Related
I have a table called tbl_A with the following schema:
After insert, I have the following data in tbl_A:
Now the question is how to write a query for the following scenario:
Put (1) in front of any employee who was present three days consecutively
Put (0) in front of employee who was not present three days consecutively
The output screen shoot:
I think we should use case statement, but I am not able to check three consecutive days from date. I hope I am helped in this
Thank you
select name, case when max(cons_days) >= 3 then 1 else 0 end as presence
from (
select name, count(*) as cons_days
from tbl_A, (values (0),(1),(2)) as a(dd)
group by name, adate + dd
)x
group by name
With a self-join on name and available = 'Y', we create an inner table with different combinations of dates for a given name and take a count of those entries in which the dates of the two instances of the table are less than 2 units apart i.e. for each value of a date adate, it will check for entries with its own value adate as well as adate + 1 and adate + 2. If all 3 entries are present, the count will be 3 and you will have a flag with value 1 for such names(this is done in the outer query). Try the below query:
SELECT Z.NAME,
CASE WHEN Z.CONSEQ_AVAIL >= 3 THEN 1 ELSE 0 END AS YOUR_FLAG
FROM
(
SELECT A.NAME,
SUM(CASE WHEN B.ADATE >= A.ADATE AND B.ADATE <= A.ADATE + 2 THEN 1 ELSE 0 END) AS CONSEQ_AVAIL
FROM
TABL_A A INNER JOIN TABL_A B
ON A.NAME = B.NAME AND A.AVAILABLE = 'Y' AND B.AVAILABLE = 'Y'
GROUP BY A.NAME
) Z;
Due to the complexity of the problem, I have not been able to test it out. If something is really wrong, please let me know and I will be happy to take down my answer.
--Below is My Approch
select Name,
Case WHen Max_Count>=3 Then 1 else 0 end as Presence
from
(
Select Name,MAx(Coun) as Max_Count
from
(
select Name, (count(*) over (partition by Name,Ref_Date)) as Coun from
(
select Name,adate + row_number() over (partition by Name order by Adate desc) as Ref_Date
from temp
where available='Y'
)
) group by Name
);
select name as employee , case when sum(diff) > =3 then 1 else 0 end as presence
from
(select id, name, Available,Adate, lead(Adate,1) over(order by name) as lead,
case when datediff(day, Adate,lead(Adate,1) over(order by name)) = 1 then 1 else 0 end as diff
from table_A
where Available = 'Y') A
group by name;
I need to get the first and last record for a user if one of the key fields is different over time using a Hive table:
This is some sample data:
UserID EntryDate Activity
a3324 1/1/16 walk
a3324 1/2/16 walk
a3324 1/3/16 walk
a3324 1/4/16 run
a5613 1/1/16 walk
a5613 1/2/16 walk
a5613 1/3/16 walk
a5613 1/4/16 walk
And I'm looking for output preferably like this:
a3324 1/1/16 walk 1/4/16 run
Or at least like this:
a3324 walk run
I start writing code like this:
SELECT UserID, MINIMUM(EntryDate), MAXIMUM(EntryDate), Activity
FROM
SELECT UserID, DISTINCT Activity
GROUP BY UserID
HAVING Count(Activity) > 1
But I know that's not it.
I'd also like to be able to specify the cases where the original activity was Walk and the second activity was Run perhaps in the Where clause.
Can you help with an approach?
