Reduce records in a slowly changing dimension table

  • I have a table that tracks groups over time using a start and end date columns. Some of the groups were loaded with a daily snapshot method where new records were inserted with equal start and end dates rather than the end date being incremented to show continuing membership in a group.

    I have some sample data below. 

    I would like to delete all the records showing member 2 belong to group B and insert 1 record having a start date of 2000-01-01 and an end date of 2000-01-04. The 4 records showing member 4 belong to group D for different periods should be replaced with 2 records: 2017-01-01 - 2017-01-02 and 2017-01-04 - 2017-01-05

    Thanks if you can help. I am working in SQL Server 2008 (SP3)


    IF OBJECT_ID('dbo.GroupMembers', 'U') IS NOT NULL DROP TABLE dbo.GroupMembers;

    CREATE TABLE dbo.GroupMembers
    (
        GroupCode VARCHAR(1) NOT NULL,
        GroupMember INT NOT NULL,
        StartDate DATE NOT NULL,
        EndDate DATE NULL
    )

    --Some of my groups have proper ranges for start and end dates
    INSERT INTO dbo.GroupMembers VALUES ('A', 1, '20000101', '20170101')

    --Some of my groups have were populated with a snapshot for every day
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000101', '20000101')
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000102', '20000102')
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000103', '20000103')
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000104', '20000104')

    --Other groups are open ended and that is fine
    INSERT INTO dbo.GroupMembers VALUES ('C', 3, '20000104', NULL)

    --Some members may be removed and then re-added
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170101', '20170101')
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170102', '20170102')
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170104', '20170104')
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170105', '20170105')

    SELECT * FROM dbo.GroupMembers

  • So, are you looking for something like this?

    IF OBJECT_ID('dbo.GroupMembers', 'U') IS NOT NULL DROP TABLE dbo.GroupMembers;

    CREATE TABLE dbo.GroupMembers
    (
        GroupCode VARCHAR(1) NOT NULL,
        GroupMember INT NOT NULL,
        StartDate DATE NOT NULL,
        EndDate DATE NULL
    );


    --Some of my groups have proper ranges for start and end dates
    INSERT INTO dbo.GroupMembers VALUES ('A', 1, '20000101', '20170101');

    --Some of my groups have were populated with a snapshot for every day
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000101', '20000101');
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000102', '20000102');
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000103', '20000103');
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000104', '20000104');

    --Other groups are open ended and that is fine
    INSERT INTO dbo.GroupMembers VALUES ('C', 3, '20000104', NULL);

    --Some members may be removed and then re-added
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170101', '20170101');
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170102', '20170102');
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170104', '20170104');
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170105', '20170105');


    SELECT * FROM dbo.GroupMembers;

    WITH base AS (
    SELECT
      [GroupCode]
      , [GroupMember]
      , [StartDate]
      , [EndDate]
      , rn = ROW_NUMBER() OVER (PARTITION BY [GroupCode], [GroupMember] ORDER BY [StartDate])
    FROM
      dbo.GroupMembers
    ), Interim AS (
    SELECT
      .[GroupCode]
      , .[GroupMember]
      , .[StartDate]
      , .[EndDate]
      , .[rn]
      , [GroupDate] = DATEADD(DAY,-.[rn],.[StartDate])
    FROM
      base b
    )
    SELECT
      .[GroupCode]
      , .[GroupMember]
      , [StartDate] = MIN(.[StartDate])
      , [EndDate] = MAX(.[EndDate])
    FROM
      [Interim] i
    GROUP BY
      .[GroupCode]
      , .[GroupMember]
      , .[GroupDate];

  • Yes that is exactly what I am looking for. This should help me eliminate 99% of the records. Thanks so much.

  • I neglected to provide sample data that would illustrate a particular scenario and naturally the code Lynn provided does not address it. The scenario is some groups may have both existing ranges and the snapshot records (records where the start and end date is the same). When there are 2 date range records with no gaps between I would like to consolidate these multiple records

    --New sample data that illustrates what I neglected to show earlier; Some members may have existing ranges and snapshot records
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171201', '20171215');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171216', '20171216');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171217', '20171217');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171218', '20171218');

    In the spirit of showing you the code I have worked on I have what may be a 'solution' but it is rather inelegant. I used a calendar table to un-consolidate the data before consolidating it with the code Lynn provided. 
    Below will have the new sample data; Lynn's code and then I jam the unconsolidation into Lynn's code as a subquery.
    Any suggestions on how to do this properly? It seems silly to -un-consolidate the data.
    IF OBJECT_ID('dbo.GroupMembers', 'U') IS NOT NULL DROP TABLE dbo.GroupMembers;
    CREATE TABLE dbo.GroupMembers
    (
      GroupCode VARCHAR(1) NOT NULL,
      GroupMember INT NOT NULL,
      StartDate DATE NOT NULL,
      EndDate DATE NULL
    );

