[gentoo-catalyst] server maintenance w/ catalyst

[William Kilian]

Hello,

I have a lot of questions about using catalyst to automate server 
maintenance. Here's the scenario:

I administer several servers and want to be able to perform upgrades for 
closing security bugs and adding necessary features without waiting for 
compiles. I want the compiles to start at midnight or 4am, and be ready 
for completed for me to accept or reject the newly built packages based 
on my test suites.

All the servers run basically the same software so that administration 
is consistent across machines, but hardware varies from pentium mmx's to 
dual xeons. So apparently I should build a generic x86 stage1 or just 
use a precompiled one, then build custom stage2/3s from it with spec 
files that only differ by their CFLAGS, etc. I can build those on one 
machine (I have a nice dual-xeon ready and waiting), then perform the 
installs by setting up my fs's, booting to a floppy and wget -O - 
http://build-box/stage3.subarch.tar.bz2 | tar -xjPps -C /    ... or 
something like that. But then I suppose I should use tinderboxes to 
build upgrade binary packages so that I can install binary packages on 
the slow machines and do all my building in one, very secure, place.

That's the scenario. Questions:

Where do I put all my custom use flags?
I specify use flags on a per-package basis most of the time, in 
/etc/portage/packages.use. I also have to use custom masks, etc. to 
install certain packages that I need. I can't tell when or how I'm 
supposed to specify all those customizations. As far as I can tell, only 
stuff that will eventually go in /etc/make.conf belongs in a catalyst 
spec file. If I should just add the /etc/portage files after the stage1 
is built, when and where do I put add them? In the stage1 image before 
doing the stage2?

Should I really build my own stage1 or just use stage1s from gentoo mirrors?

What's the best way to perform upgrades and incorporate them into my 
stage3 images?
Obviously I will want to build upgrade packages using tinderboxes with 
package caching. I would use a separate package cache for each subarch. 
Can I do an emerge -u style upgrade to a package in a tinderbox if the 
stage3 for the tinderbox already has an old version of the package 
installed? Will I need to uninstall the old version from the stage3 
first? Would it be better to simply use package-caching and build a 
whole new series of stages 1, 2, and 3 if I'm upgrading packages, then 
install the binary packages on existing systems?

What's the best way to automate custom configurations of system files 
(e.g. /etc/profile)?
Should I build my own ebuild or customize existing ebuilds using a 
/usr/local/portage overlay?

I have more questions about using catalyst to automate server 
maintenance as much as possible, but answers to some of the above 
questions would help a lot.

Thanks,

William Kilian

--
gentoo-catalyst@g.o mailing list