vm01 gains a primary IPv6 next to its IPv4, one standalone IPv6 address
joins the dataset; covers ip_version on both sources, combination with
filters, the IPv6-preferring behaviour of netbox's combined primary IP,
and the invalid value error path.
inventory.ip_version (or NETBOX_INVENTORY_IP_VERSION): both (default),
v4 or v6; the ip-addresses source filters by family server-side, the
virtual-machines source picks primary_ip4/primary_ip6 for ansible_host.
seed.py populates the stack through the API: a read-only user with a
self-provisioned token, tenants, sites, clusters, a VRF, virtual
machines (one with a primary IP) and standalone addresses; the tests
run the real inventory script against every source/filter combination,
every error path and the ansible --list/--host protocol.
compose stack modelled on the production netbox deployment (same pinned
images, worker, healthchecks) minus traefik: the API is served straight
on localhost and nothing survives an e2e-down.
one filter option plus its value, set in config or via
NETBOX_INVENTORY_FILTER / NETBOX_INVENTORY_FILTER_VALUE:
- tenant (both sources), resolved by name server-side
- named (ip-addresses), only addresses with a DNS name, deduplicated
- vrf (ip-addresses), resolved by name server-side
- domain (both sources), hostname ends with the domain
- cluster, site (virtual-machines), resolved by name server-side
hosts land in a single "netbox" group with hostvars under _meta; IP
addresses use the DNS name as inventory hostname when set, virtual
machines expose their primary IP as ansible_host when they have one.
config.yaml next to the script (or $NETBOX_INVENTORY_CONFIG) with environment
variable overrides, requests session with token auth, paginated list fetching
and readable error reporting.
