setrefs.c contains logic to discard no-op SubqueryScan nodes, that is,
ones that have no qual to check and copy the input targetlist unchanged.
(Formally it's not very nice to be applying such optimizations so late
in the planner, but there are practical reasons for it; mostly that we
can't unify relids between the subquery and the parent query until we
flatten the rangetable during setrefs.c.) This behavior falsifies our
previous cost estimates, since we would've charged cpu_tuple_cost per
row just to pass data through the node. Most of the time that's little
enough to not matter, but there are cases where this effect visibly
changes the plan compared to what you would've gotten with no
sub-select.
To improve the situation, make the callers of cost_subqueryscan tell
it whether they think the targetlist is trivial. cost_subqueryscan
already has the qual list, so it can check the other half of the
condition easily. It could make its own determination of tlist
triviality too, but doing so would be repetitive (for callers that
may call it several times) or unnecessarily expensive (for callers
that can determine this more cheaply than a general test would do).
This isn't a 100% solution, because createplan.c also does things
that can falsify any earlier estimate of whether the tlist is
trivial. However, it fixes nearly all cases in practice, if results
for the regression tests are anything to go by.
setrefs.c also contains logic to discard no-op Append and MergeAppend
nodes. We did have knowledge of that behavior at costing time, but
somebody failed to update it when a check on parallel-awareness was
added to the setrefs.c logic. Fix that while we're here.
These changes result in two minor changes in query plans shown in
our regression tests. Neither is relevant to the purposes of its
test case AFAICT.
Patch by me; thanks to Richard Guo for review.
Discussion: https://postgr.es/m/
2581077.
1651703520@sss.pgh.pa.us
{
Query *parse = root->parse;
Query *subquery = rte->subquery;
+ bool trivial_pathtarget;
Relids required_outer;
pushdown_safety_info safetyInfo;
double tuple_fraction;
*/
set_subquery_size_estimates(root, rel);
+ /*
+ * Also detect whether the reltarget is trivial, so that we can pass that
+ * info to cost_subqueryscan (rather than re-deriving it multiple times).
+ * It's trivial if it fetches all the subplan output columns in order.
+ */
+ if (list_length(rel->reltarget->exprs) != list_length(subquery->targetList))
+ trivial_pathtarget = false;
+ else
+ {
+ trivial_pathtarget = true;
+ foreach(lc, rel->reltarget->exprs)
+ {
+ Node *node = (Node *) lfirst(lc);
+ Var *var;
+
+ if (!IsA(node, Var))
+ {
+ trivial_pathtarget = false;
+ break;
+ }
+ var = (Var *) node;
+ if (var->varno != rti ||
+ var->varattno != foreach_current_index(lc) + 1)
+ {
+ trivial_pathtarget = false;
+ break;
+ }
+ }
+ }
+
/*
* For each Path that subquery_planner produced, make a SubqueryScanPath
* in the outer query.
/* Generate outer path using this subpath */
add_path(rel, (Path *)
create_subqueryscan_path(root, rel, subpath,
+ trivial_pathtarget,
pathkeys, required_outer));
}
/* Generate outer path using this subpath */
add_partial_path(rel, (Path *)
create_subqueryscan_path(root, rel, subpath,
+ trivial_pathtarget,
pathkeys,
required_outer));
}
*
* 'baserel' is the relation to be scanned
* 'param_info' is the ParamPathInfo if this is a parameterized path, else NULL
+ * 'trivial_pathtarget' is true if the pathtarget is believed to be trivial.
*/
void
cost_subqueryscan(SubqueryScanPath *path, PlannerInfo *root,
- RelOptInfo *baserel, ParamPathInfo *param_info)
+ RelOptInfo *baserel, ParamPathInfo *param_info,
+ bool trivial_pathtarget)
{
Cost startup_cost;
Cost run_cost;
path->path.startup_cost = path->subpath->startup_cost;
path->path.total_cost = path->subpath->total_cost;
+ /*
+ * However, if there are no relevant restriction clauses and the
+ * pathtarget is trivial, then we expect that setrefs.c will optimize away
+ * the SubqueryScan plan node altogether, so we should just make its cost
+ * and rowcount equal to the input path's.
