feat(VMware): add option overwrite_host_interface_type

[tfoks]

The VMware source derives a host pNIC's type from its current link speed. If the NIC has no link, it uses the first valid link speed, or other. The check_redfish source sets the type from the NIC hardware. With both sources on the same ESXi hosts, they disagree on every run:

INFO: Interface 'vmnic1 (de-mucesx54)' attribute 'type' changed from '10gbase-x-sfpp' to '1000base-t'
INFO: Interface 'vmnic4 (de-mucesx54)' attribute 'type' changed from '10gbase-x-sfpp' to '100base-tx'

The first case is an SFP+ port running at 1G; the second is an SFP+ port without link. Within one run, the revert logic cancels the PATCH again. What repeats is the log line, and the final type depends on the order of the sources.

This adds an option to the VMware source, as suggested in the review:

; define if the type of a host interface derived from its link speed overwrites a physical
; interface type in NetBox. Set to False if a source which knows the hardware (i.e.
; check_redfish) sets the type of the host interfaces
;overwrite_host_interface_type = True

The default True keeps the current behaviour, so vCenter-only setups still get a wrong type corrected. With False, the type of an existing host interface is kept if it's a physical type. virtual, bridge, lag, other or no type are still set by vCenter. New interfaces get the type vCenter derives.

Changes:

• module/sources/vmware/config.py, settings-example.ini: new option overwrite_host_interface_type, default True.
• module/sources/vmware/connection.py: with the option False, don't send a type for an existing host interface that has a physical type. NetBox returns the type as {"value": ..., "label": ...}, so the value is read from that dict.
• tests/test_vmware_host_nic_type.py: second-run tests. With False, a physical type is kept and other is still updated. With the default, a wrong type is still set from the link speed. The first test fails on development.

Tested with the vcsim captures; see the tests above. Against NetBox 4.7.2 with a vCenter source and the check_redfish inventory files of HPE and Lenovo hosts, the first version of this PR stopped the type changes in the dry-run. Confirmed with overwrite_host_interface_type: False on our test and prod instances: the dry-run logs no type changes for host interfaces and no errors. pytest: 374 passed, 8 skipped (vcsim installed).

🤖 Generated with Claude Code

[semx]

The conflict between the two sources is real, but the speed heuristic changes behaviour for everyone:

• vCenter-only setups no longer get a too fast type corrected. Checked on 4.7.2 with vcsim: development sets vmnic0 back from 400gbase-x-qsfpdd to 1000base-t, this branch keeps 400G.
• A port without link: vCenter derives other, its valid_types value is a string, so the guard is skipped and the SFP+ type is still overwritten, the second case from the description.
• The FC types are in Gbit (1..128) and the Ethernet ones in Mbit, so with sync_host_fc_adapters they're compared across families.

I'd make this explicit instead: an overwrite_host_interface_type option like overwrite_device_interface_name, default True (current behaviour), and with False an existing physical type is kept. That covers redfish plus vCenter without guessing from speeds.

On the description: within one run the PATCH is cancelled again by the revert logic, what repeats is the log line, and the final type depends on the order of the sources.

[tfoks]

Agreed. The speed comparison was guessing, and your three cases show it. I replaced it with overwrite_host_interface_type, modelled on overwrite_device_interface_name:

• Default True: unchanged behaviour. A test checks that a too-fast type is still corrected from the link speed.
• False: an existing physical type is kept. virtual, bridge, lag, other or no type are still set by vCenter. A port without link no longer matters, and nothing compares FC and Ethernet speeds.
• Updated the description: within one run the PATCH is cancelled again, what repeats is the log line, and the final type depends on the source order.

The title is now feat(VMware): add option overwrite_host_interface_type.

[semx]

Much simpler this way, thanks. I tried it on the vcsim setup: set two vmnics to SFP+ and one to other through the API. With the option off the SFP+ types stayed, other got the link speed type, and the next run didn't touch anything. With the default, vCenter put 1000base-t back like before.

[tfoks]

Confirmed with overwrite_host_interface_type: False against NetBox 4.7.2 with a vCenter source plus the check_redfish inventory files of HPE and Lenovo hosts, on both our test and prod instances: the dry-run logs no type changes for host interfaces and no errors. I updated the description to match.