Multi-WAN, High Availability, policy routing. Failover breaks connections

dayer

I've also tested:

only with a gateway, but without a default gateway for the system (in order to use policy routing)

/root: netstat -r 
Routing tables

Internet:
Destination        Gateway            Flags      Netif Expire
8.8.4.4            192.168.1.1        UGHS   lagg0_vl
[...]

add a floating rule at the top with:
- action: pass
- quick: checked
- interface: select all
- direction: out
- address family: IPv4
- Protocol: Any
- Source: This Firewall (self)
- Destination: any
- Gateway: GW1 - 192.168.1.1

and a ping from the firewall to outdoor doesn't know the path:


/root: ping -S 127.0.0.1 8.8.8.8
PING 8.8.8.8 (8.8.8.8) from 127.0.0.1: 56 data bytes
ping: sendto: No route to host

However a ping from the WAN1 IP goes well:


/root: ping -S 192.168.1.111 8.8.8.8
PING 8.8.8.8 (8.8.8.8) from 192.168.1.111: 56 data bytes
64 bytes from 8.8.8.8: icmp_seq=0 ttl=57 time=3.357 ms

I think this could be related with some thread and the bug #5476 :(

dayer

I've simulated and simplified that situation with pfSense 2.4.0-RC using three virtual machines with VirtualBox and the host computer doing NAT (see the attachment file).
However the problem is the same.

Settings:

Pfsense1 as master. Pfsense2 as backup.
WLAN1 as default gateway (GW1), in the system routing table
WLAN2 as gateway (GW2) for traffic from the LAN to outside (with policy routing)
There are Outbound NAT from LAN in WAN1 (HA IP WAN1) and WAN2 (HA IP WAN2).

Reproduce:

I put a non-stop ping from PC to 8.8.8.8. The traffic flows through GW2. It's OK.
Disable CARP in Pfsense1. Now Pfsense2 is the master unit.
The states related to this ping are also in Pfsense2.
The ping begins to fail. There's no response.
With tcpdump in Pfsense2 I see:
- The packets from PC arrive to LAN interface correctly.
- The packets try leaving the firewall through WAN1 (¿¿¿why don't continue across WAN2???) and using the HA IP from WAN2 as source (because it keeps the NAT information according to the states)
If:
- I close the ping and relaunch it, the pings goes OK through WAN2.
- I enable CARP in Pfsense1 and it's the master unit again, the pings packets go through WAN2 again.

From my point of view, when a Pfsense is the new master unit and there's traffic flowing, it try to route the traffic only according to routing table (or static routes) and it ignores the policy routing.

This is the normal behavior in Pfsense? Do I need any special rules? Could it be a bug?

PD: I've fixed the thread subject.
PD2: I've also posted the first scenario in reddit and there's another person with the same behaviour.

pfsense-multi-wan-ha.png_thumb

dayer

I attach some information I'm using to explain the problem also in mailing list thread:

Configuration files from a scenario with pfSense 2.4-0-RC over two virtual machines
Screenshots for an example

00-initial-pf1.png_thumb

00-initial-pf2.png_thumb

01-gateways.png_thumb

02-rules.png_thumb

03-outbound-nat.png_thumb

04-ping-pf1-master.png_thumb

05-pf2-new-master.png_thumb

06-incorrect-routing.png_thumb

07-restart-ping.png_thumb
config.zip

luckman212

Interesting problem. Sorry I don't have a solution right now, but I'll be following this thread.

If, in your "Reproduce" section above, instead of stopping the ping and restarting it in step#6 - you go to Diag>States>Reset States and kill all states, does the ping start succeeding again?

dayer

Thank you for your interest :)

I've followed your suggestion related to the setp #6:

If I kill all states the ping starts succeeding again :)
If I kill all states related to the destination IP, the ping starts succeeding again :)
If I kill the states related to the destination IP in WAN2 , the ping continues failing :(
If I kill the states related to the destination IP in LAN , the ping starts succeeding again :)

And in the successfully situations, If Pfsense1 recover the master rol the ping continues succeeding.
I think it's a problem related to keep the routes in established states at LAN interface.

Could it be considered a bug?

luckman212

Ok at least you're a little closer to identifiying the culprit. I don't unfortunately have too much experience with CARP. So I will have to let someone else respond as to whether this is a config issue or a bug, and if it's a bug - whether the bug lies in pfSense or FreeBSD itself. If it's the latter, we'll have to file it upstream.

dayer

One more thing.
I've done the last tests again, with SSH instead of ping, but I haven't achieved recover the SSH.