Thanks
You can use lag /lead to get a solution
SELECT * FROM (
select UserID ,EntryDate , Activityslec,
lead(Activityslec, 1) over (UserID ,EntryDate ) as nextActivityslec
from table) as A
where Activityslec <> nextActivityslec
SELECT
t.UserId
,MIN(CASE WHEN t.RowNumAsc = 1 THEN t.EntryDate END) as MinEntryDate
,MIN(CASE WHEN t.RowNumAsc = 1 THEN t.Activity END) as MinActivity
,MAX(CASE WHEN t.RowNumDesc = 1 THEN t.EntryDate END) as MaxEntryDate
,MAX(CASE WHEN t.RowNumDesc = 1 THEN t.Activity END) as MaxActivity
FROM
(
SELECT
UserId
,EntryDate
,Activity
,ROW_NUMBER() OVER (PARTITION BY UserId ORDER BY EntryDate) as RowNumAsc
,ROW_NUMBER() OVER (PARTITION BY UserId ORDER BY EntryDate DESC) as RowNumDesc
FROM
Table
) t
WHERE
t.RowNumAsc = 1
OR t.RowNumDesc = 1
GROUP BY
t.UserId
Looks like window functions are supported (https://cwiki.apache.org/confluence/display/Hive/LanguageManual+WindowingAndAnalytics) so using 2 row numbers 1 for EntryDate Ascending and another for Descending with Conditional Aggregation should get you to the answer.
And if you don't want to use Analytic Functions (window functions) you can use self left joins and conditional aggregation:
SELECT
t.UserId
,MIN(CASE WHEN mn.UserId IS NULL THEN t.EntryDate END) as MinEntryDate
,MIN(CASE WHEN mn.UserId IS NULL THEN t.Activity END) as MinActivity
,MAX(CASE WHEN mx.UserId IS NULL THEN t.EntryDate END) as MaxEntryDate
,MAX(CASE WHEN mx.UserId IS NULL THEN t.Activity END) as MaxActivity
FROM
Table t
LEFT JOIN Table mn
ON t.UserId = mn.UserId
AND t.EntryDate > mn.EntryDate
LEFT JOIN Table mx
ON t.UserId = mx.UserId
AND t.EntryDate < mx.EntryDate
WHERE
mn.UserId IS NULL
OR mx.UserId IS NULL
GROUP BY
t.UserId
Or a correlated Sub Query way:
SELECT
UserId
,MIN(EntryDate) as MinEntryDate
,(SELECT
Activity
FROM
Activity a
WHERE
u.UserId = a.UserId
AND a.EntryDate = MIN(u.EntryDate)
LIMIT 1
) as MinActivity
,MAX(EntryDate) as MaxEntryDate
,(SELECT
Activity
FROM
Activity a
WHERE
u.UserId = a.UserId
AND a.EntryDate = MAX(u.EntryDate)
LIMIT 1
) as MaxActivity
FROM
Activity u
GROUP BY
UserId
I have a response column that stores 2 different values for a same product based on question 1 and question 2. That creates 2 rows for each product but I want only one row for each product.
Example:
select Product, XNumber from MyTable where QuestionID IN ('Q1','Q2')
result shows:
Product XNumber
Bat abc
Bat abc12
I want it to display like below:
Product Xnumber1 Xnumber2
Bat abc abc12
Please help.
Thanks.
If you always have two different values you can try this:
SELECT a.Product, a.XNumber as XNumber1, b.XNumber as XNumber2
FROM MyTable a
INNER JOIN MyTable b
ON a.Product = b.Product
WHERE a.QuestionId = 'Q1'
AND b.QuestionId = 'Q2'
I assume that XNumber1 is the result for Q1 and Xnumber2 is the result for Q2.
This will work best if you don't have answers for both Q1 and Q2 for all ids
SELECT a.Product, b.XNumber as XNumber1, c.XNumber as XNumber2
FROM (SELECT DISTINCT Product FROM MyTable) a
LEFT JOIN MyTable b ON a.Product = b.Product AND b.QuestionID = 'Q1'
LEFT JOIN MyTable c ON a.Product = c.Product AND c.QuestionID = 'Q2'
This is one way to achieve your expected results. However, it relies on knowing that only xNumber abc and abc12 are the values. If this is not the case, then a dynamic pivot would be likely needed.
