Winning TSQL Challenge 12

It’s a source of great pleasure for me that I am included in the winners for TSQL challenge 12.

Here’s my post describing my solution and here’s the analysis by Jacob Sebastian, the founder and president of TSQL Challenges.

TSQL Challenge 12: Completing sequence by inserting missing rows

TSQL Challenge 12 was a relatively easier one. The participants were given month-wise score values and were asked to complete the sequence by creating entries for missing month.

Here’s the sample input:

YearMonth   Score
----------- -----------
200903      100
200803      95
200802      99
200801      100
200711      100

And here’s the desired output. Notice that the score of last month is replicated in each of the missing rows:

YearMonth   Score
----------- -----------
200908      100
200907      100
200906      100
200905      100
200904      100
200903      100
200902      95
200901      95
200812      95
200811      95
200810      95
200809      95
200808      95
200807      95
200806      95
200805      95
200804      95
200803      95
200802      99
200801      100
200712      100
200711      100

The script to generate sample data is provided below:

DECLARE @Scores TABLE
(
	YearMonth	INT,
	Score		INT
)

INSERT @Scores VALUES(200903, 100)
INSERT @Scores VALUES(200803, 95)
INSERT @Scores VALUES(200802, 99)
INSERT @Scores VALUES(200801 ,100)
INSERT @Scores VALUES(200711, 100)

Solution

I once blogged about creating a sequence of numbers/dates using a recursive CTE in this post. The same technique can be used here. However, since the YearMonth column in the sample data is integer, we have two choices:

  1. Convert it to DateTime and apply T-SQL DateTime functions
  2. Leave it as Integer and apply some intelligent arithmetic

I am providing both the solutions here. Note that the first solution will be slower due to overhead of casting and applying T-SQL scalar functions.

The first solution: Converting to DateTime

;with cte as
(
    select score, Cast(Cast(YearMonth as varchar)+'01' as datetime) as dateVal
        from @scores
    union all
    select score, dateadd(month, 1, dateval)
        from cte
        where not exists
            --the resultant YearMonth value should not lie in the original table
            ( select 1 from @scores s where s.YearMonth = cast( left(convert(varchar, dateadd(month, 1, cte.dateval), 112), 6) as int) )
            --stop at current month
            and dateadd(month, 1, cte.dateval) < getdate()
)

select left(convert(varchar, dateval, 112), 6) as yearmonth, score
from cte
order by dateval desc

Explanation:
Here I am simply converting the integer YearMonth column to a datetime dateval column by appending 01 to the end (so a 200901 becomes 20090101 that can easily be cast to a dateTime) and then finding subsequent dates by adding one month in each CTE iteration.

The second solution: Integer Arithmetic

;with cte as
(
    select YearMonth, Score
		from @Scores
    union all
    select YearMonth + YearMonth % 100 / 12 * 88 + 1 as YearMonth, Score
        from cte
        where not exists
            --the resultant YearMonth value should not lie in the original table
            ( select s.YearMonth from @Scores s where s.YearMonth = (cte.YearMonth + cte.YearMonth % 100 / 12 * 88 + 1) )
            --stop at current month
            and cte.YearMonth < month(getdate()) + year(getdate())*100
)

select *
from cte
order by YearMonth desc

Explanation:
The important point is to increment the value of YearMonth correctly. So 200811 should get incremented to 200812 but 200812 should get incremented to 200901. This isn’t difficult if we introduce a case statement like this:

YearMonth + (Case When YearMonth%100 < 12 Then 1 Else 89 End)

But I wanted to do this purely using arithmetic with no Case statements, so I came up with this formula:

(YearMonth % 100 / 12 * 88) + 1

Note that the factor (YearMonth % 100 / 12 * 88) will reduce to zero for all values from January to November, i.e. from 200801 to 200811.

I hope you enjoyed the solution.

Update:
It was pointed out in one of the comments by Rakesh that the solution could reach the default limit of recursion which is 100. In order to avoid this, we need to add option (maxrecursion 12,000) in the final select statement. Then, we can have 10,000 years missing between two adjacent entries. Thanks, Rakesh.

