Hi Lucas! I was looking into this patch while trying to solve my issue, and I had a question. Could you please explain why selection_fields was added to the tests for the load balancer (effectively switching hasing mode from dp_hash to hash)?
My initial assumption was that this might help address the issue where, in stateless mode (with use_stateless_nat), selecting a backend via select() on a per-packet basis can lead to broken sessions when the number of backends changes. I was thinking that using hash instead of dp_hash could make backend selection more consistent with respect to backend count changes—at least in OVS with hash mode, where consistency can be preserved if OpenFlow groups aren’t recreated but only have buckets updated (though I might be misunderstanding this part). Perhaps everything I described above has nothing to do with the reason you added it, but I just wanted to express my assumptions. However, in OVN, backend changes result in the group being deleted and recreated from scratch, so this didn’t seem to be right one idea Thanks for your help! _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
