A savepoint is a way to split a transaction into smaller blocks that can be rolled back independently of each other. Thanks to savepoints, you can divide a big transaction (one transaction with multiple statements) into smaller chunks, allowing a subset of the bigger transaction to fail without having the overall transaction fail. PostgreSQL does not handle transaction nesting, so you cannot issue a nested set of BEGIN, nor COMMIT/ROLLBACK statements. Savepoints allow PostgreSQL to mimic the nesting of transaction blocks.
Savepoints are marked with a mnemonic name, which you can use to commit or rollback. The name must be unique within the transaction, and if you reuse the same over and over, the previous savepoints with the same name will be discarded. Let's see an example:
forumdb=> BEGIN;
BEGIN
forumdb=> INSERT INTO tags( tag ) VALUES ( 'Eclipse IDE' );
INSERT 0 1
forumdb=> SAVEPOINT other_tags;
SAVEPOINT
forumdb=> INSERT INTO tags( tag ) VALUES ( &apos...
