See comment in ExposureProcessor.download_and_load_exposure in pipeline/pipeline_exposure_processor.py:
# TODO : this Exposure object is going to be copied into every processor subprocess
# *Ideally* no data was loaded, only headers, so the amount of memory used is
# not significant, but we should investigate/verify this, and deal with it if
# that is not the case.
See comment in
ExposureProcessor.download_and_load_exposure
inpipeline/pipeline_exposure_processor.py
: