Partition switching on a nonaligned table?

Question

hi folks,
i was wondering if it was possible to switch between partitions when utilized a non aligned table?
DECLARE @StartMonth DATE= '2014-01-01';
CREATE PARTITION FUNCTION MonthlyProductPF (DATETIME)
AS RANGE RIGHT FOR VALUES
(@StartMonth
, DATEADD(mm,1,@StartMonth)
, DATEADD(mm,2,@StartMonth)
, DATEADD(mm,3,@StartMonth)
, DATEADD(mm,4,@StartMonth)
, DATEADD(mm,5,@StartMonth)
, DATEADD(mm,6,@StartMonth)
, DATEADD(mm,7,@StartMonth)
, DATEADD(mm,8,@StartMonth)
, DATEADD(mm,9,@StartMonth)
, DATEADD(mm,10,@StartMonth)
, DATEADD(mm,11,@StartMonth)
CREATE PARTITION SCHEME MonthlyProductPS AS PARTITION MonthlyProductPF TO (
MonthlyProductFG1
,MonthlyProductFG2
,MonthlyProductFG3
,MonthlyProductFG4
,MonthlyProductFG5
,MonthlyProductFG6
,MonthlyProductFG7
,MonthlyProductFG8
,MonthlyProductFG9
,MonthlyProductFG10
,MonthlyProductFG11
,MonthlyProductFG12
,MonthlyProductFG13
,MonthlyProductFG14
I have a table with the following clustered index that sit on PRIMARY
ALTER TABLE [dbo].[Product] ADD CONSTRAINT [PK_Product] PRIMARY KEY CLUSTERED ([Id] ASC) ON [PRIMARY]
GO
and then I got a non clustered index sitting on the
CREATE NONCLUSTERED INDEX [IX_Product_OrderDate] ON [core].[OrderProduct]
[OrderDate] DESC
) ON [MonthlyProductPS](OrderDate)
GO
when I run:
alter table dbo.Product switch partition 5 to ph.ProductToPurge partition 14
I get an error:
Warning: The specified partition 5 for the table 'db.dbo.Product' was ignored in ALTER TABLE SWITCH statement because the table is not partitioned.
Msg 7733, Level 16, State 3, Line 1
'ALTER TABLE SWITCH' statement failed. The index 'IX_Product_OrderDate' is partitioned while table 'db.dbo.Product' is not partitioned.
I have also tried to disable the PK and got the error:
The ALTER TABLE SWITCH statement failed. The table "db.dbo.Product" has a disabled clustered index.
any suggestions on getting pass these error?
Thanks,

Answer 1

Partition switching is only possible when all indexes are placed on the partition scheme. Or at least a partition scheme using the same partitioning function.
The reason for this is simple to understand. Say that you have table that is partitioned on OrderDate. Then you have an index on CustomerID which is placed on PRIMARY. Now you want to make a partition switch to get rid of the orders from Dec 2013. But if
you were permitted to this in this situation, the index on CustomerID would still have pointers to the order from Dec 2013. This may sounds benign it isn't. This query:
   SELECT COUNT(*) FROM Orders
would return different results, depending on which index the optimizer decides to use.
In your case, it is the other way round, it is the index that is partitioned, but the table that is not. That is a little, well, unusual.
But if you want to be able to do paritition switching to age out old data fast, you need all indexes to be aligned, and the table you switch with needs to have the same set of indexes. Only when this is fulfilled, you do a metadata-only operation like a
partition switch.
Erland Sommarskog, SQL Server MVP, [email protected]