We have a cube which is partitioned by month, and receives new data
files once a day. Data is loaded into a "holding" table, a temporary
partition is build based on that holding table, and then the temporary
partition is merged into the current month. This process had been
working just fine for the past year or more.

Recently, we started having problems with query performance for
particular queries, and decided to switch to usage-based optimizations
(replaced the current 500 aggregations with 68, and got a higher
percentage). This has worked out just fine for the query performance,
but now we've started having problems while merging the temporary
partition into the current month.

Normally, the merge step has been taking between 10 & 20 minutes to
perform per day. Now what's happening is that most of the time, the
merge step never finishes. If I watch the OLAP data folder, I can see
the temporary partition files being created, I can see the "'S" files
being created & growing as they normally would during the merge, but
at the point in time where they would normally disappear (replacing
the current month's partition), they don't. The Analysis Services
service keeps using 50-100% of the CPU, but nothing ever happens (I've
waited as long as 32 hours). None of the files in the OLAP data
folder are updated beyond the 10 minutes after the merge started, and
the OLAP service is doing something the entire time.

I've tried rebuilding the partitions, deleting the cube (and ensuring
that all files were deleted from the OLAP data folder), and rebuilding
from a script, but the same thing keeps happening. Worse yet, it will
sometimes work correctly. Usually after I've rebuilt the current
month, I can get 1 or 2 files to process, but then the same problem
comes back.

I've spent my last 2 weekends rebuilding this cube, and it's getting
old real fast. It's a fairly large cube, taking 10 hours/per month to
process, and currently containing 4 months.

Has anyone seen this type of behavior? Any help would be greatly
appreciated.

I too am facing the same issue. But in my case i do not
have partitions neither do i have large amounts of data. A
normal cube/dimension processing hangs with the "'S" files
getting created. And as you mentioned the appearance of
this problem is random.

Could this be a bug?