[ILUG] Xen and failover

[Niall O Broin]

After being a long time fence sitter, I've at last started to use Xen  
and it's every bit as nifty as the many users I've spoken to say it  
is (including some inhabitants of this parish who use it in  
production on some quite big systems).

I'm now looking at deploying some production servers using Xen, and  
one major reason for this is the ability to fail over to a Xen  
instance on a different box.

In the case in point, failover will not be seamless, as the live and  
standby boxes will be in different data centres, but we have no PI  
space, hence we will be failing over to different IPs. From the POV  
of service consumers, this will be handled by relatively short DNS  
timeouts (while failure of this application is painful, nobody's  
going to die, so the short timeouts method is acceptable).

My question now is how to handle this from the POV of the service  
provider, specifically the change of IP. As I see it, we have two  
choices.

1) Manipulate the LVM disk on the standby server after is it has been  
updated to give it the correct IP.

2) Run the domUs in rfc1918 space, and use iptables on the dom0s to  
route packets as needed.

1) has the advantage of simplifying the dom0, whereas 2) has the  
advantage of simplifying the sync process.

Opinions? And especially, is there a 3) (or even a 4) ) that hasn't  
occurred to me?



Niall