CREATE INDEX

The CREATE INDEX creates an index for a table. improve your database’s performance by helping SQL locate data without having to look through every row of a table.

Indexes are automatically created for a table’s and columns. When querying a table, CockroachDB uses the fastest index. For more information about that process, see Index Selection in CockroachDB.

The following types cannot be included in an index key, but can be stored (and used in a covered query) using the clause:

• The computed type, even if it is constructed from indexed fields

To create an index on the schemaless data in a column or on the data in an , use a .

The CREATE INDEX statement performs a schema change. For more information about how online schema changes work in CockroachDB, see .

Required privileges

The user must have the CREATE on the table.

Synopsis

Standard index

GIN index

Parameters

Parameter	Description
UNIQUE	Apply the to the indexed columns. This causes the system to check for existing duplicate values on index creation. It also applies the UNIQUE constraint at the table level, so the system checks for duplicate values when inserting or updating data.
INVERTED	Create a on the schemaless data in the specified column. You can also use the PostgreSQL-compatible syntax USING GIN. For more details, see .
VECTOR	Create a on the specifed column. For more details, refer to .
IF NOT EXISTS	Create a new index only if an index of the same name does not already exist; if one does exist, do not return an error.
opt\_index\_name index\_name	The name of the index to create, which must be unique to its table and follow these . If you do not specify a name, CockroachDB uses the format \_\_key/idx. key indicates the index applies the UNIQUE constraint; idx indicates it does not. Example: accounts\_balance\_idx
table\_name	The name of the table you want to create the index on.
USING name	An optional clause for compatibility with third-party tools. Accepted values for name are btree, gin, and gist, with btree for a standard secondary index, gin as the PostgreSQL-compatible syntax for a GIN index, gist for a , and cspann for a . hnsw is aliased to cspann for compatibility with [pgvector](https://github. com/pgvector/pgvector) syntax.
name	The name of the column you want to index. For , you can use the crdb\_region column within the index in the event the original index may contain non-unique entries across multiple, unique regions.
ASC or DESC	Sort the column in ascending (ASC) or descending (DESC) order in the index. How columns are sorted affects query results, particularly when using LIMIT. Default: ASC
STORING...	Store (but do not sort) each column whose name you include. For information on when to use STORING, see Store Columns. Note that columns that are part of a table’s cannot be specified as STORING columns in secondary indexes on the table. COVERING and INCLUDE are aliases for STORING and work identically.
opt\_partition\_by	An option that lets you . As of CockroachDB v21.1 and later, most users should use . Indexes against regional by row tables are automatically partitioned, so explicit index partitioning is not required.
opt\_where\_clause	An optional WHERE clause that defines the predicate boolean expression of a .
opt\_index\_visible	An optional VISIBLE, NOT VISIBLE, or VISIBILITY clause that indicates that an . If not visible, the index will not be used in queries unless it is specifically selected with an or the property is overridden with the . For examples, see . Indexes that are not visible are still used to enforce UNIQUE and FOREIGN KEY . For more considerations, see .
USING HASH	Creates a .
WITH storage\_parameter	A comma-separated list of . Supported parameters include fillfactor, s2\_max\_level, s2\_level\_mod, s2\_max\_cells, geometry\_min\_x, geometry\_max\_x, geometry\_min\_y, and geometry\_max\_y. The fillfactor parameter is a no-op, allowed for PostgreSQL-compatibility. For details, see . For an example, see .
CONCURRENTLY	Optional, no-op syntax for PostgreSQL compatibility. All indexes are created concurrently in CockroachDB.

Viewing schema changes

This schema change statement is registered as a job. You can view long-running jobs with .

Examples

Setup

To follow along, run to start a temporary, in-memory cluster with the sample dataset preloaded:

$ cockroach demo

Create standard indexes

To create the most efficient indexes, we recommend reviewing:

• Index Selection in CockroachDB

Single-column indexes

Single-column indexes sort the values of a single column.

> CREATE INDEX ON users (name);

Because each query can only use one index, single-column indexes are not typically as useful as multiple-column indexes.

Multiple-column indexes

Multiple-column indexes sort columns in the order you list them.

> CREATE INDEX ON users (name, city);

To create the most useful multiple-column indexes, we recommend reviewing our .

Unique indexes

Unique indexes do not allow duplicate values among their columns.

> CREATE UNIQUE INDEX ON users (name, id);

This also applies the at the table level, similar to . The preceding example is equivalent to:

> ALTER TABLE users ADD CONSTRAINT users_name_id_key UNIQUE (name, id);

Primary key columns that are not specified within a unique index are automatically marked as in the table and in .

Create GIN indexes

You can create on schemaless data in a column.

> CREATE INDEX ON promo_codes USING GIN (rules);

The following syntax is equivalent:

> CREATE INVERTED INDEX ON promo_codes (rules);

Create trigram indexes

You can create on STRING columns by specifying the gin_trgm_ops or gist_trgm_ops opclass:

CREATE INDEX ON rides USING GIN (vehicle_city gin_trgm_ops);

The following syntax is equivalent:

CREATE INVERTED INDEX ON rides(vehicle_city gin_trgm_ops);

GIN and GiST indexes are implemented identically on CockroachDB. GIN and GIST are therefore synonymous when defining a trigram index.

Create spatial indexes

You can create on GEOMETRY and GEOGRAPHY columns. Spatial indexes are a special type of .

To create a spatial index on a GEOMETRY column:

CREATE INDEX geom_idx_1 ON some_spatial_table USING GIST(geom);

Unlike GIN indexes, spatial indexes do not support an alternate CREATE INVERTED INDEX... syntax. Only the syntax shown here is supported.

