Question

Let’s say that I have a table of items, and for each item, there can be additional information stored for it, which goes into a second table. The additional information is referenced by a FK in the first table, which can be NULL (if the item doesn’t have additional info).

TABLE item (
    ...
    item_addtl_info_id INTEGER
)

CONSTRAINT fk_item_addtl_info FOREIGN KEY (item_addtl_info)
    REFERENCES addtl_info (addtl_info_id)

TABLE addtl_info (
    addtl_info_id INTEGER NOT NULL
    GENERATED BY DEFAULT 
    AS IDENTITY (
        INCREMENT BY 1
        NO CACHE
        ),
    addtl_info_text VARCHAR(100)
    ...
    CONSTRAINT pk_addtl_info PRIMARY KEY (addtl_info_id)
)

What is the “best practice” to update an item’s additional info (in IBM DB2 SQL, preferably)?

It should be an UPSERT operation, meaning that if additional info does not yet exist then a new record is created in the second table, but if it does, then it is only updated, and the FK in the first table does not change.

So imperatively, this is the logic:

UPSERT(item, item_info):
CASE WHEN item.item_addtl_info_id IS NULL THEN
    INSERT INTO addtl_info (item_info)
    UPDATE item.item_addtl_info_id (addtl_info.addtl_info_id)
                                               ^^^^^^^^^^^^^
ELSE
    UPDATE addtl_info (item_info)
END

My main problem is how to get the newly inserted addtl_info row’s id (underlined above). In a stored proc I can request the id from a sequence and store it in a variable, but maybe there is a more straightforward way. Isn’t it something that comes up all the time when programming databases?

I mean, I’m really not interested in what the id of the addtl_info record is as long as it remains unique and is referenced properly. So using sequences seems a bit of an overkill to me in this case.

As a matter of fact, this UPSERT operation should be part of the SQL language as a standard operation (maybe it is, and I just don’t know about it?)…

Answer 1

The syntax I was looking for is:

SELECT * FROM NEW TABLE ( INSERT INTO phone_book VALUES ( 'Peter Doe','555-2323' ) )

from Wikipedia (http://en.wikipedia.org/wiki/Insert_%28SQL%29)

This is how to refer to the record that was just inserted in the table.

My colleague called this construct an “in-place trigger”, which what it really is…

Here is the first version that I put together as a compound SQL statement:

begin atomic
declare addtl_id integer;
set addtl_id = (select item_addtl_info_id from item where item.item_id = XXX);
if addtl_id is null
then
    set addtl_id = (select addtl_info_id from new table 
                    (insert into addtl_info
                        (addtl_info_text)
                        values ('My brand new additional info')
                    )
                  );
    update item set item.item_addtl_info_id = addtl_id
    where item.item_id = XXX;

else
   update addtl_info set addtl_info_text = 'My updated additional info'
   where addtl_info.addtl_info_id = addtl_id;
end if;
end

XXX being equal to the item id to be updated – this code can now be easily inserted into a sproc, and XXX can be converted to an input parameter.

I also tried using MERGE INTO, but I couldn’t figure out a syntax for updating a table different from what was specified as the target.