v37-0001-Add-field-to-store-partitioned-relids-to-Append-.patch
application/octet-stream
Filename: v37-0001-Add-field-to-store-partitioned-relids-to-Append-.patch
Type: application/octet-stream
Part: 0
Patch
Format: format-patch
Series: patch v37-0001
Subject: Add field to store partitioned relids to Append/MergeAppend
| File | + | − |
|---|---|---|
| src/backend/optimizer/plan/createplan.c | 26 | 10 |
| src/backend/optimizer/plan/planner.c | 3 | 0 |
| src/backend/optimizer/util/appendinfo.c | 134 | 0 |
| src/backend/partitioning/partprune.c | 12 | 112 |
| src/include/nodes/plannodes.h | 14 | 0 |
| src/include/optimizer/appendinfo.h | 3 | 0 |
| src/include/partitioning/partprune.h | 2 | 1 |
From dfc41510ef3ebec38e7a56b639ffa41193109b43 Mon Sep 17 00:00:00 2001
From: amitlan <amitlangote09@gmail.com>
Date: Thu, 9 Mar 2023 11:26:06 +0900
Subject: [PATCH v37 1/3] Add field to store partitioned relids to
Append/MergeAppend
A future commit would like to move the timing of locking relations
referenced in a cached plan to ExecInitNode() traversal of the plan
tree from the current loop-over-rangetable in AcquireExecutorLocks().
Given that partitioned tables (their RT indexes) would not be
accessible via the new way of finding the relations to lock, add a
field to Append/MergeAppend to track them separately.
This refactors the code to look up partitioned parent relids from a
given list of leaf partition subpaths of an Append/MergeAppend out
of make_partition_pruneinfo() into its own function called
add_append_subpath_partrelids(). Though, the code needs to be
generalized to the cases where child rels can be joinrels or
upper (grouping) rels. Also, to make it easier to traverse the parent
chain of a child grouping rel, this makes its RelOptInfo.parent to be
set, which is already done for baserels and joinrels.
---
src/backend/optimizer/plan/createplan.c | 36 +++++--
src/backend/optimizer/plan/planner.c | 3 +
src/backend/optimizer/util/appendinfo.c | 134 ++++++++++++++++++++++++
src/backend/partitioning/partprune.c | 124 +++-------------------
src/include/nodes/plannodes.h | 14 +++
src/include/optimizer/appendinfo.h | 3 +
src/include/partitioning/partprune.h | 3 +-
7 files changed, 194 insertions(+), 123 deletions(-)
diff --git a/src/backend/optimizer/plan/createplan.c b/src/backend/optimizer/plan/createplan.c
index 910ffbf1e1..794cdb5e3b 100644
--- a/src/backend/optimizer/plan/createplan.c
+++ b/src/backend/optimizer/plan/createplan.c
@@ -25,6 +25,7 @@
#include "nodes/extensible.h"
#include "nodes/makefuncs.h"
#include "nodes/nodeFuncs.h"
+#include "optimizer/appendinfo.h"
#include "optimizer/clauses.h"
#include "optimizer/cost.h"
#include "optimizer/optimizer.h"
@@ -1209,6 +1210,7 @@ create_append_plan(PlannerInfo *root, AppendPath *best_path, int flags)
Oid *nodeCollations = NULL;
bool *nodeNullsFirst = NULL;
bool consider_async = false;
+ List *allpartrelids = NIL;
/*
* The subpaths list could be empty, if every child was proven empty by
@@ -1350,18 +1352,24 @@ create_append_plan(PlannerInfo *root, AppendPath *best_path, int flags)
++nasyncplans;
}
+ /* Populate partitioned parent relids. */
+ allpartrelids = add_append_subpath_partrelids(root, subpath, rel,
+ allpartrelids);
+
subplans = lappend(subplans, subplan);
}
+ plan->allpartrelids = allpartrelids;
+
/* Set below if we find quals that we can use to run-time prune */
plan->part_prune_index = -1;
/*
- * If any quals exist, they may be useful to perform further partition
- * pruning during execution. Gather information needed by the executor to
- * do partition pruning.
+ * If scanning partitions, check if there are quals that may be useful to
+ * perform further partition pruning during execution. Gather information
+ * needed by the executor to do partition pruning.