SELECT product, max(case when XNumber = 'abc' then xNumber end) as XNumber1,
max(Case when xNumber = 'abc12' then xNumber end) as xNumber2
FROM MyTable
GROUP BY Product
The problem is that SQL needs to know how many columns will be in the result at the time it compiles the SQL. Since the number of columns could be dependent on the data itself (2 rows vs 5 rows) it can't complete the request. Using Dynamic SQL you can find out the number of rows, then pass those values in as the column names which is why the dynamic SQL works.
This will get you two columns, the first will be the product, and the 2nd will be a comma delimited list of xNumbers.
SELECT DISTINCT T.Product,
xNumbers = Stuff((SELECT DISTINCT ', ' + T1.XNumber
FROM MyTable T1
WHERE t.Product = T1.Product
FOR XML PATH ('')),1,1,'')
FROM MyTable T
To get what you want, we need to know how many columns there will be, what to name them, and how to determine which value goes into which column
Been using rank() a lot in current code we have been working on at my day job. So this fun variant came to mind for your solution.
Using rank to get the 1st, 2nd, and 3rd possible item identifier then grouping them to create a simulated pivot
DECLARE #T TABLE (PRODUCT VARCHAR(50), XNumber VARCHAR(50))
INSERT INTO #T VALUES
('Bat','0-12345-98765-6'),
('Bat','0-12345-98767-2'),
('Bat','0-12345-98768-1'),
('Ball','0-12345-98771-6'),
('Ball','0-12345-98772-7'),
('Ball','0-12345-98777-9'),
('Hat','0-12345-98711-6'),
('Hat','0-12345-98712-3'),
('Tee','0-12345-98465-1')
SELECT
PRODUCT,
MAX(CASE WHEN I = 1 THEN XNumber ELSE '' END) AS Xnumber1,
MAX(CASE WHEN I = 2 THEN XNumber ELSE '' END) AS Xnumber2,
MAX(CASE WHEN I = 3 THEN XNumber ELSE '' END) AS Xnumber3
FROM
(
SELECT
PRODUCT,
XNumber,
RANK() OVER(PARTITION BY PRODUCT ORDER BY XNumber) AS I
FROM #T
) AS DATA
GROUP BY
PRODUCT
Not sure where to start... But basically I have a report table, an account table, and an account history table. The account history table will have zero or more records, where each record is the state of the account cancelled flag after it changed.
There is other stuff going on, but basically i am looking to return the account detail data, with the state of account cancelled bit on the start date and enddate as different columns.
What is the best way to do this?
I have the following working query below
(Idea) Should I do seperate joins on history table, 1 for each date?
I guess I could do it in three separate queries ( Get Begin Snapshot, End Snapshot, Normal Report query with a join to each snapshot)
something else?
Expected output:
AccountID, OtherData, StartDateCancelled, EndDateCancelled
Test Tables:
DECLARE #Report TABLE (ReportID INT, StartDate DATETIME, EndDate DATETIME)
DECLARE #ReportAccountDetail TABLE( ReportID INT, Accountid INT, Cancelled BIT )
DECLARE #AccountHistory TABLE( AccountID INT, ModifiedDate DATETIME, Cancelled BIT )
INSERT INTO #Report
SELECT 1,'1/1/2011', '2/1/2011'
--
INSERT INTO #ReportAccountDetail
SELECT 1 AS ReportID, 1 AS AccountID, 0 AS Cancelled
UNION
SELECT 1,2,0
UNION
SELECT 1,3,1
UNION
SELECT 1,4,1
--
INSERT INTO #AccountHistory
SELECT 2 AS CustomerID, '1/2/2010' AS ModifiedDate, 1 AS Cancelled
UNION--
SELECT 3, '2/1/2011', 1
UNION--
SELECT 4, '1/1/2010', 1
UNION
SELECT 4, '2/1/2010', 0
UNION
SELECT 4, '2/1/2011', 1
Current Query:
SELECT Accountid, OtherData,
MAX(CASE WHEN BeginRank = 1 THEN CASE WHEN BeginHistoryExists = 1 THEN HistoryCancelled ELSE DefaultCancel END ELSE NULL END ) AS StartDateCancelled,
