On Thu, Mar 13, 2025 at 9:33 AM Xavier Simonart <xsimo...@redhat.com> wrote:
> Avoid using two similar load-balancers such as: > ovn-nbctl lb-add lb4-no-aff [[fd11::a]]:8081 [[fd11::2]]:80 > ovn-nbctl lb-add lb40-no-aff [[fd11::a]]:8081 [[fd11::2]]:80 > > Using such load balancers on the same switch would result in similar > openflows > in table OFTABLE_CHK_LB_HAIRPIN, containing learn actions, and differing > only > by the cookie. One of the two flows will be installed in OVS by OVN (e.g. > the flow > with the cookie from lb4-no-aff). > Any flows learned by that learn action will be deleted by OVS when the > corresponding flow in table OFTABLE_CHK_LB_HAIRPIN is deleted. > > In case of recompute, there is no guarantee that the same openflow is > installed: > the flow with the cookie from lb40-no-aff might now be installed. This > would result > in the deletion of the learned flow in OVS. > > If one of the two similar load balancers is deleted, the same behaviour > might happen: > the learned flow in table OFTABLE_CHK_LB_HAIRPIN_REPLY might be deleted, > depending > of which load balancer was preferred. > > Describing the behaviour so that, if one considers that this is a bug and > that mutiple > similar load balancers can be added and result in no flow deletion during > recompute, > a JIRA should be entered. > > THis wa identified by comparing flows from IP and recompute and frequent > failures of test > "load balancing affinity sessions - IPv6". > nit: s/THis/This/ > > Signed-off-by: Xavier Simonart <xsimo...@redhat.com> > --- > tests/system-ovn-kmod.at | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/tests/system-ovn-kmod.at b/tests/system-ovn-kmod.at > index 17607fbf2..5b010a1a2 100644 > --- a/tests/system-ovn-kmod.at > +++ b/tests/system-ovn-kmod.at > @@ -550,9 +550,9 @@ NS_CHECK_EXEC([foo1], [nc -z fd12::b 8081]) > > NS_CHECK_EXEC([foo1], [ip -6 neigh add fd11::b lladdr 00:00:01:01:02:03 > dev foo1], [0]) > check ovn-nbctl --wait=sb lb-add lb4 [[fd11::a]]:8080 [[fd11::2]]:80 > -check ovn-nbctl --wait=sb lb-add lb40 [[fd11::a]]:8080 [[fd11::2]]:80 > +check ovn-nbctl --wait=sb lb-add lb40 [[fd11::c]]:8080 [[fd11::2]]:80 > check ovn-nbctl --wait=sb lb-add lb4-no-aff [[fd11::a]]:8081 > [[fd11::2]]:80 > -check ovn-nbctl --wait=sb lb-add lb40-no-aff [[fd11::a]]:8081 > [[fd11::2]]:80 > +check ovn-nbctl --wait=sb lb-add lb40-no-aff [[fd11::c]]:8081 > [[fd11::2]]:80 > check ovn-nbctl --wait=sb set load_balancer lb4 > options:affinity_timeout=60 options:hairpin_snat_ip="fd11::b" > check ovn-nbctl --wait=sb set load_balancer lb40 > options:affinity_timeout=60 options:hairpin_snat_ip="fd11::b" > check ovn-nbctl ls-lb-add foo lb4 > -- > 2.47.1 > > _______________________________________________ > dev mailing list > d...@openvswitch.org > https://mail.openvswitch.org/mailman/listinfo/ovs-dev > > Otherwise looks good to me, thanks. Acked-by: Ales Musil <amu...@redhat.com> _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
