ * One analysis pass over the JSON column.
+ *
+ * Performs one analysis pass on the JSON documents, and passes them to the
+ * custom analyzefunc.
+ */
+static void
+jsonAnalyzePass(JsonAnalyzeContext *ctx,
+ void (*analyzefunc)(JsonAnalyzeContext *, Jsonb *, void *),
+ void *analyzearg,
+ JsonPathDocBitmap *bitmap)
+{
+ MemoryContext tmpcxt = AllocSetContextCreate(CurrentMemoryContext,
+ "Json Analyze Pass Context",
+ ALLOCSET_DEFAULT_MINSIZE,
+ ALLOCSET_DEFAULT_INITSIZE,
+ ALLOCSET_DEFAULT_MAXSIZE);
+ MemoryContext oldcxt = MemoryContextSwitchTo(tmpcxt);
+ int row_num = -1;
+
+ ctx->null_cnt = 0;
+ ctx->analyzed_cnt = 0;
+ ctx->total_width = 0;
+
+ /* Loop over the jsonbs. */
+ for (int i = 0; i < (bitmap ? bitmap->size : ctx->samplerows); i++)
+ {
+ Datum value;
+ Jsonb *jb;
+ Size width;
+ bool isnull;
+
+ vacuum_delay_point();
+
+ if (bitmap)
+ {
+ if (bitmap->is_list)
+ row_num = bitmap->data.list[i];
+ else if (!jsonStatsBitmapNext(bitmap, &row_num))
+ break;
+ }
+ else
+ row_num = i;
+
+ value = ctx->fetchfunc(ctx->stats, row_num, &isnull);
+
+ if (isnull)
+ {
+ /* json is null, just count that */
+ ctx->null_cnt++;
+ continue;
+ }
+
+ width = toast_raw_datum_size(value);
+
+ ctx->total_width += VARSIZE_ANY(DatumGetPointer(value)); /* FIXME raw width? */
+
+ /* Skip too-large values. */
+#define JSON_WIDTH_THRESHOLD (100 * 1024)
+
+ if (width > JSON_WIDTH_THRESHOLD)
+ continue;
+
+ ctx->analyzed_cnt++;
+
+ jb = DatumGetJsonbP(value);
+
+ if (!ctx->single_pass)
+ MemoryContextSwitchTo(oldcxt);
+
+ ctx->current_rownum = row_num;
+ analyzefunc(ctx, jb, analyzearg);
+
+ if (!ctx->single_pass)
+ oldcxt = MemoryContextSwitchTo(tmpcxt);
+
+ MemoryContextReset(tmpcxt);
+ }
+
+ MemoryContextSwitchTo(oldcxt);
+}
+
+/*
+ * compute_json_stats() -- compute statistics for a json column
+ */
+static void
+compute_json_stats(VacAttrStats *stats, AnalyzeAttrFetchFunc fetchfunc,
+ int samplerows, double totalrows)
+{
+ JsonAnalyzeContext ctx;
+ JsonPathEntry **paths;
+ Jsonb **pstats;
+ int npaths;
+ int root_analyzed_cnt;
+ int root_null_cnt;
+ double root_total_width;
+
+ jsonAnalyzeInit(&ctx, stats, fetchfunc, samplerows, totalrows,
+ false /* FIXME make GUC or simply remove */);
+
+ /*
+ * Collect and analyze JSON path values in single or multiple passes.
+ * Sigle-pass collection is faster but consumes much more memory than
+ * collecting and analyzing by the one path at pass.
+ */
+ if (ctx.single_pass)
+ {
+ /* Collect all values of all paths */
+ jsonAnalyzePass(&ctx, jsonAnalyzeCollectPaths, (void *)(intptr_t) true, NULL);
+
+ root_analyzed_cnt = ctx.analyzed_cnt;
+ root_null_cnt = ctx.null_cnt;
+ root_total_width = ctx.total_width;
+
+ /*
+ * Now that we're done with processing the documents, we sort the paths
+ * we extracted and calculate stats for each of them.
+ *
+ * XXX I wonder if we could do this in two phases, to maybe not collect
+ * (or even accumulate) values for paths that are not interesting.
+ */
+ paths = jsonAnalyzeSortPaths(&ctx, &npaths);
+ pstats = palloc(sizeof(*pstats) * npaths);
+
+ for (int i = 0; i < npaths; i++)
+ {
+ JsonPathAnlStats *astats = (JsonPathAnlStats *) paths[i];
+ JsonPathAnlStats *parent = (JsonPathAnlStats *) paths[i]->parent;
+ JsonPathParentStats parent_stats;
+
+ if (parent)
+ {
+ parent_stats.freq = parent->freq;
+ parent_stats.count = parent->vstats.jsons.values.count;
+ parent_stats.narrays = parent->vstats.narrays;
+ }
+
+ pstats[i] = jsonAnalyzePath(&ctx, astats,
+ parent ? &parent_stats : NULL);
+ }
+ }
+ else
+ {
+ MemoryContext oldcxt;
+ MemoryContext tmpcxt = AllocSetContextCreate(CurrentMemoryContext,
+ "Json Analyze Tmp Context",
+ ALLOCSET_DEFAULT_MINSIZE,
+ ALLOCSET_DEFAULT_INITSIZE,
+ ALLOCSET_DEFAULT_MAXSIZE);
+ JsonPathParentStats *stack;
+
+ elog(DEBUG1, "analyzing %s attribute \"%s\"",
+ stats->attrtypid == JSONBOID ? "jsonb" : "json",
+ NameStr(stats->attr->attname));
+
+ elog(DEBUG1, "collecting json paths");
+
+ oldcxt = MemoryContextSwitchTo(tmpcxt);
+
+ /* Collect all paths first without accumulating any Values, sort them */
+ jsonAnalyzePass(&ctx, jsonAnalyzeCollectPaths, (void *)(intptr_t) false, NULL);
+ paths = jsonAnalyzeSortPaths(&ctx, &npaths);
+ pstats = MemoryContextAlloc(oldcxt, sizeof(*pstats) * npaths);
+ stack = MemoryContextAlloc(oldcxt, sizeof(*stack) * (ctx.maxdepth + 1));
+
+ root_analyzed_cnt = ctx.analyzed_cnt;
+ root_null_cnt = ctx.null_cnt;
+ root_total_width = ctx.total_width;
+
+ /*
+ * Next, process each path independently to save memory (we don't want
+ * to accumulate all values for all paths, with a lot of duplicities).
+ */
+ MemoryContextReset(tmpcxt);
+
+ for (int i = 0; i < npaths; i++)
+ {
+ JsonPathEntry *path = paths[i];
+ JsonPathAnlStats astats_tmp;
+ JsonPathAnlStats *astats;
+
+ if (!i)
+ astats = ctx.root;
+ else
+ {
+ astats = &astats_tmp;
+ jsonStatsAnlInit(astats);
+ astats->path = *path;
+ }
+
+ elog(DEBUG1, "analyzing json path (%d/%d) %s",
+ i + 1, npaths, path->pathstr);
+
+ jsonAnalyzePass(&ctx, jsonAnalyzeCollectPath, astats,
+ /* root has no bitmap */
+ i > 0 ? &((JsonPathAnlDocs *) path)->bitmap : NULL);
+
+ pstats[i] = jsonAnalyzePath(&ctx, astats,
+ path->depth ? &stack[path->depth - 1] : NULL);
+
+ /* Save parent stats in the stack */
+ stack[path->depth].freq = astats->freq;
+ stack[path->depth].count = astats->vstats.jsons.values.count;
+ stack[path->depth].narrays = astats->vstats.narrays;
+
+ MemoryContextReset(tmpcxt);
+ }
+
+ MemoryContextSwitchTo(oldcxt);
+
+ MemoryContextDelete(tmpcxt);
+ }
+
+ /* We can only compute real stats if we found some non-null values. */
+ if (root_null_cnt >= samplerows)
+ {
+ /* We found only nulls; assume the column is entirely null */
+ stats->stats_valid = true;
+ stats->stanullfrac = 1.0;
+ stats->stawidth = 0; /* "unknown" */
+ stats->stadistinct = 0.0; /* "unknown" */
+ }
+ else if (!root_analyzed_cnt)
+ {
+ int nonnull_cnt = samplerows - root_null_cnt;
+
+ /* We found some non-null values, but they were all too wide */
+ stats->stats_valid = true;
+ /* Do the simple null-frac and width stats */
+ stats->stanullfrac = (double) root_null_cnt / (double) samplerows;
+ stats->stawidth = root_total_width / (double) nonnull_cnt;
+ /* Assume all too-wide values are distinct, so it's a unique column */
+ stats->stadistinct = -1.0 * (1.0 - stats->stanullfrac);
+ }
+ else
+ {
+ VacAttrStats *jsstats = &ctx.root->vstats.jsons.stats;
+ int i;
+ int empty_slot = -1;
+
+ stats->stats_valid = true;
+
+ stats->stanullfrac = jsstats->stanullfrac;
+ stats->stawidth = jsstats->stawidth;
+ stats->stadistinct = jsstats->stadistinct;
+
+ /*
+ * We need to store the statistics the statistics slots. We simply
+ * store the regular stats in the first slots, and then we put the
+ * JSON stats into the first empty slot.
+ */
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ /* once we hit an empty slot, we're done */
+ if (!jsstats->staop[i])
+ {
+ empty_slot = i; /* remember the empty slot */
+ break;
+ }
+
+ stats->stakind[i] = jsstats->stakind[i];
+ stats->staop[i] = jsstats->staop[i];
+ stats->stanumbers[i] = jsstats->stanumbers[i];
+ stats->stavalues[i] = jsstats->stavalues[i];
+ stats->statypid[i] = jsstats->statypid[i];
+ stats->statyplen[i] = jsstats->statyplen[i];
+ stats->statypbyval[i] = jsstats->statypbyval[i];
+ stats->statypalign[i] = jsstats->statypalign[i];
+ stats->numnumbers[i] = jsstats->numnumbers[i];
+ stats->numvalues[i] = jsstats->numvalues[i];
+ }
+
+ Assert((empty_slot >= 0) && (empty_slot < STATISTIC_NUM_SLOTS));
+
+ stats->stakind[empty_slot] = STATISTIC_KIND_JSON;
+ stats->staop[empty_slot] = InvalidOid;
+ stats->numnumbers[empty_slot] = 1;
+ stats->stanumbers[empty_slot] = MemoryContextAlloc(stats->anl_context,
+ sizeof(float4));
+ stats->stanumbers[empty_slot][0] = 0.0; /* nullfrac */
+ stats->stavalues[empty_slot] =
+ jsonAnalyzeMakeStats(&ctx, pstats, npaths,
+ &stats->numvalues[empty_slot]);
+
+ /* We are storing jsonb values */
+ stats->statypid[empty_slot] = JSONBOID;
+ get_typlenbyvalalign(stats->statypid[empty_slot],
+ &stats->statyplen[empty_slot],
+ &stats->statypbyval[empty_slot],
+ &stats->statypalign[empty_slot]);
+ }
+}
+
+/*
+ * json_typanalyze -- typanalyze function for jsonb
+ */
+Datum
+jsonb_typanalyze(PG_FUNCTION_ARGS)
+{
+ VacAttrStats *stats = (VacAttrStats *) PG_GETARG_POINTER(0);
+ Form_pg_attribute attr = stats->attr;
+
+ /* If the attstattarget column is negative, use the default value */
+ /* NB: it is okay to scribble on stats->attr since it's a copy */
+ if (attr->attstattarget < 0)
+ attr->attstattarget = default_statistics_target;
+
+ stats->compute_stats = compute_json_stats;
+ /* see comment about the choice of minrows in commands/analyze.c */
+ stats->minrows = 300 * attr->attstattarget;
+
+ PG_RETURN_BOOL(true);
+}
diff --git a/src/backend/utils/adt/jsonpath_exec.c b/src/backend/utils/adt/jsonpath_exec.c
index eff3734b6ab..94ae6e1f385 100644
--- a/src/backend/utils/adt/jsonpath_exec.c
+++ b/src/backend/utils/adt/jsonpath_exec.c
@@ -1723,7 +1723,7 @@ executeLikeRegex(JsonPathItem *jsp, JsonbValue *str, JsonbValue *rarg,
cxt->cflags = jspConvertRegexFlags(jsp->content.like_regex.flags);
}
- if (RE_compile_and_execute(cxt->regex, str->val.string.val,
+ if (RE_compile_and_execute(cxt->regex, unconstify(char *, str->val.string.val),
str->val.string.len,
cxt->cflags, DEFAULT_COLLATION_OID, 0, NULL))
return jpbTrue;
diff --git a/src/include/catalog/pg_operator.dat b/src/include/catalog/pg_operator.dat
index 8e0e65ad275..9805bb15038 100644
--- a/src/include/catalog/pg_operator.dat
+++ b/src/include/catalog/pg_operator.dat
@@ -3175,7 +3175,7 @@
{ oid => '3211', oid_symbol => 'JsonbObjectFieldOperator',
descr => 'get jsonb object field',
oprname => '->', oprleft => 'jsonb', oprright => 'text', oprresult => 'jsonb',
- oprcode => 'jsonb_object_field' },
+ oprcode => 'jsonb_object_field', oprstat => 'jsonb_stats' },
