How to query this output in SQL server

Question

I have a table with data like this:

CREATE TABLE Test
    (CustName varchar(20), Country varchar(20), RecordedTime datetime, CurrNo tinyint);

INSERT INTO Test
          


        

Answer 1

You have a bunch of if/and/buts going on in your original request, but maybe having an ALTERNATE VIEW of the data might be a better solution that you just did not think of or have been offered, so I will.

Why have individual lines showing both the before and after, why not do on a single returned line. If there is no change, why show? Maybe that is a later question.

To get my solution, I have a simple WITH from the test data that I am using as a baseline for my query. You have your own CurrNo which may not always start with 1, and could even possibly skip a sequence number by some accident. So my WITH declaration gets a sequential row number by customer and ordered by the CurrNo. This will ALWAYS return per customer rows 1, 2, 3, 4 even if the CurrNo values may be 5, 8, 9, 11 (exaggerated starting number and accidental skipped values). You could even change the order by to the RecordedTime to ensure logging based on the DATE per customer activity.

;with baseData as
(
   select 
         T.*,
         ROW_NUMBER() OVER (PARTITION BY CustName ORDER BY CurrNo) AS CustOrder
      from
         Test T
)

So now, with the query. I have my first table which is the basis for finding any "changes" within the system. You will always start with these records. Apply a date restrictive filter as you need. The LEFT-JOIN will always be on the customer name PLUS whatever the row number is PLUS ONE joined to the NEXT of the same baseline data. So, if just looking at customer Alex and Jerry from your data, you would have the following

Customer  Country    Time                  CurrNo   CustOrder
Alex      Australia  2018-06-01 08:00:00   1        1
Alex      China      2018-06-01 10:00:00   2        2
Alex      India      2018-06-01 10:05:00   3        3
Alex      Japan      2018-06-01 11:00:00   4        4

Jerry     Cuba       2018-06-12 00:00:00   4        1
Jerry     Brazil     2018-06-12 00:05:00   5        2
Jerry     India      2018-06-12 00:10:00   7        3
Jerry     USA        2018-06-12 00:15:00   9        4

So you can see the natural row number normalization between each. So Now, my left-join is by customer AND the CustOrder column which will not skip gaps, so I will get records like

Customer  CustOrder   NextCustOrder        CurrNo      NextCurrNo
Alex      1           2                    1           1
Alex      2           3                    2           2
Alex      3           4                    3           3
Alex      4           (no 5th record)      4           4

Jerry     1           2                    4           5
Jerry     2           3                    5           7
Jerry     3           4                    7           9
Jerry     4           (no 5th record)      9           (no next record)

Finally, I am getting the data from the first record which will always exist and IF A CORRESPONDING NEXT record exists, it will show that new changed TO value. That next record will be the next row and it's possible changed TO value and so on..

select 
        bd.CustName,
        case when bd.CurrNo = 1 then 'ADD' else 'CHANGE' end as Audit,
        bd.Country as CurrentValue,
        bd.RecordedTime,
        bdNext.Country as ChangedValue,
        bdNext.RecordedTime ChangedTime,
        bd.CurrNo,
        bd.CustOrder
    from 
        baseData bd
            LEFT JOIN baseData bdNext
                on bd.CustName = bdNext.CustName
                AND bd.CustOrder +1 = bdNext.CustOrder;


CustName  Audit    CurrentValue   RecordedTime              ChangedValue   ChangedTime               CurrNo   CustOrder
Alex      ADD      Australia      2018-06-01 08:00:00.000   China          2018-06-01 10:00:00.000   1        1
Alex      CHANGE   China          2018-06-01 10:00:00.000   India          2018-06-01 10:05:00.000   2        2
Alex      CHANGE   India          2018-06-01 10:05:00.000   Japan          2018-06-01 11:00:00.000   3        3
Alex      CHANGE   Japan          2018-06-01 11:00:00.000   NULL           NULL                      4        4

