Orginizing Fabrics by Ansible Folders

[theotherguy2175]

Is it possible to orginize fabrics by folders?

Meaning in some fashion have directories based on each fabric. For instance we could build our CORE network in a directory that hold it's own group_vars and hostvars, and the same thing for the DATA_CENTER, and CAMPUS?

[ClausHolbechArista]

Sorry for the late reply.

It is a common pattern to build different fabrics as separate Ansible inventories.

An example could be:

avd
├── global_vars
│   ├── management             # Our 'global_vars' plugin will parse one layer of folders for YAML files.
│   │   ├── snmp_settings.yml
│   │   ├── local_users.yml
│   │   └── <...>
│   ├── network_services
│   │   ├── network_services_keys.yml  # Contains only the 'network_services_keys' key
│   │   └── <...>
│   └── <...>
├── playbooks
│   ├── build.yml
│   ├── deploy.yml
│   └── <...>
├── DATA_CENTER
│   ├── group_vars
│   │   ├── DATA_CENTER           # Ansible will parse one layer of folders for YAML files.
│   │   │   ├── misc.yml          # Contains 'fabric_name' and other common settings
│   │   │   └── <...>
│   │   ├── NETWORK_SERVICES
│   │   │   ├── tenant_a.yml      # Contains only the 'tenant_a' key (defined under 'network_services_keys')
│   │   │   └── <...>
│   │   ├── DC1
│   │   │   ├── misc.yml          # Contains 'dc_name' and other common settings
│   │   │   └── <...>
│   │   ├── DC1_SUPER_SPINES
│   │   │   ├── super_spine.yml   # Contains only the 'super_spine' and 'type' keys
│   │   │   ├── l3_edge.yml       # Contains only the 'l3_edge' key
│   │   │   └── <...>
│   │   ├── DC1_POD1
│   │   │   ├── misc.yml          # Contains 'pod_name' and other common settings
│   │   │   └── <...>
│   │   ├── DC1_POD1_SPINES
│   │   │   ├── spines.yml        # Contains only the 'spines' and 'type' keys
│   │   │   └── <...>
│   │   ├── DC1_POD1_L3LEAFS
│   │   │   ├── l3leaf.yml        # Contains only the 'l3leaf' and 'type' keys
│   │   │   └── <...>
│   │   ├── DC1_POD1_RACK1
│   │   │   ├── servers.yml       # Contains only connected endpoints for this rack
│   │   │   └── <...>
│   │   └── <...>
│   ├── host_vars
│   │   ├── dc1-pod1-spine1       # Ansible will parse one layer of folders for YAML files.
│   │   │   └── <...>             
│   │   └── <...>             
│   └── inventory.yml
├── CAMPUS
│   ├── group_vars
│   │   ├── CAMPUS                # Ansible will parse one layer of folders for YAML files.
│   │   │   ├── misc.yml          # Contains 'fabric_name' and other common settings
│   │   │   └── <...>
│   │   ├── NETWORK_SERVICES
│   │   │   ├── tenant_a.yml      # Contains only the 'tenant_a' key (defined under 'network_services_keys')
│   │   │   └── <...>
│   │   ├── CAMPUS_CORE
│   │   │   ├── spine.yml         # Contains only the 'spine' and 'type' keys
│   │   │   └── <...>
│   │   ├── BUILDING1
│   │   │   ├── misc.yml          # Contains 'dc_name' and other common settings
│   │   │   └── <...>
│   │   ├── BUILDING1_CORE
│   │   │   ├── l3spine.yml       # Contains only the 'l3spine' and 'type' keys
│   │   │   └── <...>
│   │   ├── BUILDING1_FLOOR1
│   │   │   ├── l2leaf.yml         # Contains only the 'l2leaf' and 'type' keys
│   │   │   ├── network_ports.yml  # Contains only the 'network_ports' key
│   │   │   ├── misc.yml           # Contains 'pod_name' and other common settings
│   │   │   └── <...>
│   │   └── <...>
│   ├── host_vars
│   │   ├── bld1-fl1-sw1           # Ansible will parse one layer of folders for YAML files.
│   │   │   └── <...>             
│   │   └── <...>             
│   └── inventory.yml
├── ansible.cfg
└── Makefile

Your Makefile could contain make targets like:

.PHONY: build-dc
build-dc: ## Build DC inventory
<tab>ansible-playbook playbooks/build.yml -i DATA_CENTER/inventory.yml

All this is just some example I made up for this answer, but we will work on an example for all this as time allows.

[theotherguy2175]

This is a huge help and on the surface this works great. The only problem I am having is it would be nice to have 1 AVD project for our whole network. And when testing as long as the fabric was built in the respective folders everything works. But for instance if we had a CORE network with P and PE nodes, I'm not sure how devices could be part of the fabric for DATA_CENTER for instance.

When I tried running commands to include multiple inventories and yaml files I got an error saying AVD can't build separate fabrics. Is it not possible to somehow have P and PE in a CORE folder and then have CE devices in other folders but still use the same fabric?

[ClausHolbechArista]

I would like to understand your pain points in more detail. Why do you need the devices to be in the same AVD project / build?

[theotherguy2175]

Just from a simplicity standpoint. Even the globar_vars would be nice. Instead of having 4-5 repos, it could be just 1. But as mentioned having them in different folders makes organization easier, but AVD doesn't seem to have the capability to do a EVPN/VXLAN fabric for example for different folders.

[ClausHolbechArista]

We are inheriting some limitations of Ansible around this. Ansible can have one folder of group_vars and host_vars and one inventory.yml file, so there is not much we can do around that.