Monthly Archives: July 2015

SQL Server Infernals – Circle 6: Environment Pollutors


Don’t tell me that you didn’t see it coming: at some point, Developers end up being put to hell by a DBA!

I don’t want to enter the DBA/Developer wars, but some sins committed by Developers really deserve a ticket to the SQL Server hell. In particular, some of those sins are perpetrated when not even a single line of code is written yet and they have to do with the way the development environment is set up.

What they say in Heaven

Before starting a software project, the angelic developers set up their environment in the best of all ways, with proper environment isolation and definition. In particular they will have:

  • Development Environment: this is the place (ideally a dev’s desktop) where the development work is performed. It should resemble the production environment as much as possible.
  • Test Environment (QA): This is where the testers ensure the quality of the application, open bugs and review bug fixes. It should be identical to the production environment (in Heaven it is).
  • User Acceptance Test Environment (UAT): this is where the clients test the quality of third-party applications, request features and file bugs.
  • Staging Environment (Pre-Production): this environment is used to assemble, test and review newer versions of the database before it is moved into production. The hardware mirrors that of the production environment.
  • Production Environment: This is where the real database lives. It can be updated from the staging environment, when available, as well as new functionality and bug fixes release from UAT or staging environment.

If your organization or the project are small, you probably don’t need all of these environments. In Heaven, where time and money are not a constraint, they have all of them and they’re all identical to production. Heh, Heaven is Heaven after all…

Environmental sinners will face SQL Server’s judgement

Setting up your development environment in the wrong way can harm SQL Server (and your software) in many ways, right from the start of the project, throughout its whole lifetime. Let’s see some of the most common sins:


A notable example of Polluted Database

  1. Using the production environment for development: frankly, I don’t think this sin needs any further explanation. On the other hand, don’t assume that nobody’s doing it, despite we’re in 2015: lots of damned developers’ souls confess this sin while entering the SQL Server hell!
  1. Using the test environment for development: Again, this seems so obvious that there should be no need to discuss it: development is development and test is test. The test environment(s) should be used to test the application, not to see it breaking every minute because you changed something. Developing the code and testing it are two different things and, even if you happen to be in charge of both, this is not a good reason to confuse the two tasks.
  1. Using a shared instance for development: Back in the old days, when I was working as an ASP classic developer in a software house, we had a shared development environment on a central IIS server, where everyone saved their code on a shared folder and just had to hit F5 in Internet Explorer to see the changes immediately in action.
    If you think this model is foolish you’re 100% right, but in the 90s’ we didn’t know better. However, while everyone today agrees that it’s a terrible idea for code, you will still find hordes of developers not completely convinced that it’s an equally terrible idea as far as the database is concerned. Having a shared development database greatly simplifies the process of creating a consistent development database, which is a problem only if you have no authoritative source to build it from (which brings us to the next sin).
  1. No source control: Nobody in their right mind would start a software project today without using source control, yet source control for the database is still an esoteric topic, despite the plethora of tools to accomplish this task.
  1. Granting sysadmin rights to the application: If you’re using a local development instance (and you should), you probably are the administrator of that instance. Hey, nothing wrong with that, unless you use windows authentication in your application. In that case, whenever you debug the application in Visual Studio (or whatever you’re using), the application impersonates you (a sysadmin) when hitting the database, so there is no need to grant any permission in order to let the app perform anything on the instance.
    So, what happens when you’re done with development and you have to deploy in test (or, worse, production)? Exactly: nothing works, because (hopefully) the application won’t run with sysadmin privileges in production. At that point, extracting the complete lists of permissions needed by the application is an overwhelming task that you could have happily avoided by developing with a non-privileged user in the first place. When using a regular user, each time the application needs additional permissions, you simply have to add a GRANT statement to the deployment script, which also acts as the documentation the DBA will ask for.
    If you fail to provide this documentation, two things could happen: a) the DBA may refuse to deploy the database b) you could end up needing sysadmin privileges, which means a dedicated instance, which could in turn bring us back to a).
  1. Developing on a different version/edition from production: if your application is targeting SQL Server 2008 R2, developing on SQL Server 2012 could mean that you will discover incompatible T-SQL features after development. The same can be said for the SQL Server edition: if you are using a Developer Edition for development but you are targeting Standard Edition, you will discover the use of enterprise-only features when it’s too late. You can save yourself all the pain by using in development the same exact SQL Server version and edition you are targeting in production.

In the next episodes of SQL Server Infernals I’m afraid I will have to put more developers to hell. If you’re a developer, stay tuned to find out if your soul is a at risk! If you’re a DBA, stay tuned to enjoy seeing more developers damned!

