* planner.c
* The query optimizer external interface.
*
- * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1996-2006, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California
*
*
* IDENTIFICATION
- * $PostgreSQL: pgsql/src/backend/optimizer/plan/planner.c,v 1.188 2005/06/05 22:32:56 tgl Exp $
+ * $PostgreSQL: pgsql/src/backend/optimizer/plan/planner.c,v 1.204 2006/07/26 00:34:48 momjian Exp $
*
*-------------------------------------------------------------------------
*/
#include <limits.h>
#include "catalog/pg_operator.h"
-#include "catalog/pg_type.h"
#include "executor/executor.h"
#include "executor/nodeAgg.h"
#include "miscadmin.h"
#include "nodes/makefuncs.h"
-#ifdef OPTIMIZER_DEBUG
-#include "nodes/print.h"
-#endif
#include "optimizer/clauses.h"
#include "optimizer/cost.h"
#include "optimizer/pathnode.h"
#include "optimizer/subselect.h"
#include "optimizer/tlist.h"
#include "optimizer/var.h"
-#include "parser/parsetree.h"
+#ifdef OPTIMIZER_DEBUG
+#include "nodes/print.h"
+#endif
#include "parser/parse_expr.h"
#include "parser/parse_oper.h"
-#include "utils/selfuncs.h"
+#include "parser/parsetree.h"
#include "utils/syscache.h"
/* Expression kind codes for preprocess_expression */
-#define EXPRKIND_QUAL 0
-#define EXPRKIND_TARGET 1
-#define EXPRKIND_RTFUNC 2
-#define EXPRKIND_LIMIT 3
-#define EXPRKIND_ININFO 4
+#define EXPRKIND_QUAL 0
+#define EXPRKIND_TARGET 1
+#define EXPRKIND_RTFUNC 2
+#define EXPRKIND_LIMIT 3
+#define EXPRKIND_ININFO 4
+#define EXPRKIND_APPINFO 5
static Node *preprocess_expression(PlannerInfo *root, Node *expr, int kind);
static void preprocess_qual_conditions(PlannerInfo *root, Node *jtnode);
-static Plan *inheritance_planner(PlannerInfo *root, List *inheritlist);
+static Plan *inheritance_planner(PlannerInfo *root);
static Plan *grouping_planner(PlannerInfo *root, double tuple_fraction);
+static double preprocess_limit(PlannerInfo *root,
+ double tuple_fraction,
+ int64 *offset_est, int64 *count_est);
static bool choose_hashed_grouping(PlannerInfo *root, double tuple_fraction,
Path *cheapest_path, Path *sorted_path,
- List *sort_pathkeys, List *group_pathkeys,
double dNumGroups, AggClauseCounts *agg_counts);
static bool hash_safe_grouping(PlannerInfo *root);
static List *make_subplanTargetList(PlannerInfo *root, List *tlist,
* these global state variables must be saved and restored.
*
* Query level and the param list cannot be moved into the per-query
- * PlannerInfo structure since their whole purpose is communication
- * across multiple sub-queries. Also, boundParams is explicitly info
- * from outside the query, and so is likewise better handled as a global
- * variable.
+ * PlannerInfo structure since their whole purpose is communication across
+ * multiple sub-queries. Also, boundParams is explicitly info from outside
+ * the query, and so is likewise better handled as a global variable.
*
* Note we do NOT save and restore PlannerPlanId: it exists to assign
- * unique IDs to SubPlan nodes, and we want those IDs to be unique for
- * the life of a backend. Also, PlannerInitPlan is saved/restored in
+ * unique IDs to SubPlan nodes, and we want those IDs to be unique for the
+ * life of a backend. Also, PlannerInitPlan is saved/restored in
* subquery_planner, not here.
*/
save_PlannerQueryLevel = PlannerQueryLevel;
if (isCursor)
{
/*
- * We have no real idea how many tuples the user will ultimately
- * FETCH from a cursor, but it seems a good bet that he doesn't
- * want 'em all. Optimize for 10% retrieval (you gotta better
- * number? Should this be a SETtable parameter?)
+ * We have no real idea how many tuples the user will ultimately FETCH
+ * from a cursor, but it seems a good bet that he doesn't want 'em
+ * all. Optimize for 10% retrieval (you gotta better number? Should
+ * this be a SETtable parameter?)
*/
tuple_fraction = 0.10;
}
int saved_planid = PlannerPlanId;
PlannerInfo *root;
Plan *plan;
- bool hasOuterJoins;
List *newHaving;
- List *lst;
ListCell *l;
/* Set up for a new level of subquery */
/* Create a PlannerInfo data structure for this subquery */
root = makeNode(PlannerInfo);
root->parse = parse;
+ root->in_info_list = NIL;
+ root->append_rel_list = NIL;
/*
- * Look for IN clauses at the top level of WHERE, and transform them
- * into joins. Note that this step only handles IN clauses originally
- * at top level of WHERE; if we pull up any subqueries in the next
- * step, their INs are processed just before pulling them up.
+ * Look for IN clauses at the top level of WHERE, and transform them into
+ * joins. Note that this step only handles IN clauses originally at top
+ * level of WHERE; if we pull up any subqueries in the next step, their
+ * INs are processed just before pulling them up.
*/
- root->in_info_list = NIL;
if (parse->hasSubLinks)
parse->jointree->quals = pull_up_IN_clauses(root,
parse->jointree->quals);
* this query.
*/
parse->jointree = (FromExpr *)
- pull_up_subqueries(root, (Node *) parse->jointree, false);
+ pull_up_subqueries(root, (Node *) parse->jointree, false, false);
/*
- * Detect whether any rangetable entries are RTE_JOIN kind; if not, we
- * can avoid the expense of doing flatten_join_alias_vars(). Also
- * check for outer joins --- if none, we can skip
- * reduce_outer_joins(). This must be done after we have done
+ * Detect whether any rangetable entries are RTE_JOIN kind; if not, we can
+ * avoid the expense of doing flatten_join_alias_vars(). Also check for
+ * outer joins --- if none, we can skip reduce_outer_joins() and some
+ * other processing. This must be done after we have done
* pull_up_subqueries, of course.
