v35-0002-Speed-up-searches-for-child-EquivalenceMembers.patch
application/octet-stream
Filename: v35-0002-Speed-up-searches-for-child-EquivalenceMembers.patch
Type: application/octet-stream
Part: 2
Patch
Same data as JSON:
GET /api/v1/attachments/:id/patch
the parsed metadata as JSON — format, series position, per-file stats; never the diff bytes.
API reference →
Format: format-patch
Series: patch v35-0002
Subject: Speed up searches for child EquivalenceMembers
| File | + | − |
|---|---|---|
| contrib/postgres_fdw/postgres_fdw.c | 7 | 8 |
| src/backend/optimizer/path/equivclass.c | 352 | 83 |
| src/backend/optimizer/path/indxpath.c | 9 | 6 |
| src/backend/optimizer/path/pathkeys.c | 6 | 3 |
| src/backend/optimizer/util/relnode.c | 32 | 1 |
| src/include/nodes/pathnodes.h | 79 | 0 |
| src/include/optimizer/paths.h | 6 | 0 |
| src/tools/pgindent/typedefs.list | 2 | 0 |
From e851d7220d45251fb932fe4e1bb874f63705fcf0 Mon Sep 17 00:00:00 2001
From: Yuya Watari <watari.yuya@gmail.com>
Date: Fri, 25 Aug 2023 10:43:36 +0900
Subject: [PATCH v35 2/5] Speed up searches for child EquivalenceMembers
Traditionally, child EquivalenceMembers were in the
EquivalenceClass->ec_members. When we wanted to find some child members
matching a request, we had to perform a linear search. This search
became heavy when tables had many partitions, leading to much planning
time.
After this commit, child EquivalenceMembers no longer exist in
ec_members. Instead, RelOptInfos have them. This change demonstrates a
significant performance improvement in planning time.
---
contrib/postgres_fdw/postgres_fdw.c | 15 +-
src/backend/optimizer/path/equivclass.c | 435 +++++++++++++++++++-----
src/backend/optimizer/path/indxpath.c | 15 +-
src/backend/optimizer/path/pathkeys.c | 9 +-
src/backend/optimizer/util/relnode.c | 33 +-
src/include/nodes/pathnodes.h | 79 +++++
src/include/optimizer/paths.h | 6 +
src/tools/pgindent/typedefs.list | 2 +
8 files changed, 493 insertions(+), 101 deletions(-)
diff --git a/contrib/postgres_fdw/postgres_fdw.c b/contrib/postgres_fdw/postgres_fdw.c
index 6beae0fa37f..adcf6417468 100644
--- a/contrib/postgres_fdw/postgres_fdw.c
+++ b/contrib/postgres_fdw/postgres_fdw.c
@@ -7828,14 +7828,13 @@ conversion_error_callback(void *arg)
EquivalenceMember *
find_em_for_rel(PlannerInfo *root, EquivalenceClass *ec, RelOptInfo *rel)
{
- ListCell *lc;
-
PgFdwRelationInfo *fpinfo = (PgFdwRelationInfo *) rel->fdw_private;
+ EquivalenceMemberIterator it;
+ EquivalenceMember *em;
- foreach(lc, ec->ec_members)
+ setup_eclass_member_iterator_with_children(&it, root, ec, rel->relids);
+ while ((em = eclass_member_iterator_next(&it)) != NULL)
{
- EquivalenceMember *em = (EquivalenceMember *) lfirst(lc);
-
/*
* Note we require !bms_is_empty, else we'd accept constant
* expressions which are not suitable for the purpose.
@@ -7846,6 +7845,7 @@ find_em_for_rel(PlannerInfo *root, EquivalenceClass *ec, RelOptInfo *rel)
is_foreign_expr(root, rel, em->em_expr))
return em;
}
+ dispose_eclass_member_iterator(&it);
return NULL;
}
@@ -7899,9 +7899,8 @@ find_em_for_rel_target(PlannerInfo *root, EquivalenceClass *ec,
if (em->em_is_const)
continue;
- /* Ignore child members */
- if (em->em_is_child)
- continue;
+ /* Child members should not exist in ec_members */
+ Assert(!em->em_is_child);
/* Match if same expression (after stripping relabel) */
em_expr = em->em_expr;
diff --git a/src/backend/optimizer/path/equivclass.c b/src/backend/optimizer/path/equivclass.c
index 7fe7cdff468..8cce165e95d 100644
--- a/src/backend/optimizer/path/equivclass.c
+++ b/src/backend/optimizer/path/equivclass.c
@@ -33,10 +33,14 @@
#include "utils/lsyscache.h"
+static EquivalenceMember *make_eq_member(EquivalenceClass *ec,
+ Expr *expr, Relids relids,
+ JoinDomain *jdomain,
+ EquivalenceMember *parent,
+ Oid datatype);
static EquivalenceMember *add_eq_member(EquivalenceClass *ec,
Expr *expr, Relids relids,
JoinDomain *jdomain,
- EquivalenceMember *parent,
Oid datatype);
static void generate_base_implied_equalities_const(PlannerInfo *root,
EquivalenceClass *ec);
@@ -68,6 +72,10 @@ static bool reconsider_outer_join_clause(PlannerInfo *root,
static bool reconsider_full_join_clause(PlannerInfo *root,
OuterJoinClauseInfo *ojcinfo);
static JoinDomain *find_join_domain(PlannerInfo *root, Relids relids);
