Multi-WAN, High Availability, policy routing. Failover breaks connections
-
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?
-
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?
-
-
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.
-
One more thing.
I've done the last tests again, with SSH instead of ping, but I haven't achieved recover the SSH. -
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.
-
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? -
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 KiBStates 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 BPersistent 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 KiBClient 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 KiBLeave 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 KiBSame results using temporary enabling and disabling of CARP on the primary.
-
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 KiBStates 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 BI 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.
-
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 GatewayPolicy 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 WAN2NAT for internal traffic to outside:
WAN1 internals * * * xxx.xxx.xxx.xxy * NAT from internal networks WAN2 internals * * * 192.168.1.100 * NAT from internal networksinternals 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 KiBStates 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 BPersistent 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 BClient 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 KiBLeave 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 KiBStates 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 BPersistent 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 BWith 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 156States 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 KiBLeave 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 KiBI'm going to check our differences.
-
@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. -
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.
-
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 KiBStates 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 BPersistent 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 BIt'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 -
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.
-
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 noneGateways (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 192.168.56.1 192.168.56.1I 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 KiBStates 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 BEnter 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 KiBStates 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 KiBBut 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 192.168.56.1 192.168.56.1Then, 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 KiBand the SSH client is like frozen until I leave Persistent CARP Maintenance Mode and pfsense1 recovers the master role.
-
-
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.
-
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 noneI'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 OKI 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,
Zsoltedit: typos, some clarification added
-
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.
-
Hi again,
I've just spotted that 2.3.5 is out, so I've done a fresh install again and tried to simplify the setup as much as possible to reproduce this issue.
I omitted a few irrelevant steps (DNS config, setting up GW groups, etc) from the procedure described in my previous post which resulted this:- 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 - Set up HA for syncing state and configs with relevant FW rules)
- Set up CARP IPs (WAN1-VIP, WAN2-VIP, LAN-VIP) with appropriate netmask
- change Outbound NAT Mode to Manual
- remove auto-created SYNC interface related rules
- change NAT address to WAN1-VIP on rules with interface WAN1
- change NAT address to WAN2-VIP on rules with interface WAN2
- the actual outbound NAT rules are (description removed)
Intf Source SPrt Dst DPrt NAT Address NATPrt Static Port WAN1 127.0.0.0/8 * * 500 WAN1 addres * KeepSrcStatic WAN1 127.0.0.0/8 * * * WAN1 addres * RandomizeSrcPort WAN2 127.0.0.0/8 * * 500 WAN2 address * KeepSrcStatic WAN2 127.0.0.0/8 * * * WAN2 address * RandomizeSrcPort WAN1 192.168.25.0/24 * * 500 WAN1 address * KeepSrcStatic WAN1 192.168.25.0/24 * * * 213.XX.YY.8 * RandomizeSrcPort WAN2 192.168.25.0/24 * * 500 WAN2 address * KeepSrcStatic WAN2 192.168.25.0/24 * * * 192.168.0.10 * RandomizeSrcPort- create FW rule in LAN with routing ssh traffic via WAN2:
Protocol Src Prt Dst Prt Gateway Queue IPv4 TCP * * * 22 WAN2 noneThe result is the same as described in my previous post.
defGW policy route SSH session after fail-over WAN1 disabled OK WAN1 GW:WAN1GW OK WAN1 GW:WAN2GW Freezes WAN2 disabled OK WAN2 GW:WAN1GW Freezes WAN2 GW:WAN2GW OKI've tested with opening an ssh session to an external host.
In case the FW rule directs the outgoing LAN ssh traffic via a gateway other than the default gateway, then after the HA fails over to the secondary node the open ssh session freezes.
I can open a new ssh session via the new master (secondary node), but moving the VIP back to the primary node this newly opened ssh session freezes and the previously opened session starts responding again.
I still see in tcpdump, that when the ssh session freezes, the traffic leaves the firewall on default GW's WAN interface with the other WAN interface's VIP address.Actually I haven't found much related posts where HA, multi-WAN and policy based routing is involved.
We might miss something obvious, but actually I haven't found much related posts/tutorials/howtos where HA, multi-WAN and policy based routing is involved.edit: remove duplicated part
-
WAN2 192.168.25.0/24 * * * 192.168.0.10 * RandomizeSrcPort
What is in front of that? If stateful, it will need a new state too.
-
Hi Derelict,
Tanks for your reply, but I am afraid I do not understand your question.
192.168.25.0/24 is my LAN subnet.
I've opened the ssh session from this subnet to an external host.
The quoted NAT rule supposed to do the outbound NAT on WAN2 arriving from my LAN.I'm attaching the actual config.
HTH,
Zsolt