{ oid => '3477', oid_symbol => 'JsonbObjectFieldTextOperator',
descr => 'get jsonb object field as text',
oprname => '->>', oprleft => 'jsonb', oprright => 'text', oprresult => 'text',
@@ -3183,7 +3183,7 @@
{ oid => '3212', oid_symbol => 'JsonbArrayElementOperator',
descr => 'get jsonb array element',
oprname => '->', oprleft => 'jsonb', oprright => 'int4', oprresult => 'jsonb',
- oprcode => 'jsonb_array_element' },
+ oprcode => 'jsonb_array_element', oprstat => 'jsonb_stats' },
{ oid => '3481', oid_symbol => 'JsonbArrayElementTextOperator',
descr => 'get jsonb array element as text',
oprname => '->>', oprleft => 'jsonb', oprright => 'int4', oprresult => 'text',
@@ -3191,7 +3191,8 @@
{ oid => '3213', oid_symbol => 'JsonbExtractPathOperator',
descr => 'get value from jsonb with path elements',
oprname => '#>', oprleft => 'jsonb', oprright => '_text',
- oprresult => 'jsonb', oprcode => 'jsonb_extract_path' },
+ oprresult => 'jsonb', oprcode => 'jsonb_extract_path',
+ oprstat => 'jsonb_stats' },
{ oid => '3206', oid_symbol => 'JsonbExtractPathTextOperator',
descr => 'get value from jsonb as text with path elements',
oprname => '#>>', oprleft => 'jsonb', oprright => '_text',
@@ -3229,23 +3230,23 @@
{ oid => '3246', oid_symbol => 'JsonbContainsOperator', descr => 'contains',
oprname => '@>', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '<@(jsonb,jsonb)', oprcode => 'jsonb_contains',
- oprrest => 'matchingsel', oprjoin => 'matchingjoinsel' },
+ oprrest => 'jsonb_sel', oprjoin => 'matchingjoinsel' },
{ oid => '3247', oid_symbol => 'JsonbExistsOperator', descr => 'key exists',
oprname => '?', oprleft => 'jsonb', oprright => 'text', oprresult => 'bool',
- oprcode => 'jsonb_exists', oprrest => 'matchingsel',
+ oprcode => 'jsonb_exists', oprrest => 'jsonb_sel',
oprjoin => 'matchingjoinsel' },
{ oid => '3248', oid_symbol => 'JsonbExistsAnyOperator', descr => 'any key exists',
oprname => '?|', oprleft => 'jsonb', oprright => '_text', oprresult => 'bool',
- oprcode => 'jsonb_exists_any', oprrest => 'matchingsel',
+ oprcode => 'jsonb_exists_any', oprrest => 'jsonb_sel',
oprjoin => 'matchingjoinsel' },
{ oid => '3249', oid_symbol => 'JsonbExistsAllOperator', descr => 'all keys exist',
oprname => '?&', oprleft => 'jsonb', oprright => '_text', oprresult => 'bool',
- oprcode => 'jsonb_exists_all', oprrest => 'matchingsel',
+ oprcode => 'jsonb_exists_all', oprrest => 'jsonb_sel',
oprjoin => 'matchingjoinsel' },
{ oid => '3250', oid_symbol => 'JsonbContainedOperator', descr => 'is contained by',
oprname => '<@', oprleft => 'jsonb', oprright => 'jsonb', oprresult => 'bool',
oprcom => '@>(jsonb,jsonb)', oprcode => 'jsonb_contained',
- oprrest => 'matchingsel', oprjoin => 'matchingjoinsel' },
+ oprrest => 'jsonb_sel', oprjoin => 'matchingjoinsel' },
{ oid => '3284', descr => 'concatenate',
oprname => '||', oprleft => 'jsonb', oprright => 'jsonb',
oprresult => 'jsonb', oprcode => 'jsonb_concat' },
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index d8e8715ed1c..5c8d65aa0d1 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -11747,4 +11747,15 @@
prorettype => 'bytea', proargtypes => 'pg_brin_minmax_multi_summary',
prosrc => 'brin_minmax_multi_summary_send' },
+# jsonb statistics
+{ oid => '8526', descr => 'jsonb typanalyze',
+ proname => 'jsonb_typanalyze', provolatile => 's', prorettype => 'bool',
+ proargtypes => 'internal', prosrc => 'jsonb_typanalyze' },
+{ oid => '8527', descr => 'jsonb selectivity estimation',
+ proname => 'jsonb_sel', provolatile => 's', prorettype => 'float8',
+ proargtypes => 'internal oid internal int4', prosrc => 'jsonb_sel' },
+{ oid => '8528', descr => 'jsonb statsistics estimation',
+ proname => 'jsonb_stats', provolatile => 's', prorettype => 'void',
+ proargtypes => 'internal internal int4 internal', prosrc => 'jsonb_stats' },
+
]
diff --git a/src/include/catalog/pg_statistic.h b/src/include/catalog/pg_statistic.h
index cdf74481398..c4f53ebd1ba 100644
--- a/src/include/catalog/pg_statistic.h
+++ b/src/include/catalog/pg_statistic.h
@@ -277,6 +277,8 @@ DECLARE_FOREIGN_KEY((starelid, staattnum), pg_attribute, (attrelid, attnum));
*/
#define STATISTIC_KIND_BOUNDS_HISTOGRAM 7
+#define STATISTIC_KIND_JSON 8
+
#endif /* EXPOSE_TO_CLIENT_CODE */
#endif /* PG_STATISTIC_H */
diff --git a/src/include/catalog/pg_type.dat b/src/include/catalog/pg_type.dat
index df458794635..b867db42f45 100644
--- a/src/include/catalog/pg_type.dat
+++ b/src/include/catalog/pg_type.dat
@@ -445,7 +445,7 @@
typname => 'jsonb', typlen => '-1', typbyval => 'f', typcategory => 'U',
typsubscript => 'jsonb_subscript_handler', typinput => 'jsonb_in',
typoutput => 'jsonb_out', typreceive => 'jsonb_recv', typsend => 'jsonb_send',
- typalign => 'i', typstorage => 'x' },
+ typanalyze => 'jsonb_typanalyze', typalign => 'i', typstorage => 'x' },
{ oid => '4072', array_type_oid => '4073', descr => 'JSON path',
typname => 'jsonpath', typlen => '-1', typbyval => 'f', typcategory => 'U',
typinput => 'jsonpath_in', typoutput => 'jsonpath_out',
diff --git a/src/include/utils/json_selfuncs.h b/src/include/utils/json_selfuncs.h
new file mode 100644
index 00000000000..9a36567ae65
--- /dev/null
+++ b/src/include/utils/json_selfuncs.h
@@ -0,0 +1,113 @@
+/*-------------------------------------------------------------------------
+ *
+ * json_selfuncs.h
+ * JSON cost estimation functions.