Derelict

Could it be considered a bug?

Probably a configuration issue or an issue in your virtual environment. I have done countless tests like you are doing. Up to and including failing over during live video streams, etc and state sync works great.

I just pinged through from behind my 2.4.0-RC HA Multi-WAN VM pair last night when I updated it to 2.4.1-DEV. Worked great failing over and back.

dayer

Thank you very much for your tests.

Related to the failing over during live video streams, where those using the default gateway or… a secondary gateway according to a firewall rule or a failover gateway group? This difference is important because I've only found the problem if I'm using a not default gateway while I change the master unit.

I've done several tests, with virtual and physical machines, with LACP and without them, but the behavior always has been the same :-
I attached my config in the reply #3 a few days ago to try clearing any doubt about the settings.

Please, do you know a more detailed guide than Multi-WAN + Configuring pfSense Hardware Redundancy (CARP) from PFSenseDocs or the High Availability » Multi-WAN with HA from The pfSense Book for this purpose?
Could you share your tests configuration?

Derelict

Policy routing all LAN pings out WAN2. WAN is the default gateway. Pinging 8.8.8.8:

States on Primary/MASTER:

LAN icmp 172.25.236.227:29353 -> 8.8.8.8:29353 0:0 139 / 139 11 KiB / 11 KiB
WAN2 icmp 172.25.227.17:38069 (172.25.236.227:29353) -> 8.8.8.8:38069 0:0 139 / 139 11 KiB / 11 KiB

States on Secondary/BACKUP:

LAN icmp 172.25.236.227:29353 -> 8.8.8.8:29353 0:0 0 / 0 0 B / 0 B
WAN2 icmp 172.25.227.17:38069 (172.25.236.227:29353) -> 8.8.8.8:38069 0:0 0 / 0 0 B / 0 B

Persistent CARP maintenance mode on Primary:

States on secondary start seeing the traffic.

LAN icmp 172.25.236.227:29353 -> 8.8.8.8:29353 0:0 23 / 23 2 KiB / 2 KiB
WAN2 icmp 172.25.227.17:38069 (172.25.236.227:29353) -> 8.8.8.8:38069 0:0 23 / 23 2 KiB / 2 KiB

Client dropped three pings then continued. That's about right for a failover event.

States still exists on primary but are not seeing any traffic:

LAN icmp 172.25.236.227:29353 -> 8.8.8.8:29353 0:0 239 / 239 20 KiB / 20 KiB
WAN2 icmp 172.25.227.17:38069 (172.25.236.227:29353) -> 8.8.8.8:38069 0:0 239 / 239 20 KiB / 20 KiB

Leave Persistent CARP mantenance mode on Primary. States on primary are seeing the traffic again (239 now 262):

LAN icmp 172.25.236.227:29353 -> 8.8.8.8:29353 0:0 262 / 262 21 KiB / 21 KiB
WAN2 icmp 172.25.227.17:38069 (172.25.236.227:29353) -> 8.8.8.8:38069 0:0 262 / 262 21 KiB / 21 KiB

Same results using temporary enabling and disabling of CARP on the primary.

Derelict

And FWIW, here's a TCP session. Policy routed all outbound SSH out WAN2:

States on Primary:

LAN tcp 172.25.236.227:46380 -> 192.168.223.6:22 ESTABLISHED:ESTABLISHED 75 / 81 7 KiB / 10 KiB
WAN2 tcp 172.25.227.17:21325 (172.25.236.227:46380) -> 192.168.223.6:22 ESTABLISHED:ESTABLISHED 75 / 81 7 KiB / 10 KiB

States on Secondary:

LAN tcp 172.25.236.227:46380 -> 192.168.223.6:22 ESTABLISHED:ESTABLISHED 0 / 0 0 B / 0 B
WAN2 tcp 172.25.227.17:21325 (172.25.236.227:46380) -> 192.168.223.6:22 ESTABLISHED:ESTABLISHED 0 / 0 0 B / 0 B

I failed back and forth a couple times. At no point did the ssh session drop. Noticed a couple delays in output but TCP did its thing and no data was lost in the session.

dayer

Thank you Derelict :)

I've been simulating your tests.

