Handle derived metrics related to multiple models

[Vitor-Avila]

Current scenario

Currently, trying to sync a derived metric that relies on metrics from multiple models would fail with below error:

Metric {metricName} cannot be calculated because it depends on multiple models

This error prevents the sync from finishing successfully.

Suggested solution

Handle this scenario - some suggestions:

• Notify the user as a warning that the metric wasn't created because of this reason, but properly finish the operation;
• Create the metric on all target/involved datasets
• Ability to specify the desired dataset that should receive this metric on Preset, using the meta field for the metric.

[dweaver33]

Are we saying that currently if we use a derived metric, that none of the metrics will be deployed to Preset?

[Vitor-Avila]

Derived metrics in general are supported. The problem is when the metric derives from metrics from another models. For example, imagine you have:

• model_a with metric_a
• model_b with metric_b
• model_c with a derived metric that uses metric_a and/or metric_b

Currently the CLI wouldn't be able to sync model_c, since it's using a metric that is associated with another model, so likely the columns used by this metric doesn't exist on this model and therefore it wouldn't be possible to create a metric like that on Superset (since the metric is created on the dataset level).