+ *
+ *
+ * Portions Copyright (c) 2016-2022, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ * src/include/utils/json_selfuncs.h
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#ifndef JSON_SELFUNCS_H_
+#define JSON_SELFUNCS_H 1
+
+#include "postgres.h"
+#include "access/htup.h"
+#include "utils/jsonb.h"
+#include "utils/lsyscache.h"
+#include "utils/selfuncs.h"
+
+#define JSON_PATH_ROOT "$"
+#define JSON_PATH_ROOT_LEN 1
+
+#define JSON_PATH_ROOT_ARRAY "$[*]"
+#define JSON_PATH_ROOT_ARRAY_LEN 4
+
+typedef enum
+{
+ JsonPathStatsValues,
+ JsonPathStatsArrayLength,
+ JsonPathStatsObjectLength
+} JsonPathStatsType;
+
+typedef struct JsonStatData JsonStatData, *JsonStats;
+
+/* Per-path JSON stats */
+typedef struct JsonPathStatsData
+{
+ JsonStats data; /* pointer to per-column control structure */
+ Datum *datum; /* pointer to JSONB datum with stats data */
+ const char *path; /* path string, points directly to JSONB data */
+ int pathlen; /* path length */
+ JsonPathStatsType type; /* type of stats (values, lengths etc.) */
+} JsonPathStatsData, *JsonPathStats;
+
+/* Per-column JSON stats */
+struct JsonStatData
+{
+ HeapTuple statsTuple; /* original pg_statistic tuple */
+ AttStatsSlot attslot; /* data extracted from STATISTIC_KIND_JSON
+ * slot of statsTuple */
+ RelOptInfo *rel; /* Relation, or NULL if not identifiable */
+ Datum *pathdatums; /* path JSONB datums */
+ JsonPathStatsData *paths; /* cached paths */
+ int npaths; /* number of paths */
+ float4 nullfrac; /* NULL fraction */
+ const char *prefix; /* global path prefix which needs to be used
+ * for searching in pathdatums */
+ int prefixlen; /* path prefix length */
+ bool acl_ok; /* ACL check is Ok */
+};
+
+typedef enum JsonStatType
+{
+ JsonStatJsonb,
+ JsonStatJsonbWithoutSubpaths,
+ JsonStatText,
+ JsonStatFloat4,
+ JsonStatString,
+ JsonStatNumeric,
+ JsonStatFreq,
+} JsonStatType;
+
+extern bool jsonStatsInit(JsonStats stats, const VariableStatData *vardata);
+extern void jsonStatsRelease(JsonStats data);
+
+extern JsonPathStats jsonStatsGetPathByStr(JsonStats stats,
+ const char *path, int pathlen);
+
+extern JsonPathStats jsonPathStatsGetSubpath(JsonPathStats stats,
+ const char *subpath);
+
+extern bool jsonPathStatsGetNextSubpathStats(JsonPathStats stats,
+ JsonPathStats *keystats,
+ bool keysOnly);
+
+extern JsonPathStats jsonPathStatsGetArrayLengthStats(JsonPathStats pstats);
+extern JsonPathStats jsonPathStatsGetObjectLengthStats(JsonPathStats pstats);
+
+extern float4 jsonPathStatsGetFreq(JsonPathStats pstats, float4 defaultfreq);
+
+extern float4 jsonPathStatsGetTypeFreq(JsonPathStats pstats,
+ JsonbValueType type, float4 defaultfreq);
+
+extern float4 jsonPathStatsGetAvgArraySize(JsonPathStats pstats);
+
+extern Selectivity jsonPathStatsGetArrayIndexSelectivity(JsonPathStats pstats,
+ int index);
+
+extern Selectivity jsonSelectivity(JsonPathStats stats, Datum scalar, Oid oper);
+
+extern void jsonPathAppendEntry(StringInfo path, const char *entry);
+
+extern bool jsonAnalyzeBuildSubPathsData(Datum *pathsDatums,
+ int npaths, int index,
+ const char *path, int pathlen,
+ bool includeSubpaths, float4 nullfrac,
+ Datum *pvals, Datum *pnums);
+
+#endif /* JSON_SELFUNCS_H */
diff --git a/src/test/regress/expected/jsonb_stats.out b/src/test/regress/expected/jsonb_stats.out
new file mode 100644
index 00000000000..c7b1e644099
--- /dev/null
+++ b/src/test/regress/expected/jsonb_stats.out
@@ -0,0 +1,713 @@
+CREATE OR REPLACE FUNCTION explain_jsonb(sql_query text)
+RETURNS TABLE(explain_line json) AS
+$$
+BEGIN
+ RETURN QUERY EXECUTE 'EXPLAIN (ANALYZE, FORMAT json) ' || sql_query;
+END;
+$$ LANGUAGE plpgsql;
+CREATE OR REPLACE FUNCTION get_plan_and_actual_rows(sql_query text)
+RETURNS TABLE(plan integer, actual integer) AS
+$$
+ SELECT
+ (plan->>'Plan Rows')::integer plan,
+ (plan->>'Actual Rows')::integer actual
+ FROM (
+ SELECT explain_jsonb(sql_query) #> '{0,Plan,Plans,0}'
+ ) p(plan)
+$$ LANGUAGE sql;
+CREATE OR REPLACE FUNCTION check_estimate(sql_query text, accuracy real)
+RETURNS boolean AS
+$$
+ SELECT plan BETWEEN actual / (1 + accuracy) AND (actual + 1) * (1 + accuracy)
+ FROM (SELECT * FROM get_plan_and_actual_rows(sql_query)) x
+$$ LANGUAGE sql;
+CREATE OR REPLACE FUNCTION check_estimate2(sql_query text, accuracy real)
+RETURNS TABLE(min integer, max integer) AS
+$$
+ SELECT (actual * (1 - accuracy))::integer, ((actual + 1) * (1 + accuracy))::integer
+ FROM (SELECT * FROM get_plan_and_actual_rows(sql_query)) x
+$$ LANGUAGE sql;
+CREATE TABLE jsonb_stats_test(js jsonb);
+INSERT INTO jsonb_stats_test SELECT NULL FROM generate_series(1, 1000);
+INSERT INTO jsonb_stats_test SELECT 'null' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT 'true' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT 'false' FROM generate_series(1, 500);
+INSERT INTO jsonb_stats_test SELECT '12345' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT (1000 * (i % 10))::text::jsonb FROM generate_series(1, 400) i;
+INSERT INTO jsonb_stats_test SELECT i::text::jsonb FROM generate_series(1, 500) i;
+INSERT INTO jsonb_stats_test SELECT '"foo"' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT format('"bar%s"', i % 10)::jsonb FROM generate_series(1, 400) i;
+INSERT INTO jsonb_stats_test SELECT format('"baz%s"', i)::jsonb FROM generate_series(1, 500) i;
+INSERT INTO jsonb_stats_test SELECT '{}' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', 'bar') FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', 'baz' || (i % 10)) FROM generate_series(1, 300) i;
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', i % 10) FROM generate_series(1, 200) i;
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('"foo \"bar"', i % 10) FROM generate_series(1, 200) i;
+INSERT INTO jsonb_stats_test SELECT '[]' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT '["foo"]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[12345]' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT '[["foo"]]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[{"key": "foo"}]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, "foo"]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, 12345]' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT '[null, ["foo"]]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, {"key": "foo"}]' FROM generate_series(1, 200);
+-- Build random variable-length integer arrays
+SELECT setseed(0.0);
+ setseed
+---------
+
+(1 row)
+
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array())
+FROM generate_series(1, 1000);
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int))
+FROM generate_series(1, 4000);
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int,
+ floor(random() * 10)::int))
+FROM generate_series(1, 3000);
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int,
+ floor(random() * 10)::int,
+ floor(random() * 10)::int))
+FROM generate_series(1, 2000);
+ANALYZE jsonb_stats_test;
+CREATE OR REPLACE FUNCTION check_jsonb_stats_test_estimate(sql_condition text, accuracy real)
+RETURNS boolean AS
+$$
+ SELECT check_estimate('SELECT count(*) FROM jsonb_stats_test WHERE ' || sql_condition, accuracy)
+$$ LANGUAGE sql;
+DROP FUNCTION IF EXISTS check_jsonb_stats_test_estimate2(text, real);
+NOTICE: function check_jsonb_stats_test_estimate2(text,pg_catalog.float4) does not exist, skipping
+CREATE OR REPLACE FUNCTION check_jsonb_stats_test_estimate2(sql_condition text, accuracy real)
+RETURNS TABLE(plan integer, actual integer) AS
+$$
+ SELECT get_plan_and_actual_rows('SELECT count(*) FROM jsonb_stats_test WHERE ' || sql_condition)
+$$ LANGUAGE sql;
+-- Check NULL estimate
+SELECT check_jsonb_stats_test_estimate($$js IS NULL$$, 0.03);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key}' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{1000000}' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key1' -> 'bad_key2' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key1,bad_key2}' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key1' -> 1 IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key1,1}' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 -> 'foo' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{1000000,foo}' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' = '123'$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 = '123'$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check null eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = 'null'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> 'null'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check boolean eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = 'true'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> 'true'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js = 'false'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> 'false'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check numeric eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '12345'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js#>'{}' = '12345'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js = '3000'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js = '1234'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '6000'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check numeric range estimate
