Archives of the TeradataForum
Message Posted: Tue, 13 Nov 2002 @ 03:59:46 GMT
Glad to hear everyone out here taking questions. You all are awesome and I value your opinions.
I've recently heard of some shops using the following ETL practice globally (in all cases)
1. multiload daily transactions to a temporary table
2. use sql to transform in parallel (awesome idea)
3. use fastexport to bring it out of the database again to file
4. multiload file (#3) straight to the production fact table
My questions are regarding steps 3 & 4. I've traditionally seen steps 1 & 2, with step 2 using an insert/select to apply the data.
Under what circumstances or configurations does the 4 step process make sense?
I can see the benefit of no transient journaling and the ability to leave your nusi's in place on the fact table.
On the negative side, I can also see a real consumption of the utility slots, an exclusive lock being taken during the apply phase, and the potential of pooching the production fact table by using sloppy restart procedures during an mload restart.
|Copyright 2016 - All Rights Reserved|
|Last Modified: 28 Jun 2020|