ALTER TABLE
Synopsis
Use the ALTER TABLE
statement to change the definition of a table.
Syntax
alter_table ::= ALTER TABLE [ IF EXISTS ] table_expr
alter_table_action [ , ... ]
alter_table_action ::= ADD [ COLUMN ] [ IF NOT EXISTS ] column_name
data_type [ alter_column_constraint [ ... ] ]
| RENAME TO table_name
| DROP [ COLUMN ] [ IF EXISTS ] column_name
[ RESTRICT | CASCADE ]
| ALTER [ COLUMN ] column_name [ SET DATA ]
TYPE data_type [ COLLATE collation ]
[ USING expression ]
| ADD alter_table_constraint
| DROP CONSTRAINT constraint_name
[ RESTRICT | CASCADE ]
| RENAME [ COLUMN ] column_name TO column_name
| RENAME CONSTRAINT constraint_name TO
constraint_name
| DISABLE ROW LEVEL SECURITY
| ENABLE ROW LEVEL SECURITY
| FORCE ROW LEVEL SECURITY
| SET TABLESPACE tablespace_name
| NO FORCE ROW LEVEL SECURITY
alter_table_constraint ::= [ CONSTRAINT constraint_name ]
{ CHECK ( expression )
| UNIQUE ( column_names )
index_parameters
| FOREIGN KEY ( column_names )
references_clause }
[ DEFERRABLE | NOT DEFERRABLE ]
[ INITIALLY DEFERRED
| INITIALLY IMMEDIATE ]
alter_column_constraint ::= [ CONSTRAINT constraint_name ]
{ NOT NULL
| NULL
| CHECK ( expression )
| DEFAULT expression
| UNIQUE index_parameters
| references_clause }
[ DEFERRABLE | NOT DEFERRABLE ]
[ INITIALLY DEFERRED
| INITIALLY IMMEDIATE ]
table_expr ::= [ ONLY ] table_name [ * ]
alter_table
alter_table_action
alter_table_constraint
alter_column_constraint
table_expr
Table inheritance is not yet supported
YSQL in the present "latest" YugabyteDB does not yet support the "table inheritance" feature that is described in the PostgreSQL documentation. The attempt to create a table that inherits another table causes the 0A000 (feature_not_supported) error with the message "INHERITS not supported yet". This means that the syntax that the table_expr
rule allows doesn't not yet bring any useful meaning.
It says that you can write, for example, this:
alter table t * add column y text;
or this:
alter table only t add column y text;
These variants are useful only when at least one other table inherits t
. But as yet, no table can inherit t
. This means that if the unadorned variant alter table t...
runs without error, then each of these variants will run without error too. But the effect of each is the same as that of the unadorned variant. Until inheritance is supported, use a bare table_name.
Semantics
alter_table_action
Specify one of the following actions.
ADD [ COLUMN ] [ IF NOT EXISTS ] column_name data_type constraint
Add the specified column with the specified data type and constraint.
RENAME TO table_name
Rename the table to the specified table name.
Note
Renaming a table is a non blocking metadata change operation.SET TABLESPACE tablespace_name
Asynchronously change the tablespace of an existing table. The tablespace change will immediately reflect in the config of the table, however the tablet move by the load balancer happens in the background. While the load balancer is performing the move it is perfectly safe from a correctness perspective to do reads and writes, however some query optimization that happens based on the data location may be off while data is being moved.
Example
yugabyte=# ALTER TABLE bank_transactions_eu SET TABLESPACE eu_central_1_tablespace;
NOTICE: Data movement for table bank_transactions_eu is successfully initiated.
DETAIL: Data movement is a long running asynchronous process and can be monitored by checking the tablet placement in http://<YB-Master-host>:7000/tables
ALTER TABLE
Tables can be moved to the default tablespace using:
ALTER TABLE table_name SET TABLESPACE pg_default;
DROP [ COLUMN ] [ IF EXISTS ] column_name [ RESTRICT | CASCADE ]
Drop the named column from the table.
RESTRICT
— Remove only the specified column.CASCADE
— Remove the specified column and any dependent objects.
