1 /*-------------------------------------------------------------------------
4 * Definitions for planner's internal data structures.
7 * Portions Copyright (c) 1996-2015, PostgreSQL Global Development Group
8 * Portions Copyright (c) 1994, Regents of the University of California
10 * src/include/nodes/relation.h
12 *-------------------------------------------------------------------------
17 #include "access/sdir.h"
18 #include "lib/stringinfo.h"
19 #include "nodes/params.h"
20 #include "nodes/parsenodes.h"
21 #include "storage/block.h"
26 * Set of relation identifiers (indexes into the rangetable).
28 typedef Bitmapset *Relids;
31 * When looking for a "cheapest path", this enum specifies whether we want
32 * cheapest startup cost or cheapest total cost.
34 typedef enum CostSelector
36 STARTUP_COST, TOTAL_COST
40 * The cost estimate produced by cost_qual_eval() includes both a one-time
41 * (startup) cost, and a per-tuple cost.
43 typedef struct QualCost
45 Cost startup; /* one-time cost */
46 Cost per_tuple; /* per-evaluation cost */
50 * Costing aggregate function execution requires these statistics about
51 * the aggregates to be executed by a given Agg node. Note that the costs
52 * include the execution costs of the aggregates' argument expressions as
53 * well as the aggregate functions themselves.
55 typedef struct AggClauseCosts
57 int numAggs; /* total number of aggregate functions */
58 int numOrderedAggs; /* number w/ DISTINCT/ORDER BY/WITHIN GROUP */
59 QualCost transCost; /* total per-input-row execution costs */
60 Cost finalCost; /* total per-aggregated-row costs */
61 Size transitionSpace; /* space for pass-by-ref transition data */
67 * Global information for planning/optimization
69 * PlannerGlobal holds state for an entire planner invocation; this state
70 * is shared across all levels of sub-Queries that exist in the command being
74 typedef struct PlannerGlobal
78 ParamListInfo boundParams; /* Param values provided to planner() */
80 List *subplans; /* Plans for SubPlan nodes */
82 List *subroots; /* PlannerInfos for SubPlan nodes */
84 Bitmapset *rewindPlanIDs; /* indices of subplans that require REWIND */
86 List *finalrtable; /* "flat" rangetable for executor */
88 List *finalrowmarks; /* "flat" list of PlanRowMarks */
90 List *resultRelations; /* "flat" list of integer RT indexes */
92 List *relationOids; /* OIDs of relations the plan depends on */
94 List *invalItems; /* other dependencies, as PlanInvalItems */
96 int nParamExec; /* number of PARAM_EXEC Params used */
98 Index lastPHId; /* highest PlaceHolderVar ID assigned */
100 Index lastRowMarkId; /* highest PlanRowMark ID assigned */
102 bool transientPlan; /* redo plan when TransactionXmin changes? */
104 bool hasRowSecurity; /* row security applied? */
107 /* macro for fetching the Plan associated with a SubPlan node */
108 #define planner_subplan_get_plan(root, subplan) \
109 ((Plan *) list_nth((root)->glob->subplans, (subplan)->plan_id - 1))
114 * Per-query information for planning/optimization
116 * This struct is conventionally called "root" in all the planner routines.
117 * It holds links to all of the planner's working state, in addition to the
118 * original Query. Note that at present the planner extensively modifies
119 * the passed-in Query data structure; someday that should stop.
122 typedef struct PlannerInfo
126 Query *parse; /* the Query being planned */
128 PlannerGlobal *glob; /* global info for current planner run */
130 Index query_level; /* 1 at the outermost Query */
132 struct PlannerInfo *parent_root; /* NULL at outermost Query */
134 List *plan_params; /* list of PlannerParamItems, see below */
137 * simple_rel_array holds pointers to "base rels" and "other rels" (see
138 * comments for RelOptInfo for more info). It is indexed by rangetable
139 * index (so entry 0 is always wasted). Entries can be NULL when an RTE
140 * does not correspond to a base relation, such as a join RTE or an
141 * unreferenced view RTE; or if the RelOptInfo hasn't been made yet.
143 struct RelOptInfo **simple_rel_array; /* All 1-rel RelOptInfos */
144 int simple_rel_array_size; /* allocated size of array */
147 * simple_rte_array is the same length as simple_rel_array and holds
148 * pointers to the associated rangetable entries. This lets us avoid
149 * rt_fetch(), which can be a bit slow once large inheritance sets have
152 RangeTblEntry **simple_rte_array; /* rangetable as an array */
155 * all_baserels is a Relids set of all base relids (but not "other"
156 * relids) in the query; that is, the Relids identifier of the final join
157 * we need to form. This is computed in make_one_rel, just before we
158 * start making Paths.
163 * nullable_baserels is a Relids set of base relids that are nullable by
164 * some outer join in the jointree; these are rels that are potentially
165 * nullable below the WHERE clause, SELECT targetlist, etc. This is
166 * computed in deconstruct_jointree.
168 Relids nullable_baserels;
171 * join_rel_list is a list of all join-relation RelOptInfos we have
172 * considered in this planning run. For small problems we just scan the
173 * list to do lookups, but when there are many join relations we build a
174 * hash table for faster lookups. The hash table is present and valid
175 * when join_rel_hash is not NULL. Note that we still maintain the list
176 * even when using the hash table for lookups; this simplifies life for
179 List *join_rel_list; /* list of join-relation RelOptInfos */
180 struct HTAB *join_rel_hash; /* optional hashtable for join relations */
183 * When doing a dynamic-programming-style join search, join_rel_level[k]
184 * is a list of all join-relation RelOptInfos of level k, and
185 * join_cur_level is the current level. New join-relation RelOptInfos are
186 * automatically added to the join_rel_level[join_cur_level] list.
187 * join_rel_level is NULL if not in use.
189 List **join_rel_level; /* lists of join-relation RelOptInfos */
190 int join_cur_level; /* index of list being extended */
192 List *init_plans; /* init SubPlans for query */
194 List *cte_plan_ids; /* per-CTE-item list of subplan IDs */
196 List *multiexpr_params; /* List of Lists of Params for
197 * MULTIEXPR subquery outputs */
199 List *eq_classes; /* list of active EquivalenceClasses */
201 List *canon_pathkeys; /* list of "canonical" PathKeys */
203 List *left_join_clauses; /* list of RestrictInfos for
204 * mergejoinable outer join clauses
205 * w/nonnullable var on left */
207 List *right_join_clauses; /* list of RestrictInfos for
208 * mergejoinable outer join clauses
209 * w/nonnullable var on right */
211 List *full_join_clauses; /* list of RestrictInfos for
212 * mergejoinable full join clauses */
214 List *join_info_list; /* list of SpecialJoinInfos */
216 List *lateral_info_list; /* list of LateralJoinInfos */
218 List *append_rel_list; /* list of AppendRelInfos */
220 List *rowMarks; /* list of PlanRowMarks */
222 List *placeholder_list; /* list of PlaceHolderInfos */
224 List *query_pathkeys; /* desired pathkeys for query_planner(), and
225 * actual pathkeys after planning */
227 List *group_pathkeys; /* groupClause pathkeys, if any */
228 List *window_pathkeys; /* pathkeys of bottom window, if any */
229 List *distinct_pathkeys; /* distinctClause pathkeys, if any */
230 List *sort_pathkeys; /* sortClause pathkeys, if any */
232 List *minmax_aggs; /* List of MinMaxAggInfos */
234 List *initial_rels; /* RelOptInfos we are now trying to join */
236 MemoryContext planner_cxt; /* context holding PlannerInfo */
238 double total_table_pages; /* # of pages in all tables of query */
240 double tuple_fraction; /* tuple_fraction passed to query_planner */
241 double limit_tuples; /* limit_tuples passed to query_planner */
243 bool hasInheritedTarget; /* true if parse->resultRelation is an
244 * inheritance child rel */
245 bool hasJoinRTEs; /* true if any RTEs are RTE_JOIN kind */
246 bool hasLateralRTEs; /* true if any RTEs are marked LATERAL */
247 bool hasDeletedRTEs; /* true if any RTE was deleted from jointree */
248 bool hasHavingQual; /* true if havingQual was non-null */
249 bool hasPseudoConstantQuals; /* true if any RestrictInfo has
250 * pseudoconstant = true */
251 bool hasRecursion; /* true if planning a recursive WITH item */
253 /* These fields are used only when hasRecursion is true: */
254 int wt_param_id; /* PARAM_EXEC ID for the work table */
255 struct Plan *non_recursive_plan; /* plan for non-recursive term */
257 /* These fields are workspace for createplan.c */
258 Relids curOuterRels; /* outer rels above current node */
259 List *curOuterParams; /* not-yet-assigned NestLoopParams */
261 /* optional private data for join_search_hook, e.g., GEQO */
262 void *join_search_private;
267 * In places where it's known that simple_rte_array[] must have been prepared
268 * already, we just index into it to fetch RTEs. In code that might be
269 * executed before or after entering query_planner(), use this macro.