Jerry     CHANGE   Cuba           2018-06-12 00:00:00.000   Brazil         2018-06-12 00:05:00.000   4        1
Jerry     CHANGE   Brazil         2018-06-12 00:05:00.000   India          2018-06-12 00:10:00.000   5        2
Jerry     CHANGE   India          2018-06-12 00:10:00.000   USA            2018-06-12 00:15:00.000   7        3
Jerry     CHANGE   USA            2018-06-12 00:15:00.000   NULL           NULL                      9        4

If you don't want the "last value" for a record because no changes after, just change the LEFT JOIN to INNER JOIN to guarantee something changed after. But that would fail if you wanted to see all new "ADD" records without changes. You could apply that as a where clause something like

where
      bd.CurrNo = 1
   OR bdNext.CustOrder IS NOT NULL

Additionally you could add a date filter such as

where
      bd.RecordedTime >= '2018-06-10'
  AND (    bd.CurrNo = 1
        OR bdNext.CustOrder IS NOT NULL )

As suggested in comment by Ronen Ariely, the above WHERE clause with the date would be applied to the WITH baseData as component, adding a

  where
     T.RecordedTime >= '2018-06-10'

to pre-filter the data before it gets to the rest of the join activity

Answer 2

Good day,

Please check if the below solution solves all your needs. I tested it with your data and with some more rows, but it is always best to re-check. In first glance it seems like it returns the requested result. I will add some explanation later on

The query I am using it this:

DECLARE @Date DATE = '2018-06-12';
with MyCTE as (
    SELECT 
        t.CustName,t.Country,t.RecordedTime,t.CurrNo, D = CONVERT(DATE, RecordedTime)
        ,RN_D = ROW_NUMBER() 
            OVER (partition by t.CustName order by t.CurrNo desc)
        ,RN = ROW_NUMBER() 
            OVER (partition by t.CustName order by t.CurrNo)
        ,RN_Old = ROW_NUMBER() 
            OVER (partition by t.CustName, (CASE WHEN CONVERT(DATE, RecordedTime) < @Date then 0 else 1 END) order by t.CurrNo desc)
        ,Cnt = COUNT(*) 
            OVER (partition by t.CustName)
        ,CntToday = COUNT(CASE WHEN CONVERT(DATE, RecordedTime) = @Date THEN 1 ELSE NULL END) 
            OVER (partition by t.CustName)
    FROM Test t
    where 
        -- returns rows untill current date
        CONVERT (DATE, RecordedTime) <= @Date 
        -- only if relevnat to current date
        and EXISTS (
            SELECT * FROM test t0 
            where CONVERT (DATE, RecordedTime) = @Date and t0.CustName = t.CustName
        )
)
,MyCTE2 as (
    select
        CustName, Country, RecordedTime, D, CurrNo, RN_D, RN, Cnt, t2.c, History, CntToday, RN_Old
    from MyCTE t1
    left JOIN (select * from (values(1, 'NEW'),(1, 'BEFORE')) t2(c, History) ) t2 
        on t1.CurrNo = t2.c
            and CntToday > 1
            and D = @Date
    where 
        RN_D = 1 
        or (RN = 1 and D = @Date) 
        or (RN_Old = 1 and D < @Date)
)
,MyCTE3 as (
    select CustName, Country, RecordedTime
        -- unmarke the bellow comment in order to get the accessories columns I used
        -- This is recommended to understand the line-of-thinking
        --, D, c, RN_D, RN, CurrNo, Cnt, CntToday, RN_Old
        , History = CASE
            WHEN CurrNo = 1 and Cnt = 1 then 'NEW'
            WHEN RN_D = 1 then 'CURRENT'
            else ISNULL(History,'BEFORE')
        END
    from MyCTE2
)
select CustName, Country, RecordedTime--, D, c, RN_D, RN, CurrNo, Cnt, CntToday, RN_Old
    ,Audit = CASE when History='New' then 'ADD' else 'CHANGE'  END
    , History
from MyCTE3

