> There is no excuse for not shutting down properly upon receipt of TERM.

This has been my belief for many years, but I find that it fails the
"does it scale?" test.

Suppose you have stateful database daemons that need to perform complicated
flushing on receiving TERM. How long is that going to take? What if the data
is on an NFS server and the daemons are deliberately designed to cache lots
of it?
(Think web server farm!)

Now add more daemons. How quickly can you contrive a realistic situation
where SIGTERM'ing everything causes the databases to all fight each other
for I/O and core, and the total response to SIGTERM takes more than 15
seconds.

Todd Whitesel
toddpw @ best.com