Going in Production - Expert Cube Development with SSAS Multidimensional Models - page 296

The script has detected that the following changes need to be made and made them:

• The 2003 yearly partition needs to be extended by one month to include July

2003. Since this month is already present as a monthly partition, it is merged
into the 2003 partition. This operation does not require that the 2003 partition
be reprocessed: Analysis Services is able to merge two already-processed
partitions without the need for any further processed.
When merging two partitions, Analysis Services will use the aggregation
design of the destination partition (in this case, the 2003 yearly partition),
but will disable any aggregations that are not present in the source partition
(July 2003 here). A ProcessDefault on the destination partition will rebuild
the invalidated aggregations.

• A new monthly partition for July 2004 is needed, and so it has been created.

After having performed the merge operation, the script updates all the annotations
inside the partition and also the SELECT statement the partition is bound to. This
ensures that the next time this partition is processed it contains the correct data.

There is no way to delete data from within an existing partition without reprocessing
it. In the previous scenario, when we created a new monthly partition for July 2004,
we might also have wanted to delete one month of data from the yearly partition for
2001 to maintain a consistent time window of data in the measure group. This would
have meant we had to reprocess the 2001 partition, though - a potentially expensive
operation. Instead, what the script does is only delete a partition at the start of the
time window when it contains a whole year's data. This has the added benefit that
any 'Same Period Previous Year' calculations that use the ParallelPeriod MDX
function described in Chapter 6 , Adding Calculations to the Cube , will always work
properly for the second year in the time window.

Managing processing
During development, we can process cubes or dimensions whenever we want;
in a production environment, processing needs more thought though. On a very
simple project, we might be able to get away with doing a Full Process on our entire
Analysis Services database if it only takes a few minutes to complete. However, we
usually have large data volumes to manage and a limited amount of time to perform
any processing, so a Full Process simply isn't feasible - we need to think carefully
about how we can only process the data that needs to be processed, and do that
processing in the most efficient way possible.