271 #define planner_rt_fetch(rti, root) \
272 ((root)->simple_rte_array ? (root)->simple_rte_array[rti] : \
273 rt_fetch(rti, (root)->parse->rtable))
278 * Per-relation information for planning/optimization
280 * For planning purposes, a "base rel" is either a plain relation (a table)
281 * or the output of a sub-SELECT or function that appears in the range table.
282 * In either case it is uniquely identified by an RT index. A "joinrel"
283 * is the joining of two or more base rels. A joinrel is identified by
284 * the set of RT indexes for its component baserels. We create RelOptInfo
285 * nodes for each baserel and joinrel, and store them in the PlannerInfo's
286 * simple_rel_array and join_rel_list respectively.
288 * Note that there is only one joinrel for any given set of component
289 * baserels, no matter what order we assemble them in; so an unordered
290 * set is the right datatype to identify it with.
292 * We also have "other rels", which are like base rels in that they refer to
293 * single RT indexes; but they are not part of the join tree, and are given
294 * a different RelOptKind to identify them. Lastly, there is a RelOptKind
295 * for "dead" relations, which are base rels that we have proven we don't
296 * need to join after all.
298 * Currently the only kind of otherrels are those made for member relations
299 * of an "append relation", that is an inheritance set or UNION ALL subquery.
300 * An append relation has a parent RTE that is a base rel, which represents
301 * the entire append relation. The member RTEs are otherrels. The parent
302 * is present in the query join tree but the members are not. The member
303 * RTEs and otherrels are used to plan the scans of the individual tables or
304 * subqueries of the append set; then the parent baserel is given Append
305 * and/or MergeAppend paths comprising the best paths for the individual
306 * member rels. (See comments for AppendRelInfo for more information.)
308 * At one time we also made otherrels to represent join RTEs, for use in
309 * handling join alias Vars. Currently this is not needed because all join
310 * alias Vars are expanded to non-aliased form during preprocess_expression.
312 * Parts of this data structure are specific to various scan and join
313 * mechanisms. It didn't seem worth creating new node types for them.
315 * relids - Set of base-relation identifiers; it is a base relation
316 * if there is just one, a join relation if more than one
317 * rows - estimated number of tuples in the relation after restriction
318 * clauses have been applied (ie, output rows of a plan for it)
319 * width - avg. number of bytes per tuple in the relation after the
320 * appropriate projections have been done (ie, output width)
321 * consider_startup - true if there is any value in keeping paths for
322 * this rel on the basis of having cheap startup cost
323 * reltargetlist - List of Var and PlaceHolderVar nodes for the values
324 * we need to output from this relation.
325 * List is in no particular order, but all rels of an
326 * appendrel set must use corresponding orders.
327 * NOTE: in an appendrel child relation, may contain
328 * arbitrary expressions pulled up from a subquery!
329 * pathlist - List of Path nodes, one for each potentially useful
330 * method of generating the relation
331 * ppilist - ParamPathInfo nodes for parameterized Paths, if any
332 * cheapest_startup_path - the pathlist member with lowest startup cost
333 * (regardless of ordering) among the unparameterized paths;
334 * or NULL if there is no unparameterized path
335 * cheapest_total_path - the pathlist member with lowest total cost
336 * (regardless of ordering) among the unparameterized paths;
337 * or if there is no unparameterized path, the path with lowest
338 * total cost among the paths with minimum parameterization
339 * cheapest_unique_path - for caching cheapest path to produce unique
340 * (no duplicates) output from relation; NULL if not yet requested
341 * cheapest_parameterized_paths - best paths for their parameterizations;
342 * always includes cheapest_total_path, even if that's unparameterized
344 * If the relation is a base relation it will have these fields set:
346 * relid - RTE index (this is redundant with the relids field, but
347 * is provided for convenience of access)
348 * rtekind - distinguishes plain relation, subquery, or function RTE
349 * min_attr, max_attr - range of valid AttrNumbers for rel
350 * attr_needed - array of bitmapsets indicating the highest joinrel
351 * in which each attribute is needed; if bit 0 is set then
352 * the attribute is needed as part of final targetlist
353 * attr_widths - cache space for per-attribute width estimates;
354 * zero means not computed yet
355 * lateral_vars - lateral cross-references of rel, if any (list of
356 * Vars and PlaceHolderVars)
357 * lateral_relids - required outer rels for LATERAL, as a Relids set
358 * (for child rels this can be more than lateral_vars)
359 * lateral_referencers - relids of rels that reference this one laterally
360 * indexlist - list of IndexOptInfo nodes for relation's indexes
361 * (always NIL if it's not a table)
362 * pages - number of disk pages in relation (zero if not a table)
363 * tuples - number of tuples in relation (not considering restrictions)
364 * allvisfrac - fraction of disk pages that are marked all-visible
365 * subplan - plan for subquery (NULL if it's not a subquery)
366 * subroot - PlannerInfo for subquery (NULL if it's not a subquery)
367 * subplan_params - list of PlannerParamItems to be passed to subquery
368 * fdwroutine - function hooks for FDW, if foreign table (else NULL)
369 * fdw_handler - OID of FDW handler, if foreign table (else InvalidOid)
370 * fdw_private - private state for FDW, if foreign table (else NULL)
372 * Note: for a subquery, tuples, subplan, subroot are not set immediately
373 * upon creation of the RelOptInfo object; they are filled in when
374 * set_subquery_pathlist processes the object. Likewise, fdwroutine
375 * and fdw_private are filled during initial path creation.
377 * For otherrels that are appendrel members, these fields are filled
378 * in just as for a baserel.
380 * The presence of the remaining fields depends on the restrictions
381 * and joins that the relation participates in:
383 * baserestrictinfo - List of RestrictInfo nodes, containing info about
384 * each non-join qualification clause in which this relation
385 * participates (only used for base rels)
386 * baserestrictcost - Estimated cost of evaluating the baserestrictinfo
387 * clauses at a single tuple (only used for base rels)
388 * joininfo - List of RestrictInfo nodes, containing info about each
389 * join clause in which this relation participates (but
390 * note this excludes clauses that might be derivable from
391 * EquivalenceClasses)
392 * has_eclass_joins - flag that EquivalenceClass joins are possible
394 * Note: Keeping a restrictinfo list in the RelOptInfo is useful only for
395 * base rels, because for a join rel the set of clauses that are treated as
396 * restrict clauses varies depending on which sub-relations we choose to join.
397 * (For example, in a 3-base-rel join, a clause relating rels 1 and 2 must be
398 * treated as a restrictclause if we join {1} and {2 3} to make {1 2 3}; but
399 * if we join {1 2} and {3} then that clause will be a restrictclause in {1 2}
400 * and should not be processed again at the level of {1 2 3}.) Therefore,
401 * the restrictinfo list in the join case appears in individual JoinPaths
402 * (field joinrestrictinfo), not in the parent relation. But it's OK for
403 * the RelOptInfo to store the joininfo list, because that is the same
404 * for a given rel no matter how we form it.
406 * We store baserestrictcost in the RelOptInfo (for base relations) because
407 * we know we will need it at least once (to price the sequential scan)
408 * and may need it multiple times to price index scans.
411 typedef enum RelOptKind
415 RELOPT_OTHER_MEMBER_REL,
419 typedef struct RelOptInfo
423 RelOptKind reloptkind;
425 /* all relations included in this RelOptInfo */
426 Relids relids; /* set of base relids (rangetable indexes) */
428 /* size estimates generated by planner */
429 double rows; /* estimated number of result tuples */
430 int width; /* estimated avg width of result tuples */
432 /* per-relation planner control flags */
433 bool consider_startup; /* keep cheap-startup-cost paths? */
435 /* materialization information */
436 List *reltargetlist; /* Vars to be output by scan of relation */
437 List *pathlist; /* Path structures */
438 List *ppilist; /* ParamPathInfos used in pathlist */
439 struct Path *cheapest_startup_path;
440 struct Path *cheapest_total_path;
441 struct Path *cheapest_unique_path;
442 List *cheapest_parameterized_paths;
444 /* information about a base rel (not set for join rels!) */
446 Oid reltablespace; /* containing tablespace */
447 RTEKind rtekind; /* RELATION, SUBQUERY, or FUNCTION */
448 AttrNumber min_attr; /* smallest attrno of rel (often <0) */
449 AttrNumber max_attr; /* largest attrno of rel */
450 Relids *attr_needed; /* array indexed [min_attr .. max_attr] */
451 int32 *attr_widths; /* array indexed [min_attr .. max_attr] */
452 List *lateral_vars; /* LATERAL Vars and PHVs referenced by rel */
453 Relids lateral_relids; /* minimum parameterization of rel */
454 Relids lateral_referencers; /* rels that reference me laterally */
455 List *indexlist; /* list of IndexOptInfo */
456 BlockNumber pages; /* size estimates derived from pg_class */
459 /* use "struct Plan" to avoid including plannodes.h here */
460 struct Plan *subplan; /* if subquery */
461 PlannerInfo *subroot; /* if subquery */
462 List *subplan_params; /* if subquery */
463 /* use "struct FdwRoutine" to avoid including fdwapi.h here */
464 struct FdwRoutine *fdwroutine; /* if foreign table */
465 Oid fdw_handler; /* if foreign table */
466 void *fdw_private; /* if foreign table */
468 /* used by various scans and joins: */
469 List *baserestrictinfo; /* RestrictInfo structures (if base
471 QualCost baserestrictcost; /* cost of evaluating the above */
472 List *joininfo; /* RestrictInfo structures for join clauses
473 * involving this rel */
474 bool has_eclass_joins; /* T means joininfo is incomplete */
479 * Per-index information for planning/optimization
481 * indexkeys[], indexcollations[], opfamily[], and opcintype[]
482 * each have ncolumns entries.
