Hi Maciej!
Thanks for reporting this !
Yes, we already have a problem with a "sticky" UDP socket for DNS
resolution in k8s environment with ephemeral IPs for CoreDNS nodes.
You can see more details from this GitHub issue:
https://github.com/haproxy/haproxy/issues/2654
Maybe the udp socket should be recreated if any error (including dns
timeout) occurs?
At a glance it seems to be a good idea. In order to address this
correctly, could you, please, open a feature request at GitHub (type
feature). Like this we could discuss with others, if from the
architecture point of view it is possible and there are no other drawbacks.
Kind regards,
--
Bien cordialement,
Valentine
