Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

Are these two queries the same - GROUP BY vs. DISTINCT?

These two queries seem to return the same results. Is that coincidental or are they really the same?

1.

SELECT t.ItemNumber,
  (SELECT TOP 1 ItemDescription
   FROM Transactions
   WHERE ItemNumber = t.ItemNumber
   ORDER BY DateCreated DESC) AS ItemDescription
FROM Transactions t
GROUP BY t.ItemNumber

2.

SELECT DISTINCT(t.ItemNumber),
  (SELECT TOP 1 ItemDescription
   FROM Transactions
   WHERE ItemNumber = t.ItemNumber
   ORDER BY DateCreated DESC) AS ItemDescription
FROM Transactions t

A bit of explanation: I'm trying to get a distinct list of items from a table full of transactions. For each item, I'm looking for the ItemNumber (the identifying field) and the most recent ItemDescription.

like image 826
MCS Avatar asked Jul 28 '10 15:07

MCS


5 Answers

Your example #2 had me scratching me head for a while - I thought to myself: "You can't DISTINCT a single column, what would that mean?" - until I realised what is going on.

When you have

SELECT DISTINCT(t.ItemNumber)

you are not, despite appearances, actually asking for distinct values of t.ItemNumber! Your example #2 actually gets parsed the same as

SELECT DISTINCT
  (t.ItemNumber)
  ,
  (SELECT TOP 1 ItemDescription
   FROM Transactions
   WHERE ItemNumber = t.ItemNumber
   ORDER BY DateCreated DESC) AS ItemDescription
FROM Transactions t

with syntactically-correct but superfluous parentheses around t.ItemNumber. It is to the result-set as a whole that DISTINCT applies.

In this case, since your GROUP BY groups by the column that actually varies, you get the same results. I'm actually slightly surprised that SQL Server doesn't (in the GROUP BY example) insist that the subqueried column is mentioned in the GROUP BY list.

like image 97
AakashM Avatar answered Oct 18 '22 07:10

AakashM


Same results but the second one seems to have a more expensive sort step to apply the DISTINCT on my quick test.

Both were beaten out of sight by ROW_NUMBER though...

with T as
(
SELECT ItemNumber, 
       ItemDescription,
       ROW_NUMBER() OVER ( PARTITION BY ItemNumber ORDER BY DateCreated DESC) AS RN
FROM Transactions
)
SELECT * FROM T
WHERE RN=1

edit ...which in turn was thumped by Joe's solution on my test setup.

Plans

Test Setup

CREATE TABLE Transactions
(
ItemNumber INT not null,
ItemDescription VARCHAR(50) not null,
DateCreated DATETIME not null
)

INSERT INTO Transactions
SELECT 
number, NEWID(),DATEADD(day, cast(rand(CAST(newid() as varbinary))*10000 
  as int),getdate()) 
FROM master.dbo.spt_values

ALTER TABLE dbo.Transactions ADD CONSTRAINT
    PK_Transactions PRIMARY KEY CLUSTERED 
    (ItemNumber,DateCreated) 
like image 32
Martin Smith Avatar answered Oct 18 '22 07:10

Martin Smith


Based on the data & simple queries, both will return the same results. However, the fundamental operations are very different.

DISTINCT, as AakashM beat me to pointing out, is applied to all column values, including those from subselects and computed columns. All DISTINCT does is remove duplicates, based on all columns involved, from visibility. This is why it's generally considered a hack, because people will use it to get rid of duplicates without understanding why the query is returning them in the first place (because they should be using IN or EXISTS rather than a join, typically). PostgreSQL is the only database I know of with a DISTINCT ON clause, which does work as the OP probably intended.

A GROUP BY clause is different - it's primary use is for grouping for accurate aggregate function use. To server that function, column values will be unique values based on what's defined in the GROUP BY clause. This query would never need DISTINCT, because the values of interest are already unique.

Conclusion

This is a poor example, because it portrays DISTINCT and GROUP BY as equals when they are not.

like image 3
OMG Ponies Avatar answered Oct 18 '22 07:10

OMG Ponies


If you're running at least 2005 and can use a CTE, this is a little cleaner IMHO.

EDIT: As pointed out in Martin's answer, this also performs much better.

;with cteMaxDate as (
    select t.ItemNumber, max(DateCreated) as MaxDate
        from Transactions t
        group by t.ItemNumber
)
SELECT t.ItemNumber, t.ItemDescription
    FROM cteMaxDate md
        inner join Transactions t
            on md.ItemNumber = t.ItemNumber
                and md.MaxDate = t.DateCreated
like image 3
Joe Stefanelli Avatar answered Oct 18 '22 07:10

Joe Stefanelli


Yes, they will return the same results.

like image 2
Mike M. Avatar answered Oct 18 '22 07:10

Mike M.