To make it simpler to test I insert the entire query into table function

DROP FUNCTION IF EXISTS dbo.F
GO
CREATE FUNCTION dbo.F(@Date DATE)
RETURNS TABLE AS RETURN (

--DECLARE @Date DATE = '2018-06-12';
with MyCTE as (
    SELECT 
        t.CustName,t.Country,t.RecordedTime,t.CurrNo, D = CONVERT(DATE, RecordedTime)
        ,RN_D = ROW_NUMBER() 
            OVER (partition by t.CustName order by t.CurrNo desc)
        ,RN = ROW_NUMBER() 
            OVER (partition by t.CustName order by t.CurrNo)
        ,RN_Old = ROW_NUMBER() 
            OVER (partition by t.CustName, (CASE WHEN CONVERT(DATE, RecordedTime) < @Date then 0 else 1 END) order by t.CurrNo desc)
        ,Cnt = COUNT(*) 
            OVER (partition by t.CustName)
        ,CntToday = COUNT(CASE WHEN CONVERT(DATE, RecordedTime) = @Date THEN 1 ELSE NULL END) 
            OVER (partition by t.CustName)
    FROM Test t
    where 
        -- returns rows untill current date
        CONVERT (DATE, RecordedTime) <= @Date 
        -- only if relevnat to current date
        and EXISTS (
            SELECT * FROM test t0 
            where CONVERT (DATE, RecordedTime) = @Date and t0.CustName = t.CustName
        )
)
,MyCTE2 as (
    select
        CustName, Country, RecordedTime, D, CurrNo, RN_D, RN, Cnt, t2.c, History, CntToday, RN_Old
    from MyCTE t1
    left JOIN (select * from (values(1, 'NEW'),(1, 'BEFORE')) t2(c, History) ) t2 
        on t1.CurrNo = t2.c
            and CntToday > 1
            and D = @Date
    where 
        RN_D = 1 
        or (RN = 1 and D = @Date) 
        or (RN_Old = 1 and D < @Date)
)
,MyCTE3 as (
    select CustName, Country, RecordedTime
        -- unmarke the bellow comment in order to get the accessories columns I used
        -- This is recommended to understand the line-of-thinking
        --, D, c, RN_D, RN, CurrNo, Cnt, CntToday, RN_Old
        , History = CASE
            WHEN CurrNo = 1 and Cnt = 1 then 'NEW'
            WHEN RN_D = 1 then 'CURRENT'
            else ISNULL(History,'BEFORE')
        END
    from MyCTE2
)
select CustName, Country, RecordedTime--, D, c, RN_D, RN, CurrNo, Cnt, CntToday, RN_Old
    ,Audit = CASE when History='New' then 'ADD' else 'CHANGE'  END
    , History
from MyCTE3
--order by CustName, RecordedTime
)
GO

Using the function it simpler to make multiple test, but probably in the production you will want to use the direct query

-- Test
select * from F('2018-06-01') order by CustName , RecordedTime
select * from F('2018-06-02') order by CustName , RecordedTime
select * from F('2018-06-03') order by CustName , RecordedTime
select * from F('2018-06-10') order by CustName , RecordedTime
select * from F('2018-06-11') order by CustName , RecordedTime
select * from F('2018-06-12') order by CustName , RecordedTime
select * from F('2018-06-13') order by CustName , RecordedTime
select * from F('2018-06-14') order by CustName , RecordedTime

/**************** Update at 2018-08-19 14:05 Israel Time ****************/

I notice that some more information is important to add for the sake of the participate in the thread. I hope this will be useful

NOTE! tested on Microsoft SQL Server 2017 Developer Edition

First, let's compare percentage of resources uses according to the Execution Plans of three queries: (1) My solution, (2) maulik kansara seconds solution after update the first solution, and (3) maulik kansara first solution

Now lets check the image of the EP of maulik kansara seconds solution:

This query scans the table 11 times!

** Important! EP is not the only parameter that suggestes which query we should choose, but this is probably the first information we should check. In addition we should check IO statistics and Time statistics, and more...

Credit: The image was taken using sentryone tool. There is a free version which can gives most of what DBAs need. I am using the full version which I got for free as Microsoft MVP, so Thanks ;-)

Answer 3

From what I understand from the different scenario's, they can be achieved by basically only 2 rules.

• Rule 1: The records for a CustName with the RecordedTime all on the same day.
• Rule 2: The records for a CustName with the RecordedTime over more than one day.

With a few window functions and an OR in the outer query, we can have two rules to target all scenarios.

And then use CASE WHEN to calculate the "Audit" and the "History".