*/
- if (enable_partition_pruning)
+ if (enable_partition_pruning && allpartrelids != NIL)
{
List *prunequal;
@@ -1381,7 +1389,8 @@ create_append_plan(PlannerInfo *root, AppendPath *best_path, int flags)
if (prunequal != NIL)
plan->part_prune_index = make_partition_pruneinfo(root, rel,
best_path->subpaths,
- prunequal);
+ prunequal,
+ allpartrelids);
}
plan->appendplans = subplans;
@@ -1425,6 +1434,7 @@ create_merge_append_plan(PlannerInfo *root, MergeAppendPath *best_path,
List *subplans = NIL;
ListCell *subpaths;
RelOptInfo *rel = best_path->path.parent;
+ List *allpartrelids = NIL;
/*
* We don't have the actual creation of the MergeAppend node split out
@@ -1514,18 +1524,23 @@ create_merge_append_plan(PlannerInfo *root, MergeAppendPath *best_path,
subplan = (Plan *) sort;
}
+ allpartrelids = add_append_subpath_partrelids(root, subpath, rel,
+ allpartrelids);
+
subplans = lappend(subplans, subplan);
}
+ node->allpartrelids = allpartrelids;
+
/* Set below if we find quals that we can use to run-time prune */
node->part_prune_index = -1;
/*
- * If any quals exist, they may be useful to perform further partition
- * pruning during execution. Gather information needed by the executor to
- * do partition pruning.
+ * If scanning partitions, check if there are quals that may be useful to
+ * perform further partition pruning during execution. Gather information
+ * needed by the executor to do partition pruning.
*/
- if (enable_partition_pruning)
+ if (enable_partition_pruning && allpartrelids != NIL)
{
List *prunequal;
@@ -1537,7 +1552,8 @@ create_merge_append_plan(PlannerInfo *root, MergeAppendPath *best_path,
if (prunequal != NIL)
node->part_prune_index = make_partition_pruneinfo(root, rel,
best_path->subpaths,
- prunequal);
+ prunequal,
+ allpartrelids);
}
node->mergeplans = subplans;
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index a1873ce26d..62b3ec96cc 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -7801,8 +7801,11 @@ create_partitionwise_grouping_paths(PlannerInfo *root,
agg_costs, gd, &child_extra,
&child_partially_grouped_rel);
+ /* Mark as child of grouped_rel. */
+ child_grouped_rel->parent = grouped_rel;
if (child_partially_grouped_rel)
{
+ child_partially_grouped_rel->parent = grouped_rel;
partially_grouped_live_children =
lappend(partially_grouped_live_children,
child_partially_grouped_rel);
diff --git a/src/backend/optimizer/util/appendinfo.c b/src/backend/optimizer/util/appendinfo.c
index 9d377385f1..4876742ab2 100644
--- a/src/backend/optimizer/util/appendinfo.c
+++ b/src/backend/optimizer/util/appendinfo.c
@@ -40,6 +40,7 @@ static void make_inh_translation_list(Relation oldrelation,
AppendRelInfo *appinfo);
static Node *adjust_appendrel_attrs_mutator(Node *node,
adjust_appendrel_attrs_context *context);
+static List *add_part_relids(List *allpartrelids, Bitmapset *partrelids);
/*
@@ -1031,3 +1032,136 @@ distribute_row_identity_vars(PlannerInfo *root)
}
}
}
+
+/*
+ * add_append_subpath_partrelids
+ * Look up a child subpath's rel's partitioned parent relids up to
+ * parentrel and add the bitmapset containing those into
+ * 'allpartrelids'
+ */
+List *
+add_append_subpath_partrelids(PlannerInfo *root, Path *subpath,
+ RelOptInfo *parentrel,
+ List *allpartrelids)
+{
+ RelOptInfo *prel = subpath->parent;
+ Relids partrelids = NULL;
+
+ /* Nothing to do if there's no parent to begin with. */
+ if (!IS_OTHER_REL(prel))
+ return allpartrelids;
+
+ /*
+ * Traverse up to the pathrel's topmost partitioned parent, collecting
+ * parent relids as we go; but stop if we reach parentrel. (Normally, a
+ * pathrel's topmost partitioned parent is either parentrel or a UNION ALL
+ * appendrel child of parentrel. But when handling partitionwise joins of
+ * multi-level partitioning trees, we can see an append path whose
+ * parentrel is an intermediate partitioned table.)
+ */
+ do
+ {
+ Relids parent_relids = NULL;
+
+ /*
+ * For simple child rels, we can simply get the parent relid from
+ * prel->parent. But for partitionwise join and aggregate child rels,
+ * while we can use prel->parent to move up the tree, parent relids to
+ * add into 'partrelids' must be found the hard way through the
+ * AppendInfoInfos, because 1) a joinrel's relids may point to RTE_JOIN
+ * entries, 2) topmost parent grouping rel's relids field is left NULL.