+static void add_eclass_child_members_to_iterator(EquivalenceMemberIterator *it,
+ PlannerInfo *root,
+ EquivalenceClass *ec,
+ RelOptInfo *child_rel);
static Bitmapset *get_eclass_indexes_for_relids(PlannerInfo *root,
Relids relids);
static Bitmapset *get_common_eclass_indexes(PlannerInfo *root, Relids relids1,
@@ -268,7 +276,8 @@ process_equivalence(PlannerInfo *root,
{
EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc2);
- Assert(!cur_em->em_is_child); /* no children yet */
+ /* Child members should not exist in ec_members */
+ Assert(!cur_em->em_is_child);
/*
* Match constants only within the same JoinDomain (see
@@ -373,7 +382,7 @@ process_equivalence(PlannerInfo *root,
{
/* Case 3: add item2 to ec1 */
em2 = add_eq_member(ec1, item2, item2_relids,
- jdomain, NULL, item2_type);
+ jdomain, item2_type);
ec1->ec_sources = lappend(ec1->ec_sources, restrictinfo);
ec1->ec_min_security = Min(ec1->ec_min_security,
restrictinfo->security_level);
@@ -390,7 +399,7 @@ process_equivalence(PlannerInfo *root,
{
/* Case 3: add item1 to ec2 */
em1 = add_eq_member(ec2, item1, item1_relids,
- jdomain, NULL, item1_type);
+ jdomain, item1_type);
ec2->ec_sources = lappend(ec2->ec_sources, restrictinfo);
ec2->ec_min_security = Min(ec2->ec_min_security,
restrictinfo->security_level);
@@ -422,9 +431,9 @@ process_equivalence(PlannerInfo *root,
ec->ec_max_security = restrictinfo->security_level;
ec->ec_merged = NULL;
em1 = add_eq_member(ec, item1, item1_relids,
- jdomain, NULL, item1_type);
+ jdomain, item1_type);
em2 = add_eq_member(ec, item2, item2_relids,
- jdomain, NULL, item2_type);
+ jdomain, item2_type);
root->eq_classes = lappend(root->eq_classes, ec);
@@ -510,11 +519,16 @@ canonicalize_ec_expression(Expr *expr, Oid req_type, Oid req_collation)
}
/*
- * add_eq_member - build a new EquivalenceMember and add it to an EC
+ * make_eq_member
+ *
+ * Build a new EquivalenceMember without adding it to an EC. If 'parent'
+ * parameter is NULL, the result will be a parent member, otherwise a child
+ * member. Note that child EquivalenceMembers should not be added to its
+ * parent EquivalenceClass.
*/
static EquivalenceMember *
-add_eq_member(EquivalenceClass *ec, Expr *expr, Relids relids,
- JoinDomain *jdomain, EquivalenceMember *parent, Oid datatype)
+make_eq_member(EquivalenceClass *ec, Expr *expr, Relids relids,
+ JoinDomain *jdomain, EquivalenceMember *parent, Oid datatype)
{
EquivalenceMember *em = makeNode(EquivalenceMember);
@@ -525,6 +539,7 @@ add_eq_member(EquivalenceClass *ec, Expr *expr, Relids relids,
em->em_datatype = datatype;
em->em_jdomain = jdomain;
em->em_parent = parent;
+ em->em_ec = ec;
if (bms_is_empty(relids))
{
@@ -545,11 +560,30 @@ add_eq_member(EquivalenceClass *ec, Expr *expr, Relids relids,
{
ec->ec_relids = bms_add_members(ec->ec_relids, relids);
}
- ec->ec_members = lappend(ec->ec_members, em);
return em;
}
+/*
+ * add_eq_member - build a new parent EquivalenceMember and add it to an EC
+ *
+ * Note: We don't have a function to add a child member like
+ * add_child_eq_member() because how to do it depends on the relations they
+ * are translated from. See add_child_rel_equivalences(),
+ * add_child_join_rel_equivalences() and add_setop_child_rel_equivalences()
+ * to see how to add child members.
+ */
+static EquivalenceMember *
+add_eq_member(EquivalenceClass *ec, Expr *expr, Relids relids,
+ JoinDomain *jdomain, Oid datatype)
+{
+ EquivalenceMember *em = make_eq_member(ec, expr, relids, jdomain,
+ NULL, datatype);
+
+ ec->ec_members = lappend(ec->ec_members, em);
+ return em;
+}
+
/*
* get_eclass_for_sort_expr
@@ -616,7 +650,8 @@ get_eclass_for_sort_expr(PlannerInfo *root,
foreach(lc1, root->eq_classes)
{
EquivalenceClass *cur_ec = (EquivalenceClass *) lfirst(lc1);
- ListCell *lc2;
+ EquivalenceMemberIterator it;
+ EquivalenceMember *cur_em;
/*
* Never match to a volatile EC, except when we are looking at another
@@ -631,10 +666,9 @@ get_eclass_for_sort_expr(PlannerInfo *root,
if (!equal(opfamilies, cur_ec->ec_opfamilies))
continue;
- foreach(lc2, cur_ec->ec_members)
+ setup_eclass_member_iterator_with_children(&it, root, cur_ec, rel);
+ while ((cur_em = eclass_member_iterator_next(&it)) != NULL)
{
- EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc2);
-
/*
* Ignore child members unless they match the request.
