In procedures invoked by the CALL command as well as in anonymous code blocks (DO command), it is possible to end transactions using the commands COMMIT and ROLLBACK. A new transaction is started automatically after a transaction is ended using these commands, so there is no separate START TRANSACTION command. (Note that BEGIN and END have different meanings in PL/pgSQL.)

Here is a simple example:

CREATE PROCEDURE transaction_test1()
LANGUAGE plpgsql
AS $$
BEGIN
    FOR i IN 0..9 LOOP
        INSERT INTO test1 (a) VALUES (i);
        IF i % 2 = 0 THEN
            COMMIT;
        ELSE
            ROLLBACK;
        END IF;
    END LOOP;
END;
$$;

CALL transaction_test1();

A new transaction starts out with default transaction characteristics such as transaction isolation level. In cases where transactions are committed in a loop, it might be desirable to start new transactions automatically with the same characteristics as the previous one. The commands COMMIT AND CHAIN and ROLLBACK AND CHAIN accomplish this.

Transaction control is only possible in CALL or DO invocations from the top level or nested CALL or DO invocations without any other intervening command. For example, if the call stack is CALL proc1() → CALL proc2() → CALL proc3(), then the second and third procedures can perform transaction control actions. But if the call stack is CALL proc1() → SELECT func2() → CALL proc3(), then the last procedure cannot do transaction control, because of the SELECT in between.

PL/pgSQL does not support savepoints (SAVEPOINT/ROLLBACK TO SAVEPOINT/RELEASE SAVEPOINT commands). Typical usage patterns for savepoints can be replaced by blocks with exception handlers (see Section 43.6.8). Under the hood, a block with exception handlers forms a subtransaction, which means that transactions cannot be ended inside such a block.

Special considerations apply to cursor loops. Consider this example:

CREATE PROCEDURE transaction_test2()
LANGUAGE plpgsql
AS $$
DECLARE
    r RECORD;
BEGIN
    FOR r IN SELECT * FROM test2 ORDER BY x LOOP
        INSERT INTO test1 (a) VALUES (r.x);
        COMMIT;
    END LOOP;
END;
$$;

CALL transaction_test2();

Normally, cursors are automatically closed at transaction commit. However, a cursor created as part of a loop like this is automatically converted to a holdable cursor by the first COMMIT or ROLLBACK. That means that the cursor is fully evaluated at the first COMMIT or ROLLBACK rather than row by row. The cursor is still removed automatically after the loop, so this is mostly invisible to the user. But one must keep in mind that any table or row locks taken by the cursor's query will no longer be held after the first COMMIT or ROLLBACK.

Transaction commands are not allowed in cursor loops driven by commands that are not read-only (for example UPDATE ... RETURNING).