Example
Set up and populate a parents-children pair of tables:
drop table if exists children cascade;
drop table if exists parents cascade;
-- The column "b" models a (natural) business unique key.
create table parents(
k int primary key,
b int not null,
v text not null,
constraint parents_b_unq unique(b));
create table children(
parents_b int not null,
k int not null,
v text not null,
constraint children_pk primary key(parents_b, k),
constraint children_fk foreign key(parents_b)
references parents(b)
match full
on delete cascade
on update restrict);
insert into parents(k, b, v) values (1, 10, 'dog'), (2, 20, 'cat'), (3, 30, 'frog');
insert into children(parents_b, k, v) values
(10, 1, 'dog-child-a'),
(10, 2, 'dog-child-b'),
(10, 3, 'dog-child-c'),
(20, 1, 'cat-child-a'),
(20, 2, 'cat-child-b'),
(20, 3, 'cat-child-c'),
(30, 1, 'frog-child-a'),
(30, 2, 'frog-child-b'),
(30, 3, 'frog-child-c');
select p.v as "p.v", c.v as "c.v"
from parents p inner join children c on c.parents_b = p.b
order by p.b, c.k;
This is the result:
p.v | c.v
------+--------------
dog | dog-child-a
dog | dog-child-b
dog | dog-child-c
cat | cat-child-a
cat | cat-child-b
cat | cat-child-c
frog | frog-child-a
frog | frog-child-b
frog | frog-child-c
The \d children
meta-command shows that it has a foreign key that's a dependent object on the column b
in the parents
table:
Indexes:
"children_pk" PRIMARY KEY, lsm (parents_b HASH, k ASC)
Foreign-key constraints:
"children_fk" FOREIGN KEY (parents_b) REFERENCES parents(b) MATCH FULL ON UPDATE RESTRICT ON DELETE CASCADE
This is a contrived example. It is unusual practice (and normally bad practice) to make a foreign key constraint target anything but the column list upon which the parent table's primary key constraint is defined. But there are sometimes defensible reasons to do this.
Now try to drop the column parents.b
:
do $body$
declare
message text not null := '';
detail text not null := '';
begin
-- Causes error 'cos "cascade" is required.
alter table parents drop column b;
assert false, 'Should not get here';
exception
-- Error 2BP01
when dependent_objects_still_exist then
get stacked diagnostics
message = message_text,
detail = pg_exception_detail;
assert message = 'cannot drop column b of table parents because other objects depend on it', 'Bad message';
assert detail = 'constraint children_fk on table children depends on column b of table parents', 'Bad detail';
end;
$body$;
It finishes without error, showing that the bare alter table parents drop column b
, without cascade
, fails and causes the message and hint that the code presents. Now repeat the attempt with cascade
and observe the result:
alter table parents drop column b cascade;
It quietly succeeds. Now \d children
shows that the foreign key constraint children_fk
has been transitively dropped.
ADD alter_table_constraint
Add the specified constraint to the table.
PRIMARY KEY
constraint results in a full table rewrite and full rewrite of all indexes associated with the table.
This happens because of the clustered storage by primary key that YugabyteDB uses to store rows and indexes.
Tables without a PRIMARY KEY
have a hidden one underneath and rows are stored clustered on it. The secondary indexes of the table
link to this hidden PRIMARY KEY
.
While the tables and indexes are being rewritten, you may lose any modifications made to the table.
For reference, the same semantics as Alter type with table rewrite apply.
[alter_column_type]
Change the type of an existing column. The following semantics apply:
- If data on disk is required to change, a full table rewrite is needed.
- If the optional
COLLATE
clause is not specified, the default collation for the new column type will be used. - If the optional
USING
clause is not provided, the default conversion for the new column value will be the same as an assignment cast from the old type to the new type. - A
USING
clause must be included when there is no implicit assignment cast available from the old type to the new type. - Alter type is not supported for partitioned tables. See #16980.
- Alter type is not supported for tables with rules (limitation inherited from PostgreSQL).
- Alter type is not supported for tables with CDC streams, or xCluster replication when it requires data on disk to change. See #16625.
Alter type without table-rewrite
If the change doesn't require data on disk to change, concurrent DMLs to the table can be safely performed as shown in the following example:
CREATE TABLE test (id BIGSERIAL PRIMARY KEY, a VARCHAR(50));
ALTER TABLE test ALTER COLUMN a TYPE VARCHAR(51);
Alter type with table rewrite
If the change requires data on disk to change, a full table rewrite will be done and the following semantics apply:
- The action creates an entirely new table under the hood, and concurrent DMLs may not be reflected in the new table which can lead to correctness issues.
