SQL Server - recursive references (loop, join, insert?) - sql-server

Question

I would appreciate if you could give me any hints regarding the fastest solution of the following SQL Server challenge:
Let's say I have a table with DATE, CLIENT and his several characteristics in other columns. I need to calculate COLUMN_1 and COLUMN_2 but:
COLUMN_1 uses the client's characteristics as of current DATE and as of previous DATE and COLUMN_1 value from the previous DATE (recursive referencing)
COLUMN_2 additionally uses COLUMN_1 value as of current date (therefore I would like to refer to its final value, not the particular 'case when' that implements the column logic)
How do I replicate this logic most efficiently in SQL Server?
I was thinking about the loop that goes over DATA and for each DATA, joins previous DATA, calculates firstly COLUMN_1, then COLUMN_2 (but how to make sure that the values in COLUMN_1 are accessible for COLUMN_2?)
Regards,
Bart

Answer 1

Without a specific example you we will not be able to tell you which solution would be the most efficient, especially when you are looking for a solution you describe as recursive. You might not need a full recursive solution if you could use window functions instead.
In sql server 2012+ you have access to lead() and lag() which you can use to get the previous and next values for a column based on a partition and order.
select
client
, date
, nextdate = lead(date) over (partition by client order by date)
, prevdate = lag(date) over (partition by client order by date)
, column1 = 'do stuff with lead/lag'
, column2 = 'do stuff with lead/lag'
from t
rextester example: http://rextester.com/FFHU71709
returns:
+--------+------------+------------+------------+------------------------+------------------------+
| client | date | nextdate | prevdate | column1 | column2 |
+--------+------------+------------+------------+------------------------+------------------------+
| 1 | 2017-01-01 | 2017-01-02 | NULL | do stuff with lead/lag | do stuff with lead/lag |
| 1 | 2017-01-02 | 2017-01-03 | 2017-01-01 | do stuff with lead/lag | do stuff with lead/lag |
| 1 | 2017-01-03 | NULL | 2017-01-02 | do stuff with lead/lag | do stuff with lead/lag |
| 2 | 2017-01-02 | 2017-01-04 | NULL | do stuff with lead/lag | do stuff with lead/lag |
| 2 | 2017-01-04 | 2017-01-06 | 2017-01-02 | do stuff with lead/lag | do stuff with lead/lag |
| 2 | 2017-01-06 | NULL | 2017-01-04 | do stuff with lead/lag | do stuff with lead/lag |
+--------+------------+------------+------------+------------------------+------------------------+
One way to simulate lead/lag prior to sql server 2012 is with outer apply()
select
client
, date
, nextdate
, prevdate
, column1 = 'do stuff with lead/lag'
, column2 = 'do stuff with lead/lag'
from t
outer apply (
select top 1 nextdate = i.date
from t i
where i.client = t.client
and i.date > t.date
order by i.date asc
) n
outer apply (
select top 1 prevdate = i.date
from t i
where i.client = t.client
and i.date < t.date
order by i.date desc
) p
rextester demo: http://rextester.com/GGS1299
returns:
+--------+------------+------------+------------+---------------------------------+---------------------------------+
| client | date | nextdate | prevdate | column1 | column2 |
+--------+------------+------------+------------+---------------------------------+---------------------------------+
| 1 | 2017-01-01 | 2017-01-02 | NULL | do stuff with nextdate/prevdate | do stuff with nextdate/prevdate |
| 1 | 2017-01-02 | 2017-01-03 | 2017-01-01 | do stuff with nextdate/prevdate | do stuff with nextdate/prevdate |
| 1 | 2017-01-03 | NULL | 2017-01-02 | do stuff with nextdate/prevdate | do stuff with nextdate/prevdate |
| 2 | 2017-01-02 | 2017-01-04 | NULL | do stuff with nextdate/prevdate | do stuff with nextdate/prevdate |
| 2 | 2017-01-04 | 2017-01-06 | 2017-01-02 | do stuff with nextdate/prevdate | do stuff with nextdate/prevdate |
| 2 | 2017-01-06 | NULL | 2017-01-04 | do stuff with nextdate/prevdate | do stuff with nextdate/prevdate |
+--------+------------+------------+------------+---------------------------------+---------------------------------+
For solutions that absolutely require recursion, then you probably need to use a recursive cte.
;with cte as (
-- non recursive cte to add `nextdate` for recursive join
select
t.client
, t.date
, nextdate = x.date
from t
outer apply (
select top 1 i.date
from t i
where i.client = t.client
and i.date > t.date
order by i.date asc
) x
)
, r_cte as (
--anchor rows / starting rows
select
client
, date
, nextdate
, prevDate = convert(date, null)
, column1 = convert(varchar(64),null)
, column2 = convert(varchar(64),null)
from cte t
where not exists (
select 1
from cte as i
where i.client = t.client
and i.date < t.date
)
union all
--recursion starts here
select
c.client
, c.date
, c.nextdate
, prevDate = p.date
, column1 = convert(varchar(64),'do recursive stuff with p.column1')
, column2 = convert(varchar(64),'do recursive stuff with p.column2')
from cte c
inner join r_cte p
on c.client = p.client
and c.date = p.nextdate
)
select *
from r_cte
rextester demo: http://rextester.com/LKH38243
returns:
+--------+------------+------------+------------+-----------------------------------+-----------------------------------+
| client | date | nextdate | prevdate | column1 | column2 |
+--------+------------+------------+------------+-----------------------------------+-----------------------------------+
| 1 | 2017-01-01 | 2017-01-02 | NULL | NULL | NULL |
| 2 | 2017-01-02 | 2017-01-04 | NULL | NULL | NULL |
| 2 | 2017-01-04 | 2017-01-06 | 2017-01-02 | do recursive stuff with p.column1 | do recursive stuff with p.column2 |
| 2 | 2017-01-06 | NULL | 2017-01-04 | do recursive stuff with p.column1 | do recursive stuff with p.column2 |
| 1 | 2017-01-02 | 2017-01-03 | 2017-01-01 | do recursive stuff with p.column1 | do recursive stuff with p.column2 |
| 1 | 2017-01-03 | NULL | 2017-01-02 | do recursive stuff with p.column1 | do recursive stuff with p.column2 |
+--------+------------+------------+------------+-----------------------------------+-----------------------------------+
Reference
Recursive Queries Using Common Table Expressions (cte)

Answer 2

If using SQL2012 or later, look at the features LAG & LEAD
For example, if you want to use the previous row's value in conjunction with this row's value - LAG like this:
DECLARE #T TABLE (DateCol DATETIME, StringCol VARCHAR(10))
INSERT INTO #T (DateCol, StringCol) VALUES ('2017-01-01','A'), ('2017-01-02','B'), ('2017-01-03','C'), ('2017-01-04','D'), ('2017-01-05','E')
SELECT DateCol, StringCol, PreviousRowStringcol = LAG(StringCol,1,NULL) OVER (ORDER BY DateCol) FROM #T