When running under python 3.5, we are seeing the neutron-ns-metadata-proxy fail repeatedly on Ocata RC1 master.
This is causing instances to fail to boot under a python3.5 devstack.
A gate example is here:
http://logs.openstack.org/99/407099/25/check/gate-rally-dsvm-py35-neutron-neutron-ubuntu-xenial/4741e0d/logs/screen-q-l3.txt.gz?level=ERROR#_2017-02-02_11_41_52_029
2017-02-02 11:41:52.029 29906 ERROR neutron.agent.linux.external_process [-] metadata-proxy for router with uuid 79af72b9-6b17-4864-8088-5dc96b9271df not found. The process should not have died
Running this locally I see the debug output of the configuration settings and it immediately exits with no error output.
To reproduce:
Stack a fresh devstack with the "USE_PYTHON3=True" setting in your localrc (NOTE: There are other python3x devstack bugs that may reconfigure your host in bad ways once you do this. Plan to only stack with this setting on a throw away host or one you plan to use for Python3.x going forward)
Once this devstack is up and running, setup a neuron network and subnet, then boot a cirros instance on that new subnet.
Check the cirros console.log to see that it cannot find a metadata datasource (Due to this change disabling configdrive: https://github.com/openstack-dev/devstack/commit/7682ea88a6ab8693b215646f16748dbbc2476cc4).
Check the q-l3.txt log to see the repeated "The process should not have died" messages.
You will also note that the cirros instance did not receive it's ssh keys and is requiring password login due to the missing datasource.
Fix proposed to branch: master
Review: https:/