+ */
+ if (IS_SIMPLE_REL(prel))
+ {
+ prel = prel->parent;
+ /* Stop once we reach the root partitioned rel. */
+ if (!IS_PARTITIONED_REL(prel))
+ break;
+ parent_relids = bms_add_members(parent_relids, prel->relids);
+ }
+ else
+ {
+ AppendRelInfo **appinfos;
+ int nappinfos,
+ i;
+
+ appinfos = find_appinfos_by_relids(root, prel->relids,
+ &nappinfos);
+ for (i = 0; i < nappinfos; i++)
+ {
+ AppendRelInfo *appinfo = appinfos[i];
+
+ parent_relids = bms_add_member(parent_relids,
+ appinfo->parent_relid);
+ }
+ pfree(appinfos);
+ prel = prel->parent;
+ }
+ /* accept this level as an interesting parent */
+ partrelids = bms_add_members(partrelids, parent_relids);
+ if (prel == parentrel)
+ break; /* don't traverse above parentrel */
+ } while (IS_OTHER_REL(prel));
+
+ if (partrelids == NULL)
+ return allpartrelids;
+
+ return add_part_relids(allpartrelids, partrelids);
+}
+
+/*
+ * add_part_relids
+ * Add new info to a list of Bitmapsets of partitioned relids.
+ *
+ * Within 'allpartrelids', there is one Bitmapset for each topmost parent
+ * partitioned rel. Each Bitmapset contains the RT indexes of the topmost
+ * parent as well as its relevant non-leaf child partitions. Since (by
+ * construction of the rangetable list) parent partitions must have lower
+ * RT indexes than their children, we can distinguish the topmost parent
+ * as being the lowest set bit in the Bitmapset.
+ *
+ * 'partrelids' contains the RT indexes of a parent partitioned rel, and
+ * possibly some non-leaf children, that are newly identified as parents of
+ * some subpath rel passed to make_partition_pruneinfo(). These are added
+ * to an appropriate member of 'allpartrelids'.
+ *
+ * Note that the list contains only RT indexes of partitioned tables that
+ * are parents of some scan-level relation appearing in the 'subpaths' that
+ * make_partition_pruneinfo() is dealing with. Also, "topmost" parents are
+ * not allowed to be higher than the 'parentrel' associated with the append
+ * path. In this way, we avoid expending cycles on partitioned rels that
+ * can't contribute useful pruning information for the problem at hand.
+ * (It is possible for 'parentrel' to be a child partitioned table, and it
+ * is also possible for scan-level relations to be child partitioned tables
+ * rather than leaf partitions. Hence we must construct this relation set
+ * with reference to the particular append path we're dealing with, rather
+ * than looking at the full partitioning structure represented in the
+ * RelOptInfos.)
+ */
+static List *
+add_part_relids(List *allpartrelids, Bitmapset *partrelids)
+{
+ Index targetpart;
+ ListCell *lc;
+
+ /* We can easily get the lowest set bit this way: */
+ targetpart = bms_next_member(partrelids, -1);
+ Assert(targetpart > 0);
+
+ /* Look for a matching topmost parent */
+ foreach(lc, allpartrelids)
+ {
+ Bitmapset *currpartrelids = (Bitmapset *) lfirst(lc);
+ Index currtarget = bms_next_member(currpartrelids, -1);
+
+ if (targetpart == currtarget)
+ {
+ /* Found a match, so add any new RT indexes to this hierarchy */
+ currpartrelids = bms_add_members(currpartrelids, partrelids);
+ lfirst(lc) = currpartrelids;
+ return allpartrelids;
+ }
+ }
+ /* No match, so add the new partition hierarchy to the list */
+ return lappend(allpartrelids, partrelids);
+}
diff --git a/src/backend/partitioning/partprune.c b/src/backend/partitioning/partprune.c
index 510145e3c0..3557e07082 100644
--- a/src/backend/partitioning/partprune.c
+++ b/src/backend/partitioning/partprune.c
@@ -138,7 +138,6 @@ typedef struct PruneStepResult
} PruneStepResult;
-static List *add_part_relids(List *allpartrelids, Bitmapset *partrelids);
static List *make_partitionedrel_pruneinfo(PlannerInfo *root,
RelOptInfo *parentrel,
List *prunequal,
@@ -221,33 +220,32 @@ static void partkey_datum_from_expr(PartitionPruneContext *context,
* of scan paths for its child rels.
