Dynamic Sorting within SQL Stored Procedures

Question

This is an issue that I've spent hours researching in the past. It seems to me to be something that should have been addressed by modern RDBMS solutions but as yet I have not found anything that really addresses what I see to be an incredibly common need in any Web or Windows application with a database back-end.

I speak of dynamic sorting. In my fantasy world, it should be as simple as something like:

ORDER BY @sortCol1, @sortCol2

This is the canonical example given by newbie SQL and Stored Procedure developers all over forums across the Internet. "Why isn't this possible?" they ask. Invariably, somebody eventually comes along to lecture them about the compiled nature of stored procedures, of execution plans in general, and all sorts of other reasons why it isn't possible to put a parameter directly into an ORDER BY clause.

---

I know what some of you are already thinking: "Let the client do the sorting, then." Naturally, this offloads the work from your database. In our case though, our database servers aren't even breaking a sweat 99% of the time and they aren't even multi-core yet or any of the other myriad improvements to system architecture that happen every 6 months. For this reason alone, having our databases handle sorting wouldn't be a problem. Additionally, databases are very good at sorting. They are optimized for it and have had years to get it right, the language for doing it is incredibly flexible, intuitive, and simple and above all any beginner SQL writer knows how to do it and even more importantly they know how to edit it, make changes, do maintenance, etc. When your databases are far from being taxed and you just want to simplify (and shorten!) development time this seems like an obvious choice.

Then there's the web issue. I've played around with JavaScript that will do client-side sorting of HTML tables, but they inevitably aren't flexible enough for my needs and, again, since my databases aren't overly taxed and can do sorting really really easily, I have a hard time justifying the time it would take to re-write or roll-my-own JavaScript sorter. The same generally goes for server-side sorting, though it is already probably much preferred over JavaScript. I'm not one that particularly likes the overhead of DataSets, so sue me.

But this brings back the point that it isn't possible — or rather, not easily. I've done, with prior systems, an incredibly hack way of getting dynamic sorting. It wasn't pretty, nor intuitive, simple, or flexible and a beginner SQL writer would be lost within seconds. Already this is looking to be not so much a "solution" but a "complication."

---

The following examples are not meant to expose any sort of best practices or good coding style or anything, nor are they indicative of my abilities as a T-SQL programmer. They are what they are and I fully admit they are confusing, bad form, and just plain hack.

We pass an integer value as a parameter to a stored procedure (let's call the parameter just "sort") and from that we determine a bunch of other variables. For example... let's say sort is 1 (or the default):

DECLARE @sortCol1 AS varchar(20)
DECLARE @sortCol2 AS varchar(20)
DECLARE @dir1 AS varchar(20)
DECLARE @dir2 AS varchar(20)
DECLARE @col1 AS varchar(20)
DECLARE @col2 AS varchar(20)

SET @col1 = 'storagedatetime';
SET @col2 = 'vehicleid';

IF @sort = 1                -- Default sort.
BEGIN
    SET @sortCol1 = @col1;
    SET @dir1 = 'asc';
    SET @sortCol2 = @col2;
    SET @dir2 = 'asc';
END
ELSE IF @sort = 2           -- Reversed order default sort.
BEGIN
    SET @sortCol1 = @col1;
    SET @dir1 = 'desc';
    SET @sortCol2 = @col2;
    SET @dir2 = 'desc';
END

You can already see how if I declared more @colX variables to define other columns I could really get creative with the columns to sort on based on the value of "sort"... to use it, it usually ends up looking like the following incredibly messy clause:

ORDER BY
    CASE @dir1
        WHEN 'desc' THEN
            CASE @sortCol1
                WHEN @col1 THEN [storagedatetime]
                WHEN @col2 THEN [vehicleid]
            END
    END DESC,
    CASE @dir1
        WHEN 'asc' THEN
            CASE @sortCol1
                WHEN @col1 THEN [storagedatetime]
                WHEN @col2 THEN [vehicleid]
            END
    END,
    CASE @dir2
        WHEN 'desc' THEN
            CASE @sortCol2
                WHEN @col1 THEN [storagedatetime]
                WHEN @col2 THEN [vehicleid]
            END
    END DESC,
    CASE @dir2
        WHEN 'asc' THEN
            CASE @sortCol2
                WHEN @col1 THEN [storagedatetime]
                WHEN @col2 THEN [vehicleid]
            END
    END

Obviously this is a very stripped down example. The real stuff, since we usually have four or five columns to support sorting on, each with possible secondary or even a third column to sort on in addition to that (for example date descending then sorted secondarily by name ascending) and each supporting bi-directional sorting which effectively doubles the number of cases. Yeah... it gets hairy really quick.

The idea is that one could "easily" change the sort cases such that vehicleid gets sorted before the storagedatetime... but the pseudo-flexibility, at least in this simple example, really ends there. Essentially, each case that fails a test (because our sort method doesn't apply to it this time around) renders a NULL value. And thus you end up with a clause that functions like the following:

ORDER BY NULL DESC, NULL, [storagedatetime] DESC, blah blah

You get the idea. It works because SQL Server effectively ignores null values in order by clauses. This is incredibly hard to maintain, as anyone with any basic working knowledge of SQL can probably see. If I've lost any of you, don't feel bad. It took us a long time to get it working and we still get confused trying to edit it or create new ones like it. Thankfully it doesn't need changing often, otherwise it would quickly become "not worth the trouble."

