On 2016/09/27 18:09, Ashutosh Bapat wrote: >>> I tried to debug the problem somewhat. In set_append_rel_pathlist(), >>> it finds that at least one child has a parameterized path as the >>> cheapest path, so it doesn't create an unparameterized path for append >>> rel. At the same time there is a parameterization common to all the >>> children, so it doesn't create any path. There seem to be two problems >>> here >>> 1. The children from second level onwards may not be getting >>> parameterized for lateral references. That seems unlikely but >>> possible. > > Did you check this? We may be missing on creating index scan paths > with parameterization. If we fix this, we don't need to > re-parameterize Append.
You're right. How about the attached patch that fixes the problem along these lines? The problem seems to be that multi-level inheritance sets (partitioned tables) are not handled in create_lateral_join_info(), which means that lateral_relids and lateral_referencers of the root relation are not being propagated to the partitions below level 1. I'm getting concerned about one thing though - for a given *regular* inheritance set, the root->append_rel_list would be scanned only once; But for a *partitioned table* inheritance set, it would be scanned for every partitioned table in the set (ie, the root table and internal partitions). Thanks, Amit
commit d69aeabcfcb58f349602a1b7392c611045c37465 Author: amit <amitlangot...@gmail.com> Date: Tue Sep 27 20:01:44 2016 +0900 Consider multi-level partitioned tables in create_lateral_join_info(). diff --git a/src/backend/optimizer/plan/initsplan.c b/src/backend/optimizer/plan/initsplan.c index 84ce6b3..74734f2 100644 --- a/src/backend/optimizer/plan/initsplan.c +++ b/src/backend/optimizer/plan/initsplan.c @@ -14,6 +14,7 @@ */ #include "postgres.h" +#include "catalog/pg_class.h" #include "catalog/pg_type.h" #include "nodes/nodeFuncs.h" #include "optimizer/clauses.h" @@ -623,8 +624,19 @@ create_lateral_join_info(PlannerInfo *root) for (rti = 1; rti < root->simple_rel_array_size; rti++) { RelOptInfo *brel = root->simple_rel_array[rti]; + RangeTblEntry *rte = root->simple_rte_array[rti]; - if (brel == NULL || brel->reloptkind != RELOPT_BASEREL) + if (brel == NULL) + continue; + + /* + * If an "other rel" RTE is a "partitioned table", we must propagate + * the lateral info inherited from the parent to its children. That's + * because they are not linked directly with the parent via + * AppendRelInfo's (see expand_inherited_rte_internal()). + */ + if (brel->reloptkind != RELOPT_BASEREL && + rte->relkind != RELKIND_PARTITIONED_TABLE) continue; if (root->simple_rte_array[rti]->inh)
-- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers