I was the person with issues with 10's of l3-agents not working properly on Liberty. With about 100 l3-agents we had issues with them just falling over without anything going on. We moved to 20 l3-agents and with that config we've had good luck unless you restart all of them at the same time. My understanding is that in Liberty there is a single thread per neutrons-server (we run 3) that handles agent traffic and that this was supposed to be fixed in Mitaka and newer, but I haven't seen that confirmed.
On Tue, Nov 1, 2016 at 4:38 PM, Leon Zachery (lzachery) <[email protected]> wrote: > Hi: > > At the neutron operator feedback session during the operator summit [1] and > the operator feedback session during the design summit [2], 2 issues were > raised on which > we’d like to get some further clarification. > > The first issue was the issue raised on the scale issues with thousands of > agents, from [1] above: > > Issues with lots of agents (~1000) and rabbitMQ > > We've had problems on liberty with 10's of l3-agents, supposed to be better > on mitaka+ > > Related to this bug: https://bugs.launchpad.net/mos/+bug/1494416 > > > The second issue is the hardware router integration issue, from [2] above: > > Enabling end users to create their own networks serviced by a hardware > router > > Currently using provider networks, but that loses self service > > > > [1] https://etherpad.openstack.org/p/BCN-ops-neutron-pain-points > [2] > https://etherpad.openstack.org/p/ocata-neutron-end-user-operator-feedback > > Feel free to unicast or broadcast a response and I will follow up with you. > > Thanks, > Leon > > > > _______________________________________________ > OpenStack-operators mailing list > [email protected] > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators > _______________________________________________ OpenStack-operators mailing list [email protected] http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
