I'm reviving that subject following the discussion with cloomis today.
I think we manage to distinguish our needs and expectations, I will be trying to rephrase it here to avoid further confusions. I'm going to start with what we currently have :

• a per actor configuration file in $ICS_ACTOR_DIR/etc/
• InstData class in pfs_utils to persist/load mhs keywords to/from disk.

What we want is to move as much information as possible for the etc file to a common repo.
yaml is very likely to be right choice, so that we can factorize configuration shared among identical module/unit and specify per-instance dedicated values.
This/these file(s) are edited/written by human in most (all?) cases.

The persistence mechanism is completely machine-oriented, it's read and written by a program.
It can be seen as an instrument snapshot, so that some variable/state can be reloaded if needed for instance after :

• software reboot
• some power cycle
• temporary hardware issue...

Those two mechanisms are actually different and so their time constant.
I hope it clarifies the big picture.