1 /*-------------------------------------------------------------------------
4 * Selectivity functions for standard operators, and assorted
5 * infrastructure for selectivity and cost estimation.
8 * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
9 * Portions Copyright (c) 1994, Regents of the University of California
11 * src/include/utils/selfuncs.h
13 *-------------------------------------------------------------------------
19 #include "access/htup.h"
20 #include "nodes/relation.h"
24 * Note: the default selectivity estimates are not chosen entirely at random.
25 * We want them to be small enough to ensure that indexscans will be used if
26 * available, for typical table densities of ~100 tuples/page. Thus, for
27 * example, 0.01 is not quite small enough, since that makes it appear that
28 * nearly all pages will be hit anyway. Also, since we sometimes estimate
29 * eqsel as 1/num_distinct, we probably want DEFAULT_NUM_DISTINCT to equal
33 /* default selectivity estimate for equalities such as "A = b" */
34 #define DEFAULT_EQ_SEL 0.005
36 /* default selectivity estimate for inequalities such as "A < b" */
37 #define DEFAULT_INEQ_SEL 0.3333333333333333
39 /* default selectivity estimate for range inequalities "A > b AND A < c" */
40 #define DEFAULT_RANGE_INEQ_SEL 0.005
42 /* default selectivity estimate for pattern-match operators such as LIKE */
43 #define DEFAULT_MATCH_SEL 0.005
45 /* default number of distinct values in a table */
46 #define DEFAULT_NUM_DISTINCT 200
48 /* default selectivity estimate for boolean and null test nodes */
49 #define DEFAULT_UNK_SEL 0.005
50 #define DEFAULT_NOT_UNK_SEL (1.0 - DEFAULT_UNK_SEL)
54 * Clamp a computed probability estimate (which may suffer from roundoff or
55 * estimation errors) to valid range. Argument must be a float variable.
57 #define CLAMP_PROBABILITY(p) \
66 /* Return data from examine_variable and friends */
67 typedef struct VariableStatData
69 Node *var; /* the Var or expression tree */
70 RelOptInfo *rel; /* Relation, or NULL if not identifiable */
71 HeapTuple statsTuple; /* pg_statistic tuple, or NULL if none */
72 /* NB: if statsTuple!=NULL, it must be freed when caller is done */
73 void (*freefunc) (HeapTuple tuple); /* how to free statsTuple */
74 Oid vartype; /* exposed type of expression */
75 Oid atttype; /* actual type (after stripping relabel) */
76 int32 atttypmod; /* actual typmod (after stripping relabel) */
77 bool isunique; /* matches unique index or DISTINCT clause */
78 bool acl_ok; /* result of ACL check on table or column */
81 #define ReleaseVariableStats(vardata) \
83 if (HeapTupleIsValid((vardata).statsTuple)) \
84 (* (vardata).freefunc) ((vardata).statsTuple); \
90 Pattern_Type_Like, Pattern_Type_Like_IC,
91 Pattern_Type_Regex, Pattern_Type_Regex_IC
96 Pattern_Prefix_None, Pattern_Prefix_Partial, Pattern_Prefix_Exact
97 } Pattern_Prefix_Status;
100 * deconstruct_indexquals is a simple function to examine the indexquals
101 * attached to a proposed IndexPath. It returns a list of IndexQualInfo
102 * structs, one per qual expression.
106 RestrictInfo *rinfo; /* the indexqual itself */
107 int indexcol; /* zero-based index column number */
108 bool varonleft; /* true if index column is on left of qual */
109 Oid clause_op; /* qual's operator OID, if relevant */
110 Node *other_operand; /* non-index operand of qual's operator */
114 * genericcostestimate is a general-purpose estimator that can be used for
115 * most index types. In some cases we use genericcostestimate as the base
116 * code and then incorporate additional index-type-specific knowledge in
117 * the type-specific calling function. To avoid code duplication, we make
118 * genericcostestimate return a number of intermediate values as well as
119 * its preliminary estimates of the output cost values. The GenericCosts
120 * struct includes all these values.
122 * Callers should initialize all fields of GenericCosts to zero. In addition,
123 * they can set numIndexTuples to some positive value if they have a better
124 * than default way of estimating the number of leaf index tuples visited.
