When many SQL Server customers are asked why they haven’t kept up to date with either SQL Server versions, or more importantly, patches to SQL Server, the answers usually boil down to two reasons:
They are just slack
Their vendors won’t support the later version or patch level
Many SQL Server application vendors don’t keep up to date with testing of their applications on released versions or patches for SQL Server.
While I can understand a hesitation to quickly support later versions of the product, refusing to support later patches of supported versions is particularly concerning. Worse, actively telling customers to avoid installing security patches is deeply troubling.
Preventing clients from installing security patches is simply not reasonable.
If there is a proven issue with a patch, that’s understandable. But if the main reason is that the vendor just hasn’t done the work to test the patch, I believe that vendors who do this need to bear liability for any ensuing issues that occur, regardless of their license agreement that might try to exclude consequential damages from use or inability to use their products.
I keep doing work at sites where none of the database code is stored in version control (source control) systems. I keep wondering why that is.
At a recent site, all the source code was in individual files just sitting in a single folder. That’s just not sensible.
I’m left wondering why it is that almost every team that I see working with higher-level languages just assumes that some form of source control would be used, yet it’s almost the opposite when I’m working with data teams.
Having decent source control makes such a difference:
No more overwriting changes and losing them.
No more wondering what changed between versions, or who changed them.
And so on and so on.
There seems to have never been a culture of source control among DBAs; and database developers are somewhere in between these two worlds.
One aspect of this is tooling.
Vendors like Red-Gate do a reasonable job with their source control offerings for T-SQL but some clients want a “pure-Microsoft” solution for some reason.
In earlier versions of SQL Server Management Studio (SSMS), there was support for an SCCI (Source Code Control Interface) provider add-on. That would let you connect SQL Server script projects to source control. Sadly, that disappeared in recent versions of SSMS. I gather that there might be a way to attach the Visual Studio Team Explorer to it but I haven’t pursued that and I really hope that a standard interface will return soon. I feel that SSMS should interface directly with both TFS and Git as part of a default install. Having tools like this without source code interfaces built in, helps to push an inappropriate direction.
If however, you are using SQL Server Database Tools (SSDT) to build your databases, then both TFS and Git are standard connections from Team Explorer.
I just find that I can’t do my database development work directly in SSDT. I find very few people do that. Most use SSMS for development.
I’d love to hear others’ thoughts on why this culture has evolved this way, and how to change it.
It’s likely that everyone who uses SQL Server Management Studio (SSMS) knows how to tell how long a query ran for. You can see it in the bottom right of the status bar when a query finishes.
But one question that often comes up with a long-running query is when did my query finish?
That’s not in the status bar and many people don’t seem to be aware that you can find it out.
It’s part of the data in the Properties window. So when you come to a query window where the query has finished, and you’re wondering when it finished, hit F4 (or right-click in the window and click Properties), and you’ll see this info shown:
I’ve recently been working at a site at present where a new case-sensitive SQL Server system is being implemented. I cannot begin to describe what a poor idea I think this is.
In the end, all that a case sensitive system allows you to do is to have:
Multiple identifiers exist
They exist in the same scope
The names of the identifiers differ only by case
You’d have a hard time convincing me that that would ever be a good idea.
At least not for a system used by typical humans. No sensible person is ever going to be comfortable with “John Smith” being a different name to “john smith”. And similarly, do you really want a single database table with a CustomerID column, a CustomerId column, and a customerID column?
Well I certainly don’t.
In the new system that I mentioned, there are columns ending in ID, Id, etc. They haven’t even been consistent in the naming of their case-sensitive objects.
And yes, I hear the “but what about private variables vs properties in languages like C#?” complaint:
Age is the object’s property
age is where the object stores the value assigned to the property
Surely we can come up with a better naming convention than that. I’ve lost count of how many times I’ve seen people using the property when they meant the private variable or vice versa. It’s just not sensible.
Now before I hear complaints that case matters, be clear that I’m not talking about case preservation; that’s an entirely different thing. Yes, if I defined a column as CustomerName, I don’t care if I query it by customername, Customername, etc. I want it coming back as CustomerName ie: however I defined it. Case preservation is a virtue; it’s case sensitivity that I see as an almost always painful and unnecessary thing.
