Quoth prasad on Mon, May 08, 2006 at 09:08:38AM -0700:
> - We have implemented the above dependencies in pre Solaris 10 using
> run level scripts and exited when there was a fatal failure by doing
> a "exec ksh -o vi" which took us to a console prompt where we could
> debug the issue
>
> In Solaris 10 FCS and SMF world, we have kludged our run level scripts by
> doing the following:
> - sourced /lib/svc/share/smf_include.sh so that we can write to the
> console using "echo msg | smf_console"
> - since S99dtlogin is not a SMF service yet, we could exit to the
> console prompt using "exec /bin/ksh -o vi </dev/console >/dev/console
> 2>&1"
>
> One of the problems we face now is that when we exit it to the console
> prompt during a fatal error, it seems to be unfriendly for debugging
> i.e. can't use vi for long (maybe it has some buffer) and the "console
> login message" appears constantly.
Indeed, interactive rc scripts don't work anymore. For your goals,
instead of exec'ing a shell on failure, I would echo a message to
smf_console and then
while true; do
sleep 1800
done
Eventually the service will time out and svc.startd will put it into
maintenance. I believe system/console-login is independent of
multi-user, so you should get a login prompt, and then if you saw the
error message you could begin debugging. If console-login depends on
multi-user somehow, then when it enters maintenance, svc.startd should
start sulogin, which will let you log in and debug.
Sleeping until timeout is inefficient, but unfortunately svcadm mark -I
maintenance doesn't work right now. You could kill all of the processes
in the contract, if you're willing to do some scripting.
> In future releases of Solaris 10, dtlogin maybe a service which means
> that my fatal exits in my rc2.d scripts would not work? What are my
> options to implement my custom scripts?
In snv_39, at least, cde-login depends on multi-user, which runs the
rc2.d scripts. So if it goes into maintenance, then dtlogin will not be
started.
David
