2012-10-05 by Stefan Urbanek

Cubes 0.10 Released

After a while, here is an update to Cubes - Python Lightweight OLAP framework for multidimensional modeling. There are some changes included that were mentioned in the EruoPython talk such as table_rows and cross_table.

I recommend to look at updated examples in the Github repository. The Flask example is now "real" example instead of "sandbox" example and you can see how to generate a simple table for dimension hierarchy browsing.

There is also a more complex example with star-like schema dataset in the cubes-examples github repository. Follow the instructions in README files how to make it running.

There are some backward incompatible changes in this release – until 1.0 the "point" releases will contain this kind of changes, as it is still evolving. You can find more information below.

Quick Summary

  • Way how model is constructed has changed. Designated methods are create_model() or load_model()
  • Dimension defition can have a "template". For example:
      "name": "contract_date",
      "template": "date"
  • added table_rows() and cross_table() to aggregation result for more convenient table creation. The table_rows takes care of providing appropriate dimension key and label for browsed level.
  • added simple_model(cube_name, dimension_names, measures)

Incompatibilities: use create_model() instead of Model(**dict), if you were using just load_model(), you are fine.

New Features

  • To address issue #8 create_model(dict) was added as replacement for Model(**dict). Model() from now on will expect correctly constructed model objects. create_model() will be able to handle various simplifications and defaults during the construction process.
  • added info attribute to all model objects. It can be used to store custom, application or front-end specific information
  • preliminary implementation of cross_table() (interface might be changed)
  • AggregationResult.table_rows() - new method that iterates through drill-down rows and returns a tuple with key, label, path, and rest of the fields.
  • dimension in model description can specify another template dimension – all properties from the template will be inherited in the new dimension. All dimension properties specified in the new dimension completely override the template specification
  • added point_cut_for_dimension
  • added simple_model(cube_name, dimensions, measures) – creates a single-cube model with flat dimensions from a list of dimension names and measures from a list of measure names. For example:
model = simple_model("contracts", ["year","contractor", "type"], ["amount"])

Slicer Server:

  • /cell – return cell details (replaces /details)


  • creation of a model from dictionary through Model(dict) is depreciated, use create_model(dict) instead. All initialization code will be moved there. Depreciation warnings were added. Old functionality retained for the time being. (important)
  • Replaced Attribute.full_name() with Attribute.ref()
  • Removed Dimension.attribute_reference() as same can be achieved with dim(attr).ref()
  • AggregationResult.drilldown renamed to AggregationResults.cells (important)

Planned Changes:

  • str(Attribute) will return ref() instead of attribute name as it is more useful


  • order of dimensions is now preserved in the Model


Sources can be found on github. Read the documentation.

Join the Google Group for discussion, problem solving and announcements.

Submit issues and suggestions on github

IRC channel #databrewery on irc.freenode.net

If you have any questions, comments, requests, do not hesitate to ask.

2012-06-12 by Stefan Urbanek

Cubes and Slicer are going to EuroPython 2012

Cubes is going to EuroPython 2012.

EDIT: Added "Need help?".

There are going to be two sessions. First there will be talk introducing to light-weight OLAP with Cubes and Slicer, on Friday at 9:45 in room Tagliatelle (add to calendar). Afterwards there will be longer, more in-depth and hands-on training about Slicing and Dicing with Cubes on Friday at 14:30 in room Pizza Napoli (add to calendar)

In the talk I will introduce the framework and explain reasons for it's existence. Then I will dig into architecture, features and briefly show examples how to use it for slicing and dicing. Newbies are welcome.

The training will go into more details and the participants will learn:

  • how to prepare data for aggregated browsing - star and snowflake schemas
  • how to create a logical model, define cubes, dimensions and hierarchies
  • how to browse aggregated data and how to slice and dice cubes from within Python
  • how to create a WSGI OLAP server ("in 15 minutes" style) for aggregated data browsing and how to use it in your web application for providing (browsable) data to end-user reports
  • how to provide localized reporting

If the time permits, we can look at the anatomy of the framework and see how to implement a backend for another kind of data store.

I will be focusing on the existing SQL (relational OLAP) backend.