*/
@@ -653,6 +687,7 @@ get_eclass_for_sort_expr(PlannerInfo *root,
equal(expr, cur_em->em_expr))
return cur_ec; /* Match! */
}
+ dispose_eclass_member_iterator(&it);
}
/* No match; does caller want a NULL result? */
@@ -690,7 +725,7 @@ get_eclass_for_sort_expr(PlannerInfo *root,
expr_relids = pull_varnos(root, (Node *) expr);
newem = add_eq_member(newec, copyObject(expr), expr_relids,
- jdomain, NULL, opcintype);
+ jdomain, opcintype);
/*
* add_eq_member doesn't check for volatile functions, set-returning
@@ -764,15 +799,16 @@ find_ec_member_matching_expr(PlannerInfo *root, EquivalenceClass *ec,
Expr *expr,
Relids relids)
{
- ListCell *lc;
+ EquivalenceMemberIterator it;
+ EquivalenceMember *em;
/* We ignore binary-compatible relabeling on both ends */
while (expr && IsA(expr, RelabelType))
expr = ((RelabelType *) expr)->arg;
- foreach(lc, ec->ec_members)
+ setup_eclass_member_iterator_with_children(&it, root, ec, relids);
+ while ((em = eclass_member_iterator_next(&it)) != NULL)
{
- EquivalenceMember *em = (EquivalenceMember *) lfirst(lc);
Expr *emexpr;
/*
@@ -799,6 +835,7 @@ find_ec_member_matching_expr(PlannerInfo *root, EquivalenceClass *ec,
if (equal(emexpr, expr))
return em;
}
+ dispose_eclass_member_iterator(&it);
return NULL;
}
@@ -841,7 +878,8 @@ find_computable_ec_member(PlannerInfo *root,
bool require_parallel_safe)
{
List *exprvars;
- ListCell *lc;
+ EquivalenceMemberIterator it;
+ EquivalenceMember *em;
/*
* Pull out the Vars and quasi-Vars present in "exprs". In the typical
@@ -855,9 +893,9 @@ find_computable_ec_member(PlannerInfo *root,
PVC_INCLUDE_PLACEHOLDERS |
PVC_INCLUDE_CONVERTROWTYPES);
- foreach(lc, ec->ec_members)
+ setup_eclass_member_iterator_with_children(&it, root, ec, relids);
+ while ((em = eclass_member_iterator_next(&it)) != NULL)
{
- EquivalenceMember *em = (EquivalenceMember *) lfirst(lc);
List *emvars;
ListCell *lc2;
@@ -901,6 +939,7 @@ find_computable_ec_member(PlannerInfo *root,
return em; /* found usable expression */
}
+ dispose_eclass_member_iterator(&it);
return NULL;
}
@@ -1162,7 +1201,8 @@ generate_base_implied_equalities_const(PlannerInfo *root,
Oid eq_op;
RestrictInfo *rinfo;
- Assert(!cur_em->em_is_child); /* no children yet */
+ Assert(!cur_em->em_is_child); /* Child members should not exist in
+ * ec_members */
if (cur_em == const_em)
continue;
eq_op = select_equality_operator(ec,
@@ -1231,7 +1271,8 @@ generate_base_implied_equalities_no_const(PlannerInfo *root,
EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc);
int relid;
- Assert(!cur_em->em_is_child); /* no children yet */
+ Assert(!cur_em->em_is_child); /* Child members should not exist in
+ * ec_members */
if (!bms_get_singleton_member(cur_em->em_relids, &relid))
continue;
Assert(relid < root->simple_rel_array_size);
@@ -1564,7 +1605,8 @@ generate_join_implied_equalities_normal(PlannerInfo *root,
List *new_members = NIL;
List *outer_members = NIL;
List *inner_members = NIL;
- ListCell *lc1;
+ EquivalenceMemberIterator it;
+ EquivalenceMember *cur_em;
/*
* First, scan the EC to identify member values that are computable at the
@@ -1575,10 +1617,9 @@ generate_join_implied_equalities_normal(PlannerInfo *root,
* as well as to at least one input member, plus enforce at least one
* outer-rel member equal to at least one inner-rel member.
*/
- foreach(lc1, ec->ec_members)
+ setup_eclass_member_iterator_with_children(&it, root, ec, join_relids);
+ while ((cur_em = eclass_member_iterator_next(&it)) != NULL)
{
- EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc1);
-
/*
* We don't need to check explicitly for child EC members. This test
* against join_relids will cause them to be ignored except when
@@ -1594,6 +1635,7 @@ generate_join_implied_equalities_normal(PlannerInfo *root,
else
new_members = lappend(new_members, cur_em);
}
+ dispose_eclass_member_iterator(&it);
/*
* First, select the joinclause if needed. We can equate any one outer
@@ -1611,6 +1653,7 @@ generate_join_implied_equalities_normal(PlannerInfo *root,
Oid best_eq_op = InvalidOid;
int best_score = -1;
RestrictInfo *rinfo;
+ ListCell *lc1;
foreach(lc1, outer_members)
{
@@ -1685,6 +1728,7 @@ generate_join_implied_equalities_normal(PlannerInfo *root,
List *old_members = list_concat(outer_members, inner_members);
EquivalenceMember *prev_em = NULL;
RestrictInfo *rinfo;
+ ListCell *lc1;
/* For now, arbitrarily take the first old_member as the one to use */
if (old_members)
@@ -1692,7 +1736,7 @@ generate_join_implied_equalities_normal(PlannerInfo *root,
foreach(lc1, new_members)
{
- EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc1);
+ cur_em = (EquivalenceMember *) lfirst(lc1);
if (prev_em != NULL)
{
@@ -2181,7 +2225,8 @@ reconsider_outer_join_clause(PlannerInfo *root, OuterJoinClauseInfo *ojcinfo,
{
EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc2);
- Assert(!cur_em->em_is_child); /* no children yet */
+ Assert(!cur_em->em_is_child); /* Child members should not exist
+ * in ec_members */
if (equal(outervar, cur_em->em_expr))
{
match = true;
@@ -2308,7 +2353,8 @@ reconsider_full_join_clause(PlannerInfo *root, OuterJoinClauseInfo *ojcinfo)
foreach(lc2, cur_ec->ec_members)
{
coal_em = (EquivalenceMember *) lfirst(lc2);
- Assert(!coal_em->em_is_child); /* no children yet */
+ Assert(!coal_em->em_is_child); /* Child members should not exist
+ * in ec_members */
if (IsA(coal_em->em_expr, CoalesceExpr))
{
CoalesceExpr *cexpr = (CoalesceExpr *) coal_em->em_expr;
@@ -2526,8 +2572,8 @@ exprs_known_equal(PlannerInfo *root, Node *item1, Node *item2, Oid opfamily)
{
EquivalenceMember *em = (EquivalenceMember *) lfirst(lc2);
- if (em->em_is_child)
- continue; /* ignore children here */
+ Assert(!em->em_is_child); /* Child members should not exist in
+ * ec_members */
if (equal(item1, em->em_expr))
item1member = true;
else if (equal(item2, em->em_expr))
@@ -2598,8 +2644,8 @@ match_eclasses_to_foreign_key_col(PlannerInfo *root,
EquivalenceMember *em = (EquivalenceMember *) lfirst(lc2);
Var *var;
- if (em->em_is_child)
- continue; /* ignore children here */
+ /* Child members should not exist in ec_members */
+ Assert(!em->em_is_child);
/* EM must be a Var, possibly with RelabelType */
var = (Var *) em->em_expr;
@@ -2696,6 +2742,7 @@ add_child_rel_equivalences(PlannerInfo *root,
Relids top_parent_relids = child_rel->top_parent_relids;
Relids child_relids = child_rel->relids;
int i;
+ ListCell *lc;
/*
* EC merging should be complete already, so we can use the parent rel's
@@ -2708,7 +2755,6 @@ add_child_rel_equivalences(PlannerInfo *root,
while ((i = bms_next_member(parent_rel->eclass_indexes, i)) >= 0)
{
EquivalenceClass *cur_ec = (EquivalenceClass *) list_nth(root->eq_classes, i);
- int num_members;
/*
* If this EC contains a volatile expression, then generating child
@@ -2721,29 +2767,15 @@ add_child_rel_equivalences(PlannerInfo *root,
/* Sanity check eclass_indexes only contain ECs for parent_rel */
Assert(bms_is_subset(top_parent_relids, cur_ec->ec_relids));
- /*
- * We don't use foreach() here because there's no point in scanning
- * newly-added child members, so we can stop after the last
- * pre-existing EC member.
