4 CREATE INDEX — define a new index
8 CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] name ] ON [ ONLY ]
9 table_name [ USING method ]
10 ( { column_name | ( expression ) } [ COLLATE collation ] [ opclass [ ( opcla
11 ss_parameter = value [, ... ] ) ] ] [ ASC | DESC ] [ NULLS { FIRST | LAST } ] [,
13 [ INCLUDE ( column_name [, ...] ) ]
14 [ NULLS [ NOT ] DISTINCT ]
15 [ WITH ( storage_parameter [= value] [, ... ] ) ]
16 [ TABLESPACE tablespace_name ]
21 CREATE INDEX constructs an index on the specified column(s) of the
22 specified relation, which can be a table or a materialized view.
23 Indexes are primarily used to enhance database performance (though
24 inappropriate use can result in slower performance).
26 The key field(s) for the index are specified as column names, or
27 alternatively as expressions written in parentheses. Multiple fields
28 can be specified if the index method supports multicolumn indexes.
30 An index field can be an expression computed from the values of one or
31 more columns of the table row. This feature can be used to obtain fast
32 access to data based on some transformation of the basic data. For
33 example, an index computed on upper(col) would allow the clause WHERE
34 upper(col) = 'JIM' to use an index.
36 PostgreSQL provides the index methods B-tree, hash, GiST, SP-GiST, GIN,
37 and BRIN. Users can also define their own index methods, but that is
40 When the WHERE clause is present, a partial index is created. A partial
41 index is an index that contains entries for only a portion of a table,
42 usually a portion that is more useful for indexing than the rest of the
43 table. For example, if you have a table that contains both billed and
44 unbilled orders where the unbilled orders take up a small fraction of
45 the total table and yet that is an often used section, you can improve
46 performance by creating an index on just that portion. Another possible
47 application is to use WHERE with UNIQUE to enforce uniqueness over a
48 subset of a table. See Section 11.8 for more discussion.
50 The expression used in the WHERE clause can refer only to columns of
51 the underlying table, but it can use all columns, not just the ones
52 being indexed. Presently, subqueries and aggregate expressions are also
53 forbidden in WHERE. The same restrictions apply to index fields that
56 All functions and operators used in an index definition must be
57 “immutable”, that is, their results must depend only on their arguments
58 and never on any outside influence (such as the contents of another
59 table or the current time). This restriction ensures that the behavior
60 of the index is well-defined. To use a user-defined function in an
61 index expression or WHERE clause, remember to mark the function
62 immutable when you create it.
67 Causes the system to check for duplicate values in the table
68 when the index is created (if data already exist) and each time
69 data is added. Attempts to insert or update data which would
70 result in duplicate entries will generate an error.
72 Additional restrictions apply when unique indexes are applied to
73 partitioned tables; see CREATE TABLE.
76 When this option is used, PostgreSQL will build the index
77 without taking any locks that prevent concurrent inserts,
78 updates, or deletes on the table; whereas a standard index build
79 locks out writes (but not reads) on the table until it's done.
80 There are several caveats to be aware of when using this option
81 — see Building Indexes Concurrently below.
83 For temporary tables, CREATE INDEX is always non-concurrent, as
84 no other session can access them, and non-concurrent index
88 Do not throw an error if a relation with the same name already
89 exists. A notice is issued in this case. Note that there is no
90 guarantee that the existing index is anything like the one that
91 would have been created. Index name is required when IF NOT
95 The optional INCLUDE clause specifies a list of columns which
96 will be included in the index as non-key columns. A non-key
97 column cannot be used in an index scan search qualification, and
98 it is disregarded for purposes of any uniqueness or exclusion
99 constraint enforced by the index. However, an index-only scan
100 can return the contents of non-key columns without having to
101 visit the index's table, since they are available directly from
102 the index entry. Thus, addition of non-key columns allows
103 index-only scans to be used for queries that otherwise could not
106 It's wise to be conservative about adding non-key columns to an
107 index, especially wide columns. If an index tuple exceeds the
108 maximum size allowed for the index type, data insertion will
109 fail. In any case, non-key columns duplicate data from the
110 index's table and bloat the size of the index, thus potentially
111 slowing searches. Furthermore, B-tree deduplication is never
112 used with indexes that have a non-key column.
114 Columns listed in the INCLUDE clause don't need appropriate
115 operator classes; the clause can include columns whose data
116 types don't have operator classes defined for a given access
119 Expressions are not supported as included columns since they
120 cannot be used in index-only scans.
122 Currently, the B-tree, GiST and SP-GiST index access methods
123 support this feature. In these indexes, the values of columns
124 listed in the INCLUDE clause are included in leaf tuples which
125 correspond to heap tuples, but are not included in upper-level
126 index entries used for tree navigation.
129 The name of the index to be created. No schema name can be
130 included here; the index is always created in the same schema as
131 its parent table. The name of the index must be distinct from
132 the name of any other relation (table, sequence, index, view,
133 materialized view, or foreign table) in that schema. If the name
134 is omitted, PostgreSQL chooses a suitable name based on the
135 parent table's name and the indexed column name(s).
138 Indicates not to recurse creating indexes on partitions, if the
139 table is partitioned. The default is to recurse.
142 The name (possibly schema-qualified) of the table to be indexed.
145 The name of the index method to be used. Choices are btree,
146 hash, gist, spgist, gin, brin, or user-installed access methods
147 like bloom. The default method is btree.
150 The name of a column of the table.
153 An expression based on one or more columns of the table. The
154 expression usually must be written with surrounding parentheses,
155 as shown in the syntax. However, the parentheses can be omitted
156 if the expression has the form of a function call.
159 The name of the collation to use for the index. By default, the
160 index uses the collation declared for the column to be indexed
161 or the result collation of the expression to be indexed. Indexes
162 with non-default collations can be useful for queries that
163 involve expressions using non-default collations.
166 The name of an operator class. See below for details.
169 The name of an operator class parameter. See below for details.
172 Specifies ascending sort order (which is the default).
175 Specifies descending sort order.
178 Specifies that nulls sort before non-nulls. This is the default
179 when DESC is specified.
182 Specifies that nulls sort after non-nulls. This is the default
183 when DESC is not specified.
187 Specifies whether for a unique index, null values should be
188 considered distinct (not equal). The default is that they are
189 distinct, so that a unique index could contain multiple null
193 The name of an index-method-specific storage parameter. See
194 Index Storage Parameters below for details.
197 The tablespace in which to create the index. If not specified,
198 default_tablespace is consulted, or temp_tablespaces for indexes
202 The constraint expression for a partial index.
204 Index Storage Parameters
206 The optional WITH clause specifies storage parameters for the index.
207 Each index method has its own set of allowed storage parameters.
209 The B-tree, hash, GiST and SP-GiST index methods all accept this
212 fillfactor (integer) #
213 Controls how full the index method will try to pack index pages.
214 For B-trees, leaf pages are filled to this percentage during
215 initial index builds, and also when extending the index at the
216 right (adding new largest key values). If pages subsequently
217 become completely full, they will be split, leading to
218 fragmentation of the on-disk index structure. B-trees use a
219 default fillfactor of 90, but any integer value from 10 to 100
222 B-tree indexes on tables where many inserts and/or updates are
223 anticipated can benefit from lower fillfactor settings at CREATE
224 INDEX time (following bulk loading into the table). Values in
225 the range of 50 - 90 can usefully “smooth out” the rate of page
226 splits during the early life of the B-tree index (lowering
227 fillfactor like this may even lower the absolute number of page
228 splits, though this effect is highly workload dependent). The
229 B-tree bottom-up index deletion technique described in
230 Section 65.1.4.2 is dependent on having some “extra” space on
231 pages to store “extra” tuple versions, and so can be affected by
232 fillfactor (though the effect is usually not significant).
234 In other specific cases it might be useful to increase
235 fillfactor to 100 at CREATE INDEX time as a way of maximizing
236 space utilization. You should only consider this when you are
237 completely sure that the table is static (i.e. that it will
238 never be affected by either inserts or updates). A fillfactor
239 setting of 100 otherwise risks harming performance: even a few
240 updates or inserts will cause a sudden flood of page splits.
242 The other index methods use fillfactor in different but roughly
243 analogous ways; the default fillfactor varies between methods.
245 B-tree indexes additionally accept this parameter:
247 deduplicate_items (boolean) #
248 Controls usage of the B-tree deduplication technique described
249 in Section 65.1.4.3. Set to ON or OFF to enable or disable the
250 optimization. (Alternative spellings of ON and OFF are allowed
251 as described in Section 19.1.) The default is ON.
