How about not running MERGE, UPDATE, or DELETE commands at all, and instead converting them into faster and more scalable INSERT /*+ APPEND PARALLEL */ commands, using table renames or partition-management commands (such as EXCHANGE PARTITION) to "publish" new data from old? http://evdbt.com/videos/ Hope this helps... Thanks! On 8/19/2013 9:18 AM, fmh wrote: > I usually prepare for such ops ... > - Instance prep: disable archive logs, use no logging ,... etc. > - Session prep: enable PQ, PDML, ...etc > - SQL Prep: rewrite it e.g. updateble inline views vs. correlated queries. > - Use BULK SQL if need to. > I'm running an update on a partitioned table of 44M rows. It has been 3 > days now and there seems end is not near. > > 1) The update session is the only FG session in the db. v$session_longops > show the query involved but all rows have 'sofar=totalwork'. The view has > not updated since 2 days ago. How else can progress be monitored? > > 2) I did enable PDML ,but on 10.2.0.3, we hit this bug. The workaround did > not work to disable the recycle bin. This was a major set back. Bug 4896424 > - Parallel DML can fail with ORA-60 (Doc ID 4896424.8) > > > If you've had any experience with such operations, please provide your > comments/feedback to the above inquiries. -- //www.freelists.org/webpage/oracle-l