Connectivity_tests task fails with bonded admin interface in balance-rr mode

[Alexander Zatserklyany]

Deploy cluster with bonded admin interface in balance-rr mode and Neutron VLAN fails with "Error running RPC method granular_deploy"

Steps to reproduce
------------------
1. Create cluster with Neutron VLAN network provider
2. Add 3 nodes with controller and mongo roles
3. Add a node with compute and cinder roles
4. Setup bonding for admin and management interfaces in balance-rr mode; repeat for all nodes
5. Run network verification
6. Deploy the changes

Expected results:
All nodes are in 'Ready' status

Actual results:
Deploy failed.
See log http://paste.openstack.org/show/485258/

[slava valyavskiy (slava-val-al)]

It's really hard to figure out problem without puppet log from nodes. Could you please provide it?

[Alexander Zatserklyany (zatserklyany)]

Additional information:
If step 4
"4. Setup bonding for admin and management interfaces in balance-rr mode; repeat for all nodes"
change to
"4. Setup bonding for admin and management interfaces in active-backup mode; repeat for all nodes"
the cluster will be deployed and OSTF pass

[Stanislaw Bogatkin (sbogatkin)]

Seems like duplicate of https://bugs.launchpad.net/fuel/+bug/1493290

[Stanislaw Bogatkin (sbogatkin)]

Hi Alex.
There are several questions I have to you:
1. Did you setup vlan for management network on nodes? I asked because my configuration just not allow bond tagged and untagged networks.
2. Could you, please, attach astute.yaml from primary controller?
3. Could you, please, tell about your end bonding scheme? Which network cards you have attached to which bridges and where network traffic from management interface goes?
4. Could you provide credentials to failed env to look on it?

[Matthew Mosesohn (raytrac3r)]

This is not started yet. Moving to 8.0 updates

[Dmitriy Novakovskiy (dnovakovskiy)]

The root cause + user impact is the following: if user configures NIC with balance-rr bonding mode (typical for scenarios with SDN controller) - the deployment fails. If user does LACP bonding mode (typical for scenarios with non-SDN networking) - all's fine. So the ultimate impact with - until this bug is fixed we can't deploy Fuel+SDN in Production cases (where bonding with round robbin is needed).

We have communicated with a number of teams and explained the impact to them, unless we hear about critical installation of Fuel+Contrail/NSX that is planned for 1st month after GA - this bug can be moved to MU1 (not 9.0).