That is may be because you have valid routes in main table
I can't fully agree with:
>> No, it can't be rubbish, it has to be real. The load-balance engine looks up the main table for default routes to establish default route for each defined load-balanced interface and it's respective routing table.
because i don't have any valid routes in MAIN routing table. Only in policy routing tables and it still works.
That is because Linux make routing desicion twice.
APPLICATION -> Routing -> OUTPUT Mangle -> OUTPUT NAT -> OUTPUT Filter -> Routing -> POSTROUTING Mangle -> POSTROUTING NAT -> out
The First routing (MAIN table) can have any route. The Second Policy Routing Table defines finle route.
I have a default route in MAIN table wich point to nonexistant host in LAN.
In case you have valid route on WAN, then this rules should be trigged on ICMP and DNS:
-A UBNT_WLBO_VPS -o eth0 -p icmp -j MARK --set-xmark 0x800000/0x7f800000
-A UBNT_WLBO_VPS -o eth1 -p icmp -j MARK --set-xmark 0x1000000/0x7f800000
-A UBNT_WLBO_VPS -o eth0 -p udp -m udp --dport 53 -j MARK --set-xmark 0x800000/0x7f800000
-A UBNT_WLBO_VPS -o eth1 -p udp -m udp --dport 53 -j MARK --set-xmark 0x1000000/0x7f800000
-A UBNT_WLBO_VPS -m mark ! --mark 0x0/0x7f800000 -j RETURN
After that it will exit from Load-Balance and will not reach LB rules:
-A UBNT_WLBO_VPS -m state --state NEW -m mark --mark 0x0/0x7f800000 -m dyn_random --prob-name "VPS_0" -j MARK --set-xmark 0x800000/0x7f800000
-A UBNT_WLBO_VPS -m state --state NEW -m mark --mark 0x0/0x7f800000 -j MARK --set-xmark 0x1000000/0x7f800000
But it still should work with other types of traffic.