WAN is the default gateway:

WAN1 (default)      WAN1 	xxx.xxx.xxx.xxx     8.8.8.8     WAN1 Gateway
WAN2                WAN2 	192.168.1.1         8.8.4.4     WAN2 Gateway

Policy routing all LAN traffic to outside go through WAN2:

States      Protocol    Source  Port    Destination     Port    Gateway     Queue   Schedule    Description
5 /3.80 MiB IPv4*       *       *       internals       *       *           none                Internal traffic to default gateway
0 /254 KiB  IPv4*       *       *       *               *       WAN2        none                The rest to WAN2

NAT for internal traffic to outside:

WAN1    internals 	* 	* 	* 	xxx.xxx.xxx.xxy     * 		NAT from internal networks
WAN2    internals 	* 	* 	* 	192.168.1.100       * 		NAT from internal networks

internals is an alias with internal networks.

Pinging 208.123.73.69

States on Primary/MASTER:

LAN     icmp    172.16.103.2:10618 -> 208.123.73.69:10618                           0:0     69 / 69     6 KiB / 6 KiB 	
WAN2    icmp    192.168.1.100:57632 (172.16.103.2:10618) -> 208.123.73.69:57632     0:0     69 / 69     6 KiB / 6 KiB

States on Secondary/BACKUP:


LAN 	icmp 	172.16.103.2:10618 -> 208.123.73.69:10618                           0:0     0 / 0 	0 B / 0 B 	
WAN2 	icmp 	192.168.1.100:57632 (172.16.103.2:10618) -> 208.123.73.69:57632     0:0     0 / 0 	0 B / 0 B

Persistent CARP maintenance mode on Primary:

States on secondary start seeing the traffic but something goes wrong. The number of packets observed matching the state from the destination side is zero.

LAN 	icmp 	172.16.103.2:10618 -> 208.123.73.69:10618 	                        0:0 	58 / 0 	5 KiB / 0 B 	
WAN2 	icmp 	192.168.1.100:57632 (172.16.103.2:10618) -> 208.123.73.69:57632 	0:0 	58 / 0 	5 KiB / 0 B

Client doesn't get ping reponses.

States still exists on primary but are not seeing any traffic (it's reasonable):

LAN 	icmp 	172.16.103.2:10618 -> 208.123.73.69:10618 	                        0:0 	368 / 368 	30 KiB / 30 KiB 	
WAN2 	icmp 	192.168.1.100:57632 (172.16.103.2:10618) -> 208.123.73.69:57632 	0:0 	368 / 368 	30 KiB / 30 KiB

Leave Persistent CARP maintenance mode on Primary. States on primary are seeing the traffic again (368 now 372):

LAN 	icmp 	172.16.103.2:10618 -> 208.123.73.69:10618 	                        0:0 	372 / 372 	31 KiB / 31 KiB 	
WAN2 	icmp 	192.168.1.100:57632 (172.16.103.2:10618) -> 208.123.73.69:57632 	0:0 	372 / 372 	31 KiB / 31 KiB

SSH to aaa.bbb.ccc.ddd (I've replaced the public IP for security reasons):

States on Primary:

LAN 	tcp 	172.16.103.2:43290 -> aaa.bbb.ccc.ddd:22                        ESTABLISHED:ESTABLISHED 	138 / 126 	12 KiB / 20 KiB 	
WAN2 	tcp 	192.168.1.100:54741 (172.16.103.2:43290) -> aaa.bbb.ccc.ddd:22  ESTABLISHED:ESTABLISHED 	138 / 126 	12 KiB / 20 KiB

States on Secondary:


LAN 	tcp 	172.16.103.2:43290 -> aaa.bbb.ccc.ddd:22                        ESTABLISHED:ESTABLISHED 	0 / 0 	0 B / 0 B 	
WAN2 	tcp 	192.168.1.100:54741 (172.16.103.2:43290) -> aaa.bbb.ccc.ddd:22  ESTABLISHED:ESTABLISHED 	0 / 0 	0 B / 0 B

Persistent CARP maintenance mode on Primary:

States on secondary start seeing the traffic, but something appears wrong:


LAN 	tcp 	172.16.103.2:43290 -> aaa.bbb.ccc.ddd:22                        ESTABLISHED:ESTABLISHED 	13 / 3 	2 KiB / 708 B 	
WAN2 	tcp 	192.168.1.100:54741 (172.16.103.2:43290) -> aaa.bbb.ccc.ddd:22  ESTABLISHED:ESTABLISHED 	13 / 3 	2 KiB / 708 B

With tcpdump in WAN1, I see the Secondary firewall is routing through WAN1 using the WAN2 VIP address like for NAT:

14:09:03.973890 IP 192.168.1.100.54741 > aaa.bbb.ccc.ddd.22: Flags [.], ack 1150910396, win 593, options [nop,nop,TS val 3297622232 ecr 1263162549], length 0
14:09:07.569891 IP 192.168.1.100.54741 > aaa.bbb.ccc.ddd.22: Flags [.], ack 1, win 593, options [nop,nop,TS val 3297625828 ecr 1263163448,nop,nop,sack 1 {4294967113:1}], length 0
14:09:08.810847 IP 192.168.1.100.54741 > aaa.bbb.ccc.ddd.22: Flags [P.], seq 0:52, ack 1, win 593, options [nop,nop,TS val 3297627069 ecr 1263163448], length 52
14:09:08.978668 IP 192.168.1.100.54741 > aaa.bbb.ccc.ddd.22: Flags [P.], seq 52:104, ack 1, win 593, options [nop,nop,TS val 3297627237 ecr 1263163448], length 52
14:09:09.035602 IP 192.168.1.100.54741 > aaa.bbb.ccc.ddd.22: Flags [P.], seq 52:104, ack 1, win 593, options [nop,nop,TS val 3297627294 ecr 1263163448], length 52
14:09:09.122214 IP 192.168.1.100.54741 > aaa.bbb.ccc.ddd.22: Flags [P.], seq 104:156, ack 1, win 593, options [nop,nop,TS val 3297627380 ecr 1263163448], length 52
14:09:09.180582 IP 192.168.1.100.54741 > aaa.bbb.ccc.ddd.22: Flags [P.], seq 0:156, ack 1, win 593, options [nop,nop,TS val 3297627439 ecr 1263163448], length 156

States still exists on primary but are not seeing any traffic (it's reasonable):

LAN 	tcp 	172.16.103.2:43290 -> aaa.bbb.ccc.ddd:22                        ESTABLISHED:ESTABLISHED 	232 / 220 	17 KiB / 34 KiB 	
WAN2 	tcp 	192.168.1.100:54741 (172.16.103.2:43290) -> aaa.bbb.ccc.ddd:22  ESTABLISHED:ESTABLISHED 	232 / 220 	17 KiB / 34 KiB

Leave Persistent CARP mantenance mode on Primary. States on primary are seeing the traffic again (232 now 294) and SSH terminal replies again:


LAN 	tcp 	172.16.103.2:43290 -> aaa.bbb.ccc.ddd:22                        ESTABLISHED:ESTABLISHED 	294 / 291 	22 KiB / 80 KiB 	
WAN2 	tcp 	192.168.1.100:54741 (172.16.103.2:43290) -> aaa.bbb.ccc.ddd:22  ESTABLISHED:ESTABLISHED 	294 / 291 	22 KiB / 80 KiB

I'm going to check our differences.

dayer

@Derelict, please, what VM has you tested? VirtualBox? VMware? Could you share your two XML config files?

I've tested that situation with VirtualBox (before and after a factory default, to remake and check the settings) and with two identical physical machines also. No success.
We're considering use pfSense and be sure this functionality works in pfSense is important to us.

Derelict

XenServer. No, I do not have the configurations from those tests any more.

The real key is how you are testing. Note that TRex was generating about 350K states there.

The hypervisor used will be of no consequence to what gets policy routed where.

dayer

Thank you, Derelict.

I can't find why a established state in pfSense1 isn't routed successfully when pfSense2 is the master.
I'm talking about this example (rules here):

Pinging 208.123.73.69

States on Primary/MASTER:

LAN     icmp    172.16.103.2:10618 -> 208.123.73.69:10618                           0:0     69 / 69     6 KiB / 6 KiB 	
WAN2    icmp    192.168.1.100:57632 (172.16.103.2:10618) -> 208.123.73.69:57632     0:0     69 / 69     6 KiB / 6 KiB

States on Secondary/BACKUP:


LAN 	icmp 	172.16.103.2:10618 -> 208.123.73.69:10618                           0:0     0 / 0 	0 B / 0 B 	
WAN2 	icmp 	192.168.1.100:57632 (172.16.103.2:10618) -> 208.123.73.69:57632     0:0     0 / 0 	0 B / 0 B

Persistent CARP maintenance mode on Primary:

States on secondary start seeing the traffic but something goes wrong. The number of packets observed matching the state from the destination side is zero.