+SELECT check_jsonb_stats_test_estimate($$js < '0'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js < '100'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js < '1000'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js < '3456'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js < '10000'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js < '100000'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js > '100' AND js < '600'$$, 0.5);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js > '6800' AND js < '12000'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check string eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '"foo"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js = '"bar7"'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js = '"baz1234"'$$, 10);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '"bar4"'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check string range estimate
+SELECT check_jsonb_stats_test_estimate($$js > '"foo"'$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js > '"bar"'$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js > '"baz"'$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check object eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '{}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js > '{}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check object key eq estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"bar"'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"baz"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"baz5"'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js #> '{foo}' = '"bar"'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check object key range estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' >= '"baz2"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' < '"baz9"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' >= '"baz2"' AND js -> 'foo' < '"baz9"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check array eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '[]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js >= '[]' AND js < '{}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check variable-length array element eq estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 = '1'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 = '6'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 = '8'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 = '1'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check variable-length array element range estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 < '7'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 < '7'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 < '7'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 < '7'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check variable-length array containment estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[]'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1]'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[100]'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2]'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 100]'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2, 100]'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2, 3]'$$, 5);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '1'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '100'$$, 10);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 @> '1'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 @> '1'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 @> '1'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 @> '1'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 @> '[1]'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": []}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1]}'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [100]}'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2]}'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 100]}'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2, 100]}'$$, 1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2, 3]}'$$, 3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- check misc containment
+SELECT check_jsonb_stats_test_estimate($$js @> '"foo"'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '12345'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[12345]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '["foo"]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[["foo", "bar"]]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[["foo"]]'$$, 0.2);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[{"key": "foo"}]'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js @> '[null]'$$, 0.3);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check object key null estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' IS NULL$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' IS NOT NULL$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> '"foo \"bar"' IS NOT NULL$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NOT NULL$$, 0.01);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+-- Check object key existence
+SELECT check_jsonb_stats_test_estimate($$js ? 'bad_key'$$, 10);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ? 'foo'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ? 'array'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ?| '{foo,bad_key}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ?| '{foo,array}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ?& '{foo,bad_key}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
+SELECT check_jsonb_stats_test_estimate($$js ?& '{foo,bar}'$$, 0.1);
+ check_jsonb_stats_test_estimate
+---------------------------------
+ t
+(1 row)
+
diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out
index ac468568a1a..97b6f002c45 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -2504,6 +2504,38 @@ pg_stats_ext_exprs| SELECT cn.nspname AS schemaname,
LEFT JOIN pg_namespace sn ON ((sn.oid = s.stxnamespace)))
JOIN LATERAL ( SELECT unnest(pg_get_statisticsobjdef_expressions(s.oid)) AS expr,
unnest(sd.stxdexpr) AS a) stat ON ((stat.expr IS NOT NULL)));
+pg_stats_json| SELECT n.nspname AS schemaname,
+ c.relname AS tablename,
+ a.attname,
+ (paths.path ->> 'path'::text) AS json_path,
+ s.stainherit AS inherited,
+ (((paths.path -> 'json'::text) ->> 'nullfrac'::text))::real AS null_frac,
+ (((paths.path -> 'json'::text) ->> 'width'::text))::real AS avg_width,
+ (((paths.path -> 'json'::text) ->> 'distinct'::text))::real AS n_distinct,
+ ARRAY( SELECT val.value AS val
+ FROM jsonb_array_elements((((paths.path -> 'json'::text) -> 'mcv'::text) -> 'values'::text)) val(value)) AS most_common_vals,
+ ARRAY( SELECT ((num.value)::text)::real AS num
+ FROM jsonb_array_elements((((paths.path -> 'json'::text) -> 'mcv'::text) -> 'numbers'::text)) num(value)) AS most_common_freqs,
+ ARRAY( SELECT val.value AS val
+ FROM jsonb_array_elements((((paths.path -> 'json'::text) -> 'histogram'::text) -> 'values'::text)) val(value)) AS histogram_bounds,
+ ARRAY( SELECT ((val.value)::text)::integer AS val
+ FROM jsonb_array_elements((((paths.path -> 'array_length'::text) -> 'mcv'::text) -> 'values'::text)) val(value)) AS most_common_array_lengths,
+ ARRAY( SELECT ((num.value)::text)::real AS num
+ FROM jsonb_array_elements((((paths.path -> 'array_length'::text) -> 'mcv'::text) -> 'numbers'::text)) num(value)) AS most_common_array_length_freqs,
+ (((paths.path -> 'json'::text) ->> 'correlation'::text))::real AS correlation
+ FROM (((pg_statistic s
+ JOIN pg_class c ON ((c.oid = s.starelid)))
+ JOIN pg_attribute a ON (((c.oid = a.attrelid) AND (a.attnum = s.staattnum))))
+ LEFT JOIN pg_namespace n ON ((n.oid = c.relnamespace))),
+ LATERAL ( SELECT unnest((((
+ CASE
+ WHEN (s.stakind1 = 8) THEN s.stavalues1
+ WHEN (s.stakind2 = 8) THEN s.stavalues2
+ WHEN (s.stakind3 = 8) THEN s.stavalues3
+ WHEN (s.stakind4 = 8) THEN s.stavalues4
+ WHEN (s.stakind5 = 8) THEN s.stavalues5
+ ELSE NULL::anyarray
+ END)::text)::jsonb[])[2:]) AS path) paths;
pg_tables| SELECT n.nspname AS schemaname,
c.relname AS tablename,
pg_get_userbyid(c.relowner) AS tableowner,