+ *
+ * Note: if reduce_outer_joins manages to eliminate all outer joins,
+ * root->hasOuterJoins is not reset currently. This is OK since its
+ * purpose is merely to suppress unnecessary processing in simple cases.
*/
root->hasJoinRTEs = false;
- hasOuterJoins = false;
+ root->hasOuterJoins = false;
foreach(l, parse->rtable)
{
RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
root->hasJoinRTEs = true;
if (IS_OUTER_JOIN(rte->jointype))
{
- hasOuterJoins = true;
+ root->hasOuterJoins = true;
/* Can quit scanning once we find an outer join */
break;
}
}
}
+ /*
+ * Expand any rangetable entries that are inheritance sets into "append
+ * relations". This can add entries to the rangetable, but they must be
+ * plain base relations not joins, so it's OK (and marginally more
+ * efficient) to do it after checking for join RTEs. We must do it after
+ * pulling up subqueries, else we'd fail to handle inherited tables in
+ * subqueries.
+ */
+ expand_inherited_tables(root);
+
/*
* Set hasHavingQual to remember if HAVING clause is present. Needed
- * because preprocess_expression will reduce a constant-true condition
- * to an empty qual list ... but "HAVING TRUE" is not a semantic no-op.
+ * because preprocess_expression will reduce a constant-true condition to
+ * an empty qual list ... but "HAVING TRUE" is not a semantic no-op.
*/
root->hasHavingQual = (parse->havingQual != NULL);
+ /* Clear this flag; might get set in distribute_qual_to_rels */
+ root->hasPseudoConstantQuals = false;
+
/*
* Do expression preprocessing on targetlist and quals.
*/
root->in_info_list = (List *)
preprocess_expression(root, (Node *) root->in_info_list,
EXPRKIND_ININFO);
+ root->append_rel_list = (List *)
+ preprocess_expression(root, (Node *) root->append_rel_list,
+ EXPRKIND_APPINFO);
/* Also need to preprocess expressions for function RTEs */
foreach(l, parse->rtable)
}
/*
- * In some cases we may want to transfer a HAVING clause into WHERE.
- * We cannot do so if the HAVING clause contains aggregates (obviously)
- * or volatile functions (since a HAVING clause is supposed to be executed
+ * In some cases we may want to transfer a HAVING clause into WHERE. We
+ * cannot do so if the HAVING clause contains aggregates (obviously) or
+ * volatile functions (since a HAVING clause is supposed to be executed
* only once per group). Also, it may be that the clause is so expensive
* to execute that we're better off doing it only once per group, despite
* the loss of selectivity. This is hard to estimate short of doing the
* entire planning process twice, so we use a heuristic: clauses
- * containing subplans are left in HAVING. Otherwise, we move or copy
- * the HAVING clause into WHERE, in hopes of eliminating tuples before
+ * containing subplans are left in HAVING. Otherwise, we move or copy the
+ * HAVING clause into WHERE, in hopes of eliminating tuples before
* aggregation instead of after.
*
* If the query has explicit grouping then we can simply move such a
- * clause into WHERE; any group that fails the clause will not be
- * in the output because none of its tuples will reach the grouping
- * or aggregation stage. Otherwise we must have a degenerate
- * (variable-free) HAVING clause, which we put in WHERE so that
- * query_planner() can use it in a gating Result node, but also keep
- * in HAVING to ensure that we don't emit a bogus aggregated row.
- * (This could be done better, but it seems not worth optimizing.)
+ * clause into WHERE; any group that fails the clause will not be in the
+ * output because none of its tuples will reach the grouping or
+ * aggregation stage. Otherwise we must have a degenerate (variable-free)
+ * HAVING clause, which we put in WHERE so that query_planner() can use it
+ * in a gating Result node, but also keep in HAVING to ensure that we
+ * don't emit a bogus aggregated row. (This could be done better, but it
+ * seems not worth optimizing.)
*
* Note that both havingQual and parse->jointree->quals are in
- * implicitly-ANDed-list form at this point, even though they are
- * declared as Node *.
+ * implicitly-ANDed-list form at this point, even though they are declared
+ * as Node *.
*/
newHaving = NIL;
foreach(l, (List *) parse->havingQual)
parse->havingQual = (Node *) newHaving;
/*
- * If we have any outer joins, try to reduce them to plain inner
- * joins. This step is most easily done after we've done expression
+ * If we have any outer joins, try to reduce them to plain inner joins.
+ * This step is most easily done after we've done expression
* preprocessing.
*/
- if (hasOuterJoins)
+ if (root->hasOuterJoins)
reduce_outer_joins(root);
/*
- * See if we can simplify the jointree; opportunities for this may
- * come from having pulled up subqueries, or from flattening explicit
- * JOIN syntax. We must do this after flattening JOIN alias
- * variables, since eliminating explicit JOIN nodes from the jointree
- * will cause get_relids_for_join() to fail. But it should happen
- * after reduce_outer_joins, anyway.
- */
- parse->jointree = (FromExpr *)
- simplify_jointree(root, (Node *) parse->jointree);
-
- /*
- * Do the main planning. If we have an inherited target relation,
- * that needs special processing, else go straight to
- * grouping_planner.
+ * Do the main planning. If we have an inherited target relation, that
+ * needs special processing, else go straight to grouping_planner.
*/
if (parse->resultRelation &&
- (lst = expand_inherited_rtentry(root, parse->resultRelation)) != NIL)
- plan = inheritance_planner(root, lst);
+ rt_fetch(parse->resultRelation, parse->rtable)->inh)
+ plan = inheritance_planner(root);
else
plan = grouping_planner(root, tuple_fraction);
/*
* If any subplans were generated, or if we're inside a subplan, build
- * initPlan list and extParam/allParam sets for plan nodes, and attach
- * the initPlans to the top plan node.
+ * initPlan list and extParam/allParam sets for plan nodes, and attach the
+ * initPlans to the top plan node.
*/
if (PlannerPlanId != saved_planid || PlannerQueryLevel > 1)
SS_finalize_plan(plan, parse->rtable);
preprocess_expression(PlannerInfo *root, Node *expr, int kind)
{
/*
- * Fall out quickly if expression is empty. This occurs often enough
- * to be worth checking. Note that null->null is the correct conversion
- * for implicit-AND result format, too.
