views:

113

answers:

7

I'm using the same SQL pattern over and over, and I know there has to be a better way, but I'm having trouble piecing it together. Here's a simple version of the pattern, where I'm pulling back the student's information and the last book they checked out, if one exists:

SELECT TStudents.*,
       BookName = (SELECT TOP 1 BookName 
                     FROM TBookCheckouts 
                    WHERE StudentID = TStudents.ID 
                 ORDER BY DateCheckedOut DESC),
       BookAuthor = (SELECT TOP 1 BookAuthor 
                       FROM TBookCheckouts 
                      WHERE StudentID = TStudents.ID 
                   ORDER BY DateCheckedOut DESC),
       BookCheckout = (SELECT TOP 1 DateCheckedOut 
                         FROM TBookCheckouts 
                         WHERE StudentID = TStudents.ID 
                     ORDER BY DateCheckedOut DESC)
   FROM TStudents

(For the sake of this example, please ignore the fact that TBookCheckouts should probably be split into TCheckouts and TBooks)

What I'm trying to illustrate: I tend to have a lot of subqueries for columns from the same table. I also tend to need to sort those subqueried tables by a date to get the most recent record, so it's not quite as simple (at least to me) as doing a LEFT JOIN. Notice, though, that except for which field is being returned, I'm essentially doing the same subquery 3 times. SQL Server may be smart enough to optimize that, but I'm thinking not (I definitely need to get better at reading execution plans...).

While there might be advantages to structuring it this way (sometimes this ends up being more readable, if I have tons of subqueries and sub-tables), it doesn't seem like this is particularly efficient.

I've looked into doing a LEFT JOIN from a derived table, possibly incorporating a ROW_NUMBER() and PARTITION BY, but I just can't seem to piece it all together.

A: 

Try

    ;WITH LatestCheckouts
    AS
    (
        SELECT  DISTINCT
                A.StudentID
            ,   A.BookName   
            ,   A.BookAuthor
            ,   A.DateCheckedOut
        FROM    TBookCheckouts A
            INNER JOIN
        (   
            SELECT  StudentID
            ,   DateCheckedOut =  MAX(DateCheckedOut)
             FROM TBookCheckouts
            GROUP  BY
                StudentID
        ) B

        ON A.StudentID = B.StudentID
        AND A.DateCheckedOut =  B.DateCheckedOut
    )       
    SELECT students.*
        ,  BookName     = checkouts.BookName
        ,  BookAuthor   = checkouts.BookAuthor
        ,  BookCheckout = checkouts.DateCheckedOut

    FROM    TStudents students
        LEFT JOIN
         LatestCheckouts checkouts
    ON  students.ID = checkouts.StudentID
Noel Abrahams
Close, but this won't give me the TOP 1 part. I need one record per student.
Jon Smock
@Jon, amended to get latest checkout. Did that work?
Noel Abrahams
@Noel - I get `Incorrect syntax near ')'.` when running it.
LittleBobbyTables
@LittleBobbyTables, fixed the compilation errors. I believet this should now work.
Noel Abrahams
The query returns the student multiple times if they have multiple books checked out in the same day.
LittleBobbyTables
Yes, it needed a DISTINCT which I've added. But some of the other solutions are more compact, so really just for the record.
Noel Abrahams
A: 

If you wanted to get into using a Common Table Expression, you could the following query. It doesn't gain you anything, in this case, but for future:

;with LatestBookOut as 
(
    SELECT  C.StudentID, BookID, Title, Author, DateCheckedOut AS BookCheckout 
    FROM    CheckedOut AS C
    INNER JOIN ( SELECT StudentID, 
                        MAX(DateCheckedOut) AS DD 
                FROM Checkedout 
                GROUP BY StudentID) StuMAX                 
    ON StuMAX.StudentID = C.StudentID 
    AND StuMAX.DD = C.DateCheckedOut  
)

SELECT    B.BookCheckout,
        BookId, 
        Title,    
        Author, 
        S.*

FROM    LatestBookOut AS B
INNER JOIN Student  AS S ON S.ID = B.StudentID 
p.campbell
Won't this only return one record total? I'm looking for 1 record per student.
Jon Smock
@Jon: here's an update. I've kept the same names from my test database, and not used the same names, as it appears you're obfuscated/mocked in your question. Hope this helps!
p.campbell
A: 
create table BookCheckout(StudentID int, CheckoutDate date, BookName varchar(10))

insert into BookCheckout values (1, '1.1.2010', 'a');
insert into BookCheckout values (1, '2.1.2010', 'b');
insert into BookCheckout values (1, '3.1.2010', 'c');
insert into BookCheckout values (2, '1.1.2010', 'd');
insert into BookCheckout values (2, '2.1.2010', 'e');

select *
from BookCheckout bc1
where CheckoutDate = (
    Select MAX(CheckoutDate) 
    from BookCheckout bc2
    where bc2.StudentID= bc1.StudentID)

