I have a SSIS job that loads daily data from CSV files into a raw staging table in a SQL Server database (technically is an Azure SQL database). Then after that, a stored procedure is executed to copy all rows from the raw table into a production table, whilst performing the necessary castings, handling of dates, etc. Here is a simplified example of the stored procedure code:

USE [example_database]

/****** Object:  StoredProcedure [dbo].[Append_Data]    Script Date: 19/07/2016 11:21:34 ******/


create procedure [dbo].[Append_Data] as
--Code to delete updated records 

--Code to perform the append of incremental records 
insert into dbo.[Admissions]
SELECT [refno]
      ,cast([modified_date] as datetime)
  FROM [RAW].[Admissions]


I need some help to amend this code so it can handle duplicates as follows: Based on the column that uniquely identifies each row (named refno -- a candidate key), I need to replace any existing records with the same refno in the production key, as long as the modified date column (named date_modified -- a datetime column) is bigger (newer) than what is already in the production table.

Here is an example:


refno, name, value, date_modified
1, Alex, 2112.34, 08/Jul/2016 15:37:00 
2, Geddy, 674.91, 08/Jul/2016 15:37:00
3, Neil, 914.73, 08/Jul/2016 15:37:00


refno, name, value, date_modified
1, Alex, 1001.93, 10/Jul/2016 12:27:00 
2, Geddy, 674.91, 08/Jul/2016 15:37:00
3, Neil, 425.85, 01/Jul/2016 05:13:00
4, Hugh, 9807, 01/Jan/2016 13:42:00
  • The first row should get replaced in the production table because a row with the same refno exists in the raw table, and it has a newer modified date.
  • The second row has not been modified, so we leave it as is. Even though the matching row does exist in the target production table, we can see that the date_modified is the same so we ignore it.
  • The third row has been modified, but the modified date is earlier than what we have in the production table, so we ignore it.
  • The fourth row in the staging table is new (i.e. a row with that refno does not exist in the production table yet) so we copy it over.

Does that make sense? We don't tend to get many rows per day (about 1000 tops) and it is a straightforward, denormalised table (hence the stored procedure rather than going nuts in SSIS :) ).

Any help is greatly appreciated.


2 Answers 2

merge production as target  
using raw as source  
on target.refno = source.refno
when matched and souce.date > target.date 
     then update set target.date = souce.date, target.value= souce.value 
when not matched  
     then insert ...



Thanks for the suggestion, but I want to replace rows -- not update field by field in every modified row. So after some research I think that the best thing I would have to do is to first delete all the old rows that will be replaced with new data.

              FROM TableB
              WHERE TableB.refno = TableA.refno
                AND TableB.date_modified > TableA.date_modified

Then after this I can then load everything that is in the raw table into the production table. That will then include the new rows as well.

PS: Also added a truncate for my raw (staging) table at the end of my stored procedure in order to get it ready for the next load.

  • I don't see how it should matter whether you actually replace (delete, then insert) a row or update it column by column if the end result is the same for all intents and purposes. That said, your approach might be better than using MERGE if we take into consideration that MERGE is known to have had issues, some of which may still be unresolved. (Although I'm not sure how many of those apply to Azure SQL Database as I've mostly heard of them with regard to SQL Server.)
    – Andriy M
    Jul 20, 2016 at 6:38
  • You could still use UPDATE + INSERT, though, and, as I've just said, the result would be the same.
    – Andriy M
    Jul 20, 2016 at 6:40
  • Hi Andriy. The deletion of the rows that must be replaced instead of updating them seems to result in a cleaner and easier code. I will post my final store procedure code to complement this post as soon as I get a chance.
    – pmdci
    Jul 20, 2016 at 7:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.