2 $PostgreSQL: pgsql/doc/src/sgml/ref/create_index.sgml,v 1.62 2007/04/06 22:33:41 tgl Exp $
3 PostgreSQL documentation
6 <refentry id="SQL-CREATEINDEX">
8 <refentrytitle id="sql-createindex-title">CREATE INDEX</refentrytitle>
9 <refmiscinfo>SQL - Language Statements</refmiscinfo>
13 <refname>CREATE INDEX</refname>
14 <refpurpose>define a new index</refpurpose>
17 <indexterm zone="sql-createindex">
18 <primary>CREATE INDEX</primary>
23 CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] <replaceable class="parameter">name</replaceable> ON <replaceable class="parameter">table</replaceable> [ USING <replaceable class="parameter">method</replaceable> ]
24 ( { <replaceable class="parameter">column</replaceable> | ( <replaceable class="parameter">expression</replaceable> ) } [ <replaceable class="parameter">opclass</replaceable> ] [ ASC | DESC ] [ NULLS { FIRST | LAST } ] [, ...] )
25 [ WITH ( <replaceable class="PARAMETER">storage_parameter</replaceable> = <replaceable class="PARAMETER">value</replaceable> [, ... ] ) ]
26 [ TABLESPACE <replaceable class="parameter">tablespace</replaceable> ]
27 [ WHERE <replaceable class="parameter">predicate</replaceable> ]
32 <title>Description</title>
35 <command>CREATE INDEX</command> constructs an index <replaceable
36 class="parameter">index_name</replaceable> on the specified table.
37 Indexes are primarily used to enhance database performance (though
38 inappropriate use can result in slower performance).
42 The key field(s) for the index are specified as column names,
43 or alternatively as expressions written in parentheses.
44 Multiple fields can be specified if the index method supports
49 An index field can be an expression computed from the values of
50 one or more columns of the table row. This feature can be used
51 to obtain fast access to data based on some transformation of
52 the basic data. For example, an index computed on
53 <literal>upper(col)</> would allow the clause
54 <literal>WHERE upper(col) = 'JIM'</> to use an index.
58 <productname>PostgreSQL</productname> provides the index methods
59 B-tree, hash, GiST, and GIN. Users can also define their own index
60 methods, but that is fairly complicated.
64 When the <literal>WHERE</literal> clause is present, a
65 <firstterm>partial index</firstterm> is created.
66 A partial index is an index that contains entries for only a portion of
67 a table, usually a portion that is more useful for indexing than the
68 rest of the table. For example, if you have a table that contains both
69 billed and unbilled orders where the unbilled orders take up a small
70 fraction of the total table and yet that is an often used section, you
71 can improve performance by creating an index on just that portion.
72 Another possible application is to use <literal>WHERE</literal> with
73 <literal>UNIQUE</literal> to enforce uniqueness over a subset of a
74 table. See <xref linkend="indexes-partial"> for more discussion.
78 The expression used in the <literal>WHERE</literal> clause can refer
79 only to columns of the underlying table, but it can use all columns,
80 not just the ones being indexed. Presently, subqueries and
81 aggregate expressions are also forbidden in <literal>WHERE</literal>.
82 The same restrictions apply to index fields that are expressions.
86 All functions and operators used in an index definition must be
87 <quote>immutable</>, that is, their results must depend only on
88 their arguments and never on any outside influence (such as
89 the contents of another table or the current time). This restriction
90 ensures that the behavior of the index is well-defined. To use a
91 user-defined function in an index expression or <literal>WHERE</literal>
92 clause, remember to mark the function immutable when you create it.
97 <title>Parameters</title>
101 <term><literal>UNIQUE</literal></term>
104 Causes the system to check for
105 duplicate values in the table when the index is created (if data
106 already exist) and each time data is added. Attempts to
107 insert or update data which would result in duplicate entries
108 will generate an error.
114 <term><literal>CONCURRENTLY</literal></term>
117 When this option is used, <productname>PostgreSQL</> will build the
118 index without taking any locks that prevent concurrent inserts,
119 updates, or deletes on the table; whereas a standard index build
120 locks out writes (but not reads) on the table until it's done.
121 There are several caveats to be aware of when using this option
122 — see <xref linkend="SQL-CREATEINDEX-CONCURRENTLY"
123 endterm="SQL-CREATEINDEX-CONCURRENTLY-title">.
129 <term><replaceable class="parameter">name</replaceable></term>
132 The name of the index to be created. No schema name can be included
133 here; the index is always created in the same schema as its parent
140 <term><replaceable class="parameter">table</replaceable></term>
143 The name (possibly schema-qualified) of the table to be indexed.
149 <term><replaceable class="parameter">method</replaceable></term>
152 The name of the index method to be used. Choices are
153 <literal>btree</literal>, <literal>hash</literal>,
154 <literal>gist</literal>, and <literal>gin</>. The
155 default method is <literal>btree</literal>.
