It does all residues (this appears to be the ClassicRelax algorithm).
For backbone, you know it does all residues because:
# 4. create a MoveMap, all backbone torsions free
movemap = MoveMap()
movemap.set_bb(True)
For sidechains, you know it does all residues because:
# 8. setup a PackRotamersMover
to_pack = standard_packer_task(starting_pose)
to_pack.restrict_to_repacking() # prevents design, packing only
to_pack.or_include_current(True) # considers the original sidechains
packmover = PackRotamersMover(scorefxn, to_pack)
Admittedly, #8 doesn’t _say_ all residues, but the default packer task is all residues, and this code does not restrict that.
If you wanted to relax a subset, alter the MoveMap created at 4, and the PackerTask created at 8 to allow movement at your subset of interest.
“Does the PyJobDistributor have any control over the # of residues to perturb? Or is only for parallel processing?”
No, and hope it never does – this would be a flagrant violation of encapsulation and make the code much harder to use. PyJobDistributor just lets you run the code in parallel in an easier-than-manual fashion. (Same for the C++ job distributors).
.
