That all makes good sense.

I'll look into 6NF, but I think that may over-complicate things (as you suggested).

The real goal here is just to store the data. Since new...

True.

This is a non-production "research" ETL database.

The table in question won't ever be updated by a "end user" so to speak.

However, your point is valid...

I may look at dropping the...

Well adding an index will take forever, just dropped them. Did the deletes then re-created.

Since this table will have deletes perhaps once per quarter, I think that's the way...

Erland Sommarskog (7/25/2013)

---

So the problem is exactly the foreign keys. CASCADE or NO ACTION, if you delete a row from a table, and there is an FK constraint from a...

As I mentioned I took a backup of the database before dropping that table.

I restored that backup to my local, same problem persists there.

So I can examine the issue now...

Now that's a...

Interesting point about pre-generating the values.

That does make sense about getting the job done.

I'll play around with that and post back a sample table structure and some data later.

As for...

Apologies, I should have been more clear. I wasn't hoping for a tsql direct solution, more just a general discussion.

As for it being "random", I agree it makes no...

I basically gave up on the RunID approach. Spoke with businesses users, just signing off in a test environment is good enough. Will use snapshots to allow for...

Thanks for the link!

I understand the backing out of the run for insert only, but if we also updated records... We couldn't back those out (without an audit log of...

Sean Lange (5/1/2012)

---

Are you talking about opening a transaction and keeping it open on your production server?

Do you have a different code base for these users so they can view...

Maxer (5/1/2012)

---

How crazy is using a snapshot? Data diff will only be a few gigs per load.

Start snapshot, run load, then let users poke aroind for an hour amd...

How crazy is using a snapshot? Data diff will only be a few gigs per load.

Start snapshot, run load, then let users poke aroind for an hour amd lastly...

Well, No Lock could really screw up your results though if dirty reads are not acceptable (exact counts, financial related work, multiple queries that rely on related results to tally/total/sum...

I was wondering about that.

From what I understand it will basically just take a snapshot/temp-copy of the data in a differential sense. So other operations can be performed against...