+ * Fall out quickly if expression is empty. This occurs often enough to
+ * be worth checking. Note that null->null is the correct conversion for
+ * implicit-AND result format, too.
*/
if (expr == NULL)
return NULL;
/*
* If the query has any join RTEs, replace join alias variables with
* base-relation variables. We must do this before sublink processing,
- * else sublinks expanded out from join aliases wouldn't get
- * processed.
+ * else sublinks expanded out from join aliases wouldn't get processed.
*/
if (root->hasJoinRTEs)
expr = flatten_join_alias_vars(root, expr);
* with AND directly under AND, nor OR directly under OR.
*
* Because this is a relatively expensive process, we skip it when the
- * query is trivial, such as "SELECT 2+2;" or "INSERT ... VALUES()".
- * The expression will only be evaluated once anyway, so no point in
+ * query is trivial, such as "SELECT 2+2;" or "INSERT ... VALUES()". The
+ * expression will only be evaluated once anyway, so no point in
* pre-simplifying; we can't execute it any faster than the executor can,
* and we will waste cycles copying the tree. Notice however that we
- * still must do it for quals (to get AND/OR flatness); and if we are
- * in a subquery we should not assume it will be done only once.
+ * still must do it for quals (to get AND/OR flatness); and if we are in a
+ * subquery we should not assume it will be done only once.
*/
if (root->parse->jointree->fromlist != NIL ||
kind == EXPRKIND_QUAL ||
expr = SS_process_sublinks(expr, (kind == EXPRKIND_QUAL));
/*
- * XXX do not insert anything here unless you have grokked the
- * comments in SS_replace_correlation_vars ...
+ * XXX do not insert anything here unless you have grokked the comments in
+ * SS_replace_correlation_vars ...
*/
/* Replace uplevel vars with Param nodes */
expr = SS_replace_correlation_vars(expr);
/*
- * If it's a qual or havingQual, convert it to implicit-AND format.
- * (We don't want to do this before eval_const_expressions, since the
- * latter would be unable to simplify a top-level AND correctly. Also,
+ * If it's a qual or havingQual, convert it to implicit-AND format. (We
+ * don't want to do this before eval_const_expressions, since the latter
+ * would be unable to simplify a top-level AND correctly. Also,
* SS_process_sublinks expects explicit-AND format.)
*/
if (kind == EXPRKIND_QUAL)
(int) nodeTag(jtnode));
}
-/*--------------------
+/*
* inheritance_planner
* Generate a plan in the case where the result relation is an
* inheritance set.
*
- * We have to handle this case differently from cases where a source
- * relation is an inheritance set. Source inheritance is expanded at
- * the bottom of the plan tree (see allpaths.c), but target inheritance
- * has to be expanded at the top. The reason is that for UPDATE, each
- * target relation needs a different targetlist matching its own column
- * set. (This is not so critical for DELETE, but for simplicity we treat
- * inherited DELETE the same way.) Fortunately, the UPDATE/DELETE target
- * can never be the nullable side of an outer join, so it's OK to generate
- * the plan this way.
- *
- * inheritlist is an integer list of RT indexes for the result relation set.
+ * We have to handle this case differently from cases where a source relation
+ * is an inheritance set. Source inheritance is expanded at the bottom of the
+ * plan tree (see allpaths.c), but target inheritance has to be expanded at
+ * the top. The reason is that for UPDATE, each target relation needs a
+ * different targetlist matching its own column set. Also, for both UPDATE
+ * and DELETE, the executor needs the Append plan node at the top, else it
+ * can't keep track of which table is the current target table. Fortunately,
+ * the UPDATE/DELETE target can never be the nullable side of an outer join,
+ * so it's OK to generate the plan this way.
*
* Returns a query plan.
- *--------------------
*/
static Plan *
-inheritance_planner(PlannerInfo *root, List *inheritlist)
+inheritance_planner(PlannerInfo *root)
{
Query *parse = root->parse;
int parentRTindex = parse->resultRelation;
- Oid parentOID = getrelid(parentRTindex, parse->rtable);
- int mainrtlength = list_length(parse->rtable);
List *subplans = NIL;
List *tlist = NIL;
+ PlannerInfo subroot;
ListCell *l;
- foreach(l, inheritlist)
+ subroot.parse = NULL; /* catch it if no matches in loop */
+
+ parse->resultRelations = NIL;
+
+ foreach(l, root->append_rel_list)
{
- int childRTindex = lfirst_int(l);
- Oid childOID = getrelid(childRTindex, parse->rtable);
- PlannerInfo subroot;
+ AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
Plan *subplan;
+ /* append_rel_list contains all append rels; ignore others */
+ if (appinfo->parent_relid != parentRTindex)
+ continue;
+
+ /* Build target-relations list for the executor */
+ parse->resultRelations = lappend_int(parse->resultRelations,
+ appinfo->child_relid);
+
/*
- * Generate modified query with this rel as target. We have to
- * be prepared to translate varnos in in_info_list as well as in
- * the Query proper.
+ * Generate modified query with this rel as target. We have to be
+ * prepared to translate varnos in in_info_list as well as in the
+ * Query proper.
*/
memcpy(&subroot, root, sizeof(PlannerInfo));
subroot.parse = (Query *)
- adjust_inherited_attrs((Node *) parse,
- parentRTindex, parentOID,
- childRTindex, childOID);
+ adjust_appendrel_attrs((Node *) parse,
+ appinfo);
subroot.in_info_list = (List *)
- adjust_inherited_attrs((Node *) root->in_info_list,
- parentRTindex, parentOID,
- childRTindex, childOID);
+ adjust_appendrel_attrs((Node *) root->in_info_list,
+ appinfo);
+ /* There shouldn't be any OJ info to translate, as yet */
+ Assert(subroot.oj_info_list == NIL);
/* Generate plan */
subplan = grouping_planner(&subroot, 0.0 /* retrieve all tuples */ );
subplans = lappend(subplans, subplan);
- /*
- * XXX my goodness this next bit is ugly. Really need to think about
- * ways to rein in planner's habit of scribbling on its input.
