Intersect/Except/Join Comparison Question

  • OK, maybe I am missing something here. I was reviewing an forum discussion about ignoring duplicates,

    http://www.sqlservercentral.com/Forums/Topic856838-392-2.aspx. One of the posts appeared to be interesting, so I thought I would try some speed/IO tests comparing the various insert methods. However, I appear to be having problems with comparing the EXCEPT to the LEFT OUTER JOIN and the results I get back.

    Here is the modified version of the post from the forum:

    ---- 0: Create some tables, data in tables

    -- New Transactions

    CREATE TABLE #Transactions

    (

    TranId BIGINT PRIMARY KEY,

    TranDate DATETIME,

    TranCode VARCHAR(4),

    Amount MONEY

    );

    -- Existing Transactions

    CREATE TABLE #VendorExport

    (

    TranId BIGINT PRIMARY KEY,

    TranDate DATETIME,

    TranCode VARCHAR(4),

    Amount MONEY

    );

    -- Generate one million existing transactions (takes about three seconds)

    WITH Data (rn)

    AS (

    SELECT TOP (1000000)

    ROW_NUMBER() OVER (ORDER BY C1.object_id, C1.column_id)

    FROM master.sys.columns C1,

    master.sys.columns C2,

    master.sys.columns C3

    )

    INSERT #VendorExport(TranID, TranDate, TranCode, Amount)

    SELECT rn,

    DATEADD(HOUR, Data.rn, '19950101'),

    CASE rn % 2 WHEN 0 THEN 'room' WHEN 1 THEN 'tax' ELSE NULL END,

    RAND(CHECKSUM(NEWID())) * 25 + 1

    FROM Data;

    -- Generate new data, with some overlaps

    -- (The original generate script used TOP (10000) )

    WITH Data (rn)

    AS (

    SELECT TOP (1000)

    ROW_NUMBER() OVER (ORDER BY (SELECT NULL))

    FROM master.sys.columns C1,

    master.sys.columns C2

    )

    INSERT #Transactions (TranId, TranDate, TranCode, Amount)

    SELECT rn * 100,

    DATEADD(HOUR, Data.rn, '19950101'),

    CASE rn % 2 WHEN 0 THEN 'room' WHEN 1 THEN 'tax' ELSE NULL END,

    rn % 25

    FROM Data;

    -- Generate new data, with some overlaps

    --- (Added this, since the original script did not appear to generate "new" data?)

    WITH Data (rn)

    AS (

    SELECT TOP (1000)

    ROW_NUMBER() OVER (ORDER BY (SELECT NULL))

    FROM master.sys.columns C1,

    master.sys.columns C2

    )

    INSERT #Transactions (TranId, TranDate, TranCode, Amount)

    SELECT rn * 10002,

    DATEADD(HOUR, Data.rn, '19950101'),

    CASE rn % 2 WHEN 0 THEN 'room' WHEN 1 THEN 'tax' ELSE NULL END,

    rn % 25

    FROM Data;

    ---- 1: Select results using EXCEPT

    SELECT TranID, TranDate, TranCode, Amount

    FROM #Transactions

    EXCEPT

    SELECT TranID, TranDate, TranCode, Amount

    FROM #VendorExport;

    ---- 2: Select results using LEFT OUTER JOIN

    SELECT T.TranID, T.TranDate, T.TranCode, T.Amount

    FROM #Transactions T

    LEFT OUTER JOIN #VendorExport V

    ON T.TranID = V.TranID

    WHERE V.TranID IS NULL

    ---- 3: Select results using INNER JOIN

    SELECT T.TranID, T.TranDate, T.TranCode, T.Amount

    FROM #Transactions T

    INNER JOIN #VendorExport V

    ON T.TranID = V.TranID

    ---- 4: Select results using INTERSECT

    SELECT TranID, TranDate, TranCode, Amount

    FROM #Transactions

    INTERSECT

    SELECT TranID, TranDate, TranCode, Amount

    FROM #VendorExport;

    ---- Tidy up

    DROP TABLE

    #Transactions,

    #VendorExport;

    Step 1 (EXCEPT) generates 2000 records;

    Step 2 (LEFT OUTER JOIN) generates 901 records;

    Step 3 (INNER JOIN) generates 1099 records;

    Step 4 (INTERSECT) has 0 records.

