I recently saw wait events on the single instance database, the root cause of the problem was caused by the undo event “Wait for a undo record”.
There are cases where parallel process recovery is not as fast as serial process recovery, because the PQ secondary can interfere with each other’s operation by competing for the same resource. With such a process rollback, the performance may be worse in parallel compared to serial rollback.
Due to this contention and the perceived slowness and ‘hang’-like symptoms (the database may seem to hang, the SMON and parallel query secondary may seem to be taking up all the available CPU), DBA intervention can be performed. If a large transaction is terminated, the cleaning process may also take time.
alter system set fast_start_parallel_rollback = false;