MAX(CASE WHEN EndRank = 1 THEN CASE WHEN EndHistoryExists = 1 THEN HistoryCancelled ELSE DefaultCancel END ELSE NULL END ) AS EndDateCancelled
FROM
(
SELECT c.Accountid,
'OtherData' AS OtherData,
--lots of other data
ROW_NUMBER() OVER (PARTITION BY c.AccountID ORDER BY
CASE WHEN ch.ModifiedDate <= Report.StartDate THEN 1 ELSE 0 END DESC, ch.ModifiedDate desc) AS BeginRank,
CASE WHEN ch.ModifiedDate <= Report.StartDate THEN 1 ELSE 0 END AS BeginHistoryExists,
ROW_NUMBER() OVER ( PARTITION BY c.AccountID ORDER BY
CASE WHEN ch.ModifiedDate <= Report.EndDate THEN 1 ELSE 0 END DESC, ch.ModifiedDate desc) AS EndRank,
CASE WHEN ch.ModifiedDate <= Report.EndDate THEN 1 ELSE 0 END AS EndHistoryExists,
CAST( ch.Cancelled AS INT) AS HistoryCancelled,
0 AS DefaultCancel
FROM
#Report AS Report
INNER JOIN #ReportAccountDetail AS C ON Report.ReportID = C.ReportID
--Others joins related for data to return
LEFT JOIN #AccountHistory AS CH ON CH.AccountID = C.AccountID
WHERE Report.ReportID = 1
) AS x
GROUP BY AccountID, OtherData
Welcome input on writing stack overflow questions. Thanks!
ROW_NUMBER() often suprises me and out-performs my expectations. In this case, however, I'd be tempted to just use correlated sub-queries. At least, I'd test them against the alternatives.
Note: I would also use real tables, with real indexes, and a realistic volume of fake data. (If it's worth posting this question, I'm assuming that it's worth testing this realistically.)
SELECT
[Report].ReportID,
[Account].AccountID,
[Account].OtherData,
ISNULL((SELECT TOP 1 Cancelled FROM AccountHistory WHERE AccountID = [Account].AccountID AND ModifiedDate <= [Report].StartDate ORDER BY ModifiedDate DESC), 0) AS StartDateCancelled,
ISNULL((SELECT TOP 1 Cancelled FROM AccountHistory WHERE AccountID = [Account].AccountID AND ModifiedDate <= [Report].EndDate ORDER BY ModifiedDate DESC), 0) AS EndDateCancelled
FROM
Report AS [Report]
LEFT JOIN
ReportAccountDetail AS [Account]
ON [Account].ReportID = [Report].ReportID
ORDER BY
[Report].ReportID,
[Account].AccountID
Note: For whatever reason, I've found that TOP 1 and ORDER BY is faster than MAX().
In terms of your suggested answer, I'd modify it slightly to just use ISNULL instead of trying to make the Exists columns work.
I'd also join on the "other data" after all of the working out, rather than inside the inner-most query, so as to avoid having to group by all the "other data".
WITH
HistoricData AS
(
SELECT
Report.ReportID,
c.Accountid,
c.OtherData,
ROW_NUMBER() OVER (PARTITION BY c.ReportID, c.AccountID ORDER BY CASE WHEN ch.ModifiedDate <= Report.StartDate THEN 1 ELSE 0 END DESC, ch.ModifiedDate DESC) AS BeginRank,
ROW_NUMBER() OVER (PARTITION BY c.ReportID, c.AccountID ORDER BY ch.ModifiedDate DESC) AS EndRank,
CH.Cancelled
FROM
#Report AS Report
INNER JOIN
#ReportAccountDetail AS C
ON Report.ReportID = C.ReportID
LEFT JOIN
#AccountHistory AS CH
ON CH.AccountID = C.AccountID
AND CH.ModifiedDate <= Report.EndDate
)
,
FlattenedData AS
(
SELECT
ReportID,
Accountid,
OtherData,
ISNULL(MAX(CASE WHEN BeginRank = 1 THEN Cancelled END), 0) AS StartDateCancelled,
ISNULL(MAX(CASE WHEN EndRank = 1 THEN Cancelled END), 0) AS EndDateCancelled
FROM
[HistoricData]
GROUP BY
ReportID,
AccountID,
OtherData
)
SELECT
*
FROM
[FlattenedData]
LEFT JOIN
[OtherData]
ON Whatever = YouLike
WHERE
[FlattenedData].ReportID = 1
And a final possible version...
