spontaneous rollback and retry?

I'm trying to make sense of what I'm seeing for a process that's issuing a parallel merge statement. For about the first 45 minutes I see the undo blocks allocated rising (v$transaction.used_ublk). Then the number slow goes back down to zero over the next 45 minutes. Both "transaction rollbacks" and "rollback changes - undo records applied" statistic goes up as well. At that point, Oracle seems to be stopping all the parallel sessions, starting a new transaction (v$transaction.start_scnb changes) and creating new parallel sessions. Then undo allocation starts to go back up again. It's been doing this for several hours longer than this process should really take. This process is not in any sort of loop however so it appears that Oracle is doing all of these "retries" on it's own.

What exactly is Oracle doing here?

Thanks,
Richard Kuhler Received on Wed Jan 07 2004 - 17:47:44 CST