484 * sortopfamily[], reverse_sort[], and nulls_first[] likewise have
485 * ncolumns entries, if the index is ordered; but if it is unordered,
486 * those pointers are NULL.
488 * Zeroes in the indexkeys[] array indicate index columns that are
489 * expressions; there is one element in indexprs for each such column.
491 * For an ordered index, reverse_sort[] and nulls_first[] describe the
492 * sort ordering of a forward indexscan; we can also consider a backward
493 * indexscan, which will generate the reverse ordering.
495 * The indexprs and indpred expressions have been run through
496 * prepqual.c and eval_const_expressions() for ease of matching to
497 * WHERE clauses. indpred is in implicit-AND form.
499 * indextlist is a TargetEntry list representing the index columns.
500 * It provides an equivalent base-relation Var for each simple column,
501 * and links to the matching indexprs element for each expression column.
503 typedef struct IndexOptInfo
507 Oid indexoid; /* OID of the index relation */
508 Oid reltablespace; /* tablespace of index (not table) */
509 RelOptInfo *rel; /* back-link to index's table */
511 /* index-size statistics (from pg_class and elsewhere) */
512 BlockNumber pages; /* number of disk pages in index */
513 double tuples; /* number of index tuples in index */
514 int tree_height; /* index tree height, or -1 if unknown */
516 /* index descriptor information */
517 int ncolumns; /* number of columns in index */
518 int *indexkeys; /* column numbers of index's keys, or 0 */
519 Oid *indexcollations; /* OIDs of collations of index columns */
520 Oid *opfamily; /* OIDs of operator families for columns */
521 Oid *opcintype; /* OIDs of opclass declared input data types */
522 Oid *sortopfamily; /* OIDs of btree opfamilies, if orderable */
523 bool *reverse_sort; /* is sort order descending? */
524 bool *nulls_first; /* do NULLs come first in the sort order? */
525 bool *canreturn; /* which index cols can be returned in an
527 Oid relam; /* OID of the access method (in pg_am) */
529 RegProcedure amcostestimate; /* OID of the access method's cost fcn */
531 List *indexprs; /* expressions for non-simple index columns */
532 List *indpred; /* predicate if a partial index, else NIL */
534 List *indextlist; /* targetlist representing index columns */
536 bool predOK; /* true if predicate matches query */
537 bool unique; /* true if a unique index */
538 bool immediate; /* is uniqueness enforced immediately? */
539 bool hypothetical; /* true if index doesn't really exist */
540 bool amcanorderbyop; /* does AM support order by operator result? */
541 bool amoptionalkey; /* can query omit key for the first column? */
542 bool amsearcharray; /* can AM handle ScalarArrayOpExpr quals? */
543 bool amsearchnulls; /* can AM search for NULL/NOT NULL entries? */
544 bool amhasgettuple; /* does AM have amgettuple interface? */
545 bool amhasgetbitmap; /* does AM have amgetbitmap interface? */
552 * Whenever we can determine that a mergejoinable equality clause A = B is
553 * not delayed by any outer join, we create an EquivalenceClass containing
554 * the expressions A and B to record this knowledge. If we later find another
555 * equivalence B = C, we add C to the existing EquivalenceClass; this may
556 * require merging two existing EquivalenceClasses. At the end of the qual
557 * distribution process, we have sets of values that are known all transitively
558 * equal to each other, where "equal" is according to the rules of the btree
559 * operator family(s) shown in ec_opfamilies, as well as the collation shown
560 * by ec_collation. (We restrict an EC to contain only equalities whose
561 * operators belong to the same set of opfamilies. This could probably be
562 * relaxed, but for now it's not worth the trouble, since nearly all equality
563 * operators belong to only one btree opclass anyway. Similarly, we suppose
564 * that all or none of the input datatypes are collatable, so that a single
565 * collation value is sufficient.)
567 * We also use EquivalenceClasses as the base structure for PathKeys, letting
568 * us represent knowledge about different sort orderings being equivalent.
569 * Since every PathKey must reference an EquivalenceClass, we will end up
570 * with single-member EquivalenceClasses whenever a sort key expression has
571 * not been equivalenced to anything else. It is also possible that such an
572 * EquivalenceClass will contain a volatile expression ("ORDER BY random()"),
573 * which is a case that can't arise otherwise since clauses containing
574 * volatile functions are never considered mergejoinable. We mark such
575 * EquivalenceClasses specially to prevent them from being merged with
576 * ordinary EquivalenceClasses. Also, for volatile expressions we have
577 * to be careful to match the EquivalenceClass to the correct targetlist
578 * entry: consider SELECT random() AS a, random() AS b ... ORDER BY b,a.
579 * So we record the SortGroupRef of the originating sort clause.
581 * We allow equality clauses appearing below the nullable side of an outer join
582 * to form EquivalenceClasses, but these have a slightly different meaning:
583 * the included values might be all NULL rather than all the same non-null
584 * values. See src/backend/optimizer/README for more on that point.
586 * NB: if ec_merged isn't NULL, this class has been merged into another, and
587 * should be ignored in favor of using the pointed-to class.
589 typedef struct EquivalenceClass
593 List *ec_opfamilies; /* btree operator family OIDs */
594 Oid ec_collation; /* collation, if datatypes are collatable */
595 List *ec_members; /* list of EquivalenceMembers */
596 List *ec_sources; /* list of generating RestrictInfos */
597 List *ec_derives; /* list of derived RestrictInfos */
598 Relids ec_relids; /* all relids appearing in ec_members, except
599 * for child members (see below) */
600 bool ec_has_const; /* any pseudoconstants in ec_members? */
601 bool ec_has_volatile; /* the (sole) member is a volatile expr */
602 bool ec_below_outer_join; /* equivalence applies below an OJ */
603 bool ec_broken; /* failed to generate needed clauses? */
604 Index ec_sortref; /* originating sortclause label, or 0 */
605 struct EquivalenceClass *ec_merged; /* set if merged into another EC */
609 * If an EC contains a const and isn't below-outer-join, any PathKey depending
610 * on it must be redundant, since there's only one possible value of the key.
612 #define EC_MUST_BE_REDUNDANT(eclass) \
613 ((eclass)->ec_has_const && !(eclass)->ec_below_outer_join)
616 * EquivalenceMember - one member expression of an EquivalenceClass
618 * em_is_child signifies that this element was built by transposing a member
619 * for an appendrel parent relation to represent the corresponding expression
620 * for an appendrel child. These members are used for determining the
621 * pathkeys of scans on the child relation and for explicitly sorting the
622 * child when necessary to build a MergeAppend path for the whole appendrel
623 * tree. An em_is_child member has no impact on the properties of the EC as a
624 * whole; in particular the EC's ec_relids field does NOT include the child
625 * relation. An em_is_child member should never be marked em_is_const nor
626 * cause ec_has_const or ec_has_volatile to be set, either. Thus, em_is_child
627 * members are not really full-fledged members of the EC, but just reflections
628 * or doppelgangers of real members. Most operations on EquivalenceClasses
629 * should ignore em_is_child members, and those that don't should test
630 * em_relids to make sure they only consider relevant members.
632 * em_datatype is usually the same as exprType(em_expr), but can be
633 * different when dealing with a binary-compatible opfamily; in particular
634 * anyarray_ops would never work without this. Use em_datatype when
635 * looking up a specific btree operator to work with this expression.
637 typedef struct EquivalenceMember
641 Expr *em_expr; /* the expression represented */
642 Relids em_relids; /* all relids appearing in em_expr */
643 Relids em_nullable_relids; /* nullable by lower outer joins */
644 bool em_is_const; /* expression is pseudoconstant? */
645 bool em_is_child; /* derived version for a child relation? */
646 Oid em_datatype; /* the "nominal type" used by the opfamily */
652 * The sort ordering of a path is represented by a list of PathKey nodes.
