We created experimental Neutron DVR+HA multinode job with [1].
DVR+HA needs a three node setup with two dvr_snat nodes
and one dvr node. So in this setup, we set
1) controller node's agent_mode to dvr_snat
2) first compute to dvr and other to dvr_snat
[1] https://review.openstack.org/#/c/383742/
Change-Id: I4e08a97158dc4538c6f021c049c6e60fb7293459
We switched from GRE to VXLAN in the implementation as of
I951e7e1bd98eeaf347e85296c12e027e225c2c97 so the diagrams should
match accordingly.
Change-Id: I0b0cef3a7217530573e694de54bf6dc68e295856
Provide more info on why the mutlinode network setup exists as it does
today. This should provide better background on why certain choices were
made and general info on how it all works.
Change-Id: Ie04867a95f7e340dd3c1253aef0591bbf75bab8a
venkata anil