A savepoint is a marker that defines the beginning of a nested transaction. This marker can be later used to commit or roll back just the effects of the nested transaction without affecting the progress of the enclosing transaction.
CockroachDB supports general purpose savepoints for nested transactions, in addition to continued support for special-purpose retry savepoints.
Rollbacks to savepoints over DDL statements are only supported if you're rolling back to a savepoint created at the beginning of the transaction.
No privileges are required to create a savepoint. However, privileges are required for each statement within a transaction.
|name||The name of the savepoint. Nested transactions can use any name for the savepoint. Retry savepoints default to using the name
Savepoints and row locks
CockroachDB supports exclusive row locks.
- In PostgreSQL, row locks are released/cancelled upon
ROLLBACK TO SAVEPOINT.
- In CockroachDB, row locks are preserved upon
ROLLBACK TO SAVEPOINT.
This is an architectural difference in v20.2 that may or may not be lifted in a later CockroachDB version.
The code of client applications that rely on row locks must be reviewed and possibly modified to account for this difference. In particular, if an application is relying on
ROLLBACK TO SAVEPOINT to release row locks and allow a concurrent transaction touching the same rows to proceed, this behavior will not work with CockroachDB.
Savepoints and high priority transactions
ROLLBACK TO SAVEPOINT (for either regular savepoints or "restart savepoints" defined with
cockroach_restart) causes a "feature not supported" error after a DDL statement in a
HIGH PRIORITY transaction, in order to avoid a transaction deadlock. For more information, see GitHub issue #46414.
The examples below use the following table:
> CREATE TABLE kv (k INT PRIMARY KEY, v INT);
To establish a savepoint inside a transaction:
> SAVEPOINT foo;
Due to the rules for identifiers in our SQL grammar,
SAVEPOINT foo and
SAVEPOINT Foo define the same savepoint, whereas
SAVEPOINT "Foo" defines another.
To roll back a transaction partially to a previously established savepoint:
> ROLLBACK TO SAVEPOINT foo;
To forget a savepoint, and keep the effects of statements executed after the savepoint was established, use
> RELEASE SAVEPOINT foo;
For example, the transaction below will insert the values
(3,3) into the table, but not
> BEGIN; INSERT INTO kv VALUES (1,1); SAVEPOINT my_savepoint; INSERT INTO kv VALUES (2,2); ROLLBACK TO SAVEPOINT my_savepoint; INSERT INTO kv VALUES (3,3); COMMIT;
Savepoints for nested transactions
Transactions can be nested using named savepoints.
RELEASE SAVEPOINT and
ROLLBACK TO SAVEPOINT can both refer to a savepoint "higher" in the nesting hierarchy. When this occurs, all of the savepoints "under" the nesting are automatically released / rolled back too. Specifically:
When a previous savepoint is rolled back, the statements entered after that savepoint are also rolled back.
When a previous savepoint is released, it commits; the statements entered after that savepoint are also committed.
For more information about nested transactions, see Nested transactions.
Multi-level rollback with
ROLLBACK TO SAVEPOINT
Savepoints can be arbitrarily nested, and rolled back to the outermost level so that every subsequent statement is rolled back.
For example, this transaction does not insert anything into the table. Both
INSERTs are rolled back:
> BEGIN; SAVEPOINT foo; INSERT INTO kv VALUES (5,5); SAVEPOINT bar; INSERT INTO kv VALUES (6,6); ROLLBACK TO SAVEPOINT foo; COMMIT;
Multi-level commit with
Changes committed by releasing a savepoint commit all of the statements entered after that savepoint.
For example, the following transaction inserts both
(4,4) into the table when it releases the outermost savepoint:
> BEGIN; SAVEPOINT foo; INSERT INTO kv VALUES (2,2); SAVEPOINT bar; INSERT INTO kv VALUES (4,4); RELEASE SAVEPOINT foo; COMMIT;
Multi-level rollback and commit in the same transaction
Changes partially committed by a savepoint release can be rolled back by an outer savepoint.
