a two-member LAG drops all traffic when one member's link goes down #1515

Closed
opened 2026-11-15 18:01:03 +00:00 by jaylouisw · 0 comments
Owner

_forwardL2 builds its usable-member list and filters out STP-blocked ports, wrong-VLAN ports and the ingress port - but not ports whose link is down. The hash then picks among all members, may pick the dead one, and _deliver drops the frame with reason link down.

Measured: a two-member channel-group 1 between two switches, with member 1's cable pulled, loses the ping entirely. Reproduced isolated.

Expected: a LAG with one live member carries traffic. Losing one member of a bundle is the case bundles exist for.

Why it matters beyond the lab: "no single cable failure kills HA" is one of the intent checks a plan is supposed to be scored against, and the most common HA implementation in a homelab is a two-member LAG. As it stands the simulator fails that check for a simulator reason, which makes the check worthless the first time it is used.

Where: engine/simulator.js _forwardL2 usable-port list; _deliver rejection path.

Suggested fix: exclude ifaceStatus-down members from the hash, and if the bundle has no usable member, fail with a code that names the bundle rather than dropping the frame silently.

Found by the w7 intent-check study, probe6.js.

`_forwardL2` builds its usable-member list and filters out STP-blocked ports, wrong-VLAN ports and the ingress port - but not ports whose link is down. The hash then picks among all members, may pick the dead one, and `_deliver` drops the frame with reason `link down`. Measured: a two-member `channel-group 1` between two switches, with member 1's cable pulled, loses the ping entirely. Reproduced isolated. **Expected:** a LAG with one live member carries traffic. Losing one member of a bundle is the case bundles exist for. **Why it matters beyond the lab:** "no single cable failure kills HA" is one of the intent checks a plan is supposed to be scored against, and the most common HA implementation in a homelab is a two-member LAG. As it stands the simulator fails that check for a simulator reason, which makes the check worthless the first time it is used. **Where:** `engine/simulator.js` `_forwardL2` usable-port list; `_deliver` rejection path. **Suggested fix:** exclude `ifaceStatus`-down members from the hash, and if the bundle has no usable member, fail with a code that names the bundle rather than dropping the frame silently. Found by the w7 intent-check study, probe6.js.
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jaylouisw/packetforge#1515
No description provided.