Worse, if you’ve ever tested an application against a case-sensitive server, you’ll understand the challenges involved. It’s hard to get case-sensitive code correct.
I gather that SQL Server Management Studio has a current bug that arises when you remove and re-add a database from an availability group on a case-sensitive server. Why? It appears that they aliased a table with A in one place, and used the alias as a in another place. It’s really nonsense to have a situation where that matters but it highlights the other big issue. It makes for fragile applications.
Do you really want to be the one who’s testing all your applications and 3rd party utilities to find out if they’ve tested case-sensitivity properly? Do the tools that you use have a better testing regime than SSMS? I’ll bet that most don’t. And what that means is that you get to spend your life wading through obscure tooling issues.
Every now and then, customers ask me if they really need to run DBCC CHECKDB. There was even a question that came up about this on a private mailing list full of people who really should already understand why. There is no mystery here.
DBCC CHECKDB checks the database for physical readability (are the pages intact and can they be read from the I/O system). This makes people wonder that if the I/O subsystem is already doing this, why does SQL Server need to do this?
Importantly, DBCC CHECKDB checks for logical consistency as well.
That’s incredibly important.
SQL Server is a fabulous database but it has no way of coping with internal structures that are messed up. For example, links on pages to other pages, but now pointing to the wrong pages.
Fortunately, because SQL Server is such a good database, having to deal with logical corruption caused by the database engine messing up is a truly rare event. That could happen though if there was a serious bug in SQL Server.
But even if SQL Server is perfect, it could be led astray. A good example would be a multi-path I/O system. Imagine this scenario:
The database has two paths to the I/O subsystem (multi-path I/O)
It writes a page out one path
It reads the same page from the other path
The I/O subsystem returns an older version of the page from its cache, not the one that was just written
That sort of thing could easily lead to internal corruption of the database. So could issues with your memory subsystem.
Fortunately, the SQL Server team again have your back. It is designed to help to detect some of these issues too. It probably can’t fix them but it might help to find them. For example, in the multi-path I/O problem mentioned above, SQL Server keeps a list of checksums of recently-written pages. If it reads one of those pages and it gets a page with a valid checksum but a different checksum to the one that it recently wrote, it lets you know that something went very wrong.
Even cleverness like this though isn’t going to help if the culprit is the storage subsystem having a cache bug, and so on, and so on.
Finally, DBCC CHECKDB is so important because it lets you find issues quickly. Otherwise, the logical problem would get much, much worse and be difficult, if not impossible to correct. You don’t want to be the DBA who has a large number of backups but every one of them has the same corrupted data.
Ever since I started working with SQL Server back in 1992, the #1 requested feature that I kept hearing about was a separate date data type. SQL Server had a smalldatetime and a datetime but these included both date and time within the same type.
Having date and time often led to odd bugs or performance issues, where people didn’t realize that times were included and tried to work with the values as dates.
In SQL Server 2005, the date data type was meant to be the poster-child for the new SQL CLR integration and there was a shiny new date data type built using it. But when we (the MVPs) looked at it, we all disliked it immensely. The main problem was that it didn’t work with all the existing date-related functions. So the date data type was removed from the product before it shipped.
In SQL Server 2008, we got new intrinsic (standard built-in) data types for date and time (and a few interesting other ones), and they worked with the date-related functions. This was what was needed.
However, there’s a lot of code out there that was built before this data type existed, or was simply built with the wrong data type.
As part of our free SDU Tools collection, we have provided two ways to find columns that might be worth changing:
ListPotentialDateColumns looks at the names of the columns. If the names suggest that it’s a date, but the data type is one that includes time, it’ll flag the column as a potential. You can see it being used here:
In each case, the column name suggests that it’s a date, but the data type says otherwise. That runs quite quickly as it’s just looking at the column names and data types.
Harder to find are columns that are declared with a datetime, smalldatetime, datetime2, etc. yet don’t contain any time values. To make these earlier to locate, we’ve built another tool.
