1 /*-------------------------------------------------------------------------
4 * Routines to plan a single query
6 * What's in a name, anyway? The top-level entry point of the planner/
7 * optimizer is over in planner.c, not here as you might think from the
8 * file name. But this is the main code for planning a basic join operation,
9 * shorn of features like subselects, inheritance, aggregates, grouping,
10 * and so on. (Those are the things planner.c deals with.)
12 * Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group
13 * Portions Copyright (c) 1994, Regents of the University of California
17 * $Header: /cvsroot/pgsql/src/backend/optimizer/plan/planmain.c,v 1.69 2002/06/20 20:29:31 momjian Exp $
19 *-------------------------------------------------------------------------
23 #include <sys/types.h>
25 #include "optimizer/clauses.h"
26 #include "optimizer/cost.h"
27 #include "optimizer/pathnode.h"
28 #include "optimizer/paths.h"
29 #include "optimizer/planmain.h"
30 #include "optimizer/tlist.h"
31 #include "parser/parsetree.h"
32 #include "utils/memutils.h"
35 static Plan *subplanner(Query *root, List *flat_tlist,
36 double tuple_fraction);
39 /*--------------------
41 * Generate a plan for a basic query, which may involve joins but
42 * not any fancier features.
44 * tlist is the target list the query should produce (NOT root->targetList!)
45 * tuple_fraction is the fraction of tuples we expect will be retrieved
47 * Note: the Query node now also includes a query_pathkeys field, which
48 * is both an input and an output of query_planner(). The input value
49 * signals query_planner that the indicated sort order is wanted in the
50 * final output plan. The output value is the actual pathkeys of the
51 * selected path. This might not be the same as what the caller requested;
52 * the caller must do pathkeys_contained_in() to decide whether an
53 * explicit sort is still needed. (The main reason query_pathkeys is a
54 * Query field and not a passed parameter is that the low-level routines
55 * in indxpath.c need to see it.) The pathkeys value passed to query_planner
56 * has not yet been "canonicalized", since the necessary info does not get
57 * computed until subplanner() scans the qual clauses. We canonicalize it
58 * inside subplanner() as soon as that task is done. The output value
59 * will be in canonical form as well.
61 * tuple_fraction is interpreted as follows:
62 * 0 (or less): expect all tuples to be retrieved (normal case)
63 * 0 < tuple_fraction < 1: expect the given fraction of tuples available
64 * from the plan to be retrieved
65 * tuple_fraction >= 1: tuple_fraction is the absolute number of tuples
66 * expected to be retrieved (ie, a LIMIT specification)
67 * Note that while this routine and its subroutines treat a negative
68 * tuple_fraction the same as 0, grouping_planner has a different
71 * Returns a query plan.
75 query_planner(Query *root,
77 double tuple_fraction)
84 * If the query has an empty join tree, then it's something easy like
85 * "SELECT 2+2;" or "INSERT ... VALUES()". Fall through quickly.
87 if (root->jointree->fromlist == NIL)
89 root->query_pathkeys = NIL; /* signal unordered result */
91 /* Make childless Result node to evaluate given tlist. */
92 return (Plan *) make_result(tlist, root->jointree->quals,
97 * Pull out any non-variable WHERE clauses so these can be put in a
98 * toplevel "Result" node, where they will gate execution of the whole
99 * plan (the Result will not invoke its descendant plan unless the
100 * quals are true). Note that any *really* non-variable quals will
101 * have been optimized away by eval_const_expressions(). What we're
102 * mostly interested in here is quals that depend only on outer-level
103 * vars, although if the qual reduces to "WHERE FALSE" this path will
106 root->jointree->quals = (Node *)
107 pull_constant_clauses((List *) root->jointree->quals,
111 * Create a target list that consists solely of (resdom var) target
112 * list entries, i.e., contains no arbitrary expressions.
114 * All subplan nodes will have "flat" (var-only) tlists.
116 * This implies that all expression evaluations are done at the root of
117 * the plan tree. Once upon a time there was code to try to push
118 * expensive function calls down to lower plan nodes, but that's dead
119 * code and has been for a long time...
121 var_only_tlist = flatten_tlist(tlist);
124 * Choose the best access path and build a plan for it.
126 subplan = subplanner(root, var_only_tlist, tuple_fraction);
129 * Build a result node to control the plan if we have constant quals,
130 * or if the top-level plan node is one that cannot do expression
131 * evaluation (it won't be able to evaluate the requested tlist).
132 * Currently, the only plan node we might see here that falls into
133 * that category is Append.
135 * XXX future improvement: if the given tlist is flat anyway, we don't
136 * really need a Result node.
138 if (constant_quals || IsA(subplan, Append))
141 * The result node will also be responsible for evaluating the
142 * originally requested tlist.