161 <term><replaceable class="parameter">column</replaceable></term>
164 The name of a column of the table.
170 <term><replaceable class="parameter">expression</replaceable></term>
173 An expression based on one or more columns of the table. The
174 expression usually must be written with surrounding parentheses,
175 as shown in the syntax. However, the parentheses can be omitted
176 if the expression has the form of a function call.
182 <term><replaceable class="parameter">opclass</replaceable></term>
185 The name of an operator class. See below for details.
191 <term><literal>ASC</></term>
194 Specifies ascending sort order (which is the default).
200 <term><literal>DESC</></term>
203 Specifies descending sort order.
209 <term><literal>NULLS FIRST</></term>
212 Specifies that nulls sort before non-nulls. This is the default
213 when <literal>DESC</> is specified.
219 <term><literal>NULLS LAST</></term>
222 Specifies that nulls sort after non-nulls. This is the default
223 when <literal>DESC</> is not specified.
229 <term><replaceable class="parameter">storage_parameter</replaceable></term>
232 The name of an index-method-specific storage parameter. See
239 <term><replaceable class="parameter">tablespace</replaceable></term>
242 The tablespace in which to create the index. If not specified,
243 <xref linkend="guc-default-tablespace"> is used, or the database's
244 default tablespace if <varname>default_tablespace</> is an empty
251 <term><replaceable class="parameter">predicate</replaceable></term>
254 The constraint expression for a partial index.
261 <refsect2 id="SQL-CREATEINDEX-storage-parameters">
262 <title id="SQL-CREATEINDEX-storage-parameters-title">Index Storage Parameters</title>
265 The <literal>WITH</> clause can specify <firstterm>storage parameters</>
266 for indexes. Each index method can have its own set of allowed storage
267 parameters. The built-in index methods all accept a single parameter:
273 <term><literal>FILLFACTOR</></term>
276 The fillfactor for an index is a percentage that determines how full
277 the index method will try to pack index pages. For B-trees, leaf pages
278 are filled to this percentage during initial index build, and also
279 when extending the index at the right (largest key values). If pages
280 subsequently become completely full, they will be split, leading to
281 gradual degradation in the index's efficiency. B-trees use a default
282 fillfactor of 90, but any value from 10 to 100 can be selected.
283 If the table is static then fillfactor 100 is best to minimize the
284 index's physical size, but for heavily updated tables a smaller
285 fillfactor is better to minimize the need for page splits. The
286 other index methods use fillfactor in different but roughly analogous
287 ways; the default fillfactor varies between methods.
296 <refsect2 id="SQL-CREATEINDEX-CONCURRENTLY">
297 <title id="SQL-CREATEINDEX-CONCURRENTLY-title">Building Indexes Concurrently</title>
299 <indexterm zone="SQL-CREATEINDEX-CONCURRENTLY">
300 <primary>index</primary>
301 <secondary>building concurrently</secondary>
305 Creating an index can interfere with regular operation of a database.
306 Normally <productname>PostgreSQL</> locks the table to be indexed against
307 writes and performs the entire index build with a single scan of the
308 table. Other transactions can still read the table, but if they try to
309 insert, update, or delete rows in the table they will block until the
310 index build is finished. This could have a severe effect if the system is
311 a live production database. Large tables can take many hours to be
312 indexed, and even for smaller tables, an index build can lock out writers
313 for periods that are unacceptably long for a production system.
317 <productname>PostgreSQL</> supports building indexes without locking
318 out writes. This method is invoked by specifying the
319 <literal>CONCURRENTLY</> option of <command>CREATE INDEX</>.
320 When this option is used,
321 <productname>PostgreSQL</> must perform two scans of the table, and in
322 addition it must wait for all existing transactions to terminate. Thus
323 this method requires more total work than a standard index build and takes
324 significantly longer to complete. However, since it allows normal
325 operations to continue while the index is built, this method is useful for
326 adding new indexes in a production environment. Of course, the extra CPU
327 and I/O load imposed by the index creation might slow other operations.
331 If a problem arises during the second scan of the table, such as a
332 uniqueness violation in a unique index, the <command>CREATE INDEX</>
333 command will fail but leave behind an <quote>invalid</> index. This index
334 will be ignored for querying purposes because it might be incomplete;
335 however it will still consume update overhead. The <application>psql</>
336 <command>\d</> command will mark such an index as <literal>INVALID</>:
341 Column | Type | Modifiers
342 --------+---------+-----------
345 "idx" btree (col) INVALID
348 The recommended recovery
349 method in such cases is to drop the index and try again to perform
350 <command>CREATE INDEX CONCURRENTLY</>. (Another possibility is to rebuild
351 the index with <command>REINDEX</>. However, since <command>REINDEX</>
352 does not support concurrent builds, this option is unlikely to seem
357 Another caveat when building a unique index concurrently is that the
358 uniqueness constraint is already being enforced against other transactions
359 when the second table scan begins. This means that constraint violations
360 could be reported in other queries prior to the index becoming available
361 for use, or even in cases where the index build eventually fails. Also,
362 if a failure does occur in the second scan, the <quote>invalid</> index
363 continues to enforce its uniqueness constraint afterwards.
