具有一对多关系的多表联接

Question

Using SQL Server 2008.

I have multiple Locations which each contain multiple Departments which each contain multiple Items which can have zero to many Scans. Each Scan relates to a specific Operation which may or may not have a cutoff time. Each Item also belongs to a specific Package which belongs to a specific Job. Each job contains one or more Packages which contains one or more items.

+=============+                         +=============+
|  Locations  |                         |     Jobs    |
+=============+                         +=============+
      ^                                       ^
      |                                       |
+=============+     +=============+     +=============+
| Departments | <-- |    Items    | --> |   Packages  |
+=============+     +=============+     +=============+
                          ^
                          |
                    +=============+     +=============+
                    |    Scans    | --> | Operations  |
                    +=============+     +=============+

What I am attempting to do is get a count of Scans for a Job grouped by Location and Scan date. The tricky part is that I only want to count the first Scan by date/time per Item where the cutoff time for the Operation is not null. (NOTE: the scans definitely will NOT be in date/time order in the table.)

The query I have is getting me the correct results but it is painfully slow when the number of Items for a Job reaches 75,000 or so. I am pushing for a new server -- I know our hardware is lacking -- but I am wondering if there is something I am doing in the query that is bogging it down as well.

From what little I can glean from the execution plan, most of the cost of the query seems to be in the sub-query to find the first Scan for each Item. It does an index scan (0%) on an Operations table index (ID, Cutoff) and then a lazy spool (19%). It does an index seek (61%) on a Scans table index (ItemID, DateTime, OperationID, ID). The subsequent nested loops (inner join) is only 2% and the Top operator is 0%. (Not that I really understand much of what I just typed but I am trying to provide as much info as possible...)

Here is the query:

SELECT
    Departments.LocationID
    , DATEADD(dd, 0, DATEDIFF(dd, 0, Scans.DateTime))
    , COUNT(Scans.ItemID) AS [COUNT]
FROM
    Items           
    INNER JOIN Scans
        ON Scans.ID = 
    (
        SELECT TOP 1
            Scans.ID 
        FROM
            Scans
        INNER JOIN Operations
            ON Scans.OperationID = Operations.ID
        WHERE
            Operations.Cutoff IS NOT NULL
            AND Scans.ItemID = Items.ID             
        ORDER BY
            Scans.DateTime
    )
    INNER JOIN Operations
        ON Scans.OperationID = Operations.ID
    INNER JOIN Packages
        ON Items.PackageID = Packages.ID
    INNER JOIN Departments
        ON Items.DepartmentID = Departments.ID      
WHERE
    Packages.JobID = @ID        
GROUP BY
    Departments.LocationID 
    , DATEADD(dd, 0, DATEDIFF(dd, 0, Scans.DateTime));

Which will return a sampling of results like so:

8   2012-06-08 00:00:00.000 11842
21  2012-06-07 00:00:00.000 502
11  2012-06-12 00:00:00.000 1841
15  2012-06-11 00:00:00.000 4314
16  2012-06-09 00:00:00.000 278
23  2012-06-12 00:00:00.000 1345
6   2012-06-06 00:00:00.000 2005
20  2012-06-08 00:00:00.000 352
14  2012-06-07 00:00:00.000 2408
8   2012-06-11 00:00:00.000 290
19  2012-06-10 00:00:00.000 85
20  2012-06-11 00:00:00.000 5484
7   2012-06-10 00:00:00.000 2389
16  2012-06-06 00:00:00.000 6762
18  2012-06-09 00:00:00.000 4473
14  2012-06-10 00:00:00.000 2364
1   2012-06-11 00:00:00.000 1531
22  2012-06-08 00:00:00.000 14534
5   2012-06-10 00:00:00.000 11908
9   2012-06-12 00:00:00.000 47
19  2012-06-07 00:00:00.000 559
7   2012-06-07 00:00:00.000 2576

Here's the execution plan (not sure what I changed since the original post but the cost % are slightly different. The bottleneck still seems to be in the same area though):

Answer 1

I am a little leery about marking this as the answer as I am sure we can still squeeze a little juice out of the query. But this did knock my test run from 22 seconds down to 6 seconds (with an added index on Scans: OperationID including DateTime and ItemID):

WITH CTE AS 
(
    SELECT
        Items.ItemID AS ID          
        , Scans.DateTime AS [DateTime]
        , Operations.Cutoff AS Cutoff           
        , ROW_NUMBER() OVER (PARTITION BY Items.ID ORDER BY Scans.DateTime) AS RN
        FROM
            Items
            INNER JOIN Scans            
                ON Items.ID = Scans.ItemID
            INNER JOIN Operations
                ON Scans.OperationID = Operations.ID
            INNER JOIN Packages
                ON Items.PackageID = Packages.ID
        WHERE
            Operations.Cutoff IS NOT NULL
            AND Packages.JobID = @ID                        
)
SELECT
    Departments.LocationID
    , CTE.DateTime
    , COUNT(Items.ID) AS COUNT
FROM
    Items           
    INNER JOIN CTE
        ON Items.ID = CTE.ID
        AND CTE.RN = 1
    INNER JOIN Packages
        ON Items.PackageID = Packages.ID
    INNER JOIN Departments
        ON Items.DepartmentID = Departments.ID      
WHERE
    Packages.JobID = @ID
GROUP BY
    Departments.LocationID 
    , CTE.DateTime

Answer 2

Its hard to say for sure, but something like this may behave better. I replaced your nested lookup with a ROW_NUMBER call. The problem in your original query is that nested lookup- its killing you.

Note I don't have SQL in front of me, so I cannot test it, but I think it is logically equivalent.

SELECT
    Departments.LocationID
    , DATEADD(dd, 0, DATEDIFF(dd, 0, Scans.DateTime))
    , COUNT(Scans.ItemID) AS [COUNT]
FROM
    Items           
    INNER JOIN Scans
        ON Scans.ItemID = Items.ID
    INNER JOIN Operations
        ON Scans.OperationID = Operations.ID
    INNER JOIN Packages
        ON Items.PackageID = Packages.ID
    INNER JOIN Departments
        ON Items.DepartmentID = Departments.ID      
WHERE
    Operations.Cutoff IS NOT NULL
    AND
    Packages.JobID = @ID
    AND
    ROW_NUMBER () OVER (PARTITION BY Items.ID ORDER BY Scans.DateTime) = 1
GROUP BY
    Departments.LocationID 
    , DATEADD(dd, 0, DATEDIFF(dd, 0, Scans.DateTime));

Answer 3

I'm curious - could you please run CROSS APPLY version?

SELECT
    Departments.LocationID
    , DATEADD(dd, 0, DATEDIFF(dd, 0, CA_Scans.DateTime))
    , COUNT(CA_Scans.ItemID) AS [COUNT]
FROM
    Items 
    CROSS APPLY
    (
        SELECT TOP 1
            Scans.ID,
            Scans.OperationID,
            Scans.DateTime
        FROM
            Scans
        INNER JOIN Operations
            ON Scans.OperationID = Operations.ID
        WHERE
            Operations.Cutoff IS NOT NULL
            AND Scans.ItemID = Items.ID             
        ORDER BY
            Scans.DateTime
    ) CA_Scans
    INNER JOIN Operations
        ON CA_Scans.OperationID = Operations.ID
    INNER JOIN Packages
        ON Items.PackageID = Packages.ID
    INNER JOIN Departments
        ON Items.DepartmentID = Departments.ID      
WHERE
    Packages.JobID = @ID        
GROUP BY
    Departments.LocationID 
    , DATEADD(dd, 0, DATEDIFF(dd, 0, CA_Scans.DateTime));