+ *
+ * Note: there are some edge cases where createplan.c will apply a
+ * different targetlist to the SubqueryScan node, thus falsifying our
+ * current estimate of whether the target is trivial, and making the cost
+ * estimate (though not the rowcount) wrong. It does not seem worth the
+ * extra complication to try to account for that exactly, especially since
+ * that behavior falsifies other cost estimates as well.
+ */
+ if (qpquals == NIL && trivial_pathtarget)
+ return;
+
get_restriction_qual_cost(root, baserel, param_info, &qpqual_cost);
startup_cost = qpqual_cost.startup;
/*
* See if it's safe to get rid of the Append entirely. For this to be
* safe, there must be only one child plan and that child plan's parallel
- * awareness must match that of the Append's. The reason for the latter
- * is that if the Append is parallel aware and the child is not, then the
- * calling plan may execute the non-parallel aware child multiple times.
+ * awareness must match the Append's. The reason for the latter is that
+ * if the Append is parallel aware and the child is not, then the calling
+ * plan may execute the non-parallel aware child multiple times. (If you
+ * change these rules, update create_append_path to match.)
*/
if (list_length(aplan->appendplans) == 1)
{
/*
* See if it's safe to get rid of the MergeAppend entirely. For this to
* be safe, there must be only one child plan and that child plan's
- * parallel awareness must match that of the MergeAppend's. The reason
- * for the latter is that if the MergeAppend is parallel aware and the
- * child is not then the calling plan may execute the non-parallel aware
- * child multiple times.
+ * parallel awareness must match the MergeAppend's. The reason for the
+ * latter is that if the MergeAppend is parallel aware and the child is
+ * not, then the calling plan may execute the non-parallel aware child
+ * multiple times. (If you change these rules, update
+ * create_merge_append_path to match.)
*/
if (list_length(mplan->mergeplans) == 1)
{
Index varno,
bool hack_constants,
List *input_tlist,
- List *refnames_tlist);
+ List *refnames_tlist,
+ bool *trivial_tlist);
static List *generate_append_tlist(List *colTypes, List *colCollations,
bool flag,
List *input_tlists,
Path *subpath;
Path *path;
List *tlist;
+ bool trivial_tlist;
Assert(subquery != NULL);
rtr->rtindex,
true,
subroot->processed_tlist,
- refnames_tlist);
+ refnames_tlist,
+ &trivial_tlist);
rel->reltarget = create_pathtarget(root, tlist);
/* Return the fully-fledged tlist to caller, too */
* soon too, likely.)
*/
path = (Path *) create_subqueryscan_path(root, rel, subpath,
+ trivial_tlist,
NIL, NULL);
add_path(rel, path);
partial_subpath = linitial(final_rel->partial_pathlist);
partial_path = (Path *)
create_subqueryscan_path(root, rel, partial_subpath,
+ trivial_tlist,
NIL, NULL);
add_partial_path(rel, partial_path);
}
!tlist_same_collations(*pTargetList, colCollations, junkOK))
{
PathTarget *target;
+ bool trivial_tlist;
ListCell *lc;
*pTargetList = generate_setop_tlist(colTypes, colCollations,
0,
false,
*pTargetList,
- refnames_tlist);
+ refnames_tlist,
+ &trivial_tlist);
target = create_pathtarget(root, *pTargetList);
/* Apply projection to each path */
* hack_constants: true to copy up constants (see comments in code)
* input_tlist: targetlist of this node's input node
* refnames_tlist: targetlist to take column names from
+ * trivial_tlist: output parameter, set to true if targetlist is trivial
*/
static List *
generate_setop_tlist(List *colTypes, List *colCollations,
Index varno,
bool hack_constants,
List *input_tlist,
- List *refnames_tlist)
+ List *refnames_tlist,
+ bool *trivial_tlist)
{
List *tlist = NIL;
int resno = 1;
TargetEntry *tle;
Node *expr;
+ *trivial_tlist = true; /* until proven differently */
+
forfour(ctlc, colTypes, cclc, colCollations,
itlc, input_tlist, rtlc, refnames_tlist)
{
* this only at the first level of subquery-scan plans; we don't want
* phony constants appearing in the output tlists of upper-level
* nodes!
