X-Git-Url: https://granicus.if.org/sourcecode?a=blobdiff_plain;f=src%2Fbackend%2Foptimizer%2Fplan%2Fplanner.c;h=b542fef61e26a697c1fbd44dec2be43147456b70;hb=addc42c339208d6a7a1d652fbf388e8aea7f80b9;hp=2c1081f2677ac7ddefaf9c24cbc63316c708dc76;hpb=ea0b5c856922c08ef2ce6e2eb26f6c68d20bf957;p=postgresql diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c index 2c1081f267..b542fef61e 100644 --- a/src/backend/optimizer/plan/planner.c +++ b/src/backend/optimizer/plan/planner.c @@ -3,12 +3,12 @@ * planner.c * The query optimizer external interface. * - * Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group + * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * * * IDENTIFICATION - * $Header: /cvsroot/pgsql/src/backend/optimizer/plan/planner.c,v 1.131 2002/11/26 03:01:58 tgl Exp $ + * $PostgreSQL: pgsql/src/backend/optimizer/plan/planner.c,v 1.184 2005/04/11 23:06:55 tgl Exp $ * *------------------------------------------------------------------------- */ @@ -17,7 +17,10 @@ #include +#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 @@ -33,36 +36,39 @@ #include "optimizer/subselect.h" #include "optimizer/tlist.h" #include "optimizer/var.h" -#include "parser/analyze.h" #include "parser/parsetree.h" #include "parser/parse_expr.h" -#include "rewrite/rewriteManip.h" -#include "utils/lsyscache.h" +#include "parser/parse_oper.h" #include "utils/selfuncs.h" +#include "utils/syscache.h" + + +ParamListInfo PlannerBoundParamList = NULL; /* current boundParams */ /* Expression kind codes for preprocess_expression */ -#define EXPRKIND_TARGET 0 -#define EXPRKIND_WHERE 1 -#define EXPRKIND_HAVING 2 +#define EXPRKIND_QUAL 0 +#define EXPRKIND_TARGET 1 +#define EXPRKIND_RTFUNC 2 +#define EXPRKIND_LIMIT 3 +#define EXPRKIND_ININFO 4 -static Node *pull_up_subqueries(Query *parse, Node *jtnode, - bool below_outer_join); -static bool is_simple_subquery(Query *subquery); -static bool has_nullable_targetlist(Query *subquery); -static void resolvenew_in_jointree(Node *jtnode, int varno, List *subtlist); -static Node *preprocess_jointree(Query *parse, Node *jtnode); static Node *preprocess_expression(Query *parse, Node *expr, int kind); static void preprocess_qual_conditions(Query *parse, Node *jtnode); static Plan *inheritance_planner(Query *parse, List *inheritlist); static Plan *grouping_planner(Query *parse, double tuple_fraction); +static bool choose_hashed_grouping(Query *parse, 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(Query *parse); static List *make_subplanTargetList(Query *parse, List *tlist, - AttrNumber **groupColIdx); -static Plan *make_groupsortplan(Query *parse, - List *groupClause, - AttrNumber *grpColIdx, - Plan *subplan); + AttrNumber **groupColIdx, bool *need_tlist_eval); +static void locate_grouping_columns(Query *parse, + List *tlist, + List *sub_tlist, + AttrNumber *groupColIdx); static List *postprocess_setop_tlist(List *new_tlist, List *orig_tlist); @@ -72,19 +78,25 @@ static List *postprocess_setop_tlist(List *new_tlist, List *orig_tlist); * *****************************************************************************/ Plan * -planner(Query *parse) +planner(Query *parse, bool isCursor, int cursorOptions, + ParamListInfo boundParams) { + double tuple_fraction; Plan *result_plan; Index save_PlannerQueryLevel; - List *save_PlannerParamVar; + List *save_PlannerParamList; + ParamListInfo save_PlannerBoundParamList; /* * The planner can be called recursively (an example is when * eval_const_expressions tries to pre-evaluate an SQL function). So, * these global state variables must be saved and restored. * - * These vars cannot be moved into the Query structure since their whole - * purpose is communication across multiple sub-Queries. + * Query level and the param list cannot be moved into the Query + * 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 @@ -92,26 +104,56 @@ planner(Query *parse) * subquery_planner, not here. */ save_PlannerQueryLevel = PlannerQueryLevel; - save_PlannerParamVar = PlannerParamVar; + save_PlannerParamList = PlannerParamList; + save_PlannerBoundParamList = PlannerBoundParamList; /* Initialize state for handling outer-level references and params */ PlannerQueryLevel = 0; /* will be 1 in top-level subquery_planner */ - PlannerParamVar = NIL; + PlannerParamList = NIL; + PlannerBoundParamList = boundParams; + + /* Determine what fraction of the plan is likely to be scanned */ + 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?) + */ + tuple_fraction = 0.10; + } + else + { + /* Default assumption is we need all the tuples */ + tuple_fraction = 0.0; + } /* primary planning entry point (may recurse for subqueries) */ - result_plan = subquery_planner(parse, -1.0 /* default case */ ); + result_plan = subquery_planner(parse, tuple_fraction); Assert(PlannerQueryLevel == 0); + /* + * If creating a plan for a scrollable cursor, make sure it can run + * backwards on demand. Add a Material node at the top at need. + */ + if (isCursor && (cursorOptions & CURSOR_OPT_SCROLL)) + { + if (!ExecSupportsBackwardScan(result_plan)) + result_plan = materialize_finished_plan(result_plan); + } + /* executor wants to know total number of Params used overall */ - result_plan->nParamExec = length(PlannerParamVar); + result_plan->nParamExec = list_length(PlannerParamList); /* final cleanup of the plan */ set_plan_references(result_plan, parse->rtable); /* restore state for outer planner, if any */ PlannerQueryLevel = save_PlannerQueryLevel; - PlannerParamVar = save_PlannerParamVar; + PlannerParamList = save_PlannerParamList; + PlannerBoundParamList = save_PlannerBoundParamList; return result_plan; } @@ -143,44 +185,66 @@ subquery_planner(Query *parse, double tuple_fraction) { List *saved_initplan = PlannerInitPlan; int saved_planid = PlannerPlanId; + bool hasOuterJoins; Plan *plan; List *newHaving; List *lst; + ListCell *l; /* Set up for a new level of subquery */ PlannerQueryLevel++; PlannerInitPlan = NIL; /* - * Check to see if any subqueries in the rangetable can be merged into - * this query. + * 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. */ - parse->jointree = (FromExpr *) - pull_up_subqueries(parse, (Node *) parse->jointree, false); + parse->in_info_list = NIL; + if (parse->hasSubLinks) + parse->jointree->quals = pull_up_IN_clauses(parse, + parse->jointree->quals); /* - * If so, we may have created opportunities to simplify the jointree. + * Check to see if any subqueries in the rangetable can be merged into + * this query. */ parse->jointree = (FromExpr *) - preprocess_jointree(parse, (Node *) parse->jointree); + pull_up_subqueries(parse, (Node *) parse->jointree, false); /* - * Detect whether any rangetable entries are RTE_JOIN kind; if not, - * we can avoid the expense of doing flatten_join_alias_vars(). - * This must be done after we have done pull_up_subqueries, of course. + * 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 + * pull_up_subqueries, of course. */ parse->hasJoinRTEs = false; - foreach(lst, parse->rtable) + hasOuterJoins = false; + foreach(l, parse->rtable) { - RangeTblEntry *rte = (RangeTblEntry *) lfirst(lst); + RangeTblEntry *rte = (RangeTblEntry *) lfirst(l); if (rte->rtekind == RTE_JOIN) { parse->hasJoinRTEs = true; - break; + if (IS_OUTER_JOIN(rte->jointype)) + { + hasOuterJoins = true; + /* Can quit scanning once we find an outer join */ + break; + } } } + /* + * 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. + */ + parse->hasHavingQual = (parse->havingQual != NULL); + /* * Do expression preprocessing on targetlist and quals. */ @@ -191,95 +255,118 @@ subquery_planner(Query *parse, double tuple_fraction) preprocess_qual_conditions(parse, (Node *) parse->jointree); parse->havingQual = preprocess_expression(parse, parse->havingQual, - EXPRKIND_HAVING); + EXPRKIND_QUAL); + + parse->limitOffset = preprocess_expression(parse, parse->limitOffset, + EXPRKIND_LIMIT); + parse->limitCount = preprocess_expression(parse, parse->limitCount, + EXPRKIND_LIMIT); + + parse->in_info_list = (List *) + preprocess_expression(parse, (Node *) parse->in_info_list, + EXPRKIND_ININFO); /* Also need to preprocess expressions for function RTEs */ - foreach(lst, parse->rtable) + foreach(l, parse->rtable) { - RangeTblEntry *rte = (RangeTblEntry *) lfirst(lst); + RangeTblEntry *rte = (RangeTblEntry *) lfirst(l); if (rte->rtekind == RTE_FUNCTION) rte->funcexpr = preprocess_expression(parse, rte->funcexpr, - EXPRKIND_TARGET); - /* These are not targetlist items, but close enough... */ + EXPRKIND_RTFUNC); } /* - * Check for ungrouped variables passed to subplans in targetlist and - * HAVING clause (but not in WHERE or JOIN/ON clauses, since those are - * evaluated before grouping). We can't do this any earlier because - * we must use the preprocessed targetlist for comparisons of grouped - * expressions. - */ - if (parse->hasSubLinks && - (parse->groupClause != NIL || parse->hasAggs)) - check_subplans_for_ungrouped_vars(parse); - - /* - * A HAVING clause without aggregates is equivalent to a WHERE clause - * (except it can only refer to grouped fields). Transfer any - * agg-free clauses of the HAVING qual into WHERE. This may seem like - * wasting cycles to cater to stupidly-written queries, but there are - * other reasons for doing it. Firstly, if the query contains no aggs - * at all, then we aren't going to generate an Agg plan node, and so - * there'll be no place to execute HAVING conditions; without this - * transfer, we'd lose the HAVING condition entirely, which is wrong. - * Secondly, when we push down a qual condition into a sub-query, it's - * easiest to push the qual into HAVING always, in case it contains - * aggs, and then let this code sort it out. + * 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 + * 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.) * * Note that both havingQual and parse->jointree->quals are in * implicitly-ANDed-list form at this point, even though they are - * declared as Node *. Also note that contain_agg_clause does not - * recurse into sub-selects, which is exactly what we need here. + * declared as Node *. */ newHaving = NIL; - foreach(lst, (List *) parse->havingQual) + foreach(l, (List *) parse->havingQual) { - Node *havingclause = (Node *) lfirst(lst); + Node *havingclause = (Node *) lfirst(l); - if (contain_agg_clause(havingclause)) + if (contain_agg_clause(havingclause) || + contain_volatile_functions(havingclause) || + contain_subplans(havingclause)) + { + /* keep it in HAVING */ newHaving = lappend(newHaving, havingclause); - else + } + else if (parse->groupClause) + { + /* move it to WHERE */ parse->jointree->quals = (Node *) lappend((List *) parse->jointree->quals, havingclause); + } + else + { + /* put a copy in WHERE, keep it in HAVING */ + parse->jointree->quals = (Node *) + lappend((List *) parse->jointree->quals, + copyObject(havingclause)); + newHaving = lappend(newHaving, havingclause); + } } 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 + * preprocessing. + */ + if (hasOuterJoins) + reduce_outer_joins(parse); + + /* + * 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(parse, (Node *) parse->jointree); + /* * 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_inherted_rtentry(parse, parse->resultRelation, false)) - != NIL) + (lst = expand_inherited_rtentry(parse, parse->resultRelation)) != NIL) plan = inheritance_planner(parse, lst); else plan = grouping_planner(parse, tuple_fraction); /* * If any subplans were generated, or if we're inside a subplan, build - * subPlan, extParam and locParam lists for plan nodes. + * initPlan list and extParam/allParam sets for plan nodes, and attach + * the initPlans to the top plan node. */ if (PlannerPlanId != saved_planid || PlannerQueryLevel > 1) - { - (void) SS_finalize_plan(plan, parse->rtable); - - /* - * At the moment, SS_finalize_plan doesn't handle initPlans and so - * we assign them to the topmost plan node. - */ - plan->initPlan = PlannerInitPlan; - /* Must add the initPlans' extParams to the topmost node's, too */ - foreach(lst, plan->initPlan) - { - SubPlan *subplan = (SubPlan *) lfirst(lst); - - plan->extParam = set_unioni(plan->extParam, - subplan->plan->extParam); - } - } + SS_finalize_plan(plan, parse->rtable); /* Return to outer subquery context */ PlannerQueryLevel--; @@ -289,423 +376,6 @@ subquery_planner(Query *parse, double tuple_fraction) return plan; } -/* - * pull_up_subqueries - * Look for subqueries in the rangetable that can be pulled up into - * the parent query. If the subquery has no special features like - * grouping/aggregation then we can merge it into the parent's jointree. - * - * below_outer_join is true if this jointree node is within the nullable - * side of an outer join. This restricts what we can do. - * - * A tricky aspect of this code is that if we pull up a subquery we have - * to replace Vars that reference the subquery's outputs throughout the - * parent query, including quals attached to jointree nodes above the one - * we are currently processing! We handle this by being careful not to - * change the jointree structure while recursing: no nodes other than - * subquery RangeTblRef entries will be replaced. Also, we can't turn - * ResolveNew loose on the whole jointree, because it'll return a mutated - * copy of the tree; we have to invoke it just on the quals, instead. - */ -static Node * -pull_up_subqueries(Query *parse, Node *jtnode, bool below_outer_join) -{ - if (jtnode == NULL) - return NULL; - if (IsA(jtnode, RangeTblRef)) - { - int varno = ((RangeTblRef *) jtnode)->rtindex; - RangeTblEntry *rte = rt_fetch(varno, parse->rtable); - Query *subquery = rte->subquery; - - /* - * Is this a subquery RTE, and if so, is the subquery simple - * enough to pull up? (If not, do nothing at this node.) - * - * If we are inside an outer join, only pull up subqueries whose - * targetlists are nullable --- otherwise substituting their tlist - * entries for upper Var references would do the wrong thing (the - * results wouldn't become NULL when they're supposed to). XXX - * This could be improved by generating pseudo-variables for such - * expressions; we'd have to figure out how to get the pseudo- - * variables evaluated at the right place in the modified plan - * tree. Fix it someday. - * - * Note: even if the subquery itself is simple enough, we can't pull - * it up if there is a reference to its whole tuple result. - * Perhaps a pseudo-variable is the answer here too. - */ - if (rte->rtekind == RTE_SUBQUERY && is_simple_subquery(subquery) && - (!below_outer_join || has_nullable_targetlist(subquery)) && - !contain_whole_tuple_var((Node *) parse, varno, 0)) - { - int rtoffset; - List *subtlist; - List *rt; - - /* - * First, recursively pull up the subquery's subqueries, so - * that this routine's processing is complete for its jointree - * and rangetable. NB: if the same subquery is referenced - * from multiple jointree items (which can't happen normally, - * but might after rule rewriting), then we will invoke this - * processing multiple times on that subquery. OK because - * nothing will happen after the first time. We do have to be - * careful to copy everything we pull up, however, or risk - * having chunks of structure multiply linked. - */ - subquery->jointree = (FromExpr *) - pull_up_subqueries(subquery, (Node *) subquery->jointree, - below_outer_join); - - /* - * Now make a modifiable copy of the subquery that we can run - * OffsetVarNodes and IncrementVarSublevelsUp on. - */ - subquery = copyObject(subquery); - - /* - * Adjust level-0 varnos in subquery so that we can append its - * rangetable to upper query's. - */ - rtoffset = length(parse->rtable); - OffsetVarNodes((Node *) subquery, rtoffset, 0); - - /* - * Upper-level vars in subquery are now one level closer to their - * parent than before. - */ - IncrementVarSublevelsUp((Node *) subquery, -1, 1); - - /* - * Replace all of the top query's references to the subquery's - * outputs with copies of the adjusted subtlist items, being - * careful not to replace any of the jointree structure. - * (This'd be a lot cleaner if we could use - * query_tree_mutator.) - */ - subtlist = subquery->targetList; - parse->targetList = (List *) - ResolveNew((Node *) parse->targetList, - varno, 0, subtlist, CMD_SELECT, 0); - resolvenew_in_jointree((Node *) parse->jointree, varno, subtlist); - Assert(parse->setOperations == NULL); - parse->havingQual = - ResolveNew(parse->havingQual, - varno, 0, subtlist, CMD_SELECT, 0); - - foreach(rt, parse->rtable) - { - RangeTblEntry *rte = (RangeTblEntry *) lfirst(rt); - - if (rte->rtekind == RTE_JOIN) - rte->joinaliasvars = (List *) - ResolveNew((Node *) rte->joinaliasvars, - varno, 0, subtlist, CMD_SELECT, 0); - } - - /* - * Now append the adjusted rtable entries to upper query. (We - * hold off until after fixing the upper rtable entries; no - * point in running that code on the subquery ones too.) - */ - parse->rtable = nconc(parse->rtable, subquery->rtable); - - /* - * Pull up any FOR UPDATE markers, too. (OffsetVarNodes - * already adjusted the marker values, so just nconc the - * list.) - */ - parse->rowMarks = nconc(parse->rowMarks, subquery->rowMarks); - - /* - * Miscellaneous housekeeping. - */ - parse->hasSubLinks |= subquery->hasSubLinks; - /* subquery won't be pulled up if it hasAggs, so no work there */ - - /* - * Return the adjusted subquery jointree to replace the - * RangeTblRef entry in my jointree. - */ - return (Node *) subquery->jointree; - } - } - else if (IsA(jtnode, FromExpr)) - { - FromExpr *f = (FromExpr *) jtnode; - List *l; - - foreach(l, f->fromlist) - lfirst(l) = pull_up_subqueries(parse, lfirst(l), - below_outer_join); - } - else if (IsA(jtnode, JoinExpr)) - { - JoinExpr *j = (JoinExpr *) jtnode; - - /* Recurse, being careful to tell myself when inside outer join */ - switch (j->jointype) - { - case JOIN_INNER: - j->larg = pull_up_subqueries(parse, j->larg, - below_outer_join); - j->rarg = pull_up_subqueries(parse, j->rarg, - below_outer_join); - break; - case JOIN_LEFT: - j->larg = pull_up_subqueries(parse, j->larg, - below_outer_join); - j->rarg = pull_up_subqueries(parse, j->rarg, - true); - break; - case JOIN_FULL: - j->larg = pull_up_subqueries(parse, j->larg, - true); - j->rarg = pull_up_subqueries(parse, j->rarg, - true); - break; - case JOIN_RIGHT: - j->larg = pull_up_subqueries(parse, j->larg, - true); - j->rarg = pull_up_subqueries(parse, j->rarg, - below_outer_join); - break; - case JOIN_UNION: - - /* - * This is where we fail if upper levels of planner - * haven't rewritten UNION JOIN as an Append ... - */ - elog(ERROR, "UNION JOIN is not implemented yet"); - break; - default: - elog(ERROR, "pull_up_subqueries: unexpected join type %d", - j->jointype); - break; - } - } - else - elog(ERROR, "pull_up_subqueries: unexpected node type %d", - nodeTag(jtnode)); - return jtnode; -} - -/* - * is_simple_subquery - * Check a subquery in the range table to see if it's simple enough - * to pull up into the parent query. - */ -static bool -is_simple_subquery(Query *subquery) -{ - /* - * Let's just make sure it's a valid subselect ... - */ - if (!IsA(subquery, Query) || - subquery->commandType != CMD_SELECT || - subquery->resultRelation != 0 || - subquery->into != NULL || - subquery->isPortal) - elog(ERROR, "is_simple_subquery: subquery is bogus"); - - /* - * Can't currently pull up a query with setops. Maybe after querytree - * redesign... - */ - if (subquery->setOperations) - return false; - - /* - * Can't pull up a subquery involving grouping, aggregation, sorting, - * or limiting. - */ - if (subquery->hasAggs || - subquery->groupClause || - subquery->havingQual || - subquery->sortClause || - subquery->distinctClause || - subquery->limitOffset || - subquery->limitCount) - return false; - - /* - * Don't pull up a subquery that has any set-returning functions in - * its targetlist. Otherwise we might well wind up inserting - * set-returning functions into places where they mustn't go, such as - * quals of higher queries. - */ - if (expression_returns_set((Node *) subquery->targetList)) - return false; - - /* - * Hack: don't try to pull up a subquery with an empty jointree. - * query_planner() will correctly generate a Result plan for a - * jointree that's totally empty, but I don't think the right things - * happen if an empty FromExpr appears lower down in a jointree. Not - * worth working hard on this, just to collapse SubqueryScan/Result - * into Result... - */ - if (subquery->jointree->fromlist == NIL) - return false; - - return true; -} - -/* - * has_nullable_targetlist - * Check a subquery in the range table to see if all the non-junk - * targetlist items are simple variables (and, hence, will correctly - * go to NULL when examined above the point of an outer join). - * - * A possible future extension is to accept strict functions of simple - * variables, eg, "x + 1". - */ -static bool -has_nullable_targetlist(Query *subquery) -{ - List *l; - - foreach(l, subquery->targetList) - { - TargetEntry *tle = (TargetEntry *) lfirst(l); - - /* ignore resjunk columns */ - if (tle->resdom->resjunk) - continue; - - /* Okay if tlist item is a simple Var */ - if (tle->expr && IsA(tle->expr, Var)) - continue; - - return false; - } - return true; -} - -/* - * Helper routine for pull_up_subqueries: do ResolveNew on every expression - * in the jointree, without changing the jointree structure itself. Ugly, - * but there's no other way... - */ -static void -resolvenew_in_jointree(Node *jtnode, int varno, List *subtlist) -{ - if (jtnode == NULL) - return; - if (IsA(jtnode, RangeTblRef)) - { - /* nothing to do here */ - } - else if (IsA(jtnode, FromExpr)) - { - FromExpr *f = (FromExpr *) jtnode; - List *l; - - foreach(l, f->fromlist) - resolvenew_in_jointree(lfirst(l), varno, subtlist); - f->quals = ResolveNew(f->quals, - varno, 0, subtlist, CMD_SELECT, 0); - } - else if (IsA(jtnode, JoinExpr)) - { - JoinExpr *j = (JoinExpr *) jtnode; - - resolvenew_in_jointree(j->larg, varno, subtlist); - resolvenew_in_jointree(j->rarg, varno, subtlist); - j->quals = ResolveNew(j->quals, - varno, 0, subtlist, CMD_SELECT, 0); - - /* - * We don't bother to update the colvars list, since it won't be - * used again ... - */ - } - else - elog(ERROR, "resolvenew_in_jointree: unexpected node type %d", - nodeTag(jtnode)); -} - -/* - * preprocess_jointree - * Attempt to simplify a query's jointree. - * - * If we succeed in pulling up a subquery then we might form a jointree - * in which a FromExpr is a direct child of another FromExpr. In that - * case we can consider collapsing the two FromExprs into one. This is - * an optional conversion, since the planner will work correctly either - * way. But we may find a better plan (at the cost of more planning time) - * if we merge the two nodes. - * - * NOTE: don't try to do this in the same jointree scan that does subquery - * pullup! Since we're changing the jointree structure here, that wouldn't - * work reliably --- see comments for pull_up_subqueries(). - */ -static Node * -preprocess_jointree(Query *parse, Node *jtnode) -{ - if (jtnode == NULL) - return NULL; - if (IsA(jtnode, RangeTblRef)) - { - /* nothing to do here... */ - } - else if (IsA(jtnode, FromExpr)) - { - FromExpr *f = (FromExpr *) jtnode; - List *newlist = NIL; - List *l; - - foreach(l, f->fromlist) - { - Node *child = (Node *) lfirst(l); - - /* Recursively