- */
- num_members = list_length(cur_ec->ec_members);
- for (int pos = 0; pos < num_members; pos++)
+ foreach(lc, cur_ec->ec_members)
{
- EquivalenceMember *cur_em = (EquivalenceMember *) list_nth(cur_ec->ec_members, pos);
+ EquivalenceMember *cur_em = lfirst_node(EquivalenceMember, lc);
if (cur_em->em_is_const)
continue; /* ignore consts here */
- /*
- * We consider only original EC members here, not
- * already-transformed child members. Otherwise, if some original
- * member expression references more than one appendrel, we'd get
- * an O(N^2) explosion of useless derived expressions for
- * combinations of children. (But add_child_join_rel_equivalences
- * may add targeted combinations for partitionwise-join purposes.)
- */
- if (cur_em->em_is_child)
- continue; /* ignore children here */
+ /* Child members should not exist in ec_members */
+ Assert(!cur_em->em_is_child);
/*
* Consider only members that reference and can be computed at
@@ -2759,6 +2791,7 @@ add_child_rel_equivalences(PlannerInfo *root,
/* OK, generate transformed child version */
Expr *child_expr;
Relids new_relids;
+ EquivalenceMember *child_em;
if (parent_rel->reloptkind == RELOPT_BASEREL)
{
@@ -2788,9 +2821,11 @@ add_child_rel_equivalences(PlannerInfo *root,
top_parent_relids);
new_relids = bms_add_members(new_relids, child_relids);
- (void) add_eq_member(cur_ec, child_expr, new_relids,
- cur_em->em_jdomain,
- cur_em, cur_em->em_datatype);
+ child_em = make_eq_member(cur_ec, child_expr, new_relids,
+ cur_em->em_jdomain,
+ cur_em, cur_em->em_datatype);
+ child_rel->eclass_child_members =
+ lappend(child_rel->eclass_child_members, child_em);
/* Record this EC index for the child rel */
child_rel->eclass_indexes = bms_add_member(child_rel->eclass_indexes, i);
@@ -2840,7 +2875,7 @@ add_child_join_rel_equivalences(PlannerInfo *root,
while ((i = bms_next_member(matching_ecs, i)) >= 0)
{
EquivalenceClass *cur_ec = (EquivalenceClass *) list_nth(root->eq_classes, i);
- int num_members;
+ ListCell *lc;
/*
* If this EC contains a volatile expression, then generating child
@@ -2853,25 +2888,15 @@ add_child_join_rel_equivalences(PlannerInfo *root,
/* Sanity check on get_eclass_indexes_for_relids result */
Assert(bms_overlap(top_parent_relids, cur_ec->ec_relids));
- /*
- * We don't use foreach() here because there's no point in scanning
- * newly-added child members, so we can stop after the last
- * pre-existing EC member.
- */
- num_members = list_length(cur_ec->ec_members);
- for (int pos = 0; pos < num_members; pos++)
+ foreach(lc, cur_ec->ec_members)
{
- EquivalenceMember *cur_em = (EquivalenceMember *) list_nth(cur_ec->ec_members, pos);
+ EquivalenceMember *cur_em = lfirst_node(EquivalenceMember, lc);
if (cur_em->em_is_const)
continue; /* ignore consts here */
- /*
- * We consider only original EC members here, not
- * already-transformed child members.
- */
- if (cur_em->em_is_child)
- continue; /* ignore children here */
+ /* Child members should not exist in ec_members */
+ Assert(!cur_em->em_is_child);
/*
* We may ignore expressions that reference a single baserel,
@@ -2886,6 +2911,8 @@ add_child_join_rel_equivalences(PlannerInfo *root,
/* Yes, generate transformed child version */
Expr *child_expr;
Relids new_relids;
+ EquivalenceMember *child_em;
+ int j;
if (parent_joinrel->reloptkind == RELOPT_JOINREL)
{
@@ -2916,9 +2943,38 @@ add_child_join_rel_equivalences(PlannerInfo *root,
top_parent_relids);
new_relids = bms_add_members(new_relids, child_relids);
- (void) add_eq_member(cur_ec, child_expr, new_relids,
- cur_em->em_jdomain,
- cur_em, cur_em->em_datatype);
+ child_em = make_eq_member(cur_ec, child_expr, new_relids,
+ cur_em->em_jdomain,
+ cur_em, cur_em->em_datatype);
+ child_joinrel->eclass_child_members =
+ lappend(child_joinrel->eclass_child_members, child_em);
+
+ /*
+ * Update the corresponding inverted indexes.
+ */
+ j = -1;
+ while ((j = bms_next_member(child_joinrel->relids, j)) >= 0)
+ {
+ EquivalenceClassIndexes *indexes =
+ &root->eclass_indexes_array[j];
+
+ /*
+ * We do not need to update the inverted index of the top
+ * parent relations. This is because EquivalenceMembers
+ * that have only such parent relations as em_relids are
+ * already present in the ec_members, and so cannot be
+ * candidates for additional iteration by
+ * EquivalenceMemberIterator. Since the iterator needs
+ * EquivalenceMembers whose em_relids has child relations,
+ * skipping the update of this inverted index allows for
+ * faster iteration.