    --Some of my groups have proper ranges for start and end dates
    INSERT INTO dbo.GroupMembers VALUES ('A', 1, '20170101', '20170201');

    --Some of my groups have were populated with a snapshot for every day
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000101', '20000101');
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000102', '20000102');
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000103', '20000103');
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000104', '20000104');

    --Other groups are open ended and that is fine
    INSERT INTO dbo.GroupMembers VALUES ('C', 3, '20000104', NULL);

    --Some members may be removed and then re-added
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170101', '20170101');
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170102', '20170102');
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170104', '20170104');
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170105', '20170105');

    --Some members may have existing ranges and snapshot records
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171201', '20171215');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171216', '20171216');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171217', '20171217');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171218', '20171218');

    --SELECT * FROM dbo.GroupMembers;

    --This wil have multiple E records
    WITH
        base AS
        (
        SELECT
            GroupCode,
            GroupMember,
            StartDate,
            EndDate,
            rn = ROW_NUMBER() OVER (PARTITION BY GroupCode, GroupMember ORDER BY StartDate)
        FROM dbo.GroupMembers
        )
        ,
        Interim AS
        (
        SELECT
             b.GroupCode,
             b.GroupMember,
             b.StartDate,
             b.EndDate,
             b.rn,
             GroupDate = DATEADD(DAY,-b.rn,b.StartDate)
        FROM base b
    )

    SELECT
    i.GroupCode,
    i.GroupMember,
    StartDate = MIN(i.StartDate),
    EndDate = MAX(i.EndDate)
    FROM Interim i
    GROUP BY
    i.GroupCode,
    i.GroupMember,
    i.GroupDate
    ORDER BY
    i.GroupCode; ;

    --If I explode the date ranges into the separate records in a subquery then the E records will be consolidated
    WITH
        base AS
        (
        SELECT
            GroupCode,
            GroupMember,
            StartDate,
            EndDate,
            rn = ROW_NUMBER() OVER (PARTITION BY GroupCode, GroupMember ORDER BY StartDate)
        FROM
        (
            SELECT
                #T1.GroupCode,
                #T1.GroupMember,
                C.CalendarDate AS StartDate,
                C.CalendarDate AS EndDate
            FROM #T1
            INNER JOIN dbo.Calendar C
             ON C.CalendarDate BETWEEN #T1.StartDate AND #T1.EndDate

            UNION ALL

             SELECT
                GroupCode,
                GroupMember,
                StartDate,
                EndDate
            FROM #T1
            WHERE EndDate IS NULL
        )Unconsolidation
        )
        ,
        Interim AS
        (
        SELECT
             b.GroupCode,
             b.GroupMember,
             b.StartDate,
             b.EndDate,
             b.rn,
             GroupDate = DATEADD(DAY,-b.rn,b.StartDate)
        FROM base b
    )

    SELECT
    i.GroupCode,
    i.GroupMember,
    StartDate = MIN(i.StartDate),
    EndDate = MAX(i.EndDate)
    FROM Interim i
    GROUP BY
    i.GroupCode,
    i.GroupMember,
    i.GroupDate
    ORDER BY
    i.GroupCode;

  • Chrissy321 - Tuesday, December 26, 2017 5:00 PM

    I neglected to provide sample data that would illustrate a particular scenario and naturally the code Lynn provided does not address it. The scenario is some groups may have both existing ranges and the snapshot records (records where the start and end date is the same). When there are 2 date range records with no gaps between I would like to consolidate these multiple records

    --New sample data that illustrates what I neglected to show earlier; Some members may have existing ranges and snapshot records
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171201', '20171215');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171216', '20171216');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171217', '20171217');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171218', '20171218');

    In the spirit of showing you the code I have worked on I have what may be a 'solution' but it is rather inelegant. I used a calendar table to un-consolidate the data before consolidating it with the code Lynn provided. 
    Below will have the new sample data; Lynn's code and then I jam the unconsolidation into Lynn's code as a subquery.
    Any suggestions on how to do this properly? It seems silly to -un-consolidate the data.
    IF OBJECT_ID('dbo.GroupMembers', 'U') IS NOT NULL DROP TABLE dbo.GroupMembers;
    CREATE TABLE dbo.GroupMembers
    (
      GroupCode VARCHAR(1) NOT NULL,
      GroupMember INT NOT NULL,
      StartDate DATE NOT NULL,
      EndDate DATE NULL
    );

    --Some of my groups have proper ranges for start and end dates
    INSERT INTO dbo.GroupMembers VALUES ('A', 1, '20170101', '20170201');

    --Some of my groups have were populated with a snapshot for every day
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000101', '20000101');
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000102', '20000102');
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000103', '20000103');
    INSERT INTO dbo.GroupMembers VALUES ('B', 2, '20000104', '20000104');

    --Other groups are open ended and that is fine
    INSERT INTO dbo.GroupMembers VALUES ('C', 3, '20000104', NULL);

