Onyx MLAG

Context: NVIDIA Onyx on Mellanox MSN3700 switches, MLAG pairs — the OS that ran the 100G pairs before the estate moved to Cumulus and NVUE. Same hardware, older school. This entry is the dated bites; the recipes — MLAG port-channel shapes, upgrade procedure, LLDP setup, CLI corners, management-address rollback — are in Evergreen. Hostnames, addresses and VLAN layouts genericized; the date is from the log lines in the notes.

lacp-individual: you cannot have both

lacp-individual enable force lets a port-channel member act alone when the host speaks no LACP. Without it, PXE does not work — the boot NIC has no LACP. With it, a running Red Hat 9 bond failover breaks. The notes' wording, translated: hold the beak and the tail comes loose.

The follow-up bug: when acting as individual, the switch picks the member port randomly. The fix pins it:

# show lacp interfaces ethernet 1/20
#   Port State: Down, Aggregation State: Aggregation, Defaulted
interface ethernet 1/20 lacp port-priority 32767

Lowest numerical priority wins — so the same member acts individual every time, and PXE becomes deterministic.

Bugs

Configuration error, interface Eth1/25 configured as a router port when adding an interface to an MLAG channel-group. The port had a leftover no switchport force. The fix is not intuitive — clear it by typing the opposite twice:
```
interface ethernet 1/25 no switchport force
interface ethernet 1/25 switchport force
interface ethernet 1/25 switchport
```
clusterd restarts on management address change — I/F address changed from ... restarting in the logs. Expected when renumbering mgmt0, alarming if you did not.

This saved you a night? I do this for a living: info@wirt.ee.