Instead of just logging errors let pipeline run until the end, collect errors and fail if there are any errors which we can't ignore.