On Fri, 12 Feb 2010, Dave Miner wrote: > As you may recall, we reviewed a first draft of this design in December/early > January. I've finally published a revision which addresses the comments I > received, primarily from Jan and Ethan. > > While not exactly frozen, implementation work is underway based on this > design so any remaining issues should be raised soon to allow the best chance > to address them. > > http://hub.opensolaris.org/bin/view/Project+caiman/AI+Image+Management
This looks pretty good. I just have a few questions - - With this proposal, pxegrub is going to start fetching the menu.lst from the path specified for pxegrub itself, instead of a /tftpboot location. Does this change the user interaction at all (on the server side) wrt tweaking the menu.lst for a service in order to, say, accomplish hands free automated install? - It does seem that the system.conf proposed for SPARC would work quite nicely even on x86. It would certainly simplify our boot process significantly -- currently we reference install.conf for SPARC and boot arguments for x86. The other advantage of a system.conf for x86 would be that it would eliminate the need to reboot a system to start an automated installation with service discovery; if the system initially booted to console-login. The only iffy part would seem to be the use of auto-expanding GRUB keywords (such as the proposed $CONSOLE) and not being expanded correctly if conveyed via the .conf file. However, there's nothing preventing such keywords from being passed via the menu.lst file instead. - Example 2, last line. Won't you need "boot net:dhcp - install" to trigger an install given some of the recent changes Ethan made or is this proposal reverting that behavior? Alok