Customized examples

You might use the training session (and not only the session) to solve your problem - just bring your own sample data, if you like.

Do you have any data that you would like to slice and dice? Have a database schema and do not know how to create a logical model? You can send me a data sample or a schema, so I can prepare examples based on problem you are solving.

Please, do not send any confidential data or schemas under NDA.


Need help?

If you have any questions or would like to help with your data: from data preparation, through data modeling to slicing and dicing. You can grab me during the whole event. If you can not find me, just tweet me: @Stiivi.


If anyone is interested in participating in the project, he is welcome. Here are some features that are either out of scope of my skills and I would like to cooperate with someone more professional, or I do not have available resources to do that:

I am also very open to new feature suggestions and feature change requests. Just little note: Cubes is meant to be small and simple. At least for now. There are plenty of complex and feature-rich solutions out there. If we can make new, more complex features as non-critical, optional plug-ins, that would be great.

Links and Calendar Events

You can add the talks to your calendar by following the links:

2012-05-29 by Stefan Urbanek

Cubes 0.9.1: Ranges, denormalization and query cell

The new minor release of Cubes – light-weight Python OLAP framework – brings range cuts, denormalization with the slicer tool and cells in /report query, together with fixes and important changes.

See the second part of this post for the full list.

Range Cuts

Range cuts were implemented in the SQL Star Browser. They are used as follows:


cut = RangeCut("date", [2010], [2012,5,10])
cut_hi = RangeCut("date", None, [2012,5,10])
cut_low = RangeCut("date", [2010], None)

To specify a range in slicer server where keys are sortable:


Open ranges:


Denormalization with slicer Tool

Now it is possible to denormalize tour data with the slicer tool. You do not have to denormalize using python script. Data are denormalized in a way how denormalized browser would expect them to be. You can tune the process using command line switches, if you do not like the defaults.

Denormalize all cubes in the model:

$ slicer denormalize slicer.ini

Denormalize only one cube::

$ slicer denormalize -c contracts slicer.ini

Create materialized denormalized view with indexes::

$ slicer denormalize --materialize --index slicer.ini

Example slicer.ini:

denormalized_view_prefix = mft_
denormalized_view_schema = denorm_views

# This switch is used by the browser:
use_denormalization = yes

For more information see Cubes slicer tool documentation

Cells in Report

Use cell to specify all cuts (type can be range, point or set):

    "cell": [
            "dimension": "date",
            "type": "range",
            "from": [2010,9],
            "to": [2011,9]
    "queries": {
        "report": {
            "query": "aggregate",
            "drilldown": {"date":"year"}

For more information see the slicer server documentation.

New Features

  • cut_from_string(): added parsing of range and set cuts from string; introduced requirement for key format: Keys should now have format "alphanumeric character or underscore" if they are going to be converted to strings (for example when using slicer HTTP server)
  • cut_from_dict(): create a cut (of appropriate class) from a dictionary description
  • Dimension.attribute(name): get attribute instance from name
  • added exceptions: CubesError, ModelInconsistencyError, NoSuchDimensionError, NoSuchAttributeError, ArgumentError, MappingError, WorkspaceError and BrowserError


  • implemented RangeCut conditions

Slicer Server:

  • /report JSON now accepts cell with full cell description as dictionary, overrides URL parameters

Slicer tool:

  • denormalize option for (bulk) denormalization of cubes (see the the slicer documentation for more information)


  • important: all /report JSON requests should now have queries wrapped in the key queries. This was originally intended way of use, but was not correctly implemented. A descriptive error message is returned from the server if the key queries is not present. Despite being rather a bug-fix, it is listed here as it requires your attention for possible change of your code.
  • warn when no backend is specified during slicer context creation


  • Better handling of missing optional packages, also fixes #57 (now works without slqalchemy and without werkzeug as expected)
  • see change above about /report and queries
  • push more errors as JSON responses to the requestor, instead of just failing with an exception


Sources can be found on github. Read the documentation.

Join the Google Group for discussion, problem solving and announcements.

Submit issues and suggestions on github

IRC channel #databrewery on irc.freenode.net

If you have any questions, comments, requests, do not hesitate to ask.

