优化同一个表中的多个选择

Question

I want to optimize the query that is consumed by a report. Unfortunately, I cannot modify the report, so I have to provide a specificly formatted dataset.

So, let's say I have a table that looks like this(In practice, it has 25 columns and 20k rows):

Name    Description           Price     MiscColumn1    MiscColumn2    
Tea      test description      10       misc1            misc2   
Coffee    test desc            20       misc3            misc4
Water      test                20       misc1            misc2

So, I need to transform this dataset to look like this:

Type  Name    Description           Price     MiscColumn1    MiscColumn2 
  1    Tea     test description      NULL       NULL           NULL  
  1    Coffee   test desc            NULL       NULL           NULL 
  1    Water     test                NULL       NULL           NULL
  2    NULL      NULL                 10        NULL           NULL
  2    NULL      NULL                 20        NULL           NULL  
  3    NULL      NULL                NULL       misc1          misc2
  3    NULL      NULL                NULL       misc3          misc4  

So, basically what I need to do is to select 3 groups of distinct records back into the dataset.

What I currently do is:

Create #tempTable  

And then do a 3 separate distinct selects like this:

insert into #tempTable (Name, Description)  
select distinct Name, DEscription from myTable  
 insert into #tempTable (Price)  
select distinct Price from myTable   

But it is really slow and can take up to 5 seconds to complete with my data.

Also, I was trying to use UNION, but I didn't gain any performance improvement.

Answer 1

You can do this in a single statement, which should involve a single scan, like this:

SELECT DISTINCT
   X.*
FROM
   dbo.MyTable T
   CROSS APPLY (VALUES
      (1, T.Name, T.Description, NULL, NULL, NULL),
      (2, NULL, NULL, T.Price, NULL, NULL),
      (3, NULL, NULL, NULL, T.MiscColumn1, T.MiscColumn2)
   ) X (Type, Name, Description, Price, MiscColumn1, MiscColumn2)
;

See a Live Demo at SQL Fiddle

Note that you don't need a temporary table--you can do your 15 joins and then in the CROSS APPLY simply refer to the table that each column comes from.

That brings up a point. Your data is coming from 15 tables! If any of the Type groupings of values come from a distinct subset of tables, then this is probably not the best way to do it! Let's say, for example, that MiscColumn1 and MiscColumn2 come from 2 tables that have no columns represented in another group. In that case, it will be much better to remove those 2 tables from the main query, and UNION ALL SELECT just the 2 columns from those tables separately.

I'm saying this based on the possibly mistaken impression I am getting that your reporting platform is going to do its own joining of various related data. If so, then you shouldn't try to put together the unified view of all the data, then break it back down again--that is putting extra work on the system for no reason. The need for the DISTINCT in the above query highlights the extra memory, I/O, and CPU that will be required to materialize the trimmed-down result set you need. If there's any way to get around that, I think you should do it.