653 * An empty list implies no known ordering. Otherwise the first item
654 * represents the primary sort key, the second the first secondary sort key,
655 * etc. The value being sorted is represented by linking to an
656 * EquivalenceClass containing that value and including pk_opfamily among its
657 * ec_opfamilies. The EquivalenceClass tells which collation to use, too.
658 * This is a convenient method because it makes it trivial to detect
659 * equivalent and closely-related orderings. (See optimizer/README for more
662 * Note: pk_strategy is either BTLessStrategyNumber (for ASC) or
663 * BTGreaterStrategyNumber (for DESC). We assume that all ordering-capable
664 * index types will use btree-compatible strategy numbers.
666 typedef struct PathKey
670 EquivalenceClass *pk_eclass; /* the value that is ordered */
671 Oid pk_opfamily; /* btree opfamily defining the ordering */
672 int pk_strategy; /* sort direction (ASC or DESC) */
673 bool pk_nulls_first; /* do NULLs come before normal values? */
680 * All parameterized paths for a given relation with given required outer rels
681 * link to a single ParamPathInfo, which stores common information such as
682 * the estimated rowcount for this parameterization. We do this partly to
683 * avoid recalculations, but mostly to ensure that the estimated rowcount
684 * is in fact the same for every such path.
686 * Note: ppi_clauses is only used in ParamPathInfos for base relation paths;
687 * in join cases it's NIL because the set of relevant clauses varies depending
688 * on how the join is formed. The relevant clauses will appear in each
689 * parameterized join path's joinrestrictinfo list, instead.
691 typedef struct ParamPathInfo
695 Relids ppi_req_outer; /* rels supplying parameters used by path */
696 double ppi_rows; /* estimated number of result tuples */
697 List *ppi_clauses; /* join clauses available from outer rels */
702 * Type "Path" is used as-is for sequential-scan paths, as well as some other
703 * simple plan types that we don't need any extra information in the path for.
704 * For other path types it is the first component of a larger struct.
706 * "pathtype" is the NodeTag of the Plan node we could build from this Path.
707 * It is partially redundant with the Path's NodeTag, but allows us to use
708 * the same Path type for multiple Plan types when there is no need to
709 * distinguish the Plan type during path processing.
711 * "param_info", if not NULL, links to a ParamPathInfo that identifies outer
712 * relation(s) that provide parameter values to each scan of this path.
713 * That means this path can only be joined to those rels by means of nestloop
714 * joins with this path on the inside. Also note that a parameterized path
715 * is responsible for testing all "movable" joinclauses involving this rel
716 * and the specified outer rel(s).
718 * "rows" is the same as parent->rows in simple paths, but in parameterized
719 * paths and UniquePaths it can be less than parent->rows, reflecting the
720 * fact that we've filtered by extra join conditions or removed duplicates.
722 * "pathkeys" is a List of PathKey nodes (see above), describing the sort
723 * ordering of the path's output rows.
729 NodeTag pathtype; /* tag identifying scan/join method */
731 RelOptInfo *parent; /* the relation this path can build */
732 ParamPathInfo *param_info; /* parameterization info, or NULL if none */
734 /* estimated size/costs for path (see costsize.c for more info) */
735 double rows; /* estimated number of result tuples */
736 Cost startup_cost; /* cost expended before fetching any tuples */
737 Cost total_cost; /* total cost (assuming all tuples fetched) */
739 List *pathkeys; /* sort ordering of path's output */
740 /* pathkeys is a List of PathKey nodes; see above */
743 /* Macro for extracting a path's parameterization relids; beware double eval */
744 #define PATH_REQ_OUTER(path) \
745 ((path)->param_info ? (path)->param_info->ppi_req_outer : (Relids) NULL)
748 * IndexPath represents an index scan over a single index.
750 * This struct is used for both regular indexscans and index-only scans;
751 * path.pathtype is T_IndexScan or T_IndexOnlyScan to show which is meant.
753 * 'indexinfo' is the index to be scanned.
755 * 'indexclauses' is a list of index qualification clauses, with implicit
756 * AND semantics across the list. Each clause is a RestrictInfo node from
757 * the query's WHERE or JOIN conditions. An empty list implies a full
760 * 'indexquals' has the same structure as 'indexclauses', but it contains
761 * the actual index qual conditions that can be used with the index.
762 * In simple cases this is identical to 'indexclauses', but when special
763 * indexable operators appear in 'indexclauses', they are replaced by the
764 * derived indexscannable conditions in 'indexquals'.
766 * 'indexqualcols' is an integer list of index column numbers (zero-based)
767 * of the same length as 'indexquals', showing which index column each qual
768 * is meant to be used with. 'indexquals' is required to be ordered by
769 * index column, so 'indexqualcols' must form a nondecreasing sequence.
770 * (The order of multiple quals for the same index column is unspecified.)
772 * 'indexorderbys', if not NIL, is a list of ORDER BY expressions that have
773 * been found to be usable as ordering operators for an amcanorderbyop index.
774 * The list must match the path's pathkeys, ie, one expression per pathkey
775 * in the same order. These are not RestrictInfos, just bare expressions,
776 * since they generally won't yield booleans. Also, unlike the case for
777 * quals, it's guaranteed that each expression has the index key on the left
778 * side of the operator.
780 * 'indexorderbycols' is an integer list of index column numbers (zero-based)
781 * of the same length as 'indexorderbys', showing which index column each
782 * ORDER BY expression is meant to be used with. (There is no restriction
783 * on which index column each ORDER BY can be used with.)
785 * 'indexscandir' is one of:
786 * ForwardScanDirection: forward scan of an ordered index
787 * BackwardScanDirection: backward scan of an ordered index
788 * NoMovementScanDirection: scan of an unordered index, or don't care
789 * (The executor doesn't care whether it gets ForwardScanDirection or
790 * NoMovementScanDirection for an indexscan, but the planner wants to
791 * distinguish ordered from unordered indexes for building pathkeys.)
793 * 'indextotalcost' and 'indexselectivity' are saved in the IndexPath so that
794 * we need not recompute them when considering using the same index in a
795 * bitmap index/heap scan (see BitmapHeapPath). The costs of the IndexPath
796 * itself represent the costs of an IndexScan or IndexOnlyScan plan type.
799 typedef struct IndexPath
802 IndexOptInfo *indexinfo;
807 List *indexorderbycols;
808 ScanDirection indexscandir;
810 Selectivity indexselectivity;
814 * BitmapHeapPath represents one or more indexscans that generate TID bitmaps
815 * instead of directly accessing the heap, followed by AND/OR combinations
816 * to produce a single bitmap, followed by a heap scan that uses the bitmap.
817 * Note that the output is always considered unordered, since it will come
818 * out in physical heap order no matter what the underlying indexes did.
820 * The individual indexscans are represented by IndexPath nodes, and any
821 * logic on top of them is represented by a tree of BitmapAndPath and
822 * BitmapOrPath nodes. Notice that we can use the same IndexPath node both
823 * to represent a regular (or index-only) index scan plan, and as the child
824 * of a BitmapHeapPath that represents scanning the same index using a
825 * BitmapIndexScan. The startup_cost and total_cost figures of an IndexPath
826 * always represent the costs to use it as a regular (or index-only)
827 * IndexScan. The costs of a BitmapIndexScan can be computed using the
828 * IndexPath's indextotalcost and indexselectivity.
830 typedef struct BitmapHeapPath
833 Path *bitmapqual; /* IndexPath, BitmapAndPath, BitmapOrPath */
837 * BitmapAndPath represents a BitmapAnd plan node; it can only appear as
838 * part of the substructure of a BitmapHeapPath. The Path structure is
839 * a bit more heavyweight than we really need for this, but for simplicity
840 * we make it a derivative of Path anyway.
842 typedef struct BitmapAndPath
845 List *bitmapquals; /* IndexPaths and BitmapOrPaths */
846 Selectivity bitmapselectivity;
850 * BitmapOrPath represents a BitmapOr plan node; it can only appear as
851 * part of the substructure of a BitmapHeapPath. The Path structure is
852 * a bit more heavyweight than we really need for this, but for simplicity
853 * we make it a derivative of Path anyway.
855 typedef struct BitmapOrPath
858 List *bitmapquals; /* IndexPaths and BitmapAndPaths */
859 Selectivity bitmapselectivity;
863 * TidPath represents a scan by TID
865 * tidquals is an implicitly OR'ed list of qual expressions of the form
866 * "CTID = pseudoconstant" or "CTID = ANY(pseudoconstant_array)".
867 * Note they are bare expressions, not RestrictInfos.