+ *
+ * Note that copying a constant doesn't in itself require us to mark
+ * the tlist nontrivial; see trivial_subqueryscan() in setrefs.c.
*/
if (hack_constants && inputtle->expr && IsA(inputtle->expr, Const))
expr = (Node *) inputtle->expr;
expr,
colType,
"UNION/INTERSECT/EXCEPT");
+ *trivial_tlist = false; /* the coercion makes it not trivial */
}
/*
* will reach the executor without any further processing.
*/
if (exprCollation(expr) != colColl)
+ {
expr = applyRelabelType(expr,
exprType(expr), exprTypmod(expr), colColl,
COERCE_IMPLICIT_CAST, -1, false);
+ *trivial_tlist = false; /* the relabel makes it not trivial */
+ }
tle = makeTargetEntry((Expr *) expr,
(AttrNumber) resno++,
pstrdup("flag"),
true);
tlist = lappend(tlist, tle);
+ *trivial_tlist = false; /* the extra entry makes it not trivial */
}
return tlist;
Assert(!parallel_aware || pathnode->path.parallel_safe);
/*
- * If there's exactly one child path, the Append is a no-op and will be
- * discarded later (in setrefs.c); therefore, we can inherit the child's
- * size and cost, as well as its pathkeys if any (overriding whatever the
- * caller might've said). Otherwise, we must do the normal costsize
+ * If there's exactly one child path then the output of the Append is
+ * necessarily ordered the same as the child's, so we can inherit the
+ * child's pathkeys if any, overriding whatever the caller might've said.
+ * Furthermore, if the child's parallel awareness matches the Append's,
+ * then the Append is a no-op and will be discarded later (in setrefs.c).
+ * Then we can inherit the child's size and cost too, effectively charging
+ * zero for the Append. Otherwise, we must do the normal costsize
* calculation.
*/
if (list_length(pathnode->subpaths) == 1)
{
Path *child = (Path *) linitial(pathnode->subpaths);
- pathnode->path.rows = child->rows;
- pathnode->path.startup_cost = child->startup_cost;
- pathnode->path.total_cost = child->total_cost;
+ if (child->parallel_aware == parallel_aware)
+ {
+ pathnode->path.rows = child->rows;
+ pathnode->path.startup_cost = child->startup_cost;
+ pathnode->path.total_cost = child->total_cost;
+ }
+ else
+ cost_append(pathnode, root);
+ /* Must do this last, else cost_append complains */
pathnode->path.pathkeys = child->pathkeys;
}
else
/*
* Now we can compute total costs of the MergeAppend. If there's exactly
- * one child path, the MergeAppend is a no-op and will be discarded later
- * (in setrefs.c); otherwise we do the normal cost calculation.
+ * one child path and its parallel awareness matches that of the
+ * MergeAppend, then the MergeAppend is a no-op and will be discarded
+ * later (in setrefs.c); otherwise we do the normal cost calculation.
*/
- if (list_length(subpaths) == 1)
+ if (list_length(subpaths) == 1 &&
+ ((Path *) linitial(subpaths))->parallel_aware ==
+ pathnode->path.parallel_aware)
{
pathnode->path.startup_cost = input_startup_cost;
pathnode->path.total_cost = input_total_cost;
* create_subqueryscan_path
* Creates a path corresponding to a scan of a subquery,
* returning the pathnode.
+ *
+ * Caller must pass trivial_pathtarget = true if it believes rel->reltarget to
+ * be trivial, ie just a fetch of all the subquery output columns in order.
+ * While we could determine that here, the caller can usually do it more
+ * efficiently (or at least amortize it over multiple calls).
