Add incremental count to group of rows

Question

Post reply

Add incremental count to group of rows

simon.letts

Hall of Fame

Points: 3876
More actions
January 9, 2008 at 5:10 am

#70690

Hi, I need to write a T-SQL(2000) extract based on existing tables to add a sequence number to each row in a group of like transactions. The table has a PK on the ID column only. After that there is a transaction_code column which will have a number which can be duplicated if there were more than one items sold within that transaction. ie
id trans_code
dfg1 5554
dfg2 5554
dfg3 5554
dfg4 5553
dfg5 5553
So in my select statement I need to specify a seq code for each row in a group of transactions, so I'd get 3 rows back from trans 5554 with a seq 1, 2, 3 and 2 rows back for code 5553 with a seq of 1,2 and so on. I cant change the existing tables, this is just a query which will show a seq number for the rows retrieved.
thank-you

Viewing 15 posts - 1 through 15 (of 26 total)

You must be logged in to reply to this topic. Login to reply

Ken McKelvey SSCoach Points: 18856 More actions · Answer 1

In SQL2000 you will need to use a #temp table:

http://www.sqlservercentral.com/Forums/Topic397675-8-1.aspx

simon.letts Hall of Fame Points: 3876 More actions · Answer 2

Hi thanks for that but I cant see that the result of that query is the same as I need. Have you an actual example of what I need please.

Thanks

Matt Miller (4) SSC Guru Points: 124210 More actions · Answer 3

Simon - could you post some more specifics as to the DDL of that table (i.e. column names and data types). Ken's correct and the example gives you what you'd need, but it will help if the example looked like your table. If you post your DDL we can try to adapt the logic to your table.

----------------------------------------------------------------------------------
Your lack of planning does not constitute an emergency on my part...unless you're my manager...or a director and above...or a really loud-spoken end-user..All right - what was my emergency again?

Ken McKelvey SSCoach Points: 18856 More actions · Answer 4

1. Create a #temp table with an IDENTITY column

2. Insert your data into the #temp table

3. Join the #temp table to a derived table containing the MINIMUM(ID) for each tans_code so that seq (IDNo) can be calculated.

Among outher things, this is what the link shows!

simon.letts Hall of Fame Points: 3876 More actions · Answer 5

Thanks guys, here is some code. As you'll see its part of an overall extract not just a one off query in itself..

SELECT a.trans_code

,asize.adjustmentsize_id

,'LG'

,s.movement_type

,s.store_code

, -- Here is where I also need to display the increment value per trans_code

...................other DDL

FROMadjustment a,

adjustmentsize asize,

itemcolour ic,

itemcoloursize ics,

SAP_GM_Codes s

WHEREa.trans_code = asize.trans_code

ANDics.itemcolour_id = ic.itemcolour_id

ANDasize.itemcoloursize_id = ics.itemcoloursize_id

ANDic.itemcolour_ref not in ('20816','20817','62700','64543','22432','22433','22434','22435','01618','01002','01005','01000')

ANDs.movement_type = 'J'

ANDs.movement_code = a.trans_code

ANDasize.itemcoloursize_id IS NOT NULL

So for the trans_code, instead of just seeing

trans_code

5554

5553

I'd like

trans_code

5554 1

5554 2

5554 3

5553 1

5553 2

Thanks v much

Matt Miller (4) SSC Guru Points: 124210 More actions · Answer 6

Here's an example using somethig approximating your data. You weren't really clear on the table layout, etc... so it's going to be tough to know.

Anyway - see if this helps

--Set up test data

if object_ID('adjustment','u')>0

drop table adjustment

create table adjustment (iD int identity(1,1) primary key clustered, transcode varchar(20), dateval datetime)

go

insert adjustment(transcode, dateval)