2012-05-14 by Stefan Urbanek

Cubes 0.9 Released

The new version of Cubes – light-weight Python OLAP framework – brings new StarBrowser, which we discussed in previous blog posts:

The new SQL backend is written from scratch, it is much cleaner, transparent, configurable and open for future extensions. Also allows direct browsing of star/snowflake schema without denormalization, therefore you can use Cubes on top of a read-only database. See DenormalizedMapper and SnowflakeMapper for more information.

Just to name a few new features: multiple aggregated computations (min, max,...), cell details, optional/configurable denormalization.

Important Changes

Summary of most important changes that might affect your code:

Slicer: Change all your slicer.ini configuration files to have [workspace] section instead of old [db] or [backend]. Depreciation warning is issued, will work if not changed.

Model: Change dimensions in model to be an array instead of a dictionary. Same with cubes. Old style: "dimensions" = { "date" = ... } new style: "dimensions" = [ { "name": "date", ... } ]. Will work if not changed, just be prepared.

Python: Use Dimension.hierarchy() instead of Dimension.default_hierarchy.

New Features

  • slicer_context() - new method that holds all relevant information from configuration. can be reused when creating tools that work in connected database environment
  • added Hierarchy.all_attributes() and .key_attributes()
  • Cell.rollup_dim() - rolls up single dimension to a specified level. this might later replace the Cell.rollup() method
  • Cell.drilldown() - drills down the cell
  • create_workspace(backend,model, **options) - new top-level method for creating a workspace by specifying backend name. Easier to create browsers (from possible browser pool) programmatically. The backend name might be full module name path or relative to the cubes.backends, for example sql.star for new or sql.browser for old SQL browser.
  • get_backend() - get backend by name

  • AggregationBrowser.cell_details(): New method returning values of attributes representing the cell. Preliminary implementation, return value might change.

  • AggregationBrowser.cut_details(): New method returning values of attributes representing a single cut. Preliminary implementation, return value might change.

  • Dimension.validate() now checks whether there are duplicate attributes

  • Cube.validate() now checks whether there are duplicate measures or details

SQL backend:

  • new StarBrowser implemented:
    • StarBrowser supports snowflakes or denormalization (optional)
    • for snowflake browsing no write permission is required (does not have to be denormalized)
  • new DenormalizedMapper for mapping logical model to denormalized view
  • new SnowflakeMapper for mapping logical model to a snowflake schema
  • ddl_for_model() - get schema DDL as string for model
  • join finder and attribute mapper are now just Mapper - class responsible for finding appropriate joins and doing logical-to-physical mappings
  • coalesce_attribute() - new method for coalescing multiple ways of describing a physical attribute (just attribute or table+schema+attribute)
  • dimension argument was removed from all methods working with attributes (the dimension is now required attribute property)
  • added create_denormalized_view() with options: materialize, create_index, keys_only

Slicer tool/server:

  • slicer ddl - generate schema DDL from model
  • slicer test - test configuration and model against database and report list of issues, if any
  • Backend options are now in [workspace], removed configurability of custom backend section. Warning are issued when old section names [db] and [backend] are used
  • server responds to /details which is a result of AggregationBrowser.cell_details()


  • added simple Flask based web example - dimension aggregation browser


  • in Model: dimension and cube dictionary specification during model initialization is depreciated, list should be used (with explicitly mentioned attribute "name") -- important
  • important: Now all attribute references in the model (dimension attributes, measures, ...) are required to be instances of Attribute() and the attribute knows it's dimension
  • removed hierarchy argument from Dimension.all_attributes() and .key_attributes()
  • renamed builder to denormalizer
  • Dimension.default_hierarchy is now depreciated in favor of Dimension.hierarchy() which now accepts no arguments or argument None - returning default hierarchy in those two cases
  • metadata are now reused for each browser within one workspace - speed improvement.


  • Slicer version should be same version as Cubes: Original intention was to have separate server, therefore it had its own versioning. Now there is no reason for separate version, moreover it can introduce confusion.
  • Proper use of database schema in the Mapper


Sources can be found on github. Read the documentation.

Join the Google Group for discussion, problem solving and announcements.

