From c2f94809f72a8c89b67326a000e5bbb6db52ffbf Mon Sep 17 00:00:00 2001 From: Yuya Watari Date: Fri, 25 Aug 2023 10:44:25 +0900 Subject: [PATCH v24 2/3] Introduce indexes for RestrictInfo This commit adds indexes to speed up searches for RestrictInfos. When there are many child partitions and we have to perform planning for join operations on the tables, we have to handle a large number of RestrictInfos, resulting in a long planning time. This commit adds ec_[source|derive]_indexes to speed up the search. We can use the indexes to filter out unwanted RestrictInfos, and improve the planning performance. Author: David Rowley and me, and includes rebase by Alena Rybakina [1]. [1] https://www.postgresql.org/message-id/72d292a1-06ff-432a-a803-af5053786444%40yandex.ru --- src/backend/nodes/outfuncs.c | 6 +- src/backend/nodes/readfuncs.c | 2 + src/backend/optimizer/path/costsize.c | 3 +- src/backend/optimizer/path/equivclass.c | 519 ++++++++++++++++++++-- src/backend/optimizer/plan/analyzejoins.c | 175 ++++++-- src/backend/optimizer/plan/planner.c | 2 + src/backend/optimizer/prep/prepjointree.c | 2 + src/backend/optimizer/util/appendinfo.c | 2 + src/backend/optimizer/util/inherit.c | 7 + src/backend/optimizer/util/restrictinfo.c | 5 + src/include/nodes/parsenodes.h | 6 + src/include/nodes/pathnodes.h | 41 +- src/include/optimizer/paths.h | 21 +- 13 files changed, 693 insertions(+), 98 deletions(-) diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c index 2c30bba212..c61110c200 100644 --- a/src/backend/nodes/outfuncs.c +++ b/src/backend/nodes/outfuncs.c @@ -463,8 +463,8 @@ _outEquivalenceClass(StringInfo str, const EquivalenceClass *node) WRITE_NODE_FIELD(ec_opfamilies); WRITE_OID_FIELD(ec_collation); WRITE_NODE_FIELD(ec_members); - WRITE_NODE_FIELD(ec_sources); - WRITE_NODE_FIELD(ec_derives); + WRITE_BITMAPSET_FIELD(ec_source_indexes); + WRITE_BITMAPSET_FIELD(ec_derive_indexes); WRITE_BITMAPSET_FIELD(ec_relids); WRITE_BOOL_FIELD(ec_has_const); WRITE_BOOL_FIELD(ec_has_volatile); @@ -567,6 +567,8 @@ _outRangeTblEntry(StringInfo str, const RangeTblEntry *node) WRITE_BOOL_FIELD(inh); WRITE_BOOL_FIELD(inFromCl); WRITE_NODE_FIELD(securityQuals); + WRITE_BITMAPSET_FIELD(eclass_source_indexes); + WRITE_BITMAPSET_FIELD(eclass_derive_indexes); } static void diff --git a/src/backend/nodes/readfuncs.c b/src/backend/nodes/readfuncs.c index b1e2f2b440..e3518c602b 100644 --- a/src/backend/nodes/readfuncs.c +++ b/src/backend/nodes/readfuncs.c @@ -431,6 +431,8 @@ _readRangeTblEntry(void) READ_BOOL_FIELD(inh); READ_BOOL_FIELD(inFromCl); READ_NODE_FIELD(securityQuals); + READ_BITMAPSET_FIELD(eclass_source_indexes); + READ_BITMAPSET_FIELD(eclass_derive_indexes); READ_DONE(); } diff --git a/src/backend/optimizer/path/costsize.c b/src/backend/optimizer/path/costsize.c index 8b76e98529..ae64b169f8 100644 --- a/src/backend/optimizer/path/costsize.c +++ b/src/backend/optimizer/path/costsize.c @@ -5784,7 +5784,8 @@ get_foreign_key_join_selectivity(PlannerInfo *root, if (ec && ec->ec_has_const) { EquivalenceMember *em = fkinfo->fk_eclass_member[i]; - RestrictInfo *rinfo = find_derived_clause_for_ec_member(ec, + RestrictInfo *rinfo = find_derived_clause_for_ec_member(root, + ec, em); if (rinfo) diff --git a/src/backend/optimizer/path/equivclass.c b/src/backend/optimizer/path/equivclass.c index 1931b01cbc..17fffc4087 100644 --- a/src/backend/optimizer/path/equivclass.c +++ b/src/backend/optimizer/path/equivclass.c @@ -42,6 +42,10 @@ static EquivalenceMember *add_eq_member(EquivalenceClass *ec, Expr *expr, Relids relids, JoinDomain *jdomain, Oid datatype); +static void add_eq_source(PlannerInfo *root, EquivalenceClass *ec, + RestrictInfo *rinfo); +static void add_eq_derive(PlannerInfo *root, EquivalenceClass *ec, + RestrictInfo *rinfo); static bool is_exprlist_member(Expr *node, List *exprs); static void generate_base_implied_equalities_const(PlannerInfo *root, EquivalenceClass *ec); @@ -320,7 +324,6 @@ process_equivalence(PlannerInfo *root, /* If case 1, nothing to do, except add to sources */ if (ec1 == ec2) { - ec1->ec_sources = lappend(ec1->ec_sources, restrictinfo); ec1->ec_min_security = Min(ec1->ec_min_security, restrictinfo->security_level); ec1->ec_max_security = Max(ec1->ec_max_security, @@ -331,6 +334,8 @@ process_equivalence(PlannerInfo *root, /* mark the RI as usable with this pair of EMs */ restrictinfo->left_em = em1; restrictinfo->right_em = em2; + + add_eq_source(root, ec1, restrictinfo); return true; } @@ -351,8 +356,10 @@ process_equivalence(PlannerInfo *root, * be found. */ ec1->ec_members = list_concat(ec1->ec_members, ec2->ec_members); - ec1->ec_sources = list_concat(ec1->ec_sources, ec2->ec_sources); - ec1->ec_derives = list_concat(ec1->ec_derives, ec2->ec_derives); + ec1->ec_source_indexes = bms_join(ec1->ec_source_indexes, + ec2->ec_source_indexes); + ec1->ec_derive_indexes = bms_join(ec1->ec_derive_indexes, + ec2->ec_derive_indexes); ec1->ec_relids = bms_join(ec1->ec_relids, ec2->ec_relids); ec1->ec_has_const |= ec2->ec_has_const; /* can't need to set has_volatile */ @@ -364,10 +371,9 @@ process_equivalence(PlannerInfo *root, root->eq_classes = list_delete_nth_cell(root->eq_classes, ec2_idx); /* just to avoid debugging confusion w/ dangling pointers: */ ec2->ec_members = NIL; - ec2->ec_sources = NIL; - ec2->ec_derives = NIL; + ec2->ec_source_indexes = NULL; + ec2->ec_derive_indexes = NULL; ec2->ec_relids = NULL; - ec1->ec_sources = lappend(ec1->ec_sources, restrictinfo); ec1->ec_min_security = Min(ec1->ec_min_security, restrictinfo->security_level); ec1->ec_max_security = Max(ec1->ec_max_security, @@ -378,13 +384,14 @@ process_equivalence(PlannerInfo *root, /* mark the RI as usable with this pair of EMs */ restrictinfo->left_em = em1; restrictinfo->right_em = em2; + + add_eq_source(root, ec1, restrictinfo); } else if (ec1) { /* Case 3: add item2 to ec1 */ em2 = add_eq_member(ec1, item2, item2_relids, jdomain, item2_type); - ec1->ec_sources = lappend(ec1->ec_sources, restrictinfo); ec1->ec_min_security = Min(ec1->ec_min_security, restrictinfo->security_level); ec1->ec_max_security = Max(ec1->ec_max_security, @@ -395,13 +402,14 @@ process_equivalence(PlannerInfo *root, /* mark the RI as usable with this pair of EMs */ restrictinfo->left_em = em1; restrictinfo->right_em = em2; + + add_eq_source(root, ec1, restrictinfo); } else