    The questions:

    a) Why the 1099 records in Step 3 are the same as the first 1099 records in Step 1?

    b) Why does the INNER JOIN return 1099 records but the INTERSECT returns 0 records?

    Argue for your limitations, and sure enough they're yours (Richard Bach, Illusions)

  • Just to make sure I understood the Intersect/Except operators, I tried a variation of the examples from BOL. I put debug(s) around it all just to run various tests without having to re-generate the TEMP tables all the time:

    DECLARE @DEBUG INT;

    SET @DEBUG = 1;

    IF (@DEBUG = 0)

    BEGIN

    CREATE TABLE #TableA (col1 int);

    CREATE TABLE #TableB (col1 int);

    CREATE TABLE #TableC (col1 int);

    INSERT INTO #TableA

    SELECT NULL UNION

    SELECT 1 UNION

    SELECT 2 UNION

    SELECT 2 UNION

    SELECT 2 UNION

    SELECT 2 UNION

    SELECT 3 UNION

    SELECT 4 UNION

    SELECT 4;

    INSERT INTO #TableA VALUES (NULL);

    INSERT INTO #TableA VALUES (NULL);

    INSERT INTO #TableB

    SELECT NULL UNION

    SELECT 1 UNION

    SELECT 3 UNION

    SELECT 4 UNION

    SELECT 4;

    INSERT INTO #Tablec

    SELECT 2 UNION

    SELECT 2 UNION

    SELECT 2 UNION

    SELECT 4 UNION

    SELECT 4;

    SELECT * from #TableA;

    SELECT * FROM #TableB;

    SELECT * FROM #TableC;

    END;

    IF (@Debug = 1)

    BEGIN

    SELECT col1

    FROM #TableA

    EXCEPT

    SELECT col1 FROM #TableB;

    SELECT A.Col1

    FROM #TableA A

    LEFT OUTER JOIN #TableB B

    ON A.Col1 = B.Col1

    WHERE B.Col1 IS NULL;

    SELECT *

    FROM #TableA

    INTERSECT

    SELECT * FROM #TableB;

    SELECT A.Col1

    FROM #TableA A

    INNER JOIN #TableB B

    ON A.Col1 = B.Col1

    END

    IF (@DEBUG = 10)

    DROP TABLE #TableA, #TableB, #TableC;

    All of the results here are as expected. But that still leaves the questions as to why the first tests had differences..... 🙁

    Argue for your limitations, and sure enough they're yours (Richard Bach, Illusions)

  • The reason your results are not corresponding to what you would have expected them to be, is that you're not understanding the operations correctly.

    EXCEPT and INTERSECT compare the *entire* data set.

    EXCEPT returns all the rows which exist in Table 1 but do NOT exist in table 2 - for *ALL* columns.

    Take the following information as example data for what I'm going to describe below:

    SELECT * FROM #Transactions WHERE tranID = 100

    TranIdTranDateTranCodeAmount

    1001995-01-01 01:00:00.000tax1.00

    -----------------

    SELECT * FROM #VendorExport WHERE tranID = 100

    TranIdTranDateTranCodeAmount

    1001995-01-05 04:00:00.000room3.5587

    When doing the EXCEPT query, the row would be returned. This is because although they have the same TranID, they are different in their TranDate, TranCode, and Amount fields.

    When doing the LEFT OUTER JOIN query, the row would not be returned. This is because the LEFT OUTER JOIN query is looking to match on TranID alone, and since both tables share the TranID in question, the "WHERE v.TranID IS NULL" part of the query would cause it to be excluded.

    When doing the INNER JOIN query, the row would be returned. This is because the INNER JOIN is joining only when the TranID matches. The reason why you are seeing the "same" information as the records from Query 1, is because you are looking only at the columns from #Transactions. If you were to show the columns from #VendorExport, then you would see the different results brought back by the JOIN.

    Finally, when doing the INTERSECT query, the row would not be returned. The reason why the INNER JOIN query returns 1099 rows while the INTERSECT query returns 0 rows, is because of the way INTERSECT (and EXCEPT, for that matter) works. INTERSECT returns the opposite result set of EXCEPT. It returns for you all the rows which are *exactly* the same between the two tables (while EXCEPT returns the rows which are *not* exactly the same).

Viewing 3 posts - 1 through 2 (of 2 total)

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