Question

I’m writing tsql for SQL Server 2008. I’ve got two tables with roughly 2 million rows each. The Source table gets updated daily and changes are pushed to the Destination table based on a last_edit date. If this date is newer in source than destination then update the destination row. If a new row exists in source compared to destination insert it into destination. This is really only a one way process that I’m concerned with, from source to destination. The source and destination table use a unique identifier across 4 columns, serialid, itemid, systemcode, and role.

My table are modeled similar to the script below. There are many data columns but I’ve limited it to 3 in this example. I’m looking for 2 outputs. 1 set of data with rows to update and 1 set of data with rows to add.

CREATE TABLE [dbo].[TABLE_DEST](
  [SERIALID] [nvarchar](20) NOT NULL,
  [ITEMID] [nvarchar](20) NOT NULL,
  [SYSTEMCODE] [nvarchar](20) NOT NULL,
  [ROLE] [nvarchar](10) NOT NULL,
  [LAST_EDIT] [datetime] NOT NULL],
  [DATA_COLUMN_1] [nvarchar](10) NOT NULL,
  [DATA_COLUMN_2] [nvarchar](10) NOT NULL,
  [DATA_COLUMN_3] [nvarchar](10) NOT NULL
)

CREATE TABLE [dbo].[TABLE_SOURCE](
 [SERIALID] [nvarchar](20) NOT NULL,
 [ITEMID] [nvarchar](20) NOT NULL,
 [SYSTEMCODE] [nvarchar](20) NOT NULL,
 [ROLE] [nvarchar](10) NOT NULL,
 [LAST_EDIT] [datetime] NOT NULL],
 [DATA_COLUMN_1] [nvarchar](10) NOT NULL,
 [DATA_COLUMN_2] [nvarchar](10) NOT NULL,
 [DATA_COLUMN_3] [nvarchar](10) NOT NULL
)

Here’s what I’ve got for the update dataset.

select s.*
from table_dest (nolock) inner join table_source s (nolock) 
  on s.SYSTEMCODE = fd.SYSTEMCODE1Y
    and s.ROLE = d.ROLE
    and s.SERIALID = d.SERIALID
    and s.ITEMID = d.ITEMID
    and s.LAST_EDIT > d.LAST_EDIT

I don’t know how best to accomplish finding the rows to add. But the solution has to be pretty efficient for the database.

Answer 1

Unmatched rows can be found with left/right join and checking target table keys for null:

select s.*, case when d.key1 is null then 'insert' else 'update' end [action]
from [table_dest] d right join [table_source] s on (d.key1 = s.key1 /* etc.. */)

If you need these rows just to perform respective operations then there is special feature for you:

merge [table_dest] d
using [table_source] s on (d.key1 = s.key1 /* etc.. */)
when mathed then
   update set d.a = s.a
when not matched by target then
   insert (key1, .., a) values (s.key1, ..., s.a);