- *
- * Planning of the subquery might have modified the rangetable,
- * either by addition of RTEs due to expansion of inherited source
- * tables, or by changes of the Query structures inside subquery
- * RTEs. We have to ensure that this gets propagated back to the
- * master copy. However, if we aren't done planning yet, we also
- * need to ensure that subsequent calls to grouping_planner have
- * virgin sub-Queries to work from. So, if we are at the last
- * list entry, just copy the subquery rangetable back to the master
- * copy; if we are not, then extend the master copy by adding
- * whatever the subquery added. (We assume these added entries
- * will go untouched by the future grouping_planner calls. We are
- * also effectively assuming that sub-Queries will get planned
- * identically each time, or at least that the impacts on their
- * rangetables will be the same each time. Did I say this is ugly?)
- */
- if (lnext(l) == NULL)
- parse->rtable = subroot.parse->rtable;
- else
- {
- int subrtlength = list_length(subroot.parse->rtable);
-
- if (subrtlength > mainrtlength)
- {
- List *subrt;
-
- subrt = list_copy_tail(subroot.parse->rtable, mainrtlength);
- parse->rtable = list_concat(parse->rtable, subrt);
- mainrtlength = subrtlength;
- }
- }
-
/* Save preprocessed tlist from first rel for use in Append */
if (tlist == NIL)
tlist = subplan->targetlist;
}
- /* Save the target-relations list for the executor, too */
- parse->resultRelations = inheritlist;
+ /*
+ * Planning might have modified the rangetable, due to changes of the
+ * Query structures inside subquery RTEs. We have to ensure that this
+ * gets propagated back to the master copy. But can't do this until we
+ * are done planning, because all the calls to grouping_planner need
+ * virgin sub-Queries to work from. (We are effectively assuming that
+ * sub-Queries will get planned identically each time, or at least that
+ * the impacts on their rangetables will be the same each time.)
+ *
+ * XXX should clean this up someday
+ */
+ parse->rtable = subroot.parse->rtable;
/* Mark result as unordered (probably unnecessary) */
root->query_pathkeys = NIL;
{
Query *parse = root->parse;
List *tlist = parse->targetList;
+ int64 offset_est = 0;
+ int64 count_est = 0;
Plan *result_plan;
List *current_pathkeys;
List *sort_pathkeys;
+ double dNumGroups = 0;
+
+ /* Tweak caller-supplied tuple_fraction if have LIMIT/OFFSET */
+ if (parse->limitCount || parse->limitOffset)
+ tuple_fraction = preprocess_limit(root, tuple_fraction,
+ &offset_est, &count_est);
if (parse->setOperations)
{
List *set_sortclauses;
/*
- * Construct the plan for set operations. The result will not
- * need any work except perhaps a top-level sort and/or LIMIT.
+ * If there's a top-level ORDER BY, assume we have to fetch all the
+ * tuples. This might seem too simplistic given all the hackery below
+ * to possibly avoid the sort ... but a nonzero tuple_fraction is only
+ * of use to plan_set_operations() when the setop is UNION ALL, and
+ * the result of UNION ALL is always unsorted.
*/
- result_plan = plan_set_operations(root,
+ if (parse->sortClause)
+ tuple_fraction = 0.0;
+
+ /*
+ * Construct the plan for set operations. The result will not need
+ * any work except perhaps a top-level sort and/or LIMIT.
+ */
+ result_plan = plan_set_operations(root, tuple_fraction,
&set_sortclauses);
/*
- * Calculate pathkeys representing the sort order (if any) of the
- * set operation's result. We have to do this before overwriting
- * the sort key information...
+ * Calculate pathkeys representing the sort order (if any) of the set
+ * operation's result. We have to do this before overwriting the sort
+ * key information...
*/
current_pathkeys = make_pathkeys_for_sortclauses(set_sortclauses,
- result_plan->targetlist);
+ result_plan->targetlist);
current_pathkeys = canonicalize_pathkeys(root, current_pathkeys);
/*
- * We should not need to call preprocess_targetlist, since we must
- * be in a SELECT query node. Instead, use the targetlist
- * returned by plan_set_operations (since this tells whether it
- * returned any resjunk columns!), and transfer any sort key
- * information from the original tlist.
+ * We should not need to call preprocess_targetlist, since we must be
+ * in a SELECT query node. Instead, use the targetlist returned by
+ * plan_set_operations (since this tells whether it returned any
+ * resjunk columns!), and transfer any sort key information from the
+ * original tlist.
*/
Assert(parse->commandType == CMD_SELECT);
AttrNumber *groupColIdx = NULL;
bool need_tlist_eval = true;
QualCost tlist_cost;
- double sub_tuple_fraction;
Path *cheapest_path;
Path *sorted_path;
Path *best_path;
- double dNumGroups = 0;
long numGroups = 0;
AggClauseCounts agg_counts;
int numGroupCols = list_length(parse->groupClause);
tlist = preprocess_targetlist(root, tlist);
/*
- * Generate appropriate target list for subplan; may be different
- * from tlist if grouping or aggregation is needed.
+ * Generate appropriate target list for subplan; may be different from
+ * tlist if grouping or aggregation is needed.
*/
sub_tlist = make_subplanTargetList(root, tlist,
- &groupColIdx, &need_tlist_eval);
+ &groupColIdx, &need_tlist_eval);
/*
- * Calculate pathkeys that represent grouping/ordering
- * requirements
+ * Calculate pathkeys that represent grouping/ordering requirements.
+ * Stash them in PlannerInfo so that query_planner can canonicalize
+ * them.
*/
- group_pathkeys = make_pathkeys_for_sortclauses(parse->groupClause,
- tlist);
- sort_pathkeys = make_pathkeys_for_sortclauses(parse->sortClause,
- tlist);
+ root->group_pathkeys =
+ make_pathkeys_for_sortclauses(parse->groupClause, tlist);
+ root->sort_pathkeys =
+ make_pathkeys_for_sortclauses(parse->sortClause, tlist);
/*
* Will need actual number of aggregates for estimating costs.
* Note: we do not attempt to detect duplicate aggregates here; a
* somewhat-overestimated count is okay for our present purposes.
