tripleo-quickstarts undercloud deployments fails due to neutron db sync

[Luke Short]

Description
===========
Using tripleo-quickstart, the Undercloud fails to deploy because the Neutron MySQL tables are not created properly.

Steps to reproduce
==================
# git clone https://github.com/openstack/tripleo-quickstart.git
# cd tripleo-quickstart
# bash quickstart.sh -v --clean --teardown all --release stable/newton 127.0.0.2

Expected result
===============
An all-in-one deployment of OpenStack should be created.

Actual result
=============
The Playbook fails at the "Install the undercloud" step.

The exact problem is shown in the undercloud_install.log. The neutron-db-sync cannot properly create the neutron tables. I also tried deploying Ocata and there was a similar problem with creating the neutron tables. This causes the rest of the installation to fail.

2017-08-24 17:41:07 | 2017-08-24 17:41:07 - Notice: /Stage[main]/Neutron::Db::Sync/Exec[neutron-db-sync]/returns: oslo_db.exception.DBError: (pymysql.err.InternalError) (1054, u"Unknown column 'networks.shared' in 'field list'") [SQL: u'SELECT networks.id AS networks_id, networks.tenant_id AS networks_tenant_id, networks.shared AS networks_shared \nFROM networks \nWHERE networks.shared = 1']

Environment
===========
Computer:
x1 KVM virtual machine with nested virtualization enabled

OS:
CentOS 7.3

Specs:
20 CPU cores
30GB RAM
200GB space

Logs & Configs
==============

The Undercloud installation log is attached for reference.

[Luke Short (ekultails)]

When I try the same process with the trunk/ocata release it also fails relating to the Neutron database.

# bash quickstart.sh -v --clean --teardown all --release trunk/ocata 127.0.0.2

Here's the main error from this run:

2017-08-27 19:26:01 | 2017-08-27 19:26:01,197 INFO: Notice: /Stage[main]/Neutron::Db::Sync/Exec[neutron-db-sync]/returns: oslo_db.exception.DBError: (pymysql.err.InternalError) (1050, u"Table 'ml2_geneve_allocations' already exists") [SQL: u'\nCREATE TABLE ml2_geneve_allocations (\n\tgeneve_vni INTEGER NOT NULL, \n\tallocated BOOL NOT NULL DEFAULT false, \n\tPRIMARY KEY (geneve_vni), \n\tCHECK (allocated IN (0, 1))\n)ENGINE=InnoDB\n\n']

[Luke Short (ekultails)]

For the trunk/ocata deploy, here are all of the databases that were created, the tables created for the "neutron" database, and the structure of the neutron.ml2_geneve_allocations table: http://paste.openstack.org/show/619572/

[Luke Short (ekultails)]

Okay, I went back digging through the stable/newton undercloud_install.log that is attached to this report and I noticed a few things that I've pasted here: http://paste.openstack.org/show/619959/

(1) The Nova database fails to be fully created due to a timeout.

(2) The Neutron database also fails to be fully created due to a timeout. It looks like Puppet attempts to create the database again but then it fails because most of those tables already exist.

Because of #2 I am thinking that perhaps neutron-db-manage should be modified to run "CREATE IF NOT EXISTS" instead of a "CREATE" for tables. That's outside of the scope of tripleo, though. As for tripleo-quickstart, is there any way I can manually increase these timeouts? It seems that if the commands are allowed to run longer it may complete. I am running it inside of a virtual machine so there is likely a slight hit to performance.

[Luke Short (ekultails)]

The server hardware I have been testing on is 8 years old. The overhead of trying to use nested virtualization made this impossible to deploy. Even when deploying to bare-metal the installation was too slow to complete successfully.

This bug can be closed as invalid.