Until we've according DHCP/DNS/VM infrastructure that's the easiest
solution...
Thanks for the template, Richard.
From: Michael Prokop <mprokop@sipwise.com>
This is hackish™ and we should figure out a way how to get
rid of this kill-foo in the long run.
Actually the installation still works but if you care about the
error messages or want to debug the resulting system in *live*
mode then it's annoying if you're not aware of what to do...
From: Michael Prokop <mprokop@sipwise.com>
We want to have /etc/fstab being up2date on the live
system after installation, so force processing the
according uevents between kernel and userspace...
From: Michael Prokop <mprokop@sipwise.com>
Booting with Debian installation and bootoption puppet=192.168.51.25
sets up the host for puppet node deployment.
From: Michael Prokop <mprokop@sipwise.com>
For whatever reason the swap partition isn't enabled by default
anymore. Used to work in all my previous test runs (which
confused me already - haha), I've no idea why it's failing now
again. So re-add the mkswap command so we can at least release
2.4.1 now. Meh… ☠
From: Michael Prokop <mprokop@sipwise.com>
parted enables the swap partition during partition creation time
already, so we'd have to rewrite the partition stuff as well.
Turns out the difference between VM sizes (no swap vs. swap) is just
about 30MB, so let's drop the efforts regarding removal-but-enabling-swap-afterwards
and instead we get a even more homogeneous solution. :)
From: Michael Prokop <mprokop@sipwise.com>
As reported by Andrew CE systems tend to run in OOM conditions
pretty soon sometimes. So lets enable a swap partition in all
our systems by default.
To avoid a bigger VM image we create an init script on CE systems
which enables the swap partition during bootup and then destroys
itself (the script, not the swap partition :)).
From: Michael Prokop <mprokop@sipwise.com>
bootoption ngcprelease is too confusing, so drop it.
instead ngcpautobuildrelease does exactly what it says:
ngcpautobuildrelease=2.4-rc0
will skip generation of sources.list inside ngcp-installer and instead
use the defaults, except for the release it will use:
deb http://deb.sipwise.com/autobuild/ release-2.4-rc0 main
so we can test release candidates of upcoming stable releases without
manual intervention or adjusting ngcp-installer for each single test
release.
When installing a trunk release we also take the most recent version
of the installer, the installer will use the sources.list.trunk file
then anway. \o/
From: Michael Prokop <mprokop@sipwise.com>
we need a decent way to test unfinished releases during the release
cycle, using ngcprelease=2.4-rc1 allows us to test the installer
without any manual intervention against the 2.4-rc1 release on jenkins
From: Michael Prokop <mprokop@sipwise.com>
Building the offline mirror on request is slow and we don't
use it anymore since we provide the mirror on the ISO itself,
so instead test for /srv/mirror/debian only.
While at it don't hardcode "squeeze" in the mirror test but
instead use ${DEBIAN_RELEASE}.
From: Michael Prokop <mprokop@sipwise.com>
Instead of providing the installation details to the screen and
sleep a certain amount let's log the data to
/tmp/installer-settings.txt instead.
While at it also make sure bootoption 'ngcpce' is used so it's
available in the logo screen as well as inside the script.
From: Michael Prokop <mprokop@sipwise.com>
When deploying inside Kantan we want to be able to skip certain
stuff while providing data to the outside host. The bootoption
'kantan' provides that option and since we want to be able to
halt the system after installation also support the 'ngcphalt'
bootoption.
From: Michael Prokop <mprokop@sipwise.com>
Using bootoption "ngcpvers=2.1" should install our platform
version 2.1 and needs to automatically be aware of installing
Debian/lenny.
I'm not entirely happy with hardcoding release vs. platform vs.
installer version, but that's the best we can do for now.
From: Michael Prokop <mprokop@sipwise.com>
We are providing it (and depend on it) on our pro installations
already, but if we're deploying a CE edition on a physical
server system and network doesn't work after deploying it's
just a waste of time. So just install it on all our own systems
by default.
Ack-ed by: Daniel
From: Michael Prokop <mprokop@sipwise.com>
define default values at beginning of file
move setting hostname to default after cmdline/profile scanning
load "ip=<client-ip>:<srv-ip>: ..." in array variable
re-orded and additional commandline output
remove more hardcoded parameters (devs, IPs, hostnames)
full profile for sppro/spce
Otherwise we're creating broken systems which don't work as supposed.
And while at it do not append variables to files but instead override
any existing values, otherwise we might have broken config files.
From: Michael Prokop <mprokop@sipwise.com>
With the current bridge setup for installing pro systems on Proxmox
we have to use ip=....:eth1:off for getting into the internet[tm].
Since the internal network device needs to be eth0 then we've
to check for the specificed ip=... device and use the other[tm] one
for configuration.
From: Michael Prokop <mprokop@sipwise.com>
Michael Prokop
Andrew Pogrebennyk
Erhard Rank