Ports of same physical network on the same hypervisor cannot talk to each other

[Han Zhou]

The model we have now for physical network connection is: for each neutron port, create a dedicated lswitch connect to external bridge. This model works well for inter-hypervisor traffic, but not for intra-hypervisor, since packets cannot go back to the inport.

It seems to me the right model should be put the ports of same neutron network to same lswitch (datapath), and connect to physcal bridge through same patch port.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/266186

[Russell Bryant (russellb)]

relevant thread starts here: http://openvswitch.org/pipermail/dev/2016-January/064413.html

[Russell Bryant (russellb)]

I posted a fix: http://openvswitch.org/pipermail/dev/2016-January/064559.html

[Han Zhou (zhouhan)]

Russell, thanks for your quick work! I will review the fix.

[Han Zhou (zhouhan)]

Update:

Russell provided a fix [1] in OVN based on existing model, which is waiting to be merged.

Based on Russell's fix I posted a follow-up patch [2] with simplified model, pending for review.

[1] http://patchwork.ozlabs.org/bundle/russellb/localnet/

[2] http://patchwork.ozlabs.org/patch/576145/

[Han Zhou (zhouhan)]

Update:

Russell provided a fix [1] in OVN based on existing model, which already got merged. So will close this bug first.

Based on Russell's fix I posted a follow-up patch [2] with simplified model, under review. Will open a new one to track the model change.

[1] http://patchwork.ozlabs.org/bundle/russellb/localnet/

[2] http://patchwork.ozlabs.org/patch/578656/