SUMMARY: Jumpstart

Thanks to:
Darren Dunham		        Jed Dobson
Jesse Adam			Thomas 'Mike' Michlmayr
Graham Wood			Kevin Amorin
Gwendolynn ferch Elydyr		Bertrand_Hutin
Ed Schernau			Paul Greidanus

The suggestion by Jed and Bertrand is what we are looking for -
mconnect.  Jed even wrote a little example script, thanks Jed.  Do a
'man mconnect' and see what I mean.  I didn't even know it existed.  The
only problem is that it doesn't exist in the mini-root and it's
dynamiclly linked.  So we are going to talk to Sun and see if we can get
a staticly linked copy that we can throw in the mini-root (unless of
course someone has a better idea...).

The ever-popular tcpwrappers also came in by Jed, which is a probability
since we intend to install these on our machines anyway for security. 
Why not use it for this as well?  We may very well do just that.

Paul suggested netcat as a possibility.  I think this deserves some
attention as it seems like a good tool.  Originally it was thought to
use this from the client side but a brief perusal of the readme showed
it also being used in server mode.  I like this idea better.  We will
look into this even if the static mconnect comes through as it could be
used for more than this.

The next one in line is to run in.rarpd and rpc.bootparamd in debug mode
and see what we get.  Chances are it would work and not be too noisy
since not much in our network uses bootp.  However, it is not as elegant
or simple as mconnect (providing we can get a binary in mini-root of
course) nor as versatile as netcat appears to be.

Some people were wondering why a begin script wouldn't work.  Mainly
because (outside of mconnect) you don't have much to work with from the
client side, you're dependant on what's in the mini-root (if you're even
willing to wait that long) and that's mounted read-only anyway so forget
spooling.  Also, I want to be notified when a client starts a
jumpstart.  When does a jumpstart really begin?  Good question since, as
Darren pointed out, it uses many technologies.  For us it begins
anywhere before the nfs mount of the mini-root; pick a technology, any
technology....  We want to be notifed from the beginning since interns
are the ones that actually start the jumpstarts with 'boot net -
install'.

~JK



Jeff Kennedy wrote:
> 
> We want to be notified via e-mail when a client starts a jumpstart and
> when it finishes.  The finish part is easy; finish script.  But the
> begin part is more difficult since you can't run it in a begin script as
> sendmail doesn't exist yet.  One idea was load the mini-root and use
> it's sendmail to fire one off but the problem is that the mini-root is
> mounted read-only, which means the spooler queue is also read-only,
> hence no mail again.
> 
> Another option was server side notification; write a wrapper for tftp in
> inetd.conf.  Maybe, but someone else suggested a listen daemon that
> listens on the port and captures information.  This would be enough to
> figure out which client was requesting the jumpstart and could be
> multi-purpose, unlike the tftp wrapper.
> 
> Does anyone out there have a mechanism to be notified when a client
> *starts* a jumpstart?  I'm talking at the IP request or sooner, during
> the actual process is too late for us.  The ultimate goal is to know
> when a client starts so we know to expect a notification of completion.
> That way we know for sure the client finished his jumpstart correctly,
> no hiccups.  If we don't get a start notification at the IP level or
> sooner, it's possible that a client could barf before we even know it's
> trying.
> 
> Thanks.

-- 
=====================
Jeff Kennedy
Unix Administrator
AMCC
jlkennedy@amcc.com
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers