> Mike Gerdts wrote: > > On 9/13/07, Danek Duvall <danek.duvall at sun.com> > wrote: > In most cases, these sorts of actions can be > postponed until either the > service is started (for a simple case) or the machine > is rebooted. Very > few third-party packages should need to perform > unique actions before > the machine reboots. > > The problem is that getting scripting correct is > difficult enough when > one is talking about running in a normal context. > Getting the scripting > ight when we're modifying a local zone of a diskless > client on an > alternate architecture down-reved boot server is > downright hard since > so much of the installation environment affects the > script... we want to > defer these sorts of actions until the service > starts.
It would seem to me that scripting during a service restart is no better than a package (and possibly worse). Once you've run the script you're in an undefined state and if a package backout occurs then you're in an even more undefined state as there's no way of triggering script backout. If you snapshot just the filesystems supposedly touched by the package, then you're ok as long as the script doesn't go outside those filesystems as you can do a zfs rollback and reboot. If your script goes out of the snapshotted area or touches a database, you're hosed. I have no solutions here, just thinking out loud.... Paul This message posted from opensolaris.org
