Solve “Target model is invalid” in GoodData

Fundamental to data modeling is that you need a fact table or a table with a unique key per row. That is no different in GoodData. In this blog post, I tackle a specific error that you might run into if your dataset doesn’t meet the necessary requirements.

When I tried to publish my data source (BigQuery) into a logical data model (LDM), GoodData showed me the following error.

I highlighted a specific part of the error that is of relevance:

Target model is invalid.: [A dataset must have either a fact or an anchor with a label or a valid grain. Dataset […] does not contain any fact. Add an anchor with a label or a valid grain to dataset […] If the dataset has a Fact Table Grain, try to query the model with the parameter includeGrain=true (model/view?includeGrain=true)]: The request could not be understood by the server due to malformed syntax.

When you generate the output stage in the data integration console, you get a query (or multiple) that generates a view within your data warehouse. It’s this view that GoodData will use to load the data.

-- gd_view_event_stream_gooddata --
--------------------------------------------------
CREATE OR REPLACE VIEW <dataset>.gd_view_event_stream_gooddata AS SELECT
		browser AS a__browser,
		browserLanguage AS a__browserlanguage,
		browserVersion AS a__browserversion,
		deviceCategory AS a__devicecategory,
		[...]

FROM <project>.<dataset>.view_event_stream_gooddata;

You might have noticed that in the query, GoodData creates a prefix for every column name. These prefixes indicate that the column is of a certain type. The prefixes that we need to create a valid data source is either cp__ or f__. If that’s not the case, you’re gonna have to make modifications to the view.

Solution one: create facts

The first (rather suboptimal) solution is to create a fact. If there’s no real “fact” in your data, for example when every row is a hit or an event, you can simply add a column of 1’s to your table. Like this:

1 as f__hitcount

Now, GoodData will no longer produce an error when you create a logical data model (LDM).

Solution Two: Create anchors

A more correct way is to generate a unique key for every row. Instead of counting the facts, GoodData’s visualizations will count unique anchors. You can achieve this by adding a cp__ (connection point) column to the view. Many SQL syntaxes have a function to generate such a key — and so do BigQuery and Snowflake.

GENERATE_UUID() as cp__id -- Google BigQuery
UUID_STRING() as cp__id -- Snowflake

If you want to generate a UUID in RedShift, follow this tutorial.

Now, you can start analyzing your data in GoodData.