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) 1996-2000, PostgreSQL, Inc
13 * Portions Copyright (c) 1994, Regents of the University of California
17 * $Header: /cvsroot/pgsql/src/backend/optimizer/plan/planmain.c,v 1.61 2000/10/05 19:11:29 tgl 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, union_planner has a different interpretation.
70 * Returns a query plan.
74 query_planner(Query *root,
76 double tuple_fraction)
83 * If the query has an empty join tree, then it's something easy like
84 * "SELECT 2+2;" or "INSERT ... VALUES()". Fall through quickly.
86 if (root->jointree->fromlist == NIL)
88 root->query_pathkeys = NIL; /* signal unordered result */
90 /* Make childless Result node to evaluate given tlist. */
91 return (Plan *) make_result(tlist, root->jointree->quals,
96 * Pull out any non-variable WHERE clauses so these can be put in a
97 * toplevel "Result" node, where they will gate execution of the whole
98 * plan (the Result will not invoke its descendant plan unless the
99 * quals are true). Note that any *really* non-variable quals will
100 * have been optimized away by eval_const_expressions(). What we're
101 * mostly interested in here is quals that depend only on outer-level
102 * vars, although if the qual reduces to "WHERE FALSE" this path will
105 root->jointree->quals = (Node *)
106 pull_constant_clauses((List *) root->jointree->quals,
110 * Create a target list that consists solely of (resdom var) target
111 * list entries, i.e., contains no arbitrary expressions.
113 * All subplan nodes will have "flat" (var-only) tlists.
115 * This implies that all expression evaluations are done at the root of
116 * the plan tree. Once upon a time there was code to try to push
117 * expensive function calls down to lower plan nodes, but that's dead
118 * code and has been for a long time...
120 var_only_tlist = flatten_tlist(tlist);
123 * Choose the best access path and build a plan for it.
125 subplan = subplanner(root, var_only_tlist, tuple_fraction);
128 * Build a result node to control the plan if we have constant quals.
134 * The result node will also be responsible for evaluating the
135 * originally requested tlist.
137 subplan = (Plan *) make_result(tlist,
138 (Node *) constant_quals,
145 * Replace the toplevel plan node's flattened target list with the
146 * targetlist given by my caller, so that expressions are
149 subplan->targetlist = tlist;
158 * Subplanner creates an entire plan consisting of joins and scans
159 * for processing a single level of attributes.
161 * flat_tlist is the flattened target list
162 * tuple_fraction is the fraction of tuples we expect will be retrieved
164 * See query_planner() comments about the interpretation of tuple_fraction.
169 subplanner(Query *root,
171 double tuple_fraction)
175 RelOptInfo *final_rel;
181 * Examine the targetlist and qualifications, adding entries to
182 * base_rel_list as relation references are found (e.g., in the
183 * qualification, the targetlist, etc.). Restrict and join clauses
184 * are added to appropriate lists belonging to the mentioned
185 * relations. We also build lists of equijoined keys for pathkey
188 root->base_rel_list = NIL;
189 root->join_rel_list = NIL;
190 root->equi_key_list = NIL;
192 build_base_rel_tlists(root, flat_tlist);
194 (void) distribute_quals_to_rels(root, (Node *) root->jointree);
197 * Make sure we have RelOptInfo nodes for all relations to be joined.
199 joined_rels = add_missing_rels_to_query(root, (Node *) root->jointree);
202 * Check that the join tree includes all the base relations used in
203 * the query --- otherwise, the parser or rewriter messed up.
205 foreach(brel, root->base_rel_list)
207 RelOptInfo *baserel = (RelOptInfo *) lfirst(brel);
208 int relid = lfirsti(baserel->relids);
210 if (! ptrMember(baserel, joined_rels))
211 elog(ERROR, "Internal error: no jointree entry for rel %s (%d)",
212 rt_fetch(relid, root->rtable)->eref->relname, relid);
216 * Use the completed lists of equijoined keys to deduce any implied
217 * but unstated equalities (for example, A=B and B=C imply A=C).
219 generate_implied_equalities(root);
222 * We should now have all the pathkey equivalence sets built, so it's
223 * now possible to convert the requested query_pathkeys to canonical
226 root->query_pathkeys = canonicalize_pathkeys(root, root->query_pathkeys);
229 * Ready to do the primary planning.
231 final_rel = make_one_rel(root);
234 elog(ERROR, "subplanner: failed to construct a relation");
236 #ifdef NOT_USED /* fix xfunc */
239 * Perform Predicate Migration on each path, to optimize and correctly
240 * assess the cost of each before choosing the cheapest one. -- JMH,
243 * Needn't do so if the top rel is pruneable: that means there's no
244 * expensive functions left to pull up. -- JMH, 11/22/92
246 if (XfuncMode != XFUNC_OFF && XfuncMode != XFUNC_NOPM &&
247 XfuncMode != XFUNC_NOPULL && !final_rel->pruneable)
251 foreach(pathnode, final_rel->pathlist)
253 if (xfunc_do_predmig((Path *) lfirst(pathnode)))
254 set_cheapest(final_rel);
260 * Now that we have an estimate of the final rel's size, we can
261 * convert a tuple_fraction specified as an absolute count (ie, a
262 * LIMIT option) into a fraction of the total tuples.
264 if (tuple_fraction >= 1.0)
265 tuple_fraction /= final_rel->rows;
268 * Determine the cheapest path, independently of any ordering
269 * considerations. We do, however, take into account whether the
270 * whole plan is expected to be evaluated or not.
272 if (tuple_fraction <= 0.0 || tuple_fraction >= 1.0)
273 cheapestpath = final_rel->cheapest_total_path;
276 get_cheapest_fractional_path_for_pathkeys(final_rel->pathlist,
280 Assert(cheapestpath != NULL);
283 * Select the best path and create a subplan to execute it.
285 * If no special sort order is wanted, or if the cheapest path is already
286 * appropriately ordered, we use the cheapest path found above.
288 if (root->query_pathkeys == NIL ||
289 pathkeys_contained_in(root->query_pathkeys,
290 cheapestpath->pathkeys))
292 root->query_pathkeys = cheapestpath->pathkeys;
293 resultplan = create_plan(root, cheapestpath);
298 * Otherwise, look to see if we have an already-ordered path that is
299 * cheaper than doing an explicit sort on the cheapest-total-cost
302 cheapestpath = final_rel->cheapest_total_path;
304 get_cheapest_fractional_path_for_pathkeys(final_rel->pathlist,
305 root->query_pathkeys,
309 Path sort_path; /* dummy for result of cost_sort */
311 cost_sort(&sort_path, root->query_pathkeys,
312 final_rel->rows, final_rel->width);
313 sort_path.startup_cost += cheapestpath->total_cost;
314 sort_path.total_cost += cheapestpath->total_cost;
315 if (compare_fractional_path_costs(presortedpath, &sort_path,
316 tuple_fraction) <= 0)
318 /* Presorted path is cheaper, use it */
319 root->query_pathkeys = presortedpath->pathkeys;
320 resultplan = create_plan(root, presortedpath);
323 /* otherwise, doing it the hard way is still cheaper */
327 * Nothing for it but to sort the cheapest-total-cost path --- but we
328 * let the caller do that. union_planner has to be able to add a sort
329 * node anyway, so no need for extra code here. (Furthermore, the
330 * given pathkeys might involve something we can't compute here, such
331 * as an aggregate function...)
333 root->query_pathkeys = cheapestpath->pathkeys;
334 resultplan = create_plan(root, cheapestpath);