Some Task Groups don't Expand when Clicked

[MatrixManAtYrService]

Apache Airflow version

2.2.0b2 (beta snapshot)

Operating System

debian (docker)

Versions of Apache Airflow Providers

n/a

Deployment

Astronomer

Deployment details

astro dev start

Dockerfile:

FROM quay.io/astronomer/ap-airflow-dev:2.2.0-buster-44417

dag:
https://gist.github.com/MatrixManAtYrService/a3ee53dab27edf8249d0ff0dfd11c76e

The image above was generated with the command:

airflow dags show structure_hash | grep digraph -A999 | dot -Tpng  > output.png

What happened

I have other DAGs with task groups, and they expand/collapse just fine, but this one refuses to expand when I click its blue box.

Instead my browser's js console shows an error:

Uncaught SyntaxError: Invalid or unexpected token
dagre-d3.min.js:1642 Uncaught TypeError: Cannot convert undefined or null to object
    at Graph._removeFromParentsChildList (dagre-d3.min.js:1642)
    at Graph.setParent (dagre-d3.min.js:1642)
    at graph.d7b2753a50a36b9c46ed.js:1485
    at Array.forEach (<anonymous>)
    at expandGroup (graph.d7b2753a50a36b9c46ed.js:1470)
    at SVGGElement.<anonymous> (graph.d7b2753a50a36b9c46ed.js:1044)
    at SVGGElement.__onclick (d3.min.js:1)

What you expected to happen

The blue boxen expand when clicked, ultimately exposing the structure show in the Deployment Details section.

How to reproduce

Load the dag mentioned above, navigate to its graph view, try to expand the task groups

Anything else

Sorry the DAG is so weird. I spent a little time trying to cobble together components that reproduced this more minimally, but I didn't stumble across whatever secret sauce makes this one special.

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[bbovenzi]

Interestingly, they render just fine if they start expanded and if you close it. The only problem is opening them. I'll keep investigating.

[MatrixManAtYrService]

I spent a while trying to reduce DAG complexity towards a more minimal case, and I was able to get them quite minimal indeed while still reproducing the issue. But then later when I tried starting from nothing and building up to the minimal case I couldn't reproduce it--which is why I ended up posting the whole thing.

My feeling was that something was being remembered between iterations that I didn't expect to be. I didn't take careful notes during this exploration though, so it remains just a feeling.

[bbovenzi]

Just tested with the new graph view, #29413. This issue no longer happens.