How to copy tables avoiding cursors in SQL?

Question

I want to write script in SQL that will copy these 2 tables(A,B) to other 2 tables(C,D) with the same structure as A,B accordingly.

IMPORTANT:

Tables C,D are NOT necessary empty
Several processes may call script simultaneously

Table A has foreign key(fk_a_b) of table B

   ________________________  _________________
   |        Table A       |  |   Table B     |  
   |______________________|  |_______________|
   | id     FK_A_B   name |  | id    visible |
   | ----- -------- ------|  | ----- --------|
   | 1      21       n1   |  | 21     true   |
   | 5      32       n2   |  | 32     false  |
   ------------------------  -----------------

Let say that after copying table B to D this is what I get

   ________________
   |   Table D    |  
   |______________|
   | id   visible |
   | ----- -------|
   | 51    true   |
   | 52    false  |
   ----------------

Now, when I'll copy table A to C I need to know, somehow, that ID=21 maps now to ID=51, and ID=32 to ID=52. Finally, the table C will be:

   ________________________
   |        Table C       |
   |______________________|
   | id     FK_C_D   name |
   | ----- -------- ------|
   | 61      51       n1  |
   | 62      52       n2  |
   ------------------------

Because several processes may call script simultaneously, I CAN'T alter table A,B to add some helper columns. So, to achieve this I used CURSOR. I copied row by row of table B and managed temp table to map OldId to NewId(21->51,32->52) and then used this temp table to copy table A.

I've read that CURSOR is bad practice. So, is there another way to do it?

Thank you

You **can** keep same identity values from your source tables if you want. Is there any reason you want to create new ids? — d-live, May 30 '11 at 09:23
@d-live, I wrote that table are **NOT** empty. The identity values from source table can be already used in target tables — theateist, May 31 '11 at 05:36

score 7 · Accepted Answer · edited Jan 18 '21 at 12:30

You can use the output clause with the merge statement to get a mapping between source id and target id. Described in this question. Using merge..output to get mapping between source.id and target.id

Here is some code that you can test. I use table variables instead of real tables.

Setup sample data:

-- @A and @B is the source tables
declare @A as table
(
  id int,
  FK_A_B int,
  name varchar(10)
)

declare @B as table
(
  id int,
  visible bit
)  

-- Sample data in @A and @B
insert into @B values (21, 1),(32, 0)
insert into @A values (1, 21, 'n1'),(5, 32, 'n2')


-- @C and @D is the target tables with id as identity columns
declare @C as table
(
  id int identity,
  FK_C_D int not null,
  name varchar(10)
)

declare @D as table
(
  id int identity,
  visible bit
)  

-- Sample data already in @C and @D
insert into @D values (1),(0)
insert into @C values (1, 'x1'),(1, 'x2'),(2, 'x3')

Copy data:

-- The @IdMap is a table that holds the mapping between
-- the @B.id and @D.id (@D.id is an identity column)
declare @IdMap table(TargetID int, SourceID int)

-- Merge from @B to @D.
merge @D as D             -- Target table
using @B as B             -- Source table
on 0=1                    -- 0=1 means that there are no matches for merge
when not matched then
  insert (visible) values(visible)    -- Insert to @D
output inserted.id, B.id into @IdMap; -- Capture the newly created inserted.id and
                                      -- map that to the source (@B.id)

-- Add rows to @C from @A with a join to
-- @IdMap to get the new id for the FK relation
insert into @C(FK_C_D, name)
select I.TargetID, A.name 
from @A as A
  inner join @IdMap as I
    on A.FK_A_B = I.SourceID

Result:

select *
from @D as D
  inner join @C as C
    on D.id = C.FK_C_D

id          visible id          FK_C_D      name
----------- ------- ----------- ----------- ----------
1           1       1           1           x1
1           1       2           1           x2
2           0       3           2           x3
3           1       4           3           n1
4           0       5           4           n2

You can test the code here: https://data.stackexchange.com/stackoverflow/q/101643/using-merge-to-map-source-id-to-target-id

I will read about it more deeply, but at first glance it seems more complicated. Does it though more efficient then cursors? — theateist, May 30 '11 at 09:28
@theateist - It is hard to generalize about performance but in this case I think it is safe to say that performance is a lot better with merge than it is with cursors. — Mikael Eriksson, May 30 '11 at 09:30
@theateist - It is perhaps not obvious in the linked question, but you don't need to use a loop when using merge. In your case you will have one merge statement for B->D with the output of the mapped id's. and one insert statement A-C that joins the output table from the merge to find the new id's. — Mikael Eriksson, May 30 '11 at 09:33
it definitely looks complicated but it worth knowing. it is very useful. i'll go out on a limb and say that it will be more efficient than a cursor — Beno, May 30 '11 at 09:34
@Alex - The map between B and D in the merge is `on 0=1` meaning that there are no matches in the merge so all rows from `B` will be inserted to `D`. — Mikael Eriksson, May 30 '11 at 11:20

score 0 · Answer 2 · answered May 30 '11 at 09:40

E.g. SQL Server adds rowguid fields to tables that included in some publications for merge replication. I think, such approach can be used in your task. The idea is to add a couple of GUID fields that will play the role pf global identifiers, so we can use them in both pairs of master-datails tables

score 0 · Answer 3 · answered May 30 '11 at 09:55

You can do something like this:

if object_id('tempdb..#TableB') is not null
    drop table #TableB

select identity(int) RowId, *
into #TableB
from TableB

if object_id('tempdb..#TableDIds') is not null
    drop table #TableDIds
create table  #TableDIds (RowId int identity(1,1), Id int)

insert TableD
output inserted.Id into #TableDIds
select Visible
from #TableB
order by RowId

insert TableC
select tdi.Id, ta.name
from TableA ta
    join #TableB tb on
        ta.FK_A_B = tb.Id
    join #TableDIds tdi on
        tdi.RowId = tb.RowId

I used the following setup:

create table TableB
(
    Id int not null primary key,
    Visible bit not null
)

create table TableA
(
    Id int not null, 
    FK_A_B int not null foreign key references TableB(Id), 
    Name varchar(10) not null
)

create table TableD
(
    Id int identity(1,1) primary key,
    Visible bit not null
)

create table TableC
(
    Id int identity(1,1), 
    FK_C_D int not null references TableD(Id), 
    Name varchar(10) not null
)

insert TableB
values
    (21, 1),
    (32, 0)

insert TableA
values
    (1, 21, 'n1'),
    (5, 32, 'n2')

score 0 · Answer 4 · answered Jun 01 '11 at 10:01

0

I hope this helps

http://pastebin.com/embed_iframe.php?i=fkUd4RPY

answered Jun 01 '11 at 10:01

Nitin Sawant

7,278
9
52
98

How to copy tables avoiding cursors in SQL?

4 Answers4

Linked