select 'aaa1', getdate()-rand()*20 UNION ALL

select 'bbb1', getdate()-rand()*20 UNION ALL

select 'ccc1', getdate()-rand()*20 UNION ALL

select 'ddd1', getdate()-rand()*20 UNION ALL

select 'eee1', getdate()-rand()*20 UNION ALL

select 'fff1', getdate()-rand()*20 UNION ALL

select 'hhh1', getdate()-rand()*20 UNION ALL

select 'mmm1', getdate()-rand()*20 UNION ALL

select 'aaa1', getdate()-rand()*20 UNION ALL

select 'bbb1', getdate()-rand()*20 UNION ALL

select 'ccc1', getdate()-rand()*20 UNION ALL

select 'ddd1', getdate()-rand()*20 UNION ALL

select 'eee1', getdate()-rand()*20 UNION ALL

select 'fff1', getdate()-rand()*20 UNION ALL

select 'hhh1', getdate()-rand()*20 UNION ALL

select 'mmm1', getdate()-rand()*20 UNION ALL

select 'aaa1', getdate()-rand()*20 UNION ALL

select 'bbb1', getdate()-rand()*20 UNION ALL

select 'ccc1', getdate()-rand()*20 UNION ALL

select 'ddd1', getdate()-rand()*20 UNION ALL

select 'eee1', getdate()-rand()*20 UNION ALL

select 'fff1', getdate()-rand()*20 UNION ALL

select 'hhh1', getdate()-rand()*20 UNION ALL

select 'mmm1', getdate()-rand()*20 UNION ALL

select 'aaa1', getdate()-rand()*20 UNION ALL

select 'bbb1', getdate()-rand()*20 UNION ALL

select 'ccc1', getdate()-rand()*20 UNION ALL

select 'ddd1', getdate()-rand()*20 UNION ALL

select 'eee1', getdate()-rand()*20 UNION ALL

select 'fff1', getdate()-rand()*20 UNION ALL

select 'hhh1', getdate()-rand()*20 UNION ALL

select 'mmm1', getdate()-rand()*20

--go 30

--done with test data setup

--Now - for the seqno part

declare @prevtrans varchar(20)

declare @dummy int

declare @i int

drop table #tmpTransWithSeq

--create the temp table

--needs to include all fields to return plus the original PK for ordering plus SEQNO

create table #tmpTransWithSeq

(

TransCode varchar(20),

ID int,

seqno int

)

--create a clustered index to make the SEQNO work

--needs to be created with transcode on the LEFT, since that's the grouping

create index UC_tmpTransWithSeq on #tmpTransWithSeq(TRANSCODE,ID)

--now, insert the results you need

insert #tmpTransWithSeq

(

TransCode,

ID)

select TransCode, ID from adjustment

--now run the update to create the sequence numbers

--but first - set up the variables

select @prevtrans='',@i=0, @dummy=0

update #tmpTransWithSeq

set

seqno=@i+1,

@i=CASE when @prevtrans=transcode then @i+1 else 0 end,

@dummy=@i,

@prevtrans=transcode

from #tmpTransWithSeq

WITH (INDEX(UC_tmpTransWithSeq))--this indexing hint set up the order

OPTION (MAXDOP 1) --use just one processor, so that the order is respected

--now show the result

select * from #tmpTransWithSeq order by transcode,seqno

----------------------------------------------------------------------------------
Your lack of planning does not constitute an emergency on my part...unless you're my manager...or a director and above...or a really loud-spoken end-user..All right - what was my emergency again?

simon.letts Hall of Fame Points: 3876 More actions · Answer 7

Thanks for that effort, it looks very complicated for what I want though.

To make it easier ...lets say that I just have a table with just one column, trans_code and it has entires such as

5554

5553

For every like entry i want to run a SELECT extract to display the row and a sequnce number within the trans_code

5554 1

5554 2

5554 3

5553 1

5553 2

However this will be running on an existing table as part of an overal select statement so I cant use a new table with identity or use an insert statement. the data (trans_code) is already there I simply want to display a sequence number, not add it to the table.

thanks

Matt Miller (4) SSC Guru Points: 124210 More actions · Answer 8

The only other way to do that without a temporary table would be a correlated sub-query, which is going to kill your performance. If your adjustments table is big - this is going to peg your server big-time.

