Backstory
At work where we're planning on deprecating a Natural Key column in one of our primary tables. The project consists of 100+ applications that link to this table/column; 400+ stored procedures that reference this column directly; and a vast array of common tables between these applications that also reference this column.
The Big Bang and Start from Scratch methods are out of the picture. We're going to deprecate this column one application at a time, certify the changes, and move on to the next... and we've got a lengthy target goal to make this effort practical.
The problem I have is that a lot of these applications have shared stored procedures and tables. If I completely convert all of Application A's tables/stored procedures Application B and C will be broken until converted. These in turn may break applications D, E, F...Etc. I've already got a strategy implemented for Code classes and Stored Procedures, the part I'm stuck on is the transitioning state of the database.
Here's a basic example of what we have:
Users
---------------------------
Code varchar(32) natural key
Access
---------------------------
UserCode varchar(32) foreign key
AccessLevel int
And we're aiming now just for transitional state like this:
Users
---------------------------
Code varchar(32)
Id int surrogate key
Access
---------------------------
UserCode varchar(32)
UserID int foreign key
AccessLevel int
The idea being during the transitional phase un-migrated applications and stored procedures will still be able to access all the appropriate data and new ones can start pushing to the correct columns -- Once the migration is complete for all stored procedures and applications we can finally drop the extra columns.
I wanted to use SQL Server's triggers to automatically intercept any new Insert/Update's and do something like the following on each of the af开发者_如何学编程fected tables:
CREATE TRIGGER tr_Access_Sync
ON Access
INSTEAD OF INSERT(, UPDATE)
AS
BEGIN
DIM @code as Varchar(32)
DIM @id as int
SET @code = (SELECT inserted.code FROM inserted)
SET @id = (SELECT inserted.code FROM inserted)
-- This is a migrated application; find the appropriate legacy key
IF @code IS NULL AND @id IS NOT NULL
SELECT Code FROM Users WHERE Users.id = @id
-- This is a legacy application; find the appropriate surrogate key
IF @id IS NULL AND @code IS NOT NULL
SELECT Code FROM Users WHERE Users.id = @id
-- Impossible code:
UPDATE inserted SET inserted.code=@code, inserted.id=@id
END
Question
The 2 huge problems I'm having so far are:
- I can't do an "AFTER INSERT" because NULL constraints will make the insert fail.
- The "impossible code" I mentioned is how I'd like to cleanly proxy the original query; If the original query has x, y, z columns in it or just x, I ideally would like the same trigger to do these. And if I add/delete another column, I'd like the trigger to remain functional.
Anyone have a code example where this could be possible, or even an alternate solution for keeping these columns properly filled even when only one of values is passed to SQL?
Tricky business...
OK, first of all: this trigger will NOT work in many circumstances:
SET @code = (SELECT inserted.code FROM inserted)
SET @id = (SELECT inserted.code FROM inserted)
The trigger can be called with a set of rows in the Inserted
pseudo-table - which one are you going to pick here?? You need to write your trigger in such a fashion that it will work even when you get 10 rows in the Inserted
table. If a SQL statement inserts 10 rows, your trigger will not be fired ten times - one for each row - but only once for the whole batch - you need to take that into account!
Second point: I would try to make the ID's IDENTITY
fields - then they'll always get a value - even for "legacy" apps. Those "old" apps should provide a legacy key instead - so you should be fine there. The only issue I see and don't know how you handle those are inserts from an already converted app - do they provide an "old-style" legacy key as well? If not - how quickly do you need to have such a key?
What I'm thinking about would be a "cleanup job" that would run over the table and get all the rows with a NULL legacy key and then provide some meaningful value for it. Make this a regular stored procedure and execute it every e.g. day, four hours, 30 minutes - whatever suits your needs. Then you don't have to deal with triggers and all the limitations they have.
Wouldn't it be possible to make the schema changes 'bigbang' but create views over the top of those tables that 'hide' the change?
I think you might find you are simply putting off the breakages to a later point in time: "We're going to deprecate this column one application at a time" - it might be my naivety but I can't see how that's ever going to work.
Surely, a worse mess can occur when different applications are doing things differently?
After sleeping on the problem, this seems to be the most generic/re-usable solution I could come up with within the SQL Syntax. It works fine even if both columns have a NOT NULL restraint, even if you don't reference the "other" column at all in your insert.
CREATE TRIGGER tr_Access_Sync
ON Access
INSTEAD OF INSERT
AS
BEGIN
/*-- Create a temporary table to modify because "inserted" is read-only */
/*-- "temp" is actually "#temp" but it throws off stackoverflow's syntax highlighting */
SELECT * INTO temp FROM inserted
/*-- If for whatever reason the secondary table has it's own identity column */
/*-- we need to get rid of it from our #temp table to do an Insert later with identities on */
ALTER TABLE temp DROP COLUMN oneToManyIdentity
UPDATE temp
SET
UserCode = ISNULL(UserCode, (SELECT UserCode FROM Users U WHERE U.UserID = temp.UserID)),
UserID = ISNULL(UserID, (SELECT UserID FROM Users U WHERE U.UserCode = temp.UserCode))
INSERT INTO Access SELECT * FROM temp
END
精彩评论