Not sure if this is a bug, but I have found in testing the Virtual Routers dump 
their arp cash almost every 5-10 seconds. 

This makes them loose connectivity if they live behind a perimeter firewall 
cluster running floating IPs for gateway addresses (for example: Conntrack, 
CRM/Pacemaker, VRRP).  Any vm using a shared network connection to the floating 
gateway has no issues, but isolated networks requiring the VR to be an initial 
gateway will loose connectivity as the Debian VR aggressively flushes its ARP 
cache. 

You can even watch connections stop then start when the gateway cluster sends 
its unsolicited ARP broadcasts, and then within a few seconds, stop again.

Can we look into this?

My setup is below

Gateway cluster running: 
4 floating IPs on eth4 (Cloud public network gateways, 1 per zone, all VRs 
point to these)
36 floating IPs on eth0 (SNAT addresses into cloud)
2 floating IPs on eth1 (Management Gateways)
HB on eth2 (Cluster heartbeat)

*No issues connecting VMs directly to the gateway cluster with shared 
networks(VLAN)

*connection issues when using isolated networks routed through VR to gateway 
cluster

-kelcey

Sent from my iPhone


Reply via email to