Savepoints - Can My DB?

Database	Status	Syntax Family	Min Version
PostgreSQL	full	sql-standard	-
MySQL	full	sql-standard	-
MariaDB	full	sql-standard	-
Oracle	full	sql-standard	-
SQL Server	full	tsql-save	-
SQLite	full	sql-standard	-
Redshift	none	-	-
DB2	full	sql-standard	-
Snowflake	none	-	-
BigQuery	none	-	-

Compare Databases

Database Details

PostgreSQL full

BEGIN;
INSERT INTO orders (id, total) VALUES (1, 100);
SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 999);
-- Oops, rollback just the item insert
ROLLBACK TO SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 42);
COMMIT;

Reusing a savepoint name replaces the previous marker of that name.
Required for error recovery in PL/pgSQL: any error aborts the transaction otherwise, and BEGIN ... EXCEPTION blocks are built on savepoints under the hood.

Docs: SAVEPOINT

MySQL full

START TRANSACTION;
INSERT INTO orders (id, total) VALUES (1, 100);
SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 999);
ROLLBACK TO SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 42);
COMMIT;

InnoDB (and other transactional engines) only; MyISAM has no savepoints.
DDL causes an implicit commit, which discards any open savepoints.

Docs: SAVEPOINT

MariaDB full

START TRANSACTION;
INSERT INTO orders (id, total) VALUES (1, 100);
SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 999);
ROLLBACK TO SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 42);
COMMIT;

InnoDB only. DDL implicit-commits and discards savepoints, same as MySQL.

Docs: SAVEPOINT

Oracle full

INSERT INTO orders (id, total) VALUES (1, 100);
SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 999);
ROLLBACK TO SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 42);
COMMIT;

No RELEASE SAVEPOINT; savepoints release implicitly at COMMIT.
Transactions start implicitly on the first statement, so there is no BEGIN keyword.

Docs: SAVEPOINT

SQL Server full

BEGIN TRANSACTION;
INSERT INTO orders (id, total) VALUES (1, 100);
SAVE TRANSACTION sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 999);
ROLLBACK TRANSACTION sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 42);
COMMIT;

Syntax is SAVE TRANSACTION <name> and ROLLBACK TRANSACTION <name>, not the SQL-standard SAVEPOINT / ROLLBACK TO.
No RELEASE equivalent; markers are dropped at COMMIT.

Docs: SAVE TRANSACTION

SQLite full

BEGIN;
INSERT INTO orders (id, total) VALUES (1, 100);
SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 999);
ROLLBACK TO SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 42);
RELEASE SAVEPOINT sp1;
COMMIT;

A SAVEPOINT outside an explicit BEGIN starts a transaction implicitly.
RELEASE of the outermost savepoint commits the transaction up to that point.

Docs: SAVEPOINT

Redshift none

No SAVEPOINT, despite the PostgreSQL lineage.
For partial rollback, split the work across separate transactions or use staging tables.

Docs: SQL Commands (Redshift)

DB2 full

BEGIN;
INSERT INTO orders (id, total) VALUES (1, 100);
SAVEPOINT sp1 ON ROLLBACK RETAIN CURSORS;
INSERT INTO order_items (order_id, product_id) VALUES (1, 999);
ROLLBACK TO SAVEPOINT sp1;
INSERT INTO order_items (order_id, product_id) VALUES (1, 42);
COMMIT;

SAVEPOINT requires an ON ROLLBACK clause: RETAIN CURSORS or RETAIN LOCKS.
ROLLBACK TO SAVEPOINT releases locks acquired since the savepoint unless ON ROLLBACK RETAIN LOCKS was set.

Docs: SAVEPOINT

Snowflake none

Transactions expose only COMMIT and ROLLBACK. For nested error handling, use exception blocks in Snowflake Scripting.

Docs: Transaction Commands

BigQuery none

Multi-statement transactions support only COMMIT and ROLLBACK. No partial rollback.

Docs: Transactions in BigQuery