ListPotentialDateColumnsByValue looks through all the values in a column. If the column only contains dates, it reports it:
Now that’s obviously a bit slower as it has too look through all the data. After running this (for example), we know that the ValidFrom column in Sales.BuyingGroups only contains dates, without any time component. It’s likely that it should have been a date column instead.
You can see both of these in action here:
You’ll find more information on our free SDU Tools here:
Many database developers (and other developers) seem to regard the endless use of abbreviations as some badge of honor. Don’t be one of these people.
Avoid abbreviations almost all the time.
I’ve written before about my dislike for the EOMONTH T-SQL statement. Given the same version introduced names like DATETIMEOFFSETFROMPARTS, surely we didn’t have to save 3 characters and could have had ENDOFMONTH. (I heard it was named this way to match the Excel function but matching something from another language that was created a long time ago isn’t the right answer here).
I’d also note that I’d rather have seen it as END_OF_MONTH. It’s a real pity that T-SQL has names like DATETIMEFROMPARTS but also names like NODE_FROM_PARTS all introduced in recent years. Again, I think there’s a need for more cranky old dudes in the development team.
Now we can’t change the language directly, but we can make better decisions when using the language.
For example, when using the DATEADD function, we have two choices for parameters. One is a set of (sometimes fairly obscure) abbreviated values. The other is a set of actual English words.
Please choose to use words not abbreviations.
Don’t write this code like this:
Are you confident about what each of those means? Instead write this:
The same applies to column names. Don’t write Mbr when you could have written Member.
When I’m looking at a database schema for the first time, there are a number of tell-tale signs that give me the hint that the developers really haven’t done much work with SQL Server before. They’ve made a “newbie” mistake.
One of those is the extensive use of the float data type.
Most times that I see this, the developers have come from a C or Java background and they assume that something that needs a decimal point in it, needs to be float. There are some situations where float makes sense, but 99% of the time what they should have used was decimal.
float is used to store approximate values, not exact values. It has a precision from 1 to 53 digits.
real is similar but is an IEEE standard floating point value, equivalent to float(24).
Neither should be used for storing monetary values. Because the values cannot be stored precisely, people who use float end up with values that don’t match, columns of values that don’t quite add up, or totals that are a cent out, etc. They spend their lives trying to round values to fix the issue, and usually don’t get it right.
Here’s an example. What values should this code print?
You’d expect the values 0.0, 0.1, 0.2 and so on up to 10.0. But that’s not what you get. The query would continue until the maximum value for the data type was exceeded (a long time). If you stop the query, you’ll see odd values:
Worse, note that our stop value of 10 is actually shown, but it didn’t stop:
The problem is that while the value 10 can be stored accurately in float, the value 0.1 can’t be. In decimal, we have recurring fractions. 1/3 is 0.33333 recurring. We can’t write it precisely in decimal. In binary though, 0.1 has the same issue.
So even though we had a test of WHILE @Value <> 10.0, the value never exactly equalled 10.0. So why does it show 10 in the Messages tab? That’s because SQL Server Management Studio (SSMS) rounds the values that it prints. It’s not showing us the actual value.
We could fix this by substracting @Value from 10 and taking the absolute value of the result, then comparing it to a small increment. But who wants to write code like that?
In SQL Server, decimal, numeric, money, and smallmoney are the data types with decimal places that store values precisely. numeric is basically a synonym for decimal. money and smallmoney are old Sybase data types that have fixed scale, and have a funky relationship with currency symbols when converting strings. I generally don’t use those. (There are some arguments for them in gigantic data warehouses where their smaller storage size might help but with row compression, the reasons for that are quickly disappearing). It’s one of the problems with backwards compatibility in SQL Server.
You can’t blame people for using a data type called money for storing amounts of money. But it’s generally not the right answer.
Let’s now look at the query from before if we change to decimal:
When executed, it stops exactly as expected:
Decimal (and numeric) require a precision and a scale. These should be chosen appropriately to store the values that you need. You need to keep rounding in mind when you calculate decimal values.
As I mentioned earlier, there are places where float and/or real make sense, but they are typically scientific calculations, not business calculations.