Skip to content
@jennifer/routeros

Bonding (link aggregation)

File: src/topics/bonding.j. Path: /interface/bonding (BONDING_PATH).

Background

A bond makes several physical links act as one logical interface. Two motivations, two modes:

  • Throughput + redundancy with a managed switch: LACP (802.3ad, constant BOND_LACP). Both ends negotiate the bundle; if a cable dies, the bundle shrinks and keeps working. Important expectation management: one flow (one download, one backup stream) still travels one link - LACP raises total capacity across many flows, not the speed of a single transfer.
  • Pure redundancy with any switch: active-backup (BOND_ACTIVE_BACKUP). One link carries traffic, the others stand by; no switch cooperation needed, which also makes it the right mode when the two cables go to two different switches.

The bond is an interface like any other afterwards - typically it goes straight into the LAN bridge, or carries VLANs.

Struct

jennifer
mt.Bond {
    id, name,
    slaves,               # member interfaces, comma-separated
    mode, primary,        # primary only meaningful in active-backup
    transmitHashPolicy,   # how 802.3ad spreads flows
    mac, mtu, running, disabled, comment
}

Functions

FunctionPurpose
addLacpBond(c, name, slaves) → id802.3ad aggregation (switch must match)
addFailoverBond(c, name, slaves, primary) → idactive-backup redundancy
addBond(c, name, slaves, mode) → idany of the seven modes
bonds(c) → list of Bondall bonds
bondByName(c, name) → Bondone bond
setBondSlaves(c, name, slaves)change membership
removeBond(c, name)delete (members become plain ports)
enableBond(c, name) / disableBond(c, name)switch on / off

Slave lists are validated before anything is sent: at least two distinct, existing interfaces, none already serving another bond; a failover bond's primary must be one of its slaves.

Examples

A two-cable LACP trunk to the core switch, feeding the LAN bridge:

jennifer
mt.addLacpBond($c, "bondtrunk", "ether1,ether2");
mt.addBridgePort($c, "brlan", "bondtrunk");
# on the switch: create a LAG/port-channel with LACP on those two ports

Redundant uplink across two independent switches (no LACP possible):

jennifer
mt.addFailoverBond($c, "bonduplink", "ether1,ether2", "ether1");
mt.addIpAddress($c, "10.0.0.2/24", "bonduplink");

Grow a bundle later:

jennifer
mt.setBondSlaves($c, "bondtrunk", "ether1,ether2,ether3");

Check the bundle's health:

jennifer
def b as mt.Bond init mt.bondByName($c, "bondtrunk");
io.printf("%s (%s) over %s running=%t\n", $b.name, $b.mode, $b.slaves, $b.running);
# per-member link state comes from the ethernet topic:
def ls as mt.LinkStatus init mt.linkStatus($c, "ether1");
io.printf("member ether1: %s at %s\n", $ls.status, $ls.rate);

Pitfalls

  • LACP needs both ends configured. A bond in 802.3ad against a switch without a matching LAG stays down (that is LACP protecting you from a loop). For "it must just work with any switch", use addFailoverBond.
  • One flow, one link. A single large transfer over a 2×1G LACP bond still peaks at 1G. More parallel flows spread better; the spread policy is transmit-hash-policy (generic set, e.g. "layer-3-and-4").
  • Member ports must be plain: not bridge ports, not carrying addresses. routeros checks they exist and are not in another bond; clearing other roles is on you.
  • running == true means at least one member is up - a degraded LACP bundle still shows running. Check the members' linkStatus for the full picture.
  • Removing a bond orphans whatever referenced it (bridge port, addresses) - tear those down first.