2 63.1. Basic API Structure for Indexes #
4 Each index access method is described by a row in the pg_am system
5 catalog. The pg_am entry specifies a name and a handler function for
6 the index access method. These entries can be created and deleted using
7 the CREATE ACCESS METHOD and DROP ACCESS METHOD SQL commands.
9 An index access method handler function must be declared to accept a
10 single argument of type internal and to return the pseudo-type
11 index_am_handler. The argument is a dummy value that simply serves to
12 prevent handler functions from being called directly from SQL commands.
13 The result of the function must be a palloc'd struct of type
14 IndexAmRoutine, which contains everything that the core code needs to
15 know to make use of the index access method. The IndexAmRoutine struct,
16 also called the access method's API struct, includes fields specifying
17 assorted fixed properties of the access method, such as whether it can
18 support multicolumn indexes. More importantly, it contains pointers to
19 support functions for the access method, which do all of the real work
20 to access indexes. These support functions are plain C functions and
21 are not visible or callable at the SQL level. The support functions are
22 described in Section 63.2.
24 The structure IndexAmRoutine is defined thus:
25 typedef struct IndexAmRoutine
30 * Total number of strategies (operators) by which we can traverse/search
31 * this AM. Zero if AM does not have a fixed set of strategy assignments.
34 /* total number of support functions that this AM uses */
36 /* opclass options support function number or 0 */
38 /* does AM support ORDER BY indexed column's value? */
40 /* does AM support ORDER BY result of an operator on indexed column? */
42 /* does AM support hashing using API consistent with the hash AM? */
44 /* do operators within an opfamily have consistent equality semantics? */
45 bool amconsistentequality;
46 /* do operators within an opfamily have consistent ordering semantics? */
47 bool amconsistentordering;
48 /* does AM support backward scanning? */
50 /* does AM support UNIQUE indexes? */
52 /* does AM support multi-column indexes? */
54 /* does AM require scans to have a constraint on the first index column? */
56 /* does AM handle ScalarArrayOpExpr quals? */
58 /* does AM handle IS NULL/IS NOT NULL quals? */
60 /* can index storage data type differ from column data type? */
62 /* can an index of this type be clustered on? */
64 /* does AM handle predicate locks? */
66 /* does AM support parallel scan? */
68 /* does AM support parallel build? */
69 bool amcanbuildparallel;
70 /* does AM support columns included with clause INCLUDE? */
72 /* does AM use maintenance_work_mem? */
73 bool amusemaintenanceworkmem;
74 /* does AM summarize tuples, with at least all tuples in the block
75 * summarized in one summary */
77 /* OR of parallel vacuum flags */
78 uint8 amparallelvacuumoptions;
79 /* type of data stored in index, or InvalidOid if variable */
82 /* interface functions */
83 ambuild_function ambuild;
84 ambuildempty_function ambuildempty;
85 aminsert_function aminsert;
86 aminsertcleanup_function aminsertcleanup; /* can be NULL */
87 ambulkdelete_function ambulkdelete;
88 amvacuumcleanup_function amvacuumcleanup;
89 amcanreturn_function amcanreturn; /* can be NULL */
90 amcostestimate_function amcostestimate;
91 amgettreeheight_function amgettreeheight; /* can be NULL */
92 amoptions_function amoptions;
93 amproperty_function amproperty; /* can be NULL */
94 ambuildphasename_function ambuildphasename; /* can be NULL */
95 amvalidate_function amvalidate;
96 amadjustmembers_function amadjustmembers; /* can be NULL */
97 ambeginscan_function ambeginscan;
98 amrescan_function amrescan;
99 amgettuple_function amgettuple; /* can be NULL */
100 amgetbitmap_function amgetbitmap; /* can be NULL */
101 amendscan_function amendscan;
