This is yet another time that a blog post has come about from a question by a developer. They’re good guys, I guess, they keep me on my toes.
This time it was with change logging. We didn’t have Change Data Capture (CDC), or Temporal Tables enabled (have you seen the YouTube videos by Bert Wagner ( blog | twitter ) on these?). What we did have was “manual logging” and no, I’m not even talking about Triggers.
What we had was
INSERT statements, directly after a
MERGE statement, that inserted into a table variable a hard-coded name of the column, the old value, and the new value.
Is that what I would do? Doesn’t matter, it was there before I got there, seems to work, and is low down on the list of priorities to change.
The question was, every time that they needed to add a column to a table, and change log it, they had to add multiple lines to the change tracking procedure and the procedure was getting gross and hard to maintain.
Something to do with DRYness?
You know the drill by now, I quite like to play along so let us facilitate that (from now on I’m going to use Gist, formatting with native WordPress is starting to annoy me).
This will create our table and, luckily, all of it’s columns are important enough to warrant capturing when they get changed.
Old, Way WHERE old=way
Let’s take a look at the code that they were using, shall we?
And the results?
You can probably see the problem here.
Hey! It’s legacy code, let’s focus on just 1 problem at at time!
The main issue that I was asked about was every time a column was deemed important and needed to be added to the list, they had to insert another
INSERT INTO @ChangeLogTemp... and they thought that it wasn’t sustainable in the long run.
Hmmm it also comes across as very RBAR doesn’t it? Every time we want to include another column to the change tracking, we have to add them row by agonizing row. The script is already big enough, if we keep adding more, it will get massive!
Set based is 90% of the time the right way to go but how do we do set based solutions on the same table?
New JOIN Way ON new = way
The first thing I do is to change that table variable into a temp table. Stats, indexes (if necessary), and I can query the results as we go along. Much better!
The second thing is that, whether by luck or by design, the legacy code has the same naming conventions for the columns; new column values are have the prefix “New%” in the column name and old columns have the “Old%” prefix.
This works for us because we can now split the new columns into 2 derived tables, New and Old, and that way we have the differences.
Have you ever tried to find the differences between two consecutive rows of data? It’s fiendishly difficult.
WHERE Column1 on row1 != Column1 on row2 apparently just does not work, le sigh.
I say “little brother” because the whole document talks about PIVOT, with only brief mentions of UNPIVOT in the notes.
If you’re writing documentation like this, please stop.
UNPIVOT we can create a table of our rows around our ID and Column names…
… and with this, we can join on our ID and Column names and get to our more intuitive
WHERE OldValue != NewValue.
Bringing it all together!
And it works!
It’s not great though.
The whole thing was supposed to be to reduce the amount of changes required when they need to include or exclude columns. All in all though, it’s just 6 lines less. Not exactly the great return that you’d expect.
Yeah, true with the old way for every column we want to add we have to add an extra 6 lines while the new way adds 2.
That means for 1,024 columns:
- The old way could have at least 6,144 lines per table. (1024 * 6)
- The new way could have at least 2,048 lines per table (not explaining this calculation 😡 )
So, is there anything else that we can do?
I’ve talked before about T-SQL automation with Dynamic SQL and this should be a good candidate for that.
What can we make dynamic here though? How about…
- The new and old columns bit?
FOR ColumnName IN([Column1], [Column2], [Column3], [Column4], [Column5], [Column6])bit?
CAST(ISNULL([Old/NewColumn], '') AS nvarcharbit?
Explain it to me.
- The new and old columns.
Well, temp tables exist in the tempdb database, they just get a suffix of a lot of underscores and a hex value.
So to get our column names, we can just query the
sys.columns catalog views in [tempdb] and we should have what we need.
FOR ColumnName IN (
I’ve talked before about concatenating values so we can use that to generate this part of the script.
This is basically the same as the above. Don’t be put off by needing to add
CAST(ISNULL( before the column names, it’s not as complex as you’d think.
Now that we have our dynamic bits, let’s create the full statements.
Full Dynamic Script
Results are good!
Overall, the script is longer at nearly double the lines but where it shines is when adding new columns.
To include new columns, just add them to the table; to exclude them, just add in a filter clause.
So, potentially, if every column in this table is to be tracked and we add columns all the way up to 1,024 columns, this code will not increase.
Old way: at least 6,144.
New way: at least 2,048.
Dynamic: no change
Like the script, this was a massive post. Back at the start, I said that a developer came to me because they wanted to get more DRY (?) and stop needing to add more content to the stored procedure.
Do you think the developer used this?
I can’t say that I blame them, it’s slightly ugly and unwieldy, and I wrote it so I should love it.
Yet if something was to go wrong and the need was there to open the procedure and troubleshoot it, the first person to open this up is going to let out a groan of despair!
So this request turned into a proof of concept and nothing more. No skin off my back, I have a growing list of tasks to accomplish by 5 minutes ago. Better get back to them.