simplify the child... */ - child = preprocess_jointree(parse, child); - /* Now, is it a FromExpr? */ - if (child && IsA(child, FromExpr)) - { - /* - * Yes, so do we want to merge it into parent? Always do - * so if child has just one element (since that doesn't - * make the parent's list any longer). Otherwise we have - * to be careful about the increase in planning time - * caused by combining the two join search spaces into - * one. Our heuristic is to merge if the merge will - * produce a join list no longer than GEQO_RELS/2. - * (Perhaps need an additional user parameter?) - */ - FromExpr *subf = (FromExpr *) child; - int childlen = length(subf->fromlist); - int myothers = length(newlist) + length(lnext(l)); - - if (childlen <= 1 || (childlen + myothers) <= geqo_rels / 2) - { - newlist = nconc(newlist, subf->fromlist); - f->quals = make_and_qual(subf->quals, f->quals); - } - else - newlist = lappend(newlist, child); - } - else - newlist = lappend(newlist, child); - } - f->fromlist = newlist; - } - else if (IsA(jtnode, JoinExpr)) - { - JoinExpr *j = (JoinExpr *) jtnode; - - /* Can't usefully change the JoinExpr, but recurse on children */ - j->larg = preprocess_jointree(parse, j->larg); - j->rarg = preprocess_jointree(parse, j->rarg); - } - else - elog(ERROR, "preprocess_jointree: unexpected node type %d", - nodeTag(jtnode)); - return jtnode; -} - /* * preprocess_expression * Do subquery_planner's preprocessing work for an expression, @@ -715,24 +385,31 @@ preprocess_jointree(Query *parse, Node *jtnode) static Node * preprocess_expression(Query *parse, Node *expr, int kind) { + /* + * 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. + */ + if (parse->hasJoinRTEs) + expr = flatten_join_alias_vars(parse, expr); + /* * Simplify constant expressions. * - * Note that at this point quals have not yet been converted to - * implicit-AND form, so we can apply eval_const_expressions directly. + * Note: this also flattens nested AND and OR expressions into N-argument + * form. All processing of a qual expression after this point must be + * careful to maintain AND/OR flatness --- that is, do not generate a tree + * with AND directly under AND, nor OR directly under OR. */ expr = eval_const_expressions(expr); /* - * If it's a qual or havingQual, canonicalize it, and convert it to - * implicit-AND format. - * - * XXX Is there any value in re-applying eval_const_expressions after - * canonicalize_qual? + * If it's a qual or havingQual, canonicalize it. */ - if (kind != EXPRKIND_TARGET) + if (kind == EXPRKIND_QUAL) { - expr = (Node *) canonicalize_qual((Expr *) expr, true); + expr = (Node *) canonicalize_qual((Expr *) expr); #ifdef OPTIMIZER_DEBUG printf("After canonicalize_qual()\n"); @@ -742,20 +419,25 @@ preprocess_expression(Query *parse, Node *expr, int kind) /* Expand SubLinks to SubPlans */ if (parse->hasSubLinks) - expr = SS_process_sublinks(expr); + 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 ... + */ /* Replace uplevel vars with Param nodes */ if (PlannerQueryLevel > 1) expr = SS_replace_correlation_vars(expr); /* - * If the query has any join RTEs, try to replace join alias variables - * with base-relation variables, to allow quals to be pushed down. We - * must do this after sublink processing, since it does not recurse - * into sublinks. + * 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 (parse->hasJoinRTEs) - expr = flatten_join_alias_vars(expr, parse->rtable, false); + if (kind == EXPRKIND_QUAL) + expr = (Node *) make_ands_implicit((Expr *) expr); return expr; } @@ -777,12 +459,12 @@ preprocess_qual_conditions(Query *parse, Node *jtnode) else if (IsA(jtnode, FromExpr)) { FromExpr *f = (FromExpr *) jtnode; - List *l; + ListCell *l; foreach(l, f->fromlist) preprocess_qual_conditions(parse, lfirst(l)); - f->quals = preprocess_expression(parse, f->quals, EXPRKIND_WHERE); + f->quals = preprocess_expression(parse, f->quals, EXPRKIND_QUAL); } else if (IsA(jtnode, JoinExpr)) { @@ -791,11 +473,11 @@ preprocess_qual_conditions(Query *parse, Node *jtnode) preprocess_qual_conditions(parse, j->larg); preprocess_qual_conditions(parse, j->rarg); - j->quals = preprocess_expression(parse, j->quals, EXPRKIND_WHERE); + j->quals = preprocess_expression(parse, j->quals, EXPRKIND_QUAL); } else - elog(ERROR, "preprocess_qual_conditions: unexpected node type %d", - nodeTag(jtnode)); + elog(ERROR, "unrecognized node type: %d", + (int) nodeTag(jtnode)); } /*-------------------- @@ -824,13 +506,14 @@ inheritance_planner(Query *parse, List *inheritlist) { int parentRTindex = parse->resultRelation; Oid parentOID = getrelid(parentRTindex, parse->rtable); + int mainrtlength = list_length(parse->rtable); List *subplans = NIL; List *tlist = NIL; - List *l; + ListCell *l; foreach(l, inheritlist) { - int childRTindex = lfirsti(l); + int childRTindex = lfirst_int(l); Oid childOID = getrelid(childRTindex, parse->rtable); Query *subquery; Plan *subplan; @@ -842,6 +525,42 @@ inheritance_planner(Query *parse, List *inheritlist) /* Generate plan */ subplan = grouping_planner(subquery, 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 = subquery->rtable; + else + { + int subrtlength = list_length(subquery->rtable); + + if (subrtlength > mainrtlength) + { + List *subrt; + + subrt = list_copy_tail(subquery->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; @@ -850,6 +569,9 @@ inheritance_planner(Query *parse, List *inheritlist) /* Save the target-relations list for the executor, too */ parse->resultRelations = inheritlist; + /* Mark result as unordered (probably unnecessary) */ + parse->query_pathkeys = NIL; + return (Plan *) make_append(subplans, true, tlist); } @@ -863,16 +585,14 @@ inheritance_planner(Query *parse, List *inheritlist) * tuple_fraction is the fraction of tuples we expect will be retrieved * * tuple_fraction is interpreted as follows: - * < 0: determine fraction by inspection of query (normal case) - * 0: expect all tuples to be retrieved + * 0: expect all tuples to be retrieved (normal case) * 0 < tuple_fraction < 1: expect the given fraction of tuples available * from the plan to be retrieved * tuple_fraction >= 1: tuple_fraction is the absolute number of tuples * expected to be retrieved (ie, a LIMIT specification) - * The normal case is to pass -1, but some callers pass values >= 0 to - * override this routine's determination of the appropriate fraction. * - * Returns a query plan. + * Returns a query plan. Also, parse->query_pathkeys is returned as the + * actual output ordering of the plan (in pathkey format). *-------------------- */ static Plan * @@ -885,11 +605,23 @@ grouping_planner(Query *parse, double tuple_fraction) 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. */ - result_plan = plan_set_operations(parse); + result_plan = plan_set_operations(parse, + &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... + */ + current_pathkeys = make_pathkeys_for_sortclauses(set_sortclauses, + result_plan->targetlist); + current_pathkeys = canonicalize_pathkeys(parse, current_pathkeys); /* * We should not need to call preprocess_targetlist, since we must @@ -907,19 +639,12 @@ grouping_planner(Query *parse, double tuple_fraction) * already, but let's make sure). */ if (parse->rowMarks) - elog(ERROR, "SELECT FOR UPDATE is not allowed with UNION/INTERSECT/EXCEPT"); - - /* - * We set current_pathkeys NIL indicating we do not know sort - * order. This is correct when the top set operation is UNION - * ALL, since the appended-together results are unsorted even if - * the subplans were sorted. For other set operations we could be - * smarter --- room for future improvement! - */ - current_pathkeys = NIL; + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("SELECT FOR UPDATE is not allowed with UNION/INTERSECT/EXCEPT"))); /* - * Calculate