367 Concurrent builds of expression indexes and partial indexes are supported.
368 Errors occurring in the evaluation of these expressions could cause
369 behavior similar to that described above for unique constraint violations.
373 Regular index builds permit other regular index builds on the
374 same table to occur in parallel, but only one concurrent index build
375 can occur on a table at a time. In both cases, no other types of schema
376 modification on the table are allowed meanwhile. Another difference
377 is that a regular <command>CREATE INDEX</> command can be performed within
378 a transaction block, but <command>CREATE INDEX CONCURRENTLY</> cannot.
387 See <xref linkend="indexes"> for information about when indexes can
388 be used, when they are not used, and in which particular situations
393 Currently, only the B-tree and GiST index methods support
394 multicolumn indexes. Up to 32 fields can be specified by default.
395 (This limit can be altered when building
396 <productname>PostgreSQL</productname>.) Only B-tree currently
397 supports unique indexes.
401 An <firstterm>operator class</firstterm> can be specified for each
402 column of an index. The operator class identifies the operators to be
403 used by the index for that column. For example, a B-tree index on
404 four-byte integers would use the <literal>int4_ops</literal> class;
405 this operator class includes comparison functions for four-byte
406 integers. In practice the default operator class for the column's data
407 type is usually sufficient. The main point of having operator classes
408 is that for some data types, there could be more than one meaningful
409 ordering. For example, we might want to sort a complex-number data
410 type either by absolute value or by real part. We could do this by
411 defining two operator classes for the data type and then selecting
412 the proper class when making an index. More information about
413 operator classes is in <xref linkend="indexes-opclass"> and in <xref
418 For index methods that support ordered scans (currently, only B-tree),
419 the optional clauses <literal>ASC</>, <literal>DESC</>, <literal>NULLS
420 FIRST</>, and/or <literal>NULLS LAST</> can be specified to reverse
421 the normal sort direction of the index. Since an ordered index can be
422 scanned either forward or backward, it is not normally useful to create a
423 single-column <literal>DESC</> index — that sort ordering is already
424 available with a regular index. The value of these options is that
425 multicolumn indexes can be created that match the sort ordering requested
426 by a mixed-ordering query, such as <literal>SELECT ... ORDER BY x ASC, y
427 DESC</>. The <literal>NULLS</> options are useful if you need to support
428 <quote>nulls sort low</> behavior, rather than the default <quote>nulls
429 sort high</>, in queries that depend on indexes to avoid sorting steps.
433 Use <xref linkend="sql-dropindex" endterm="sql-dropindex-title">
438 Prior releases of <productname>PostgreSQL</productname> also had an
439 R-tree index method. This method has been removed because
440 it had no significant advantages over the GiST method.
441 If <literal>USING rtree</> is specified, <command>CREATE INDEX</>
442 will interpret it as <literal>USING gist</>, to simplify conversion
443 of old databases to GiST.
448 <title>Examples</title>
451 To create a B-tree index on the column <literal>title</literal> in
452 the table <literal>films</literal>:
454 CREATE UNIQUE INDEX title_idx ON films (title);
459 To create an index on the expression <literal>lower(title)</>,
460 allowing efficient case-insensitive searches:
462 CREATE INDEX lower_title_idx ON films ((lower(title)));
467 To create an index with non-default sort ordering of nulls:
469 CREATE INDEX title_idx_nulls_low ON films (title NULLS FIRST);
474 To create an index with non-default fill factor:
476 CREATE UNIQUE INDEX title_idx ON films (title) WITH (fillfactor = 70);
481 To create an index on the column <literal>code</> in the table
482 <literal>films</> and have the index reside in the tablespace
483 <literal>indexspace</>:
485 CREATE INDEX code_idx ON films(code) TABLESPACE indexspace;
491 Is this example correct?
494 To create a GiST index on a point attribute so that we
495 can efficiently use box operators on the result of the
498 CREATE INDEX pointloc
499 ON points USING GIST (point2box(location) box_ops);
501 WHERE point2box(points.pointloc) = boxes.box;
507 To create an index without locking out writes to the table:
509 CREATE INDEX CONCURRENTLY sales_quantity_index ON sales_table (quantity);
516 <title>Compatibility</title>
519 <command>CREATE INDEX</command> is a
520 <productname>PostgreSQL</productname> language extension. There
521 are no provisions for indexes in the SQL standard.
526 <title>See Also</title>
528 <simplelist type="inline">
529 <member><xref linkend="sql-alterindex" endterm="sql-alterindex-title"></member>
530 <member><xref linkend="sql-dropindex" endterm="sql-dropindex-title"></member>