LAN 	icmp 	172.16.103.2:10618 -> 208.123.73.69:10618 	                        0:0 	58 / 0 	5 KiB / 0 B 	
WAN2 	icmp 	192.168.1.100:57632 (172.16.103.2:10618) -> 208.123.73.69:57632 	0:0 	58 / 0 	5 KiB / 0 B

It's like the policy routing is ignored for this kind of situation and the firewall is trying route the established traffic through the default gateway (and use the NAT for a connection to exit through another WAN).
I've tried also with floating rules, without success.
There's some command or log to check what decisions takes pfSense with search state or packet

Derelict

Except policy routing is not ignored in that scenario.

It looks like it is not policy routing that is your problem, since you see the traffic going out thoses states.

It looks like your problem is whatever is upstream is refusing to accept the CARP VIP (and MAC address) moving from primary to secondary.

Policy routing only affects outbound traffic. It can't do anything about problems with the reply traffic.

dayer

Thank you Derelict.
I understand your point of view. However, if…

It looks like your problem is whatever is upstream is refusing to accept the CARP VIP (and MAC address) moving from primary to secondary.

I can't understand why I only see this behavior when the gateway from LAN traffic to outside is different from the default gateway. If the gateway from LAN traffic to outside is the same from the default gateway, everything goes well.

That is:

LAN: 192.168.2.0/24
WAN1: 192.168.1.0/24
WAN2: 192.168.56.0/24

Rules for LAN:

States      Protocol    Source  Port    Destination     Port    Gateway     Queue   Schedule    Description     Actions
1 /427 B    IPv4 *      *       *       LAN net         *       *           none
1 /1.04 MiB IPv4 *      *       *       *               *       GW1         none

Gateways (default gateway = gateway for LAN to outside):

Name            Interface   Gateway         Monitor IP
GW1 (default)   WAN1        192.168.1.1     192.168.1.1
GW2             WAN2        192.168.56.1    192.168.56.1

I try with SSH and it's goes well.

States relalted to xx.xxx.xxx.xxx in pfsense1 (master):


LAN     tcp     192.168.2.1:60626 -> xx.xxx.xxx.xxx:22522                           ESTABLISHED:ESTABLISHED     146 / 130   11 KiB / 20 KiB 	
WAN1    tcp     192.168.1.20:62445 (192.168.2.1:60626) -> xx.xxx.xxx.xxx:22522      ESTABLISHED:ESTABLISHED     146 / 130   11 KiB / 20 KiB

States related to xx.xxx.xxx.xxx in pfsense2 (backup):

LAN     tcp     192.168.2.1:60626 -> xx.xxx.xxx.xxx:22522                           ESTABLISHED:ESTABLISHED     0 / 0       0 B / 0 B 	
WAN1    tcp     192.168.1.20:62445 (192.168.2.1:60626) -> xx.xxx.xxx.xxx:22522      ESTABLISHED:ESTABLISHED     0 / 0       0 B / 0 B

Enter Persistent CARP Maintenance Mode

States related to xx.xxx.xxx.xxx in pfsense1 (backup):

LAN     tcp     192.168.2.1:60626 -> xx.xxx.xxx.xxx:22522                           ESTABLISHED:ESTABLISHED     339 / 321   21 KiB / 48 KiB 	
WAN1    tcp     192.168.1.20:62445 (192.168.2.1:60626) -> xx.xxx.xxx.xxx:22522      ESTABLISHED:ESTABLISHED     339 / 321   21 KiB / 48 KiB

States related to xx.xxx.xxx.xxx in pfsense2 (master):

LAN     tcp     192.168.2.1:60626 -> xx.xxx.xxx.xxx:22522                           ESTABLISHED:ESTABLISHED     111 / 111   6 KiB / 16 KiB 	
WAN1    tcp     192.168.1.20:62445 (192.168.2.1:60626) -> xx.xxx.xxx.xxx:22522      ESTABLISHED:ESTABLISHED     111 / 111   6 KiB / 16 KiB

