[CentOS] automated remote cloning

Mon Jan 3 03:19:38 UTC 2011
Nico Kadel-Garcia <nkadel at gmail.com>

On Sun, Jan 2, 2011 at 11:55 AM, Les Mikesell <lesmikesell at gmail.com> wrote:
> On 1/1/11 5:50 AM, Geoff Galitz wrote:
>> Good morning/day and Happy New Year.
>> We have a geographically distributed environment (marketing speak: cloud) where
>> we regularly need to migrate individual systems to new hardware (for bigger
>> disks or for better geographical placement, for example). We currently use
>> Cobbler to do our base installs automatically and I am now looking at
>> integrating Clonezilla with Cobbler. The goal is take an *individual* system
>> that has been customized and migrate it in an automated fashion.
>> We currently do this using Cobbler and then running rsync and mysqldump in a
>> script along with other system -> userland module configurations (such as PHP
>> modules) but the process is tedious and generally a PITA. We also must have the
>> ability to run this across different datacenters on different continents.
>> Any pointers to good automated solutions?
>> References:
>> https://fedorahosted.org/cobbler/wiki/ClonezillaIntegration
> Clonezilla is probably the only thing that doesn't lock you into a single
> OS/distribution.  It works as long as the hardware is similar and the target
> disk is as large or larger than the source, but unless you use DHCP for IP
> addressing on all NICs you can expect problems with losing your static configs
> when cloning.   I don't think it would be efficient to clone a large number of
> individualized copies remotely from a central location though. You'd probably
> want a generic image you could cache locally, followed by a scripted update of
> the differences.   Clonezilla also does not have a handy way to clone an
> existing machine to one with larger disks letting you specify new
> arbitrarily-sized partitions (although it can resize them all proportionally).
> --
>   Les Mikesell
>    lesmikesell at gmail.com

That's what the post-install wrapper scripts are for, and definitely
why enterprise class setups need DHCP reservation based setups, so
that re-installs automatically get the right network setups.

CentOS and RHEL 5 DHCP do have a flaw. Their DHCP is so old that it
does not follow the "domain search order" directives of DHCP RFC's
that already existed when they were published, but weren't in the main
DHCP codeline. So if you have "example.com", "mail.example.com", and
"test.example.com" in your local searchable domains, you won't get it
in your resolv.conf unless you manually insert the entirely
undocumented "SEARCH=" settings in /etc/sysconfig/network. Neither
system-config-network nor NetworkManager get this right.

And yes, I reported this as a feature request for documentation of
this DHCP configuration requirement on my RHEL systems.