config-download missing data when deploying with metalsmith

[John Fulton]

Using TripleO master as of the last week ofMay 2020, if I deploy with metalsmith [1] using 1 compute and 1 controller everything is fine. However If I scale to 3 controllers and 2 computes, then the following data is missing:

1. The tripleo-ansible-inventory command only has 1 controller and 1 compute. The other 2 controllers and other one compute are missing so they never get configured

2. The config-download host_vars directory contains compute-0 and controller-0 but not other hosts

3. If I modify my deployment to include 3 extra CephStorage nodes then 0 CephStorage nodes are in the resulting config-download output (not even one CephStorage related entry in the inventory or host_vars or group_vars).

[1] https://docs.openstack.org/project-deploy-guide/tripleo-docs/latest/provisioning/baremetal_provision.html

[John Fulton (jfulton-org)]

How to reproduce:

1. Provisioning hardware:

    openstack overcloud node provision \
              --stack $STACK \
              --output deployed-metal-big.yaml \
              metal-big.yaml

Where metal-big.yaml is defined as:

 https://github.com/fultonj/victoria/blob/43daa3d986003507e98ab4ff3adb79083db23d42/substandard/metal-big.yaml

2. Deploy overcloud:

ln -s /usr/share/openstack-tripleo-heat-templates templates
openstack overcloud -v deploy \
         --stack oc0
         --templates ~/templates/ \
         -n ../network-data.yaml \
         -e ~/templates/environments/deployed-server-environment.yaml \
         -e deployed-metal-big.yaml \
         -e ~/templates/environments/net-multiple-nics.yaml \
         -e ~/templates/environments/network-isolation.yaml \
         -e ~/templates/environments/network-environment.yaml \
         -e ~/templates/environments/disable-telemetry.yaml \
         -e ~/templates/environments/low-memory-usage.yaml \
         -e ~/templates/environments/enable-swap.yaml \
         -e ~/templates/environments/podman.yaml \
         -e ~/generated-container-prepare.yaml \
         -e ~/domain.yaml \
         -e overrides.yaml \
         --stack-only \
         --libvirt-type qemu

Where overrides.yaml and network-data.yaml are:

 https://github.com/fultonj/victoria/blob/43daa3d986003507e98ab4ff3adb79083db23d42/substandard/overrides.yaml
 https://github.com/fultonj/victoria/blob/43daa3d986003507e98ab4ff3adb79083db23d42/network-data.yaml

3. Download the config and see that host_vars only has two hosts:

    openstack overcloud config download \
              --name oc0 \
              --config-dir config-download

4. Genereate an inventory and observe that it only contains control-0 and compute-0:

    tripleo-ansible-inventory --static-yaml-inventory \
                                  tripleo-ansible-inventory.yaml \
                                  --stack oc0

[John Fulton (jfulton-org)]

The deployed-metal-big.yaml output by 'openstack overcloud node provision' is at:

https://github.com/fultonj/victoria/blob/43daa3d986003507e98ab4ff3adb79083db23d42/substandard/deployed-metal-big.yaml

[John Fulton (jfulton-org)]

I'm able to reproduce this using Master on July 1 as described in comment #3 even if I follow steps 1 and 2 only and don't pass '--stack-only'. When config-download is run the the default process the inventory at the following path still missing hosts:

 /home/stack/config-download/<STACK>/tripleo-ansible-inventory.yaml

[John Fulton (jfulton-org)]

I'm no longer able to reproduce this bug using the main branch. I think my problem was the hostnamne map.

https://github.com/fultonj/victoria/commit/594232d6337f32e82e1c5fc50c142dfd9973469e#diff-4cb67750a36e0efdff239c53b6d616894d573c85e26c9200c55643f0f6e0d544