If you're talking about a "normal size" table - don't be surprised if performance on the following sucks...

select

t1.*,

seqno=(

select count(*)

from adjustment t2

where t1.transcode=t2.transcode and

t1.id>=t2.id

)

from adjustment t1

order by transcode,seqno

----------------------------------------------------------------------------------
Your lack of planning does not constitute an emergency on my part...unless you're my manager...or a director and above...or a really loud-spoken end-user..All right - what was my emergency again?

Jeff Moden SSC Guru Points: 1004406 More actions · Answer 9

Just in case you want to know why Matt says the performance will suck...

http://www.sqlservercentral.com/articles/T-SQL/61539/

--Jeff Moden

RBAR is pronounced "ree-bar" and is a "Modenism" for Row-By-Agonizing-Row.
First step towards the paradigm shift of writing Set Based code:
________Stop thinking about what you want to do to a ROW... think, instead, of what you want to do to a COLUMN.

Change is inevitable... Change for the better is not.

Helpful Links:
How to post code problems
How to Post Performance Problems
Create a Tally Function (fnTally)

simon.letts Hall of Fame Points: 3876 More actions · Answer 10

Thanks for all your help guys. Matt, your script did the job fine!!!

However we are currently testing against a limited set of rows so I will insist we do further tests against a full copy of the live DB so I can assess performance.

thanks again

Sergiy SSC Guru Points: 110208 More actions · Answer 11

simon.letts (1/10/2008)
Thanks for all your help guys. Matt, your script did the job fine!!!
However we are currently testing against a limited set of rows so I will insist we do further tests against a full copy of the live DB so I can assess performance.
thanks again

Just many rows not suppose to be a problem.

You should expect performance issues when you have many rows per group.

And, of course, indexing must be done well.

_____________
Code for TallyGenerator

Peter E. Kierstead SSCarpal Tunnel Points: 4114 More actions · Answer 12

Nobody mentioned that you could use UPDATE. Add in integer column and try this:

Declare @PrevGroupingColumn VarChar(256),@i Int

Select @PrevGroupingColumn='' -- Any value that would not occur in the grouping column.

Update dbo.YourTable Set

@i=Case when GroupingColumn=@PrevGroupingColumn then @i+1 else 1 End,

SequenceColumn=@i,@PrevGroupingColumn=GroupingColumn

Adjust the above to match your column names and data types and consider performance if YourTable is large...

PeteK
I have CDO. It's like OCD but all the letters are in alphabetical order... as they should be.

Jeff Moden SSC Guru Points: 1004406 More actions · Answer 13

Careful... that's only guaranteed to work correctly if one of two things happen...

1. You have a rock solid clustered index on the correct columns and you have an "anchor" column (thanks, Matt Miller) in the update AND you force the index with a WITH(INDEX()). Otherwise, you could end up with what Gail Shaw (GilaMonster) refers to as a "Merry-Go-Round" index that will give an incorrect answer.

2. You have a TOP 100 PERCENT/ORDER BY as a "control" sub-query in the FROM clause of the update. Of course, that will greatly reduce the performance compared to method 1 above although it's still likely to be faster than a Cursor or While Loop.

Both methods above, as well as the Cursor or While Loop method will be faster than using a correlated sub-query with a "Triangular Join"... please refer to the following article for what the impact on performance is for "Triangular Joins"...

http://www.sqlservercentral.com/articles/T-SQL/61539/

I will say that, due to caching, if the groups of rows to be "ranked" or counted are very small, the correlated sub-query method won't make a huge impact on performance an may probably suffice for most not concerned with scalability of the data.

There is a very high performance alternate method to doing this type of grouped ranking which uses a Tally table... I'll see if I can dig it up...

--Jeff Moden

RBAR is pronounced "ree-bar" and is a "Modenism" for Row-By-Agonizing-Row.
First step towards the paradigm shift of writing Set Based code:
________Stop thinking about what you want to do to a ROW... think, instead, of what you want to do to a COLUMN.

Change is inevitable... Change for the better is not.

Helpful Links:
How to post code problems
How to Post Performance Problems
Create a Tally Function (fnTally)

simon.letts Hall of Fame Points: 3876 More actions · Answer 14

Thanks again everyone for your valued input. Fortunately although the table is large the actual groups to be ranked are always less than 5.