*
- * Note: think not that we can turn off hasAggs if we find no aggs.
- * It is possible for constant-expression simplification to remove
- * all explicit references to aggs, but we still have to follow
- * the aggregate semantics (eg, producing only one output row).
+ * Note: think not that we can turn off hasAggs if we find no aggs. It
+ * is possible for constant-expression simplification to remove all
+ * explicit references to aggs, but we still have to follow the
+ * aggregate semantics (eg, producing only one output row).
*/
if (parse->hasAggs)
{
/*
* Figure out whether we need a sorted result from query_planner.
*
- * If we have a GROUP BY clause, then we want a result sorted
- * properly for grouping. Otherwise, if there is an ORDER BY
- * clause, we want to sort by the ORDER BY clause. (Note: if we
- * have both, and ORDER BY is a superset of GROUP BY, it would be
- * tempting to request sort by ORDER BY --- but that might just
- * leave us failing to exploit an available sort order at all.
- * Needs more thought...)
+ * If we have a GROUP BY clause, then we want a result sorted properly
+ * for grouping. Otherwise, if there is an ORDER BY clause, we want
+ * to sort by the ORDER BY clause. (Note: if we have both, and ORDER
+ * BY is a superset of GROUP BY, it would be tempting to request sort
+ * by ORDER BY --- but that might just leave us failing to exploit an
+ * available sort order at all. Needs more thought...)
*/
if (parse->groupClause)
- root->query_pathkeys = group_pathkeys;
+ root->query_pathkeys = root->group_pathkeys;
else if (parse->sortClause)
- root->query_pathkeys = sort_pathkeys;
+ root->query_pathkeys = root->sort_pathkeys;
else
root->query_pathkeys = NIL;
/*
- * Adjust tuple_fraction if we see that we are going to apply
- * limiting/grouping/aggregation/etc. This is not overridable by
- * the caller, since it reflects plan actions that this routine
- * will certainly take, not assumptions about context.
+ * Generate the best unsorted and presorted paths for this Query (but
+ * note there may not be any presorted path). query_planner will also
+ * estimate the number of groups in the query, and canonicalize all
+ * the pathkeys.
*/
- if (parse->limitCount != NULL)
- {
- /*
- * A LIMIT clause limits the absolute number of tuples
- * returned. However, if it's not a constant LIMIT then we
- * have to punt; for lack of a better idea, assume 10% of the
- * plan's result is wanted.
- */
- double limit_fraction = 0.0;
-
- if (IsA(parse->limitCount, Const))
- {
- Const *limitc = (Const *) parse->limitCount;
- int32 count = DatumGetInt32(limitc->constvalue);
-
- /*
- * A NULL-constant LIMIT represents "LIMIT ALL", which we
- * treat the same as no limit (ie, expect to retrieve all
- * the tuples).
- */
- if (!limitc->constisnull && count > 0)
- {
- limit_fraction = (double) count;
- /* We must also consider the OFFSET, if present */
- if (parse->limitOffset != NULL)
- {
- if (IsA(parse->limitOffset, Const))
- {
- int32 offset;
-
- limitc = (Const *) parse->limitOffset;
- offset = DatumGetInt32(limitc->constvalue);
- if (!limitc->constisnull && offset > 0)
- limit_fraction += (double) offset;
- }
- else
- {
- /* OFFSET is an expression ... punt ... */
- limit_fraction = 0.10;
- }
- }
- }
- }
- else
- {
- /* LIMIT is an expression ... punt ... */
- limit_fraction = 0.10;
- }
+ query_planner(root, sub_tlist, tuple_fraction,
+ &cheapest_path, &sorted_path, &dNumGroups);
- if (limit_fraction > 0.0)
- {
- /*
- * If we have absolute limits from both caller and LIMIT,
- * use the smaller value; if one is fractional and the
- * other absolute, treat the fraction as a fraction of the
- * absolute value; else we can multiply the two fractions
- * together.
- */
- if (tuple_fraction >= 1.0)
- {
- if (limit_fraction >= 1.0)
- {
- /* both absolute */
- tuple_fraction = Min(tuple_fraction, limit_fraction);
- }
- else
- {
- /* caller absolute, limit fractional */
- tuple_fraction *= limit_fraction;
- if (tuple_fraction < 1.0)
- tuple_fraction = 1.0;
- }
- }
- else if (tuple_fraction > 0.0)
- {
- if (limit_fraction >= 1.0)
- {
- /* caller fractional, limit absolute */
- tuple_fraction *= limit_fraction;
- if (tuple_fraction < 1.0)
- tuple_fraction = 1.0;
- }
- else
- {
- /* both fractional */
- tuple_fraction *= limit_fraction;
- }
- }
- else
- {
- /* no info from caller, just use limit */
- tuple_fraction = limit_fraction;
- }
- }
- }
+ group_pathkeys = root->group_pathkeys;
+ sort_pathkeys = root->sort_pathkeys;
/*
- * With grouping or aggregation, the tuple fraction to pass to
- * query_planner() may be different from what it is at top level.
+ * If grouping, decide whether we want to use hashed grouping.
*/
- sub_tuple_fraction = tuple_fraction;
-
if (parse->groupClause)
{
- /*
- * In GROUP BY mode, we have the little problem that we don't
- * really know how many input tuples will be needed to make a
- * group, so we can't translate an output LIMIT count into an
- * input count. For lack of a better idea, assume 25% of the
- * input data will be processed if there is any output limit.
- * However, if the caller gave us a fraction rather than an
- * absolute count, we can keep using that fraction (which
- * amounts to assuming that all the groups are about the same
- * size).
- */
- if (sub_tuple_fraction >= 1.0)
- sub_tuple_fraction = 0.25;
-
- /*
- * If both GROUP BY and ORDER BY are specified, we will need
- * two levels of sort --- and, therefore, certainly need to
- * read all the input tuples --- unless ORDER BY is a subset
- * of GROUP BY. (We have not yet canonicalized the pathkeys,
- * so must use the slower noncanonical comparison method.)
