Enhancement: use static IP, not DHCP

[mykaul]

Why are we using DHCP if we are setting the IP anyway? Why not just use static IP? Surely it'll make libvirt's network configuration smaller (and thus faster) and I expect the hosts to boot faster too).

Requires a small change in the mapping from mac:ip to mac:{ip,gw} (assuming the subnet is /24) and accordingly in the jinja template and then some code removal not to configure the libvirt's DHCP service.

(Not sure about IPv6, I assume need to be added somewhere too?)

[ifireball]

AFAIK you will still need 'dnsmasq' to provide DNS, so the savings as far as runtime resources go will probably be negligible. And you will create a new problem of having to know how to inject a static IP address reliably for any kind of OS any Lago user may decide to boot up. Libgustfs makes this look easy, and cloud-init may also help, but the potential for issues is far bigger IMO then just supporting DHCP.

[mykaul]

1. I don't care much about any OS - as we could change EL to use it. The rest could continue to use DHCP - it's in the Jinja template, no? Note that as we are already writing the ifcfg file, it's a matter of writing it with static and not DHCP.
2. Perhaps correct about setting the DNS - it may still take 5-6 secs still per network to come up, but I expect hosts to come up faster.

[nvgoldin]

I can try running a simple test and look at 'systemd-analyze' output in the CI to see if that + 'lago start' time is better. If we agree this does improve boot up time - I suggest adding a 'static' network type, which will do what @mykaul suggested(allocate IPs & inject), and if we don't have a good way of injecting it to a specific distro, we can fail early on 'init'. The Jinja'ing part should be simple.