Block construction in PL/pgSQL is extended by the optional autonomous keyword. It is possible to treat the whole function body as an autonomous transaction:

CREATE FUNCTION foo(x integer) RETURNS integer AS $$
BEGIN AUTONOMOUS
  RETURN x;
END;
$$ LANGUAGE plpgsql;

or create a separate BEGIN/END block:

CREATE OR REPLACE FUNCTION myaudit() RETURNS boolean AS $$
BEGIN AUTONOMOUS
        BEGIN AUTONOMOUS
        INSERT INTO audit_schedule VALUES ('new audit',now());
        END;
        ... -- do audit itself
        RETURN true;
END;
$$ LANGUAGE plpgsql;

If an exception is raised inside a BEGIN AUTONOMOUS block, this autonomous transaction is aborted, and the standard exception handling procedure is started, unwinding the stack and executing exception handlers until the exception is caught. Thus, exception handling for autonomous transactions in PL/pgSQL is done in the same way as for Postgres Pro subtransactions.

If an error is caught by an EXCEPTION clause, local variables of the PL/pgSQL function remain as they were when the error occurred, but all changes to persistent database state within the block are rolled back.