- */
- if (parse->groupClause && parse->sortClause &&
- !noncanonical_pathkeys_contained_in(sort_pathkeys,
- group_pathkeys))
- sub_tuple_fraction = 0.0;
- }
- else if (parse->hasAggs)
- {
- /*
- * Ungrouped aggregate will certainly want all the input
- * tuples.
- */
- sub_tuple_fraction = 0.0;
- }
- else if (parse->distinctClause)
- {
- /*
- * SELECT DISTINCT, like GROUP, will absorb an unpredictable
- * number of input tuples per output tuple. Handle the same
- * way.
- */
- if (sub_tuple_fraction >= 1.0)
- sub_tuple_fraction = 0.25;
- }
-
- /*
- * Generate the best unsorted and presorted paths for this Query
- * (but note there may not be any presorted path).
- */
- query_planner(root, sub_tlist, sub_tuple_fraction,
- &cheapest_path, &sorted_path);
-
- /*
- * We couldn't canonicalize group_pathkeys and sort_pathkeys
- * before running query_planner(), so do it now.
- */
- group_pathkeys = canonicalize_pathkeys(root, group_pathkeys);
- sort_pathkeys = canonicalize_pathkeys(root, sort_pathkeys);
-
- /*
- * If grouping, estimate the number of groups. (We can't do this
- * until after running query_planner(), either.) Then decide
- * whether we want to use hashed grouping.
- */
- if (parse->groupClause)
- {
- List *groupExprs;
- double cheapest_path_rows;
-
- /*
- * Beware of the possibility that cheapest_path->parent is NULL.
- * This could happen if user does something silly like
- * SELECT 'foo' GROUP BY 1;
- */
- if (cheapest_path->parent)
- cheapest_path_rows = cheapest_path->parent->rows;
- else
- cheapest_path_rows = 1; /* assume non-set result */
-
- groupExprs = get_sortgrouplist_exprs(parse->groupClause,
- parse->targetList);
- dNumGroups = estimate_num_groups(root,
- groupExprs,
- cheapest_path_rows);
- /* Also want it as a long int --- but 'ware overflow! */
- numGroups = (long) Min(dNumGroups, (double) LONG_MAX);
-
use_hashed_grouping =
choose_hashed_grouping(root, tuple_fraction,
cheapest_path, sorted_path,
- sort_pathkeys, group_pathkeys,
dNumGroups, &agg_counts);
+
+ /* Also convert # groups to long int --- but 'ware overflow! */
+ numGroups = (long) Min(dNumGroups, (double) LONG_MAX);
}
/*
* Select the best path. If we are doing hashed grouping, we will
- * always read all the input tuples, so use the cheapest-total
- * path. Otherwise, trust query_planner's decision about which to use.
+ * always read all the input tuples, so use the cheapest-total path.
+ * Otherwise, trust query_planner's decision about which to use.
*/
if (use_hashed_grouping || !sorted_path)
best_path = cheapest_path;
best_path = sorted_path;
/*
- * Check to see if it's possible to optimize MIN/MAX aggregates.
- * If so, we will forget all the work we did so far to choose a
- * "regular" path ... but we had to do it anyway to be able to
- * tell which way is cheaper.
+ * Check to see if it's possible to optimize MIN/MAX aggregates. If
+ * so, we will forget all the work we did so far to choose a "regular"
+ * path ... but we had to do it anyway to be able to tell which way is
+ * cheaper.
*/
result_plan = optimize_minmax_aggregates(root,
tlist,
if (result_plan != NULL)
{
/*
- * optimize_minmax_aggregates generated the full plan, with
- * the right tlist, and it has no sort order.
+ * optimize_minmax_aggregates generated the full plan, with the
+ * right tlist, and it has no sort order.
*/
current_pathkeys = NIL;
}
* GROUP BY without aggregation, so insert a group node (plus
* the appropriate sort node, if necessary).
*
- * Add an explicit sort if we couldn't make the path come
- * out the way the GROUP node needs it.
+ * Add an explicit sort if we couldn't make the path come out
+ * the way the GROUP node needs it.
*/
if (!pathkeys_contained_in(group_pathkeys, current_pathkeys))
{
* This is a degenerate case in which we are supposed to emit
* either 0 or 1 row depending on whether HAVING succeeds.
* Furthermore, there cannot be any variables in either HAVING
- * or the targetlist, so we actually do not need the FROM table
- * at all! We can just throw away the plan-so-far and generate
- * a Result node. This is a sufficiently unusual corner case
- * that it's not worth contorting the structure of this routine
- * to avoid having to generate the plan in the first place.
+ * or the targetlist, so we actually do not need the FROM
+ * table at all! We can just throw away the plan-so-far and
+ * generate a Result node. This is a sufficiently unusual
+ * corner case that it's not worth contorting the structure of
+ * this routine to avoid having to generate the plan in the
+ * first place.
*/
result_plan = (Plan *) make_result(tlist,
parse->havingQual,
} /* end of if (setOperations) */
/*
- * If we were not able to make the plan come out in the right order,
- * add an explicit sort step.
+ * If we were not able to make the plan come out in the right order, add
+ * an explicit sort step.
*/
if (parse->sortClause)
{
result_plan = (Plan *) make_unique(result_plan, parse->distinctClause);
/*
- * If there was grouping or aggregation, leave plan_rows as-is
- * (ie, assume the result was already mostly unique). If not,
- * it's reasonable to assume the UNIQUE filter has effects
- * comparable to GROUP BY.
+ * If there was grouping or aggregation, leave plan_rows as-is (ie,
+ * assume the result was already mostly unique). If not, use the
+ * number of distinct-groups calculated by query_planner.
*/
if (!parse->groupClause && !root->hasHavingQual && !parse->hasAggs)
- {
- List *distinctExprs;
-
- distinctExprs = get_sortgrouplist_exprs(parse->distinctClause,
- parse->targetList);
- result_plan->plan_rows = estimate_num_groups(root,
- distinctExprs,
- result_plan->plan_rows);
- }
+ result_plan->plan_rows = dNumGroups;
}
/*
* Finally, if there is a LIMIT/OFFSET clause, add the LIMIT node.
