We are running AS 2000 (latest and greatest) on an 8 processor, 16 GB
RAM, 64 Bit Itanium server. We were processing some partitions in
parallel when we received the following error while processing (using
the DSO library):


File error [during the Open operation] 'Projection_3fact.data The
system cannot find the file specified. '

It seems the .data file for one of our partitions just disappeared.
Any ideas how this could happen?

Here is the relevant section of the processing log...
************************************************** ****
10/7/2005 1:11:06 AM 19647 Writing aggregations and indexes of
Partition 'Projection_3' (segment 38)
10/7/2005 1:11:06 AM 19647 Writing aggregations and indexes of
Partition 'Projection_3' (segment 39)
10/7/2005 1:11:06 AM 19647 Writing aggregations and indexes of
Partition 'Projection_3' (segment 40)
10/7/2005 1:11:06 AM 19647 Completed Processing Partition
'Projection_3'. End time: 1:11:06 AM Duration: 0:01:51 Rows
processed: 2560734
10/7/2005 1:11:06 AM 19647 Merging partitions of Partition
'Projection_3' Start time: 1:11:06 AM
10/7/2005 1:11:06 AM 19647 Initializing Partition
'Projection_3'
10/7/2005 1:11:07 AM 19647 Merging partitions of Cube
'Projection' Start time: 1:11:07 AM
10/7/2005 1:11:07 AM 19647 Merge in progress: Merging
partition '~Projection_3' into 'Projection_3'
10/7/2005 1:11:07 AM 19647 Analysis Server Error: File
error [during the Open operation] 'Projection_3fact.data The system
cannot find the file specified. ' ; Time:10/7/2005 1:11:07 AM
10/7/2005 1:11:07 AM 19647 Analysis Server Error: File
error [during the Open operation] 'Projection_3fact.data The system
cannot find the file specified. ' ; Time:10/7/2005 1:11:07 AM
10/7/2005 1:11:07 AM 19647 File error [during the Open
operation] 'Projection_3fact.data The system cannot find the file
specified. ' ; Time:10/7/2005 1:11:07 AM
10/7/2005 1:11:07 AM 19647 Processing Cube 'Projection' failed. No
changes have been made to the database.

Restore from tape?

<scott_gunn (AT) email (DOT) com> wrote

Do you know what is causing the "Initializing Partition" line in the
log?

It's been a few years since I wrote a Parallel processing engine for one
of my AS2k clients, but from memory...

1) I found the safest thing to do was to process all shared dimensions
on a single thread (or process) unless you can check all dependances -
processing a virtual dimension on one thread would try to kick off the
processing of a real dimension which may already be processing on
another thread, which would lead to a bit of a mess.

2) I also seem to remember doing an explicit "ProcessBuildStructure" on
all the cubes in a single thread, before I processed any partitions.
This would process any private dimensions and I think it also
initialised some base structures in the cube AND clears out the current
structure.

If you do not do this step I think the first partition you try to
process will attempt to implicitly build the structure and if you get
another "thread" trying to do the same thing you get issues.

This may be what you seeing, a second spinning up and performing an
implicit ProcessBuildStructure and clobbering the processing done on the
first thread.

Parallel processing in AS2k is a bit of a hack - roll on November and
AS2k5

--
Regards
Darren Gosbell [MCSD]
<dgosbell_at_yahoo_dot_com>
Blog: http://www.geekswithblogs.net/darrengosbell

We were able to do a full process on that partition to restore it.

Thanks Darren. The logic I'm doing here is an incremental update. I'm
not processing an entire cube. In fact, I can't do that here because
the cube is "live" and answering queries as this is going on.

Also, it has run successfully many times and only failed this once. It
always says "Initializing Partition xxx" before the merge.

The logic goes like this:
- Go through all 7 cubes and find the partitions for this customer
- Put those partitions in a queue
- Start a number of separate processes (2 in this case) to do the
actual cube processing
- Each process takes a partition out of the queue, builds a temporary
partition with the new data, and then merges the temporary partition
into the partition it got from the queue.

I'm not sure what part is responsible for the "Initializing Partition"
line in the log. I do have to change the SourceTableFilter on the
partition (and an Update()) before the merge. I suspect that's what's
causing the initialization, but I do all of those things in each
process protected by an OS Mutex. Only the actual Process() or Merge()
call is done outside of the OS Mutex, so I think I've covered my bases
there...