255 Turning deduplicate_items off via ALTER INDEX prevents future
256 insertions from triggering deduplication, but does not in itself
257 make existing posting list tuples use the standard tuple
260 GiST indexes additionally accept this parameter:
263 Controls whether the buffered build technique described in
264 Section 65.2.4.1 is used to build the index. With OFF buffering
265 is disabled, with ON it is enabled, and with AUTO it is
266 initially disabled, but is turned on on-the-fly once the index
267 size reaches effective_cache_size. The default is AUTO. Note
268 that if sorted build is possible, it will be used instead of
269 buffered build unless buffering=ON is specified.
271 GIN indexes accept these parameters:
273 fastupdate (boolean) #
274 Controls usage of the fast update technique described in
275 Section 65.4.4.1. ON enables fast update, OFF disables it. The
280 Turning fastupdate off via ALTER INDEX prevents future
281 insertions from going into the list of pending index entries,
282 but does not in itself flush existing entries. You might want to
283 VACUUM the table or call the gin_clean_pending_list function
284 afterward to ensure the pending list is emptied.
286 gin_pending_list_limit (integer) #
287 Overrides the global setting of gin_pending_list_limit for this
288 index. This value is specified in kilobytes.
290 BRIN indexes accept these parameters:
292 pages_per_range (integer) #
293 Defines the number of table blocks that make up one block range
294 for each entry of a BRIN index (see Section 65.5.1 for more
295 details). The default is 128.
297 autosummarize (boolean) #
298 Defines whether a summarization run is queued for the previous
299 page range whenever an insertion is detected on the next one
300 (see Section 65.5.1.1 for more details). The default is off.
302 Building Indexes Concurrently
304 Creating an index can interfere with regular operation of a database.
305 Normally PostgreSQL locks the table to be indexed against writes and
306 performs the entire index build with a single scan of the table. Other
307 transactions can still read the table, but if they try to insert,
308 update, or delete rows in the table they will block until the index
309 build is finished. This could have a severe effect if the system is a
310 live production database. Very large tables can take many hours to be
311 indexed, and even for smaller tables, an index build can lock out
312 writers for periods that are unacceptably long for a production system.
314 PostgreSQL supports building indexes without locking out writes. This
315 method is invoked by specifying the CONCURRENTLY option of CREATE
316 INDEX. When this option is used, PostgreSQL must perform two scans of
317 the table, and in addition it must wait for all existing transactions
318 that could potentially modify or use the index to terminate. Thus this
319 method requires more total work than a standard index build and takes
320 significantly longer to complete. However, since it allows normal
321 operations to continue while the index is built, this method is useful
322 for adding new indexes in a production environment. Of course, the
323 extra CPU and I/O load imposed by the index creation might slow other
326 In a concurrent index build, the index is actually entered as an
327 “invalid” index into the system catalogs in one transaction, then two
328 table scans occur in two more transactions. Before each table scan, the
329 index build must wait for existing transactions that have modified the
330 table to terminate. After the second scan, the index build must wait
331 for any transactions that have a snapshot (see Chapter 13) predating
332 the second scan to terminate, including transactions used by any phase
333 of concurrent index builds on other tables, if the indexes involved are
334 partial or have columns that are not simple column references. Then
335 finally the index can be marked “valid” and ready for use, and the
336 CREATE INDEX command terminates. Even then, however, the index may not
337 be immediately usable for queries: in the worst case, it cannot be used
338 as long as transactions exist that predate the start of the index
341 If a problem arises while scanning the table, such as a deadlock or a
342 uniqueness violation in a unique index, the CREATE INDEX command will
343 fail but leave behind an “invalid” index. This index will be ignored
344 for querying purposes because it might be incomplete; however it will
345 still consume update overhead. The psql \d command will report such an
349 Column | Type | Collation | Nullable | Default
350 --------+---------+-----------+----------+---------
353 "idx" btree (col) INVALID
355 The recommended recovery method in such cases is to drop the index and
356 try again to perform CREATE INDEX CONCURRENTLY. (Another possibility is
357 to rebuild the index with REINDEX INDEX CONCURRENTLY).
