SQL Server maximum 8KB per row?

I just happened to read the Maximum Capacity Specification for SQL Server 2008 and saw a ma开发者_JAVA百科ximum of 8060bytes per row? What the... Only 8KB per row allowed? (Yes, I saw "row-overflow storage" special handling, I'm talking about standard behavior)

Did I misunderstand something here? I'm sure I have, because I'm sure I saw binary objects with several MB sizes stored inside SQL Server databases. Does this ominous per row really mean a table row as in one row, multiple columns?

So when I have three nvarchar columns with each 4000 characters in there (suppose three legal documents written in textboxes...) - the server spits out a warning?

Yes, you'll get a warning on CREATE TABLE, an error on INSERT or UPDATE

LOB types (nvarchar(max), varchar(max) and varbinary(max) allow 2Gb-1 bytes which is how you'd store large chunks of data and is what you'd have seen before.

• For a single field > 4000 characters/8000 bytes I'd use nvarchar(max)

• For 3 x nvarchar(4000) in one row I'd consider one of:

  • my design is wrong
  • nvarchar(max) for one or more column
  • 1:1 child table for the "least populated" columns

2008 will handle the overflow while in 2000, it would simply refuse to insert a record that overflowed. However, it is still best to design with this in mind because a significant number of records overflowed might cause some performance issues in querying. In the case you described, I might consider a related table with a column for document type, a large field for document and and a foreign key to the intial table. If however it is unlikey that all three columns would be filled in the same record or at the max values, then the design might be fine. You have to know your data to determine which is best. Another consideration is to continue as you have now until you have problems and then replace with a separate document table. You could even refactor by renaming the existing table and creating a new one and then creating a view with the existing tablename that pulls the data from the new structure. This could keep alot of your code from breaking although you would still have to adjust any insert or update statements.