REQUEST: Network Interface Failover and multi-DNS resolution

[Carlos Rodrigues]

Doncho N. Gunchev wrote:
>     Just put them both in /etc/resolv.conf, or better - fix them to
> know of each other's zones. You can also run local DNS server, which
> you can tell where to look for what, but this is not a workstation's
> function - fix/link your DNS servers/network. At least, you can use
> /etc/hosts as it was in the beginning...

Putting them both in resolv.conf doesn't work, as the second one is 
queried only when the first is down.

Running a local DNS server is just plain ugly (and while I can do it 
just fine, what I'm proposing is automating this stuff).

Fixing the servers to know about each other just can't be done. Each one 
of them is inside his respective LAN, which has NAT to the outside. We 
could put a single server just above them but that would miss the whole 
point of having the two networks separate (they are supposed to be self 
contained).

My best take at solving this is having a flag or configuration option 
somewhere that changes the resolver's "query DNS server only when 
previous is down" behavior to "query DNS server when first gives a 
negative answer". And then have the distro's network scripts adding the 
DNS servers that come from the other interfaces by DHCP added to 
resolv.conf if the second behavior is selected.

> 
>>   2. that there be some failover for network connections.
>>
>>By failover I mean being able to define one interface as "primary". The 
>>primary interface would set the default gateway and all that 
>>global-unique stuff (including resolv.conf, without feature 1.).
>>When that interface goes down, those global settings are changed to the 
>>ones provided by another active interface. If the primary interface goes 
>>up again, it restores the initial configuration.
> 
> 
>     Why not just use DHCP?

Both interfaces use DHCP. But when I unplug the one from which the DNS 
servers and default gateway were obtained, nothing changes. My point is 
that the second interface should take over completely (the required 
settings came through DHCP when it was activated). If the first 
interface comes back, it puts things back to normal.


>>This would be very useful for cases such as a laptop with wired and 
>>wireless networking. The "wired" connection would be the primary 
>>interface. The "wireless" connection would take over if the "wired" one 
>>goes down (they may be different networks, e.g. we have a totally open 
>>and untrusted wireless lan and our linux users can't just unplug the 
>>cable and move around, they are forced to restart the interfaces).
> 
> 
>     You can add 2/more default gateways. In this case linux uses 'Dead
> Gateway Detection' - see http://www.ssi.bg/~ja/dgd-usage.txt for example.
> I've never used two network interfaces with DHCP, so I've never tried
> this trick with DHCP, but you can :)
> ...

Well, I didn't knew that. However my point is Fedora's (and other 
distros) network scripts should do this (or could do this), possibly 
with some configuration options in system-config-network (something like 
  an "interface takeover" checkbox and another on an interface to set it 
as primary). You can't just ask people to go around adding routes and 
tweaking stuff by hand when some of them don't even know what routes 
are... That will only make them complain about how it used to work just 
fine in Windows, and I don't know about you but I just hate to hear that.

Carlos