Submit issues and suggestions on github

IRC channel #databrewery on irc.freenode.net

If you have any questions, comments, requests, do not hesitate to ask.

2012-05-12 by Stefan Urbanek

Star Browser, Part 3: Aggregations and Cell Details

Last time I was talking about joins and denormalisation in the Star Browser. This is the last part about the star browser where I will describe the aggregation and what has changed, compared to the old browser.

The Star Browser is new aggregation browser in for the Cubes – lightweight Python OLAP Framework. Next version v0.9 will be released next week.


sum is not the only aggregation. The new browser allows to have other aggregate functions as well, such as min, max.

You can specify the aggregations for each measure separately:

    "name": "amount",
    "aggregations": ["sum", "min", "max"]

The resulting aggregated attribute name will be constructed from the measure name and aggregation suffix, for example the mentioned amount will have three aggregates in the result: amount_sum, amount_min and amount_max.

Source code reference: see StarQueryBuilder.aggregations_for_measure

Aggregation Result

Result of aggregation is a structure containing: summary - summary for the aggregated cell, drilldown - drill down cells, if was desired, and total_cell_count - total cells in the drill down, regardless of pagination.

Cell Details

When we are browsing the cube, the cell provides current browsing context. For aggregations and selections to happen, only keys and some other internal attributes are necessary. Those can not be presented to the user though. For example we have geography path (country, region) as ['sk', 'ba'], however we want to display to the user Slovakia for the country and Bratislava for the region. We need to fetch those values from the data store. Cell details is basically a human readable description of the current cell.

For applications where it is possible to store state between aggregation calls, we can use values from previous aggregations or value listings. Problem is with web applications - sometimes it is not desirable or possible to store whole browsing context with all details. This is exact the situation where fetching cell details explicitly might come handy.

Note: The Original browser added cut information in the summary, which was ok when only point cuts were used. In other situations the result was undefined and mostly erroneous.

The cell details are now provided separately by method AggregationBrowser.cell_details(cell) which has Slicer HTTP equivalent /details or {"query":"detail", ...} in /report request. The result is a list of

For point cuts, the detail is a list of dictionaries for each level. For example our previously mentioned path ['sk', 'ba'] would have details described as:

        "geography.country_code": "sk",
        "geography.country_name": "Slovakia",
        "geography.something_more": "..."
        "_key": "sk",
        "_label": "Slovakia"
        "geography.region_code": "ba",
        "geography.region_name": "Bratislava",
        "geography.something_even_more": "...",
        "_key": "ba",
        "_label": "Bratislava"

You might have noticed the two redundant keys: _key and _label - those contain values of a level key attribute and level label attribute respectively. It is there to simplify the use of the details in presentation layer, such as templates. Take for example doing only one-dimensional browsing and compare presentation of "breadcrumbs":

labels = [detail["_label"] for detail in cut_details]

Which is equivalent to:

levels = dimension.hierarchy.levels()
labels = []
for i, detail in enumerate(cut_details):

Note that this might change a bit: either full detail will be returned or just key and label, depending on an option argument (not yet decided).


The Star Browser is being created with SQL pre-aggregation in mind. This is not possible in the old browser, as it is not flexible enough. It is planned to be integrated when all basic features are finished.

Proposed access from user's perspective will be through configuration options: use_preaggregation, preaggregation_prefix, preaggregation_schema and a method for cube pre-aggregation will be available through the slicer tool.


The new browser has better internal structure resulting in increased flexibility for future extensions. It fixes not so good architectural decisions of the old browser.

New and fixed features:

  • direct star/snowflake schema browsing
  • improved mappings - more transparent and understandable process
  • ability to explicitly specify database schemas
  • multiple aggregations

The new backend sources are here and the mapper is here.

To do

To be done in the near future:

  • DDL generator for denormalized schema, corresponding logical schema and physical schema
  • explicit list of attributes to be selected (instead of all)
  • selection of aggregations per-request (now all specified in model are used)


See also Cubes at github, Cubes Documentation, Mailing List and Submit issues. Also there is an IRC channel #databrewery on irc.freenode.net

Previous Page | Next Page