128 /* These are the values the cost estimator must return to the planner */
129 Cost indexStartupCost; /* index-related startup cost */
130 Cost indexTotalCost; /* total index-related scan cost */
131 Selectivity indexSelectivity; /* selectivity of index */
132 double indexCorrelation; /* order correlation of index */
134 /* Intermediate values we obtain along the way */
135 double numIndexPages; /* number of leaf pages visited */
136 double numIndexTuples; /* number of leaf tuples visited */
137 double spc_random_page_cost; /* relevant random_page_cost value */
138 double num_sa_scans; /* # indexscans from ScalarArrayOps */
141 /* Hooks for plugins to get control when we ask for stats */
142 typedef bool (*get_relation_stats_hook_type) (PlannerInfo *root,
145 VariableStatData *vardata);
146 extern PGDLLIMPORT get_relation_stats_hook_type get_relation_stats_hook;
147 typedef bool (*get_index_stats_hook_type) (PlannerInfo *root,
149 AttrNumber indexattnum,
150 VariableStatData *vardata);
151 extern PGDLLIMPORT get_index_stats_hook_type get_index_stats_hook;
153 /* Functions in selfuncs.c */
155 extern void examine_variable(PlannerInfo *root, Node *node, int varRelid,
156 VariableStatData *vardata);
157 extern bool statistic_proc_security_check(VariableStatData *vardata, Oid func_oid);
158 extern bool get_restriction_variable(PlannerInfo *root, List *args,
160 VariableStatData *vardata, Node **other,
162 extern void get_join_variables(PlannerInfo *root, List *args,
163 SpecialJoinInfo *sjinfo,
164 VariableStatData *vardata1,
165 VariableStatData *vardata2,
166 bool *join_is_reversed);
167 extern double get_variable_numdistinct(VariableStatData *vardata,
169 extern double mcv_selectivity(VariableStatData *vardata, FmgrInfo *opproc,
170 Datum constval, bool varonleft,
172 extern double histogram_selectivity(VariableStatData *vardata, FmgrInfo *opproc,
173 Datum constval, bool varonleft,
174 int min_hist_size, int n_skip,
177 extern Pattern_Prefix_Status pattern_fixed_prefix(Const *patt,
181 Selectivity *rest_selec);
182 extern Const *make_greater_string(const Const *str_const, FmgrInfo *ltproc,
185 extern Selectivity boolvarsel(PlannerInfo *root, Node *arg, int varRelid);
186 extern Selectivity booltestsel(PlannerInfo *root, BoolTestType booltesttype,
187 Node *arg, int varRelid,
188 JoinType jointype, SpecialJoinInfo *sjinfo);
189 extern Selectivity nulltestsel(PlannerInfo *root, NullTestType nulltesttype,
190 Node *arg, int varRelid,
191 JoinType jointype, SpecialJoinInfo *sjinfo);
192 extern Selectivity scalararraysel(PlannerInfo *root,
193 ScalarArrayOpExpr *clause,
195 int varRelid, JoinType jointype, SpecialJoinInfo *sjinfo);
196 extern int estimate_array_length(Node *arrayexpr);
197 extern Selectivity rowcomparesel(PlannerInfo *root,
198 RowCompareExpr *clause,
199 int varRelid, JoinType jointype, SpecialJoinInfo *sjinfo);
201 extern void mergejoinscansel(PlannerInfo *root, Node *clause,
202 Oid opfamily, int strategy, bool nulls_first,
203 Selectivity *leftstart, Selectivity *leftend,
204 Selectivity *rightstart, Selectivity *rightend);
206 extern double estimate_num_groups(PlannerInfo *root, List *groupExprs,
207 double input_rows, List **pgset);
209 extern Selectivity estimate_hash_bucketsize(PlannerInfo *root, Node *hashkey,
212 extern List *deconstruct_indexquals(IndexPath *path);
213 extern void genericcostestimate(PlannerInfo *root, IndexPath *path,
216 GenericCosts *costs);
218 /* Functions in array_selfuncs.c */
220 extern Selectivity scalararraysel_containment(PlannerInfo *root,
221 Node *leftop, Node *rightop,
222 Oid elemtype, bool isEquality, bool useOr,
225 #endif /* SELFUNCS_H */