For example, the following transaction inserts only value
(5, 5). The values
(7,7) are rolled back.
> BEGIN; INSERT INTO kv VALUES (5,5); SAVEPOINT foo; INSERT INTO kv VALUES (6,6); SAVEPOINT bar; INSERT INTO kv VALUES (7,7); RELEASE SAVEPOINT bar; ROLLBACK TO SAVEPOINT foo; COMMIT;
Error recovery in nested transactions with
ROLLBACK TO SAVEPOINT
ROLLBACK TO SAVEPOINT is used after a database error, it can also cancel the error state of the transaction. Database errors move a transaction (or nested transaction) into an "Aborted" state. In this state, the transaction will not execute any further SQL statements.
You can use
ROLLBACK TO SAVEPOINT to recover from a logical error in a nested transaction. Logical errors include:
- Unique index error (duplicate row)
- Failed foreign key constraint check (row does not exist in referenced table)
- Mistakes in queries (reference a column that does not exist)
In addition, you can check the status of a nested transaction using the
SHOW TRANSACTION STATUS statement as shown below.
> BEGIN; SAVEPOINT error1; INSERT INTO kv VALUES (5,5); -- Duplicate key error
ERROR: duplicate key value (k)=(5) violates unique constraint "primary" SQLSTATE: 23505
SHOW TRANSACTION STATUS;
TRANSACTION STATUS ---------------------- Aborted (1 row)
ROLLBACK TO SAVEPOINT error1; INSERT INTO kv VALUES (6,6); COMMIT;
Savepoint name visibility
The name of a savepoint that was rolled back over is no longer visible afterward.
For example, in the transaction below, the name "bar" is not visible after it was rolled back over:
> BEGIN; SAVEPOINT foo; SAVEPOINT bar; ROLLBACK TO SAVEPOINT foo; RELEASE SAVEPOINT bar; COMMIT;
ERROR: savepoint bar does not exist SQLSTATE: 3B001
? ERROR> ROLLBACK;
Savepoints and prepared statements
Prepared statements (
EXECUTE) are not transactional. Therefore, prepared statements are not invalidated upon savepoint rollback. As a result, the prepared statement was saved and executed inside the transaction, despite the rollback to the prior savepoint:
> BEGIN; SAVEPOINT foo; PREPARE bar AS SELECT 1; ROLLBACK TO SAVEPOINT foo; EXECUTE bar; COMMIT;
?column? ------------ 1 (1 row)
Savepoints for client-side transaction retries
The example below shows basic usage of a retry savepoint.
> BEGIN; SAVEPOINT cockroach_restart; UPDATE products SET inventory = 0 WHERE sku = '8675309'; INSERT INTO orders (customer, sku, status) VALUES (1001, '8675309', 'new'); RELEASE SAVEPOINT cockroach_restart; COMMIT;
SAVEPOINT for client-side transaction retries must also include functions to execute retries with
ROLLBACK TO SAVEPOINT.
Note that you can customize the retry savepoint name to something other than
cockroach_restart with a session variable if you need to.
Customizing the retry savepoint name
Once this variable is set, the
SAVEPOINT statement will accept any name for the retry savepoint, not just
cockroach_restart. In addition, it causes every savepoint name to be equivalent to
cockroach_restart, therefore disallowing the use of nested transactions.
This feature exists to support applications that want to use the advanced client-side transaction retry protocol, but cannot customize the name of savepoints to be
cockroach_restart. For example, this may be necessary because you are using an ORM that requires its own names for savepoints.
Showing savepoint status
SHOW SAVEPOINT STATUS statement to see how many savepoints are active in the current transaction:
> SHOW SAVEPOINT STATUS;
savepoint_name | is_initial_savepoint -----------------+----------------------- foo | true bar | false baz | false (3 rows)
Note that the
is_initial_savepoint column will be true if the savepoint is the outermost savepoint in the transaction.