869 typedef struct TidPath
872 List *tidquals; /* qual(s) involving CTID = something */
876 * ForeignPath represents a potential scan of a foreign table
878 * fdw_private stores FDW private data about the scan. While fdw_private is
879 * not actually touched by the core code during normal operations, it's
880 * generally a good idea to use a representation that can be dumped by
881 * nodeToString(), so that you can examine the structure during debugging
882 * with tools like pprint().
884 typedef struct ForeignPath
891 * CustomPath represents a table scan done by some out-of-core extension.
893 * We provide a set of hooks here - which the provider must take care to set
894 * up correctly - to allow extensions to supply their own methods of scanning
895 * a relation. For example, a provider might provide GPU acceleration, a
896 * cache-based scan, or some other kind of logic we haven't dreamed up yet.
898 * CustomPaths can be injected into the planning process for a relation by
899 * set_rel_pathlist_hook functions.
901 * Core code must avoid assuming that the CustomPath is only as large as
902 * the structure declared here; providers are allowed to make it the first
903 * element in a larger structure. (Since the planner never copies Paths,
904 * this doesn't add any complication.) However, for consistency with the
905 * FDW case, we provide a "custom_private" field in CustomPath; providers
906 * may prefer to use that rather than define another struct type.
910 #define CUSTOMPATH_SUPPORT_BACKWARD_SCAN 0x0001
911 #define CUSTOMPATH_SUPPORT_MARK_RESTORE 0x0002
913 typedef struct CustomPathMethods
915 const char *CustomName;
917 /* Convert Path to a Plan */
918 struct Plan *(*PlanCustomPath) (PlannerInfo *root,
920 struct CustomPath *best_path,
923 /* Optional: print additional fields besides "private" */
924 void (*TextOutCustomPath) (StringInfo str,
925 const struct CustomPath *node);
928 typedef struct CustomPath
931 uint32 flags; /* mask of CUSTOMPATH_* flags, see above */
932 List *custom_private;
933 const CustomPathMethods *methods;
937 * AppendPath represents an Append plan, ie, successive execution of
938 * several member plans.
940 * Note: it is possible for "subpaths" to contain only one, or even no,
941 * elements. These cases are optimized during create_append_plan.
942 * In particular, an AppendPath with no subpaths is a "dummy" path that
943 * is created to represent the case that a relation is provably empty.
945 typedef struct AppendPath
948 List *subpaths; /* list of component Paths */
951 #define IS_DUMMY_PATH(p) \
952 (IsA((p), AppendPath) && ((AppendPath *) (p))->subpaths == NIL)
954 /* A relation that's been proven empty will have one path that is dummy */
955 #define IS_DUMMY_REL(r) \
956 ((r)->cheapest_total_path != NULL && \
957 IS_DUMMY_PATH((r)->cheapest_total_path))
960 * MergeAppendPath represents a MergeAppend plan, ie, the merging of sorted
961 * results from several member plans to produce similarly-sorted output.
963 typedef struct MergeAppendPath
966 List *subpaths; /* list of component Paths */
967 double limit_tuples; /* hard limit on output tuples, or -1 */
971 * ResultPath represents use of a Result plan node to compute a variable-free
972 * targetlist with no underlying tables (a "SELECT expressions" query).
973 * The query could have a WHERE clause, too, represented by "quals".
975 * Note that quals is a list of bare clauses, not RestrictInfos.
977 typedef struct ResultPath
984 * MaterialPath represents use of a Material plan node, i.e., caching of
985 * the output of its subpath. This is used when the subpath is expensive
986 * and needs to be scanned repeatedly, or when we need mark/restore ability
987 * and the subpath doesn't have it.
989 typedef struct MaterialPath
996 * UniquePath represents elimination of distinct rows from the output of
999 * This is unlike the other Path nodes in that it can actually generate
1000 * different plans: either hash-based or sort-based implementation, or a
1001 * no-op if the input path can be proven distinct already. The decision
1002 * is sufficiently localized that it's not worth having separate Path node
1003 * types. (Note: in the no-op case, we could eliminate the UniquePath node
1004 * entirely and just return the subpath; but it's convenient to have a
1005 * UniquePath in the path tree to signal upper-level routines that the input
1006 * is known distinct.)
1010 UNIQUE_PATH_NOOP, /* input is known unique already */
1011 UNIQUE_PATH_HASH, /* use hashing */
1012 UNIQUE_PATH_SORT /* use sorting */
1015 typedef struct UniquePath
1019 UniquePathMethod umethod;
1020 List *in_operators; /* equality operators of the IN clause */
1021 List *uniq_exprs; /* expressions to be made unique */
1025 * All join-type paths share these fields.
1028 typedef struct JoinPath
1034 Path *outerjoinpath; /* path for the outer side of the join */
1035 Path *innerjoinpath; /* path for the inner side of the join */
1037 List *joinrestrictinfo; /* RestrictInfos to apply to join */
1040 * See the notes for RelOptInfo and ParamPathInfo to understand why
1041 * joinrestrictinfo is needed in JoinPath, and can't be merged into the
1042 * parent RelOptInfo.
1047 * A nested-loop path needs no special fields.
1050 typedef JoinPath NestPath;
1053 * A mergejoin path has these fields.
1055 * Unlike other path types, a MergePath node doesn't represent just a single
1056 * run-time plan node: it can represent up to four. Aside from the MergeJoin
1057 * node itself, there can be a Sort node for the outer input, a Sort node
1058 * for the inner input, and/or a Material node for the inner input. We could
1059 * represent these nodes by separate path nodes, but considering how many
1060 * different merge paths are investigated during a complex join problem,
1061 * it seems better to avoid unnecessary palloc overhead.
1063 * path_mergeclauses lists the clauses (in the form of RestrictInfos)
1064 * that will be used in the merge.
1066 * Note that the mergeclauses are a subset of the parent relation's
1067 * restriction-clause list. Any join clauses that are not mergejoinable
1068 * appear only in the parent's restrict list, and must be checked by a
1069 * qpqual at execution time.
1071 * outersortkeys (resp. innersortkeys) is NIL if the outer path
1072 * (resp. inner path) is already ordered appropriately for the
1073 * mergejoin. If it is not NIL then it is a PathKeys list describing
1074 * the ordering that must be created by an explicit Sort node.
1076 * materialize_inner is TRUE if a Material node should be placed atop the
1077 * inner input. This may appear with or without an inner Sort step.
1080 typedef struct MergePath
1083 List *path_mergeclauses; /* join clauses to be used for merge */
1084 List *outersortkeys; /* keys for explicit sort, if any */
1085 List *innersortkeys; /* keys for explicit sort, if any */
1086 bool materialize_inner; /* add Materialize to inner? */
1090 * A hashjoin path has these fields.
1092 * The remarks above for mergeclauses apply for hashclauses as well.
1094 * Hashjoin does not care what order its inputs appear in, so we have
1095 * no need for sortkeys.
1098 typedef struct HashPath
1101 List *path_hashclauses; /* join clauses used for hashing */
1102 int num_batches; /* number of batches expected */
1106 * Restriction clause info.
1108 * We create one of these for each AND sub-clause of a restriction condition
1109 * (WHERE or JOIN/ON clause). Since the restriction clauses are logically
1110 * ANDed, we can use any one of them or any subset of them to filter out
1111 * tuples, without having to evaluate the rest. The RestrictInfo node itself
1112 * stores data used by the optimizer while choosing the best query plan.
1114 * If a restriction clause references a single base relation, it will appear
1115 * in the baserestrictinfo list of the RelOptInfo for that base rel.
1117 * If a restriction clause references more than one base rel, it will
1118 * appear in the joininfo list of every RelOptInfo that describes a strict
1119 * subset of the base rels mentioned in the clause. The joininfo lists are
1120 * used to drive join tree building by selecting plausible join candidates.
1121 * The clause cannot actually be applied until we have built a join rel
1122 * containing all the base rels it references, however.
1124 * When we construct a join rel that includes all the base rels referenced
1125 * in a multi-relation restriction clause, we place that clause into the
1126 * joinrestrictinfo lists of paths for the join rel, if neither left nor
1127 * right sub-path includes all base rels referenced in the clause. The clause
1128 * will be applied at that join level, and will not propagate any further up
1129 * the join tree. (Note: the "predicate migration" code was once intended to
1130 * push restriction clauses up and down the plan tree based on evaluation
1131 * costs, but it's dead code and is unlikely to be resurrected in the
1132 * foreseeable future.)
1134 * Note that in the presence of more than two rels, a multi-rel restriction
1135 * might reach different heights in the join tree depending on the join
1136 * sequence we use. So, these clauses cannot be associated directly with
1137 * the join RelOptInfo, but must be kept track of on a per-join-path basis.
