T-SQL 查询将数据插入到具有可变列数的表中

Question

I am trying to insert data from the table A, assuming the data is ordering in the way it looks, into the table B.

The way the data should be loaded into table B is by id_column. For example, the data which has id_column 1 has to be loaded into column C01. Note that not all the data has the same number of column_id like int the case of the BLOCK 1 or BLOCK 3.

The problem i am having is that the table B not always has the same amount of columns. In this case, it has 4 columns, but it could be 5 or more or less, so what i am trying is something that allow me the chance of work with different number of columns.

(When i have created previously table B, i did it thinking in the maximum id_column of table A)

Please, some kind of suggestion?

Answer 1

SQL tables represent unordered sets. Your ordering assumes an ordering in table A, so let me assume that you have such a column.

You can assign a grouping and then use conditional aggregation:

select max(case when column_id = 1 then value end) as c01,
       max(case when column_id = 2 then value end) as c02,
       max(case when column_id = 3 then value end) as c03,
       max(case when column_id = 4 then value end) as c04,
       'X' as con
from (select a.*,
             sum(case when column_id = 1 then 1 else 0 end) over (order by <ordering col>) as grp
      from a
     ) a
group by grp;

This assigns the groups by looking for column_id = 1 , which seems to indicate the start of a new block.

Answer 2

An INSERT statement can't handle a dynamic set of columns. You have to specify which columns your inserting into which means you have to SELECT a specific set of columns and in the same order that insert is in.

Having a table A like the one you have is a popular approach for dynamically created datasets and having a table like it is usually the end of of that data in the database. Instead of trying to interpret that data any other way in your database (like storing it in your table B as well), most applications just select from the table A and then parse the data out into your objects in code.

Another option I like to do for datasets that have unknown columns is to stored the unknown parts in an xml column and then have column for the stuff you do know.

An example would be

CREATE TABLE tblData
(
   BlockID int NOT NULL PRIMARY KEY,
   Data xml
)

Then in code you will build the xml block with all the random columns. And then parse it back out again in code. This gives you the flexibility of the unknown. Especially if some of those unknown columns might have additional meta data that you don't care to keep for other unknown columns.

Answer 3

Unfortunately, your only option is to do something using dynamic SQL. Here is some example code to do perform a dynamic pivot.

CREATE TABLE #tblA
(
    BLK VARCHAR(100)
    ,COLUMN_ID INT
    ,VALUE VARCHAR(100)
)

INSERT INTO #tblA
(
    BLK
    ,COLUMN_ID
    ,VALUE
)
VALUES
('BLOCK 1', 1, 'AA')
,('BLOCK 1', 2, 'BB')
,('BLOCK 1', 3, 'CC')
,('BLOCK 2', 1, 'AAAA')
,('BLOCK 2', 2, 'BBBB')
,('BLOCK 2', 3, 'CCCC')
,('BLOCK 2', 4, 'DDDD')
,('BLOCK 3', 1, 'AAAAAA')
,('BLOCK 3', 2, 'BBBBBB')
,('BLOCK 3', 3, 'CCCCCC')
,('BLOCK 4', 1, 'AAAAAAAA')
,('BLOCK 4', 2, 'BBBBBBBB')
,('BLOCK 4', 3, 'CCCCCCCC')
,('BLOCK 4', 4, 'DDDDDDDD')

--END DEMO SETUP

DECLARE @sql NVARCHAR(MAX) = N'';
DECLARE @cols NVARCHAR(MAX) = N''

--BUILD LIST OF COLUMNS TO PIVOT
SELECT @cols += ',' + QUOTENAME('CO' + CAST(COLUMN_ID AS VARCHAR(10))) FROM #tblA GROUP BY COLUMN_ID
SELECT @cols = STUFF(@cols, 1,1,'')
--SELECT @cols

SELECT @sql = 'SELECT BLK, ' + @cols
    + ' FROM (' 
    + ' SELECT BLK, ''CO'' + CAST(COLUMN_ID AS VARCHAR(10)) AS ColName, Value FROM #tblA'
    + ') data'
    + ' PIVOT('
    + ' MAX(VALUE)'
    + ' FOR ColName IN  (' + @cols + ')'
    + ') AS pvt'
PRINT @sql
EXEC(@sql)