diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/backend/optimizer/plan/setrefs.c | 71 | ||||
| -rw-r--r-- | src/include/nodes/pathnodes.h | 4 | ||||
| -rw-r--r-- | src/test/regress/expected/subselect.out | 40 | ||||
| -rw-r--r-- | src/test/regress/sql/subselect.sql | 12 |
4 files changed, 103 insertions, 24 deletions
diff --git a/src/backend/optimizer/plan/setrefs.c b/src/backend/optimizer/plan/setrefs.c index e50624c465e..3d23a2e5ac7 100644 --- a/src/backend/optimizer/plan/setrefs.c +++ b/src/backend/optimizer/plan/setrefs.c @@ -249,6 +249,7 @@ static List *set_returning_clause_references(PlannerInfo *root, Plan * set_plan_references(PlannerInfo *root, Plan *plan) { + Plan *result; PlannerGlobal *glob = root->glob; int rtoffset = list_length(glob->finalrtable); ListCell *lc; @@ -301,8 +302,44 @@ set_plan_references(PlannerInfo *root, Plan *plan) glob->appendRelations = lappend(glob->appendRelations, appinfo); } + /* If needed, create workspace for processing AlternativeSubPlans */ + if (root->hasAlternativeSubPlans) + { + root->isAltSubplan = (bool *) + palloc0(list_length(glob->subplans) * sizeof(bool)); + root->isUsedSubplan = (bool *) + palloc0(list_length(glob->subplans) * sizeof(bool)); + } + /* Now fix the Plan tree */ - return set_plan_refs(root, plan, rtoffset); + result = set_plan_refs(root, plan, rtoffset); + + /* + * If we have AlternativeSubPlans, it is likely that we now have some + * unreferenced subplans in glob->subplans. To avoid expending cycles on + * those subplans later, get rid of them by setting those list entries to + * NULL. (Note: we can't do this immediately upon processing an + * AlternativeSubPlan, because there may be multiple copies of the + * AlternativeSubPlan, and they can get resolved differently.) + */ + if (root->hasAlternativeSubPlans) + { + foreach(lc, glob->subplans) + { + int ndx = foreach_current_index(lc); + + /* + * If it was used by some AlternativeSubPlan in this query level, + * but wasn't selected as best by any AlternativeSubPlan, then we + * don't need it. Do not touch subplans that aren't parts of + * AlternativeSubPlans. + */ + if (root->isAltSubplan[ndx] && !root->isUsedSubplan[ndx]) + lfirst(lc) = NULL; + } + } + + return result; } /* @@ -1765,8 +1802,7 @@ fix_param_node(PlannerInfo *root, Param *p) * Note: caller must still recurse into the result! * * We don't make any attempt to fix up cost estimates in the parent plan - * node or higher-level nodes. However, we do remove the rejected subplan(s) - * from root->glob->subplans, to minimize cycles expended on them later. + * node or higher-level nodes. */ static Node * fix_alternative_subplan(PlannerInfo *root, AlternativeSubPlan *asplan, @@ -1778,9 +1814,8 @@ fix_alternative_subplan(PlannerInfo *root, AlternativeSubPlan *asplan, /* * Compute the estimated cost of each subplan assuming num_exec - * executions, and keep the cheapest one. Replace discarded subplans with - * NULL pointers in the global subplans list. In event of exact equality - * of estimates, we prefer the later plan; this is a bit arbitrary, but in + * executions, and keep the cheapest one. In event of exact equality of + * estimates, we prefer the later plan; this is a bit arbitrary, but in * current usage it biases us to break ties against fast-start subplans. */ Assert(asplan->subplans != NIL); @@ -1791,31 +1826,19 @@ fix_alternative_subplan(PlannerInfo *root, AlternativeSubPlan *asplan, Cost curcost; curcost = curplan->startup_cost + num_exec * curplan->per_call_cost; - if (bestplan == NULL) - { - bestplan = curplan; - bestcost = curcost; - } - else if (curcost <= bestcost) + if (bestplan == NULL || curcost <= bestcost) { - /* drop old bestplan */ - ListCell *lc2 = list_nth_cell(root->glob->subplans, - bestplan->plan_id - 1); - - lfirst(lc2) = NULL; bestplan = curplan; bestcost = curcost; } - else - { - /* drop curplan */ - ListCell *lc2 = list_nth_cell(root->glob->subplans, - curplan->plan_id - 1); - lfirst(lc2) = NULL; - } + /* Also mark all subplans that are in AlternativeSubPlans */ + root->isAltSubplan[curplan->plan_id - 1] = true; } + /* Mark the subplan we selected */ + root->isUsedSubplan[bestplan->plan_id - 1] = true; + return (Node *) bestplan; } diff --git a/src/include/nodes/pathnodes.h b/src/include/nodes/pathnodes.h index 1abe233db2b..c8a8eec6e2f 100644 --- a/src/include/nodes/pathnodes.h +++ b/src/include/nodes/pathnodes.h @@ -368,6 +368,10 @@ struct PlannerInfo Relids curOuterRels; /* outer rels above current node */ List *curOuterParams; /* not-yet-assigned NestLoopParams */ + /* These fields are workspace for setrefs.c */ + bool *isAltSubplan; /* array corresponding to glob->subplans */ + bool *isUsedSubplan; /* array corresponding to glob->subplans */ + /* optional private data for join_search_hook, e.g., GEQO */ void *join_search_private; diff --git a/src/test/regress/expected/subselect.out b/src/test/regress/expected/subselect.out index 30615dd6bc7..07426260330 100644 --- a/src/test/regress/expected/subselect.out +++ b/src/test/regress/expected/subselect.out @@ -921,6 +921,46 @@ where (exists(select 1 from tenk1 k where k.unique1 = t.unique2) or ten < 0) 10 (1 row) +-- It's possible for the same EXISTS to get resolved both ways +create temp table exists_tbl (c1 int, c2 int, c3 int) partition by list (c1); +create temp table exists_tbl_null partition of exists_tbl for values in (null); +create temp table exists_tbl_def partition of exists_tbl default; +insert into exists_tbl select x, x/2, x+1 from generate_series(0,10) x; +analyze exists_tbl; +explain (costs off) +select * from exists_tbl t1 + where (exists(select 1 from exists_tbl t2 where t1.c1 = t2.c2) or c3 < 0); + QUERY PLAN +------------------------------------------------------ + Append + -> Seq Scan on exists_tbl_null t1_1 + Filter: ((SubPlan 1) OR (c3 < 0)) + SubPlan 1 + -> Append + -> Seq Scan on exists_tbl_null t2_1 + Filter: (t1_1.c1 = c2) + -> Seq Scan on exists_tbl_def t2_2 + Filter: (t1_1.c1 = c2) + -> Seq Scan on exists_tbl_def t1_2 + Filter: ((hashed SubPlan 2) OR (c3 < 0)) + SubPlan 2 + -> Append + -> Seq Scan on exists_tbl_null t2_4 + -> Seq Scan on exists_tbl_def t2_5 +(15 rows) + +select * from exists_tbl t1 + where (exists(select 1 from exists_tbl t2 where t1.c1 = t2.c2) or c3 < 0); + c1 | c2 | c3 +----+----+---- + 0 | 0 | 1 + 1 | 0 | 2 + 2 | 1 | 3 + 3 | 1 | 4 + 4 | 2 | 5 + 5 | 2 | 6 +(6 rows) + -- -- Test case for planner bug with nested EXISTS handling -- diff --git a/src/test/regress/sql/subselect.sql b/src/test/regress/sql/subselect.sql index bd17f5d264d..e879999708b 100644 --- a/src/test/regress/sql/subselect.sql +++ b/src/test/regress/sql/subselect.sql @@ -526,6 +526,18 @@ select count(*) from tenk1 t where (exists(select 1 from tenk1 k where k.unique1 = t.unique2) or ten < 0) and thousand = 1; +-- It's possible for the same EXISTS to get resolved both ways +create temp table exists_tbl (c1 int, c2 int, c3 int) partition by list (c1); +create temp table exists_tbl_null partition of exists_tbl for values in (null); +create temp table exists_tbl_def partition of exists_tbl default; +insert into exists_tbl select x, x/2, x+1 from generate_series(0,10) x; +analyze exists_tbl; +explain (costs off) +select * from exists_tbl t1 + where (exists(select 1 from exists_tbl t2 where t1.c1 = t2.c2) or c3 < 0); +select * from exists_tbl t1 + where (exists(select 1 from exists_tbl t2 where t1.c1 = t2.c2) or c3 < 0); + -- -- Test case for planner bug with nested EXISTS handling -- |