...
select *
from cte
order by YearMonth desc
option (maxrecursion 12,000)

TSQL Challenge 11: Calculating the lowest price of an item by applying discount coupons

TSQL Challenge 11 was a practical problem. Given a a list of products and a list of discount coupons, we needed to find the minimum price for all the products based on certain rules. Here are those rules:

  • Maximum two coupons can be applied on the same product
  • The discount price can not be less than 70% of the original price
  • The total amount of the discount can not exceed 30$

Also, note that coupons are applied in a cumulative way. So the second coupon is applied on the result of the original price + first coupon.

Sample Data:
Here is some sample products data:

ID NAME    PRICE
-- ------- ---------
1  PROD 1  100,00
2  PROD 2  220,00
3  PROD 3  15,00
4  PROD 4  70,00
5  PROD 5  150,00

Here are the coupons to be applied

ID NAME         VALUE  IS_PERCENT
-- -----------  ------ ----------
1  CP 1 : -15$  15     0
2  CP 2 : -5$   5      0
3  CP 3 : -10%  10     1
4  CP 4 : -12$  12     0

And here’s the required output:

ID NAME    PRICE    DISC_PRICE  TOT_DISC  RATE    COUPON_NAMES
2.-- ------  -------- ----------- --------- ------- -------------------------
3.1  PROD 1  100.00$  73.00$      27.00$    27.00%  CP 4 : -12$ + CP 1 : -15$
4.2  PROD 2  220.00$  193.00$     27.00$    12.27%  CP 4 : -12$ + CP 1 : -15$
5.3  PROD 3  15.00$   13.50$      1.50$     10.00%  CP 3 : -10%
6.4  PROD 4  70.00$   49.50$      20.50$    29.28%  CP 1 : -15$ + CP 3 : -10%
7.5  PROD 5  150.00$  120.00$     30.00$    20.00%  CP 3 : -10% + CP 1 : -15$

Solution

Interesting enough… So lets attempt to find a solution. Read the rest of this entry »

TSQL Challenge 10: Horizontal and Vertical sorting of a result set

TSQL Challenge 10 was an interesting one. We needed to sort a result set horizontally as well as vertically. That is, given the following as an input:

input
We need to write a query that :

  1. Sort the values horizontally: Arrange the values from smallest to the largest. for example, the first row contains values “2”, “1” and “3”, it should be arranged as “1”, “2” and “3”.
  2. Sort the rows vertically: This is the regular sorting that we are familiar with.
  3. Remove duplicates: Duplicate rows should be removed from the final output.

And here’s the expected output:

output
I tackled the problem systematically in series of steps, each represented by a CTE:

  1. First, I assigned a rowID to each row of the input
  2. Then I merged the three columns into one using unpivot transformation
  3. Then I sorted that merged result grouped by rowID (horizontal sorting)
  4. Then I used pivot to split the values into three columns based on that sorting
  5. Finally I sorted the result (vertical sorting) and displayed distinct rows

Here are the CTE definitions along with the result at each step: Read the rest of this entry »

TSQL Challenge 9: Getting longest chain of consecutive alike/duplicate rows

The BeyondRelational team is working hard to present us with cool TSQL challenges. For challenge 9, the contestants were required to find the first and last IDs for consecutive rows with same values of Send and Ack states. That is, given the following as an input,

Input

We were required to produce the following output.

Output

An important part of the challenge was to write a scalable query capable of handling milllions of rows. This requirement kicked off the following simple answer to the question involving min/max subqueries:

;With Result
as
(
	Select
	(Select IsNull(Max(C.id)+1, (Select Min(id) from @tc9)) from @tc9 C
			where
				C.id <= A.id and
				(C.SendState<>A.SendState or C.AckState<>A.AckState)
			) as MinID
	,
	(Select IsNull(Min(C.id)-1, (Select Max(id) from @tc9)) from @tc9 C
			where
				C.id >= A.id and
				(C.SendState<>A.SendState or C.AckState<>A.AckState)
			) as MaxID
	,A.SendState, A.AckState
	From @tc9 A
)

Select distinct C.MinID, C.MaxID, C.SendState, C.AckState
from Result C
order by C.MinID

The above solution tries to find Min/Max IDs for every row using a subquery and hence will result in hopeless speed when executed on very large tables. We need to find another more efficient way to solve the problem.

