DragonFly users List (threaded) for 2012-02
Re: running vkernels in production
On 02/03/12 11:08, n0g0013 wrote:
> personally, i think it would be nice to have something functional
> in base for this and would be fairly confident about changing
> `init` and/or `getty` if i thought people were open to that change
This topic has come up a few times in the past as well but
unfortunately no one has put something together as a solution -
My current 'theoretical' solution would be to have a 'libpty' which
basically encapsulates the functionality of e.g. screen/tmux without
the extra session managment, etc. overhead. I hopefully plan on
working on this - but have been saying that since vkernels came
out ~5y ago :| ..
Applications could then use this thin wrapper library around pty
functionality to create 'pty sockets' in a specified location. The
library/system would also have a 'pty' command which could be used
to (re)connect/disconnect to these sockets - kind of like a single
command for an individual screen/tmux 'window'. (and would probably
also support simple single-window screen/tmux like functionality)
Then, vkernel could be modified to link against this library and
vkernel users could just specify a certain location for the vkernel
to create a 'pty socket' to 'serve' the console.
This would also be compatible with a 'notty' launch of vkernels
from init - and also would be more generic to cover the startup
script and other cases, etc.
Personally I think if someone was going to put effort into it, it
would make more sense to make vkernels 'tty agnostic' than to hook
init tty handling to support vkernels better, but whatever is
clever I suppose - the two do not conflict necessarily.