Multidimensional Sankey chart

[alesnav]

Hello there!

Sankey charts in Superset only support one source and one target, so it is not possible to create a Sankey chart like the one in the following example (random Google Images search):

I saw an old issue regarding this same request, but it was closed in 2023: #3059

It would be nice to have this capability, because many people use Sankey charts in their current dashboarding solutions (such as Splunk, for instance), so it is difficult to migrate to Superset when this is the case.

Thanks!

[manimovassagh]

This is definitely a gap. The current Sankey plugin wraps ECharts' Sankey series type, but only exposes a single source/target pair in the chart controls.

ECharts itself fully supports multi-level Sankey diagrams -- the underlying series.type: 'sankey' just needs nodes and links arrays with multiple hops. The limitation is in Superset's transformProps.ts for the Sankey plugin, which only maps two columns.

Workaround with SQL Lab + Virtual Dataset:

You can fake a multi-level Sankey by unioning your data into a flat source/target/value structure:

-- Level 1 -> Level 2
SELECT category AS source, subcategory AS target, SUM(amount) AS value
FROM my_table
GROUP BY category, subcategory

UNION ALL

-- Level 2 -> Level 3
SELECT subcategory AS source, detail AS target, SUM(amount) AS value
FROM my_table
GROUP BY subcategory, detail

Save this as a virtual dataset, then point the Sankey chart at it with source, target, and value columns. As long as the node names are consistent across levels (i.e., subcategory values match between the two unions), ECharts will render the full multi-hop flow.

Caveat: Node names must be globally unique across all levels. If "Marketing" appears as both a Level 1 and Level 3 category, you'll get unexpected links. Prefix them if needed (e.g., L1: Marketing, L3: Marketing).

For a proper fix, the Sankey plugin would need to accept an arbitrary number of dimension columns and generate the links array by chaining adjacent pairs. That would be a solid contribution if anyone's looking for a way into the Superset plugin ecosystem.