359 Another caveat when building a unique index concurrently is that the
360 uniqueness constraint is already being enforced against other
361 transactions when the second table scan begins. This means that
362 constraint violations could be reported in other queries prior to the
363 index becoming available for use, or even in cases where the index
364 build eventually fails. Also, if a failure does occur in the second
365 scan, the “invalid” index continues to enforce its uniqueness
366 constraint afterwards.
368 Concurrent builds of expression indexes and partial indexes are
369 supported. Errors occurring in the evaluation of these expressions
370 could cause behavior similar to that described above for unique
371 constraint violations.
373 Regular index builds permit other regular index builds on the same
374 table to occur simultaneously, but only one concurrent index build can
375 occur on a table at a time. In either case, schema modification of the
376 table is not allowed while the index is being built. Another difference
377 is that a regular CREATE INDEX command can be performed within a
378 transaction block, but CREATE INDEX CONCURRENTLY cannot.
380 Concurrent builds for indexes on partitioned tables are currently not
381 supported. However, you may concurrently build the index on each
382 partition individually and then finally create the partitioned index
383 non-concurrently in order to reduce the time where writes to the
384 partitioned table will be locked out. In this case, building the
385 partitioned index is a metadata only operation.
389 See Chapter 11 for information about when indexes can be used, when
390 they are not used, and in which particular situations they can be
393 Currently, only the B-tree, GiST, GIN, and BRIN index methods support
394 multiple-key-column indexes. Whether there can be multiple key columns
395 is independent of whether INCLUDE columns can be added to the index.
396 Indexes can have up to 32 columns, including INCLUDE columns. (This
397 limit can be altered when building PostgreSQL.) Only B-tree currently
398 supports unique indexes.
400 An operator class with optional parameters can be specified for each
401 column of an index. The operator class identifies the operators to be
402 used by the index for that column. For example, a B-tree index on
403 four-byte integers would use the int4_ops class; this operator class
404 includes comparison functions for four-byte integers. In practice the
405 default operator class for the column's data type is usually
406 sufficient. The main point of having operator classes is that for some
407 data types, there could be more than one meaningful ordering. For
408 example, we might want to sort a complex-number data type either by
409 absolute value or by real part. We could do this by defining two
410 operator classes for the data type and then selecting the proper class
411 when creating an index. More information about operator classes is in
412 Section 11.10 and in Section 36.16.
414 When CREATE INDEX is invoked on a partitioned table, the default
415 behavior is to recurse to all partitions to ensure they all have
416 matching indexes. Each partition is first checked to determine whether
417 an equivalent index already exists, and if so, that index will become
418 attached as a partition index to the index being created, which will
419 become its parent index. If no matching index exists, a new index will
420 be created and automatically attached; the name of the new index in
421 each partition will be determined as if no index name had been
422 specified in the command. If the ONLY option is specified, no recursion
423 is done, and the index is marked invalid. (ALTER INDEX ... ATTACH
424 PARTITION marks the index valid, once all partitions acquire matching
425 indexes.) Note, however, that any partition that is created in the
426 future using CREATE TABLE ... PARTITION OF will automatically have a
427 matching index, regardless of whether ONLY is specified.
429 For index methods that support ordered scans (currently, only B-tree),
430 the optional clauses ASC, DESC, NULLS FIRST, and/or NULLS LAST can be
431 specified to modify the sort ordering of the index. Since an ordered
432 index can be scanned either forward or backward, it is not normally
433 useful to create a single-column DESC index — that sort ordering is
434 already available with a regular index. The value of these options is
435 that multicolumn indexes can be created that match the sort ordering
436 requested by a mixed-ordering query, such as SELECT ... ORDER BY x ASC,
437 y DESC. The NULLS options are useful if you need to support “nulls sort
438 low” behavior, rather than the default “nulls sort high”, in queries
439 that depend on indexes to avoid sorting steps.
441 The system regularly collects statistics on all of a table's columns.
442 Newly-created non-expression indexes can immediately use these
443 statistics to determine an index's usefulness. For new expression
444 indexes, it is necessary to run ANALYZE or wait for the autovacuum
445 daemon to analyze the table to generate statistics for these indexes.
447 While CREATE INDEX is running, the search_path is temporarily changed
448 to pg_catalog, pg_temp.