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index 6d8f524ae9e..7117f220ed7 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -111,7 +111,7 @@ test: select_views portals_p2 foreign_key cluster dependency guc bitmapops combo
# ----------
# Another group of parallel tests (JSON related)
# ----------
-test: json jsonb json_encoding jsonpath jsonpath_encoding jsonb_jsonpath
+test: json jsonb json_encoding jsonpath jsonpath_encoding jsonb_jsonpath jsonb_stats
# ----------
# Another group of parallel tests
diff --git a/src/test/regress/sql/jsonb_stats.sql b/src/test/regress/sql/jsonb_stats.sql
new file mode 100644
index 00000000000..fac71d09ded
--- /dev/null
+++ b/src/test/regress/sql/jsonb_stats.sql
@@ -0,0 +1,249 @@
+CREATE OR REPLACE FUNCTION explain_jsonb(sql_query text)
+RETURNS TABLE(explain_line json) AS
+$$
+BEGIN
+ RETURN QUERY EXECUTE 'EXPLAIN (ANALYZE, FORMAT json) ' || sql_query;
+END;
+$$ LANGUAGE plpgsql;
+
+CREATE OR REPLACE FUNCTION get_plan_and_actual_rows(sql_query text)
+RETURNS TABLE(plan integer, actual integer) AS
+$$
+ SELECT
+ (plan->>'Plan Rows')::integer plan,
+ (plan->>'Actual Rows')::integer actual
+ FROM (
+ SELECT explain_jsonb(sql_query) #> '{0,Plan,Plans,0}'
+ ) p(plan)
+$$ LANGUAGE sql;
+
+CREATE OR REPLACE FUNCTION check_estimate(sql_query text, accuracy real)
+RETURNS boolean AS
+$$
+ SELECT plan BETWEEN actual / (1 + accuracy) AND (actual + 1) * (1 + accuracy)
+ FROM (SELECT * FROM get_plan_and_actual_rows(sql_query)) x
+$$ LANGUAGE sql;
+
+CREATE OR REPLACE FUNCTION check_estimate2(sql_query text, accuracy real)
+RETURNS TABLE(min integer, max integer) AS
+$$
+ SELECT (actual * (1 - accuracy))::integer, ((actual + 1) * (1 + accuracy))::integer
+ FROM (SELECT * FROM get_plan_and_actual_rows(sql_query)) x
+$$ LANGUAGE sql;
+
+CREATE TABLE jsonb_stats_test(js jsonb);
+
+INSERT INTO jsonb_stats_test SELECT NULL FROM generate_series(1, 1000);
+
+INSERT INTO jsonb_stats_test SELECT 'null' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT 'true' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT 'false' FROM generate_series(1, 500);
+
+INSERT INTO jsonb_stats_test SELECT '12345' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT (1000 * (i % 10))::text::jsonb FROM generate_series(1, 400) i;
+INSERT INTO jsonb_stats_test SELECT i::text::jsonb FROM generate_series(1, 500) i;
+
+INSERT INTO jsonb_stats_test SELECT '"foo"' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT format('"bar%s"', i % 10)::jsonb FROM generate_series(1, 400) i;
+INSERT INTO jsonb_stats_test SELECT format('"baz%s"', i)::jsonb FROM generate_series(1, 500) i;
+
+INSERT INTO jsonb_stats_test SELECT '{}' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', 'bar') FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', 'baz' || (i % 10)) FROM generate_series(1, 300) i;
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('foo', i % 10) FROM generate_series(1, 200) i;
+INSERT INTO jsonb_stats_test SELECT jsonb_build_object('"foo \"bar"', i % 10) FROM generate_series(1, 200) i;
+
+INSERT INTO jsonb_stats_test SELECT '[]' FROM generate_series(1, 100);
+INSERT INTO jsonb_stats_test SELECT '["foo"]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[12345]' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT '[["foo"]]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[{"key": "foo"}]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, "foo"]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, 12345]' FROM generate_series(1, 300);
+INSERT INTO jsonb_stats_test SELECT '[null, ["foo"]]' FROM generate_series(1, 200);
+INSERT INTO jsonb_stats_test SELECT '[null, {"key": "foo"}]' FROM generate_series(1, 200);
+
+-- Build random variable-length integer arrays
+SELECT setseed(0.0);
+
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array())
+FROM generate_series(1, 1000);
+
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int))
+FROM generate_series(1, 4000);
+
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int,
+ floor(random() * 10)::int))
+FROM generate_series(1, 3000);
+
+INSERT INTO jsonb_stats_test
+SELECT jsonb_build_object('array',
+ jsonb_build_array(
+ floor(random() * 10)::int,
+ floor(random() * 10)::int,
+ floor(random() * 10)::int))
+FROM generate_series(1, 2000);
+
+
+ANALYZE jsonb_stats_test;
+
+CREATE OR REPLACE FUNCTION check_jsonb_stats_test_estimate(sql_condition text, accuracy real)
+RETURNS boolean AS
+$$
+ SELECT check_estimate('SELECT count(*) FROM jsonb_stats_test WHERE ' || sql_condition, accuracy)
+$$ LANGUAGE sql;
+
+DROP FUNCTION IF EXISTS check_jsonb_stats_test_estimate2(text, real);
+
+CREATE OR REPLACE FUNCTION check_jsonb_stats_test_estimate2(sql_condition text, accuracy real)
+RETURNS TABLE(plan integer, actual integer) AS
+$$
+ SELECT get_plan_and_actual_rows('SELECT count(*) FROM jsonb_stats_test WHERE ' || sql_condition)
+$$ LANGUAGE sql;
+
+-- Check NULL estimate
+SELECT check_jsonb_stats_test_estimate($$js IS NULL$$, 0.03);
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key}' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js #> '{1000000}' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key1' -> 'bad_key2' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key1,bad_key2}' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key1' -> 1 IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js #> '{bad_key1,1}' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 -> 'foo' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js #> '{1000000,foo}' IS NULL$$, 0.01);
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' = '123'$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 1000000 = '123'$$, 0.01);
+
+-- Check null eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = 'null'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> 'null'$$, 0.1);
+
+-- Check boolean eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = 'true'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> 'true'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js = 'false'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> 'false'$$, 0.1);
+
+-- Check numeric eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '12345'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js#>'{}' = '12345'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js = '3000'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js = '1234'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js @> '6000'$$, 0.2);
+
+-- Check numeric range estimate
+SELECT check_jsonb_stats_test_estimate($$js < '0'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js < '100'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js < '1000'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js < '3456'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js < '10000'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js < '100000'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js > '100' AND js < '600'$$, 0.5);
+SELECT check_jsonb_stats_test_estimate($$js > '6800' AND js < '12000'$$, 0.1);
+
+-- Check string eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '"foo"'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js = '"bar7"'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js = '"baz1234"'$$, 10);
+SELECT check_jsonb_stats_test_estimate($$js @> '"bar4"'$$, 0.3);
+
+-- Check string range estimate
+SELECT check_jsonb_stats_test_estimate($$js > '"foo"'$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js > '"bar"'$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js > '"baz"'$$, 0.01);
+
+-- Check object eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '{}'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js > '{}'$$, 0.1);
+
+-- Check object key eq estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"bar"'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"baz"'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' = '"baz5"'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js #> '{foo}' = '"bar"'$$, 0.2);
+
+-- Check object key range estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' >= '"baz2"'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' < '"baz9"'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' >= '"baz2"' AND js -> 'foo' < '"baz9"'$$, 0.1);
+
+-- Check array eq estimate
+SELECT check_jsonb_stats_test_estimate($$js = '[]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js >= '[]' AND js < '{}'$$, 0.1);
+
+-- Check variable-length array element eq estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 = '1'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 = '6'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 = '8'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 = '1'$$, 0.2);
+
