Why

on_pipeline_error

is not executed when there's an exception in the dataset implementation? 🤔

that’s interesting

I've raised ValueError in the `_load / __init__`and my

on_pipeline_error

was not called

Full stacktrace? Maybe it error out early even before entering the pipeline?

OK, I know what the problem was. I was initializing an experiment tracker session in the

after_context_created

and it was too early in the lifecycle - I couldn't capture failures in

on_pipeline_error

if a dataset raised exception in

__init__

. Switching the session initialization from

after_context_created

to

before_pipeline_run

does the job better (as the exp. trck session will not be initialized when the exception in the dataset occurs). Everything is fine then. Sorry for the confusion sadcat2