lbaas is unreliable in the gate

[Sean Dague]

the lbaas service is quite unreliable in the gate, and causing a large number of gate resets.

http://logs.openstack.org/78/81778/1/check/check-tempest-dsvm-neutron/0a85d7c/console.html#_2014-03-20_13_43_58_324

TRACE & ERROR logs for this run here

http://logs.openstack.org/78/81778/1/check/check-tempest-dsvm-neutron/0a85d7c/logs/screen-q-lbaas.txt.gz?level=TRACE

regardless, the fact that lbaas is making lots of blind network namespace calls that are failing (and exploding) seems bad.

[Aaron Rosen (arosen)]

 I'm curious why this is done: https://github.com/openstack/neutron/blob/master/neutron/services/loadbalancer/drivers/common/agent_driver_base.py#L98 : Seems status is is PENDING_UPDATE when this is called. Raising here causes this failure.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/81831
Committed: https://git.openstack.org/cgit/openstack/tempest/commit/?id=c4aee3528ab9b2e7d29a163014292be6a08b31aa
Submitter: Jenkins
Branch: master

commit c4aee3528ab9b2e7d29a163014292be6a08b31aa
Author: Eugene Nikanorov <email address hidden>
Date: Thu Mar 20 18:17:00 2014 +0400

    Skip loadbalancer basic scenario test

    Change-Id: I5feeddba784d7ca568485b7c03a17883eff6cef9
    Partial-Bug: #1295165

[Eugene Nikanorov (enikanorov)]

The thing is that API tests for loadbalancer may cause different stack traces on the server side because API tests don't care about consistency on the agent or backend. And that is fine.

The real issue is with scenario test, which has some real connectivity problem test_client->floatingip->vip ip->haproxy

[Eugene Nikanorov (enikanorov)]

fixed by https://review.openstack.org/#/c/81537/ on Neutron side

[Attila Fazekas (afazekas)]

https://review.openstack.org/#/c/84542/ The fix is not confirmed.