Category: T-SQL

Andy Levy recommends checking out sequences:

The SEQUENCE object eliminates all of this. It gives you a simple way to just grab a new number and increment for the next caller. It’s very fast as there’s no visible table I/O, and it’s unaffected by rollbacks.

And it’s so much easier to use! You just ask for the next number in the sequence!

I like sequences, though the interesting thing is that 2020 me has created them a lot less frequently than 2012 me was sure I would. I’m glad they’re in the product, however.

One thing I should point out is that sequences are like identity columns in that you can have gaps due to user behavior, such as rolling back transactions. If you absolutely need gap-free sets of numbers, you’re back to Andy’s Method One, except everything has to be serializable and wrapped in explicit transactions.

Comments closed

Jeet Kainth explains the importance of specifying your window when using LAST_VALUE():

To return the actual last value, add the additional clause ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING. This ensures that for each row the LAST_VALUE() function looks at all rows; from the very first row to the very last row.

Click through for the example. Remember that the default is going to the current row, not the entire data set.

Comments closed

Phil Factor dives into sys.dm_exec_describe_first_result_set_for_object():

I’ve been working on a project unkindly nicknamed ‘The Gloop’ because the code is a bit amorphous. Basically, it is an approach to documenting SQL Server databases, using the facilities provided such as the metadata views and DMFs. Although it is relatively simple to record the result returned by a table-valued function I’d rather neglected the stored procedures because there was no metadata that could produce the first result set produced by a procedure or trigger.

I’d been silly because there is, of course, an Execution system DMF that does it: sys.dm_exec_describe_first_result_set_for_object(). it takes as its parameter the object_id of a procedure or trigger and describes the first result metadata for the module with that ID. It has the same result set definition as sys.dm_exec_describe_first_result_set.

I’m going to have to try this out, but I have problems with sp_describe_first_result_set and how it blows up if you use a temp table in the stored procedure whose result set you’re grabbing.

Comments closed

Andy Levy talks about one of my favorite functions in SQL Server 2017:

This is another in a group of several posts on modernizing T-SQL code with new features and functionality available in SQL Server.

SQL Server 2016 gave us the STRING_SPLIT() function, but what about the reverse – compiling a set of values into one delimited string? We only had to wait 15 months for the release of SQL Server 2017, and the STRING_AGG() function.

I had the STUFF() / FOR XML PATH trick memorized for quite some time, but that was always a solution which felt like it worked on accident. Even if the new solution weren’t faster than the old, I’d still use it.

Comments closed

Itzik Ben-Gan continues a series complexity around NULL:

When using the offset window functions LAG, LEAD, FIRST_VALUE and LAST_VALUE, sometimes you need to control the NULL treatment behavior. By default, these functions return the result of the requested expression in the requested position, irrespective of whether the result of the expression is an actual value or a NULL. However, sometimes you want to continue moving in the relevant direction, (backward for LAG and LAST_VALUE, forward for LEAD and FIRST_VALUE), and return the first non-NULL value if present, and NULL otherwise. The standard gives you control over this behavior using a NULL treatment clause with the following syntax:

offset_function(<expression>) IGNORE_NULLS | RESPECT NULLS OVER(<window specification>)

There are three good examples of functionality around handling NULL which the current implementation of T-SQL is missing.

Comments closed