Reusable Views

A view is a stored and named . By default, CockroachDB’s views are dematerialized: they do not store the results of the underlying queries. Instead, the underlying query is executed anew every time the view is used.

CockroachDB also supports materialized views. Materialized views are views that store their selection query results.

By default, views created in a database cannot reference objects in a different database. To enable cross-database references for views, set the sql.cross_db_views.enabled to true.

By default, a view checks privileges on its underlying tables using the view owner’s privileges. A view created or altered with security_invoker checks privileges using the querying user’s privileges instead.

Why use views?

There are various reasons to use views, including:

• Hide query complexity
• Limit access to underlying data
• Use invoker privileges for underlying data

Hide query complexity

When you have a complex query that, for example, joins several tables, or performs complex calculations, you can store the query as a view and then select from the view as you would from a standard table.

Example

Let’s say you’re using our , which contains two tables, episodes and quotes.

> USE startrek;

There’s a foreign key constraint between the episodes.id column and the quotes.episode column. To count the number of famous quotes per season, you could run the following join:

> SELECT episodes.season, count(*)
  FROM quotes
  JOIN episodes
  ON quotes.episode = episodes.id
  GROUP BY episodes.season;

  season | count
---------+--------
       1 |    78
       2 |    76
       3 |    46
(3 rows)

Alternatively, to make it much easier to run this complex query, you could create a view:

> CREATE VIEW quotes_per_season (season, quotes)
  AS SELECT episodes.season, count(*)
  FROM quotes
  JOIN episodes
  ON quotes.episode = episodes.id
  GROUP BY episodes.season
  ORDER BY episodes.season;

Then, executing the query is as easy as SELECTing from the view:

> SELECT * FROM quotes_per_season;

  season | quotes
---------+---------
       1 |     78
       2 |     76
       3 |     46
(3 rows)

Limit access to underlying data

When you do not want to grant a user access to all the data in one or more standard tables, you can create a view that contains only the columns and/or rows that the user should have access to and then grant the user permissions on the view.

This pattern uses the default view behavior: CockroachDB checks access to the underlying tables using the view owner’s privileges.

Example

Let’s say you have a bank database containing an accounts table:

> USE bank;

> SELECT * FROM accounts;

  id |   type   | balance |      email
-----+----------+---------+------------------
   1 | checking |    1000 | [e-mail]
   2 | savings  |   10000 | [e-mail]
   3 | checking |   15000 | [e-mail]
   4 | checking |    5000 | [e-mail]
   5 | savings  |   50000 | [e-mail]
(5 rows)

You want a particular user, bob, to be able to see the types of accounts each user has without seeing the balance in each account, so you create a view to expose just the type and email columns:

> CREATE VIEW user_accounts
  AS SELECT type, email
  FROM accounts;

You then make sure bob does not have privileges on the underlying accounts table:

> SHOW GRANTS ON accounts;

  database_name | schema_name | table_name | grantee | privilege_type
----------------+-------------+------------+---------+-----------------
  bank          | public      | accounts   | admin   | ALL
  bank          | public      | accounts   | root    | ALL
(2 rows)

Finally, you grant bob privileges on the user_accounts view:

> GRANT SELECT ON user_accounts TO bob;

Now, bob will get a permissions error when trying to access the underlying accounts table but will be allowed to query the user_accounts view:

> SELECT * FROM accounts;

pq: user bob does not have SELECT privilege on table accounts

> SELECT * FROM user_accounts;

    type   |      email
-----------+------------------
  checking | [e-mail]
  savings  | [e-mail]
  checking | [e-mail]
  checking | [e-mail]
  savings  | [e-mail]
(5 rows)

Use invoker privileges for underlying data

If you want a view to enforce privileges on the underlying tables for the querying user, create or alter the view with security_invoker. This also makes the view apply policies as the querying user instead of the view owner.

CREATE VIEW user_accounts_invoker WITH (security_invoker) AS
  SELECT type, email
  FROM accounts;

To enable this behavior on an existing view, use .

