[Openstack-operators] Best kernel options for openvswitch on network nodes on a large setup

Jean-Philippe Méthot jp.methot at planethoster.info
Mon Oct 1 21:51:55 UTC 2018


So, after some testing, we finally fixed our issue of lost connections to instances. The actual issue was that the ARP table on the network node was hitting its limit constantly and thus, discarding legitimate routes. This caused our connections to flap and the HA routers to switch node without warning. Increasing net.ipv4.neigh.default.gc_thresh1, net.ipv4.neigh.default.gc_thresh2 and net.ipv4.neigh.default.gc_thresh3 kernel values ended up fixing the issue.

Jean-Philippe Méthot
Openstack system administrator
Administrateur système Openstack
PlanetHoster inc.




> Le 28 sept. 2018 à 10:53, Jean-Philippe Méthot <jp.methot at planethoster.info> a écrit :
> 
> Thank you, I will try it next week (since today is Friday) and update this thread if it has fixed my issues. We are indeed using the latest RDO Pike, so ovsdbapp 0.4.3.1 . 
> 
> Jean-Philippe Méthot
> Openstack system administrator
> Administrateur système Openstack
> PlanetHoster inc.
> 
> 
> 
> 
>> Le 28 sept. 2018 à 03:03, Slawomir Kaplonski <skaplons at redhat.com <mailto:skaplons at redhat.com>> a écrit :
>> 
>> Hi,
>> 
>> What version of Neutron and ovsdbapp You are using? IIRC there was such issue somewhere around Pike version, we saw it in functional tests quite often. But later with new ovsdbapp version I think that this problem was somehow solved.
>> Maybe try newer version of ovsdbapp and check if it will be better.
> 
> _______________________________________________
> OpenStack-operators mailing list
> OpenStack-operators at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20181001/675248bc/attachment.html>


More information about the OpenStack-operators mailing list