T-sql Reset Row number on Field Change

Question

I’ve struggled with this for three full days now and I can’t get my head around this. It’s quite similar to a recent post of mine “t-sql sequential duration”, but not exactly the same…I want to reset the row number based on a change in column x (in my case, column “who”)…

Here’s the first query that returns the a small sample of the raw(ish) data:

SELECT      DISTINCT chr.custno, 
            CAST(LEFT(CONVERT( VARCHAR(20),chr.moddate,112),10)+ ' ' + chr.modtime AS DATETIME)as  moddate, 
            chr.who     
FROM        <TABLE> chr 
WHERE       chr.custno = 581827
            AND LEFT(chr.who, 5) = 'EMSZC'
            AND chr.[description] NOT LIKE 'Recalled and viewed this customer'
ORDER BY    chr.custno

Result:

custno      moddate             who
581827      2012-11-08 08:38:00.000     EMSZC14
581827      2012-11-08 08:41:10.000     EMSZC14
581827      2012-11-08 08:53:46.000     EMSZC14
581827      2012-11-08 08:57:04.000     EMSZC14
581827      2012-11-08 08:58:35.000     EMSZC14
581827      2012-11-08 08:59:13.000     EMSZC14
581827      2012-11-08 09:00:06.000     EMSZC14
581827      2012-11-08 09:04:39.000     EMSZC49 Reset row number to 1
581827      2012-11-08 09:05:04.000     EMSZC49
581827      2012-11-08 09:06:32.000     EMSZC49
581827      2012-11-08 09:12:03.000     EMSZC49
581827      2012-11-08 09:12:38.000     EMSZC49
581827      2012-11-08 09:14:18.000     EMSZC49
581827      2012-11-08 09:17:35.000     EMSZC14 Reset row number to 1

Second step is to add the row number (I didn’t do this in the first query because of the use of the word DISTINCT); so…

WITH c1 AS (
        SELECT      DISTINCT chr.custno
                    CAST(LEFT(CONVERT( VARCHAR(20),chr.moddate,112),10)+ ' ' + chr.modtime AS DATETIME)as moddate,
                    chr.who
        FROM        <TABLE> chr 
        WHERE       chr.custno = 581827
                    AND LEFT(chr.who, 5) = 'EMSZC'
                    AND chr.[description] NOT LIKE 'Recalled and viewed this customer'
        )

SELECT  ROW_NUMBER() OVER (PARTITION BY custno ORDER BY custno, moddate, who) AS RowID, custno, moddate, who
FROM    c1

Result:

RowID   custno      moddate                      who
1       581827      2012-11-08 08:38:00.000     EMSZC14
2       581827      2012-11-08 08:41:10.000     EMSZC14
3       581827      2012-11-08 08:53:46.000     EMSZC14
4       581827      2012-11-08 08:57:04.000     EMSZC14
5       581827      2012-11-08 08:58:35.000     EMSZC14
6       581827      2012-11-08 08:59:13.000     EMSZC14
7       581827      2012-11-08 09:00:06.000     EMSZC14
8       581827      2012-11-08 09:04:39.000     EMSZC49 Reset row number to 1
9       581827      2012-11-08 09:05:04.000     EMSZC49
10      581827      2012-11-08 09:06:32.000     EMSZC49
11      581827      2012-11-08 09:12:03.000     EMSZC49
12      581827      2012-11-08 09:12:38.000     EMSZC49
13      581827      2012-11-08 09:14:18.000     EMSZC49
14      581827      2012-11-08 09:17:35.000     EMSZC14 Reset row number to 1

The next step is where I’m stuck: the goal is to reset the RowID to 1 on each change of value in the “who” column. The following code gets an “almost there” result (and it should be noted that I stole/borrowed this code from somewhere, but now I can’t find the website):

WITH c1 AS (
        SELECT      DISTINCT chr.custno,
                    CAST(LEFT(CONVERT( VARCHAR(20),chr.moddate,112),10)+ ' ' + chr.modtime AS DATETIME)as moddate,
                    chr.who
        FROM        <TABLE> chr 
        WHERE       chr.custno = 581827
                    AND LEFT(chr.who, 5) = 'EMSZC'
                    AND chr.[description] NOT LIKE 'Recalled and viewed this customer'
        )
, c1a AS    (
            SELECT  ROW_NUMBER() OVER (PARTITION BY custno ORDER BY custno, moddate, who) AS RowID, custno, moddate, who
            FROM    c1
            )

