Support multiple tables / star schema

[pudo]

Right now, only a single fact table is supported, and aggregations of dimensions will assume that all columns for a given dimension are perfectly denormalised. The application should support JOINs for a star schema instead.

[pwalsh]

@akariv first we need to understand if we need this. Assigning to you to investigate as part of general deep dive on babbage.

[pudo]

Might be useful to separate between being able to load from multiple resources and keeping multiple tables for dimensions internally. The two are almost certainly a "false friend" equality (i.e. looks like the same thing but in practice turns out not to be one).

[pwalsh]

@pudo totally agree. seems to me that single resources in FDP speak could (should?) load into multiple tables (star schema).

I definitely do not want babbage to change to reflect the various modelling strategies in an FDP mapping: rather, that the loader can work with the mapping to write to a consistent (normalised?) babbage model.

I guess you are currently in the best position to advise on a way forward around that.

[pudo]

There was an interesting thought from @Stiivi (also relevant to #4): he proposed loading the source data facts into a JSON column on the fact table (i.e. as-is) and then generating the actual columns and star schema as the model is defined through PostgreSQL commands (i.e. inside the DB).

Otherwise you end up in that spot where each time someone changes the label for an attribute, you have to re-load 900 MB of source CSV.

[pwalsh]

I'm going to backlog this now. We don't need this now, and while there are great ideas in here, let's focus on our core tasks first.

[larjohn]

I am not sure if now is the time to pop this out of the backlog.

We are interested in using such a feature in openbudgets.eu. Other than the backend implementation, in which I can't offer any opinion yet, what is the suggested way to support this API-wise?

Would it be a call to the aggregates/facts etc endpoints without a cube path part?

[akariv]

I would say that from an API point of view, nothing needs to change.
The API should reflect to the user a denormalized data table, while using normalized tables in the backend.

[larjohn]

Yes, but how do you imagine a call of this service? This call:

http://next.openspending.org/api/3/cubes/405f5e1681d4b6673fbfc0ceb196fe47:boost-armenia-2006-2014/aggregate?drilldown=activity.program&pagesize=30&cut=classification.econ_func%3A%22Economic%22&order=adjusted.sum%3Adesc

already contains the cube name in it. Will "cube-less" queries be supported somehow?