WITH
ReportStartHistory AS
(
SELECT
*
FROM
(
SELECT
[Report].ReportID,
ROW_NUMBER() OVER (PARTITION BY [Report].ReportID, [History].AccountID ORDER BY [History].ModifiedDate) AS SequenceID,
[History].*
FROM
Report AS [Report]
INNER JOIN
AccountHistory AS [History]
ON [History].ModifiedDate <= [Report].StartDate
)
AS [data]
WHERE
SequenceID = 1
)
,
ReportEndHistory AS
(
SELECT
*
FROM
(
SELECT
[Report].ReportID,
ROW_NUMBER() OVER (PARTITION BY [Report].ReportID, [History].AccountID ORDER BY [History].ModifiedDate) AS SequenceID,
[History].*
FROM
Report AS [Report]
INNER JOIN
AccountHistory AS [History]
ON [History].ModifiedDate <= [Report].EndDate
)
AS [data]
WHERE
SequenceID = 1
)
SELECT
[Report].ReportID,
[Account].*,
ISNULL([ReportStartHistory].Cancelled, 0) AS StartDateCancelled,
ISNULL([ReportEndHistory].Cancelled, 0) AS EndDateCancelled
FROM
Report AS [Report]
INNER JOIN
Account AS [Account]
LEFT JOIN
[ReportStartHistory]
ON [ReportStartHistory].ReportID = [Report].ReportID
AND [ReportStartHistory].AccountID = [Account].AccountID
LEFT JOIN
[ReportEndHistory]
ON [ReportEndHistory].ReportID = [Report].ReportID
AND [ReportEndHistory].AccountID = [Account].AccountID
I need to pull multiple columns from a subquery which also requires a WHERE filter referencing columns of the FROM table. I have a couple of questions about this:
Is there another solution to this problem besides mine below?
Is another solution even necessary or is this solution efficient enough?
Example:
In the following example I'm writing a view to present test scores, particularly to discover failures that may need to be addressed or retaken.
I cannot simply use JOIN because I need to filter my actual subquery first (notice I'm getting TOP 1 for the "examinee", sorted either by score or date descending)
My goal is to avoid writing (and executing) essentially the same subquery repeatedly.