    --Some members may be removed and then re-added
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170101', '20170101');
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170102', '20170102');
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170104', '20170104');
    INSERT INTO dbo.GroupMembers VALUES ('D', 4, '20170105', '20170105');

    --Some members may have existing ranges and snapshot records
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171201', '20171215');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171216', '20171216');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171217', '20171217');
    INSERT INTO dbo.GroupMembers VALUES ('E', 5, '20171218', '20171218');

    --SELECT * FROM dbo.GroupMembers;

    --This wil have multiple E records
    WITH
        base AS
        (
        SELECT
            GroupCode,
            GroupMember,
            StartDate,
            EndDate,
            rn = ROW_NUMBER() OVER (PARTITION BY GroupCode, GroupMember ORDER BY StartDate)
        FROM dbo.GroupMembers
        )
        ,
        Interim AS
        (
        SELECT
             b.GroupCode,
             b.GroupMember,
             b.StartDate,
             b.EndDate,
             b.rn,
             GroupDate = DATEADD(DAY,-b.rn,b.StartDate)
        FROM base b
    )

    SELECT
    i.GroupCode,
    i.GroupMember,
    StartDate = MIN(i.StartDate),
    EndDate = MAX(i.EndDate)
    FROM Interim i
    GROUP BY
    i.GroupCode,
    i.GroupMember,
    i.GroupDate
    ORDER BY
    i.GroupCode; ;

    --If I explode the date ranges into the separate records in a subquery then the E records will be consolidated
    WITH
        base AS
        (
        SELECT
            GroupCode,
            GroupMember,
            StartDate,
            EndDate,
            rn = ROW_NUMBER() OVER (PARTITION BY GroupCode, GroupMember ORDER BY StartDate)
        FROM
        (
            SELECT
                #T1.GroupCode,
                #T1.GroupMember,
                C.CalendarDate AS StartDate,
                C.CalendarDate AS EndDate
            FROM #T1
            INNER JOIN dbo.Calendar C
             ON C.CalendarDate BETWEEN #T1.StartDate AND #T1.EndDate

            UNION ALL

             SELECT
                GroupCode,
                GroupMember,
                StartDate,
                EndDate
            FROM #T1
            WHERE EndDate IS NULL
        )Unconsolidation
        )
        ,
        Interim AS
        (
        SELECT
             b.GroupCode,
             b.GroupMember,
             b.StartDate,
             b.EndDate,
             b.rn,
             GroupDate = DATEADD(DAY,-b.rn,b.StartDate)
        FROM base b
    )

    SELECT
    i.GroupCode,
    i.GroupMember,
    StartDate = MIN(i.StartDate),
    EndDate = MAX(i.EndDate)
    FROM Interim i
    GROUP BY
    i.GroupCode,
    i.GroupMember,
    i.GroupDate
    ORDER BY
    i.GroupCode;

    You're destroying the whole reason for SCDs on this table.   You should never have gaps between "records" and, if you check the "records", you'll likely find that something changed, which is why a new "record" was required.  If you do have absolutely identical rows except for the start and end date of the row and those are adjacent to other rows, then you need to fix your SCD marking and row generation process before you get into deleting all the nice history you've built up.

    --Jeff Moden


    RBAR is pronounced "ree-bar" and is a "Modenism" for Row-By-Agonizing-Row.
    First step towards the paradigm shift of writing Set Based code:
    ________Stop thinking about what you want to do to a ROW... think, instead, of what you want to do to a COLUMN.

    Change is inevitable... Change for the better is not.


    Helpful Links:
    How to post code problems
    How to Post Performance Problems
    Create a Tally Function (fnTally)

  • Happy new year all 🙂

    Giving the original OP the benefit of the doubt, it looks like either the load strategy has been changed or different sources are employing different strategies.  It is clear from the sample data that this is abstracted from the real problem.  @ Jeff, if the SCD data is looking at continuous days attendance then a gap of one day would be perfectly acceptable and would definitely generate a new SCD record - I have used this pattern myself when tracking data quality issues to identify failure rate and mean-time-to-fix. I think the OP has the right approach to fixing the issue which is to explode out the records with the contiguous days, then find islands and gaps using a calendar table and collapse the records back down.  There may be more elegant ways to solve the problem but if it is a one-off data realignment I would not be looking to spend days on code to optimise the solution.  Do what is pragmatic and move on.

  • Thanks for your response. Yes there have been different methods of populating this table. As Jeff said one of the methods is broken and needs to be fixed. I will post up a different thread once I get a chance to take a crack at writing code to properly update a SCD table. If any one had any links/sample code on this that would be appreciated.

    My original post was looking for a one-off fix and what I posted subsequently seems adequate so i will do what is pragmatic and move on. The existing broken process is not broken in the sense that the data is incorrect but rather it is always inserting records rather than updating which resulted in performance problems. My one-off fix should not destroy any data, all the information should be preserved just represented by a smaller set of records. Thanks all.

Viewing 7 posts - 1 through 6 (of 6 total)

You must be logged in to reply to this topic. Login to reply