*/
- if (parse->limitOffset || parse->limitCount)
+ if (parse->limitCount || parse->limitOffset)
{
result_plan = (Plan *) make_limit(result_plan,
parse->limitOffset,
- parse->limitCount);
+ parse->limitCount,
+ offset_est,
+ count_est);
}
/*
- * Return the actual output ordering in query_pathkeys for possible
- * use by an outer query level.
+ * Return the actual output ordering in query_pathkeys for possible use by
+ * an outer query level.
*/
root->query_pathkeys = current_pathkeys;
return result_plan;
}
+/*
+ * preprocess_limit - do pre-estimation for LIMIT and/or OFFSET clauses
+ *
+ * We try to estimate the values of the LIMIT/OFFSET clauses, and pass the
+ * results back in *count_est and *offset_est. These variables are set to
+ * 0 if the corresponding clause is not present, and -1 if it's present
+ * but we couldn't estimate the value for it. (The "0" convention is OK
+ * for OFFSET but a little bit bogus for LIMIT: effectively we estimate
+ * LIMIT 0 as though it were LIMIT 1. But this is in line with the planner's
+ * usual practice of never estimating less than one row.) These values will
+ * be passed to make_limit, which see if you change this code.
+ *
+ * The return value is the suitably adjusted tuple_fraction to use for
+ * planning the query. This adjustment is not overridable, since it reflects
+ * plan actions that grouping_planner() will certainly take, not assumptions
+ * about context.
+ */
+static double
+preprocess_limit(PlannerInfo *root, double tuple_fraction,
+ int64 *offset_est, int64 *count_est)
+{
+ Query *parse = root->parse;
+ Node *est;
+ double limit_fraction;
+
+ /* Should not be called unless LIMIT or OFFSET */
+ Assert(parse->limitCount || parse->limitOffset);
+
+ /*
+ * Try to obtain the clause values. We use estimate_expression_value
+ * primarily because it can sometimes do something useful with Params.
+ */
+ if (parse->limitCount)
+ {
+ est = estimate_expression_value(parse->limitCount);
+ if (est && IsA(est, Const))
+ {
+ if (((Const *) est)->constisnull)
+ {
+ /* NULL indicates LIMIT ALL, ie, no limit */
+ *count_est = 0; /* treat as not present */
+ }
+ else
+ {
+ *count_est = DatumGetInt64(((Const *) est)->constvalue);
+ if (*count_est <= 0)
+ *count_est = 1; /* force to at least 1 */
+ }
+ }
+ else
+ *count_est = -1; /* can't estimate */
+ }
+ else
+ *count_est = 0; /* not present */
+
+ if (parse->limitOffset)
+ {
+ est = estimate_expression_value(parse->limitOffset);
+ if (est && IsA(est, Const))
+ {
+ if (((Const *) est)->constisnull)
+ {
+ /* Treat NULL as no offset; the executor will too */
+ *offset_est = 0; /* treat as not present */
+ }
+ else
+ {
+ *offset_est = DatumGetInt64(((Const *) est)->constvalue);
+
+ if (*offset_est < 0)
+ *offset_est = 0; /* less than 0 is same as 0 */
+ }
+ }
+ else
+ *offset_est = -1; /* can't estimate */
+ }
+ else
+ *offset_est = 0; /* not present */
+
+ if (*count_est != 0)
+ {
+ /*
+ * A LIMIT clause limits the absolute number of tuples returned.
+ * However, if it's not a constant LIMIT then we have to guess; for
+ * lack of a better idea, assume 10% of the plan's result is wanted.
+ */
+ if (*count_est < 0 || *offset_est < 0)
+ {
+ /* LIMIT or OFFSET is an expression ... punt ... */
+ limit_fraction = 0.10;
+ }
+ else
+ {
+ /* LIMIT (plus OFFSET, if any) is max number of tuples needed */
+ limit_fraction = (double) *count_est + (double) *offset_est;
+ }
+
+ /*
+ * If we have absolute limits from both caller and LIMIT, use the
+ * smaller value; likewise if they are both fractional. If one is
+ * fractional and the other absolute, we can't easily determine which
+ * is smaller, but we use the heuristic that the absolute will usually
+ * be smaller.
+ */
+ if (tuple_fraction >= 1.0)
+ {
+ if (limit_fraction >= 1.0)
+ {
+ /* both absolute */
+ tuple_fraction = Min(tuple_fraction, limit_fraction);
+ }
+ else
+ {
+ /* caller absolute, limit fractional; use caller's value */
+ }
+ }
+ else if (tuple_fraction > 0.0)
+ {
+ if (limit_fraction >= 1.0)
+ {
+ /* caller fractional, limit absolute; use limit */
+ tuple_fraction = limit_fraction;
+ }
+ else
+ {
+ /* both fractional */
+ tuple_fraction = Min(tuple_fraction, limit_fraction);
+ }
+ }
+ else
+ {
+ /* no info from caller, just use limit */
+ tuple_fraction = limit_fraction;
+ }
+ }
+ else if (*offset_est != 0 && tuple_fraction > 0.0)
+ {
+ /*
+ * We have an OFFSET but no LIMIT. This acts entirely differently
+ * from the LIMIT case: here, we need to increase rather than decrease
+ * the caller's tuple_fraction, because the OFFSET acts to cause more
+ * tuples to be fetched instead of fewer. This only matters if we got
+ * a tuple_fraction > 0, however.
+ *
+ * As above, use 10% if OFFSET is present but unestimatable.
+ */
+ if (*offset_est < 0)
+ limit_fraction = 0.10;
+ else
+ limit_fraction = (double) *offset_est;
+
+ /*
+ * If we have absolute counts from both caller and OFFSET, add them
+ * together; likewise if they are both fractional. If one is
+ * fractional and the other absolute, we want to take the larger, and
+ * we heuristically assume that's the fractional one.
+ */
+ if (tuple_fraction >= 1.0)
+ {
+ if (limit_fraction >= 1.0)
+ {
+ /* both absolute, so add them together */
+ tuple_fraction += limit_fraction;
+ }
+ else
+ {
+ /* caller absolute, limit fractional; use limit */
+ tuple_fraction = limit_fraction;
+ }
+ }
+ else
+ {
+ if (limit_fraction >= 1.0)
+ {
+ /* caller fractional, limit absolute; use caller's value */
+ }
+ else
+ {
+ /* both fractional, so add them together */
+ tuple_fraction += limit_fraction;
+ if (tuple_fraction >= 1.0)
+ tuple_fraction = 0.0; /* assume fetch all */
+ }
+ }
+ }
+
+ return tuple_fraction;
+}
+
/*
* choose_hashed_grouping - should we use hashed grouping?