How views work

Creating views

To create a view, use the statement:

> CREATE VIEW quotes_per_season (season, quotes)
  AS SELECT episodes.season, count(*)
  FROM quotes
  JOIN episodes
  ON quotes.episode = episodes.id
  GROUP BY episodes.season
  ORDER BY episodes.season;

Any is valid as operand to CREATE VIEW, not just .

Listing views

Once created, views are listed alongside regular tables in the database:

> SHOW TABLES FROM startrek;

  schema_name |    table_name     | type  | estimated_row_count
--------------+-------------------+-------+----------------------
  public      | episodes          | table |                  79
  public      | quotes            | table |                 200
  public      | quotes_per_season | view  |                   0
(3 rows)

To list just views, you can query the views table in the :

> SELECT * FROM information_schema.views;

  table_catalog | table_schema |    table_name     |                                                                               view_definition                                                                                | check_option | is_updatable | is_insertable_into | is_trigger_updatable | is_trigger_deletable | is_trigger_insertable_into
----------------+--------------+-------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+--------------+--------------+--------------------+----------------------+----------------------+-----------------------------
  startrek      | public       | quotes_per_season | SELECT episodes.season, count(*) FROM startrek.public.quotes JOIN startrek.public.episodes ON quotes.episode = episodes.id GROUP BY episodes.season ORDER BY episodes.season | NULL         | NO           | NO                 | NO                   | NO                   | NO
(1 row)

Query a view

To query a view, target it with a , for example using a , just as you would with a stored table:

> SELECT * FROM quotes_per_season;

  season | quotes
---------+---------
       1 |     78
       2 |     76
       3 |     46
(3 rows)

SELECTing a view executes the view’s stored SELECT statement, which returns the relevant data from the underlying table(s). To inspect the SELECT statement executed by the view, use the statement:

> SHOW CREATE quotes_per_season;

     table_name     |                                                                                                           create_statement
--------------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  quotes_per_season | CREATE VIEW public.quotes_per_season (season, quotes) AS SELECT episodes.season, count(*) FROM startrek.public.quotes JOIN startrek.public.episodes ON quotes.episode = episodes.id GROUP BY episodes.season ORDER BY episodes.season
(1 row)

You can also inspect the SELECT statement executed by a view by querying the views table in the :

> SELECT view_definition FROM information_schema.views WHERE table_name = 'quotes_per_season';

                                                                                view_definition
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  SELECT episodes.season, count(*) FROM startrek.public.quotes JOIN startrek.public.episodes ON quotes.episode = episodes.id GROUP BY episodes.season ORDER BY episodes.season
(1 row)

View dependencies

A view depends on the objects targeted by its underlying query. Attempting to referenced in a view’s stored query therefore results in an error:

> ALTER TABLE quotes RENAME TO sayings;

ERROR: cannot rename relation "startrek.public.quotes" because view "quotes_per_season" depends on it
SQLSTATE: 2BP01
HINT: you can drop quotes_per_season instead.

Likewise, attempting to referenced in a view’s stored query results in an error:

> DROP TABLE quotes;

ERROR: cannot drop relation "quotes" because view "quotes_per_season" depends on it
SQLSTATE: 2BP01
HINT: you can drop quotes_per_season instead.

> ALTER TABLE episodes DROP COLUMN season;

ERROR: cannot drop column "season" because view "quotes_per_season" depends on it
SQLSTATE: 2BP01
HINT: you can drop quotes_per_season instead.

You can or from a table on which a view is dependent, as long as the view does not depend on that column of the table. For example, because there is no view that depends on the num column of the episodes table, you can rename it to number:

> ALTER TABLE startrek.episodes RENAME COLUMN num TO number;

Similarly, because no view depends on the title column of the episodes table, you can drop it. Note that to drop a column with data in it, you must first set sql_safe_updates = false.

> SET sql_safe_updates = false;

> ALTER TABLE startrek.episodes DROP COLUMN title;

> SHOW COLUMNS FROM startrek.episodes;

  column_name | data_type | is_nullable | column_default | generation_expression |  indices  | is_hidden
--------------+-----------+-------------+----------------+-----------------------+-----------+------------
  id          | INT8      |    false    | NULL           |                       | {primary} |   false
  season      | INT8      |    true     | NULL           |                       | {primary} |   false
  number      | INT8      |    true     | NULL           |                       | {primary} |   false
  stardate    | DECIMAL   |    true     | NULL           |                       | {primary} |   false
(4 rows)

When or , you can use the CASCADE keyword to drop all dependent objects as well:

> DROP TABLE quotes CASCADE;

DROP TABLE

CASCADE drops all dependent objects without listing them, which can lead to inadvertent and difficult-to-recover losses. To avoid potential harm, we recommend dropping objects individually in most cases.

Renaming views

To rename a view, use the statement:

> ALTER VIEW quotes_per_season RENAME TO quotes_count;

RENAME VIEW

It is not possible to change the stored query executed by the view. Instead, you must drop the existing view and create a new view.

Replace a view

To replace a view, use :

> CREATE OR REPLACE VIEW quotes_count (season, quotes, stardate)
  AS SELECT episodes.season, count(*), episodes.stardate
  FROM quotes
  JOIN episodes
  ON quotes.episode = episodes.id
  GROUP BY episodes.season, episodes.stardate;

> SELECT * FROM quotes_count LIMIT 10;

  season | quotes | stardate
---------+--------+-----------
       1 |      5 |   1709.2
       1 |      7 |   2821.5
       1 |      2 |   3113.2
       1 |      6 |     3134
       1 |      3 |   2715.1
       1 |      7 |   3012.4
       1 |      2 |   3196.1
       2 |      4 |   3468.1
       2 |      1 |   3541.9
       2 |      5 |   4211.4
(10 rows)

Remove a view

To remove a view, use the statement:

> DROP VIEW quotes_count;

DROP VIEW

Materialized views

CockroachDB supports materialized views. A materialized view is a view that stores the results of its underlying query.

When you from a materialized view, the stored query data that is returned might be out-of-date. This contrasts with a standard (i.e., “dematerialized”) view, which runs its underlying query every time it is used, returning the latest results. In order to get the latest results from a materialized view, you must , and then select from it.

Because materialized views store query results, they offer better performance than standard views, at the expense of the additional storage required to store query results and the guarantee that the results are up-to-date.

Usage

Materialized views and standard views share similar syntax for , , , and .

To create a materialized view, use a statement.

For example, suppose that you have the loaded to a CockroachDB cluster, and populated with some workload values:

> CREATE MATERIALIZED VIEW overdrawn_accounts
  AS SELECT id, balance
  FROM bank
  WHERE balance < 0;

> SELECT * FROM overdrawn_accounts;

  id  | balance
------+----------
    1 |  -17643
    3 |   -5928
   13 |   -3700
...
(402 rows)

To show existing materialized views, use a statement:

> SHOW TABLES;

  schema_name |     table_name     |       type        | estimated_row_count
--------------+--------------------+-------------------+----------------------
  public      | bank               | table             |                1000
  public      | overdrawn_accounts | materialized view |                   0
(2 rows)

Now suppose you update the balance values of the bank table:

> UPDATE bank SET balance = 0 WHERE balance < 0;

UPDATE 402

The changes can be seen in the table with a simple SELECT statement against the table:

> SELECT id, balance
FROM bank
WHERE balance < 0;

  id | balance
-----+----------
(0 rows)

Recall that materialized views do not automatically update their stored results. Selecting from overdrawn_accounts returns stored results, which are outdated:

> SELECT * FROM overdrawn_accounts;

  id  | balance
------+----------
    1 |  -17643
    3 |   -5928
   13 |   -3700
...
(402 rows)

To update the materialized view’s results, use a statement:

> REFRESH MATERIALIZED VIEW overdrawn_accounts;

> SELECT * FROM overdrawn_accounts;

  id | balance
-----+----------
(0 rows)

You can also create or refresh materialized views using historical data with the clause. This is useful for reducing by leveraging .

> CREATE MATERIALIZED VIEW overdrawn_accounts
  AS SELECT id, balance
  FROM bank
  WHERE balance < 0
  AS OF SYSTEM TIME follower_read_timestamp();

> REFRESH MATERIALIZED VIEW overdrawn_accounts
  AS OF SYSTEM TIME follower_read_timestamp();

For more information, see and .

