This week was Canadian Thanksgiving so this post on metadata stuffing is timely.
Today Thomas LaRock (@sqlrockstar | blog ) posted a rant about our Database Design Throwdown topic on naming standards for tables and other database objects. Tom is a fan of what I call “metadata stuffing” in object names. That’s basically shoving as much additional information as one can into object names so that one does not have to go find out that information from its rightful place.
My reason for wanting to use prefixes is simple enough: I want to know if I am looking at a table or a view when reviewing code.
Karen doesn’t believe that anyone should be using object names as a place to store meta data about the objects themselves. I would like to agree with her but then we’d both be wrong.
He even created the nice graphic of my quote I use here. Thanks, Tom.
Object Confusion Abounds
That quote is indeed one I use during our debate. It’s a snarky introduction to what I think the real problem is: our tools aren’t doing enough to help us with this potential confusion of tables and views. When you are writing or looking at a query, the syntax for referencing a table or a view is exactly the same.
When a developer writes:
USE AdventureWorks; GO SELECT Name, ProductNumber, ListPrice AS Price FROM Production.Product ORDER BY Name ASC; GO |
He has no idea for certain if Product is a table or a view. The syntax is the same. And if the developer is working in a text editor of some sort, or with some native tools, there’s no tooltip or other help that the he can use to check what type of object they are querying.
Why does this matter? Tom has a great presentation where he shows the impacts of trying to make stuff work and you don’t know what you are looking at. That’s why Tom wants to do this with his objects:
FROM Production.tbl_Product |
or
FROM Production.view_Product |
So that the object type is injected into the object name. That seems so innocuous, doesn’t it? What could go wrong?
I’m here to tell you that this is a slippery slope. One of the most egregious examples of this sort of meta data stuffing I’ve run across is one that required all this meta data to be prefixed in front of every table name:
- tbl_ prefix
- Primary Systems that managed the table
- Primary subject area that the table belongs to in the data model
- Classification of the role the table plays in the database (Associative Entity, Domain Entity, Master Data, Reference Data, Log Data, etc.)
- Three letter login of the DBA responsible for administering this table (I KID YOU NOT).
So in this wonderful naming scheme, we’d get:
FROM Production.tbl_MFGR_ORDERPROC_DOM_KQL_Product |
I have found that once an organization starts thinking of stuffing, their designs become turkeys really fast. It’s ugly. Think about the tools you use, with all those nifty object lists on the left side. To find a table you need to know all that great metadata as you scroll through the list, hoping that PRODUCT Is buried in there, somewhere. And what the heck was the name of that guy that dressed funny who did all that data stuff for the company before he won the lottery?
All that metadata that should have been managed elsewhere, not prefixed in front of the “real” name of the table. In fact, it was. In the data model and in the system catalog. Every time any of that data changed (DBA assignment changes, DBA wins the lottery, whatever), we had to rename the table and change all the code and reports that referenced it. Sure we could have isolated systems by this change by using views and or aliases but that is additional complexity for no performance gain, either. Refactoring might have helped, but eventually we’d still have to change all the code and queries.
Why I Hate Metadata Stuffing
- It’s redundant data. Just like with business data, the reason we want to minimize redundant data is because we then have to worry about updating the data in multiple places. There’s cost and added risk for that.
- It changes. I don’t know about you, but I don’t have the luxury of taking down a production system just to update a change in the name of the DBA’s favourite TV show or whatever lame naming scheme someone thought up. Sure, tables can’t change into views or indexes, but all the other type of stuffings will change.
- It takes up real estate. I get all kinds of flack from developers and DBAs for the length of object names when I want the names to be meaningful. It’s funny how spelling out CUSTOMER is unacceptable, but adding the exact same characters in front of every object of its type is A-OKAY. What’s up with that? Somehow optimizing names for developers is more important than loving your data? Show me where it says that in the Project Charter.
- tbl_ is a tell for bad database design. I don’t know where this particular naming scheme originated, but when I do a database design review and I see this naming scheme, I know that the designer learned design in a one hour webinar “training course” and has not really mastered the complexities of enterprise database design and maintenance. The design will be less than best practice 90% of the time. This naming scheme is prominent in programming books, introduction to database books, presentations by non-database people, and uninformed blog posts, by far. It’s not popular with people who do professional database design. Sure, some products use this, too, but do you really want to take database design best practices from vendors? How many professional data architects do you think they have on staff? I will most likely see a database design that is highly optimized to make development go faster. Not for data integrity or loving data.
- It’s not needed “for consistency”. One of Tom’s points is that if we are going to prefix views, we have to prefix tables to be consistent. Actually, no, we don’t. If we have to bite the bullet and prefix views because our tools let us down, we can choose not to clutter up tables names just to punish those objects, too. I’m assuming that since Tom prefixes tables and views, he prefixes columns, too, right? its just being consistent. < snort>
- It gets in the way of using the data. Tables and Column names are the most user-facing parts of a database design. When we in IT insist on munging up these names with a bunch of systemese, we make it more difficult for business users to get at their data. It shows that we have optimized the database design to help a relatively small number of technical users (developers, DBAs, ETL folks) over the needs of the business. Ultimately, we build databases to manage data. For the business.
What’s the Cost, Benefit and Risk?
One of my Splendid Truths is that all design decisions should assess cost, benefit and risks. In the overall scheme of things, just prefixing “tbl_” in front of a table name isn’t that costly and it isn’t that risky. Tom assesses his designs based only on potential for performance harm according to his post. He “laughs” at my position. I’m happy that my stance on metadata stuffing brings happiness to his day. But performance is only one data point out of many for making a design decision. Usability, clarity, business goal support are other factors that a database architect needs to consider when assigning a name to an object. If we optimize something for a subsystem, we do it as the expense of other subsystems.
Our Tools Should Help Us More
Having said that, I feel the pain of people having to work with sub-standard tools or having to use tools that just refuse to help. Tom showed how SQL Server Management Studio tooltips can help. But all those command line “I don’t need any stinking help” aficionados are left on their own to know what they are looking at.
Oh, and Tom:
ProperNoun_Tom, Pronoun_you Verb_made Pronoun_me Verb_laugh Preposition_with Possessive_your Noun_post.
(See how all this stuffing gets in the way?)
One of my other Splendid Truths about database design is:
Your tools will impact your data models and database designs more than you can imagine.
We shouldn’t sit back and let that happen. Stuffing is great with Tofurky, not with databases.