+-- Check variable-length array element range estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 < '7'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 < '7'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 < '7'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 < '7'$$, 0.1);
+
+-- Check variable-length array containment estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[]'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1]'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[100]'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2]'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 100]'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2, 100]'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '[1, 2, 3]'$$, 5);
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '1'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' @> '100'$$, 10);
+
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 @> '1'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 1 @> '1'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 2 @> '1'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 3 @> '1'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js -> 'array' -> 0 @> '[1]'$$, 0.3);
+
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": []}'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1]}'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [100]}'$$, 0.3);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2]}'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 100]}'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2, 100]}'$$, 1);
+SELECT check_jsonb_stats_test_estimate($$js @> '{"array": [1, 2, 3]}'$$, 3);
+
+-- check misc containment
+SELECT check_jsonb_stats_test_estimate($$js @> '"foo"'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '12345'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '[]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '[12345]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '["foo"]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '[["foo", "bar"]]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '[["foo"]]'$$, 0.2);
+SELECT check_jsonb_stats_test_estimate($$js @> '[{"key": "foo"}]'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js @> '[null]'$$, 0.3);
+
+-- Check object key null estimate
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' IS NULL$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'foo' IS NOT NULL$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> '"foo \"bar"' IS NOT NULL$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NULL$$, 0.01);
+SELECT check_jsonb_stats_test_estimate($$js -> 'bad_key' IS NOT NULL$$, 0.01);
+
+-- Check object key existence
+SELECT check_jsonb_stats_test_estimate($$js ? 'bad_key'$$, 10);
+SELECT check_jsonb_stats_test_estimate($$js ? 'foo'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js ? 'array'$$, 0.1);
+
+SELECT check_jsonb_stats_test_estimate($$js ?| '{foo,bad_key}'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js ?| '{foo,array}'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js ?& '{foo,bad_key}'$$, 0.1);
+SELECT check_jsonb_stats_test_estimate($$js ?& '{foo,bar}'$$, 0.1);
--
2.25.1
^ permalink raw reply [nested|flat] 9+ messages in thread
* Re: Collecting statistics about contents of JSONB columns
@ 2022-05-17 11:44 Mahendra Singh Thalor <[email protected]>
parent: Nikita Glukhov <[email protected]>
0 siblings, 1 reply; 9+ messages in thread
From: Mahendra Singh Thalor @ 2022-05-17 11:44 UTC (permalink / raw)
To: Nikita Glukhov <[email protected]>; +Cc: Tomas Vondra <[email protected]>; PostgreSQL Hackers <[email protected]>; Mahendra Thalor <[email protected]>; Oleg Bartunov <[email protected]>; Thomas Munro <[email protected]>
On Fri, 11 Mar 2022 at 04:29, Nikita Glukhov <[email protected]>
wrote:
>
>
> On 04.02.2022 05:47, Tomas Vondra wrote:
>
> On 1/25/22 17:56, Mahendra Singh Thalor wrote:
> >
>
> ...
>
> For the last few days, I was trying to understand these patches, and
based on Tomas's suggestion, I was doing some performance tests.
>
> With the attached .SQL file, I can see that analyze is taking more time
with these patches.
>
> I haven't found the root cause of this but I feel that this time is due
to a loop of all the paths.
> In my test data, there is a total of 951 different-2 paths. While doing
analysis, first we check all the sample rows(30000) and we collect all the
different-2 paths (here 951), and after that for every single path, we loop
over all the sample rows again to collect stats for a particular path. I
feel that these loops might be taking time.
>
> Thanks, I've been doing some performance tests too, and you're right it
takes quite a bit of time.
>
>
> That is absolutely not surprising, I have warned about poor performance
> in cases with a large number of paths.
>
>
> I agree the slowness is largely due to extracting all paths and then
processing them one by one - which means we have to loop over the tuples
over and over. In this case there's about 850k distinct paths extracted, so
we do ~850k loops over 30k tuples. That's gotta take time.
>
> I don't know what exactly to do about this, but I already mentioned we
may need to pick a subset of paths to keep, similarly to how we pick items
for MCV. I mean, if we only saw a path once or twice, it's unlikely to be
interesting enough to build stats for it. I haven't tried, but I'd bet most
of the 850k paths might be ignored like this.
>
> The other thing we might do is making it the loops more efficient. For
example, we might track which documents contain each path (by a small
bitmap or something), so that in the loop we can skip rows that don't
contain the path we're currently processing. Or something like that.
>
> Apart from this performance issue, I haven't found any crashes or issues.
>
>
> Well, I haven't seen any crashes either, but as I mentioned for complex
documents (2 levels, many distinct keys) the ANALYZE starts consuming a lot
of memory and may get killed by OOM. For example if you generate documents
like this
>
> ./json-generate.py 30000 2 8 1000 6 1000
>
> and then run ANALYZE, that'll take ages and it very quickly gets into a
situation like this (generated from gdb by calling MemoryContextStats on
TopMemoryContext): and then run ANALYZE, that'll take ages and it very
quickly gets into a situation like this (generated from gdb by calling
MemoryContextStats on TopMemoryContext):
>
> -------------------------------------------------------------------------
> TopMemoryContext: 80776 total in 6 blocks; 13992 free (18 chunks); 66784
used
> ...
> TopPortalContext: 8192 total in 1 blocks; 7656 free (0 chunks); 536 used
> PortalContext: 1024 total in 1 blocks; 488 free (0 chunks); 536 used:
<unnamed>
> Analyze: 472726496 total in 150 blocks; 3725776 free (4 chunks);
469000720 used
> Analyze Column: 921177696 total in 120 blocks; 5123256 free (238
chunks); 916054440 used
> Json Analyze Tmp Context: 8192 total in 1 blocks; 5720 free (1
chunks); 2472 used
> Json Analyze Pass Context: 8192 total in 1 blocks; 7928 free
(0 chunks); 264 used
> JSON analyze path table: 1639706040 total in 25084 blocks;
1513640 free (33 chunks); 1638192400 used
> Vacuum: 8192 total in 1 blocks; 7448 free (0 chunks); 744 used
> ...
> Grand total: 3035316184 bytes in 25542 blocks; 10971120 free (352
chunks); 3024345064 used
> -------------------------------------------------------------------------
>
>
> Yes, that's backend 3GB of memory, out of which 1.6GB is in "analyze path
table" context, 400MB in "analyze" and 900MB in "analyze column" contexts.
I mean, that seems a bit excessive. And it grows over time, so after a
while my laptop gives up and kills the backend.
>
> I'm not sure if it's a memory leak (which would be fixable), or it's due
to keeping stats for all the extracted paths. I mean, in this particular
case we have 850k paths - even if stats are just 1kB per path, that's
850MB. This requires more investigation.
>
> Thank you for the tests and investigation.
>
> I have tried to reduce memory consumption and speed up row scanning:
>
> 1. "JSON analyze path table" context contained ~1KB JsonPathAnlStats
> structure per JSON path in the global hash table. I have moved
> JsonPathAnlStats to the stack of compute_json_stats(), and hash
> table now consumes ~70 bytes per path.
>
> 2. I have fixed copying of resulting JSONB stats into context, which
> reduced the size of "Analyze Column" context.
>
> 3. I have optimized consumption of single-pass algorithm by storing
> only value lists in the non-temporary context. That helped to
> execute "2 64 64" test case in 30 seconds. Single-pass is a
> bit faster in non-TOASTed cases, and much faster in TOASTed.
> But it consumes much more memory and still goes to OOM in the
> cases with more than ~100k paths.