But if the default gateway is not the gateway for LAN to outside:

Name            Interface   Gateway         Monitor IP
GW1             WAN1        192.168.1.1     192.168.1.1
GW2 (default)   WAN2        192.168.56.1    192.168.56.1

Then, the behavior is well until I put CARP Maintenance Mode (pfsense1 backup, pfsense2 master) and the states related to xx.xxx.xxx.xxx in pfsense2 (master) are:

LAN     tcp     192.168.2.1:60632 -> xx.xxx.xxx.xxx:22522                           ESTABLISHED:ESTABLISHED     31 / 5      5 KiB / 1 KiB 	
WAN1    tcp     192.168.1.20:49862 (192.168.2.1:60632) -> xx.xxx.xxx.xxx:22522      ESTABLISHED:ESTABLISHED     31 / 5      5 KiB / 1 KiB

and the SSH client is like frozen until I leave Persistent CARP Maintenance Mode and pfsense1 recovers the master role.

Derelict

I don't know. You have something scrwed up in your outbound NAT it looks like. I built this and it works fine. Using multiple versions of pfSense. Countless people are doing exactly the same thing.

I suggest you reinstall and start over as simply as possible. Adding nothing but what is necessary to test this concept.

You are chasing a red herring with the "not default gateway" thing. It doesn't exist. It is something else you have done.

ZsZs

Hi Dayer and Derelict,

Apparently I am in the same situation as Dayer. The network layout is the same.
I am using 2.3.4 and I made a clean install as follows (apologize for the detailed list, but there might be an obvious mistake or missing part):

the two VMs running on the same ESXi 6.0U3 host (for testing purpose)
set up WAN1 and WAN2 in different subnet:
- WAN1 public /26 (default GW)
- WAN2 internal /24 (behind a cable modem)
set up WAN1 and WAN2 with following monitoring IP addresses:
- WAN1: 8.8.8.8
- WAN2: 208.67.220.220
add two DNS servers to each WAN
Configure DNS resolver to forwarder mode
install Open-vm-tools package (no other packages have been installed)
Set up HA for syncing state and configs
Set up CARP IPs (WAN1-VIP, WAN2-VIP, LAN-VIP) with appropriate netmask
change Outbound NAT to Manual
- remove auto-created SYNC interface related outbound NAT rules
- change NAT address to WAN1-VIP on rules with interface WAN1
- change NAT address to WAN2-VIP on rules with interface WAN2
create WAN1first gateway group with WAN1GW Tier1, WAN2GW: Tier2
create WAN2first gateway group with WAN1GW Tier2, WAN2GW: Tier1
create FW rule with Policy routing for ssh traffic in LAN:

Protocol    Src Prt Dst Prt Gateway     Queue
IPv4 TCP    *   *   *   22  WAN2first   none

I've tried following policy routing scenarios by simply:

changing GW in the above rule
disabling the aboce rule
toggling default GW


defGW   policy route   SSH sesseion after failover
WAN1    disabled       OK
WAN1    GW:WAN1GW      OK
WAN1    GW:WAN2GW      Freezes
WAN1    GW:WAN1first   OK
WAN1    GW:WAN2first   Freezes

WAN2    disabled       OK
WAN2    GW:WAN1GW      Freezes
WAN2    GW:WAN2GW      OK
WAN2    GW:WAN1first   Freezes
WAN2    GW:WAN2first   OK

I had the same issue, that in case the policy routing rule points to a gateway (group) other than the default, then after the HA fail-over to the secondary node the opened session freezes.
I can open a new ssh session via the new master, but moving the VIP back to the primary node this one freezes and the previously opened session starts responding again.
I also saw in tcpdump, that when the ssh session freezes, the traffic leaves the firewall on wrong WAN interface (on the default one) with the other WAN interface's source IP address.

I appreciate any hints you might have.

Regards,
Zsolt

edit: typos, some clarification added

dayer

Hi ZsZs,

I've read your description with attention and I think it could be the same problem. Although I'll wait to Derelict point of view.
I know this part is the key:

I also saw in tcpdump, that when the ssh session freezes, the traffic leaves the firewall on wrong WAN interface with the other WAN interface's source IP address.

However, according to the last Derelict suggestion, I must repeat the test with the simplest scenario.