Luckily, T-SQL 2005 presents us with ranking funcitons that are very helpful in various scenarios. Consider assigning a row number per Send and Ack states using the following query.

;With RowNoPerStateCombination
As
(
    Select
        *,
        --assign a row number grouped per SendState/AckState combination
        ROW_NUMBER() OVER(PARTITION BY SendState,AckState ORDER BY ID) AS RowID
    From @tc9
)

Select * from RowNoPerStateCombination
order by ID

The output is

RowIDs

Notice a great pattern here: if we try to subtract this generated RowID from the primary key(ID), it is going to give us a unique result for every consecutive send/ack state combination. This is depicted using the following screenshot from Excel.
Pattern

Since we are able to produce a unique result (say it GroupID) for every consecutive send/ack state combinations, we can just pick the min and max values per GroupID per Send/Ack State combination. Here’s what I am talking about:

;With MessageGroups
as
(
    Select
        *,
        --assign a unique group number for each consequtive state combination
        ID - ROW_NUMBER() OVER(PARTITION BY SendState,AckState ORDER BY ID) AS GroupID
    From @tc9
    Where CreationDate between @startTime and @endTime
)

Select MIN(ID) as FirstIdInclusive, MAX(ID) as LastIdInclusive, SendState, AckState
From MessageGroups
Group by GroupID, SendState, AckState
Order by MIN(ID)

And we get the required output.

Output

Finally, here’s my complete solution.

--populate sample data
DECLARE @tc9 TABLE(
    ID INT IDENTITY(1,1),
    CreationDate DATETIME,
    Content NVARCHAR(10),
    SendState BIT,
    AckState BIT
)

INSERT INTO @tc9 (CreationDate,Content,SendState,AckState)
SELECT GETDATE()-1.0,'Msg #1',0,0 UNION
SELECT GETDATE()-0.9,'Msg #2',0,0 UNION
SELECT GETDATE()-0.8,'Msg #3',1,1 UNION
SELECT GETDATE()-0.7,'Msg #4',1,1 UNION
SELECT GETDATE()-0.6,'Msg #5',1,1 UNION
SELECT GETDATE()-0.5,'Msg #6',1,0 UNION
SELECT GETDATE()-0.4,'Msg #7',1,0 UNION
SELECT GETDATE()-0.3,'Msg #8',1,0 UNION
SELECT GETDATE()-0.2,'Msg #9',1,0 UNION
SELECT GETDATE()-0.1,'Msg #10',1,1

--SELECT * FROM @tc9

--solution
Declare @startTime datetime
Declare @endTime datetime

set @startTime = GetDate()-20.8
set @endTime = GetDate()

;With MessageGroups
as
(
    Select
        *,
        --assign a unique group number for each consequtive state combination
        ID - ROW_NUMBER() OVER(PARTITION BY SendState,AckState ORDER BY ID) AS GroupID
    From @tc9
    Where CreationDate between @startTime and @endTime
)

Select MIN(ID) as FirstIdInclusive, MAX(ID) as LastIdInclusive, SendState, AckState
From MessageGroups
Group by GroupID, SendState, AckState
Order by MIN(ID)

TSQL Challenge 8: Using recursive CTE for a hierarchical relationship

Last week, I submitted an entry for T-SQL challenge 8. This time, the contestants were asked to process a hierarchy using recursive CTEs without applying any filters inside the CTE. The big challenge was the condition that the CTE should not contain any filter for a specific manager.
Here is some code for populating test data.

--Populate test data
DECLARE @Employees TABLE (EmpID INT, EmpName VARCHAR(20), ReportsTo INT)
INSERT INTO @Employees(EmpID, EmpName, ReportsTo)
  SELECT 1, 'Jacob', NULL UNION ALL
  SELECT 2, 'Rui', NULL UNION ALL
  SELECT 3, 'Jacobson', NULL UNION ALL
  SELECT 4, 'Jess', 1 UNION ALL
  SELECT 5, 'Steve', 1 UNION ALL
  SELECT 6, 'Bob', 1 UNION ALL
  SELECT 7, 'Smith', 2 UNION ALL
  SELECT 8, 'Bobbey', 2 UNION ALL
  SELECT 9, 'Steffi', 3 UNION ALL
  SELECT 10, 'Bracha', 3 UNION ALL
  SELECT 11, 'John', 5 UNION ALL
  SELECT 12, 'Michael', 6 UNION ALL
  SELECT 13, 'Paul', 6 UNION ALL
  SELECT 14, 'Lana', 7 UNION ALL
  SELECT 15, 'Johnson', 7 UNION ALL
  SELECT 16, 'Mic', 8 UNION ALL
  SELECT 17, 'Stev', 8 UNION ALL
  SELECT 18, 'Paulson', 9 UNION ALL
  SELECT 19, 'Jessica', 10

