I now wrote a function to flatten the list of lists to a list. Probably right now the best way.

Using a Dask Worker, is it good to map over thousands of list items? Or should I loop within a single task? Each iteration should perform a web-request

As for whether you should map or loop - it’s a subjective question along two dimensions: 1) resource availability and 2) unit of work preference. Mapping will give you greater visibility (and performance) than looping, because each item becomes its own task. However, it comes with proportionately greater resource requirements for passing the data around and executing the operation in parallel. In general, thousands of items is no issue for a sufficiently large Dask cluster, so I’d suggest mapping up until you hit a memory constraint.