A load table is partitioned and has a set of “rolling” partitions. Multiple processes are continuously “loading” and these processes all take a “share” lock on the table through dbms_lock while they are inserting data.

When partition maintenance must occur (i.e. adding a new daily partition) that process requests an “exclusive” lock on the load table. This effectively blocks future “share” lock requests, and once all of the current “share” locks are released the partition maintenance process will have an “exclusive” lock and can add the new partition.

Of course this is all application level coordination, but with multiple processes running in a RAC environment the global lock coordination takes much longer to accomplish in a free for all than it does by temporarily halting access to the load table.

The beauty of this is that it all coordinates automagically and since it’s based on Oracle’s own locking mechanism it’s been rock solid.