SELECT  x.RowID - y.MinID + 1 AS Row,
        x.custno, x.Touch, x.moddate, x.who      
FROM    (
            SELECT  custno, who, MIN(RowID) AS MinID
            FROM    c1a
            GROUP BY custno, who
        ) AS y
        INNER JOIN c1a x ON x.custno = y.custno AND x.who = y.who

Result:

Row custno      moddate                    who
1   581827      2012-11-08 08:38:00.000     EMSZC14
2   581827      2012-11-08 08:41:10.000     EMSZC14
3   581827      2012-11-08 08:53:46.000     EMSZC14
4   581827      2012-11-08 08:57:04.000     EMSZC14
5   581827      2012-11-08 08:58:35.000     EMSZC14
6   581827      2012-11-08 08:59:13.000     EMSZC14
7   581827      2012-11-08 09:00:06.000     EMSZC14
1   581827      2012-11-08 09:04:39.000     EMSZC49 Reset row number to 1 (Hooray! It worked!)
2   581827      2012-11-08 09:05:04.000     EMSZC49
3   581827      2012-11-08 09:06:32.000     EMSZC49
4   581827      2012-11-08 09:12:03.000     EMSZC49
5   581827      2012-11-08 09:12:38.000     EMSZC49
6   581827      2012-11-08 09:14:18.000     EMSZC49
14  581827      2012-11-08 09:17:35.000     EMSZC14 Reset row number to 1 (Crappies.)

DESIRED result:

Row custno      moddate                     who
1   581827      2012-11-08 08:38:00.000     EMSZC14
2   581827      2012-11-08 08:41:10.000     EMSZC14
3   581827      2012-11-08 08:53:46.000     EMSZC14
4   581827      2012-11-08 08:57:04.000     EMSZC14
5   581827      2012-11-08 08:58:35.000     EMSZC14
6   581827      2012-11-08 08:59:13.000     EMSZC14
7   581827      2012-11-08 09:00:06.000     EMSZC14
1   581827      2012-11-08 09:04:39.000     EMSZC49 Reset row number to 1 
2   581827      2012-11-08 09:05:04.000     EMSZC49
3   581827      2012-11-08 09:06:32.000     EMSZC49
4   581827      2012-11-08 09:12:03.000     EMSZC49
5   581827      2012-11-08 09:12:38.000     EMSZC49
6   581827      2012-11-08 09:14:18.000     EMSZC49
1   581827      2012-11-08 09:17:35.000     EMSZC14 Reset row number to 1

Any assistance is appreciated. You can giggle at me, too, because I’m sure this is quite simple to resolve – I just can’t seem to get out of my own way.

Thanks.

Answer 1

The only solution I can think of is to use a cursor (ugh) and suffer through the RBAR process. NOT an elegant solution as the cursor will have to read in excess of 1m rows. Bummer.

Answer 2

Instead of:

PARTITION BY custno ORDER BY custno, moddate, who)

try:

PARTITION BY custno, who ORDER BY custno, moddate)

Answer 3

If you are on SQL Server 2012 you can use LAG to compare value with previous row and you can use SUM and OVER to record the changes.

with C1 as
(
  select custno,
         moddate,
         who,
         lag(who) over(order by moddate) as lag_who
  from chr
),
C2 as
(
  select custno,
         moddate,
         who,
         sum(case when who = lag_who then 0 else 1 end) 
            over(order by moddate rows unbounded preceding) as change 
  from C1
)
select row_number() over(partition by change order by moddate) as RowID,
       custno,
       moddate,
       who
from C2

SQL Fiddle

Update:

A version for SQL Server 2005. It uses a recursive CTE and a temp table for intermediary storage of the data you need to iterate over.

create table #tmp
(
  id int primary key,
  custno int not null,
  moddate datetime not null,
  who varchar(10) not null
);

insert into #tmp(id, custno, moddate, who)
select row_number() over(order by moddate),
       custno,
       moddate,
       who
from chr;

with C as
(
  select 1 as rowid,
         T.id,
         T.custno,
         T.moddate,
         T.who,
         cast(null as varchar(10)) as lag_who
  from #tmp as T
  where T.id = 1
  union all
  select case when T.who = C.who then C.rowid + 1 else 1 end,
         T.id,
         T.custno,
         T.moddate,
         T.who,
         C.who
  from #tmp as T
    inner join C
      on T.id = C.id + 1
)
select rowid,
       custno,
       moddate,
       who
from C
option (maxrecursion 0);

drop table #tmp;

SQL Fiddle

Answer 4

I had success with this issue by using Rank():

SELECT RANK() OVER (PARTITION BY who ORDER BY custno, moddate) AS RANK

This returned your desired results. I actually found this post trying to solve the same problem.