SQL Server Infernals – Circle 5: Inconsistent Baptists


There’s a place in the SQL Server hell where you can find poor souls wandering the paths of their circle, shouting nonsense table names or system-generated constraint names, trying to baptize everything they find on their way in a different manner. They might seem innocuous at a first glance, but beware those damned souls, as they can raise confusion and endanger performance.

What they say in Heaven

Guided by the Intelligent Designer’s hands, database architects in Heaven always name their tables, columns and all database objects following the rules in the ISO 11179 standard. However, standards aside, the most important thing they do is adhere to a single naming convention, so that every angelic DBA and developer can sing in the same language.

It has to be said that even in Heaven some angels prefer specific naming conventions and some other angels might prefer different ones (say plural or singular table names), but as soon as they start to design a database, every disagreement magically disappears and they all sing in harmony.

Damnation by namification

Some naming conventions are better than others, but many times it all comes down to personal preference. It’s a highly debatable subject and I will refrain from posting here what my preference is. If you want to learn more about naming conventions, take advice from one of the masters.

That said, some naming conventions are really bad and adopting them is a one way ticket to the SQL Server hell:


  1. Hungarian Notation: my friends in Hungary will forgive me if I say that their notation doesn’t play well with database objects. In fact, the Hungarian Notation was conceived in order to overcome the lack of proper data types in the BCPL language, putting a metadata prefix in each variable name. For instance, a variable holding a string would carry the “str” prefix, while a variable holding a long integer would carry the “l” prefix.
    SQL Server (and all modern relational databases) have proper data type support and all sorts of metadata discovery features, so there is no point in naming a table “tbl_customer” or a view “vwSales”. Moreover, if the DBA decides to break a table in two and expose its previous structure as a view (in order to prevent breaking existing code), having the “tbl” prefix in the view name completely defeats the purpose of identifying the object type by its prefix.
    Next time you’re tempted to use the Hungarian Notation ask yourself: “is my name John or DBA_John?”


    Hungary is a nice str_country.

  2. Using insanely short object names: Some legacy databases (yes, you, DB2/400) used to have a hard maximum of 10 characters for object names. It wasn’t uncommon to see table names such as “VN30SKF0OF” or “PRB10SPE4F”: good luck figuring out what those tables represented!
    Fortunately, those days are gone and today there is no single reason to use alphabet soup names for your objects. The object name is a contract between the object and its contents and it should be immediately clear what the contents are by just glancing at the name.
  1. Using insanely long object names: On the other hand, table names such as “ThisIsTheViewThatContainsOrdersWhichAreYetToBeShipped” adds nothing to clarity of the schema. “UnshippedOrders” will do just as well.
  1. Mixing Languages: if you’re fortunate enough to be a native English speaker, you have no idea what this means. In countries such as Italy or Spain, this is a real issue. Many people may end up designing different parts the database schema and each designer may be inclined to use English (the lingua franca of Information Technology) or his/her first language. Needless to say that the result is a mess.
  1. Using the “sp_” prefix for stored procedures: it’s a special case of Hungarian Notation, with severe performance implications. In his blog, Aaron Bertrand discussed the notorious negative impact of the “sp_” prefix, offering a performance comparison with charts and crunchy numbers.
    TL;DR version: SQL Server looks up objects with the sp_ prefix in the master database first, then in the user database. While it may look like a negligible performance issue, it can explode at scale.
  1. Using reserved keywords or illegal characters: While it’s still possible to include almost anything inside square brackets, the use of spaces, quotes or any other illegal character is a totally unneeded masochistic habit. Reserved keywords may also add a thrilling touch of insane confusion to your T-SQL code:
    Enough said.
  1. Using system-generated names for constraints, indexes and so on: When you don’t name your constraints and indexes explicitly, SQL Server is kind enough as to do it for you, using a semi-random system-generated name. That’s great! Uh, wait a moment: this means that two databases deployed to two different instances will contains the same index with a different name, making all your deployment scripts nearly useless. Do yourself a favor and take the time to name all your objects explicitly.
  1. No naming convention or multiple, inconsistent naming conventions: The worst of all mistakes is having multiple naming conventions, or no naming convention at all (which is equal to “as many naming conventions as objects in the database”). Naming conventions is a sort of religious subject and there are multiple valid reasons to adopt one or another: the only thing you should absolutely avoid is turning your database into a sort of Babel tower, where multiple different languages are spoken and nobody understands what the others say.