if (ec2) { /* Case 3: add item1 to ec2 */ em1 = add_eq_member(ec2, item1, item1_relids, jdomain, item1_type); - ec2->ec_sources = lappend(ec2->ec_sources, restrictinfo); ec2->ec_min_security = Min(ec2->ec_min_security, restrictinfo->security_level); ec2->ec_max_security = Max(ec2->ec_max_security, @@ -412,6 +420,8 @@ process_equivalence(PlannerInfo *root, /* mark the RI as usable with this pair of EMs */ restrictinfo->left_em = em1; restrictinfo->right_em = em2; + + add_eq_source(root, ec2, restrictinfo); } else { @@ -421,8 +431,8 @@ process_equivalence(PlannerInfo *root, ec->ec_opfamilies = opfamilies; ec->ec_collation = collation; ec->ec_members = NIL; - ec->ec_sources = list_make1(restrictinfo); - ec->ec_derives = NIL; + ec->ec_source_indexes = NULL; + ec->ec_derive_indexes = NULL; ec->ec_relids = NULL; ec->ec_has_const = false; ec->ec_has_volatile = false; @@ -444,6 +454,8 @@ process_equivalence(PlannerInfo *root, /* mark the RI as usable with this pair of EMs */ restrictinfo->left_em = em1; restrictinfo->right_em = em2; + + add_eq_source(root, ec, restrictinfo); } return true; @@ -578,6 +590,170 @@ add_eq_member(EquivalenceClass *ec, Expr *expr, Relids relids, return em; } +/* + * add_eq_source - add 'rinfo' in eq_sources for this 'ec' + */ +static void +add_eq_source(PlannerInfo *root, EquivalenceClass *ec, RestrictInfo *rinfo) +{ + int source_idx = list_length(root->eq_sources); + int i; + + ec->ec_source_indexes = bms_add_member(ec->ec_source_indexes, source_idx); + root->eq_sources = lappend(root->eq_sources, rinfo); + Assert(rinfo->eq_sources_index == -1); + rinfo->eq_sources_index = source_idx; + + i = -1; + while ((i = bms_next_member(rinfo->clause_relids, i)) >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[i]; + + rte->eclass_source_indexes = bms_add_member(rte->eclass_source_indexes, + source_idx); + } +} + +/* + * add_eq_derive - add 'rinfo' in eq_derives for this 'ec' + */ +static void +add_eq_derive(PlannerInfo *root, EquivalenceClass *ec, RestrictInfo *rinfo) +{ + int derive_idx = list_length(root->eq_derives); + int i; + + ec->ec_derive_indexes = bms_add_member(ec->ec_derive_indexes, derive_idx); + root->eq_derives = lappend(root->eq_derives, rinfo); + Assert(rinfo->eq_derives_index == -1); + rinfo->eq_derives_index = derive_idx; + + i = -1; + while ((i = bms_next_member(rinfo->clause_relids, i)) >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[i]; + + rte->eclass_derive_indexes = bms_add_member(rte->eclass_derive_indexes, + derive_idx); + } +} + +/* + * update_clause_relids + * Update RestrictInfo->clause_relids with adjusting the relevant indexes. + * The clause_relids field must be updated through this function, not by + * setting the field directly. + */ +void +update_clause_relids(PlannerInfo *root, RestrictInfo *rinfo, + Relids new_clause_relids) +{ + int i; + Relids removing_relids; + Relids adding_relids; +#ifdef USE_ASSERT_CHECKING + Relids common_relids; +#endif + + /* + * If there is nothing to do, we return immediately after setting the + * clause_relids field. + */ + if (rinfo->eq_sources_index == -1 && rinfo->eq_derives_index == -1) + { + rinfo->clause_relids = new_clause_relids; + return; + } + + /* + * Remove any references between this RestrictInfo and RangeTblEntries + * that are no longer relevant. + */ + removing_relids = bms_difference(rinfo->clause_relids, new_clause_relids); + i = -1; + while ((i = bms_next_member(removing_relids, i)) >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[i]; + + if (rinfo->eq_sources_index != -1) + { + Assert(bms_is_member(rinfo->eq_sources_index, + rte->eclass_source_indexes)); + rte->eclass_source_indexes = + bms_del_member(rte->eclass_source_indexes, + rinfo->eq_sources_index); + } + if (rinfo->eq_derives_index != -1) + { + Assert(bms_is_member(rinfo->eq_derives_index, + rte->eclass_derive_indexes)); + rte->eclass_derive_indexes = + bms_del_member(rte->eclass_derive_indexes, + rinfo->eq_derives_index); + } + } + bms_free(removing_relids); + + /* + * Add references between this RestrictInfo and RangeTblEntries that will + * be relevant. + */ + adding_relids = bms_difference(new_clause_relids, rinfo->clause_relids); + i = -1; + while ((i = bms_next_member(adding_relids, i)) >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[i]; + + if (rinfo->eq_sources_index != -1) + { + Assert(!bms_is_member(rinfo->eq_sources_index, + rte->eclass_source_indexes)); + rte->eclass_source_indexes = + bms_add_member(rte->eclass_source_indexes, + rinfo->eq_sources_index); + } + if (rinfo->eq_derives_index != -1) + { + Assert(!bms_is_member(rinfo->eq_derives_index, + rte->eclass_derive_indexes)); + rte->eclass_derive_indexes = + bms_add_member(rte->eclass_derive_indexes, + rinfo->eq_derives_index); + } + } + bms_free(adding_relids); + +#ifdef USE_ASSERT_CHECKING + /* + * Verify that all indexes are set for common Relids. + */ + common_relids = bms_intersect(rinfo->clause_relids, new_clause_relids); + i = -1; + while ((i = bms_next_member(common_relids, i)) >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[i]; + + if (rinfo->eq_sources_index != -1) + { + Assert(bms_is_member(rinfo->eq_sources_index, + rte->eclass_source_indexes)); + } + if (rinfo->eq_derives_index != -1) + { + Assert(bms_is_member(rinfo->eq_derives_index, + rte->eclass_derive_indexes)); + } + } + bms_free(common_relids); +#endif + + /* + * Since we have done everything to adjust the indexes, we can set the + * clause_relids field. + */ + rinfo->clause_relids = new_clause_relids; +} + /* * get_eclass_for_sort_expr @@ -736,8 +912,8 @@ get_eclass_for_sort_expr(PlannerInfo *root, newec->ec_opfamilies = list_copy(opfamilies); newec->ec_collation = collation; newec->ec_members = NIL; - newec->ec_sources = NIL; - newec->ec_derives = NIL; + newec->ec_source_indexes = NULL; + newec->ec_derive_indexes = NULL; newec->ec_relids = NULL; newec->ec_has_const = false; newec->ec_has_volatile = contain_volatile_functions((Node *) expr); @@ -1115,7 +1291,7 @@ relation_can_be_sorted_early(PlannerInfo *root, RelOptInfo *rel, * scanning of the quals and before Path construction begins. * * We make no attempt to avoid generating duplicate RestrictInfos here: we - * don't search ec_sources or ec_derives for matches. It doesn't really + * don't search eq_sources or eq_derives for