I started by creating a recursive CTE for the whole table like this:

;With Hierarchy(EmpName, EmpID, Level, FullyQualifiedName)
As
(
  Select E.EmpName, E.EmpID, 0, Cast('.'+E.EmpName+'.' as Varchar(MAX))
    From @Employees E
    Where E.ReportsTo is null
  Union all
  Select E.EmpName, E.EmpID, H.Level+1, H.FullyQualifiedName+'.'+E.EmpName+'.'
    from @Employees E
    inner join Hierarchy H on H.EmpID=E.ReportsTo
)

Select Space(Level*4) + H.EmpName
  from Hierarchy H
  order by H.FullyQualifiedName

Result of CTE

Notice that I am constructing a FullyQualifiedName value for each row. This value consists of full path from root to the current person. This approach helped me in filtering the records in the final select statement where I just needed to look for a '.' + ManagerName + '.' filter. Here is a select query that extracts all subordinates for a particular manager.

Select Space(Level*4) + H.EmpName
  from Hierarchy H
  where CHARINDEX('.'+(Select Top(1) E.EmpName from @Employees E Where E.EmpName=@manager)+'.', H.FullyQualifiedName) > 0
  order by H.FullyQualifiedName

The only issue that remains is that when the query is run for a non-parent employee, we get extra spaces in the beginning. E.g. when we run the query for “Paul”, we get an output like:

Spacing issue

I solved it by inserted a sub-query in the space() expression like this:

Select Space((Level-(Select Top(1) Level from Hierarchy H2 Where H2.EmpName=@manager))*4) + EmpName
  from Hierarchy H
  where CHARINDEX('.'+(Select Top(1) E.EmpName from @Employees E Where E.EmpName=@manager)+'.', H.FullyQualifiedName) > 0
  order by H.FullyQualifiedName

Instead of two subqueries, I tried to use a join but the performance using sub-queries was better so I submitted the subquery solution. Here’s the join version I am referring, it looks very nice and clean:

Select Space((H.Level-H2.Level)*4) + H.EmpName
  from Hierarchy H
  inner join Hierarchy H2 on CHARINDEX('.'+H2.EmpName+'.', H.FullyQualifiedName) > 0
  where H2.EmpName=@manager
  order by H.FullyQualifiedName

In the last, here is the complete solution:

--Populate test data
DECLARE @Employees TABLE (EmpID INT, EmpName VARCHAR(20), ReportsTo INT)
INSERT INTO @Employees(EmpID, EmpName, ReportsTo)
  SELECT 1, 'Jacob', NULL UNION ALL
  SELECT 2, 'Rui', NULL UNION ALL
  SELECT 3, 'Jacobson', NULL UNION ALL
  SELECT 4, 'Jess', 1 UNION ALL
  SELECT 5, 'Steve', 1 UNION ALL
  SELECT 6, 'Bob', 1 UNION ALL
  SELECT 7, 'Smith', 2 UNION ALL
  SELECT 8, 'Bobbey', 2 UNION ALL
  SELECT 9, 'Steffi', 3 UNION ALL
  SELECT 10, 'Bracha', 3 UNION ALL
  SELECT 11, 'John', 5 UNION ALL
  SELECT 12, 'Michael', 6 UNION ALL
  SELECT 13, 'Paul', 6 UNION ALL
  SELECT 14, 'Lana', 7 UNION ALL
  SELECT 15, 'Johnson', 7 UNION ALL
  SELECT 16, 'Mic', 8 UNION ALL
  SELECT 17, 'Stev', 8 UNION ALL
  SELECT 18, 'Paulson', 9 UNION ALL
  SELECT 19, 'Jessica', 10

