DNS anycast node monitor
rkoberman at gmail.com
Mon Apr 20 05:23:34 UTC 2015
On Thu, Apr 16, 2015 at 2:03 PM, Bob Harold <rharolde at umich.edu> wrote:
> We use ExaBGP to query the DNS process locally and advertise/withdraw
> We use BFD between the server and the router to detect link failure (or
> server crash):
> The BFD (Bidirectional Forwarding Detection) protocol is designed to
> rapid detection of communication failures between two devices. Routing
> protocols can use BFD state to rapidly (< 1 second) withdraw routes in the
> event of a failure, without having to wait for a routing protocol timeout
> (3 minutes by default for BGP).
> Seems to work well.
> Bob Harold
But that only protects from the system going down, not the DNS server
failing in some manner. We used a monitor that actually made test queries
to the server and then used bgpd to pull the anycast address down after N
consecutive failures or more than M of the last K queries. I no longer have
access to the trivial script since I retired.
It's really harder than it looks to do right and I don't think my code was
adequately rigorous, but was capable of responding to most issues. I'm sure
more heuristics really needed to be added.
Kevin Oberman, Network Engineer, Retired
E-mail: rkoberman at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the bind-users