+ */
+ if (root->append_rel_array[j] == NULL)
+ continue;
+ indexes->joinrel_indexes =
+ bms_add_member(indexes->joinrel_indexes,
+ child_joinrel->join_rel_list_index);
+ }
}
}
}
@@ -2971,7 +3027,7 @@ add_setop_child_rel_equivalences(PlannerInfo *root, RelOptInfo *child_rel,
tle->expr,
child_rel->relids,
parent_em->em_jdomain,
- parent_em,
+ /* parent_em, */ /* Will be fixed in the next commit*/
exprType((Node *) tle->expr));
lc2 = lnext(setop_pathkeys, lc2);
@@ -2987,6 +3043,219 @@ add_setop_child_rel_equivalences(PlannerInfo *root, RelOptInfo *child_rel,
list_length(root->eq_classes) - 1);
}
+/*
+ * setup_eclass_member_iterator_with_children
+ * Setup an EquivalenceMemberIterator 'it' to iterate over all parent
+ * EquivalenceMembers and child members associated with the given 'ec' that
+ * are relevant to the specified 'relids'.
+ *
+ * This iterator returns:
+ * - All parent members stored directly in ec->ec_members.
+ * - The child members whose em_relids is a subset of the given 'relids'.
+ *
+ * Note:
+ * - The iterator may return false positives, i.e., child members whose
+ * em_relids is not a subset. So the caller must check that they satisfy
+ * the desired condition.
+ * - Once used, the caller should dispose of the iterator by calling
+ * dispose_eclass_member_iterator().
+ *
+ * Parameters:
+ * root - The PlannerInfo context.
+ * ec - The EquivalenceClass from which to iterate members.
+ * relids - The Relids used to filter for relevant child members.
+ */
+void
+setup_eclass_member_iterator_with_children(EquivalenceMemberIterator *it,
+ PlannerInfo *root,
+ EquivalenceClass *ec,
+ Relids relids)
+{
+ Bitmapset *matching_indexes;
+ int i;
+
+ /*
+ * Initialize the iterator.
+ */
+ it->index = -1;
+ it->list_is_copy = false;
+ it->ec_members = ec->ec_members;
+
+ /*
+ * If there are no child relations, there is nothing to do. This
+ * effectively avoids regression for non-partitioned cases.
+ */
+ if (root->append_rel_array == NULL)
+ return;
+
+ /*
+ * EquivalenceClass->ec_members has only parent EquivalenceMembers. Child
+ * members are translated using child RelOptInfos and stored in them. This
+ * is done in add_child_rel_equivalences(),
+ * add_child_join_rel_equivalences(), and
+ * add_setop_child_rel_equivalences(). To retrieve child
+ * EquivalenceMembers of some parent, we need to know which RelOptInfos
+ * have such child members. We can know this information using indexes
+ * like EquivalenceClassIndexes->joinrel_indexes.
+ *
+ * We use an inverted index mechanism to quickly iterate over the members
+ * whose em_relids is a subset of the given 'relids'. The inverted indexes
+ * store RelOptInfo indices that have EquivalenceMembers mentioning them.
+ * Taking the union of these indexes allows to find which RelOptInfos have
+ * the EquivalenceMember we are looking for. With this method, the
+ * em_relids of the newly iterated ones overlap the given 'relids', but
+ * may not be subsets, so the caller must check that they satisfy the
+ * desired condition.
+ *
+ * The above comments are about joinrels, and for simple rels, this
+ * mechanism is simpler. It is sufficient to simply add the child
+ * EquivalenceMembers of RelOptInfo to the iterator.
+ *
+ * We need to perform these steps for each of the two types of relations.
+ */
+
+ /*
+ * Iterate over the given relids, adding child members for simple rels and
+ * taking union indexes for join rels.
+ */
+ i = -1;
+ matching_indexes = NULL;
+ while ((i = bms_next_member(relids, i)) >= 0)
+ {
+ RelOptInfo *child_rel;
+ EquivalenceClassIndexes *indexes;
+
+ /*
+ * If this relation is a parent, we don't have to do anything.
+ */
+ if (root->append_rel_array[i] == NULL)
+ continue;
+
+ /*
+ * Add child members that mention this relation to the iterator.
+ */
+ child_rel = root->simple_rel_array[i];
+ if (child_rel != NULL)
+ add_eclass_child_members_to_iterator(it, root, ec, child_rel);
+
+ /*
+ * Union indexes for join rels.
+ */
+ indexes = &root->eclass_indexes_array[i];
+ matching_indexes =
+ bms_add_members(matching_indexes, indexes->joinrel_indexes);
+ }
+
+ /*
+ * For join rels, add child members using 'matching_indexes'.
+ */
+ i = -1;
+ while ((i = bms_next_member(matching_indexes, i)) >= 0)
+ {
+ RelOptInfo *child_joinrel =
+ list_nth_node(RelOptInfo, root->join_rel_list, i);
+
+ Assert(child_joinrel != NULL);
+
+ /*
+ * If this joinrel's Relids is not a subset of the given one, then the
+ * child EquivalenceMembers it holds should never be a subset either.
+ */
+ if (bms_is_subset(child_joinrel->relids, relids))
+ add_eclass_child_members_to_iterator(it, root, ec, child_joinrel);
+#ifdef USE_ASSERT_CHECKING
+ else
+ {
+ /*
+ * Verify that the above comment is correct.
+ *
+ * NOTE: We may remove this assertion after the beta process.
+ */
+
+ ListCell *lc;
+
+ foreach(lc, child_joinrel->eclass_child_members)
+ {
+ EquivalenceMember *child_em = lfirst_node(EquivalenceMember, lc);
+
+ if (child_em->em_ec != ec)
+ continue;
+ Assert(!bms_is_subset(child_em->em_relids, relids));
+ }
+ }
+#endif
+ }
+ bms_free(matching_indexes);
+}
+
+/*
+ * add_eclass_child_members_to_iterator
+ * Add a child EquivalenceMember referencing the given child_rel to
+ * the iterator from RelOptInfo.