pathkeys that represent ordering requirements + * Calculate pathkeys that represent result ordering requirements */ sort_pathkeys = make_pathkeys_for_sortclauses(parse->sortClause, tlist); @@ -931,78 +656,29 @@ grouping_planner(Query *parse, double tuple_fraction) List *sub_tlist; List *group_pathkeys; 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; - int numAggs = 0; - int numGroupCols = length(parse->groupClause); + AggClauseCounts agg_counts; + int numGroupCols = list_length(parse->groupClause); bool use_hashed_grouping = false; - /* Preprocess targetlist in case we are inside an INSERT/UPDATE. */ - tlist = preprocess_targetlist(tlist, - parse->commandType, - parse->resultRelation, - parse->rtable); + MemSet(&agg_counts, 0, sizeof(AggClauseCounts)); - /* - * Add TID targets for rels selected FOR UPDATE (should this be - * done in preprocess_targetlist?). The executor uses the TID to - * know which rows to lock, much as for UPDATE or DELETE. - */ - if (parse->rowMarks) - { - List *l; - - /* - * We've got trouble if the FOR UPDATE appears inside - * grouping, since grouping renders a reference to individual - * tuple CTIDs invalid. This is also checked at parse time, - * but that's insufficient because of rule substitution, query - * pullup, etc. - */ - CheckSelectForUpdate(parse); - - /* - * Currently the executor only supports FOR UPDATE at top - * level - */ - if (PlannerQueryLevel > 1) - elog(ERROR, "SELECT FOR UPDATE is not allowed in subselects"); - - foreach(l, parse->rowMarks) - { - Index rti = lfirsti(l); - char *resname; - Resdom *resdom; - Var *var; - TargetEntry *ctid; - - resname = (char *) palloc(32); - snprintf(resname, 32, "ctid%u", rti); - resdom = makeResdom(length(tlist) + 1, - TIDOID, - -1, - resname, - true); - - var = makeVar(rti, - SelfItemPointerAttributeNumber, - TIDOID, - -1, - 0); - - ctid = makeTargetEntry(resdom, (Node *) var); - tlist = lappend(tlist, ctid); - } - } + /* Preprocess targetlist */ + tlist = preprocess_targetlist(parse, tlist); /* * Generate appropriate target list for subplan; may be different * from tlist if grouping or aggregation is needed. */ - sub_tlist = make_subplanTargetList(parse, tlist, &groupColIdx); + sub_tlist = make_subplanTargetList(parse, tlist, + &groupColIdx, &need_tlist_eval); /* * Calculate pathkeys that represent grouping/ordering @@ -1015,15 +691,19 @@ grouping_planner(Query *parse, double tuple_fraction) /* * Will need actual number of aggregates for estimating costs. - * Also, it's possible that optimization has eliminated all - * aggregates, and we may as well check for that here. + * + * 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). */ if (parse->hasAggs) { - numAggs = length(pull_agg_clause((Node *) tlist)) + - length(pull_agg_clause(parse->havingQual)); - if (numAggs == 0) - parse->hasAggs = false; + count_agg_clauses((Node *) tlist, &agg_counts); + count_agg_clauses(parse->havingQual, &agg_counts); } /* @@ -1044,29 +724,6 @@ grouping_planner(Query *parse, double tuple_fraction) else parse->query_pathkeys = NIL; - /* - * Figure out whether we expect to retrieve all the tuples that - * the plan can generate, or to stop early due to outside factors - * such as a cursor. If the caller passed a value >= 0, believe - * that value, else do our own examination of the query context. - */ - if (tuple_fraction < 0.0) - { - /* Initial assumption is we need all the tuples */ - tuple_fraction = 0.0; - - /* - * Check for retrieve-into-portal, ie DECLARE CURSOR. - * - * 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?) - */ - if (parse->isPortal) - tuple_fraction = 0.10; - } - /* * Adjust tuple_fraction if we see that we are going to apply * limiting/grouping/aggregation/etc. This is not overridable by @@ -1230,286 +887,255 @@ grouping_planner(Query *parse, double tuple_fraction) &cheapest_path, &sorted_path); /* - * We couldn't canonicalize group_pathkeys and sort_pathkeys before - * running query_planner(), so do it now. + * We couldn't canonicalize group_pathkeys and sort_pathkeys + * before running query_planner(), so do it now. */ group_pathkeys = canonicalize_pathkeys(parse, group_pathkeys); sort_pathkeys = canonicalize_pathkeys(parse, sort_pathkeys); /* - * Consider whether we might want to use hashed grouping. + * 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) { - /* - * Always estimate the number of groups. We can't do this until - * after running query_planner(), either. - */ - dNumGroups = estimate_num_groups(parse, - parse->groupClause, - cheapest_path->parent->rows); - /* Also want it as a long int --- but 'ware overflow! */ - numGroups = (long) Min(dNumGroups, (double) LONG_MAX); + List *groupExprs; + double cheapest_path_rows; /* - * Executor doesn't support hashed aggregation with DISTINCT - * aggregates. (Doing so would imply storing *all* the input - * values in the hash table, which seems like a certain loser.) + * 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 (!enable_hashagg) - use_hashed_grouping = false; - else if (parse->hasAggs && - (contain_distinct_agg_clause((Node *) tlist) || - contain_distinct_agg_clause(parse->havingQual))) - use_hashed_grouping = false; + if (cheapest_path->parent) + cheapest_path_rows = cheapest_path->parent->rows; else - { - /* - * Use hashed grouping if (a) we think we can fit the - * hashtable into SortMem, *and* (b) 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. - * - * In most cases we have no good way to estimate the size of - * the transition value needed by an aggregate; arbitrarily - * assume it is 100 bytes. Also set the overhead per hashtable - * entry at 64 bytes. - */ - int hashentrysize = cheapest_path->parent->width + 64 + - numAggs * 100; - - if (hashentrysize * dNumGroups <= SortMem * 1024L) - { - /* - * Okay, do the 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. - * - * These path variables are dummies that just hold cost - * fields; we don't make actual Paths for these steps. - */ - Path hashed_p; - Path sorted_p; - - cost_agg(&hashed_p, parse, - AGG_HASHED, numAggs, - numGroupCols, dNumGroups, - cheapest_path->startup_cost, - cheapest_path->total_cost, - cheapest_path->parent->rows); - /* Result of hashed agg is always unsorted */ - if (sort_pathkeys) - cost_sort(&hashed_p, parse, sort_pathkeys, - hashed_p.total_cost, - dNumGroups, - cheapest_path->parent->width); - - if (sorted_path) - { - sorted_p.startup_cost = sorted_path->startup_cost; - sorted_p.total_cost = sorted_path->total_cost; - current_pathkeys = sorted_path->pathkeys; - } - else - { - sorted_p.startup_cost = cheapest_path->startup_cost; - sorted_p.total_cost = cheapest_path->total_cost; - current_pathkeys = cheapest_path->pathkeys; - } - if (!pathkeys_contained_in(group_pathkeys, - current_pathkeys)) - { - cost_sort(&sorted_p, parse, group_pathkeys, - sorted_p.total_cost, - cheapest_path->parent->rows, - cheapest_path->parent->width); - current_pathkeys = group_pathkeys; - } - if (parse->hasAggs) - cost_agg(&sorted_p, parse, - AGG_SORTED, numAggs, - numGroupCols, dNumGroups, - sorted_p.startup_cost, - sorted_p.total_cost, - cheapest_path->parent->rows); - else - cost_group(&sorted_p, parse, - numGroupCols, dNumGroups, - sorted_p.startup_cost, - sorted_p.total_cost, - cheapest_path->parent->rows); - /* The Agg or Group node will preserve ordering */ - if (sort_pathkeys && - !pathkeys_contained_in(sort_pathkeys, - current_pathkeys)) - { - cost_sort(&sorted_p, parse, sort_pathkeys, - sorted_p.total_cost, - dNumGroups, - cheapest_path->parent->width); - } + cheapest_path_rows = 1; /* assume non-set result */ - /* - * Now make the decision using the top-level tuple - * fraction. First we have to convert an absolute - * count (LIMIT) into fractional form. - */ - if (tuple_fraction >= 1.0) - tuple_fraction /= dNumGroups; + groupExprs = get_sortgrouplist_exprs(parse->groupClause, + parse->targetList); + dNumGroups = estimate_num_groups(parse, + groupExprs, + cheapest_path_rows); + /* Also want it as a long int --- but 'ware overflow! */ + numGroups = (long) Min(dNumGroups, (double) LONG_MAX); - if (compare_fractional_path_costs(&hashed_p, &sorted_p, - tuple_fraction) <= 0) - { - /* Hashed is cheaper, so use it */ - use_hashed_grouping = true; - } - } - } + use_hashed_grouping = + choose_hashed_grouping(parse, tuple_fraction, + cheapest_path, sorted_path, + sort_pathkeys, group_pathkeys, + dNumGroups, &agg_counts); } /* - * Select the best path and create a plan to execute it. - * - * 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. + * 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. */ - if (sorted_path && !use_hashed_grouping) - { - result_plan = create_plan(parse, sorted_path); - current_pathkeys = sorted_path->pathkeys; - } + if (use_hashed_grouping || !sorted_path) + best_path = cheapest_path; else - { - result_plan = create_plan(parse, cheapest_path); - current_pathkeys = cheapest_path->pathkeys; - } + best_path = sorted_path; /* - * create_plan() returns a plan with just a "flat" tlist of required - * Vars. We want to insert the sub_tlist as the tlist of the top - * plan node. If the top-level plan node is one that cannot do - * expression evaluation, we must insert a Result node to project the - * desired tlist. - * Currently, the only plan node we might see here that falls into - * that category is Append. + * 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. */ - if (IsA(result_plan, Append)) - { - result_plan = (Plan *) make_result(sub_tlist, NULL, result_plan); - } - else + result_plan = optimize_minmax_aggregates(parse, + tlist, + best_path); + if (result_plan != NULL) { /* - * Otherwise, just replace the flat tlist with the desired tlist. + * optimize_minmax_aggregates generated the full plan, with + * the right tlist, and it has no sort order. */ - result_plan->targetlist = sub_tlist; - } - - /* - * Insert AGG or GROUP node if needed, plus an explicit sort step - * if necessary. - * - * HAVING clause, if any, becomes qual of the Agg node - */ - if (use_hashed_grouping) - { - /* Hashed aggregate plan --- no sort needed */ - result_plan = (Plan *) make_agg(parse, - tlist, - (List *) parse->havingQual, - AGG_HASHED, - numGroupCols, - groupColIdx, - numGroups, - numAggs, - result_plan); - /* Hashed aggregation produces randomly-ordered results */ current_pathkeys = NIL; } - else if (parse->hasAggs) + else { - /* Plain aggregate plan --- sort if needed */ - AggStrategy aggstrategy; + /* + * Normal case --- create a plan according to query_planner's + * results. + */ + result_plan = create_plan(parse, best_path); + current_pathkeys = best_path->pathkeys; - if (parse->groupClause) + /* + * create_plan() returns a plan with just a "flat" tlist of + * required Vars. Usually we need to insert the sub_tlist as the + * tlist of the top plan node. However, we can skip that if we + * determined that whatever query_planner chose to return will be + * good enough. + */ + if (need_tlist_eval) { - if (!pathkeys_contained_in(group_pathkeys, current_pathkeys)) + /* + * If the top-level plan node is one that cannot do expression + * evaluation, we must insert a Result node to project the + * desired tlist. + */ + if (!is_projection_capable_plan(result_plan)) { - result_plan = make_groupsortplan(parse, - parse->groupClause, - groupColIdx, - result_plan); - current_pathkeys = group_pathkeys; + result_plan = (Plan *) make_result(sub_tlist, NULL, + result_plan); + } + else + { + /* + * Otherwise, just replace the subplan's flat tlist with + * the desired tlist. + */ + result_plan->targetlist = sub_tlist; } - aggstrategy = AGG_SORTED; + /* - * The AGG node will not change the sort ordering of its - * groups, so current_pathkeys describes the result too. + * Also, account for the cost of evaluation of the sub_tlist. + * + * Up to now, we have only been dealing with "flat" tlists, + * containing just Vars. So their evaluation cost is zero + * according to the model used by cost_qual_eval() (or if you + * prefer, the cost is factored into cpu_tuple_cost). Thus we + * can avoid accounting for tlist cost throughout + * query_planner() and subroutines. But now we've inserted a + * tlist that might contain actual operators, sub-selects, etc + * --- so we'd better account for its cost. + * + * Below this point, any tlist eval cost for added-on nodes + * should be accounted for as we create those nodes. + * Presently, of the node types we can add on, only Agg and + * Group project new tlists (the rest just copy their input + * tuples) --- so make_agg() and make_group() are responsible + * for computing the added cost. */ + cost_qual_eval(&tlist_cost, sub_tlist); + result_plan->startup_cost += tlist_cost.startup; + result_plan->total_cost += tlist_cost.startup + + tlist_cost.per_tuple * result_plan->plan_rows; } else { - aggstrategy = AGG_PLAIN; - /* Result will be only one row anyway; no sort order */ - current_pathkeys = NIL; + /* + * Since we're using query_planner's tlist and not the one + * make_subplanTargetList calculated, we have to refigure any + * grouping-column indexes make_subplanTargetList computed. + */ + locate_grouping_columns(parse, tlist, result_plan->targetlist, + groupColIdx); } - result_plan = (Plan *) make_agg(parse, - tlist, - (List *) parse->havingQual, - aggstrategy, - numGroupCols, - groupColIdx, - numGroups, - numAggs, - result_plan); - } - else - { /* - * If there are no Aggs, we shouldn't have any HAVING qual anymore + * Insert AGG or GROUP node if needed, plus an explicit sort step + * if necessary. + * + * HAVING clause, if any, becomes qual of the Agg or Group node. */ - Assert(parse->havingQual == NULL); + if (use_hashed_grouping) + { + /* Hashed aggregate plan --- no sort needed */ + result_plan = (Plan *) make_agg(parse, + tlist, + (List *) parse->havingQual, + AGG_HASHED, + numGroupCols, + groupColIdx, + numGroups, + agg_counts.numAggs, + result_plan); + /* Hashed aggregation produces randomly-ordered results */ + current_pathkeys = NIL; + } + else if (parse->hasAggs) + { + /* Plain aggregate plan --- sort if needed */ + AggStrategy aggstrategy; - /* - * If we have a GROUP BY clause, insert a group node (plus the - * appropriate sort node, if necessary). - */ - if (parse->groupClause) + if (parse->groupClause) + { + if (!pathkeys_contained_in(group_pathkeys, + current_pathkeys)) + { + result_plan = (Plan *) + make_sort_from_groupcols(parse, + parse->groupClause, + groupColIdx, + result_plan); + current_pathkeys = group_pathkeys; + } + aggstrategy = AGG_SORTED; + + /* + * The AGG node will not change the sort ordering of its + * groups, so current_pathkeys describes the result too. + */ + } + else + { + aggstrategy = AGG_PLAIN; + /* Result will be only one row anyway; no sort order */ + current_pathkeys = NIL; + } + + result_plan = (Plan *) make_agg(parse, + tlist, + (List *) parse->havingQual, + aggstrategy, + numGroupCols, + groupColIdx, + numGroups, + agg_counts.numAggs, + result_plan); + } + else if (parse->groupClause) { /* - * Add an explicit sort if we couldn't make the path come out - * the way the GROUP node needs it. + * 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. */ if (!pathkeys_contained_in(group_pathkeys, current_pathkeys)) { - result_plan = make_groupsortplan(parse, - parse->groupClause, - groupColIdx, - result_plan); + result_plan = (Plan *) + make_sort_from_groupcols(parse, + parse->groupClause, + groupColIdx, + result_plan); current_pathkeys = group_pathkeys; } result_plan = (Plan *) make_group(parse, tlist, + (List *) parse->havingQual, numGroupCols, groupColIdx, dNumGroups, result_plan); /* The