MSSQL 2005 filtered index non-unique column return unique result set vs indexed view
Posted on 2010-11-15
I have an MSSQL 2005 table that has transaction-related data, which may have multiple entries per significant column. It is unique only on its sequential ID key (primary key).
I need to return, if possible, the data filtered down to the most recent entry for each key field value and was wondering if it's possible to do that with a filtered key.
id, order_num, order_ref_num, ts
Where id is an auto-incrementing sequential unique ID, order_num and order_ref_num are text fields and ts is a date/time field (timestamp.) There are occasions where multiple order_num+order_ref_num sets have the same timestamp value, because the format was changed from logging date only to logging time and date, so the uniqueness and filtering of most-recent order/ref pairs can't be gained by using timestamp.
The order_num field would relate to an order table. There may be multiple order_ref_num values associated with a particular order_num, and there may be multiple records with the same order_num/order_ref_num set but different timestamps.
Can the most-recent unique record be returned using a filtered index, or does it have to be filtered down in the query? If it can be done with a filtered index, what would be your recommendation for setting up that index. The key field would be order_ref_num, and would need to return unique order_ref_num and order_num combinations.
If not, can it be accomplished using an indexed view, and if so, how? Are complex SELECT statements available for creating an unique-valued indexed view per the requirements stated earlier? What would be the potential performance impact of an indexed view vs a filtered index?
By the way, the key fields of the order table are order_num and transaction_id, which set also is non-unique and would also have to be filtered down to the most-recent records of each key set to make them unique as well. It also has a primary key of a sequential ID.
I was looking on technet and don't see whether it's possible to do an equivalent of select distinct or similar to create a filtered index, and it doesn't really say if indexed views can use anything like select distinct either. TOP is mentioned, but isn't likely to be of any use for this scenario.