SELECT ExamineeID, LastName, FirstName, Email,
(SELECT COUNT(examineeTestID)
FROM exam.ExamineeTest tests
WHERE E.ExamineeID = ExamineeID AND TestRevisionID = 3 AND TestID = 2) Attempts,
(SELECT TOP 1 ExamineeTestID
FROM exam.ExamineeTest T
WHERE E.ExamineeID = ExamineeID AND TestRevisionID = 3 AND TestID = 2
ORDER BY Score DESC) bestExamineeTestID,
(SELECT TOP 1 Score
FROM exam.ExamineeTest T
WHERE E.ExamineeID = ExamineeID AND TestRevisionID = 3 AND TestID = 2
ORDER BY Score DESC) bestScore,
(SELECT TOP 1 DateDue
FROM exam.ExamineeTest T
WHERE E.ExamineeID = ExamineeID AND TestRevisionID = 3 AND TestID = 2
ORDER BY Score DESC) bestDateDue,
(SELECT TOP 1 TimeCommitted
FROM exam.ExamineeTest T
WHERE E.ExamineeID = ExamineeID AND TestRevisionID = 3 AND TestID = 2
ORDER BY Score DESC) bestTimeCommitted,
(SELECT TOP 1 ExamineeTestID
FROM exam.ExamineeTest T
WHERE E.ExamineeID = ExamineeID AND TestRevisionID = 3 AND TestID = 2
ORDER BY DateDue DESC) currentExamineeTestID,
(SELECT TOP 1 Score
FROM exam.ExamineeTest T
WHERE E.ExamineeID = ExamineeID AND TestRevisionID = 3 AND TestID = 2
ORDER BY DateDue DESC) currentScore,
(SELECT TOP 1 DateDue
FROM exam.ExamineeTest T
WHERE E.ExamineeID = ExamineeID AND TestRevisionID = 3 AND TestID = 2
ORDER BY DateDue DESC) currentDateDue,
(SELECT TOP 1 TimeCommitted
FROM exam.ExamineeTest T
WHERE E.ExamineeID = ExamineeID AND TestRevisionID = 3 AND TestID = 2
ORDER BY DateDue DESC) currentTimeCommitted
FROM exam.Examinee E
To answer your second question first, yes, a better way is in order, because the query you're using is hard to understand, hard to maintain, and even if the performance is acceptable now, it's a shame to query the same table multiple times when you don't need to plus the performance may not always be acceptable if your application ever grows to an appreciable size.
To answer your first question, I have a few methods for you. These assume SQL 2005 or up unless where noted.
Note that you don't need BestExamineeID and CurrentExamineeID because they will always be the same as ExamineeID unless no tests were taken and they're NULL, which you can tell from the other columns being NULL.
You can think of OUTER/CROSS APPLY as an operator that lets you move correlated subqueries from the WHERE clause into the JOIN clause. They can have an outer reference to a previously-named table, and can return more than one column. This enables you to do the job only once per logical query rather than once for each column.
SELECT
ExamineeID,
LastName,
FirstName,
Email,
B.Attempts,
BestScore = B.Score,
BestDateDue = B.DateDue,
BestTimeCommitted = B.TimeCommitted,
CurrentScore = C.Score,
CurrentDateDue = C.DateDue,
CurrentTimeCommitted = C.TimeCommitted
FROM
exam.Examinee E
OUTER APPLY ( -- change to CROSS APPLY if you only want examinees who've tested
SELECT TOP 1
Score, DateDue, TimeCommitted,
Attempts = Count(*) OVER ()
FROM exam.ExamineeTest T
WHERE
E.ExamineeID = T.ExamineeID
AND T.TestRevisionID = 3
AND T.TestID = 2
ORDER BY Score DESC
) B
OUTER APPLY ( -- change to CROSS APPLY if you only want examinees who've tested
SELECT TOP 1
Score, DateDue, TimeCommitted
FROM exam.ExamineeTest T
WHERE
E.ExamineeID = T.ExamineeID
AND T.TestRevisionID = 3
AND T.TestID = 2
ORDER BY DateDue DESC
) C
You should experiment to see if my Count(*) OVER () is better than having an additional OUTER APPLY that just gets the count. If you're not restricting the Examinee from the exam.Examinee table, it may be better to just do a normal aggregate in a derived table.
Here's another method that (sort of) goes and gets all the data in one swoop. It conceivably could perform better than other queries, except my experience is that windowing functions can get very and surprisingly expensive in some situations, so testing is in order.