* 'prunequal' is a list of potential pruning quals (i.e., restriction
* clauses that are applicable to the appendrel).
+ * 'allpartrelids' contains Bitmapsets of RT indexes of partitioned parents
+ * whose partitions' Paths are in 'subpaths'; there's one Bitmapset for every
+ * partition tree involved.
*/
int
make_partition_pruneinfo(PlannerInfo *root, RelOptInfo *parentrel,
List *subpaths,
- List *prunequal)
+ List *prunequal,
+ List *allpartrelids)
{
PartitionPruneInfo *pruneinfo;
Bitmapset *allmatchedsubplans = NULL;
- List *allpartrelids;
List *prunerelinfos;
int *relid_subplan_map;
ListCell *lc;
int i;
+ Assert(list_length(allpartrelids) > 0);
+
/*
- * Scan the subpaths to see which ones are scans of partition child
- * relations, and identify their parent partitioned rels. (Note: we must
- * restrict the parent partitioned rels to be parentrel or children of
- * parentrel, otherwise we couldn't translate prunequal to match.)
- *
- * Also construct a temporary array to map from partition-child-relation
- * relid to the index in 'subpaths' of the scan plan for that partition.
+ * Construct a temporary array to map from partition-child-relation relid
+ * to the index in 'subpaths' of the scan plan for that partition.
* (Use of "subplan" rather than "subpath" is a bit of a misnomer, but
* we'll let it stand.) For convenience, we use 1-based indexes here, so
* that zero can represent an un-filled array entry.
*/
- allpartrelids = NIL;
relid_subplan_map = palloc0(sizeof(int) * root->simple_rel_array_size);
i = 1;
@@ -256,50 +254,9 @@ make_partition_pruneinfo(PlannerInfo *root, RelOptInfo *parentrel,
Path *path = (Path *) lfirst(lc);
RelOptInfo *pathrel = path->parent;
- /* We don't consider partitioned joins here */
- if (pathrel->reloptkind == RELOPT_OTHER_MEMBER_REL)
- {
- RelOptInfo *prel = pathrel;
- Bitmapset *partrelids = NULL;
-
- /*
- * Traverse up to the pathrel's topmost partitioned parent,
- * collecting parent relids as we go; but stop if we reach
- * parentrel. (Normally, a pathrel's topmost partitioned parent
- * is either parentrel or a UNION ALL appendrel child of
- * parentrel. But when handling partitionwise joins of
- * multi-level partitioning trees, we can see an append path whose
- * parentrel is an intermediate partitioned table.)
- */
- do
- {
- AppendRelInfo *appinfo;
-
- Assert(prel->relid < root->simple_rel_array_size);
- appinfo = root->append_rel_array[prel->relid];
- prel = find_base_rel(root, appinfo->parent_relid);
- if (!IS_PARTITIONED_REL(prel))
- break; /* reached a non-partitioned parent */
- /* accept this level as an interesting parent */
- partrelids = bms_add_member(partrelids, prel->relid);
- if (prel == parentrel)
- break; /* don't traverse above parentrel */
- } while (prel->reloptkind == RELOPT_OTHER_MEMBER_REL);
-
- if (partrelids)
- {
- /*
- * Found some relevant parent partitions, which may or may not
- * overlap with partition trees we already found. Add new
- * information to the allpartrelids list.
- */
- allpartrelids = add_part_relids(allpartrelids, partrelids);
- /* Also record the subplan in relid_subplan_map[] */
- /* No duplicates please */
- Assert(relid_subplan_map[pathrel->relid] == 0);
- relid_subplan_map[pathrel->relid] = i;
- }
- }
+ /* No duplicates please */
+ Assert(relid_subplan_map[pathrel->relid] == 0);
+ relid_subplan_map[pathrel->relid] = i;
i++;
}
@@ -368,63 +325,6 @@ make_partition_pruneinfo(PlannerInfo *root, RelOptInfo *parentrel,
return list_length(root->partPruneInfos) - 1;
}
-/*
- * add_part_relids
- * Add new info to a list of Bitmapsets of partitioned relids.
- *
- * Within 'allpartrelids', there is one Bitmapset for each topmost parent
- * partitioned rel. Each Bitmapset contains the RT indexes of the topmost
- * parent as well as its relevant non-leaf child partitions. Since (by
- * construction of the rangetable list) parent partitions must have lower
- * RT indexes than their children, we can distinguish the topmost parent
- * as being the lowest set bit in the Bitmapset.
- *
- * 'partrelids' contains the RT indexes of a parent partitioned rel, and
- * possibly some non-leaf children, that are newly identified as parents of
- * some subpath rel passed to make_partition_pruneinfo(). These are added
- * to an appropriate member of 'allpartrelids'.
- *
- * Note that the list contains only RT indexes of partitioned tables that
- * are parents of some scan-level relation appearing in the 'subpaths' that
- * make_partition_pruneinfo() is dealing with. Also, "topmost" parents are
- * not allowed to be higher than the 'parentrel' associated with the append
- * path. In this way, we avoid expending cycles on partitioned rels that
- * can't contribute useful pruning information for the problem at hand.
- * (It is possible for 'parentrel' to be a child partitioned table, and it
- * is also possible for scan-level relations to be child partitioned tables
- * rather than leaf partitions. Hence we must construct this relation set
- * with reference to the particular append path we're dealing with, rather
- * than looking at the full partitioning structure represented in the
- * RelOptInfos.)
- */
-static List *
-add_part_relids(List *allpartrelids, Bitmapset *partrelids)
-{
- Index targetpart;
- ListCell *lc;
-
- /* We can easily get the lowest set bit this way: */
- targetpart = bms_next_member(partrelids, -1);
- Assert(targetpart > 0);
-
- /* Look for a matching topmost parent */
- foreach(lc, allpartrelids)
- {
- Bitmapset *currpartrelids = (Bitmapset *) lfirst(lc);
- Index currtarget = bms_next_member(currpartrelids, -1);
-
- if (targetpart == currtarget)
- {
- /* Found a match, so add any new RT indexes to this hierarchy */
- currpartrelids = bms_add_members(currpartrelids, partrelids);
- lfirst(lc) = currpartrelids;
- return allpartrelids;
- }
- }
- /* No match, so add the new partition hierarchy to the list */
- return lappend(allpartrelids, partrelids);
-}
-
/*
* make_partitionedrel_pruneinfo
* Build a List of PartitionedRelPruneInfos, one for each interesting
diff --git a/src/include/nodes/plannodes.h b/src/include/nodes/plannodes.h
index 659bd05c0c..a0bb16cff4 100644
--- a/src/include/nodes/plannodes.h
+++ b/src/include/nodes/plannodes.h
@@ -270,6 +270,13 @@ typedef struct Append
List *appendplans;
int nasyncplans; /* # of asynchronous plans */
+ /*
+ * RTIs of all partitioned tables whose children are scanned by
+ * appendplans. The list contains a bitmapset for every partition tree
+ * covered by this Append.
+ */
+ List *allpartrelids;
+
/*
* All 'appendplans' preceding this index are non-partial plans. All
* 'appendplans' from this index onwards are partial plans.
@@ -294,6 +301,13 @@ typedef struct MergeAppend
List *mergeplans;
+ /*
+ * RTIs of all partitioned tables whose children are scanned by
+ * mergeplans. The list contains a bitmapset for every partition tree
+ * covered by this MergeAppend.
+ */
+ List *allpartrelids;
+
/* these fields are just like the sort-key info in struct Sort: */
/* number of sort-key columns */
diff --git a/src/include/optimizer/appendinfo.h b/src/include/optimizer/appendinfo.h
index a05f91f77d..1621a7319a 100644
--- a/src/include/optimizer/appendinfo.h
+++ b/src/include/optimizer/appendinfo.h
@@ -46,5 +46,8 @@ extern void add_row_identity_columns(PlannerInfo *root, Index rtindex,
RangeTblEntry *target_rte,
Relation target_relation);
extern void distribute_row_identity_vars(PlannerInfo *root);
+extern List *add_append_subpath_partrelids(PlannerInfo *root, Path *subpath,
+ RelOptInfo *parentrel,
+ List *allpartrelids);
#endif /* APPENDINFO_H */
diff --git a/src/include/partitioning/partprune.h b/src/include/partitioning/partprune.h
index c0d6889d47..2d907d31d4 100644
--- a/src/include/partitioning/partprune.h
+++ b/src/include/partitioning/partprune.h
@@ -73,7 +73,8 @@ typedef struct PartitionPruneContext
extern int make_partition_pruneinfo(struct PlannerInfo *root,
struct RelOptInfo *parentrel,
List *subpaths,
- List *prunequal);
+ List *prunequal,
+ List *allpartrelids);
extern Bitmapset *prune_append_rel_partitions(struct RelOptInfo *rel);
extern Bitmapset *get_matching_partitions(PartitionPruneContext *context,
List *pruning_steps);
--
2.35.3