Why can't I connect to instance via their floating IPs from outside the compute machine?
I have carefully followed the instruction in the URL http://docs.openstack.org/admin-guide... to configure floating ip for openstack instances. I am using havana version and nova.network.manager.FlatDHCPManager as network manager.
After floating ip allocation to an instance, I am able to reach (ping, ssh) the instance's floating ip from the compute node itself. Unfortunately, I cannot access the instance via floating ip, from another server. When I run tcpdump on my public interface, it is clear that requests (e.g. ping requests) get to the interface. Unfortunately, response only occurs when I ping from the compute server itself.
As suggested in the URL http://docs.openstack.org/admin-guide... , I have set sysctl -w net.ipv4.conf.eth2.rp_filter=0. This has not solved the problem.
Setting firewall_driver=nova.virt.firewall.NoopFirewallDriver did not solve the problem either.
My situation is similar to that presented in https://ask.openstack.org/en/question... . Unfortunately, the solution was not properly explained in that thread. It only mentioned that the gateway had to be changed. Which gateway is not clear.
What else do I need to check or do in order to be able to reach the instances via floating ip, from external machines?
Still stuck with inability to access floating ip. Any help?
could you post the output of ip r and ip netns on the host where the floating IP is routed ?
Thanks dachary for the concern. In response to your request, I have added more information to my original question above (i.e. Additional info section). eth2 is the public ip interface. eth1 is the private ip interface.
I even can't ping/access the floating IP from any of the nodes. How did you achieve that?