Bug 2231273

Hi YoungCheol,

I have looked into the documentation you linked and I am not sure where the Note comes from. As far as I can tell, ML2/OVS backend also does not have ability to detect issues on the external(to openstack) network.
Is this documentation Note the only source you have indicating that ML2/OVS has such capability? I can work with documentation to have that fixed. Thank you

Hi YoungCheol,

Thanks for sharing that KCS, to be honest I did not know that ML2/OVN keepalived had that capability so I looked into it.
So basically what happens in keepalived(used with ML2/OVS) when you set ha_vrrp_health_check_interval to anything > 0 is that it configures keep_alived track_script[1] which is basically a script that keepalived runs and considers its return code into overall healthcheck(it also does some weight and priorty checks but thats not relevant here). If the script returns 0, everything is ok, otherwise something is wrong and then the failover starts.
In our case the script just pings the external network subnets gateway IPs. Here is an example from my 
system

keepalived.conf (irrelevant things are omitted):
vrrp_script ha_health_check_23 {                                                                             
    script "/var/lib/neutron/ha_confs/d9652efa-8031-46f7-948e-8710c73232ad/ha_check_script_23.sh"            
    interval 5                                                                                               
    fall 2                                                                                                   
    rise 2                                                                                                   
}  
... 
    track_script {
        ha_health_check_23
    }

where cat of ha_check_script_23.sh

#!/bin/bash -eu
ip a | grep fe80::f816:3eff:fed4:bf7b || exit 0
ping -c 1 -w 1 10.0.0.1 1>/dev/null || exit 1
ping6 -c 1 -w 1 2620:52:0:13b8::fe 1>/dev/null || exit 1


As far as I can tell this functionality does not exists in ML2/OVN and this is a valid RFE. Looking at this image[2], the OVN BFD alrgorithm only checks connections between nodes (interface 2) but does not check the interface 3. I think this is what the customer wants in ML2/OVN setup. I.e. check if connection to interface 3 is also OK and if not consider the node "not healthy"




[1] https://manpages.debian.org/testing/keepalived/keepalived.conf.5.en.html
[2] https://docs.openstack.org/networking-ovn/latest/admin/routing.html#bfd-monitoring

^ Correction I meant to say " ... I did not know that ML2/OVS keepalived had that capability..."

Hi Miro,

Thank you for your update.

Yes your understanding is correct what customer wants.

Let me know if you need additional info.

Regards,
YoungCheol.