 SELECT
  CustName, Country, 
  FORMAT(RecordedTime, 'yyyy-MMM-dd hh:mm tt') as RecordedTime,
  (CASE CurrNo 
   WHEN 1 then 'ADD'
   ELSE 'CHANGE'
   END) as [Audit],
  (CASE  
   WHEN CurrNo = 1 then 'NEW'
   WHEN ReverseCurrRN = 1 then 'CURRENT'
   ELSE 'BEFORE'
   END) as [History]
   --, CurrNo, ReverseCurrRN, MinCurrNoPerCust, MaxCurrNoPerCustDate ,MinCurrNoPerCust, MinDatePerCust, MaxDatePerCust
 FROM
 (
    SELECT CustName, Country, RecordedTime, CurrNo
      ,ROW_NUMBER() OVER (PARTITION BY CustName ORDER BY CurrNo DESC) AS ReverseCurrRN
      ,MIN(CurrNo) OVER (PARTITION BY CustName) AS MinCurrNoPerCust
      ,MAX(CurrNo) OVER (PARTITION BY CustName, cast(RecordedTime AS DATE)) AS MaxCurrNoPerCustDate
      ,CAST(MIN(RecordedTime) OVER (PARTITION BY CustName) AS DATE) AS MinDatePerCust
      ,CAST(MAX(RecordedTime) OVER (PARTITION BY CustName) AS DATE) AS MaxDatePerCust
    FROM Test t
    WHERE CAST(RecordedTime AS DATE) between DATEADD(day,-1,@Date) and @Date
 ) q
 WHERE MaxDatePerCust = @Date
 AND (
    -- Scenario 1 & 2 & 4 & 6
       (MinDatePerCust = MaxDatePerCust AND (ReverseCurrRN = 1 OR CurrNo = MinCurrNoPerCust OR (MinCurrNoPerCust = 1 AND ReverseCurrRN = 2)))
    -- Scenario 2 & 3 & 5 & 7 & 8
    OR (MinDatePerCust != MaxDatePerCust AND (ReverseCurrRN = 1 OR CurrNo = 1 OR CurrNo = MaxCurrNoPerCustDate))
 )
 ORDER BY MinDatePerCust, CustName, CurrNo;

You can test it on db<>fiddle here

Answer 4

Based on my understanding, Below is the logic for your required output

step 1

• get max(CurrNo) for given date for each custName.
• here it is cte.

step 2

• get first entered record for the mentioned date for All CustNames available in cte if any.
• in the union of above record, get last edited record for records edited less than the current date for All CustNames available in cte if any.
• here it is cte2.

step 3

• get records from #test table where currNo=maximun or currno=1 from this step, you will get last updated records and any newly added record if any for given date.
• here it is the first query after cte and cte2.

step 4

• get first added record for given date if any with Audit as change and History as Before.
• this is for custom special case requirement
• here it is the second query.

step 5

• get the first record from cte2 to get one records for previously edited record for given date.
• here it is the third query.

-

CREATE TABLE #Test
    (CustName varchar(20), Country varchar(20), RecordedTime datetime, CurrNo tinyint);


INSERT INTO #Test
    (CustName, Country, RecordedTime, CurrNo)
VALUES
    ('Alex', 'Australia', '2018-06-01 08:00:00', 1),
    ('Alex', 'China', '2018-06-01 10:00:00', 2),
    ('Alex', 'India', '2018-06-01 10:05:00', 3),
    ('Alex', 'Japan', '2018-06-01 11:00:00', 4),
    ('John', 'Australia', '2018-06-01 08:00:00', 1),
    ('John', 'China', '2018-06-02 08:00:00', 2),
    ('Bob', 'Australia', '2018-06-02 09:00:00', 1),
    ('Bob', 'Brazil', '2018-06-03 09:50:00', 2),
    ('Bob', 'Africa', '2018-06-03 11:50:00', 3),
    ('Bob', 'India', '2018-06-03 00:55:00', 4),
    ('Tim', 'Brazil', '2018-06-10 00:00:00', 2),
    ('Tim', 'Cuba', '2018-06-11 00:00:00', 3),
    ('Tim', 'India', '2018-06-11 00:05:00', 4),
    ('Jerry', 'Cuba', '2018-06-12 00:00:00', 4),
    ('Jerry', 'Brazil', '2018-06-12 00:05:00', 5),
    ('Jerry', 'India', '2018-06-12 00:10:00', 7),
    ('Jerry', 'USA', '2018-06-12 00:15:00', 9),
    ('Maulik', 'Aus', '2018-06-12 00:00:00',3),
    ('Maulik', 'Eng', '2018-06-13 00:00:00',4),
    ('Maulik', 'USA', '2018-06-14 00:00:00',5),
    ('Maulik', 'Ind', '2018-06-14 00:00:00',6);