Group node won't change sort ordering */ } - } - } /* end of if (setOperations) */ + else if (parse->hasHavingQual) + { + /* + * No aggregates, and no GROUP BY, but we have a HAVING qual. + * 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. + */ + result_plan = (Plan *) make_result(tlist, + parse->havingQual, + NULL); + } + } /* end of non-minmax-aggregate case */ + } /* end of if (setOperations) */ /* * If we were not able to make the plan come out in the right order, @@ -1518,8 +1144,13 @@ grouping_planner(Query *parse, double tuple_fraction) if (parse->sortClause) { if (!pathkeys_contained_in(sort_pathkeys, current_pathkeys)) - result_plan = make_sortplan(parse, tlist, result_plan, - parse->sortClause); + { + result_plan = (Plan *) + make_sort_from_sortclauses(parse, + parse->sortClause, + result_plan); + current_pathkeys = sort_pathkeys; + } } /* @@ -1527,18 +1158,24 @@ grouping_planner(Query *parse, double tuple_fraction) */ if (parse->distinctClause) { - result_plan = (Plan *) make_unique(tlist, result_plan, - parse->distinctClause); + 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 (!parse->groupClause && !parse->hasAggs) + if (!parse->groupClause && !parse->hasHavingQual && !parse->hasAggs) + { + List *distinctExprs; + + distinctExprs = get_sortgrouplist_exprs(parse->distinctClause, + parse->targetList); result_plan->plan_rows = estimate_num_groups(parse, - parse->distinctClause, - result_plan->plan_rows); + distinctExprs, + result_plan->plan_rows); + } } /* @@ -1546,31 +1183,205 @@ grouping_planner(Query *parse, double tuple_fraction) */ if (parse->limitOffset || parse->limitCount) { - result_plan = (Plan *) make_limit(tlist, result_plan, + result_plan = (Plan *) make_limit(result_plan, parse->limitOffset, parse->limitCount); } + /* + * Return the actual output ordering in query_pathkeys for possible + * use by an outer query level. + */ + parse->query_pathkeys = current_pathkeys; + return result_plan; } +/* + * choose_hashed_grouping - should we use hashed grouping? + */ +static bool +choose_hashed_grouping(Query *parse, double tuple_fraction, + Path *cheapest_path, Path *sorted_path, + List *sort_pathkeys, List *group_pathkeys, + double dNumGroups, AggClauseCounts *agg_counts) +{ + int numGroupCols = list_length(parse->groupClause); + double cheapest_path_rows; + int cheapest_path_width; + Size hashentrysize; + List *current_pathkeys; + Path hashed_p; + Path sorted_p; + + /* + * Check can't-do-it conditions, including whether the grouping operators + * are hashjoinable. + * + * Executor doesn't support hashed aggregation with DISTINCT aggregates. + * (Doing so would imply storing *all* the input values in the hash table, + * which seems like a certain loser.) + */ + if (!enable_hashagg) + return false; + if (agg_counts->numDistinctAggs != 0) + return false; + if (!hash_safe_grouping(parse)) + return false; + + /* + * 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; + */ + if (cheapest_path->parent) + { + cheapest_path_rows = cheapest_path->parent->rows; + cheapest_path_width = cheapest_path->parent->width; + } + else + { + 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; + /* plus space for pass-by-ref transition values... */ + hashentrysize += agg_counts->transitionSpace; + /* plus the per-hash-entry overhead */ + hashentrysize += hash_agg_entry_size(agg_counts->numAggs); + + if (hashentrysize * dNumGroups > work_mem * 1024L) + 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. + * + * 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. + */ + cost_agg(&hashed_p, parse, AGG_HASHED, agg_counts->numAggs, + numGroupCols, dNumGroups, + 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, parse, sort_pathkeys, hashed_p.total_cost, + dNumGroups, cheapest_path_width); + + if (sorted_path) + { + sorted_p.startup_cost = sorted_path->startup_cost; + sorted_p.total_cost = sorted_path->total_cost; + current_pathkeys = sorted_path->pathkeys; + } + else + { + sorted_p.startup_cost = cheapest_path->startup_cost; + sorted_p.total_cost = cheapest_path->total_cost; + current_pathkeys = cheapest_path->pathkeys; + } + if (!pathkeys_contained_in(group_pathkeys, + current_pathkeys)) + { + cost_sort(&sorted_p, parse, group_pathkeys, sorted_p.total_cost, + cheapest_path_rows, cheapest_path_width); + current_pathkeys = group_pathkeys; + } + + if (parse->hasAggs) + cost_agg(&sorted_p, parse, AGG_SORTED, agg_counts->numAggs, + numGroupCols, dNumGroups, + sorted_p.startup_cost, sorted_p.total_cost, + cheapest_path_rows); + else + cost_group(&sorted_p, parse, numGroupCols, dNumGroups, + 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, parse, sort_pathkeys, sorted_p.total_cost, + dNumGroups, cheapest_path_width); + + /* + * Now make the decision using the top-level tuple fraction. First we + * have to convert an absolute count (LIMIT) into fractional form. + */ + if (tuple_fraction >= 1.0) + tuple_fraction /= dNumGroups; + + if (compare_fractional_path_costs(&hashed_p, &sorted_p, + tuple_fraction) < 0) + { + /* Hashed is cheaper, so use it */ + return true; + } + return false; +} + +/* + * hash_safe_grouping - are grouping operators hashable? + * + * We assume hashed aggregation will work if the datatype's equality operator + * is marked hashjoinable. + */ +static bool +hash_safe_grouping(Query *parse) +{ + ListCell *gl; + + foreach(gl, parse->groupClause) + { + GroupClause *grpcl = (GroupClause *) lfirst(gl); + TargetEntry *tle = get_sortgroupclause_tle(grpcl, parse->targetList); + Operator optup; + bool oprcanhash; + + optup = equality_oper(exprType((Node *) tle->expr), true); + if (!optup) + return false; + oprcanhash = ((Form_pg_operator) GETSTRUCT(optup))->oprcanhash; + ReleaseSysCache(optup); + if (!oprcanhash) + return false; + } + return true; +} + /*--------------- * make_subplanTargetList * Generate appropriate target list when grouping is required. * - * When grouping_planner inserts Aggregate or Group plan nodes above - * the result of query_planner, we typically want to pass a different + * When grouping_planner inserts Aggregate, Group, or Result plan nodes + * above the result of query_planner, we typically want to pass a different * target list to query_planner than the outer plan nodes should have. * This routine generates the correct target list for the subplan. * * The initial target list passed from the parser already contains entries * for all ORDER BY and GROUP BY expressions, but it will not have entries * for variables used only in HAVING clauses; so we need to add those - * variables to the subplan target list. Also, if we are doing either - * grouping or aggregation, we flatten all expressions except GROUP BY items - * into their component variables; the other expressions will be computed by - * the inserted nodes rather than by the subplan. For example, - * given a query like + * variables to the subplan target list. Also, we flatten all expressions + * except GROUP BY items into their component variables; the other expressions + * will be computed by the inserted nodes rather than by the subplan. + * For example, given a query like * SELECT a+b,SUM(c+d) FROM table GROUP BY a+b; * we want to pass this targetlist to the subplan: * a,b,c,d,a+b @@ -1582,10 +1393,17 @@ grouping_planner(Query *parse, double tuple_fraction) * the extra computation to recompute a+b at the outer level; see * replace_vars_with_subplan_refs() in setrefs.c.) * + * If we are grouping or aggregating, *and* there are no non-Var grouping + * expressions, then the returned tlist is effectively dummy; we do not + * need to force it to be evaluated, because all the Vars it contains + * should be present in the output of query_planner anyway. + * * 'parse' is the query being processed. * 'tlist' is the query's target list. * 