[c-nsp] Anycast Questions

[Phil Mayers]

On 16/02/11 13:39, Gert Doering wrote:
> Hi,
>
> On Wed, Feb 16, 2011 at 09:04:49AM +0000, Phil Mayers wrote:
>>> Which makes "monitoring whether everything is fine" a somewhat more
>>> interesting challenge :-) - if the machine is up, but bgpd fails, the
>>> service might silently fall over to another instance and things like
>>> "does this anycasted DNS server still respond?" will happily report
>>> success...
>>
>> SNMP to pool the BGP neighbour status?
>
> If that would work for IPv6 BGP peers...

Ah.

>
> We're mostly not relying on SNMP to monitor things, but on ping and
> "service queries" (DNS probes, HTTP checks, ...) to the actual services,
> interface IPs, etc.

We do that too.

I did wonder at one point about using GRE tunnels and Linux policy 
routing so that I could "select" which backend of a virtual IP I was 
checking against.