The Bit Bucket

I've lost count of the number of computer systems that I've worked with over the years, that insist that everyone has at least two names. Yet there are many people in the community who only have a single name. I'm not just talking about people like Cher who might well have a stage name but then have another name on her passport. I'm talking about people who really only have one name.

Can your systems deal with that?

And if they only have one name, which column does it go into?

Need to be PC

In Australia, we have standards for this sort of thing. While many existing systems will use words like Surname, LastName, FirstName, ChristianName, etc. they aren't considered PC (politically correct or appropriate) now either. In our standards, the terms are:

Family Name
Given Names

Only the family name is mandatory. At Meteor (Metadata Online Registry), these are also defined as 40 characters long each. And only the family name is mandatory. That standard says:

Some people do not have a family name and a given name, they have only one name by which they are known. If the person has only one name, record it in the 'Family name' field and leave the 'Given names' field blank.

A Tale of a Disruptive Man with One Name

At a place that I previously worked, we had a guy apply for a job, and he only had one name. He'd legally changed his name to just a single word. He wanted nothing to do with his family any more and removed all reference to them. He was a pretty alternate sort of character but I so wanted him to come and work with us. He would have broken every system in the place.

But in an interesting twist of fate, he couldn't come to his job interview. The airline told us they were not prepared to allow him to fly in what he was wearing.

To this day, I can't imagine what you could wear to an interview but an airline wouldn't let you fly in it.

When I need to display a potentially large number of rows on a screen, I often need to show them in pages. This is called paginating the rows.

For example, if I'm showing bank transactions, I might want to show 25 per page. The challenge is that I also need to be able to ask for a specific page number. So my query becomes: give me page 7 where there are 25 rows per page. In that case, I want to skip the first 150 rows, then ask for the next 25 rows.

Now I could get SQL Server to send me all the rows at once, and I could just choose which rows to display. If there are many thousands though, that could be really wasteful, mostly in transferring all that data from SQL Server to my application. And if I want to only show a single page at a time, I really don't want to get all the rows every time and then just throw most of them away.

While we could do this with T-SQL in all versions of SQL Server, it was messy. I was really pleased when SQL Server 2012 gave us an option to help with this. It came in the form of extra clauses to the ORDER BY clause.

Take a look at the following query:

It's quite elegant. It says to skip the first 150 rows, then return the next 25 rows, exactly as we wanted. To get any particular page, we just subtract one from the page, multiply by the number of rows per page, and use that as the offset:

OFFSET (@PageNumber – 1) * @RowsPerPage ROWS
FETCH NEXT @RowsPerPage ROWS ONLY

Note that this option isn't really a performance enhancement but that's a story for another day. I like the fact that it is declarative though. You are telling SQL Server what you want, not how to do it. That's almost always a good thing.

Learning T-SQL

It's worth your while becoming proficient in SQL. If you'd like to learn a lot about T-SQL in a hurry, our Writing T-SQL Queries for SQL Server course is online, on-demand, and low cost.

This is part 4 in a series of posts:

• Part 1 covered being sure there is a locking and blocking issue
• Part 2 covered the impact of RCSI
• Part 3 looked at the impacts of indexing on locking and blocking

Today, though, I want to look at the next potential issue, and that's deadlocks.

Deadlocks

I often hear people interchangeably using the words lock, block, and deadlock.

Locks are a standard feature of systems that are designed for concurrency. Nothing nasty to see here. When they become a problem is in two situations:

• They are held too long and cause a blocking problem
• Incompatible locks are held and cause deadlocks

The simplest example of a lock is when I have an item and you want it, but you must wait until I finish with it. That's just normal with sharing something.

If I finish using it quickly, you might barely even notice. But if I hang onto it for a very long time, you experience blocking. Apart from slowing you down (which might seem to be a problem to you), it's also not actually an issue. When I finish, you'll still get it.

As soon as we have more than one item though, things can get messy. If I have item A and I want item B, but you are holding item B because you also want item A, then we have a problem. That's a deadlock.