102 ammarkpos_function ammarkpos; /* can be NULL */
103 amrestrpos_function amrestrpos; /* can be NULL */
105 /* interface functions to support parallel index scans */
106 amestimateparallelscan_function amestimateparallelscan; /* can be NULL */
107 aminitparallelscan_function aminitparallelscan; /* can be NULL */
108 amparallelrescan_function amparallelrescan; /* can be NULL */
110 /* interface functions to support planning */
111 amtranslate_strategy_function amtranslatestrategy; /* can be NULL */
112 amtranslate_cmptype_function amtranslatecmptype; /* can be NULL */
115 To be useful, an index access method must also have one or more
116 operator families and operator classes defined in pg_opfamily,
117 pg_opclass, pg_amop, and pg_amproc. These entries allow the planner to
118 determine what kinds of query qualifications can be used with indexes
119 of this access method. Operator families and classes are described in
120 Section 36.16, which is prerequisite material for reading this chapter.
122 An individual index is defined by a pg_class entry that describes it as
123 a physical relation, plus a pg_index entry that shows the logical
124 content of the index — that is, the set of index columns it has and the
125 semantics of those columns, as captured by the associated operator
126 classes. The index columns (key values) can be either simple columns of
127 the underlying table or expressions over the table rows. The index
128 access method normally has no interest in where the index key values
129 come from (it is always handed precomputed key values) but it will be
130 very interested in the operator class information in pg_index. Both of
131 these catalog entries can be accessed as part of the Relation data
132 structure that is passed to all operations on the index.
134 Some of the flag fields of IndexAmRoutine have nonobvious implications.
135 The requirements of amcanunique are discussed in Section 63.5. The
136 amcanmulticol flag asserts that the access method supports
137 multi-key-column indexes, while amoptionalkey asserts that it allows
138 scans where no indexable restriction clause is given for the first
139 index column. When amcanmulticol is false, amoptionalkey essentially
140 says whether the access method supports full-index scans without any
141 restriction clause. Access methods that support multiple index columns
142 must support scans that omit restrictions on any or all of the columns
143 after the first; however they are permitted to require some restriction
144 to appear for the first index column, and this is signaled by setting
145 amoptionalkey false. One reason that an index AM might set
146 amoptionalkey false is if it doesn't index null values. Since most
147 indexable operators are strict and hence cannot return true for null
148 inputs, it is at first sight attractive to not store index entries for
149 null values: they could never be returned by an index scan anyway.
150 However, this argument fails when an index scan has no restriction
151 clause for a given index column. In practice this means that indexes
152 that have amoptionalkey true must index nulls, since the planner might
153 decide to use such an index with no scan keys at all. A related
154 restriction is that an index access method that supports multiple index
155 columns must support indexing null values in columns after the first,
156 because the planner will assume the index can be used for queries that
157 do not restrict these columns. For example, consider an index on (a,b)
158 and a query with WHERE a = 4. The system will assume the index can be
159 used to scan for rows with a = 4, which is wrong if the index omits
160 rows where b is null. It is, however, OK to omit rows where the first
161 indexed column is null. An index access method that does index nulls
162 may also set amsearchnulls, indicating that it supports IS NULL and IS
163 NOT NULL clauses as search conditions.
165 The amcaninclude flag indicates whether the access method supports
166 “included” columns, that is it can store (without processing)
167 additional columns beyond the key column(s). The requirements of the
168 preceding paragraph apply only to the key columns. In particular, the
169 combination of amcanmulticol=false and amcaninclude=true is sensible:
170 it means that there can only be one key column, but there can also be
171 included column(s). Also, included columns must be allowed to be null,
172 independently of amoptionalkey.
174 The amsummarizing flag indicates whether the access method summarizes
175 the indexed tuples, with summarizing granularity of at least per block.
176 Access methods that do not point to individual tuples, but to block
177 ranges (like BRIN), may allow the HOT optimization to continue. This
178 does not apply to attributes referenced in index predicates, an update
179 of such an attribute always disables HOT.