1139 * RestrictInfos that represent equivalence conditions (i.e., mergejoinable
1140 * equalities that are not outerjoin-delayed) are handled a bit differently.
1141 * Initially we attach them to the EquivalenceClasses that are derived from
1142 * them. When we construct a scan or join path, we look through all the
1143 * EquivalenceClasses and generate derived RestrictInfos representing the
1144 * minimal set of conditions that need to be checked for this particular scan
1145 * or join to enforce that all members of each EquivalenceClass are in fact
1146 * equal in all rows emitted by the scan or join.
1148 * When dealing with outer joins we have to be very careful about pushing qual
1149 * clauses up and down the tree. An outer join's own JOIN/ON conditions must
1150 * be evaluated exactly at that join node, unless they are "degenerate"
1151 * conditions that reference only Vars from the nullable side of the join.
1152 * Quals appearing in WHERE or in a JOIN above the outer join cannot be pushed
1153 * down below the outer join, if they reference any nullable Vars.
1154 * RestrictInfo nodes contain a flag to indicate whether a qual has been
1155 * pushed down to a lower level than its original syntactic placement in the
1156 * join tree would suggest. If an outer join prevents us from pushing a qual
1157 * down to its "natural" semantic level (the level associated with just the
1158 * base rels used in the qual) then we mark the qual with a "required_relids"
1159 * value including more than just the base rels it actually uses. By
1160 * pretending that the qual references all the rels required to form the outer
1161 * join, we prevent it from being evaluated below the outer join's joinrel.
1162 * When we do form the outer join's joinrel, we still need to distinguish
1163 * those quals that are actually in that join's JOIN/ON condition from those
1164 * that appeared elsewhere in the tree and were pushed down to the join rel
1165 * because they used no other rels. That's what the is_pushed_down flag is
1166 * for; it tells us that a qual is not an OUTER JOIN qual for the set of base
1167 * rels listed in required_relids. A clause that originally came from WHERE
1168 * or an INNER JOIN condition will *always* have its is_pushed_down flag set.
1169 * It's possible for an OUTER JOIN clause to be marked is_pushed_down too,
1170 * if we decide that it can be pushed down into the nullable side of the join.
1171 * In that case it acts as a plain filter qual for wherever it gets evaluated.
1172 * (In short, is_pushed_down is only false for non-degenerate outer join
1173 * conditions. Possibly we should rename it to reflect that meaning?)
1175 * RestrictInfo nodes also contain an outerjoin_delayed flag, which is true
1176 * if the clause's applicability must be delayed due to any outer joins
1177 * appearing below it (ie, it has to be postponed to some join level higher
1178 * than the set of relations it actually references).
1180 * There is also an outer_relids field, which is NULL except for outer join
1181 * clauses; for those, it is the set of relids on the outer side of the
1182 * clause's outer join. (These are rels that the clause cannot be applied to
1183 * in parameterized scans, since pushing it into the join's outer side would
1184 * lead to wrong answers.)
1186 * There is also a nullable_relids field, which is the set of rels the clause
1187 * references that can be forced null by some outer join below the clause.
1189 * outerjoin_delayed = true is subtly different from nullable_relids != NULL:
1190 * a clause might reference some nullable rels and yet not be
1191 * outerjoin_delayed because it also references all the other rels of the
1192 * outer join(s). A clause that is not outerjoin_delayed can be enforced
1193 * anywhere it is computable.
1195 * In general, the referenced clause might be arbitrarily complex. The
1196 * kinds of clauses we can handle as indexscan quals, mergejoin clauses,
1197 * or hashjoin clauses are limited (e.g., no volatile functions). The code
1198 * for each kind of path is responsible for identifying the restrict clauses
1199 * it can use and ignoring the rest. Clauses not implemented by an indexscan,
1200 * mergejoin, or hashjoin will be placed in the plan qual or joinqual field
1201 * of the finished Plan node, where they will be enforced by general-purpose
1202 * qual-expression-evaluation code. (But we are still entitled to count
1203 * their selectivity when estimating the result tuple count, if we
1204 * can guess what it is...)
1206 * When the referenced clause is an OR clause, we generate a modified copy
1207 * in which additional RestrictInfo nodes are inserted below the top-level
1208 * OR/AND structure. This is a convenience for OR indexscan processing:
1209 * indexquals taken from either the top level or an OR subclause will have
1210 * associated RestrictInfo nodes.
1212 * The can_join flag is set true if the clause looks potentially useful as
1213 * a merge or hash join clause, that is if it is a binary opclause with
1214 * nonoverlapping sets of relids referenced in the left and right sides.
1215 * (Whether the operator is actually merge or hash joinable isn't checked,
1218 * The pseudoconstant flag is set true if the clause contains no Vars of
1219 * the current query level and no volatile functions. Such a clause can be
1220 * pulled out and used as a one-time qual in a gating Result node. We keep
1221 * pseudoconstant clauses in the same lists as other RestrictInfos so that
1222 * the regular clause-pushing machinery can assign them to the correct join
1223 * level, but they need to be treated specially for cost and selectivity
1224 * estimates. Note that a pseudoconstant clause can never be an indexqual
1225 * or merge or hash join clause, so it's of no interest to large parts of
1228 * When join clauses are generated from EquivalenceClasses, there may be
1229 * several equally valid ways to enforce join equivalence, of which we need
1230 * apply only one. We mark clauses of this kind by setting parent_ec to
1231 * point to the generating EquivalenceClass. Multiple clauses with the same
1232 * parent_ec in the same join are redundant.
1235 typedef struct RestrictInfo
1239 Expr *clause; /* the represented clause of WHERE or JOIN */
1241 bool is_pushed_down; /* TRUE if clause was pushed down in level */
1243 bool outerjoin_delayed; /* TRUE if delayed by lower outer join */
1245 bool can_join; /* see comment above */
1247 bool pseudoconstant; /* see comment above */
1249 /* The set of relids (varnos) actually referenced in the clause: */
1250 Relids clause_relids;
1252 /* The set of relids required to evaluate the clause: */
1253 Relids required_relids;
1255 /* If an outer-join clause, the outer-side relations, else NULL: */
1256 Relids outer_relids;
1258 /* The relids used in the clause that are nullable by lower outer joins: */
1259 Relids nullable_relids;
1261 /* These fields are set for any binary opclause: */
1262 Relids left_relids; /* relids in left side of clause */
1263 Relids right_relids; /* relids in right side of clause */
1265 /* This field is NULL unless clause is an OR clause: */
1266 Expr *orclause; /* modified clause with RestrictInfos */
1268 /* This field is NULL unless clause is potentially redundant: */
1269 EquivalenceClass *parent_ec; /* generating EquivalenceClass */
1271 /* cache space for cost and selectivity */
1272 QualCost eval_cost; /* eval cost of clause; -1 if not yet set */
1273 Selectivity norm_selec; /* selectivity for "normal" (JOIN_INNER)
1274 * semantics; -1 if not yet set; >1 means a
1275 * redundant clause */
1276 Selectivity outer_selec; /* selectivity for outer join semantics; -1 if
1279 /* valid if clause is mergejoinable, else NIL */
1280 List *mergeopfamilies; /* opfamilies containing clause operator */
1282 /* cache space for mergeclause processing; NULL if not yet set */
1283 EquivalenceClass *left_ec; /* EquivalenceClass containing lefthand */
1284 EquivalenceClass *right_ec; /* EquivalenceClass containing righthand */
1285 EquivalenceMember *left_em; /* EquivalenceMember for lefthand */
1286 EquivalenceMember *right_em; /* EquivalenceMember for righthand */
1287 List *scansel_cache; /* list of MergeScanSelCache structs */
1289 /* transient workspace for use while considering a specific join path */
1290 bool outer_is_left; /* T = outer var on left, F = on right */
1292 /* valid if clause is hashjoinable, else InvalidOid: */
1293 Oid hashjoinoperator; /* copy of clause operator */
1295 /* cache space for hashclause processing; -1 if not yet set */
1296 Selectivity left_bucketsize; /* avg bucketsize of left side */
1297 Selectivity right_bucketsize; /* avg bucketsize of right side */
1301 * Since mergejoinscansel() is a relatively expensive function, and would
1302 * otherwise be invoked many times while planning a large join tree,
1303 * we go out of our way to cache its results. Each mergejoinable
1304 * RestrictInfo carries a list of the specific sort orderings that have
1305 * been considered for use with it, and the resulting selectivities.