This is the last circle of SQL Server hell dedicated to Database Design sins: in the next episode of SQL Server Infernals we will venture into the first circle dedicated to development. Stay tuned!

SQL Server Infernals – Circle 4: Anarchic Designers


Constraints are sometimes annoying in real life, but no society can exist without rules and regulations. The same concept is found in Database Design: no good data can exist without constraints.

What they say in Heaven

Constraints define what is acceptable in the database and what does not comply with business rules. In Heaven, where the perfect database runs smoothly, no constraint is overlooked and all the data obeys to the rules of angels:

  • Every column accepts only the data it was meant for, using the appropriate data type
  • Every column that requires a value has a NOT NULL constraint
  • Every column that references a key in a different table has a FOREIGN KEY constraint
  • Every column that must comply with a business rule has a CHECK constraint
  • Every column that must be populated with a predefined value has a DEFAULT constraint
  • Every table has a PRIMARY KEY constraint
  • Every group of columns that does not accept duplicate values has a UNIQUE constraint

Chaos belongs to hell

OK: Heaven is Heaven, but what about hell? Let’s see what will get you instant damnation:


  1. Using the wrong data type: we already found out that the SQL Server hell is full of Shaky Typers. The data type is the first constraint on your data: choose it carefully.
  1. No PRIMARY KEY constraints: In the relational model, tables have primary keys. Without a primary key, a table is not even a table (exception made for staging tables and other temporary objects). Do you want duplicate data and unusable data? Go on and drop your primary key.
  1. NULL and NOT NULL used interchangeably: NOT NULL is a constraint on your data: failing to mark required columns with NOT NULL will inevitably mean that you’ll end up having missing information in your rows. At the same time, marking all columns as NOT NULL will bring garbage data in the database, because users will start using dummy data to circumvent the stupid constraint. We already met these sinners in the First Circle of the SQL Server hell.
  1. No Foreign Key constraints: Foreign Keys can be annoying, because they force you to modify the database in the correct order, but following the rules pays off. Without proper constraints, what would happen if you tried to delete from a lookup table a key referenced in other tables? Unfortunately, it would work, silently destroying the correctness of your data.
    What would happen if you tried to sneak in a row that references a non-existing key? Again, it would bring in invalid data.
  1. No CHECK constraints: Many columns have explicit or implicit constraints: failing to add them to the database schema means that values forbidden by the business rules will start to flow into the database. Some constraints are implicit, but equally important as the explicit ones. For instance:
    • an order should never be placed in a future date
    • a stock quantity should never be negative
    • a ZIP code should only contain numeric characters
    • a Social Security Number should be exactly 9 digits long
  1. Relying on the application to validate data: If I had €0.01 for every time I found invalid data in a database and the developers said “the application will guarantee consistency”, I would be blogging from my castle in Mauritius. Maybe the application can guarantee consistency for the data that it manipulates (and it won’t, trust me), but it can do nothing for other applications using the same database. Often the database is a hub for many applications, each with its own degree of complexity and each with its level of quality. Pretending that all these applications will independently guarantee that no invalid data is brought in is totally unrealistic.

 The last circle of SQL Server hell dedicated to Database Design sins is the circle of Inconsistent Baptists, those who fail to comply to sensible naming conventions. Stay tuned!

SQL Server Infernals – Circle 3: Shaky Typers


Choosing the right data type for your columns is first of all a design decision that has tremendous impact on the correctness of the database schema. It is not just about performance or space usage: the data type is the first constraint on your data and it decides what can be persisted in your columns and what is not acceptable.

Choosing the wrong data type for your columns is a mistake that might make your life as a DBA look like hell.

What they say in Heaven

Guided by angelic spells, the hands that design databases in Heaven always choose the right data type. Database architects always look at the logical schema and ask the right questions about each attribute and they always manage to understand what the attribute is used for and what it will be used for in the future.

What will put you to hell

Choosing the wrong data type is like trying to fit a square peg in a round hole. The worst thing about it is that you end up damaging the peg… ahem… the data.