matches. It doesn't really * seem worth the trouble to do so. */ void @@ -1204,6 +1380,7 @@ generate_base_implied_equalities_const(PlannerInfo *root, { EquivalenceMember *const_em = NULL; ListCell *lc; + int i; /* * In the trivial case where we just had one "var = const" clause, push @@ -1213,9 +1390,9 @@ generate_base_implied_equalities_const(PlannerInfo *root, * equivalent to the old one. */ if (list_length(ec->ec_members) == 2 && - list_length(ec->ec_sources) == 1) + bms_get_singleton_member(ec->ec_source_indexes, &i)) { - RestrictInfo *restrictinfo = (RestrictInfo *) linitial(ec->ec_sources); + RestrictInfo *restrictinfo = list_nth_node(RestrictInfo, root->eq_sources, i); distribute_restrictinfo_to_rels(root, restrictinfo); return; @@ -1273,9 +1450,9 @@ generate_base_implied_equalities_const(PlannerInfo *root, /* * If the clause didn't degenerate to a constant, fill in the correct - * markings for a mergejoinable clause, and save it in ec_derives. (We + * markings for a mergejoinable clause, and save it in eq_derives. (We * will not re-use such clauses directly, but selectivity estimation - * may consult the list later. Note that this use of ec_derives does + * may consult the list later. Note that this use of eq_derives does * not overlap with its use for join clauses, since we never generate * join clauses from an ec_has_const eclass.) */ @@ -1285,7 +1462,8 @@ generate_base_implied_equalities_const(PlannerInfo *root, rinfo->left_ec = rinfo->right_ec = ec; rinfo->left_em = cur_em; rinfo->right_em = const_em; - ec->ec_derives = lappend(ec->ec_derives, rinfo); + + add_eq_derive(root, ec, rinfo); } } } @@ -1351,7 +1529,7 @@ generate_base_implied_equalities_no_const(PlannerInfo *root, /* * If the clause didn't degenerate to a constant, fill in the * correct markings for a mergejoinable clause. We don't put it - * in ec_derives however; we don't currently need to re-find such + * in eq_derives however; we don't currently need to re-find such * clauses, and we don't want to clutter that list with non-join * clauses. */ @@ -1407,11 +1585,12 @@ static void generate_base_implied_equalities_broken(PlannerInfo *root, EquivalenceClass *ec) { - ListCell *lc; + int i = -1; - foreach(lc, ec->ec_sources) + while ((i = bms_next_member(ec->ec_source_indexes, i)) >= 0) { - RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(lc); + RestrictInfo *restrictinfo = list_nth_node(RestrictInfo, + root->eq_sources, i); if (ec->ec_has_const || bms_membership(restrictinfo->required_relids) != BMS_MULTIPLE) @@ -1452,11 +1631,11 @@ generate_base_implied_equalities_broken(PlannerInfo *root, * Because the same join clauses are likely to be needed multiple times as * we consider different join paths, we avoid generating multiple copies: * whenever we select a particular pair of EquivalenceMembers to join, - * we check to see if the pair matches any original clause (in ec_sources) - * or previously-built clause (in ec_derives). This saves memory and allows - * re-use of information cached in RestrictInfos. We also avoid generating - * commutative duplicates, i.e. if the algorithm selects "a.x = b.y" but - * we already have "b.y = a.x", we return the existing clause. + * we check to see if the pair matches any original clause in root's + * eq_sources or previously-built clause (in root's eq_derives). This saves + * memory and allows re-use of information cached in RestrictInfos. We also + * avoid generating commutative duplicates, i.e. if the algorithm selects + * "a.x = b.y" but we already have "b.y = a.x", we return the existing clause. * * If we are considering an outer join, sjinfo is the associated OJ info, * otherwise it can be NULL. @@ -1835,12 +2014,16 @@ generate_join_implied_equalities_broken(PlannerInfo *root, Relids nominal_inner_relids, RelOptInfo *inner_rel) { + Bitmapset *matching_es; List *result = NIL; - ListCell *lc; + int i; - foreach(lc, ec->ec_sources) + matching_es = get_ec_source_indexes(root, ec, nominal_join_relids); + i = -1; + while ((i = bms_next_member(matching_es, i)) >= 0) { - RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(lc); + RestrictInfo *restrictinfo = list_nth_node(RestrictInfo, + root->eq_sources, i); Relids clause_relids = restrictinfo->required_relids; if (bms_is_subset(clause_relids, nominal_join_relids) && @@ -1852,12 +2035,12 @@ generate_join_implied_equalities_broken(PlannerInfo *root, /* * If we have to translate, just brute-force apply adjust_appendrel_attrs * to all the RestrictInfos at once. This will result in returning - * RestrictInfos that are not listed in ec_derives, but there shouldn't be + * RestrictInfos that are not listed in eq_derives, but there shouldn't be * any duplication, and it's a sufficiently narrow corner case that we * shouldn't sweat too much over it anyway. * * Since inner_rel might be an indirect descendant of the baserel - * mentioned in the ec_sources clauses, we have to be prepared to apply + * mentioned in the eq_sources clauses, we have to be prepared to apply * multiple levels of Var translation. */ if (IS_OTHER_REL(inner_rel) && result != NIL) @@ -1919,10 +2102,11 @@ create_join_clause(PlannerInfo *root, EquivalenceMember *rightem, EquivalenceClass *parent_ec) { + Bitmapset *matches; RestrictInfo *rinfo; RestrictInfo *parent_rinfo = NULL; - ListCell *lc; MemoryContext oldcontext; + int i; /* * Search to see if we already built a RestrictInfo for this pair of @@ -1933,9 +2117,12 @@ create_join_clause(PlannerInfo *root, * it's not identical, it'd better have the same effects, or the operator * families we're using are broken. */ - foreach(lc, ec->ec_sources) + matches = bms_int_members(get_ec_source_indexes_strict(root, ec, leftem->em_relids), + get_ec_source_indexes_strict(root, ec, rightem->em_relids)); + i = -1; + while ((i = bms_next_member(matches, i)) >= 0) { - rinfo = (RestrictInfo *) lfirst(lc); + rinfo = list_nth_node(RestrictInfo, root->eq_sources, i); if (rinfo->left_em == leftem && rinfo->right_em == rightem && rinfo->parent_ec == parent_ec) @@ -1946,9 +2133,13 @@ create_join_clause(PlannerInfo *root, return rinfo; } - foreach(lc, ec->ec_derives) + matches = bms_int_members(get_ec_derive_indexes_strict(root, ec, leftem->em_relids), + get_ec_derive_indexes_strict(root, ec, rightem->em_relids)); + + i = -1; + while ((i = bms_next_member(matches, i)) >= 