1307 typedef struct MergeScanSelCache
1309 /* Ordering details (cache lookup key) */
1310 Oid opfamily; /* btree opfamily defining the ordering */
1311 Oid collation; /* collation for the ordering */
1312 int strategy; /* sort direction (ASC or DESC) */
1313 bool nulls_first; /* do NULLs come before normal values? */
1315 Selectivity leftstartsel; /* first-join fraction for clause left side */
1316 Selectivity leftendsel; /* last-join fraction for clause left side */
1317 Selectivity rightstartsel; /* first-join fraction for clause right side */
1318 Selectivity rightendsel; /* last-join fraction for clause right side */
1319 } MergeScanSelCache;
1322 * Placeholder node for an expression to be evaluated below the top level
1323 * of a plan tree. This is used during planning to represent the contained
1324 * expression. At the end of the planning process it is replaced by either
1325 * the contained expression or a Var referring to a lower-level evaluation of
1326 * the contained expression. Typically the evaluation occurs below an outer
1327 * join, and Var references above the outer join might thereby yield NULL
1328 * instead of the expression value.
1330 * Although the planner treats this as an expression node type, it is not
1331 * recognized by the parser or executor, so we declare it here rather than
1335 typedef struct PlaceHolderVar
1338 Expr *phexpr; /* the represented expression */
1339 Relids phrels; /* base relids syntactically within expr src */
1340 Index phid; /* ID for PHV (unique within planner run) */
1341 Index phlevelsup; /* > 0 if PHV belongs to outer query */
1345 * "Special join" info.
1347 * One-sided outer joins constrain the order of joining partially but not
1348 * completely. We flatten such joins into the planner's top-level list of
1349 * relations to join, but record information about each outer join in a
1350 * SpecialJoinInfo struct. These structs are kept in the PlannerInfo node's
1353 * Similarly, semijoins and antijoins created by flattening IN (subselect)
1354 * and EXISTS(subselect) clauses create partial constraints on join order.
1355 * These are likewise recorded in SpecialJoinInfo structs.
1357 * We make SpecialJoinInfos for FULL JOINs even though there is no flexibility
1358 * of planning for them, because this simplifies make_join_rel()'s API.
1360 * min_lefthand and min_righthand are the sets of base relids that must be
1361 * available on each side when performing the special join. lhs_strict is
1362 * true if the special join's condition cannot succeed when the LHS variables
1363 * are all NULL (this means that an outer join can commute with upper-level
1364 * outer joins even if it appears in their RHS). We don't bother to set
1365 * lhs_strict for FULL JOINs, however.
1367 * It is not valid for either min_lefthand or min_righthand to be empty sets;
1368 * if they were, this would break the logic that enforces join order.
1370 * syn_lefthand and syn_righthand are the sets of base relids that are
1371 * syntactically below this special join. (These are needed to help compute
1372 * min_lefthand and min_righthand for higher joins.)
1374 * delay_upper_joins is set TRUE if we detect a pushed-down clause that has
1375 * to be evaluated after this join is formed (because it references the RHS).
1376 * Any outer joins that have such a clause and this join in their RHS cannot
1377 * commute with this join, because that would leave noplace to check the
1378 * pushed-down clause. (We don't track this for FULL JOINs, either.)
1380 * For a semijoin, we also extract the join operators and their RHS arguments
1381 * and set semi_operators, semi_rhs_exprs, semi_can_btree, and semi_can_hash.
1382 * This is done in support of possibly unique-ifying the RHS, so we don't
1383 * bother unless at least one of semi_can_btree and semi_can_hash can be set
1384 * true. (You might expect that this information would be computed during
1385 * join planning; but it's helpful to have it available during planning of
1386 * parameterized table scans, so we store it in the SpecialJoinInfo structs.)
1388 * jointype is never JOIN_RIGHT; a RIGHT JOIN is handled by switching
1389 * the inputs to make it a LEFT JOIN. So the allowed values of jointype
1390 * in a join_info_list member are only LEFT, FULL, SEMI, or ANTI.
1392 * For purposes of join selectivity estimation, we create transient
1393 * SpecialJoinInfo structures for regular inner joins; so it is possible
1394 * to have jointype == JOIN_INNER in such a structure, even though this is
1395 * not allowed within join_info_list. We also create transient
1396 * SpecialJoinInfos with jointype == JOIN_INNER for outer joins, since for
1397 * cost estimation purposes it is sometimes useful to know the join size under
1398 * plain innerjoin semantics. Note that lhs_strict, delay_upper_joins, and
1399 * of course the semi_xxx fields are not set meaningfully within such structs.
1402 typedef struct SpecialJoinInfo
1405 Relids min_lefthand; /* base relids in minimum LHS for join */
1406 Relids min_righthand; /* base relids in minimum RHS for join */
1407 Relids syn_lefthand; /* base relids syntactically within LHS */
1408 Relids syn_righthand; /* base relids syntactically within RHS */
1409 JoinType jointype; /* always INNER, LEFT, FULL, SEMI, or ANTI */
1410 bool lhs_strict; /* joinclause is strict for some LHS rel */
1411 bool delay_upper_joins; /* can't commute with upper RHS */
1412 /* Remaining fields are set only for JOIN_SEMI jointype: */
1413 bool semi_can_btree; /* true if semi_operators are all btree */
1414 bool semi_can_hash; /* true if semi_operators are all hash */
1415 List *semi_operators; /* OIDs of equality join operators */
1416 List *semi_rhs_exprs; /* righthand-side expressions of these ops */
1420 * "Lateral join" info.
1422 * Lateral references constrain the join order in a way that's somewhat like
1423 * outer joins, though different in detail. We construct a LateralJoinInfo
1424 * for each lateral cross-reference, placing them in the PlannerInfo node's
1425 * lateral_info_list.
1427 * For unflattened LATERAL RTEs, we generate LateralJoinInfo(s) in which
1428 * lateral_rhs is the relid of the LATERAL baserel, and lateral_lhs is a set
1429 * of relids of baserels it references, all of which must be present on the
1430 * LHS to compute a parameter needed by the RHS. Typically, lateral_lhs is
1431 * a singleton, but it can include multiple rels if the RHS references a
1432 * PlaceHolderVar with a multi-rel ph_eval_at level. We disallow joining to
1433 * only part of the LHS in such cases, since that would result in a join tree
1434 * with no convenient place to compute the PHV.
1436 * When an appendrel contains lateral references (eg "LATERAL (SELECT x.col1
1437 * UNION ALL SELECT y.col2)"), the LateralJoinInfos reference the parent
1438 * baserel not the member otherrels, since it is the parent relid that is
1439 * considered for joining purposes.
1441 * If any LATERAL RTEs were flattened into the parent query, it is possible
1442 * that the query now contains PlaceHolderVars containing lateral references,
1443 * representing expressions that need to be evaluated at particular spots in
1444 * the jointree but contain lateral references to Vars from elsewhere. These
1445 * give rise to LateralJoinInfos in which lateral_rhs is the evaluation point
1446 * of a PlaceHolderVar and lateral_lhs is the set of lateral rels it needs.
1449 typedef struct LateralJoinInfo
1452 Relids lateral_lhs; /* rels needed to compute a lateral value */
1453 Relids lateral_rhs; /* rel where lateral value is needed */
1457 * Append-relation info.
1459 * When we expand an inheritable table or a UNION-ALL subselect into an
1460 * "append relation" (essentially, a list of child RTEs), we build an
1461 * AppendRelInfo for each child RTE. The list of AppendRelInfos indicates
1462 * which child RTEs must be included when expanding the parent, and each
1463 * node carries information needed to translate Vars referencing the parent
1464 * into Vars referencing that child.
1466 * These structs are kept in the PlannerInfo node's append_rel_list.
1467 * Note that we just throw all the structs into one list, and scan the
1468 * whole list when desiring to expand any one parent. We could have used
1469 * a more complex data structure (eg, one list per parent), but this would
1470 * be harder to update during operations such as pulling up subqueries,
1471 * and not really any easier to scan. Considering that typical queries
1472 * will not have many different append parents, it doesn't seem worthwhile
1473 * to complicate things.
1475 * Note: after completion of the planner prep phase, any given RTE is an
1476 * append parent having entries in append_rel_list if and only if its
1477 * "inh" flag is set. We clear "inh" for plain tables that turn out not
1478 * to have inheritance children, and (in an abuse of the original meaning
1479 * of the flag) we set "inh" for subquery RTEs that turn out to be
1480 * flattenable UNION ALL queries. This lets us avoid useless searches
1481 * of append_rel_list.
1483 * Note: the data structure assumes that append-rel members are single
1484 * baserels. This is OK for inheritance, but it prevents us from pulling
1485 * up a UNION ALL member subquery if it contains a join. While that could
1486 * be fixed with a more complex data structure, at present there's not much
1487 * point because no improvement in the plan could result.
1490 typedef struct AppendRelInfo
1495 * These fields uniquely identify this append relationship. There can be
1496 * (in fact, always should be) multiple AppendRelInfos for the same
1497 * parent_relid, but never more than one per child_relid, since a given
1498 * RTE cannot be a child of more than one append parent.