*/
static bool
choose_hashed_grouping(PlannerInfo *root, double tuple_fraction,
Path *cheapest_path, Path *sorted_path,
- List *sort_pathkeys, List *group_pathkeys,
double dNumGroups, AggClauseCounts *agg_counts)
{
int numGroupCols = list_length(root->parse->groupClause);
* Don't do it if it doesn't look like the hashtable will fit into
* work_mem.
*
- * Beware here of the possibility that cheapest_path->parent is NULL.
- * This could happen if user does something silly like
- * SELECT 'foo' GROUP BY 1;
+ * Beware here of the possibility that cheapest_path->parent is NULL. This
+ * could happen if user does something silly like SELECT 'foo' GROUP BY 1;
*/
if (cheapest_path->parent)
{
}
else
{
- cheapest_path_rows = 1; /* assume non-set result */
- cheapest_path_width = 100; /* arbitrary */
+ cheapest_path_rows = 1; /* assume non-set result */
+ cheapest_path_width = 100; /* arbitrary */
}
/* Estimate per-hash-entry space at tuple width... */
- hashentrysize = cheapest_path_width;
+ hashentrysize = MAXALIGN(cheapest_path_width) + MAXALIGN(sizeof(MinimalTupleData));
/* plus space for pass-by-ref transition values... */
hashentrysize += agg_counts->transitionSpace;
/* plus the per-hash-entry overhead */
return false;
/*
- * See if the estimated cost is no more than doing it the other way.
- * While avoiding the need for sorted input is usually a win, the fact
- * that the output won't be sorted may be a loss; so we need to do an
- * actual cost comparison.
+ * See if the estimated cost is no more than doing it the other way. While
+ * avoiding the need for sorted input is usually a win, the fact that the
+ * output won't be sorted may be a loss; so we need to do an actual cost
+ * comparison.
*
- * We need to consider
- * cheapest_path + hashagg [+ final sort]
- * versus either
- * cheapest_path [+ sort] + group or agg [+ final sort]
- * or
- * presorted_path + group or agg [+ final sort]
- * where brackets indicate a step that may not be needed. We assume
- * query_planner() will have returned a presorted path only if it's a
- * winner compared to cheapest_path for this purpose.
+ * We need to consider cheapest_path + hashagg [+ final sort] versus
+ * either cheapest_path [+ sort] + group or agg [+ final sort] or
+ * presorted_path + group or agg [+ final sort] where brackets indicate a
+ * step that may not be needed. We assume query_planner() will have
+ * returned a presorted path only if it's a winner compared to
+ * cheapest_path for this purpose.
*
* These path variables are dummies that just hold cost fields; we don't
* make actual Paths for these steps.
cheapest_path->startup_cost, cheapest_path->total_cost,
cheapest_path_rows);
/* Result of hashed agg is always unsorted */
- if (sort_pathkeys)
- cost_sort(&hashed_p, root, sort_pathkeys, hashed_p.total_cost,
+ if (root->sort_pathkeys)
+ cost_sort(&hashed_p, root, root->sort_pathkeys, hashed_p.total_cost,
dNumGroups, cheapest_path_width);
if (sorted_path)
sorted_p.total_cost = cheapest_path->total_cost;
current_pathkeys = cheapest_path->pathkeys;
}
- if (!pathkeys_contained_in(group_pathkeys,
- current_pathkeys))
+ if (!pathkeys_contained_in(root->group_pathkeys, current_pathkeys))
{
- cost_sort(&sorted_p, root, group_pathkeys, sorted_p.total_cost,
+ cost_sort(&sorted_p, root, root->group_pathkeys, sorted_p.total_cost,
cheapest_path_rows, cheapest_path_width);
- current_pathkeys = group_pathkeys;
+ current_pathkeys = root->group_pathkeys;
}
if (root->parse->hasAggs)
sorted_p.startup_cost, sorted_p.total_cost,
cheapest_path_rows);
/* The Agg or Group node will preserve ordering */
- if (sort_pathkeys &&
- !pathkeys_contained_in(sort_pathkeys, current_pathkeys))
- cost_sort(&sorted_p, root, sort_pathkeys, sorted_p.total_cost,
+ if (root->sort_pathkeys &&
+ !pathkeys_contained_in(root->sort_pathkeys, current_pathkeys))
+ cost_sort(&sorted_p, root, root->sort_pathkeys, sorted_p.total_cost,
dNumGroups, cheapest_path_width);
/*
/*
* Otherwise, start with a "flattened" tlist (having just the vars
- * mentioned in the targetlist and HAVING qual --- but not upper-
- * level Vars; they will be replaced by Params later on).
+ * mentioned in the targetlist and HAVING qual --- but not upper- level
+ * Vars; they will be replaced by Params later on).
*/
sub_tlist = flatten_tlist(tlist);
extravars = pull_var_clause(parse->havingQual, false);
/*
* If grouping, create sub_tlist entries for all GROUP BY expressions
- * (GROUP BY items that are simple Vars should be in the list
- * already), and make an array showing where the group columns are in
- * the sub_tlist.
+ * (GROUP BY items that are simple Vars should be in the list already),
+ * and make an array showing where the group columns are in the sub_tlist.
*/
numCols = list_length(parse->groupClause);
if (numCols > 0)
Assert(orig_tlist_item != NULL);
orig_tle = (TargetEntry *) lfirst(orig_tlist_item);
orig_tlist_item = lnext(orig_tlist_item);
- if (orig_tle->resjunk) /* should not happen */
+ if (orig_tle->resjunk) /* should not happen */
elog(ERROR, "resjunk output columns are not implemented");
Assert(new_tle->resno == orig_tle->resno);
new_tle->ressortgroupref = orig_tle->ressortgroupref;
projects / postgresql / blobdiff