>
> 4. I have implemented per-path document lists/bitmaps, which really
> speed up the case "2 8 1000". List is converted into bitmap when
> it becomes larger than bitmap.
>
> 5. Also I have fixed some bugs.
>
>
> All these changes you can find commit form in our GitHub repository
> on the branch jsonb_stats_20220310 [1].
>
>
> Updated results of the test:
>
> levels keys uniq keys paths master multi-pass single-pass
> ms MB ms MB
> -------------------------------------------------------------------
> 1 1 1 2 153 122 10 82 14
> 1 1 1000 1001 134 105 11 78 38
> 1 8 8 9 157 384 19 328 32
> 1 8 1000 1001 155 454 23 402 72
> 1 64 64 65 129 2889 45 2386 155
> 1 64 1000 1001 158 3990 94 1447 177
> 2 1 1 3 237 147 10 91 16
> 2 1 1000 30577 152 264 32 394 234
> 2 8 8 72 245 1943 37 1692 139
> 2 8 1000 852333 152 9175 678 OOM
> 2 64 64 4161 1784 ~1 hour 53 30018 1750
> 2 64 1000 1001001 4715 ~4 hours 1600 OOM
>
> The two last multi-pass results are too slow, because JSONBs becomes
> TOASTed. For measuring master in these tests, I disabled
> WIDTH_THRESHOLD check which skipped TOASTed values > 1KB.
>
>
> Next, I am going to try to disable all-paths collection and implement
> collection of most common paths (and/or hashed paths maybe).
Hi Nikita,
I and Tomas discussed the design for disabling all-paths collection(collect
stats for only some paths). Below are some thoughts/doubts/questions.
*Point 1)* Please can you elaborate more that how are you going to
implement this(collect stats for only some paths).
*Point 2) *As JSON stats are taking time so should we add an on/off switch
to collect JSON stats?
*Point 3)* We thought of one more design: we can give an explicit path to
collect stats for a particular path only or we can pass a subset of the
JSON values but this may require a lot of code changes like syntax and all
so we are thinking that it will be good if we can collect stats only for
some common paths(by limit or any other way)
Thoughts?
--
Thanks and Regards
Mahendra Singh Thalor
EnterpriseDB: http://www.enterprisedb.com
^ permalink raw reply [nested|flat] 9+ messages in thread
* Re: Collecting statistics about contents of JSONB columns
@ 2022-05-17 16:06 Tomas Vondra <[email protected]>
parent: Mahendra Singh Thalor <[email protected]>
0 siblings, 0 replies; 9+ messages in thread
From: Tomas Vondra @ 2022-05-17 16:06 UTC (permalink / raw)
To: Mahendra Singh Thalor <[email protected]>; Nikita Glukhov <[email protected]>; +Cc: PostgreSQL Hackers <[email protected]>; Mahendra Thalor <[email protected]>; Oleg Bartunov <[email protected]>; Thomas Munro <[email protected]>
On 5/17/22 13:44, Mahendra Singh Thalor wrote:
> ...
>
> Hi Nikita,
> I and Tomas discussed the design for disabling all-paths
> collection(collect stats for only some paths). Below are some
> thoughts/doubts/questions.
>
> *Point 1)* Please can you elaborate more that how are you going to
> implement this(collect stats for only some paths).
I think Nikita mentioned he plans to only build stats only for most
common paths, which seems generally straightforward:
1) first pass over the documents, collect distinct paths and track how
many times we saw each one
2) in the second pass extract stats only for the most common paths (e.g.
the top 100 most common ones, or whatever the statistics target says)
I guess we might store at least the frequencing for uncommon paths,
which seems somewhat useful for selectivity estimation.
I wonder if we might further optimize this for less common paths. AFAICS
one of the reasons why we process the paths one by one (in the second
pass) is to limit memory consumption. By processing a single path, we
only need to accumulate values for that path.
But if we know the path is uncommon, we know there'll be few values. For
example the path may be only in 100 documents, not the whole sample. So
maybe we might process multiple paths at once (which would mean we don't
need to detoast the JSON documents that often, etc.).
OTOH that may be pointless, because if the paths are uncommon, chances
are the subsets of documents will be different, in which case it's
probably cheaper to just process the paths one by one.
> *Point 2) *As JSON stats are taking time so should we add an on/off
> switch to collect JSON stats?
IMHO we should think about doing that. I think it's not really possible
to eliminate (significant) regressions for all corner cases, and in many
cases people don't even need this statistics (e.g. when just storing and
retrieving JSON docs, without querying contents of the docs).
I don't know how exactly to enable/disable this - it very much depends
on how we store the stats. If we store that in pg_statistic, then ALTER
TABLE ... ALTER COLUMN seems like the right way to enable/disable these
path stats. We might also have a new "json" stats and do this through
CREATE STATISTICS. Or something else, not sure.
> *Point 3)* We thought of one more design: we can give an explicit path
> to collect stats for a particular path only or we can pass a subset of
> the JSON values but this may require a lot of code changes like syntax
> and all so we are thinking that it will be good if we can collect stats
> only for some common paths(by limit or any other way)
>
I'm not sure I understand what this is saying, particularly the part
about subset of JSON values. Can you elaborate?
I can imagine specifying a list of interesting paths, and we'd only
collect stats for the matching subset of the JSON documents. So if you
have huge JSON documents with complex schema, but you only query a very
limited subset of paths, we could restrict ANALYZE to this subset.
In fact, that's what the 'selective analyze' commit [1] in Nikita's
original patch set does in principle. We'd probably need to improve this
in some ways (e.g. to allow defining the column filter not just in
ANALYZE itself). I left it out of this patch to keep the patch as simple
as possible.
But why/how exactly would we limit the "JSON values"? Can you give some
example demonstrating that in practice?
regards
[1]
https://github.com/postgrespro/postgres/commit/7ab7397450df153e5a8563c978728cb731a0df33
--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
^ permalink raw reply [nested|flat] 9+ messages in thread
end of thread, other threads:[~2022-05-17 16:06 UTC | newest]
Thread overview: 9+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2020-05-11 07:50 [PATCH v38 2/6] Introduce UniqueKey attributes on RelOptInfo struct. 一挃 <[email protected]>
2022-01-23 00:24 Re: Collecting statistics about contents of JSONB columns Nikita Glukhov <[email protected]>
2022-01-24 22:19 ` Re: Collecting statistics about contents of JSONB columns Tomas Vondra <[email protected]>
2022-01-25 16:56 ` Re: Collecting statistics about contents of JSONB columns Mahendra Singh Thalor <[email protected]>
2022-02-04 02:47 ` Re: Collecting statistics about contents of JSONB columns Tomas Vondra <[email protected]>
2022-02-04 03:00 ` Re: Collecting statistics about contents of JSONB columns Tomas Vondra <[email protected]>
2022-03-10 22:58 ` Re: Collecting statistics about contents of JSONB columns Nikita Glukhov <[email protected]>
2022-05-17 11:44 ` Re: Collecting statistics about contents of JSONB columns Mahendra Singh Thalor <[email protected]>
2022-05-17 16:06 ` Re: Collecting statistics about contents of JSONB columns Tomas Vondra <[email protected]>
This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox