Thanks @karthik_p for your time looking into this.
It's a job cluster.
I have the same error on all job cluster that I checked so far (5 out of ~20).
The cluster has nothing special about it that I can think of.
There are no GPUs, not a huge amount of workers, ...
Driver: m5d.xlarge, Workers: m5d.xlarge, 1-8 workers, On-demand and Spot, fall back to On-demand, 11.3 LTS (includes Apache Spark 3.3.0, Scala 2.12), auto