KDC worker processes project

[Greg Hudson]

On Fri, 2010-09-17 at 15:31 -0400, Nicolas Williams wrote:
> That's the flaw though...  It really should wait in an I/O loop.  The
> wait should happen as a result of processing an I/O event.

Why's that?  Even if the supervisor wanted to listen for I/O on the
routing socket, it could just have its select() interrupted by SIGCHLD
and then call waitpid() (repeatedly with WNOHANG) until it sees ECHILD.

The difficulty with Ken's plan isn't so much wait vs. select.  The
problem is that net-server.c would need to grow more complexity and more
application interconnects in order to meet the new needs of the
supervisor as well as the normal KDC and kadmind workload.  net-server.c
is already 2000 lines of pretty complicated code (and it's not badly
written; it just has a very complicated job).