+ *
+ * This function is expected to be called only from
+ * setup_eclass_member_iterator_with_children().
+ */
+static void
+add_eclass_child_members_to_iterator(EquivalenceMemberIterator *it,
+ PlannerInfo *root,
+ EquivalenceClass *ec,
+ RelOptInfo *child_rel)
+{
+ ListCell *lc;
+
+ foreach(lc, child_rel->eclass_child_members)
+ {
+ EquivalenceMember *child_em = lfirst_node(EquivalenceMember, lc);
+
+ /* Skip unwanted EquivalenceMembers */
+ if (child_em->em_ec != ec)
+ continue;
+
+ /*
+ * If this is the first time the iterator's list has been modified, we
+ * need to make a copy of it.
+ */
+ if (!it->list_is_copy)
+ {
+ it->ec_members = list_copy(it->ec_members);
+ it->list_is_copy = true;
+ }
+
+ /* Add this child EquivalenceMember to the list */
+ it->ec_members = lappend(it->ec_members, child_em);
+ }
+}
+
+/*
+ * eclass_member_iterator_next
+ * Get a next EquivalenceMember from an EquivalenceMemberIterator 'it'
+ * that was setup by setup_eclass_member_iterator_with_children(). NULL is
+ * returned if there are no members left.
+ */
+EquivalenceMember *
+eclass_member_iterator_next(EquivalenceMemberIterator *it)
+{
+ if (++it->index < list_length(it->ec_members))
+ return list_nth_node(EquivalenceMember, it->ec_members, it->index);
+ return NULL;
+}
+
+/*
+ * dispose_eclass_member_iterator
+ * Free any memory allocated by the iterator.
+ */
+void
+dispose_eclass_member_iterator(EquivalenceMemberIterator *it)
+{
+ /*
+ * XXX Should we use list_free()? I decided to use this style to take
+ * advantage of speculative execution.
+ */
+ if (unlikely(it->list_is_copy))
+ pfree(it->ec_members);
+}
+
/*
* generate_implied_equalities_for_column
@@ -3041,6 +3310,7 @@ generate_implied_equalities_for_column(PlannerInfo *root,
EquivalenceClass *cur_ec = (EquivalenceClass *) list_nth(root->eq_classes, i);
EquivalenceMember *cur_em;
ListCell *lc2;
+ EquivalenceMemberIterator it;
/* Sanity check eclass_indexes only contain ECs for rel */
Assert(is_child_rel || bms_is_subset(rel->relids, cur_ec->ec_relids));
@@ -3062,15 +3332,14 @@ generate_implied_equalities_for_column(PlannerInfo *root,
* corner cases, so for now we live with just reporting the first
* match. See also get_eclass_for_sort_expr.)
*/
- cur_em = NULL;
- foreach(lc2, cur_ec->ec_members)
+ setup_eclass_member_iterator_with_children(&it, root, cur_ec, rel->relids);
+ while ((cur_em = eclass_member_iterator_next(&it)) != NULL)
{
- cur_em = (EquivalenceMember *) lfirst(lc2);
if (bms_equal(cur_em->em_relids, rel->relids) &&
callback(root, rel, cur_ec, cur_em, callback_arg))
break;
- cur_em = NULL;
}
+ dispose_eclass_member_iterator(&it);
if (!cur_em)
continue;
@@ -3085,8 +3354,8 @@ generate_implied_equalities_for_column(PlannerInfo *root,
Oid eq_op;
RestrictInfo *rinfo;
- if (other_em->em_is_child)
- continue; /* ignore children here */
+ /* Child members should not exist in ec_members */
+ Assert(!other_em->em_is_child);
/* Make sure it'll be a join to a different rel */
if (other_em == cur_em ||
@@ -3304,8 +3573,8 @@ eclass_useful_for_merging(PlannerInfo *root,
{
EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc);
- if (cur_em->em_is_child)
- continue; /* ignore children here */
+ /* Child members should not exist in ec_members */
+ Assert(!cur_em->em_is_child);
if (!bms_overlap(cur_em->em_relids, relids))
return true;
diff --git a/src/backend/optimizer/path/indxpath.c b/src/backend/optimizer/path/indxpath.c
index a43ca16d683..c206f260633 100644
--- a/src/backend/optimizer/path/indxpath.c
+++ b/src/backend/optimizer/path/indxpath.c
@@ -190,7 +190,7 @@ static IndexClause *expand_indexqual_rowcompare(PlannerInfo *root,
IndexOptInfo *index,
Oid expr_op,
bool var_on_left);
-static void match_pathkeys_to_index(IndexOptInfo *index, List *pathkeys,
+static void match_pathkeys_to_index(PlannerInfo *root, IndexOptInfo *index, List *pathkeys,
List **orderby_clauses_p,
List **clause_columns_p);
static Expr *match_clause_to_ordering_op(IndexOptInfo *index,
@@ -934,7 +934,7 @@ build_index_paths(PlannerInfo *root, RelOptInfo *rel,
* query_pathkeys will allow an incremental sort to be considered on
* the index's partially sorted results.
*/
- match_pathkeys_to_index(index, root->query_pathkeys,
+ match_pathkeys_to_index(root, index, root->query_pathkeys,
&orderbyclauses,
&orderbyclausecols);
if (list_length(root->query_pathkeys) == list_length(orderbyclauses))
@@ -3739,7 +3739,7 @@ expand_indexqual_rowcompare(PlannerInfo *root,
* item in the given 'pathkeys' list.
*/
static void
-match_pathkeys_to_index(IndexOptInfo *index, List *pathkeys,
+match_pathkeys_to_index(PlannerInfo *root, IndexOptInfo *index, List *pathkeys,
List **orderby_clauses_p,
List **clause_columns_p)
{
@@ -3756,7 +3756,8 @@ match_pathkeys_to_index(IndexOptInfo *index, List *pathkeys,
{
PathKey *pathkey = (PathKey *) lfirst(lc1);
bool found = false;
- ListCell *lc2;
+ EquivalenceMemberIterator it;
+ EquivalenceMember *member;
/* Pathkey must request default sort order for the target opfamily */
@@ -3776,9 +3777,10 @@ match_pathkeys_to_index(IndexOptInfo *index, List *pathkeys,
* be considered to match more than one pathkey list, which is OK
* here. See also get_eclass_for_sort_expr.)