select * from #Test

   declare @selectedDate date='2018-06-14';
with cte as
(
select CustName,max(CurrNo) maxno,count(1) cnt
from #Test where datediff(day,RecordedTime,@selectedDate)=0 
group by CustName
),cte2 as (
select top 1 t.*,cnt
 from #Test t join cte c on c.CustName =t.CustName
where datediff(day,RecordedTime,@selectedDate)>0 and t.CurrNo!=c.maxno
order by t.RecordedTime  desc
union select top 1 t.*,cnt
 from #Test t join cte c on c.CustName =t.CustName
where datediff(day,RecordedTime,@selectedDate)=0 and t.CurrNo!=c.maxno
order by t.RecordedTime 
)
select t.CustName,t.Country,t.RecordedTime,cnt
,case when t.CurrNo=1 then 'ADD' else 'CHANGE' End as Audit
,case when t.CurrNo=1 then 'NEW' when t.CurrNo=c.maxno then 'CURRENT' else 'BEFORE' end History
 from #Test t join cte c on c.CustName =t.CustName
where datediff(day,RecordedTime,@selectedDate)=0 and (t.CurrNo=c.maxno or t.CurrNo=1)
union 
select top 1 t.CustName,t.Country,t.RecordedTime,cnt
,'CHANGE' Audit
,'BEFORE' History
 from #Test t join cte c on c.CustName =t.CustName
where datediff(day,RecordedTime,@selectedDate)=0 and t.CurrNo=1 and c.cnt>1
union 
select top 1 t.CustName,t.Country,t.RecordedTime,cnt
,'CHANGE' Audit
,'BEFORE' History
 from cte2 t order by RecordedTime

drop table #Test

There is a nice point by "Ronen Ariely". And below is the updated query with 2 times table scan instead of 8.

with cte as
(
select CustName,max(CurrNo) maxno,count(1) cnt
from #Test where datediff(day,RecordedTime,@selectedDate)=0 
group by CustName
),cte2 as (
select * from #Test where CustName in 
(
select distinct custname from cte
)
and datediff(day,RecordedTime,@selectedDate)>=0
),cte3 as (
select top 1 t.*,cnt
 from cte2 t join cte c on c.CustName =t.CustName
where datediff(day,RecordedTime,@selectedDate)>0 and t.CurrNo!=c.maxno
order by t.RecordedTime  desc
union select top 1 t.*,cnt
 from cte2 t join cte c on c.CustName =t.CustName
where datediff(day,RecordedTime,@selectedDate)=0 and t.CurrNo!=c.maxno
order by t.RecordedTime 
)
select t.CustName,t.Country,t.RecordedTime
,case when t.CurrNo=1 then 'ADD' else 'CHANGE' End as Audit
,case when t.CurrNo=1 then 'NEW' when t.CurrNo=c.maxno then 'CURRENT' else 'BEFORE' end History
 from cte2 t join cte c on c.CustName =t.CustName
where datediff(day,RecordedTime,@selectedDate)=0 and (t.CurrNo=c.maxno or t.CurrNo=1)
union 
select top 1 t.CustName,t.Country,t.RecordedTime
,'CHANGE' Audit
,'BEFORE' History
 from cte2 t join cte c on c.CustName =t.CustName
where datediff(day,RecordedTime,@selectedDate)=0 and t.CurrNo=1 and c.cnt>1
union 
select top 1 t.CustName,t.Country,t.RecordedTime
,'CHANGE' Audit
,'BEFORE' History
 from cte3 t order by RecordedTime