WITH Data AS (
SELECT
*,
Count(*) OVER (PARTITION BY ExamineeID) Cnt,
Row_Number() OVER (PARTITION BY ExamineeID ORDER BY Score DESC) ScoreOrder,
Row_Number() OVER (PARTITION BY ExamineeID ORDER BY DateDue DESC) DueOrder
FROM
exam.ExamineeTest
), Vals AS (
SELECT
ExamineeID,
Max(Cnt) Attempts,
Max(CASE WHEN ScoreOrder = 1 THEN Score ELSE NULL END) BestScore,
Max(CASE WHEN ScoreOrder = 1 THEN DateDue ELSE NULL END) BestDateDue,
Max(CASE WHEN ScoreOrder = 1 THEN TimeCommitted ELSE NULL END) BestTimeCommitted,
Max(CASE WHEN DueOrder = 1 THEN Score ELSE NULL END) BestScore,
Max(CASE WHEN DueOrder = 1 THEN DateDue ELSE NULL END) BestDateDue,
Max(CASE WHEN DueOrder = 1 THEN TimeCommitted ELSE NULL END) BestTimeCommitted
FROM Data
GROUP BY
ExamineeID
)
SELECT
E.ExamineeID,
E.LastName,
E.FirstName,
E.Email,
V.Attempts,
V.BestScore, V.BestDateDue, V.BestTimeCommitted,
V.CurrentScore, V.CurrentDateDue, V.CurrentTimeCommitted
FROM
exam.Examinee E
LEFT JOIN Vals V ON E.ExamineeID = V.ExamineeID
-- change join to INNER if you only want examinees who've tested
Finally, here's a SQL 2000 method:
SELECT
E.ExamineeID,
E.LastName,
E.FirstName,
E.Email,
Y.Attempts,
Y.BestScore, Y.BestDateDue, Y.BestTimeCommitted,
Y.CurrentScore, Y.CurrentDateDue, Y.CurrentTimeCommitted
FROM
exam.Examinee E
LEFT JOIN ( -- change to inner if you only want examinees who've tested
SELECT
X.ExamineeID,
X.Cnt Attempts,
Max(CASE Y.Which WHEN 1 THEN T.Score ELSE NULL END) BestScore,
Max(CASE Y.Which WHEN 1 THEN T.DateDue ELSE NULL END) BestDateDue,
Max(CASE Y.Which WHEN 1 THEN T.TimeCommitted ELSE NULL END) BestTimeCommitted,
Max(CASE Y.Which WHEN 2 THEN T.Score ELSE NULL END) CurrentScore,
Max(CASE Y.Which WHEN 2 THEN T.DateDue ELSE NULL END) CurrentDateDue,
Max(CASE Y.Which WHEN 2 THEN T.TimeCommitted ELSE NULL END) CurrentTimeCommitted
FROM
(
SELECT ExamineeID, Max(Score) MaxScore, Max(DueDate) MaxDueDate, Count(*) Cnt
FROM exam.ExamineeTest
WHERE
TestRevisionID = 3
AND TestID = 2
GROUP BY ExamineeID
) X
CROSS JOIN (SELECT 1 UNION ALL SELECT 2) Y (Which)
INNER JOIN exam.ExamineeTest T
ON X.ExamineeID = T.ExamineeID
AND (
(Y.Which = 1 AND X.MaxScore = T.MaxScore)
OR (Y.Which = 2 AND X.MaxDueDate = T.MaxDueDate)
)
WHERE
T.TestRevisionID = 3
AND T.TestID = 2
GROUP BY
X.ExamineeID,
X.Cnt
) Y ON E.ExamineeID = Y.ExamineeID
This query will return unexpected extra rows if the combination of (ExamineeID, Score) or (ExamineeID, DueDate) can return multiple rows. That's probably not unlikely with Score. If neither is unique, then you need to use (or add) some additional column that can grant uniqueness so it can used to select one row. If only Score can be duplicated then an additional pre-query that gets the max Score first, then dovetailing in with the max DueDate would combine to pull the most recent score that was a tie for the highest at the same time as getting the most recent data. Let me know if you need more SQL 2000 help.
Note: The biggest thing that is going to control whether CROSS APPLY or a ROW_NUMBER() solution is better is whether you have an index on the columns that are being looked up and whether the data is dense or sparse.
Index + you're pulling only a few examinees with lots of tests each = CROSS APPLY wins.
Index + you're pulling a huge number of examines with only a few tests each = ROW_NUMBER() wins.
No index = string concatenation/value packing method wins (not shown here).