StudentID    CheckoutDate    BookName
2    2010-01-02    e
1    2010-01-03    c    

Just add the join to TStudent and you are done. There is 1 problem left: You get multiple BookCheckouts per student if there are 2 or more Bookcheckouts for a Student with the same, max checkout date.

  select s.*, LastBookCheckout.*
  from TStudent s, 
    (select *
    from BookCheckout bc1
    where CheckoutDate = (
        Select MAX(CheckoutDate) 
        from BookCheckout bc2
        where bc2.StudentID= bc1.StudentID)) LastBookCheckout
  where s.ID = LastBookCheckout.StudentID

To avoid duplicates:

select * 
from (
  select *, RANK() over (partition by StudentID order by CheckoutDate desc,BookName) rnk
    from BookCheckout bc1) x
where rnk=1

I used "BookName" as second ordering criterion. => Use primary key instead to make it a real unique criterion.

nang
That really doesn't solve his problem as you state `You get multiple BookCheckouts per student if there are 2 or more Bookcheckouts for a Student with the same, max checkout date.`
LittleBobbyTables
@LittleBobbyTables: Added another solution without duplicates
nang
But it's missing students that have never checked out a book
LittleBobbyTables
I considered adding a left join with TStudent obvious.
nang
+3  A: 

Use:

   SELECT s.*,
          x.bookname,
          x.bookauthor,
          x.datecheckedout
     FROM TSTUDENTS s
LEFT JOIN (SELECT bc.studentid,
                  bc.bookname,
                  bc.bookauthor,
                  bc.datecheckedout,
                  ROW_NUMBER() OVER(PARTITION BY bc.studentid
                                        ORDER BY bc.datecheckedout DESC) AS rank
             FROM TSBOOKCHECKOUTS bc) x ON x.studentid = s.id
                                       AND x.rank = 1

If the student has not checkout any books, the bookname, bookauthor, and datecheckedout will be NULL.

OMG Ponies
+1 Great solution, thanks a ton!
Jon Smock
+3  A: 

If you are using SQL Server 2005 and later, you can use a ranking function like so:

With LastCheckout As
    (
    Select StudentId, BookName, BookAuthor, DateCheckedOut 
        , Row_Number() Over ( Partition By StudentId Order By DateCheckedOut Desc) As CheckoutRank
    From TBookCheckouts
    )
Select ..., LastCheckout.BookName, LastCheckout.BookAuthor, LastCheckout.DateCheckedOut
From TStudents
    Left Join LastCheckout 
        On LastCheckout.StudentId = TStudents.StudentId
                And LastCheckout.CheckoutRank = 1
Thomas
I like the CTE; I may use one, depending on the query. I wish I could give out two accepted answers.
Jon Smock
Actually I'm switching the accepted answer to you, since I think the CTE solution works much more nicely. Thanks!
Jon Smock
@Jon Smock: There's no performance improvement to using a CTE, but it is cleaner to read.
OMG Ponies
@OMG Ponies That's basically why I switched. I actually started with your solution, but once I got it working, the first thing I did was rewrite it with the CTE to clean it up. Like I said, I really wish I could give two accepted answers, especially since you both submitted with minutes of each other. I did give you a plus 1 :-P
Jon Smock
@Jon Smock, @OMG Ponies - Ty for the upvote. As OMG Ponies said, there is no perf difference. It's just a bit more readable.
Thomas
+2  A: 

On 2005 and higher, OUTER APPLY is your friend:

SELECT TStudents.*,
       t.BookName ,
       t.BookAuthor ,
       t.BookCheckout
   FROM TStudents
  OUTER APPLY(SELECT TOP 1 s.* 
                     FROM TBookCheckouts AS s
                    WHERE s.StudentID = TStudents.ID 
                 ORDER BY s.DateCheckedOut DESC) AS t
AlexKuznetsov
A: 

The answer of OMGPonies is a good one. I would write it with Common Table Expressions for readability:

WITH CheckoutsPerStudentRankedByDate AS (
    SELECT bookname, bookauthor, datecheckedout, studentid,
        ROW_NUMBER(PARTITION BY studentid ORDER BY datecheckedout DESC) AS rank
    FROM TSBOOKCHECKOUTS
)
SELECT 
    s.*, c.bookname, c.bookauthor, c.datecheckedout
FROM TSTUDENTS AS s
LEFT JOIN CheckoutsPerStudentRankedByDate AS c
    ON s.studentid = c.studentid
    AND c.rank = 1

The c.rank = 1 can be replaced by c.rank IN(1, 2) for last 2 checkouts, BETWEEN 1 AND 3 for last 3, etc...

littlegreen