*/
- foreach(lc2, pathkey->pk_eclass->ec_members)
+ setup_eclass_member_iterator_with_children(&it, root, pathkey->pk_eclass,
+ index->rel->relids);
+ while ((member = eclass_member_iterator_next(&it)) != NULL)
{
- EquivalenceMember *member = (EquivalenceMember *) lfirst(lc2);
int indexcol;
/* No possibility of match if it references other relations */
@@ -3813,6 +3815,7 @@ match_pathkeys_to_index(IndexOptInfo *index, List *pathkeys,
if (found) /* don't want to look at remaining members */
break;
}
+ dispose_eclass_member_iterator(&it);
/*
* Return the matches found so far when this pathkey couldn't be
diff --git a/src/backend/optimizer/path/pathkeys.c b/src/backend/optimizer/path/pathkeys.c
index 154eb505d75..a9419d37e2f 100644
--- a/src/backend/optimizer/path/pathkeys.c
+++ b/src/backend/optimizer/path/pathkeys.c
@@ -1151,8 +1151,8 @@ convert_subquery_pathkeys(PlannerInfo *root, RelOptInfo *rel,
Oid sub_expr_coll = sub_eclass->ec_collation;
ListCell *k;
- if (sub_member->em_is_child)
- continue; /* ignore children here */
+ /* Child members should not exist in ec_members */
+ Assert(!sub_member->em_is_child);
foreach(k, subquery_tlist)
{
@@ -1709,8 +1709,11 @@ select_outer_pathkeys_for_merge(PlannerInfo *root,
{
EquivalenceMember *em = (EquivalenceMember *) lfirst(lc2);
+ /* Child members should not exist in ec_members */
+ Assert(!em->em_is_child);
+
/* Potential future join partner? */
- if (!em->em_is_const && !em->em_is_child &&
+ if (!em->em_is_const &&
!bms_overlap(em->em_relids, joinrel->relids))
score++;
}
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index ff507331a06..7791e282e9e 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -119,16 +119,23 @@ setup_simple_rel_arrays(PlannerInfo *root)
root->simple_rte_array[rti++] = rte;
}
- /* append_rel_array is not needed if there are no AppendRelInfos */
+ /*
+ * append_rel_array and eclass_indexes_array are not needed if there are
+ * no AppendRelInfos.
+ */
if (root->append_rel_list == NIL)
{
root->append_rel_array = NULL;
+ root->eclass_indexes_array = NULL;
return;
}
root->append_rel_array = (AppendRelInfo **)
palloc0(size * sizeof(AppendRelInfo *));
+ root->eclass_indexes_array = (EquivalenceClassIndexes *)
+ palloc0(size * sizeof(EquivalenceClassIndexes));
+
/*
* append_rel_array is filled with any already-existing AppendRelInfos,
* which currently could only come from UNION ALL flattening. We might
@@ -175,11 +182,21 @@ expand_planner_arrays(PlannerInfo *root, int add_size)
repalloc0_array(root->simple_rte_array, RangeTblEntry *, root->simple_rel_array_size, new_size);
if (root->append_rel_array)
+ {
+ Assert(root->eclass_indexes_array);
root->append_rel_array =
repalloc0_array(root->append_rel_array, AppendRelInfo *, root->simple_rel_array_size, new_size);
+ root->eclass_indexes_array =
+ repalloc0_array(root->eclass_indexes_array, EquivalenceClassIndexes, root->simple_rel_array_size, new_size);
+ }
else
+ {
+ Assert(!root->eclass_indexes_array);
root->append_rel_array =
palloc0_array(AppendRelInfo *, new_size);
+ root->eclass_indexes_array =
+ palloc0_array(EquivalenceClassIndexes, new_size);
+ }
root->simple_rel_array_size = new_size;
}
@@ -234,6 +251,8 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
rel->subplan_params = NIL;
rel->rel_parallel_workers = -1; /* set up in get_relation_info */
rel->amflags = 0;
+ rel->join_rel_list_index = -1;
+ rel->eclass_child_members = NIL;
rel->serverid = InvalidOid;
if (rte->rtekind == RTE_RELATION)
{
@@ -626,6 +645,12 @@ set_foreign_rel_properties(RelOptInfo *joinrel, RelOptInfo *outer_rel,
static void
add_join_rel(PlannerInfo *root, RelOptInfo *joinrel)
{
+ /*
+ * Store the index of this joinrel to use in
+ * add_child_join_rel_equivalences().
+ */
+ joinrel->join_rel_list_index = list_length(root->join_rel_list);
+
/* GEQO requires us to append the new joinrel to the end of the list! */
root->join_rel_list = lappend(root->join_rel_list, joinrel);
@@ -741,6 +766,8 @@ build_join_rel(PlannerInfo *root,
joinrel->subplan_params = NIL;
joinrel->rel_parallel_workers = -1;
joinrel->amflags = 0;
+ joinrel->join_rel_list_index = -1;
+ joinrel->eclass_child_members = NIL;
joinrel->serverid = InvalidOid;
joinrel->userid = InvalidOid;
joinrel->useridiscurrent = false;
@@ -928,6 +955,8 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
joinrel->subroot = NULL;
joinrel->subplan_params = NIL;
joinrel->amflags = 0;
+ joinrel->join_rel_list_index = -1;
+ joinrel->eclass_child_members = NIL;
joinrel->serverid = InvalidOid;
joinrel->userid = InvalidOid;
joinrel->useridiscurrent = false;
@@ -1490,6 +1519,8 @@ fetch_upper_rel(PlannerInfo *root, UpperRelationKind kind, Relids relids)
upperrel->cheapest_total_path = NULL;
upperrel->cheapest_unique_path = NULL;
upperrel->cheapest_parameterized_paths = NIL;
+ upperrel->join_rel_list_index = -1;
+ upperrel->eclass_child_members = NIL;
root->upper_rels[kind] = lappend(root->upper_rels[kind], upperrel);
diff --git a/src/include/nodes/pathnodes.h b/src/include/nodes/pathnodes.h
index c24a1fc8514..6b3a028cc58 100644
--- a/src/include/nodes/pathnodes.h
+++ b/src/include/nodes/pathnodes.h
@@ -216,6 +216,8 @@ typedef struct PlannerInfo PlannerInfo;
#define HAVE_PLANNERINFO_TYPEDEF 1
#endif
+struct EquivalenceClassIndexes;
+
struct PlannerInfo
{
pg_node_attr(no_copy_equal, no_read, no_query_jumble)
@@ -272,6 +274,13 @@ struct PlannerInfo
*/
struct AppendRelInfo **append_rel_array pg_node_attr(read_write_ignore);
+ /*
+ * eclass_indexes_array is the same length as simple_rel_array and holds
+ * the indexes of the corresponding rels for faster EquivalenceMember
+ * lookups. See the EquivalenceClass comment for more details.
