On Wed, Jul 2, 2008 at 7:46 PM, Jordan Brown <Jordan.Brown at sun.com> wrote: > As I think I mentioned, I need a run-once mechanism. > > As it worked out, the mechanism was not far from being pretty general. Take > a look... > > The only part of it that is visible to the application that wants the "run > once" service is application-xvm-first-boot.xml. When an application wants > something run once, it creates a manifest like that. Everything in it is > boilerplate except for the instance-name (which must be unique) and the > value of $RUN, which specifies the command to execute. (The > boilerplate-to-meat ratio is unfortunate, but better than it might be.) > > runonce.xml is the framework of the service. There's not a lot more there, > but there's a bit. > > runonce is the script that does the dirty work. It runs the target program. > If the target command succeeds, runonce disables the instance. If the > target command fails then it records the failure and takes the instance into > "maintenance". > > Note that runonce could be changed to delete the instance, or whatever. > > Any thoughts?
Sometimes a runonce service needs to trigger a reboot. It looks to me as though all runonce services will be running at the same time - making such a thing somewhat risky. Is this something that you are intending to support? > #! /bin/sh > # > # Copyright 2008 Sun Microsystems, Inc. All rights reserved. > # Use is subject to license terms. > # > > # runonce > # Runs command specified as instance name. > # If it fails, returns SMF failure. > # If it succeeds, disables this SMF service and returns success. > > . /lib/svc/share/smf_include.sh > > sh -c "$RUN" One of the things that I commonly need to run on first boot is an ELF executable. Is there a reason to force a Bourne shell wrapper? As currently implemented it could cause confusion for software targeting OpenSolaris and Solaris because /bin/sh in OpenSolaris 2008.05 is really ksh93. This would mean that a Korn shell script tested on OpenSolaris will work just fine but will fail on Solaris 10. Overall I am enthusiastic about having something like this available. It is a common need and ad-hoc implementations tend to be too much effort to implement reliably. -- Mike Gerdts http://mgerdts.blogspot.com/