To rename the materialized view, use :

> ALTER MATERIALIZED VIEW overdrawn_accounts RENAME TO forgiven_accounts;

RENAME VIEW

To remove the materialized view, use :

> DROP MATERIALIZED VIEW forgiven_accounts;

DROP VIEW

Add an index to a materialized view

To speed up queries on materialized views, you can add an to the view.

1. Create a materialized view of the rides table where the revenue is less than $20.00:

   CREATE MATERIALIZED VIEW low_rev_rides AS SELECT city, vehicle_id, revenue from rides WHERE revenue < 20.00;
   

2. To see the plan for a select on this view, run:

   EXPLAIN SELECT vehicle_id, revenue FROM low_rev_rides WHERE city = 'seattle';
   
                                            info
   ---------------------------------------------------------------------------------------
     distribution: full
     vectorized: true
   
     • filter
     │ filter: city = 'seattle'
     │
     └── • scan
           missing stats
           table: low_rev_rides@low_rev_rides_pkey
           spans: FULL SCAN
   
     index recommendations: 1
     1. type: index creation
        SQL command: CREATE INDEX ON low_rev_rides (city) STORING (vehicle_id, revenue);
   (14 rows)
   

   Notice that there are no statistics collected on the view.

3. Create an index on the city column:

   CREATE INDEX ON low_rev_rides (city) STORING (vehicle_id, revenue);
   

4. To see the change in the plan after adding the index, run:

   EXPLAIN SELECT vehicle_id, revenue FROM low_rev_rides@low_rev_rides_city_idx WHERE city = 'seattle';
   
                         info
   -------------------------------------------------
     distribution: local
     vectorized: true
   
     • scan
       missing stats
       table: low_rev_rides@low_rev_rides_city_idx
       spans: [/'seattle' - /'seattle']
   (7 rows)
   

Temporary views

CockroachDB supports session-scoped temporary views. Unlike persistent views, temporary views can only be accessed from the session in which they were created, and they are dropped at the end of the session. You can create temporary views on both persistent tables and .

This feature is in and subject to change. To share feedback and/or issues, contact Support.

Temporary tables must be enabled in order to use temporary views. By default, temporary tables are disabled in CockroachDB. To enable temporary tables, set the experimental_enable_temp_tables to on.

Details

• Temporary views are automatically dropped at the end of the session.
• A temporary view can only be accessed from the session in which it was created.
• Temporary views persist across transactions in the same session.
• Temporary views cannot be converted to persistent views.
• Temporary views can be created on both persistent tables and .
• When you create a view on a temporary table, the view automatically becomes temporary.

Like , temporary views are not in the public schema. Instead, when you create the first temporary table, view, or sequence for a session, CockroachDB generates a single temporary schema (pg_temp_<id) for all of the temporary objects in the current session for a database.

Usage

To create a temporary view, add to a statement.

For example:

> CREATE TEMP VIEW temp_view (season, quotes)
  AS SELECT episodes.season, count(*)
  FROM quotes
  JOIN episodes
  ON quotes.episode = episodes.id
  GROUP BY episodes.season;

> SELECT * FROM temp_view;

  season | quotes
---------+---------
       1 |     78
       2 |     76
       3 |     46
(3 rows)

Known limitations

• Casting subqueries to (including ENUMs) in views is not supported.

• Statements within views do not currently respect hint injections. The workaround is to modify the inline hints directly in the body by replacing the view.

• The optimizer may not select the most optimal query plan when querying materialized views because CockroachDB does not on materialized views..

• CockroachDB cannot refresh materialized views inside . Trying to refresh a materialized view inside an explicit transaction will result in an error.

  1. Start with the sample bank data set:

     cockroach demo bank
     

  2. Create the materialized view described in .

  3. Start a new multi-statement transaction with :

     BEGIN TRANSACTION;
     

  4. Inside the open transaction, attempt to . This will result in an error.

     REFRESH MATERIALIZED VIEW overdrawn_accounts;
     
     ERROR: cannot refresh view in an explicit transaction
     SQLSTATE: 25000
     

See also