It also doesn't have to just be two people. I could have item A and want item B. Terry might have item B and want item C. Nga might have item C and want item D. Paula might have item D and want item A. A larger deadlock has occurred.

Resolving Deadlocks – the Old Way

In the past, I worked with systems that didn't resolve this automatically. What would typically happen is that two users would suddenly have their applications freeze. Then a little while later, another user needed something that one of the first two had, and her application would freeze too.

This would go on and on across the building. We used to call it creeping death syndrome.

Eventually, someone would get sick of waiting, and go to find an admin. The admin would work out which user caused the first problem and kill their process. Everyone would then come back to life.

Resolving Deadlocks – the SQL Server Way

In SQL Server, there was a conscious decision to not let this happen and to resolve it automatically. By the word "resolve" though, you need to read "kill one of the user's processes".

SQL Server has a process that looks for these situations and cleans them up. The only question, of course, was which process to kill.

In early versions, they used to kill whichever process had just put the last link the deadlock chain. That process would return an error explaining they had been chosen as a deadlock victim and roll back.

(Note: users don't take well to being told they're victims).

Over time though, this caused issues as long-running processes could collide with short-running processes and get rolled back. They could then have the same issue happen again and again.

So a decision was made to look at the amount of data that had been written to the transaction log by the processes, and kill the one that had done the least work. That worked way better.

That's how SQL Server currently handles deadlocks. In the next post, I'll talk about how you should plan to work around deadlocks.

A while back I wanted to do some testing of Azure Database for PostgreSQL and I needed a database to work with. So I thought I'd try to migrate a basic WideWorldImporters database across to PostgreSQL.

I wrote scripts to do the main migration but then decided that I wanted to change the names of the columns.  While a large number of people use PascalCased names in SQL Server, it's really, really common to use snake_cased names in PostgreSQL. Object names in PostgreSQL are also limited to 63 characters. I needed to alter these names while generating the scripts and I didn't want to do that manually.

To make it easy, in our free SDU Tools for developers and DBAs, we added a function PGObjectName to help.

You can see it in action in the image above, and in the video here:

To become an SDU Insider and to get our free tools and eBooks, please just visit here:

http://sdutools.sqldownunder.com

When you run a query against SQL Server, you might want the rows to be returned in a specific order. However, the first thing you should ask yourself is whether that's needed at all.

Sorting can be a very expensive operation if there is no suitable index that's already sorted the data for you. And the question is: should SQL Server be the one doing the sorting? In many cases, the client application that's reading the data might be a better place to do the sorting. That's not always true but it's worth considering. You often have way more clients than your single database server, and sorting data unnecessarily in the database server could limit your ability to scale your system to higher volumes.

Let's let SQL Server do it

However, let's assume that you do want SQL Server to sort the data. An important first concept is that there's no default order for data when you query SQL Server. I regularly see people who think that if a table has a primary key, or has a clustering key (long story we'll talk about another time), that the data will come back in that order.

It might, but you can't guarantee it. Don't assume that if it does in simple tests, that it will continue to do so in production volumes and situations.

If you want rows to come back in a specific order, you need to say which order you want.

And we do that with the ORDER BY clause. Look at these queries:

The simplest example would be to just order the rows by a single column. In the first query though, I've ordered by two columns. That means that the rows are sorted by Description first, and if the Description values are the same, they are then sorted by ProductID, within each Description.

Ordering can also be ascending (ASC) or descending (DESC). Ascending is the default and we usually don't put ASC in the queries, but there's nothing to stop you doing that if you prefer it.

When you have multiple columns in the ordering, each column can be sorted in a different order. In the second query, I've made the Description column sort descending, but then within each Description, the ProductID would be sorted ascending.

Learning T-SQL

It's worth your while becoming proficient in SQL. If you'd like to learn a lot about T-SQL in a hurry, our Writing T-SQL Queries for SQL Server course is online, on-demand, and low cost.