Thu Jun 30 23:33:29 EDT 2016
I've been trying to add "let it crash" functionality to non-Erlang
processes, and it is quite hard to make this work in the presence of
Erlang VMs dying without assuming that cleanup needs to be done before
something is started. I.e. to clean up the mess left by a previous
Is this a good way of doing things? It breaks modularity,
encapsulation: assumes things about before while we really only know
So design everything in such a way that a VM doesn't need to be
restarted. If it dies, reboot the entire machine or at least perform
a reset that performs the same relevant initializations.
The problem really is state, and while it's possible to do RAII in
erlang, it hinges on external processes cleaing up state when they die.