> >This assumes that there is a common pool of IP addresses available - >which assumes things like BGP are in place and routing packets between >multiple datacenters/providers and that you are interacting with BGP >properly. > >I'll be frank and say that the complexity and interaction with systems >that are outside of our control scares me.
David, yes this is a valid concern. So, initially I was planning to leverage the ADC like NetScaler's routing capabilities to advertise routes. When IP is transferred from zone to another zone, CloudStack will orchestrate the route advertisements. But as you reasoned, this is not the best way to go. So what I am proposing is that, let CloudStack raise the trigger (raise action event for eg.) when IP is transferred, on which Admin/external tools can act up on. > >What exactly are we getting here that we couldn't obtain with things >like having folks manage DNS much better, as I fear there are many >dragons along this path. Good question. This is purely in-practice AWS EIP use case. With DNS re-mapping there is huge failover recovery time (propagation of new DNS mapping, client cache etc) in reflecting the DNS name to new public IP. What is happening in this case is DNS name, public IP remain static, it just the back end server that changes. I donĀ¹t see testing is big concern. In some sense CloudStack is dumb in this context all its doing is just configuring NAT rule, intelligence to advertise the IP or out side of CloudStack. > >How would you propose to test such a beast? > >--David >