Closes #20972: Add support for channelized subinterfaces

[jeremystretch]

Closes: #20972

• New model fields:
  • Add channels and channel_id to Interface.
  • Add parent, channels, and channel_id to InterfaceTemplate.
• Add a unique constraint for (parent, channel_id) to Interface and InterfaceTemplate.
• Override set_cable_termination() and clear_cable_termination() on the Interface model to replicate cable assignment to channelized subinterfaces.
• Adapt the cable tracing logic & rendering to support paths terminated to subinterfaces for a cable connected to the physical parent interface.

[jeremystretch]

@claude /review Only report on issues of potential concern; we don't need commentary on anything that doesn't require further consideration.

[github-actions]

Claude finished @jeremystretch's task in 5m 22s —— View job

---

PR Review: Channelized Subinterfaces

• Gathering context and reading diff
• Analyzing for issues of concern
• Posting review

---

Issues of Concern

1. No validation that cable profile capacity covers all channel IDs

netbox/dcim/models/device_components.py:1259-1266 / netbox/dcim/cable_profiles.py:28-52

propagate_channel_cables() assigns each child cable_positions=[child.channel_id]. If the parent has more channels than the cable profile has positions (e.g. 8-channel parent on a 1C4P cable), channels 5–8 get position values that have no corresponding CableTermination. get_peer_termination() then returns None silently — the path tracer creates incomplete paths with no user-facing error or warning.

CableProfile.clean() validates the number of connectors (terminations), not positions per connector, so it doesn't catch this mismatch. A channelized interface can be cabled with a profile whose position count is lower than its channels value, and the resulting dead-end paths are invisible to the user.

Suggestion: validate in propagate_channel_cables() (or somewhere in Cable.clean() / CableTermination.save()) that parent.channels <= profile_position_count_for_the_relevant_connector. Fix this →

---

2. API test coverage gap for new fields

netbox/dcim/tests/test_api.py

channels and channel_id are added to the Interface and InterfaceTemplate serializers, but the existing test_api.py has no test data or assertions for these fields. The standard NetBox API test pattern exercises create/update/bulk operations on all serializer fields; the absence of these fields from the test fixtures means their serialization, filtering, and validation aren't exercised by CI.

---

3. N+1 queries in signal handler update_channelized_cable_paths

netbox/dcim/signals.py:202-205

for parent in Interface.objects.filter(pk__in=parent_ids, cable__isnull=False):
    if parent.channels:
        parent.propagate_channel_cables()
    rebuild_cable_paths(parent.cable)

propagate_channel_cables() accesses self.cable and self.cable.profile (line 1253). Because the query above has no select_related('cable__profile'), each iteration issues two additional queries (one to fetch the Cable, one for its profile). For a parent with a large number of simultaneously-moving channel subinterfaces this is minor, but adding .select_related('cable__profile') to the filter on line 202 would eliminate the extra round-trips.

---

4. Intermediate invalid DB state during device/module instantiation

netbox/dcim/utils.py:169-184 / netbox/dcim/models/device_component_templates.py:569-582

InterfaceTemplate.instantiate() passes channel_id=self.channel_id when building the Interface object. That interface is saved immediately (without a parent, since parents are wired in the second pass by update_interface_parents()). Because save() does not call full_clean(), the DB briefly contains channel-type interfaces with channel_id set but parent=NULL — a state that fails model validation. If an exception interrupts the process between the first save and update_interface_parents(), those orphaned interfaces remain.

This is a low-risk concern since both passes should be wrapped in a transaction, but the intermediate state bypasses the invariant enforced by InterfaceValidationMixin.clean(). One approach: defer saving channel subinterface templates until after their parents exist, following the two-pass order used for update_interface_bridges.

[jeremystretch]

1. I'm going to keep this as-is for now, as it can be argued that it's desirable behavior. We can re-evaluate once we get feedback from the v4.7 beta.
2. Addressed
3. Addressed (by adding selected_related('cable'))
4. This seems like a narrow, acceptable risk. Going to leave it for now, as mitigating it would likely require a substantial change to the instantiation mechanism.