0) { - rinfo = (RestrictInfo *) lfirst(lc); + rinfo = list_nth_node(RestrictInfo, root->eq_derives, i); if (rinfo->left_em == leftem && rinfo->right_em == rightem && rinfo->parent_ec == parent_ec) @@ -2006,7 +2197,7 @@ create_join_clause(PlannerInfo *root, rinfo->left_em = leftem; rinfo->right_em = rightem; /* and save it for possible re-use */ - ec->ec_derives = lappend(ec->ec_derives, rinfo); + add_eq_derive(root, ec, rinfo); MemoryContextSwitchTo(oldcontext); @@ -2682,16 +2873,19 @@ match_eclasses_to_foreign_key_col(PlannerInfo *root, * Returns NULL if no such clause can be found. */ RestrictInfo * -find_derived_clause_for_ec_member(EquivalenceClass *ec, +find_derived_clause_for_ec_member(PlannerInfo *root, EquivalenceClass *ec, EquivalenceMember *em) { - ListCell *lc; + int i; Assert(ec->ec_has_const); Assert(!em->em_is_const); - foreach(lc, ec->ec_derives) + + i = -1; + while ((i = bms_next_member(ec->ec_derive_indexes, i)) >= 0) { - RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc); + RestrictInfo *rinfo = list_nth_node(RestrictInfo, root->eq_derives, + i); /* * generate_base_implied_equalities_const will have put non-const @@ -3325,7 +3519,7 @@ have_relevant_eclass_joinclause(PlannerInfo *root, * surely be true if both of them overlap ec_relids.) * * Note we don't test ec_broken; if we did, we'd need a separate code - * path to look through ec_sources. Checking the membership anyway is + * path to look through eq_sources. Checking the membership anyway is * OK as a possibly-overoptimistic heuristic. * * We don't test ec_has_const either, even though a const eclass won't @@ -3413,7 +3607,7 @@ eclass_useful_for_merging(PlannerInfo *root, /* * Note we don't test ec_broken; if we did, we'd need a separate code path - * to look through ec_sources. Checking the members anyway is OK as a + * to look through eq_sources. Checking the members anyway is OK as a * possibly-overoptimistic heuristic. */ @@ -3566,3 +3760,240 @@ get_common_eclass_indexes(PlannerInfo *root, Relids relids1, Relids relids2) /* Calculate and return the common EC indexes, recycling the left input. */ return bms_int_members(rel1ecs, rel2ecs); } + +/* + * get_ec_source_indexes + * Returns a Bitmapset with indexes into root->eq_sources for all + * RestrictInfos in 'ec' that have + * bms_overlap(relids, rinfo->clause_relids) as true. + * + * Returns a newly allocated Bitmapset which the caller is free to modify. + */ +Bitmapset * +get_ec_source_indexes(PlannerInfo *root, EquivalenceClass *ec, Relids relids) +{ + Bitmapset *rel_esis = NULL; + int i = -1; + + while ((i = bms_next_member(relids, i)) >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[i]; + + rel_esis = bms_add_members(rel_esis, rte->eclass_source_indexes); + } + +#ifdef USE_ASSERT_CHECKING + /* verify the results look sane */ + i = -1; + while ((i = bms_next_member(rel_esis, i)) >= 0) + { + RestrictInfo *rinfo = list_nth_node(RestrictInfo, root->eq_sources, + i); + + Assert(bms_overlap(relids, rinfo->clause_relids)); + } +#endif + + /* bitwise-AND to leave only the ones for this EquivalenceClass */ + return bms_int_members(rel_esis, ec->ec_source_indexes); +} + +/* + * get_ec_source_indexes_strict + * Returns a Bitmapset with indexes into root->eq_sources for all + * RestrictInfos in 'ec' that have + * bms_is_subset(relids, rinfo->clause_relids) as true. + * + * Returns a newly allocated Bitmapset which the caller is free to modify. + */ +Bitmapset * +get_ec_source_indexes_strict(PlannerInfo *root, EquivalenceClass *ec, + Relids relids) +{ + Bitmapset *esis = NULL; + int i = bms_next_member(relids, -1); + + if (i >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[i]; + + /* + * bms_intersect to the first relation to try to keep the resulting + * Bitmapset as small as possible. This saves having to make a + * complete bms_copy() of one of them. One may contain significantly + * more words than the other. + */ + esis = bms_intersect(ec->ec_source_indexes, + rte->eclass_source_indexes); + + while ((i = bms_next_member(relids, i)) >= 0) + { + rte = root->simple_rte_array[i]; + esis = bms_int_members(esis, rte->eclass_source_indexes); + } + } + +#ifdef USE_ASSERT_CHECKING + /* verify the results look sane */ + i = -1; + while ((i = bms_next_member(esis, i)) >= 0) + { + RestrictInfo *rinfo = list_nth_node(RestrictInfo, root->eq_sources, + i); + + Assert(bms_is_subset(relids, rinfo->clause_relids)); + } +#endif + + return esis; +} + +/* + * get_ec_derive_indexes + * Returns a Bitmapset with indexes into root->eq_derives for all + * RestrictInfos in 'ec' that have + * bms_overlap(relids, rinfo->clause_relids) as true. + * + * Returns a newly allocated Bitmapset which the caller is free to modify. + * + * XXX is this function even needed? + */ +Bitmapset * +get_ec_derive_indexes(PlannerInfo *root, EquivalenceClass *ec, Relids relids) +{ + Bitmapset *rel_edis = NULL; + int i = -1; + + while ((i = bms_next_member(relids, i)) >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[i]; + + rel_edis = bms_add_members(rel_edis, rte->eclass_derive_indexes); + } + +#ifdef USE_ASSERT_CHECKING + /* verify the results look sane */ + i = -1; + while ((i = bms_next_member(rel_edis, i)) >= 0) + { + RestrictInfo *rinfo = list_nth_node(RestrictInfo, root->eq_derives, + i); + + Assert(bms_overlap(relids, rinfo->clause_relids)); + } +#endif + + /* bitwise-AND to leave only the ones for this EquivalenceClass */ + return bms_int_members(rel_edis, ec->ec_derive_indexes); +} + +/* + * get_ec_derive_indexes_strict + * Returns a Bitmapset with indexes into root->eq_derives for all + * RestrictInfos in 'ec' that have + * bms_is_subset(relids, rinfo->clause_relids) as true. + * + * Returns a newly allocated Bitmapset which the caller is free to modify. + */ +Bitmapset * +get_ec_derive_indexes_strict(PlannerInfo *root, EquivalenceClass *ec, + Relids relids) +{ + Bitmapset *edis = NULL; + int i = bms_next_member(relids, -1); + + if (i >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[i]; + + /* + * bms_intersect to the first relation to try to keep the resulting + * Bitmapset as small as possible. This saves having to make a + * complete bms_copy() of one of them. One may contain significantly + * more words than the other. + */ + edis = bms_intersect(ec->ec_derive_indexes, + rte->eclass_derive_indexes); + + while ((i = bms_next_member(relids, i)) >= 0) + { + rte = root->simple_rte_array[i]; + edis = bms_int_members(edis, rte->eclass_derive_indexes); + } + } + +#ifdef USE_ASSERT_CHECKING + /* verify the results look sane */ + i = -1; + while ((i = bms_next_member(edis, i)) >= 0) + { + RestrictInfo *rinfo = list_nth_node(RestrictInfo, root->eq_derives, + i); + + Assert(bms_is_subset(relids, rinfo->clause_relids)); + } +#endif + + return edis; +} + +#ifdef USE_ASSERT_CHECKING +/* + * verify_eclass_indexes + * Verify that there are no missing references between RestrictInfos and + * EquivalenceMember's indexes, namely eclass_source_indexes and + * eclass_derive_indexes. If you modify these indexes, you should check + * them with this function. + */ +void +verify_eclass_indexes(PlannerInfo *root, EquivalenceClass *ec) +{ + ListCell *lc; + + /* + * All RestrictInfos in root->eq_sources must have references to + * eclass_source_indexes. + */ + foreach(lc, root->eq_sources) + { + RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc); + int index; + int k; + + /* Deleted members are marked NULL */ + if (rinfo == NULL) + continue; + + index = list_cell_number(root->eq_sources, lc); + k = -1; + while ((k = bms_next_member(rinfo->clause_relids, k)) >= 0) + { + /* must have a reference */ + Assert(bms_is_member(index, root->simple_rte_array[k]->eclass_source_indexes)); + } + } + + /* + * All RestrictInfos in root->eq_derives must have references to + * eclass_derive_indexes. + */ + foreach(lc, root->eq_derives) + { + RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc); + int index; + int k; + + /* Deleted members are marked NULL */ + if (rinfo == NULL) + continue; + + index = list_cell_number(root->eq_derives, lc); + k = -1; + while ((k = bms_next_member(rinfo->clause_relids, k)) >= 0) + { + /* must have a reference */ + Assert(bms_is_member(index, root->simple_rte_array[k]->eclass_derive_indexes)); + } + } +} +#endif diff --git a/src/backend/optimizer/plan/analyzejoins.c b/src/backend/optimizer/plan/analyzejoins.c index 4e72741997..13e987493b 100644 --- a/src/backend/optimizer/plan/analyzejoins.c +++ b/src/backend/optimizer/plan/analyzejoins.c @@ -50,7 +50,8 @@ static bool join_is_removable(PlannerInfo *root, SpecialJoinInfo *sjinfo); static void remove_leftjoinrel_from_query(PlannerInfo *root, int relid, SpecialJoinInfo *sjinfo); -static void remove_rel_from_restrictinfo(RestrictInfo *rinfo, +static void remove_rel_from_restrictinfo(PlannerInfo *root, + RestrictInfo *rinfo, int relid, int ojrelid); static void remove_rel_from_eclass(PlannerInfo *root, EquivalenceClass *ec, int relid, int ojrelid); @@ -66,7 +67,7 @@ static bool is_innerrel_unique_for(PlannerInfo *root, JoinType jointype, List *restrictlist, List **extra_clauses); -static void replace_varno(Node *node, int from, int to); +static void replace_varno(PlannerInfo *root, Node *node, int from, int to); static Bitmapset *replace_relid(Relids relids, int oldId, int newId); static int self_join_candidates_cmp(const void *a, const void *b); @@ -385,7 +386,7 @@ remove_rel_from_query(PlannerInfo *root, RelOptInfo *rel, } /* Update lateral_vars list. */ - replace_varno((Node *) otherrel->lateral_vars, relid, subst); + replace_varno(root, (Node *) otherrel->lateral_vars, relid, subst); } /* @@ -422,7 +423,7 @@ remove_rel_from_query(PlannerInfo *root, RelOptInfo *rel, sjinf->commute_below_l = replace_relid(sjinf->commute_below_l, ojrelid, subst); sjinf->commute_below_r = replace_relid(sjinf->commute_below_r, ojrelid, subst); - replace_varno((Node *) sjinf->semi_rhs_exprs, relid, subst); + replace_varno(root, (Node *) sjinf->semi_rhs_exprs, relid, subst); } /* @@ -467,7 +468,7 @@ remove_rel_from_query(PlannerInfo *root, RelOptInfo *rel, phv->phrels = replace_relid(phv->phrels, relid, subst); phv->phrels = replace_relid(phv->phrels, ojrelid, subst); Assert(!bms_is_empty(phv->phrels)); - replace_varno((Node *) phv->phexpr, relid, subst); + replace_varno(root, (Node *) phv->phexpr, relid, subst); Assert(phv->phnullingrels == NULL); /* no need to adjust */ } } @@ -541,7 +542,7 @@ remove_leftjoinrel_from_query(PlannerInfo *root, int relid, * that any such PHV is safe (and updated its ph_eval_at), so we * can just drop those references. */ - remove_rel_from_restrictinfo(rinfo, relid, ojrelid); + remove_rel_from_restrictinfo(root, rinfo, relid, ojrelid); /* * Cross-check that the clause itself does not reference the @@ -598,16 +599,24 @@ remove_leftjoinrel_from_query(PlannerInfo *root, int relid, * we have to also clean up the sub-clauses. */ static void -remove_rel_from_restrictinfo(RestrictInfo *rinfo, int relid, int ojrelid) +remove_rel_from_restrictinfo(PlannerInfo *root, RestrictInfo *rinfo, int relid, + int ojrelid) { + Relids new_clause_relids; + + /* + * The 'new_clause_relids' passed to update_clause_relids() must be a + * different instance from the current rinfo->clause_relids, so we make a + * copy of them. + */ + new_clause_relids = bms_copy(rinfo->clause_relids); + new_clause_relids = bms_del_member(new_clause_relids, relid); + new_clause_relids = bms_del_member(new_clause_relids, ojrelid); + update_clause_relids(root, rinfo, new_clause_relids); /* - * The clause_relids probably aren't shared with anything else, but let's + * The required_relids probably aren't shared with anything else, but let's * copy them just to be sure. */ - rinfo->clause_relids = bms_copy(rinfo->clause_relids); - rinfo->clause_relids = bms_del_member(rinfo->clause_relids, relid); - rinfo->clause_relids = bms_del_member(rinfo->clause_relids, ojrelid); - /* Likewise for required_relids */ rinfo->required_relids = bms_copy(rinfo->required_relids); rinfo->required_relids = bms_del_member(rinfo->required_relids, relid); rinfo->required_relids = bms_del_member(rinfo->required_relids, ojrelid); @@ -632,14 +641,14 @@ remove_rel_from_restrictinfo(RestrictInfo *rinfo, int relid, int ojrelid) { RestrictInfo *rinfo2 = lfirst_node(RestrictInfo, lc2); - remove_rel_from_restrictinfo(rinfo2, relid, ojrelid); + remove_rel_from_restrictinfo(root, rinfo2, relid, ojrelid); } } else { RestrictInfo *rinfo2 = castNode(RestrictInfo, orarg); - remove_rel_from_restrictinfo(rinfo2, relid, ojrelid); + remove_rel_from_restrictinfo(root, rinfo2, relid, ojrelid); } } } @@ -659,6 +668,7 @@ remove_rel_from_eclass(PlannerInfo *root, EquivalenceClass *ec, int relid, int ojrelid) { ListCell *lc; + int i; /* Fix up the EC's overall relids */ ec->ec_relids = bms_del_member(ec->ec_relids, relid); @@ -685,11 +695,12 @@ remove_rel_from_eclass(PlannerInfo *root, EquivalenceClass *ec, int relid, } /* Fix up the source clauses, in case we can re-use them later */ - foreach(lc, ec->ec_sources) + i = -1; + while ((i = bms_next_member(ec->ec_source_indexes, i)) >= 0) { - RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc); + RestrictInfo *rinfo = list_nth_node(RestrictInfo, root->eq_sources, i); - remove_rel_from_restrictinfo(rinfo, relid, ojrelid); + remove_rel_from_restrictinfo(root, rinfo, relid, ojrelid); } /* @@ -697,7 +708,7 @@ remove_rel_from_eclass(PlannerInfo *root, EquivalenceClass *ec, int relid, * drop them. (At this point, any such clauses would be base restriction * clauses, which we'd not need anymore anyway.) */ - ec->ec_derives = NIL; + ec->ec_derive_indexes = NULL; } /* @@ -1436,6 +1447,7 @@ is_innerrel_unique_for(PlannerInfo *root, typedef struct { + PlannerInfo *root; int from; int to; int sublevels_up; @@ -1494,10 +1506,10 @@ replace_varno_walker(Node *node, ReplaceVarnoContext *ctx) if (bms_is_member(ctx->from, rinfo->clause_relids)) { - replace_varno((Node *) rinfo->clause, ctx->from, ctx->to); - replace_varno((Node *) rinfo->orclause, ctx->from, ctx->to); - rinfo->clause_relids = - replace_relid(rinfo->clause_relids, ctx->from, ctx->to); + replace_varno(ctx->root, (Node *) rinfo->clause, ctx->from, ctx->to); + replace_varno(ctx->root, (Node *) rinfo->orclause, ctx->from, ctx->to); + update_clause_relids(ctx->root, rinfo, + replace_relid(rinfo->clause_relids, ctx->from, ctx->to)); rinfo->left_relids = replace_relid(rinfo->left_relids, ctx->from, ctx->to); rinfo->right_relids = @@ -1547,13 +1559,14 @@ replace_varno_walker(Node *node, ReplaceVarnoContext *ctx) } static void -replace_varno(Node *node, int from, int to) +replace_varno(PlannerInfo *root, Node *node, int from, int to) { ReplaceVarnoContext ctx; if (to <= 0) return; + ctx.root = root; ctx.from = from; ctx.to = to; ctx.sublevels_up = 0; @@ -1614,10 +1627,12 @@ replace_relid(Relids relids, int oldId, int newId) * delete them. */ static void -update_eclasses(EquivalenceClass *ec, int from, int to) +update_eclasses(PlannerInfo *root, EquivalenceClass *ec, int from, int to) { List *new_members = NIL; - List *new_sources = NIL; + Bitmapset *new_source_indexes = NULL; + int i; + int j; ListCell *lc; ListCell *lc1; @@ -1636,7 +1651,7 @@ update_eclasses(EquivalenceClass *ec, int from, int to) em->em_jdomain->jd_relids = replace_relid(em->em_jdomain->jd_relids, from, to); /* We only process inner joins */ - replace_varno((Node *) em->em_expr, from, to); + replace_varno(root, (Node *) em->em_expr, from, to); foreach(lc1, new_members) { @@ -1659,31 +1674,64 @@ update_eclasses(EquivalenceClass *ec, int from, int to) list_free(ec->ec_members); ec->ec_members = new_members; - list_free(ec->ec_derives); - ec->ec_derives = NULL; + i = -1; + while ((i = bms_next_member(ec->ec_derive_indexes, i)) >= 0) + { + RestrictInfo *rinfo = list_nth_node(RestrictInfo, root->eq_derives, i); + + /* + * Remove all references between this RestrictInfo and its relating + * RangeTblEntry. + */ + j = -1; + while ((j = bms_next_member(rinfo->clause_relids, j)) >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[j]; + + Assert(bms_is_member(i, rte->eclass_derive_indexes)); + rte->eclass_derive_indexes = + bms_del_member(rte->eclass_derive_indexes, i); + } + + /* + * Can't delete the element because we would need to rebuild all + * the eq_derives indexes. But set a null to detect potential problems. + */ + list_nth_cell(root->eq_derives, i)->ptr_value = NULL; + + /* + * Since this RestrictInfo no longer exists in root->eq_derives, we + * must reset the stored index. + */ + rinfo->eq_derives_index = -1; + } + bms_free(ec->ec_derive_indexes); + ec->ec_derive_indexes = NULL; /* Update EC source expressions */ - foreach(lc, ec->ec_sources) + i = -1; + while ((i = bms_next_member(ec->ec_source_indexes, i)) >= 0) { - RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc); + RestrictInfo *rinfo = list_nth_node(RestrictInfo, root->eq_sources, i); bool is_redundant = false; if (!bms_is_member(from, rinfo->required_relids)) { - new_sources = lappend(new_sources, rinfo); + new_source_indexes = bms_add_member(new_source_indexes, i); continue; } - replace_varno((Node *) rinfo, from, to); + replace_varno(root, (Node *) rinfo, from, to); /* * After switching the clause to the remaining relation, check it for * redundancy with existing ones. We don't have to check for * redundancy with derived clauses, because we've just deleted them. */ - foreach(lc1, new_sources) + j = -1; + while ((j = bms_next_member(new_source_indexes, j)) >= 0) { - RestrictInfo *other = lfirst_node(RestrictInfo, lc1); + RestrictInfo *other = list_nth_node(RestrictInfo, root->eq_sources, j); if (!equal(rinfo->clause_relids, other->clause_relids)) continue; @@ -1695,13 +1743,46 @@ update_eclasses(EquivalenceClass *ec, int from, int to) } } - if (!is_redundant) - new_sources = lappend(new_sources, rinfo); + if (is_redundant) + { + /* + * Remove all references between this RestrictInfo and its relating + * RangeTblEntry. + */ + j = -1; + while ((j = bms_next_member(rinfo->clause_relids, j)) >= 0) + { + RangeTblEntry *rte = root->simple_rte_array[j]; + + Assert(bms_is_member(i, rte->eclass_source_indexes)); + rte->eclass_source_indexes = + bms_del_member(rte->eclass_source_indexes, i); + } + + /* + * Can't delete the element because we would need to rebuild all + * the eq_sources indexes. But set a null to detect potential problems. + */ + list_nth_cell(root->eq_sources, i)->ptr_value = NULL; + + /* + * Since this RestrictInfo no longer exists in root->eq_sources, we + * must reset the stored index. + */ + rinfo->eq_sources_index = -1; + } + else + new_source_indexes = bms_add_member(new_source_indexes, i); } - list_free(ec->ec_sources); - ec->ec_sources = new_sources; + bms_free(ec->ec_source_indexes); + ec->ec_source_indexes = new_source_indexes; ec->ec_relids = replace_relid(ec->ec_relids, from, to); + +#ifdef USE_ASSERT_CHECKING + /* Make sure that we didn't break EquivalenceClass indexes */ + verify_eclass_indexes(root, ec); +#endif } /* @@ -1771,7 +1852,7 @@ remove_self_join_rel(PlannerInfo *root, PlanRowMark *kmark, PlanRowMark *rmark, RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc); remove_join_clause_from_rels(root, rinfo, rinfo->required_relids); - replace_varno((Node *) rinfo, toRemove->relid, toKeep->relid); + replace_varno(root, (Node *) rinfo, toRemove->relid, toKeep->relid); if (bms_membership(rinfo->required_relids) == BMS_MULTIPLE) jinfo_candidates = lappend(jinfo_candidates, rinfo); @@ -1791,7 +1872,7 @@ remove_self_join_rel(PlannerInfo *root, PlanRowMark *kmark, PlanRowMark *rmark, { RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc); - replace_varno((Node *) rinfo, toRemove->relid, toKeep->relid); + replace_varno(root, (Node *) rinfo, toRemove->relid, toKeep->relid); if (bms_membership(rinfo->required_relids) == BMS_MULTIPLE) jinfo_candidates = lappend(jinfo_candidates, rinfo); @@ -1879,7 +1960,7 @@ remove_self_join_rel(PlannerInfo *root, PlanRowMark *kmark, PlanRowMark *rmark, { EquivalenceClass *ec = (EquivalenceClass *) list_nth(root->eq_classes, i); - update_eclasses(ec, toRemove->relid, toKeep->relid); + update_eclasses(root, ec, toRemove->relid, toKeep->relid); toKeep->eclass_indexes = bms_add_member(toKeep->eclass_indexes, i); } @@ -1891,7 +1972,7 @@ remove_self_join_rel(PlannerInfo *root, PlanRowMark *kmark, PlanRowMark *rmark, { Node *node = lfirst(lc); - replace_varno(node, toRemove->relid, toKeep->relid); + replace_varno(root, node, toRemove->relid, toKeep->relid); if (!list_member(toKeep->reltarget->exprs, node)) toKeep->reltarget->exprs = lappend(toKeep->reltarget->exprs, node); } @@ -1932,7 +2013,7 @@ remove_self_join_rel(PlannerInfo *root, PlanRowMark *kmark, PlanRowMark *rmark, } /* Replace varno in all the query structures */ - replace_varno((Node *) root->parse, toRemove->relid, toKeep->relid); + replace_varno(root, (Node *) root->parse, toRemove->relid, toKeep->relid); /* See remove_self_joins_one_group() */ Assert(root->parse->resultRelation != toRemove->relid); @@ -1942,9 +2023,9 @@ remove_self_join_rel(PlannerInfo *root, PlanRowMark *kmark, PlanRowMark *rmark, remove_rel_from_query(root, toRemove, toKeep->relid, NULL, NULL); /* At last, replace varno in root targetlist and HAVING clause */ - replace_varno((Node *) root->processed_tlist, + replace_varno(root, (Node *) root->processed_tlist, toRemove->relid, toKeep->relid); - replace_varno((Node *) root->processed_groupClause, + replace_varno(root, (Node *) root->processed_groupClause, toRemove->relid, toKeep->relid); replace_relid(root->all_result_relids, toRemove->relid, toKeep->relid); replace_relid(root->leaf_result_relids, toRemove->relid, toKeep->relid); @@ -2021,7 +2102,7 @@ split_selfjoin_quals(PlannerInfo *root, List *joinquals, List **selfjoinquals, * when we have cast of the same var to different (but compatible) * types. */ - replace_varno(rightexpr, bms_singleton_member(rinfo->right_relids), + replace_varno(root, rightexpr, bms_singleton_member(rinfo->right_relids), bms_singleton_member(rinfo->left_relids)); if (equal(leftexpr, rightexpr)) @@ -2062,7 +2143,7 @@ match_unique_clauses(PlannerInfo *root, RelOptInfo *outer, List *uclauses, bms_is_empty(rinfo->right_relids)); clause = (Expr *) copyObject(rinfo->clause); - replace_varno((Node *) clause, relid, outer->relid); + replace_varno(root, (Node *) clause, relid, outer->relid); iclause = bms_is_empty(rinfo->left_relids) ? get_rightop(clause) : get_leftop(clause); diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c index be4e182869..1436cdd6e3 100644 --- a/src/backend/optimizer/plan/planner.c +++ b/src/backend/optimizer/plan/planner.c @@ -648,6 +648,8 @@ subquery_planner(PlannerGlobal *glob, Query *parse, root->multiexpr_params = NIL; root->join_domains = NIL; root->eq_classes = NIL; + root->eq_sources = NIL; + root->eq_derives = NIL; root->ec_merging_done = false; root->last_rinfo_serial = 0; root->all_result_relids = diff --git a/src/backend/optimizer/prep/prepjointree.c b/src/backend/optimizer/prep/prepjointree.c index aa83dd3636..9fc0b69580 100644 --- a/src/backend/optimizer/prep/prepjointree.c +++ b/src/backend/optimizer/prep/prepjointree.c @@ -993,6 +993,8 @@ pull_up_simple_subquery(PlannerInfo *root, Node *jtnode, RangeTblEntry *rte, subroot->multiexpr_params = NIL; subroot->join_domains = NIL; subroot->eq_classes = NIL; + subroot->eq_sources = NIL; + subroot->eq_derives = NIL; subroot->ec_merging_done = false; subroot->last_rinfo_serial = 0; subroot->all_result_relids = NULL; diff --git a/src/backend/optimizer/util/appendinfo.c b/src/backend/optimizer/util/appendinfo.c index 51fdeace7d..27e4b093d7 100644 --- a/src/backend/optimizer/util/appendinfo.c +++ b/src/backend/optimizer/util/appendinfo.c @@ -491,6 +491,8 @@ adjust_appendrel_attrs_mutator(Node *node, newinfo->right_bucketsize = -1; newinfo->left_mcvfreq = -1; newinfo->right_mcvfreq = -1; + newinfo->eq_sources_index = -1; + newinfo->eq_derives_index = -1; return (Node *) newinfo; } diff --git a/src/backend/optimizer/util/inherit.c b/src/backend/optimizer/util/inherit.c index 3501fbbeed..a65a76ace9 100644 --- a/src/backend/optimizer/util/inherit.c +++ b/src/backend/optimizer/util/inherit.c @@ -483,6 +483,13 @@ expand_single_inheritance_child(PlannerInfo *root, RangeTblEntry *parentrte, */ childrte = makeNode(RangeTblEntry); memcpy(childrte, parentrte, sizeof(RangeTblEntry)); + /* + * We do not want to inherit the EquivalenceMember indexes of the parent + * to its child + */ + childrte->eclass_source_indexes = NULL; + childrte->eclass_derive_indexes = NULL; + Assert(parentrte->rtekind == RTE_RELATION); /* else this is dubious */ childrte->relid = childOID; childrte->relkind = childrel->rd_rel->relkind; diff --git a/src/backend/optimizer/util/restrictinfo.c b/src/backend/optimizer/util/restrictinfo.c index 0b406e9334..59ba503ecb 100644 --- a/src/backend/optimizer/util/restrictinfo.c +++ b/src/backend/optimizer/util/restrictinfo.c @@ -247,6 +247,9 @@ make_restrictinfo_internal(PlannerInfo *root, restrictinfo->left_hasheqoperator = InvalidOid; restrictinfo->right_hasheqoperator = InvalidOid; + restrictinfo->eq_sources_index = -1; + restrictinfo->eq_derives_index = -1; + return restrictinfo; } @@ -403,6 +406,8 @@ commute_restrictinfo(RestrictInfo *rinfo, Oid comm_op) result->right_mcvfreq = rinfo->left_mcvfreq; result->left_hasheqoperator = InvalidOid; result->right_hasheqoperator = InvalidOid; + result->eq_sources_index = -1; + result->eq_derives_index = -1; return result; } diff --git a/src/include/nodes/parsenodes.h b/src/include/nodes/parsenodes.h index baa6a97c7e..f624f41d99 100644 --- a/src/include/nodes/parsenodes.h +++ b/src/include/nodes/parsenodes.h @@ -1194,6 +1194,12 @@ typedef struct RangeTblEntry bool inh; /* inheritance requested? */ bool inFromCl; /* present in FROM clause? */ List *securityQuals; /* security barrier quals to apply, if any */ + Bitmapset *eclass_source_indexes; /* Indexes in PlannerInfo's eq_sources + * list for RestrictInfos that mention + * this relation */ + Bitmapset *eclass_derive_indexes; /* Indexes in PlannerInfo's eq_derives + * list for RestrictInfos that mention + * this relation */ } RangeTblEntry; /* diff --git a/src/include/nodes/pathnodes.h b/src/include/nodes/pathnodes.h index c359505f6a..81253d0351 100644 --- a/src/include/nodes/pathnodes.h +++ b/src/include/nodes/pathnodes.h @@ -318,6 +318,12 @@ struct PlannerInfo /* list of active EquivalenceClasses */ List *eq_classes; + /* list of source RestrictInfos used to build EquivalenceClasses */ + List *eq_sources; + + /* list of RestrictInfos derived from EquivalenceClasses */ + List *eq_derives; + /* set true once ECs are canonical */ bool ec_merging_done; @@ -1371,6 +1377,24 @@ 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. * + * At various locations in the query planner, we must search for source and + * derived RestrictInfos regarding a given EquivalenceClass. For the common + * case, an EquivalenceClass does not have a large number of RestrictInfos, + * however, in cases such as planning queries to partitioned tables, the + * number of members can become large. To maintain planning performance, we + * make use of a bitmap index to allow us to quickly find RestrictInfos in a + * given EquivalenceClass belonging to a given relation or set of relations. + * This is done by storing a list of RestrictInfos belonging to all + * EquivalenceClasses in PlannerInfo and storing a Bitmapset for each + * RelOptInfo which has a bit set for each RestrictInfo in that list which + * relates to the given relation. We also store a Bitmapset to mark all of + * the indexes in the PlannerInfo's list of RestrictInfos in the + * EquivalenceClass. We can quickly find the interesting indexes into the + * PlannerInfo's list by performing a bit-wise AND on the RelOptInfo's + * Bitmapset and the EquivalenceClasses. RestrictInfos must be looked up in + * PlannerInfo by this technique using the ec_source_indexes and + * ec_derive_indexes Bitmapsets. + * * 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. * @@ -1389,8 +1413,10 @@ typedef struct EquivalenceClass List *ec_opfamilies; /* btree operator family OIDs */ Oid ec_collation; /* collation, if datatypes are collatable */ List *ec_members; /* list of EquivalenceMembers */ - List *ec_sources; /* list of generating RestrictInfos */ - List *ec_derives; /* list of derived RestrictInfos */ + Bitmapset *ec_source_indexes; /* indexes into PlannerInfo's eq_sources + * list of generating RestrictInfos */ + Bitmapset *ec_derive_indexes; /* indexes into PlannerInfo's eq_derives + * list of derived RestrictInfos */ Relids ec_relids; /* all relids appearing in ec_members, except * for child members (see below) */ bool ec_has_const; /* any pseudoconstants in ec_members? */ @@ -2620,7 +2646,12 @@ typedef struct RestrictInfo /* number of base rels in clause_relids */ int num_base_rels pg_node_attr(equal_ignore); - /* The relids (varnos+varnullingrels) actually referenced in the clause: */ + /* + * The relids (varnos+varnullingrels) actually referenced in the clause. + * + * NOTE: This field must be updated through update_clause_relids(), not by + * setting the field directly. + */ Relids clause_relids pg_node_attr(equal_ignore); /* The set of relids required to evaluate the clause: */ @@ -2738,6 +2769,10 @@ typedef struct RestrictInfo /* hash equality operators used for memoize nodes, else InvalidOid */ Oid left_hasheqoperator pg_node_attr(equal_ignore); Oid right_hasheqoperator pg_node_attr(equal_ignore); + + /* the index within root->eq_sources and root->eq_derives */ + int eq_sources_index pg_node_attr(equal_ignore); + int eq_derives_index pg_node_attr(equal_ignore); } RestrictInfo; /* diff --git a/src/include/optimizer/paths.h b/src/include/optimizer/paths.h index 5f6a08821e..6bb740266b 100644 --- a/src/include/optimizer/paths.h +++ b/src/include/optimizer/paths.h @@ -129,6 +129,8 @@ extern bool process_equivalence(PlannerInfo *root, extern Expr *canonicalize_ec_expression(Expr *expr, Oid req_type, Oid req_collation); extern void reconsider_outer_join_clauses(PlannerInfo *root); +extern void update_clause_relids(PlannerInfo *root, RestrictInfo *rinfo, + Relids new_clause_relids); extern EquivalenceClass *get_eclass_for_sort_expr(PlannerInfo *root, Expr *expr, List *opfamilies, @@ -164,7 +166,8 @@ extern bool exprs_known_equal(PlannerInfo *root, Node *item1, Node *item2); extern EquivalenceClass *match_eclasses_to_foreign_key_col(PlannerInfo *root, ForeignKeyOptInfo *fkinfo, int colno); -extern RestrictInfo *find_derived_clause_for_ec_member(EquivalenceClass *ec, +extern RestrictInfo *find_derived_clause_for_ec_member(PlannerInfo *root, + EquivalenceClass *ec, EquivalenceMember *em); extern void add_child_rel_equivalences(PlannerInfo *root, AppendRelInfo *appinfo, @@ -199,6 +202,22 @@ extern bool eclass_useful_for_merging(PlannerInfo *root, extern bool is_redundant_derived_clause(RestrictInfo *rinfo, List *clauselist); extern bool is_redundant_with_indexclauses(RestrictInfo *rinfo, List *indexclauses); +extern Bitmapset *get_ec_source_indexes(PlannerInfo *root, + EquivalenceClass *ec, + Relids relids); +extern Bitmapset *get_ec_source_indexes_strict(PlannerInfo *root, + EquivalenceClass *ec, + Relids relids); +extern Bitmapset *get_ec_derive_indexes(PlannerInfo *root, + EquivalenceClass *ec, + Relids relids); +extern Bitmapset *get_ec_derive_indexes_strict(PlannerInfo *root, + EquivalenceClass *ec, + Relids relids); +#ifdef USE_ASSERT_CHECKING +extern void verify_eclass_indexes(PlannerInfo *root, + EquivalenceClass *ec); +#endif /* * pathkeys.c -- 2.42.0.windows.2