*/
SubqueryScanPath *
create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
+ bool trivial_pathtarget,
List *pathkeys, Relids required_outer)
{
SubqueryScanPath *pathnode = makeNode(SubqueryScanPath);
pathnode->path.pathkeys = pathkeys;
pathnode->subpath = subpath;
- cost_subqueryscan(pathnode, root, rel, pathnode->path.param_info);
+ cost_subqueryscan(pathnode, root, rel, pathnode->path.param_info,
+ trivial_pathtarget);
return pathnode;
}
case T_SubqueryScan:
{
SubqueryScanPath *spath = (SubqueryScanPath *) path;
+ Path *subpath = spath->subpath;
+ bool trivial_pathtarget;
+
+ /*
+ * If existing node has zero extra cost, we must have decided
+ * its target is trivial. (The converse is not true, because
+ * it might have a trivial target but quals to enforce; but in
+ * that case the new node will too, so it doesn't matter
+ * whether we get the right answer here.)
+ */
+ trivial_pathtarget =
+ (subpath->total_cost == spath->path.total_cost);
return (Path *) create_subqueryscan_path(root,
rel,
- spath->subpath,
+ subpath,
+ trivial_pathtarget,
spath->path.pathkeys,
required_outer);
}
RelOptInfo *baserel, List *tidrangequals,
ParamPathInfo *param_info);
extern void cost_subqueryscan(SubqueryScanPath *path, PlannerInfo *root,
- RelOptInfo *baserel, ParamPathInfo *param_info);
+ RelOptInfo *baserel, ParamPathInfo *param_info,
+ bool trivial_pathtarget);
extern void cost_functionscan(Path *path, PlannerInfo *root,
RelOptInfo *baserel, ParamPathInfo *param_info);
extern void cost_valuesscan(Path *path, PlannerInfo *root,
Relids required_outer,
double *rows);
extern SubqueryScanPath *create_subqueryscan_path(PlannerInfo *root,
- RelOptInfo *rel, Path *subpath,
- List *pathkeys, Relids required_outer);
+ RelOptInfo *rel,
+ Path *subpath,
+ bool trivial_pathtarget,
+ List *pathkeys,
+ Relids required_outer);
extern Path *create_functionscan_path(PlannerInfo *root, RelOptInfo *rel,
List *pathkeys, Relids required_outer);
extern Path *create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel,
------------------------------------------------------------------------------------------
Hash Join
Output: (cte.r).column2, ((ROW("*VALUES*".column1, "*VALUES*".column2))).column2
- Hash Cond: (((ROW("*VALUES*".column1, "*VALUES*".column2))).column1 = (cte.r).column1)
+ Hash Cond: ((cte.r).column1 = ((ROW("*VALUES*".column1, "*VALUES*".column2))).column1)
CTE cte
-> Values Scan on "*VALUES*_1"
Output: ROW("*VALUES*_1".column1, "*VALUES*_1".column2)
- -> Limit
- Output: (ROW("*VALUES*".column1, "*VALUES*".column2))
- -> Values Scan on "*VALUES*"
- Output: ROW("*VALUES*".column1, "*VALUES*".column2)
- -> Hash
+ -> CTE Scan on cte
Output: cte.r
- -> CTE Scan on cte
- Output: cte.r
+ -> Hash
+ Output: (ROW("*VALUES*".column1, "*VALUES*".column2))
+ -> Limit
+ Output: (ROW("*VALUES*".column1, "*VALUES*".column2))
+ -> Values Scan on "*VALUES*"
+ Output: ROW("*VALUES*".column1, "*VALUES*".column2)
(14 rows)
explain (verbose, costs off)
Sort Key: a.q1, a.q2, x.q1, x.q2, (a.q1)
-> Nested Loop
-> Seq Scan on int8_tbl a
- -> Hash Right Join
- Hash Cond: ((a.q1) = x.q2)
- -> Seq Scan on int4_tbl y
+ -> Hash Left Join
+ Hash Cond: (x.q2 = (a.q1))
+ -> Seq Scan on int8_tbl x
-> Hash
- -> Seq Scan on int8_tbl x
+ -> Seq Scan on int4_tbl y
(9 rows)
select * from int8_tbl a,
projects / users / gsingh / postgres.git / commitdiff