CODEWITHSUNDEEP
sqlsql-server
ajudi
ajudi

Reputation: 53

sql select last but one record from every group

Please help me with building up following query: I have a joined table from a selection query:

select nevlista.nevID as ColA, nkhk.nevktghID as ColB, nkhk.ktghelyID as ColC 
from nevlista
inner join nkhk on nevlista.nevID = nkhk.nevID
where nevlista.nevID = nkhk.nevID

This gives following result:

 ColA   ColB ColC
 90002  629 6
 90003  835 9
 90003  875 12
 90003  112 12
 90004  424 17
 90004  570 1
 90004  905 17
 90005  648 1
 90005  649 17
 90005  523 17
 and so on...

Now, I need from every group of same ID-s in ColA, the last but one highest value from ColB (and the value from ColC). (The number of same ID-s in ColA are different, usually there are 2 to 5 records with the same ID).

Upvotes: 1

Views: 375

Answers (3)

Zohar Peled
Zohar Peled

Reputation: 82534

The simplest thing to do is a cte with Row_number partitioned by colA ordered by colB desc:

;With cte as
(
    select  nevlista.nevID as ColA, 
            nkhk.nevktghID as ColB, 
            nkhk.ktghelyID as ColC,
            ROW_NUMBER() OVER(PARTITION BY nevlista.nevID ORDER BY nkhk.nevktghID DESC) as rn
    from nevlista
    inner join nkhk on nevlista.nevID = nkhk.nevID
    where nevlista.nevID = nkhk.nevID
)

select ColA, ColB, ColC
FROM CTE
WHERE rn = 2 -- assiming every nevlista.nevID appears at least twice in the cte

Upvotes: 2

Radim Bača
Radim Bača

Reputation: 10711

Consider using group by with max:

select nevlista.nevID as ColA, nkhk.nevktghID as ColB, nkhk.ktghelyID as ColC 
from nevlista
join nkhk on nevlista.nevID = nkhk.nevID
join
(
    select nevID, max(nevktghID) max_nevktghID
    from nkhk 
    group by nevID
) t1 on nkhk.nevID = t1.nevID and
        nkhk.nevktghID = t1.max_nevktghID

It can be considerably faster on SQL Server mainly if you have the following index:

CREATE NONCLUSTERED INDEX ix_nkhk _nevID_nevktghID
  ON nkhk (nevID,nevktghID) INCLUDE (ktghelyID)

Upvotes: 0

Yogesh Sharma
Yogesh Sharma

Reputation: 50173

You could use ranking function row_number() with ties method

select top(1) with ties nl.nevID as cola, n.nevktghID as colb, n.ktghelyID as colc
from nevlista nl 
inner join nkhk n on nl.nevID = n.nevID
order by row_number() over (partition by nl.nevID order by n.nevktghID  desc)

And, i suspect there would be no need to use where clause after joins

Upvotes: 2

Related Questions