450 For most index methods, the speed of creating an index is dependent on
451 the setting of maintenance_work_mem. Larger values will reduce the time
452 needed for index creation, so long as you don't make it larger than the
453 amount of memory really available, which would drive the machine into
456 PostgreSQL can build indexes while leveraging multiple CPUs in order to
457 process the table rows faster. This feature is known as parallel index
458 build. For index methods that support building indexes in parallel
459 (currently, B-tree, GIN, and BRIN), maintenance_work_mem specifies the
460 maximum amount of memory that can be used by each index build operation
461 as a whole, regardless of how many worker processes were started.
462 Generally, a cost model automatically determines how many worker
463 processes should be requested, if any.
465 Parallel index builds may benefit from increasing maintenance_work_mem
466 where an equivalent serial index build will see little or no benefit.
467 Note that maintenance_work_mem may influence the number of worker
468 processes requested, since parallel workers must have at least a 32MB
469 share of the total maintenance_work_mem budget. There must also be a
470 remaining 32MB share for the leader process. Increasing
471 max_parallel_maintenance_workers may allow more workers to be used,
472 which will reduce the time needed for index creation, so long as the
473 index build is not already I/O bound. Of course, there should also be
474 sufficient CPU capacity that would otherwise lie idle.
476 Setting a value for parallel_workers via ALTER TABLE directly controls
477 how many parallel worker processes will be requested by a CREATE INDEX
478 against the table. This bypasses the cost model completely, and
479 prevents maintenance_work_mem from affecting how many parallel workers
480 are requested. Setting parallel_workers to 0 via ALTER TABLE will
481 disable parallel index builds on the table in all cases.
485 You might want to reset parallel_workers after setting it as part of
486 tuning an index build. This avoids inadvertent changes to query plans,
487 since parallel_workers affects all parallel table scans.
489 While CREATE INDEX with the CONCURRENTLY option supports parallel
490 builds without special restrictions, only the first table scan is
491 actually performed in parallel.
493 Use DROP INDEX to remove an index.
495 Like any long-running transaction, CREATE INDEX on a table can affect
496 which tuples can be removed by concurrent VACUUM on any other table.
498 Prior releases of PostgreSQL also had an R-tree index method. This
499 method has been removed because it had no significant advantages over
500 the GiST method. If USING rtree is specified, CREATE INDEX will
501 interpret it as USING gist, to simplify conversion of old databases to
504 Each backend running CREATE INDEX will report its progress in the
505 pg_stat_progress_create_index view. See Section 27.4.4 for details.
509 To create a unique B-tree index on the column title in the table films:
510 CREATE UNIQUE INDEX title_idx ON films (title);
512 To create a unique B-tree index on the column title with included
513 columns director and rating in the table films:
514 CREATE UNIQUE INDEX title_idx ON films (title) INCLUDE (director, rating);
516 To create a B-Tree index with deduplication disabled:
517 CREATE INDEX title_idx ON films (title) WITH (deduplicate_items = off);
519 To create an index on the expression lower(title), allowing efficient
520 case-insensitive searches:
521 CREATE INDEX ON films ((lower(title)));
523 (In this example we have chosen to omit the index name, so the system
524 will choose a name, typically films_lower_idx.)
526 To create an index with non-default collation:
527 CREATE INDEX title_idx_german ON films (title COLLATE "de_DE");
529 To create an index with non-default sort ordering of nulls:
530 CREATE INDEX title_idx_nulls_low ON films (title NULLS FIRST);
532 To create an index with non-default fill factor:
533 CREATE UNIQUE INDEX title_idx ON films (title) WITH (fillfactor = 70);
535 To create a GIN index with fast updates disabled:
536 CREATE INDEX gin_idx ON documents_table USING GIN (locations) WITH (fastupdate =
539 To create an index on the column code in the table films and have the
540 index reside in the tablespace indexspace:
541 CREATE INDEX code_idx ON films (code) TABLESPACE indexspace;
543 To create a GiST index on a point attribute so that we can efficiently
544 use box operators on the result of the conversion function:
545 CREATE INDEX pointloc
546 ON points USING gist (box(location,location));
548 WHERE box(location,location) && '(0,0),(1,1)'::box;
550 To create an index without locking out writes to the table:
551 CREATE INDEX CONCURRENTLY sales_quantity_index ON sales_table (quantity);
555 CREATE INDEX is a PostgreSQL language extension. There are no
556 provisions for indexes in the SQL standard.
560 ALTER INDEX, DROP INDEX, REINDEX, Section 27.4.4
begriffs.com / projects / ai-pg / blob