reproducer for 3node incorrectly overrides the role configuration

[Alex Schultz]

I recently tried to use the quickstart reproducer script[0] for the tripleo-ci-centos-7-3nodes-multinode job. The script executed the deployment but failed on tempest with:

TASK [validate-tempest : Exit with tempest result code if configured] ***********************************************************************************************************************************************
Thursday 22 March 2018 19:43:09 +0000 (0:00:02.526) 0:01:33.783 ********
fatal: [undercloud]: FAILED! => {"changed": true, "cmd": "tail -10 tempest_output.log; exit 1", "delta": "0:00:00.007400", "end": "2018-03-22 19:43:11.806583", "failed": true, "msg": "non-zero return code", "rc":
1, "start": "2018-03-22 19:43:11.799183", "stderr": "", "stderr_lines": [], "stdout": "2018-03-22 19:42:14 | Failed to discover available identity versions when contacting http://192.168.24.15:5000//v3. Attempting
 to parse version from URL.\n2018-03-22 19:42:14 | Could not determine a suitable URL for the plugin\n2018-03-22 19:42:14 | ++ neutron net-list -c id -f value\n2018-03-22 19:42:15 | neutron CLI is deprecated and w
ill be removed in the future. Use openstack CLI instead.\n2018-03-22 19:42:17 | Failed to discover available identity versions when contacting http://192.168.24.15:5000//v3. Attempting to parse version from URL.\n
2018-03-22 19:42:17 | Could not determine a suitable URL for the plugin\n2018-03-22 19:42:17 | + neutron net-create public --router:external=True --provider:network_type flat --provider:physical_network datacentre
\n2018-03-22 19:42:18 | neutron CLI is deprecated and will be removed in the future. Use openstack CLI instead.\n2018-03-22 19:42:20 | Failed to discover available identity versions when contacting http://192.168.
24.15:5000//v3. Attempting to parse version from URL.\n2018-03-22 19:42:20 | Could not determine a suitable URL for the plugin", "stdout_lines": ["2018-03-22 19:42:14 | Failed to discover available identity versio
ns when contacting http://192.168.24.15:5000//v3. Attempting to parse version from URL.", "2018-03-22 19:42:14 | Could not determine a suitable URL for the plugin", "2018-03-22 19:42:14 | ++ neutron net-list -c id
 -f value", "2018-03-22 19:42:15 | neutron CLI is deprecated and will be removed in the future. Use openstack CLI instead.", "2018-03-22 19:42:17 | Failed to discover available identity versions when contacting ht
tp://192.168.24.15:5000//v3. Attempting to parse version from URL.", "2018-03-22 19:42:17 | Could not determine a suitable URL for the plugin", "2018-03-22 19:42:17 | + neutron net-create public --router:external=
True --provider:network_type flat --provider:physical_network datacentre", "2018-03-22 19:42:18 | neutron CLI is deprecated and will be removed in the future. Use openstack CLI instead.", "2018-03-22 19:42:20 | Fa
iled to discover available identity versions when contacting http://192.168.24.15:5000//v3. Attempting to parse version from URL.", "2018-03-22 19:42:20 | Could not determine a suitable URL for the plugin"]}

The deployment runs but basically only on a single node and deploys no services. It turns out that the deployment command on the reproducer script has '--roles-file /home/zuul/overcloud_roles.yaml' where the one that runs in the CI does not as it inherits the roles from '-r /usr/share/openstack-tripleo-heat-templates/ci/environments/multinode-3nodes.yaml' in fs032. In reseaching this, it appears that the composable_roles ansible configuration is being overridden somewhere in the configuration of the reproducer. The fs032 has composable_roles set to false.

I traced it back to the inclusion of '/opt/stack/new/tripleo-ci/toci-quickstart/config/testenv/multinode-rdocloud.yml' for the reproducer playbook call[1] which has composable_roles set to 'true'. This should not be defined in the testenv but rather defined at the featureset level.

[0] http://logs.openstack.org/56/555056/1/gate/tripleo-ci-centos-7-3nodes-multinode/ea8834f/logs/reproducer-quickstart.sh
[1] http://paste.openstack.org/show/709027/

[Alex Schultz (alex-schultz)]

This may be able to be worked around by appending '--extra-vars composable_roles=false' in the EXTRA_VARS of the reproducer-quickstart.sh

Example:
export EXTRA_VARS="\$EXTRA_VARS --extra-vars dlrn_hash_tag=35414701c176a6288fc2ad141dad0f73624dcb94_43527485 --extra-vars composable_roles=false"