144 subplan = (Plan *) make_result(tlist,
145 (Node *) constant_quals,
151 * Replace the toplevel plan node's flattened target list with the
152 * targetlist given by my caller, so that expressions are
155 subplan->targetlist = tlist;
164 * Subplanner creates an entire plan consisting of joins and scans
165 * for processing a single level of attributes.
167 * flat_tlist is the flattened target list
168 * tuple_fraction is the fraction of tuples we expect will be retrieved
170 * See query_planner() comments about the interpretation of tuple_fraction.
175 subplanner(Query *root,
177 double tuple_fraction)
179 RelOptInfo *final_rel;
184 /* init lists to empty */
185 root->base_rel_list = NIL;
186 root->other_rel_list = NIL;
187 root->join_rel_list = NIL;
188 root->equi_key_list = NIL;
191 * Construct RelOptInfo nodes for all base relations in query.
193 (void) add_base_rels_to_query(root, (Node *) root->jointree);
196 * Examine the targetlist and qualifications, adding entries to
197 * baserel targetlists for all referenced Vars. Restrict and join
198 * clauses are added to appropriate lists belonging to the mentioned
199 * relations. We also build lists of equijoined keys for pathkey
202 build_base_rel_tlists(root, flat_tlist);
204 (void) distribute_quals_to_rels(root, (Node *) root->jointree);
207 * Use the completed lists of equijoined keys to deduce any implied
208 * but unstated equalities (for example, A=B and B=C imply A=C).
210 generate_implied_equalities(root);
213 * We should now have all the pathkey equivalence sets built, so it's
214 * now possible to convert the requested query_pathkeys to canonical
217 root->query_pathkeys = canonicalize_pathkeys(root, root->query_pathkeys);
220 * Ready to do the primary planning.
222 final_rel = make_one_rel(root);
225 elog(ERROR, "subplanner: failed to construct a relation");
227 #ifdef NOT_USED /* fix xfunc */
230 * Perform Predicate Migration on each path, to optimize and correctly
231 * assess the cost of each before choosing the cheapest one. -- JMH,
234 * Needn't do so if the top rel is pruneable: that means there's no
235 * expensive functions left to pull up. -- JMH, 11/22/92
237 if (XfuncMode != XFUNC_OFF && XfuncMode != XFUNC_NOPM &&
238 XfuncMode != XFUNC_NOPULL && !final_rel->pruneable)
242 foreach(pathnode, final_rel->pathlist)
244 if (xfunc_do_predmig((Path *) lfirst(pathnode)))
245 set_cheapest(final_rel);
251 * Now that we have an estimate of the final rel's size, we can
252 * convert a tuple_fraction specified as an absolute count (ie, a
253 * LIMIT option) into a fraction of the total tuples.
255 if (tuple_fraction >= 1.0)
256 tuple_fraction /= final_rel->rows;
259 * Determine the cheapest path, independently of any ordering
260 * considerations. We do, however, take into account whether the
261 * whole plan is expected to be evaluated or not.
263 if (tuple_fraction <= 0.0 || tuple_fraction >= 1.0)
264 cheapestpath = final_rel->cheapest_total_path;
267 get_cheapest_fractional_path_for_pathkeys(final_rel->pathlist,
271 Assert(cheapestpath != NULL);
274 * Select the best path and create a subplan to execute it.
276 * If no special sort order is wanted, or if the cheapest path is already
277 * appropriately ordered, we use the cheapest path found above.
279 if (root->query_pathkeys == NIL ||
280 pathkeys_contained_in(root->query_pathkeys,
281 cheapestpath->pathkeys))
283 root->query_pathkeys = cheapestpath->pathkeys;
284 resultplan = create_plan(root, cheapestpath);
289 * Otherwise, look to see if we have an already-ordered path that is
290 * cheaper than doing an explicit sort on the cheapest-total-cost
293 cheapestpath = final_rel->cheapest_total_path;
295 get_cheapest_fractional_path_for_pathkeys(final_rel->pathlist,
296 root->query_pathkeys,
300 Path sort_path; /* dummy for result of cost_sort */
302 cost_sort(&sort_path, root, root->query_pathkeys,
303 final_rel->rows, final_rel->width);
304 sort_path.startup_cost += cheapestpath->total_cost;
305 sort_path.total_cost += cheapestpath->total_cost;
306 if (compare_fractional_path_costs(presortedpath, &sort_path,
307 tuple_fraction) <= 0)
309 /* Presorted path is cheaper, use it */
310 root->query_pathkeys = presortedpath->pathkeys;
311 resultplan = create_plan(root, presortedpath);
314 /* otherwise, doing it the hard way is still cheaper */
318 * Nothing for it but to sort the cheapest-total-cost path --- but we
319 * let the caller do that. grouping_planner has to be able to add a
320 * sort node anyway, so no need for extra code here. (Furthermore,
321 * the given pathkeys might involve something we can't compute here,
322 * such as an aggregate function...)
324 root->query_pathkeys = cheapestpath->pathkeys;
325 resultplan = create_plan(root, cheapestpath);