+ */
+ struct EquivalenceClassIndexes *eclass_indexes_array pg_node_attr(read_write_ignore);
+
/*
* all_baserels is a Relids set of all base relids (but not joins or
* "other" rels) in the query. This is computed in deconstruct_jointree.
@@ -984,6 +993,17 @@ typedef struct RelOptInfo
/* Bitmask of optional features supported by the table AM */
uint32 amflags;
+ /*
+ * information about a join rel
+ */
+ /* index in root->join_rel_list of this rel */
+ int join_rel_list_index;
+
+ /*
+ * EquivalenceMembers referencing this rel
+ */
+ List *eclass_child_members;
+
/*
* Information about foreign tables and foreign joins
*/
@@ -1403,6 +1423,12 @@ typedef struct JoinDomain
* entry: consider SELECT random() AS a, random() AS b ... ORDER BY b,a.
* So we record the SortGroupRef of the originating sort clause.
*
+ * EquivalenceClass->ec_members can only have parent members, and child members
+ * are stored in RelOptInfos, from which those child members are translated. To
+ * lookup child EquivalenceMembers, we use EquivalenceMemberIterator. See
+ * its comment for usage. The approach to lookup child members quickly is
+ * described as setup_eclass_member_iterator_with_children() comment.
+ *
* NB: if ec_merged isn't NULL, this class has been merged into another, and
* should be ignored in favor of using the pointed-to class.
*
@@ -1477,8 +1503,61 @@ typedef struct EquivalenceMember
JoinDomain *em_jdomain; /* join domain containing the source clause */
/* if em_is_child is true, this links to corresponding EM for top parent */
struct EquivalenceMember *em_parent pg_node_attr(read_write_ignore);
+ EquivalenceClass *em_ec; /* EquivalenceClass which has this member */
} EquivalenceMember;
+/*
+ * EquivalenceMemberIterator
+ *
+ * EquivalenceMemberIterator is designed to iterate over all parent
+ * EquivalenceMembers and child members associated with the given 'ec' that
+ * are relevant to the specified 'relids'. In particular, it iterates over:
+ * - All parent members stored directly in ec->ec_members.
+ * - The child members whose em_relids is a subset of the given 'relids'.
+ *
+ * Note:
+ * - The iterator may return false positives, i.e., child members whose
+ * em_relids is not a subset. So the caller must check that they satisfy
+ * the desired condition.
+ *
+ * The most common way to use this iterator is as follows:
+ * -----
+ * PlannerInfo *root = given;
+ * EquivalenceClass *ec = given;
+ * Relids rel = given;
+ * EquivalenceMemberIterator it;
+ * EquivalenceMember *em;
+ *
+ * setup_eclass_member_iterator_with_children(&it, root, ec, rel);
+ * while ((em = eclass_member_iterator_next(&it)) != NULL)
+ * {
+ * use em ...;
+ * }
+ * dispose_eclass_member_iterator(&it);
+ * -----
+ */
+typedef struct
+{
+ int index; /* current index within 'ec_members'. */
+ bool list_is_copy; /* is 'ec_members' a newly allocated one? */
+ List *ec_members; /* parent and child members */
+} EquivalenceMemberIterator;
+
+/*
+ * EquivalenceClassIndexes
+ *
+ * As mentioned in the EquivalenceClass comment, we introduce a
+ * bitmapset-based indexing mechanism for faster lookups of child
+ * EquivalenceMembers. This struct exists for each relation and holds the
+ * corresponding indexes.
+ */
+typedef struct EquivalenceClassIndexes
+{
+ Bitmapset *joinrel_indexes; /* Indexes in PlannerInfo's join_rel_list
+ * list for RelOptInfos that mention this
+ * relation */
+} EquivalenceClassIndexes;
+
/*
* PathKeys
*
diff --git a/src/include/optimizer/paths.h b/src/include/optimizer/paths.h
index 84a000a3ef1..6b80c8665b9 100644
--- a/src/include/optimizer/paths.h
+++ b/src/include/optimizer/paths.h
@@ -183,6 +183,12 @@ extern void add_setop_child_rel_equivalences(PlannerInfo *root,
RelOptInfo *child_rel,
List *child_tlist,
List *setop_pathkeys);
+extern void setup_eclass_member_iterator_with_children(EquivalenceMemberIterator *it,
+ PlannerInfo *root,
+ EquivalenceClass *ec,
+ Relids relids);
+extern EquivalenceMember *eclass_member_iterator_next(EquivalenceMemberIterator *it);
+extern void dispose_eclass_member_iterator(EquivalenceMemberIterator *it);
extern List *generate_implied_equalities_for_column(PlannerInfo *root,
RelOptInfo *rel,
ec_matches_callback_type callback,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index bfa276d2d35..c9688aa6b67 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -700,7 +700,9 @@ EphemeralNamedRelationData
EphemeralNamedRelationMetadata
EphemeralNamedRelationMetadataData
EquivalenceClass
+EquivalenceClassIndexes
EquivalenceMember
+EquivalenceMemberIterator
ErrorContextCallback
ErrorData
ErrorSaveContext
--
2.45.2.windows.1