  1. Using numeric data types for non-numeric attributes: Even if a telephone number contains only digits and it’s called telephone number, it is not a number at all. It does not allow mathematical operations and it has no order relation (saying that a telephone number is greater than another one makes no sense). In fact, a telephone number is a code you have to dial to contact a telephone extension. The same can be said for ZIP codes, which only allow numeric digits, but are nothing like a number. Storing this data in a numeric column is looking for trouble.
  1. Storing data as their human-readable representation: A Notable example is dates stored as (var)char. The string representation of a date is not a date at all: without the validation rules included in the date types, any invalid date could be saved in your column, including ‘2015-02-30’ or ‘2015-33-99’. Moreover, varchar columns do not allow date manipulation functions, such as DATEADD, DATEDIFF, YEAR, MONTH and so on. Another reason why this is a terrible idea is that dates have their own sorting rules, which you lose when you store them as strings. You also need more storage space to save a string representation of a date compared to the proper date type. If you really want to convert a date to a string, you can find many algorithms and functions to perform the conversion in this article I wrote for SQLServerCentral in 2012, but please do it in your presentation layer, not when storing the data in your tables.
    Another surprisingly common mistake in the AS/400 world is storing dates in three separate integer columns for year, month and day. I have no idea where this pattern comes from, but it definitely belongs to hell.
    While much more uncommon in the wild, the same applies to numbers: storing them as varchars is a terrible idea.
    Extra evil bonus: you get double evil points for storing dates and numbers as nvarchar: double the storage, double the pain.
  1. Using deprecated data types: (n)text and image are things of the past: get over it. The replacement (n)varchar(max) and varbinary(max) are much more powerful and flexible.
  1. Using “extended” data type just to “be safe”: This applies both to numeric and character columns: using a bigger data type just to play it safe can be a good idea at times, but not when the size of the column is well known upfront and is instead a vital constraint on the data itself. For instance, a ZIP code longer than 5 characters is obviously an error. A social security number longer than 9 digits is not valid.
    Along the same lines, storing years in a int column is only going to be a waste of storage space. The same can be said about small lookup tables with just a handful of rows in them, where the key column can be a smallint or even a tinyint: it won’t save much space in the lookup table itself, but it can save lots of space in the main tables (with many more rows) where the code is referenced.
  1. Storing fixed-size information in varchar columns: Similarly to the previous sin, when your attribute has a fixed character size, there is no point in using a varying character type. If your attribute has exactly 3 characters, why use varchar(3)?
    Extra evil bonus: varchar(1) will get you double points.
  1. Storing duration in time or datetime columns: Datetime and time represent points in time and they are not meant for storing durations. If you really want to store a duration, use a numeric column to store the number of seconds (it’s the ANSI standard unit measure for representing a duration). Even better, you could store the start/end date and time in two separate datetime columns. SQL Server 2016 also supports periods. 
  1. Getting Unicode wrong: Choosing nvarchar for attributes that will never contain Unicode data and choosing varchar for attributes that can contain Unicode data are equally evil and will get you instant damnation. For instance, a ZIP code will only contain numeric characters, so using Unicode data types will have the only outcome of wasting space. At the same time, storing customer business names or annotations in varchar columns means that you won’t be able to persist international characters. While it may appear quite unlikely that such characters will ever appear in your database, you will regret your decision when that happens (and it will). 
  1. Messing with XML: I’m not a big fan of XML in the database, but sometimes it can come handy. Storing XML data in a plain varchar column is a very bad idea. The XML data type provides validation rules that won’t allow in invalid or malformed XML and also provides functions to manipulate the XML data. Storing schema-less XML is another bad idea: if you have an XML schema use it, otherwise you will end up saving invalid data. On the other hand, using XML to go “beyond relational” and mimic Oracle’s nested tables will only get you damned. Fun times. 
  1. Using different data types in different tables for the same attribute: there’s only one thing worse than getting your data types wrong: getting them wrong in multiple places. Once you decided the data type to store an attribute, don’t change your mind when designing new tables. If it is a varchar(10), don’t use varchar(15) in your next table. Usually proper foreign key constraints help you avoid this issue, but it’s not always the case.
    If this query returns rows, chances are that you have schizophrenic columns in your database schema:

    WITH my_schema AS (
        SELECT OBJECT_NAME(c.object_id) AS table_name,
   AS column_name,
   AS type_name,
        FROM sys.columns AS c
        INNER JOIN sys.types AS t
            ON c.system_type_id = t.system_type_id
    incarnations AS (
        SELECT *,
            DENSE_RANK() OVER (
                PARTITION BY column_name
                ORDER BY type_name, max_length, precision, scale
            ) AS incarnation_number
        FROM my_schema
    incarnation_count AS (
        SELECT *,
            MAX(incarnation_number) OVER (
                PARTITION BY column_name
            ) AS incarnation_count
        FROM incarnations
    SELECT *
    FROM incarnation_count
    WHERE incarnation_count > 1
    ORDER BY incarnation_count DESC,

The lack of proper constraints will be the topic of the next post, when we will meet the anarchic designers. Stay tuned!