Cumulus, NVUE, MLAG
Context: Cumulus Linux on Mellanox MSN3700 (Spectrum-2, 32×200G QSFP56) and Spectrum-4 400G switches, NVUE, in MLAG pairs — the estate's switches after the Onyx era. This entry is the bites, collected from the running notes across the 5.6 → 5.10 trail; the reusable shapes — MLAG essentials, the parameterized port config, breakouts, ONIE installs — are in Evergreen. Hostnames, addresses and VLAN layouts below are genericized.
What bit — quick reference
| The thing you are staring at | Fix | § |
|---|---|---|
Invalid config: multiple vlan-aware bridges are not supported with mlag |
one bridge, br_default; segment with VLANs |
1 |
vlan + access + untagged conflict on validation |
one attachment mode per interface | 2 |
br_default state DOWN with a member down |
svi-force-up if the SVI must stay up |
3 |
snmpd in a management VRF answers nothing (ENETUNREACH) |
run it inside the VRF explicitly | 4 |
| a link that flaps or refuses to come up | fixed speed beats autoneg, usually | 5 |
| a 400G link that works and loses packets | mlxlink, trust the raw BER, reseat the optics |
6 |
MSTP_IN_rx_bpdu ... Clear Bridge assurance inconsistency |
per-interface stp network enabled, both switches |
7 |
Notes on a few:
-
The bridge SVI tracks its members — with one down, the SVI goes down with it.
nv set system global svi-force-up enable onwhen that is not what you want. -
snmpd as a plain systemd service does not inherit the VRF, and answers nothing — strace shows
sendmsg = -1 ENETUNREACH. Run it inside the VRF explicitly:# /etc/systemd/system/snmpd.service.d/overwrite.conf ExecStart=/usr/bin/vrf exec mgmt /usr/sbin/snmpd -LOw -u Debian-snmp \ -g Debian-snmp -I -smux -f --noTokenWarnings ProtectControlGroups=false # VRF execution needs control groups -
On links that flap or refuse to come up, forcing works — but note both directions in the notes worked:
nv set interface swp26 link speed 40Gfixed one, and on 5.6.0speed autoalso fine. The full nudge set:nv set interface swp20 link speed 10G nv set interface swp20 link auto-negotiate disabled nv set interface swp20 link duplex full nv set interface swp20 link fast-linkup enabled nv set interface swp20 link state up -
Random packet loss on a 400G link.
mlxlinkis the tool that tells the truth:sudo mlxlink -d /dev/mst/mt53104_pci_cr0 -p <port> --show_counters Recommendation : Bad signal integrity Effective Physical Errors : 0 Effective Physical BER : 15E-255 Raw Physical BER : 2E-5 Link Down Counter : 6
Effective counters clean, raw BER 2E-5 — the link "works" and loses packets. Trust the raw counters; reseat or replace the optics.
- syslog fills with
MSTP_IN_rx_bpdu ... Clear Bridge assurance inconsistencyon a bond, and the reason is one missing per-interface knob:nv set interface bond1 bridge domain br_default stp network enabled
Do it on both switches of the pair.
This saved you a night? I do this for a living: info@wirt.ee.