Hi,

I’ve set up a lab network to test Openstack with a load balanced solution that 
uses keepalived on linux, to provide load balanced & redundant services. I 
installed the ML2 plugin, and removed the security for the ports associated 
with the public and private ports of the load balanced servers with these 
commands:

neutron port-update --no-security-groups  <port UUID>
neutron port-update <port UUID> --port-security-enabled=False

This has allowed VRRP to function correctly, requests are correctly load 
balanced, and I can failover between the servers at will.

The issue I’m having, is when I try to simulate a hard crash of the server that 
is the master in the keepalived pool. If I shut it down in a way that simulates 
a power failure (I’m using ifdown to kill the interfaces, and shutdown in a 
one-liner), the server that was the backup becomes the master, takes over the 
load balanced IP, and starts responding to arp requests for that IP. I can see 
the arp responses with tcpdump on the new master, but these are obviously being 
filtered, as they don’t make it to the other servers.

So where am I going wrong? Why if I do a graceful reboot of the master, would 
the load balanced IP correctly shift to the other server, but if I simulate a 
crash, the arp responses are blocked?

Any help would be greatly appreciated.

Thanks,

Bill
_______________________________________________
Mailing list: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
Post to     : [email protected]
Unsubscribe : http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack

Reply via email to