For parallelized runs, the worker.py script should take a command line option which specifies which chunks to resume or to run from scratch, instead of having to run them all. Sometimes failure is limited to a few chunks only, no point in redoing the good ones.
Version: unspecified
Severity: enhancement