[Mondrian] Closure tables, and aggregation
rolf.lear at algorithmics.com
rolf.lear at algorithmics.com
Thu Feb 19 13:45:53 EST 2009
Hi all.
I believe there is a bug in the implementation of aggregation tables
where one of the aggregated dimensions is a parent/child hierarchy with
a closure table.
The description of the bug is as follows:
1. When loading the RolapCube data for a cube (for example, the HR cube
in FoodMart), the code encounters a Closure delcaration in the XML for
the parent/child Level.
2. This event triggers the RolapCubeLevel.init() method to create a
"closedPeer" reference to a new RolapCubeLevel based on a 'trick'
dimension called "Employees$Closure".
3. This trick dimension is created with the RolapCube as it's parent,
but it is never actually 'registered' with the RolapCube by adding it to
the RolapCube.dimensions Dimension[] array.
4. When the Aggregation Loader scans tables in the system to identify
potential Aggregation tables, it checks each candidate table against
iether the 'search patterns', or any explicitly declared AggName
definitions for the fact table.
5. in the case of an explicit AggName definition, the code identifies
which columns on the candidate aggregation table match columns on the
fact table (Fact count, measures, levels, etc.).
6. the mechanism it uses (in
mondrian.rolap.aggmatcher.Recognizer.checkLevels()) to identify levels
is to scan all available cubes, and each dimension and hierarchy in
those cubes. The documentation for the combined Closure/Aggregation
concept indicates that the Hierarchy will be called [Employees$Closure],
but, because, in step 3, the trick dimension is not actually registered
with the cube, the hierarchy is never available to be scanned. As a
result, the explicit aggregation table ends up with columns that have no
explicit usage, and causes the candidate aggregate table to be rejected.
7. because an explicit aggregation table is rejected, the
intitialization of Mondrian fails with 'too many errors'.
Since I have not been able to get things working, and have tried a
number of mechanisms, the following example will fail on just the first
error I describe above. It is possible that the naming convention of the
way it is supposed to work is different from the example I give.
To reproduce the problem I invite you to change the HR cube to:
<Table name="salary">
<AggName name="FoodMart_HR_1">
<AggFactCount
column="fact_count" />
<AggMeasure
column="salary_paid" name="[Measures].[Org Salary]" />
<AggLevel
column="employee_id" name="[Employees$Closure].[Closure]" />
</AggName>
</Table>
Then, create the table:
CREATE TABLE "FoodMart_HR_1" (
"employee_id" NUMBER(38),
"salary_paid" NUMBER,
"fact_count" INTEGER);
I am using Oracle, so forgive the DB specific statement.
There is no need to populate the table to encounter the problem.
By loading the new mondrian schema you will encounter the problem.
For the record, I have debugged this code, and tried to correlate the
actual implementation with the documentation described here:
http://mondrian.pentaho.org/documentation/aggregate_tables.php#Combined_
closure_and_aggregate_tables
and I find that the differences are significant. The actual names of the
Hierarchies and Levels in the 'trick' dimension are significantly
different to the implied values in the web page.
Finally, I have tried to resolve the problem by modifying the code. My
initial attempt was to add the dimension to the Dimension[] array on the
cube, but this has issues with the dimension naming comvention. The
second attempt is to add a new hierarchy to the existing Employees
dimension, but the problem there is that there are 'deep' problems with
the fact that the table for the closure level is 'employee_closure', but
this is not actually part of the RolapStar for the HR cube, and thus has
other issues.
Any assistance/insight is much appreciated.
Thanks in advance.
Rolf
--------------------------------------------------------------------------
This email and any files transmitted with it are confidential and proprietary to Algorithmics Incorporated and its affiliates ("Algorithmics"). If received in error, use is prohibited. Please destroy, and notify sender. Sender does not waive confidentiality or privilege. Internet communications cannot be guaranteed to be timely, secure, error or virus-free. Algorithmics does not accept liability for any errors or omissions. Any commitment intended to bind Algorithmics must be reduced to writing and signed by an authorized signatory.
--------------------------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.pentaho.org/pipermail/mondrian/attachments/20090219/d18f33ce/attachment.html
More information about the Mondrian
mailing list