- If the operation fails, it is possible that the existing table is renamed in DocDB. This may lead to issues with yb-admin commands that take table name. For example,
./bin/yb-admin list_tablets
. - If the operation fails, a new dangling table may exist in DocDB. Use
yb-admin delete_table
to drop it. - Altering the data type of a foreign key column is not supported.
- If there are concurrent DMLs, you can first rename the table to fail the DMLs, alter the column data type, and then rename the table again.
- The operation preserves split properties for hash-partitioned tables and hash-partitioned secondary indexes. For range-partitioned tables (and secondary indexes), split properties are only preserved if the altered column is not part of the table's (or secondary index's) range key.
Following is an example of alter type with table rewrite:
CREATE TABLE test (id BIGSERIAL PRIMARY KEY, a VARCHAR(50));
INSERT INTO test(a) VALUES ('1234555');
ALTER TABLE test ALTER COLUMN a TYPE VARCHAR(40);
-- try to change type to BIGINT
ALTER TABLE test ALTER COLUMN a TYPE BIGINT;
ERROR: column "a" cannot be cast automatically to type bigint
HINT: You might need to specify "USING a::bigint".
-- use USING clause to cast the values
ALTER TABLE test ALTER COLUMN a SET DATA TYPE BIGINT USING a::BIGINT;
Another option is to use a custom function as follows:
CREATE OR REPLACE FUNCTION myfunc(text) RETURNS BIGINT
AS 'select $1::BIGINT;'
LANGUAGE SQL
IMMUTABLE
RETURNS NULL ON NULL INPUT;
ALTER TABLE test ALTER COLUMN a SET DATA TYPE BIGINT USING myfunc(a);
DROP CONSTRAINT constraint_name [ RESTRICT | CASCADE ]
Drop the named constraint from the table.
RESTRICT
— Remove only the specified constraint.CASCADE
— Remove the specified constraint and any dependent objects.
PRIMARY KEY
constraint results in a full table rewrite and full rewrite of all indexes associated with the table.
This happens because of the clustered storage by primary key that YugabyteDB uses to store rows and indexes.
While the tables and indexes are being rewritten, you may lose any modifications made to the table.
For reference, the same semantics as Alter type with table rewrite apply.
RENAME [ COLUMN ] column_name TO column_name
Rename a column to the specified name.
RENAME CONSTRAINT constraint_name TO constraint_name
Rename a constraint to the specified name.
Example
Create a table with a constraint and rename the constraint:
CREATE TABLE test(id BIGSERIAL PRIMARY KEY, a TEXT);
ALTER TABLE test ADD constraint vague_name unique (a);
ALTER TABLE test RENAME CONSTRAINT vague_name TO unique_a_constraint;
ENABLE / DISABLE ROW LEVEL SECURITY
This enables or disables row level security for the table. If enabled and no policies exist for the table, then a default-deny policy is applied. If disabled, then existing policies for the table will not be applied and will be ignored. See CREATE POLICY for details on how to create row level security policies.
FORCE / NO FORCE ROW LEVEL SECURITY
This controls the application of row security policies for the table when the user is the table owner. If enabled, row level security policies will be applied when the user is the table owner. If disabled (the default) then row level security will not be applied when the user is the table owner. See CREATE POLICY for details on how to create row level security policies.
Constraints
Specify a table or column constraint.
CONSTRAINT constraint_name
Specify the name of the constraint.
Foreign key
FOREIGN KEY
and REFERENCES
specify that the set of columns can only contain values that are present in the referenced columns of the referenced table. It is used to enforce referential integrity of data.
Unique
This enforces that the set of columns specified in the UNIQUE
constraint are unique in the table, that is, no two rows can have the same values for the set of columns specified in the UNIQUE
constraint.
Check
This is used to enforce that data in the specified table meets the requirements specified in the CHECK
clause.
Default
This is used to specify a default value for the column. If an INSERT
statement does not specify a value for the column, then the default value is used. If no default is specified for a column, then the default is NULL.
Deferrable constraints
Constraints can be deferred using the DEFERRABLE
clause. Currently, only foreign key constraints
can be deferred in YugabyteDB. A constraint that is not deferrable will be checked after every row
within a statement. In the case of deferrable constraints, the checking of the constraint can be postponed
until the end of the transaction.
Constraints marked as INITIALLY IMMEDIATE
will be checked after every row within a statement.
Constraints marked as INITIALLY DEFERRED
will be checked at the end of the transaction.