1500 Index parent_relid; /* RT index of append parent rel */
1501 Index child_relid; /* RT index of append child rel */
1504 * For an inheritance appendrel, the parent and child are both regular
1505 * relations, and we store their rowtype OIDs here for use in translating
1506 * whole-row Vars. For a UNION-ALL appendrel, the parent and child are
1507 * both subqueries with no named rowtype, and we store InvalidOid here.
1509 Oid parent_reltype; /* OID of parent's composite type */
1510 Oid child_reltype; /* OID of child's composite type */
1513 * The N'th element of this list is a Var or expression representing the
1514 * child column corresponding to the N'th column of the parent. This is
1515 * used to translate Vars referencing the parent rel into references to
1516 * the child. A list element is NULL if it corresponds to a dropped
1517 * column of the parent (this is only possible for inheritance cases, not
1518 * UNION ALL). The list elements are always simple Vars for inheritance
1519 * cases, but can be arbitrary expressions in UNION ALL cases.
1521 * Notice we only store entries for user columns (attno > 0). Whole-row
1522 * Vars are special-cased, and system columns (attno < 0) need no special
1523 * translation since their attnos are the same for all tables.
1525 * Caution: the Vars have varlevelsup = 0. Be careful to adjust as needed
1526 * when copying into a subquery.
1528 List *translated_vars; /* Expressions in the child's Vars */
1531 * We store the parent table's OID here for inheritance, or InvalidOid for
1532 * UNION ALL. This is only needed to help in generating error messages if
1533 * an attempt is made to reference a dropped parent column.
1535 Oid parent_reloid; /* OID of parent relation */
1539 * For each distinct placeholder expression generated during planning, we
1540 * store a PlaceHolderInfo node in the PlannerInfo node's placeholder_list.
1541 * This stores info that is needed centrally rather than in each copy of the
1542 * PlaceHolderVar. The phid fields identify which PlaceHolderInfo goes with
1543 * each PlaceHolderVar. Note that phid is unique throughout a planner run,
1544 * not just within a query level --- this is so that we need not reassign ID's
1545 * when pulling a subquery into its parent.
1547 * The idea is to evaluate the expression at (only) the ph_eval_at join level,
1548 * then allow it to bubble up like a Var until the ph_needed join level.
1549 * ph_needed has the same definition as attr_needed for a regular Var.
1551 * The PlaceHolderVar's expression might contain LATERAL references to vars
1552 * coming from outside its syntactic scope. If so, those rels are *not*
1553 * included in ph_eval_at, but they are recorded in ph_lateral.
1555 * Notice that when ph_eval_at is a join rather than a single baserel, the
1556 * PlaceHolderInfo may create constraints on join order: the ph_eval_at join
1557 * has to be formed below any outer joins that should null the PlaceHolderVar.
1559 * We create a PlaceHolderInfo only after determining that the PlaceHolderVar
1560 * is actually referenced in the plan tree, so that unreferenced placeholders
1561 * don't result in unnecessary constraints on join order.
1564 typedef struct PlaceHolderInfo
1568 Index phid; /* ID for PH (unique within planner run) */
1569 PlaceHolderVar *ph_var; /* copy of PlaceHolderVar tree */
1570 Relids ph_eval_at; /* lowest level we can evaluate value at */
1571 Relids ph_lateral; /* relids of contained lateral refs, if any */
1572 Relids ph_needed; /* highest level the value is needed at */
1573 int32 ph_width; /* estimated attribute width */
1577 * For each potentially index-optimizable MIN/MAX aggregate function,
1578 * root->minmax_aggs stores a MinMaxAggInfo describing it.
1580 typedef struct MinMaxAggInfo
1584 Oid aggfnoid; /* pg_proc Oid of the aggregate */
1585 Oid aggsortop; /* Oid of its sort operator */
1586 Expr *target; /* expression we are aggregating on */
1587 PlannerInfo *subroot; /* modified "root" for planning the subquery */
1588 Path *path; /* access path for subquery */
1589 Cost pathcost; /* estimated cost to fetch first row */
1590 Param *param; /* param for subplan's output */
1594 * At runtime, PARAM_EXEC slots are used to pass values around from one plan
1595 * node to another. They can be used to pass values down into subqueries (for
1596 * outer references in subqueries), or up out of subqueries (for the results
1597 * of a subplan), or from a NestLoop plan node into its inner relation (when
1598 * the inner scan is parameterized with values from the outer relation).
1599 * The planner is responsible for assigning nonconflicting PARAM_EXEC IDs to
1600 * the PARAM_EXEC Params it generates.
1602 * Outer references are managed via root->plan_params, which is a list of
1603 * PlannerParamItems. While planning a subquery, each parent query level's
1604 * plan_params contains the values required from it by the current subquery.
1605 * During create_plan(), we use plan_params to track values that must be
1606 * passed from outer to inner sides of NestLoop plan nodes.
1608 * The item a PlannerParamItem represents can be one of three kinds:
1610 * A Var: the slot represents a variable of this level that must be passed
1611 * down because subqueries have outer references to it, or must be passed
1612 * from a NestLoop node to its inner scan. The varlevelsup value in the Var
1613 * will always be zero.
1615 * A PlaceHolderVar: this works much like the Var case, except that the
1616 * entry is a PlaceHolderVar node with a contained expression. The PHV
1617 * will have phlevelsup = 0, and the contained expression is adjusted
1618 * to match in level.
1620 * An Aggref (with an expression tree representing its argument): the slot
1621 * represents an aggregate expression that is an outer reference for some
1622 * subquery. The Aggref itself has agglevelsup = 0, and its argument tree
1623 * is adjusted to match in level.
1625 * Note: we detect duplicate Var and PlaceHolderVar parameters and coalesce
1626 * them into one slot, but we do not bother to do that for Aggrefs.
1627 * The scope of duplicate-elimination only extends across the set of
1628 * parameters passed from one query level into a single subquery, or for
1629 * nestloop parameters across the set of nestloop parameters used in a single
1630 * query level. So there is no possibility of a PARAM_EXEC slot being used
1631 * for conflicting purposes.
1633 * In addition, PARAM_EXEC slots are assigned for Params representing outputs
1634 * from subplans (values that are setParam items for those subplans). These
1635 * IDs need not be tracked via PlannerParamItems, since we do not need any
1636 * duplicate-elimination nor later processing of the represented expressions.
1637 * Instead, we just record the assignment of the slot number by incrementing
1638 * root->glob->nParamExec.
1640 typedef struct PlannerParamItem
1644 Node *item; /* the Var, PlaceHolderVar, or Aggref */
1645 int paramId; /* its assigned PARAM_EXEC slot number */
1649 * When making cost estimates for a SEMI or ANTI join, there are some
1650 * correction factors that are needed in both nestloop and hash joins
1651 * to account for the fact that the executor can stop scanning inner rows
1652 * as soon as it finds a match to the current outer row. These numbers
1653 * depend only on the selected outer and inner join relations, not on the
1654 * particular paths used for them, so it's worthwhile to calculate them
1655 * just once per relation pair not once per considered path. This struct
1656 * is filled by compute_semi_anti_join_factors and must be passed along
1657 * to the join cost estimation functions.
1659 * outer_match_frac is the fraction of the outer tuples that are
1660 * expected to have at least one match.
1661 * match_count is the average number of matches expected for
1662 * outer tuples that have at least one match.
1664 typedef struct SemiAntiJoinFactors
1666 Selectivity outer_match_frac;
1667 Selectivity match_count;
1668 } SemiAntiJoinFactors;
1671 * For speed reasons, cost estimation for join paths is performed in two
1672 * phases: the first phase tries to quickly derive a lower bound for the
1673 * join cost, and then we check if that's sufficient to reject the path.
1674 * If not, we come back for a more refined cost estimate. The first phase
1675 * fills a JoinCostWorkspace struct with its preliminary cost estimates
1676 * and possibly additional intermediate values. The second phase takes
1677 * these values as inputs to avoid repeating work.
1679 * (Ideally we'd declare this in cost.h, but it's also needed in pathnode.h,
1680 * so seems best to put it here.)
1682 typedef struct JoinCostWorkspace
1684 /* Preliminary cost estimates --- must not be larger than final ones! */
1685 Cost startup_cost; /* cost expended before fetching any tuples */
1686 Cost total_cost; /* total cost (assuming all tuples fetched) */
1688 /* Fields below here should be treated as private to costsize.c */
1689 Cost run_cost; /* non-startup cost components */
1691 /* private for cost_nestloop code */
1692 Cost inner_rescan_run_cost;
1693 double outer_matched_rows;
1694 Selectivity inner_scan_frac;
1696 /* private for cost_mergejoin code */
1697 Cost inner_run_cost;
1700 double outer_skip_rows;
1701 double inner_skip_rows;
1703 /* private for cost_hashjoin code */
1706 } JoinCostWorkspace;
1708 #endif /* RELATION_H */