Question

It seems like this title has been used many times before, unfortunately I don’t know how else to describe my problem. So, first off, if you have a suggestion for a better title that will help future searchers, fire away!

Anyway, my problem is trying to write the SQL to return a resultset based on a set of rules applied to the following schema:

TABLE: Tests
COLUMNS: ID (PK), Name

TABLE: TestVersions    
COLUMNS: TestID (PK), Version (PK), IsActive

TABLE: TestSessions
COLUMNS: TestID (PK), TestVersion (PK), UserID (PK), Iteration (PK), Completed, CompletionDate

There is a relationship between [Tests] and [TestVersions] on [Tests].[ID] = [TestVersions].[TestID]. There is also a relationship between [TestSessions] and [TestVersions] on [TestSessions].[TestID] = [TestVersions].[TestID] AND [TestSessions].[TestVersion] = [TestVersions].[Version]

The result set should return [Tests].[ID], [Tests].[Name] and [TestVersions].[Version] based on the following rules:

1. Any Test that has an associated record in TestSessions where Complete is false.
    
2. The maximum Version for any Test that has no associated records in TestSessions but IsActive is true.
    
3. This one is the complicated one. If a Test has associated records in TestSessions where all are completed (Complete is true) and has at least one associated record in TestVersions with IsActive true, I need to verify that the most recent CompletionDate is at least 30 days ago and, if so, return the highest Version from TestVersions.
    

Hopefully this makes sense.

Answer 1

Here’s a quick script that creates temp tables, fills them with data, and shows you how i’d approach the problem you’re having. Let me say up front that this kind of stuff is sometimes best left to the business logic layers. Also, I’m not sure that your description is very accurate because there seem to be some inconsistencies. With that said, I hope you can take away how to approach your problem, break it down into parts and solve it. Don’t worry about optimizing up front, just write clean and logical code. Optimizer will tell you if you’re making any mistakes. This assumes SQL Server 2008:

declare @Tests table
(
    ID int,
    Name varchar(100)
)

declare @TestVersions table
(
    TestID int,
    Version int,
    IsActive bit
)

declare @TestSessions table
(
    TestID int,
    TestVersion int,
    UserID varchar(100),
    Iteration int,
    Completed bit,
    CompletionDate date
)

insert into @Tests
select 1, 'one'     union all
select 2, 'two'     union all
select 3, 'three'   union all
select 4, 'four'    union all
select 5, 'five'

insert into @TestVersions
select 1, 1, 0      union all
select 2, 1, 1      union all
select 3, 1, 0      union all
select 4, 1, 0      union all
select 5, 1, 1      union all
select 1, 2, 0      union all
select 2, 2, 0      union all
select 3, 2, 1      union all
select 4, 2, 0      union all
select 1, 3, 1      union all
select 2, 3, 1

insert into @TestSessions
select 1, 1, 'a', 1, 1, GETDATE()-101       union all
select 2, 1, 'b', 1, 1, GETDATE()-11        union all
select 3, 1, 'c', 1, 0, null                union all
select 4, 1, 'd', 1, 1, GETDATE()-103       union all
select 5, 1, 'e', 1, 1, GETDATE()-101       union all
select 1, 2, 'f', 1, 1, GETDATE()-15        union all
select 2, 2, 'g', 1, 0, null                union all
select 3, 2, 'h', 1, 1, GETDATE()-17        union all
select 4, 2, 'i', 1, 0, null                union all
select 1, 3, 'j', 2, 1, GETDATE()-109       union all
select 2, 3, 'k', 2, 1, GETDATE()-101       union all
select 2, 1, 'l', 3, 1, GETDATE()-120       union all
select 3, 1, 'm', 1, 1, GETDATE()-11        union all
select 4, 1, 'n', 1, 1, GETDATE()-140       union all
select 5, 1, 'a', 1, 0, null                union all
select 1, 2, 'b', 1, 1, GETDATE()-160       union all
select 2, 2, 'c', 2, 0, null                union all
select 3, 2, 'd', 1, 1, GETDATE()-17        union all
select 4, 2, 'e', 1, 0, null                union all
select 1, 3, 'f', 2, 1, GETDATE()-4         union all
select 2, 3, 'g', 3, 1, GETDATE()-101


 select Id
        ,Name
        ,Version
   from @Tests t
            inner join
        @TestVersions v on t.ID = v.TestID

  -- any test with a session with completed = false
  where exists
        (select *
           from @TestSessions s1
          where s1.Completed = 0
            and s1.TestID = t.ID
            and s1.TestVersion = v.Version
        )
     or (
  -- any max version of a test which has no sessions
            not exists
            (select *
               from @TestSessions s1
              where s1.TestID = t.ID
                and s1.TestVersion = v.Version 
            )
            and v.Version =
            (select MAX(Version)
               from @TestVersions v2
              where v2.TestID = t.ID
            )
        )
     or (
     -- no uncompleted sessions
            not exists
            (select *
               from @TestSessions s1
              where s1.Completed = 0
                and s1.TestID = t.ID
                and s1.TestVersion = v.Version
            )
     -- we're already inner joining to versions, so we just need to check IsActive
            and IsActive = 1
     -- the most recent completion date is at least 30 days ago (not quite 30 days ago but i'm lazy)
            and GETDATE() - 30 <=
            (select MAX(CompletionDate)
               from @TestSessions s2
              where s2.TestID = t.ID
                and s2.TestVersion = v.Version
            )
     -- not sure what you mean by "return", but I'm sure you can figure out how to project what you need
        )