The group by solution that I gave for SQL 2000 will probably perform the worst, but not guaranteed. Like I said, testing is in order.
If any of my queries do give performance problems let me know and I'll see what I can do to help. I'm sure I probably have typos as I didn't work up any DDL to recreate your tables, but I did my best without trying it.
If performance really does become crucial, I would create ExamineeTestBest and ExamineeTestCurrent tables that get pushed to by a trigger on the ExamineeTest table that would always keep them updated. However, this is denormalization and probably not necessary or a good idea unless you've scaled so awfully big that retrieving results becomes unacceptably long.
It's not same subquery. It's three different subqueries.
count() on all
TOP (1) ORDER BY Score DESC
TOP (1) ORDER BY DateDue DESC
You can't avoid executing it less than 3 times.
The question is, how to make it execute no more than 3 times.
One option would be to write 3 inline table functions and use them with outer apply. Make sure they are actually inline, otherwise your performance will drop a hundred times. One of these three functions might be:
create function dbo.topexaminee_byscore(#ExamineeID int)
returns table
as
return (
SELECT top (1)
ExamineeTestID as bestExamineeTestID,
Score as bestScore,
DateDue as bestDateDue,
TimeCommitted as bestTimeCommitted
FROM exam.ExamineeTest
WHERE (ExamineeID = #ExamineeID) AND (TestRevisionID = 3) AND (TestID = 2)
ORDER BY Score DESC
)
Another option would be to do essentially the same, but with subqueries. Because you fetch data for all students anyway, there shouldn't be too much of a difference performance-wise. Create three subqueries, for example:
select bestExamineeTestID, bestScore, bestDateDue, bestTimeCommitted
from (
SELECT
ExamineeTestID as bestExamineeTestID,
Score as bestScore,
DateDue as bestDateDue,
TimeCommitted as bestTimeCommitted,
row_number() over (partition by ExamineeID order by Score DESC) as takeme
FROM exam.ExamineeTest
WHERE (TestRevisionID = 3) AND (TestID = 2)
) as foo
where foo.takeme = 1
Same for ORDER BY DateDue DESC and for all records, with respective columns being selected.
Join these three on the examineeid.
What is going to be better/more performant/more readable is up to you. Do some testing.
It looks like you can replace the three columns that are based on the alias "bestTest" with a view. All three of those subqueries have the same WHERE clause and the same ORDER BY clause.
Ditto for the subquery aliased "bestNewTest". Ditto ditto for the subquery aliased "currentTeest".
If I counted right, that would replace 8 subqueries with 3 views. You can join on the views. I think the joins would be faster, but if I were you, I'd check the execution plan of both versions.
You could use a CTE and OUTER APPLY.
;WITH testScores AS
(
SELECT ExamineeID, ExamineeTestID, Score, DateDue, TimeCommitted
FROM exam.ExamineeTest
WHERE TestRevisionID = 3 AND TestID = 2
)
SELECT ExamineeID, LastName, FirstName, Email, total.Attempts,
bestTest.*, currentTest.*
FROM exam.Examinee
LEFT OUTER JOIN
(
SELECT ExamineeID, COUNT(ExamineeTestID) AS Attempts
FROM testScores
GROUP BY ExamineeID
) AS total ON exam.Examinee.ExamineeID = total.ExamineeID
OUTER APPLY
(
SELECT TOP 1 ExamineeTestID, Score, DateDue, TimeCommitted
FROM testScores
WHERE exam.Examinee.ExamineeID = t.ExamineeID
ORDER BY Score DESC
) AS bestTest (bestExamineeTestID, bestScore, bestDateDue, bestTimeCommitted)
OUTER APPLY
(
SELECT TOP 1 ExamineeTestID, Score, DateDue, TimeCommitted
FROM testScores
WHERE exam.Examinee.ExamineeID = t.ExamineeID
ORDER BY DateDue DESC
) AS currentTest (currentExamineeTestID, currentScore, currentDateDue,
currentTimeCommitted)