SQL Help (Insert and Update)

Question

I need to use a SQL command like MERGE that is in 9i; however, I am using 8.1.7 which MERGE does not exist. I want to update table "U" with a value if the key columns match with the key columns in table "E". If they do not match, I want to insert a row into table "U". How would the SQL statement be written??

Answer 1

The short answer is: use 2 SQL statements.
I've used the 2-statement approach many times. Its funny that I've waited, and waited for the MERGE command for years. Now that its available in 9i, I haven't had an opportunity to use it. However, considering how complicated the "upsert" operation can get, I'd bet that the MERGE command is pretty complicated...
First off, to save yourself headaches make sure you take care of duplicates in your source table (table E in Douglas' question).
Now, the 2-SQL's would look something like this:
update u
set   (u.col1,
       u.col2) =
           (select e.col1,
                   e.col2
            from   e
            where  e.pk = u.pk)
where  exists
   (select 1
    from   e e2           /* notice alias e2 used for table e */
    where  e2.pk = u.pk)
insert into u
    (pk, col1, col2)
  select e.pk,
         e.col1,
         e.col2
  from   e
  where  not exists
    (select 1
     from   u  u2          /* notice alias u2 used for table u */
     where  u2.pk = e.pk)
--There are many variations on this scheme. One is to have a "status" column in the source table. When there are a small number of updates (or a small number of inserts), you can identify them and set a "status flag". Then take care of the small number of transactions (based on the flag). Now, you can delete the small number of rows and run a much simpler SQL for the majority of the rows.
Also, you may find it easier to use PL/SQL (as APC was mentioning). However, using the PL/SQL approach gets slower the more rows that are processed. An example might look something like:
begin
  for c in (select * from e)
  loop
    update u
    set    col1 = c.col1,
           col2 = c.col2
    where  e.pk = c.pk ;
    if sql%rowcount = 0
    then
      insert into u
         (pk, col1, col2)  values (c.pk, c.col1, c.col2) ;
    end if ;
  end loop ;
  commit ;
end ;
/Avoid this PL/SQL method if you have more than 10,000 rows. This "looping" approach can be optimized with "bulk-PL/SQL" available in Oracle 8i.
Typically, this PL/SQL approach is used when major non-SQL calculations must be done during the loading process...
HTH,
Stan