Yes it is a common case but I think the pipe.start_locally(run_pipeline_steps_locally=False) solved it. I started running it again and it seems to have passed the phase where it failed last time