Question

I have two tables with 10-20 million rows that have GUID primary keys and at leat 12 tables related via foreign key. The base tables have 10-20 indexes each.

We are moving from GUID to BigInt primary keys. I’m wondering if anyone has any suggestions on an approach. Right now this is the approach I’m pondering:

1. Drop all indexes and fkeys on all the tables involved.
2. Add ‘NewPrimaryKey’ column to each table
3. Make the key identity on the two base tables
4. Script the data change “update table x, set NewPrimaryKey = y where OldPrimaryKey = z
5. Rename the original primarykey to ‘oldprimarykey’
6. Rename the ‘NewPrimaryKey’ column ‘PrimaryKey’
7. Script back all the indexes and fkeys

Does this seem like a good approach?
Does anyone know of a tool or script that would help with this?

TD: Edited per additional information. See this blog post that addresses an approach when the GUID is the Primary: http://www.sqlmag.com/blogs/sql-server-questions-answered/sql-server-questions-answered/tabid/1977/entryid/12749/Default.aspx

Answer 1

It certainly sounds like this strategy would work — dropping the constraints, changing the column out from underneath them (type changes, name remains the same), and then recreating the constraints is fairly elegant.

Is the goal to ultimately drop the GUID columns? If so, you won’t actually reclaim the space unless the tables are copied or rebuilt, so maybe the following adjustment:

…
4.Script the data change “update table x, set NewPrimaryKey = y where OldPrimaryKey = z
5.Drop the original primarykey to ‘oldprimarykey’
6.Rename the ‘NewPrimaryKey’ column ‘PrimaryKey’
7.Script back all the indexes and fkeys (building clustered indexes “rebuilds” tables)
8.For all tables that don’t have clustered indexes, do something to make sure they get rebuilt and their space is reclaimed (such build and then drop a clustered index)

Needless to say, test it on a dev box before running on Production!