For advanced users, there are a number of that can be passed in using the syntax WITH (var1=val1, var2=val2) as follows:

CREATE INDEX geom_idx_2
  ON some_spatial_table USING GIST(geom)
  WITH (s2_max_cells = 20, s2_max_level = 12, s2_level_mod = 3);

Most users should not change the default spatial index settings. There is a risk that you will get worse performance by changing the default settings. For more information, see .

Create vector indexes

You can create on columns.

To create a vector index on a VECTOR column named embedding:

CREATE VECTOR INDEX ON items (embedding);

Store columns

Storing a column improves the performance of queries that retrieve (but do not filter) its values.

> CREATE INDEX ON users (city) STORING (name);

However, to use stored columns, queries must filter another column in the same index. For example, SQL can retrieve name values from the above index only when a query’s WHERE clause filters city.

An index that stores all the columns needed by a query is also known as a covering index for that query. When a query has a covering index, CockroachDB can use that index directly instead of doing an “index join” with the primary index, which is likely to be slower.

Change column sort order

To sort columns in descending order, you must explicitly set the option when creating the index. (Ascending order is the default.)

> CREATE INDEX ON users (city DESC, name);

How a column is ordered in the index will affect the ordering of the index keys, and may affect the efficiency of queries that include an ORDER BY clause.

Query specific indexes

Normally, CockroachDB selects the index that it calculates will scan the fewest rows. However, you can override that selection and specify the name of the index you want to use. To find the name, use .

> SHOW INDEX FROM users;

  table_name |   index_name        | non_unique | seq_in_index | column_name | direction | storing | implicit
+------------+---------------------+------------+--------------+-------------+-----------+---------+----------+
  users      | users_pkey          |   false    |            1 | city        | ASC       |  false  |  false
  users      | users_pkey          |   false    |            2 | id          | ASC       |  false  |  false
  users      | users_pkey          |   false    |            3 | name        | N/A       |  true   |  false
  users      | users_pkey          |   false    |            4 | address     | N/A       |  true   |  false
  users      | users_pkey          |   false    |            5 | credit_card | N/A       |  true   |  false
  users      | users_city_name_idx |    true    |            1 | city        | DESC      |  false  |  false
  users      | users_city_name_idx |    true    |            2 | name        | ASC       |  false  |  false
  users      | users_city_name_idx |    true    |            3 | id          | ASC       |  false  |   true
(8 rows)

> SELECT name FROM users@users_name_idx WHERE city='new york';

        name
+------------------+
  Catherine Nelson
  Devin Jordan
  James Hamilton
  Judy White
  Robert Murphy
(5 rows)

You can use the @primary alias to use the table’s primary key in your query if no secondary index explicitly named primary exists on that table.

Create a hash-sharded secondary index

We . If a table must be indexed on sequential keys, use . Hash-sharded indexes distribute sequential traffic uniformly across ranges, eliminating single-range and improving write performance on sequentially-keyed indexes at a small cost to read performance.

Let’s assume the events table already exists:

> CREATE TABLE events (
    product_id INT8,
    owner UUID,
    serial_number VARCHAR,
    event_id UUID,
    ts TIMESTAMP,
    data JSONB,
    PRIMARY KEY (product_id, owner, serial_number, ts, event_id)
);

You can create a hash-sharded index on an existing table:

> CREATE INDEX ON events(ts) USING HASH;

> SHOW INDEX FROM events;

  table_name |  index_name   | non_unique | seq_in_index |        column_name        | direction | storing | implicit
-------------+---------------+------------+--------------+---------------------------+-----------+---------+-----------
  events     | events_pkey   |   false    |            1 | product_id                | ASC       |  false  |  false
  events     | events_pkey   |   false    |            2 | owner                     | ASC       |  false  |  false
  events     | events_pkey   |   false    |            3 | serial_number             | ASC       |  false  |  false
  events     | events_pkey   |   false    |            4 | ts                        | ASC       |  false  |  false
  events     | events_pkey   |   false    |            5 | event_id                  | ASC       |  false  |  false
  events     | events_pkey   |   false    |            6 | data                      | N/A       |  true   |  false
  events     | events_ts_idx |    true    |            1 | crdb_internal_ts_shard_16 | ASC       |  false  |   true
  events     | events_ts_idx |    true    |            2 | ts                        | ASC       |  false  |  false
  events     | events_ts_idx |    true    |            3 | product_id                | ASC       |  false  |   true
  events     | events_ts_idx |    true    |            4 | owner                     | ASC       |  false  |   true
  events     | events_ts_idx |    true    |            5 | serial_number             | ASC       |  false  |   true
  events     | events_ts_idx |    true    |            6 | event_id                  | ASC       |  false  |   true
(12 rows)

> SHOW COLUMNS FROM events;

         column_name        | data_type | is_nullable | column_default |               generation_expression               |           indices           | is_hidden
----------------------------+-----------+-------------+----------------+---------------------------------------------------+-----------------------------+------------
  product_id                | INT8      |    false    | NULL           |                                                   | {events_pkey,events_ts_idx} |   false
  owner                     | UUID      |    false    | NULL           |                                                   | {events_pkey,events_ts_idx} |   false
  serial_number             | VARCHAR   |    false    | NULL           |                                                   | {events_pkey,events_ts_idx} |   false
  event_id                  | UUID      |    false    | NULL           |                                                   | {events_pkey,events_ts_idx} |   false
  ts                        | TIMESTAMP |    false    | NULL           |                                                   | {events_pkey,events_ts_idx} |   false
  data                      | JSONB     |    true     | NULL           |                                                   | {events_pkey}               |   false
  crdb_internal_ts_shard_16 | INT8      |    false    | NULL           | mod(fnv32(crdb_internal.datums_to_bytes(ts)), 16) | {events_ts_idx}             |   true
(7 rows)

See also