2 Chapter 63. Index Access Method Interface Definition
6 63.1. Basic API Structure for Indexes
7 63.2. Index Access Method Functions
9 63.4. Index Locking Considerations
10 63.5. Index Uniqueness Checks
11 63.6. Index Cost Estimation Functions
13 This chapter defines the interface between the core PostgreSQL system
14 and index access methods, which manage individual index types. The core
15 system knows nothing about indexes beyond what is specified here, so it
16 is possible to develop entirely new index types by writing add-on code.
18 All indexes in PostgreSQL are what are known technically as secondary
19 indexes; that is, the index is physically separate from the table file
20 that it describes. Each index is stored as its own physical relation
21 and so is described by an entry in the pg_class catalog. The contents
22 of an index are entirely under the control of its index access method.
23 In practice, all index access methods divide indexes into standard-size
24 pages so that they can use the regular storage manager and buffer
25 manager to access the index contents. (All the existing index access
26 methods furthermore use the standard page layout described in
27 Section 66.6, and most use the same format for index tuple headers; but
28 these decisions are not forced on an access method.)
30 An index is effectively a mapping from some data key values to tuple
31 identifiers, or TIDs, of row versions (tuples) in the index's parent
32 table. A TID consists of a block number and an item number within that
33 block (see Section 66.6). This is sufficient information to fetch a
34 particular row version from the table. Indexes are not directly aware
35 that under MVCC, there might be multiple extant versions of the same
36 logical row; to an index, each tuple is an independent object that
37 needs its own index entry. Thus, an update of a row always creates
38 all-new index entries for the row, even if the key values did not
39 change. (HOT tuples are an exception to this statement; but indexes do
40 not deal with those, either.) Index entries for dead tuples are
41 reclaimed (by vacuuming) when the dead tuples themselves are reclaimed.