Yet it did work.

---

My question is then: is there a better way?

I'm okay with solutions other than Stored Procedure ones, as I realize it may just not be the way to go. Preferably, I'd like to know if anyone can do it better within the Stored Procedure, but if not, how do you all handle letting the user dynamically sort tables of data (bi-directionally, too) with ASP.NET?

And thank you for reading (or at least skimming) such a long question!

PS: Be glad I didn't show my example of a stored procedure that supports dynamic sorting, dynamic filtering/text-searching of columns, pagination via ROWNUMBER() OVER, AND try...catch with transaction rollbacking on errors... "behemoth-sized" doesn't even begin to describe them.

---

Update:

• I would like to avoid dynamic SQL. Parsing a string together and running an EXEC on it defeats a lot of the purpose of having a stored procedure in the first place. Sometimes I wonder though if the cons of doing such a thing wouldn't be worth it, at least in these special dynamic sorting cases. Still, I always feel dirty whenever I do dynamic SQL strings like that — like I'm still living in the Classic ASP world.
• A lot of the reason we want stored procedures in the first place is for security. I don't get to make the call on security concerns, only suggest solutions. With SQL Server 2005 we can set permissions (on a per-user basis if need be) at the schema level on individual stored procedures and then deny any queries against the tables directly. Critiquing the pros and cons of this approach is perhaps for another question, but again it's not my decision. I'm just the lead code monkey. :)

Answer 1

When you are paging sorted results, dynamic SQL is a good option. If you're paranoid about SQL injection you can use the column numbers instead of the column name. I've done this before using negative values for descending. Something like this...

declare @o int;
set @o = -1;

declare @sql nvarchar(2000);
set @sql = N'select * from table order by ' + 
    cast(abs(@o) as varchar) + case when @o < 0 then ' desc' else ' asc' end + ';'

exec sp_executesql @sql

Then you just need to make sure the number is inside 1 to # of columns. You could even expand this to a list of column numbers and parse that into a table of ints using a function like this. Then you would build the order by clause like so...

declare @cols varchar(100);
set @cols = '1 -2 3 6';

declare @order_by varchar(200)

select @order_by = isnull(@order_by + ', ', '') + 
        cast(abs(number) as varchar) + 
        case when number < 0 then ' desc' else '' end
from dbo.iter_intlist_to_tbl(@cols) order by listpos

print @order_by

One drawback is you have to remember the order of each column on the client side. Especially, when you don't display all the columns or you display them in a different order. When the client wants to sort, you map the column names to the column order and generate the list of ints.

Answer 2

There's a couple of different ways you can hack this in.

Prerequisites:

1. Only one SELECT statement in the sp
2. Leave out any sorting (or have a default)

Then insert into a temp table:

create table #temp ( your columns )

insert #temp
exec foobar

select * from #temp order by whatever

Method #2: set up a linked server back to itself, then select from this using openquery: http://www.sommarskog.se/share_data.html#OPENQUERY

Answer 3

An argument against doing the sorting on the client side is large volume data and pagination. Once your row count gets beyond what you can easily display you're often sorting as part of a skip/take, which you probably want to run in SQL.

For Entity Framework, you could use a stored procedure to handle your text search. If you encounter the same sort issue, the solution I've seen is to use a stored proc for the search, returning only an id key set for the match. Next, re-query (with the sort) against the db using the ids in a list (contains). EF handles this pretty well, even when the ID set is pretty large. Yes, this is two round trips, but it allows you to always keep your sorting in the DB, which can be important in some situations, and prevents you from writing a boatload of logic in the stored procedure.

Answer 4

A stored procedure technique (hack?) I've used to avoid dynamic SQL for certain jobs is to have a unique sort column. I.e.,

SELECT
   name_last,
   name_first,
   CASE @sortCol WHEN 'name_last' THEN [name_last] ELSE 0 END as mySort
FROM
   table
ORDER BY 
    mySort

This one is easy to beat into submission -- you can concat fields in your mySort column, reverse the order with math or date functions, etc.

Preferably though, I use my asp.net gridviews or other objects with build-in sorting to do the sorting for me AFTER retrieving the data fro Sql-Server. Or even if it's not built-in -- e.g., datatables, etc. in asp.net.

Answer 5

Sorry I'm late to the party, but here's another option for those who really want to avoid dynamic sql, but want the flexibility it offers:

Instead of dynamically generating the sql on the fly, write code to generate a unique proc for every possible variation. Then you can write a method in code to look at the search options and have it choose the appropriate proc to call.

If you only have a few variations then you can just create the procs by hand. But if you have a lot of variations then instead of having to maintain them all, you would just maintain your proc generator instead to have it recreate them.

As an added benefit, you'll get better sql plans for better performance doing it this way too.

Answer 6

Yeah, it's a pain, and the way you're doing it looks similar to what I do:

order by
case when @SortExpr = 'CustomerName' and @SortDir = 'ASC' 
    then CustomerName end asc, 
case when @SortExpr = 'CustomerName' and @SortDir = 'DESC' 
    then CustomerName end desc,
...

This, to me, is still much better than building dynamic SQL from code, which turns into a scalability and maintenance nightmare for DBAs.

What I do from code is refactor the paging and sorting so I at least don't have a lot of repetition there with populating values for @SortExpr and @SortDir.

As far as the SQL is concerned, keep the design and formatting the same between different stored procedures, so it's at least neat and recognizable when you go in to make changes.