--Solution starts here
DECLARE @manager VARCHAR(20)
SELECT @manager = 'Jacob'

--CTE
;With Hierarchy(EmpName, EmpID, Level, FullyQualifiedName)
As
(
  Select E.EmpName, E.EmpID, 0, Cast('.'+E.EmpName+'.' as Varchar(MAX))
    From @Employees E
    Where E.ReportsTo is null
  Union all
  Select E.EmpName, E.EmpID, H.Level+1, H.FullyQualifiedName+'.'+E.EmpName+'.'
    from @Employees E
    inner join Hierarchy H on H.EmpID=E.ReportsTo
)

--Result
Select Space((Level-(Select Top(1) Level from Hierarchy H2 Where H2.EmpName=@manager))*4) + EmpName
  from Hierarchy H
  where CHARINDEX('.'+(Select Top(1) E.EmpName from @Employees E Where E.EmpName=@manager)+'.', H.FullyQualifiedName) > 0
  order by H.FullyQualifiedName

TSQL Challenge 7: Listing the 5 biggest tables on the server

A few weeks ago, I stumbled upon this blog that presents cool T-SQL challenges. I submitted an entry for challenge 7 that asked to write the shortest script to list the 5 biggest tables on a server. Here’s my solution:

create table #temp
(
	[database] nvarchar(MAX),
	[table] nvarchar(MAX),
	[rows] int,
	[reserved_size] nvarchar(100),
	[data_size] nvarchar(100),
	[index_size] nvarchar(100),
	[unused_space] nvarchar(100)
)

declare @sql nvarchar(MAX)
set @sql=replace('if !~! not in (!master!,!model!,!msdb!,!tempdb!)
  exec [~].dbo.sp_msforeachtable
    "insert into #temp([table], [rows], [reserved_size], [data_size], [index_size], [unused_space])
      exec [~].dbo.sp_spaceused !?!"','!',char(39))

EXEC sp_MSForEachDB
	@command1=@sql,
	@command2="update #temp set [database]='~' where [database] is null",
	@replacechar='~'

select top(5) [database] as base, [table], [data_size] as size, [rows] as rows
from #temp
order by Cast(LEFT([data_size],len([data_size])-3) as int) desc

drop table #temp

So, I started by creating a temporary table with columns (database, table, rows, reserved_size, data_size, index_size, unused_space) for the output. I used the two undocumented stored procedures sp_MSforeachdb and sp_MSforeachtable to iterate through all the tables in all the databases and executed sp_spaceused as described in the following pseudo code:

foreach(database db in serverDatabases)
  if (db not in 'master', 'msdb', 'model', 'tempdb')
    foreach(table t in db.Tables)
    {
       insert into #temp (table, rows, reserved_size, data_size, index_size, unused_space)
         execute sp_spaceused for table 't'

       --at this point, our #temp table will be populated with data for each table
       --but the 'database' column will be 'null', so now replace it with the name of database
       update #temp
         set [database] = 'db' where [database] is null
    }

The most important part is that I am using an update operation for storing the database name in the temporary table. Thanks to Microsoft that we can give a set of 3 commands to the above mentioned undocumented stored procedures. Another hard part was to create a single t-sql statement that iterates for all tables inside a database and execute sp_spaceused. I did this by a complex combination of single quotes, double quotes and the replace function. In the last, I am just selecting the top(5) rows ordered by size.

To enter the contest, I reduced the script length by replacing all the variable/column names with a single length identifier. Here was my final submission:

create table #t(d nvarchar(MAX),t nvarchar(MAX),r int,x nvarchar(100),s nvarchar(100),y nvarchar(100),z nvarchar(100))
declare @s nvarchar(MAX)
set @s=replace('if !~! not in (!master!,!model!,!msdb!,!tempdb!) exec [~].dbo.sp_msforeachtable "insert into #t(t, r,x,s,y,z) exec [~].dbo.sp_spaceused !?!"','!',char(39))
EXEC sp_MSForEachDB @command1=@s, @command2="update #t set d='~' where d is null", @replacechar='~'
select top(5) d as base, t as [table], s as size, r as rows from #t order by Cast(LEFT(s,len(s)-3) as int) desc
drop table #t

Let’s wait and see the solution of other players.