SQL：获取在另一个表列中具有最大值但还与某些跨表子句匹配的项

Question

This is probably a really basic question but I have spend a lot of time on this.

I have 2 Tables:

Files

| FileId | Type | FolderId |
----------------------------
|   1    | txt  |    11    |
|   2    | xml  |    15    |
|   3    | xml  |    17    |
|   4    | txt  |    19    |

Folders

| FolderId |   path    | version | typeId |        
-------------------------------------------     
| 11       | //c:/here | 1       |    1   | 
| 19       | //c:/here | 3       |    1   | 
| 15       | //c:/dummy| 6       |    1   | 
| 17       | //c:/dummy| 4       |    1   |

I want it to return: All info from Files , with the highest Folder.Version number, where the Files.type , Folders.path and Folders.typeID all match.

Essentially I need to group by Files.type , where Folders.path & Folders.typeID are the same. Then return the item with the max version field for each group.

So in the example above I would want the output:

| FileId | Type | FolderId |
----------------------------
|   2    | xml  |    15    |
|   4    | txt  |    19    |

So far I have the query below, which I think is ok for getting the items with the highest version number when TypeId and path match (I think????). But I cannot figure out how to group by Files.Type.

SELECT * FROM Files
WHERE FileId IN (
  SELECT FileId   FROM Files
  WHERE FolderId IN (  
    SELECT fo.FolderId                   
    FROM Folders fo                
    LEFT JOIN Folders fo2      
        ON fo.Path = fo2.Path AND fo.TypeId = fo2.TypeId    
        AND fo.Version < fo2.Version            
    WHERE fo2.Version IS NULL
  )
)

Can anyone explain how I can get this query done? I have a feeling I need some group by statements but I can't figure it out!

This is SQL server 2012.

Answer 1

You didn't specify a verson of SQL Server.

select Files.*
from
    Files inner join
    (
        select f1.FolderId
        from Folders f1 inner join
            (
                select TypeId, Path, max(Version) as MaxVersion
                from Folders
                group TypeId, Path
            ) as f2
            on f2.TypeId = f1.TypeId and f2.Path = f1.Path f2.MaxVersion = f1.Version
    ) Folders
        on Folders.FolderId = Files.FolderId

If you have a later edition of SQL Server then this should work too:

select Files.*
from
    Files inner join
    (
        select first_value(FolderId)
            over (partition by TypeId, Path order by Version desc) as FolderId
        from Folders
    ) Folders
        on Folders.FolderId = Files.FolderId

EDIT: Maybe this is the fix per your comment:

select Files.*
from
    Files fi inner join Folders fo on fo.FolderId = fi.FolderId
    inner join
    (
        select "Type", TypeId, Path, max(Version) as MaxVersion
        from Files fi2 inner join Folders fo2 on fo2.FolderId = fi2.FolderId
        group "Type", TypeId, Path
    ) as mv
        on      mv."Type" = fi."Type"
            and mv.TypeId = fo.TypeId and mv.Path = fo.Path
            and mv.MaxVersion = fo.Version

I'm in a rush but I think this might work on SQL Server 2012:

select distinct
    first_value(FileId)
        over (partition by "Type", TypeId, Path order by Version desc) as FileId,
    "Type",
    first_value(fo.FolderId)
        over (partition by "Type", TypeId, Path order by Version desc) as FolderId
from
    Files fi inner join Folders fo on fo.FolderId = fi.FolderId

Answer 2

Give this a shot:

select f.*
from files f
inner join folders dir on f.folderid = dir.folderid
inner join (
  select type, max(version) maxver
  from files fi
  inner join folders fo on fi.folderid = fo.folderid
  group by type
) t on f.type = t.type and dir.version = t.maxver

Result:

| fileid | type | folderid |
|--------|------|----------|
|      4 |  txt |       19 |
|      2 |  xml |       15 |

Since SQLFiddle SQL Server 2008 and 2014 were not responding well, I created an example with MySQL here: http://sqlfiddle.com/#!9/7abd4/13 . I expect the query results to be identical for SQL Server 2012.

Answer 3

Simple top-1-per-group, isn't it?

One possible variant using ROW_NUMBER .

PARTITION BY Files.Type, Folders.path, Folders.typeId define the groups, ORDER BY Folders.version DESC define which row to pick for each group.

More efficient than self-join.

WITH
CTE AS
(
    SELECT
        Files.FileId
        ,Files.Type
        ,Files.FolderId
        ,Folders.path
        ,Folders.typeId
        ,Folders.version
        ,ROW_NUMBER() OVER (
            PARTITION BY Files.Type, Folders.path, Folders.typeId 
            ORDER BY Folders.version DESC) AS rn
    FROM
        Files
        INNER JOIN Folders ON Folders.FolderId = Files.FolderId
)
SELECT *
FROM CTE
WHERE rn = 1
;