'groupColIdx' receives an array of column numbers for the GROUP BY - * expressions (if there are any) in the subplan's target list. + * expressions (if there are any) in the subplan's target list. + * 'need_tlist_eval' is set true if we really need to evaluate the + * result tlist. * * The result is the targetlist to be passed to the subplan. *--------------- @@ -1593,7 +1411,8 @@ grouping_planner(Query *parse, double tuple_fraction) static List * make_subplanTargetList(Query *parse, List *tlist, - AttrNumber **groupColIdx) + AttrNumber **groupColIdx, + bool *need_tlist_eval) { List *sub_tlist; List *extravars; @@ -1602,11 +1421,14 @@ make_subplanTargetList(Query *parse, *groupColIdx = NULL; /* - * If we're not grouping or aggregating, nothing to do here; + * If we're not grouping or aggregating, there's nothing to do here; * query_planner should receive the unmodified target list. */ - if (!parse->hasAggs && !parse->groupClause && !parse->havingQual) + if (!parse->hasAggs && !parse->groupClause && !parse->hasHavingQual) + { + *need_tlist_eval = true; return tlist; + } /* * Otherwise, start with a "flattened" tlist (having just the vars @@ -1616,7 +1438,8 @@ make_subplanTargetList(Query *parse, sub_tlist = flatten_tlist(tlist); extravars = pull_var_clause(parse->havingQual, false); sub_tlist = add_to_flat_tlist(sub_tlist, extravars); - freeList(extravars); + list_free(extravars); + *need_tlist_eval = false; /* only eval if not flat tlist */ /* * If grouping, create sub_tlist entries for all GROUP BY expressions @@ -1624,12 +1447,12 @@ make_subplanTargetList(Query *parse, * already), and make an array showing where the group columns are in * the sub_tlist. */ - numCols = length(parse->groupClause); + numCols = list_length(parse->groupClause); if (numCols > 0) { int keyno = 0; AttrNumber *grpColIdx; - List *gl; + ListCell *gl; grpColIdx = (AttrNumber *) palloc(sizeof(AttrNumber) * numCols); *groupColIdx = grpColIdx; @@ -1639,7 +1462,7 @@ make_subplanTargetList(Query *parse, GroupClause *grpcl = (GroupClause *) lfirst(gl); Node *groupexpr = get_sortgroupclause_expr(grpcl, tlist); TargetEntry *te = NULL; - List *sl; + ListCell *sl; /* Find or make a matching sub_tlist entry */ foreach(sl, sub_tlist) @@ -1650,17 +1473,16 @@ make_subplanTargetList(Query *parse, } if (!sl) { - te = makeTargetEntry(makeResdom(length(sub_tlist) + 1, - exprType(groupexpr), - exprTypmod(groupexpr), - NULL, - false), - groupexpr); + te = makeTargetEntry((Expr *) groupexpr, + list_length(sub_tlist) + 1, + NULL, + false); sub_tlist = lappend(sub_tlist, te); + *need_tlist_eval = true; /* it's not flat anymore */ } /* and save its resno */ - grpColIdx[keyno++] = te->resdom->resno; + grpColIdx[keyno++] = te->resno; } } @@ -1668,89 +1490,50 @@ make_subplanTargetList(Query *parse, } /* - * make_groupsortplan - * Add a Sort node to explicitly sort according to the GROUP BY clause. + * locate_grouping_columns + * Locate grouping columns in the tlist chosen by query_planner. * - * Note: the Sort node always just takes a copy of the subplan's tlist - * plus ordering information. (This might seem inefficient if the - * subplan contains complex GROUP BY expressions, but in fact Sort - * does not evaluate its targetlist --- it only outputs the same - * tuples in a new order. So the expressions we might be copying - * are just dummies with no extra execution cost.) + * This is only needed if we don't use the sub_tlist chosen by + * make_subplanTargetList. We have to forget the column indexes found + * by that routine and re-locate the grouping vars in the real sub_tlist. */ -static Plan * -make_groupsortplan(Query *parse, - List *groupClause, - AttrNumber *grpColIdx, - Plan *subplan) -{ - List *sort_tlist = new_unsorted_tlist(subplan->targetlist); - int keyno = 0; - List *gl; - - foreach(gl, groupClause) - { - GroupClause *grpcl = (GroupClause *) lfirst(gl); - TargetEntry *te = nth(grpColIdx[keyno] - 1, sort_tlist); - Resdom *resdom = te->resdom; - - /* - * Check for the possibility of duplicate group-by clauses --- - * the parser should have removed 'em, but the Sort executor - * will get terribly confused if any get through! - */ - if (resdom->reskey == 0) - { - /* OK, insert the ordering info needed by the executor. */ - resdom->reskey = ++keyno; - resdom->reskeyop = grpcl->sortop; - } - } - - Assert(keyno > 0); - - return (Plan *) make_sort(parse, sort_tlist, subplan, keyno); -} - -/* - * make_sortplan - * Add a Sort node to implement an explicit ORDER BY clause. - */ -Plan * -make_sortplan(Query *parse, List *tlist, Plan *plannode, List *sortcls) +static void +locate_grouping_columns(Query *parse, + List *tlist, + List *sub_tlist, + AttrNumber *groupColIdx) { - List *sort_tlist; - List *i; int keyno = 0; + ListCell *gl; /* - * First make a copy of the tlist so that we don't corrupt the - * original. + * No work unless grouping. */ - sort_tlist = new_unsorted_tlist(tlist); + if (!parse->groupClause) + { + Assert(groupColIdx == NULL); + return; + } + Assert(groupColIdx != NULL); - foreach(i, sortcls) + foreach(gl, parse->groupClause) { - SortClause *sortcl = (SortClause *) lfirst(i); - TargetEntry *tle = get_sortgroupclause_tle(sortcl, sort_tlist); - Resdom *resdom = tle->resdom; + GroupClause *grpcl = (GroupClause *) lfirst(gl); + Node *groupexpr = get_sortgroupclause_expr(grpcl, tlist); + TargetEntry *te = NULL; + ListCell *sl; - /* - * Check for the possibility of duplicate order-by clauses --- the - * parser should have removed 'em, but the executor will get - * terribly confused if any get through! - */ - if (resdom->reskey == 0) + foreach(sl, sub_tlist) { - /* OK, insert the ordering info needed by the executor. */ - resdom->reskey = ++keyno; - resdom->reskeyop = sortcl->sortop; + te = (TargetEntry *) lfirst(sl); + if (equal(groupexpr, te->expr)) + break; } - } - - Assert(keyno > 0); + if (!sl) + elog(ERROR, "failed to locate grouping columns"); - return (Plan *) make_sort(parse, sort_tlist, plannode, keyno); + groupColIdx[keyno++] = te->resno; + } } /* @@ -1760,13 +1543,14 @@ make_sortplan(Query *parse, List *tlist, Plan *plannode, List *sortcls) * We need to transpose sort key info from the orig_tlist into new_tlist. * NOTE: this would not be good enough if we supported resjunk sort keys * for results of set operations --- then, we'd need to project a whole - * new tlist to evaluate the resjunk columns. For now, just elog if we + * new tlist to evaluate the resjunk columns. For now, just ereport if we * find any resjunk columns in orig_tlist. */ static List * postprocess_setop_tlist(List *new_tlist, List *orig_tlist) { - List *l; + ListCell *l; + ListCell *orig_tlist_item = list_head(orig_tlist); foreach(l, new_tlist) { @@ -1774,19 +1558,18 @@ postprocess_setop_tlist(List *new_tlist, List *orig_tlist) TargetEntry *orig_tle; /* ignore resjunk columns in setop result */ - if (new_tle->resdom->resjunk) + if (new_tle->resjunk) continue; - Assert(orig_tlist != NIL); - orig_tle = (TargetEntry *) lfirst(orig_tlist); - orig_tlist = lnext(orig_tlist); - if (orig_tle->resdom->resjunk) - elog(ERROR, "postprocess_setop_tlist: resjunk output columns not implemented"); - Assert(new_tle->resdom->resno == orig_tle->resdom->resno); - Assert(new_tle->resdom->restype == orig_tle->resdom->restype); - new_tle->resdom->ressortgroupref = orig_tle->resdom->ressortgroupref; + 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 */ + elog(ERROR, "resjunk output columns are not implemented"); + Assert(new_tle->resno == orig_tle->resno); + new_tle->ressortgroupref = orig_tle->ressortgroupref; } - if (orig_tlist != NIL) - elog(ERROR, "postprocess_setop_tlist: resjunk output columns not implemented"); + if (orig_tlist_item != NULL) + elog(ERROR, "resjunk output columns are not implemented"); return new_tlist; }