[erlang-questions] Mnesia k-safetey

[James Laken]

Hi All,

I am trying to implement automatic k-safety with mnesia tables and 
fragments. In this context the k-safety means that each table or 
fragments have at least K number of replica across N number of machines.
- On node crash each remaining and non-overloaded node should "pick" 1/N 
of the data from the crashed node replicas.
- On node join, some tables (fragments) replicas migrated to the new 
node, to provide load-balancing.
At first sight it may looks easy, because Mnesia support online node and 
fragment addition/deletion - but doing it in automatic way, with error 
handling and fault tolerance and without race condition that makes it 
difficult (or at least, to me).

What is the correct way to implement this kind of setup? Any experience 
or advice?

Regards,
James