Thanks for the reply @Clive_A ,

It seems that "Backup Mode" addresses exactly my issue. I'm using standalone mode, though, I see both "Mode" options (with a third timed one) but no "Backup Mode".

I'll test this again focusing to see if they're old sessions as you suggested.

  @Clive_A I can confirm that new sessions are still being sent through secondary WAN3 several minutes (edit: an hour still) after primary WAN1 comes online.

I tested with both with speedtest (<1 Gbps) and traceroute (second hop goes to WAN3 ISP).

EDIT:

To be honest, keeping old sessions seems reasonable, but right now it's switching to the secondary WAN3 indefinitely. I'd rather have the choice to terminate sessions and switch to primary (good for LTE WAN) or keep old sessions and switch new ones to primary (good for connectivity if asymmetry isn't too high).

The behaviour I would expect:

• No load balancing, no failover:
  • Use a random WAN statically, in case of failure remove that WAN row from the routing table and whatever happens.
• Load balancing:
  • Randomly distribute sessions between WANs. [Worst]
  • Weighted Round-Robin between them considering fixed throughput [this is what I think the router does currently], or...
  • Even better, use the WAN with fastest throughput available (approx) when deciding (keeping track of the current usage against a fixed speed. [Possible enhancement!]
• Failover:
  • Same priority:
    • Failure of a primary WAN switches to another random primary WAN, keeps all traffic in the new WAN indefinitely.
  • Different priority:
    • If any/all primary are down (selectable as it is now), switch to secondary WAN and route back to primary when it recovers and two options:
      • Keep old sessions in the secondary to avoid connectivity interruptions. [Good for unmetered/fiber secondary.]
      • Kill old sessions and route back everything to the primary. [Good for metered/LTE/slow secondary.]

  @Clive_A  I have the same problem, can you follow up on the bug?

Hello  @Clive_A !

I see you enabled load balancing, that's a completely different test case than what I introduced in my first post, as with failover and without load balancing the WANs aren't used alternatively (unlike you've shown) unless one fails.

I'm not using load balancing because my WANs are very asymmetrical and the secondary doesn't have dual stack.

  @Clive_A  I'll test then as you suggested, with load balancing enabled.

But I think that load balancing with primary-secondary shouldn't have used the secondary until after the primary was down (as per the wording of the option was implying).