@Kaniz Fatma​ , or anyone at Databricks, why does this have to be so 'hacky' to import from a different repo in the Workspace on all the executors? Is it possible for the root of the Workspace to be on the python path by default, rather than only the repo of the running Notebook? Is that part of the product roadmap, or is this not a practical option for some reason?

I realize that creating a wheels build of a package and installing it on all the nodes would work, but I think our system engineers want to avoid installation of custom packages, so we were looking at the Repos as a potential way to use custom packages that can be reused across many notebooks, with those notebooks potentially being in different Repos, so nesting the custom package within a notebook Repo or all of those notebook Repos would not be practical.

But actually, I can't even get this suggestion by @Scott.B (Customer)​ to work.  The executors on the worker nodes just can't seem to find the package.

Just like the OP, I can import the package fine except for when I wrap it in (in my case) a call to DataFrame.applyInPandas(), rather than wrapping in a UDF.