Multi-WAN gateway failover not switching back to tier 1 gw after back online

Derelict

I would change the monitor IP in the WAN2CABLEGW to 8.8.8.8 or anything else that responds reliably and see if things improve. You can't expect any multi-WAN routing solution to perform with any semblance of continuity with flapping like that.

MrD

Actually, the monitor IP is default so it is the gateway of each wan.

I will give a try for a one like google (8.8.8.8) but the problem is not the monitor ip or the trigger, the problem is that when the deffect wan is back to normal (ping to monitor ip is better quality) the system do not switch back. I mean If I log in pfsense (hours after the problem), and watch my "gateway groups" status, they are all in green (same for gateways) but the system do not switch back to the favorite gateway.

@Derelict:

I would change the monitor IP in the WAN2CABLEGW to 8.8.8.8 or anything else that responds reliably and see if things improve. You can't expect any multi-WAN routing solution to perform with any semblance of continuity with flapping like that.

Derelict

No, the problem is your gateway is flapping about every minute due to packet loss to your monitor IP. If that was in my multi-wan group I would disable it until it was fixed. If that's "just the way it is" you will need to increase your monitoring threshholds and consider it up when it sucks like that.

MrD

OK my gateway has "troubles" for few minutes a day (not all the time).

That's precisely why I want a failover.

And this do not explain why the system do not go back to it's first wan after the first wan is seen by ths system in green. (hours after!)

If the goal of a failover is to work on connections that never have tropubles, it's non sense to me…

@jahonix:

You have one or two Gateway Groups defined? The one with time stamp 02-25-44.

What you call "WANGROUP" is easier to handle when called " PPPoE 2 UPC"
Now you need an additional "UPC 2 PPPoE" group with reversed tiers.
Add another firewall rule for that one as well and it should work.

And start with setting both "Trigger levels" to "Member Down".

jmonline

I am pretty sure this is exactly related to my issue and my most recent detailed post here:

https://forum.pfsense.org/index.php?topic=86851.msg632594#msg632594

MrD

Hello JM,

Globally it seems to be the same problem reported by most people writing in your post. I've seen a bug report but dev team consider it is not a bug but misconfiguration without explaining where is the misconfiguration… strange

@jmonline:

I am pretty sure this is exactly related to my issue and my most recent detailed post here:

https://forum.pfsense.org/index.php?topic=86851.msg632594#msg632594

Derelict

It is not a bug.

A setting that kills all states on a Tier X interface when a Tier < X interface returns to service would be a feature request.

I did not see one for this on redmine.pfsense.org.

MrD

After many readings on this subject it is the first time I read that this is normal and this is a feature request. I've read that this was the result of missconfiguration meaning that connection should go back to what it was before failover…

For example:
https://redmine.pfsense.org/issues/5090

Chris Buechler
…
I went through and re-tested multi-WAN in general on 2.2.5 (which is the same as 2.2.4 in that regard) and it fails over and back as it should just fine every time.
...
There may be some edge case but nothing here to suggest what that might be.

BUT fiew lines later, it goes another way

Chris Buechler
…
that's how it's supposed to work at this point. Sounds like you want state killing on failback, which doesn't exist at this time. feature #855 covers that

https://redmine.pfsense.org/issues/855

So the final answer is FAILOVER DO NOT GO BACK TO INITIAL STATE
This is suprising but knowing this, I stop loosing time trying different config options…

@Derelict:

It is not a bug.

A setting that kills all states on a Tier X interface when a Tier < X interface returns to service would be a feature request.

I did not see one for this on redmine.pfsense.org.

jmonline

@Derelict:

It is not a bug.

A setting that kills all states on a Tier X interface when a Tier < X interface returns to service would be a feature request.

I did not see one for this on redmine.pfsense.org.

Right, but if it's not a bug, then how do you get traffic to go back over the original interface when it returns online.

Killing the states does not always work.

I have also been able to test that a brand new device connected to the network, will still route in the same way (onto the failover interface) even if the primary wan was back online BEFORE the new device was connected.

I have also been testing this in a virtual environment and can replicate the issue. Although it is not always the same. Sometimes new states will follow the correct route (back over the primary wan) and other times they will get stuck on the backup wan. It is not consistent which doesn't make sense.

MrD

Let's be clear, to me it is a bug. But if they say no, I have no choice.

Actually, I reset all states and sometimes I change the firewall rule (time consuming!!!) If better proposition I'm interested.

Derelict

Killing the states does not always work.

Please demonstrate with evidence.

Derelict

@MrD:

@Derelict:

I did not see one for this on redmine.pfsense.org.

https://redmine.pfsense.org/issues/855

So the final answer is FAILOVER DO NOT GO BACK TO INITIAL STATE
This is suprising but knowing this, I stop loosing time trying different config options…

There. Feature #855. My redmine searching could obviously use a tuneup.

jmonline

@Derelict:

Please demonstrate with evidence.

Ok so in very basic terms since I already have quite a lot of information on this post here https://forum.pfsense.org/index.php?topic=86851.msg632594#msg632594

The connection has failed over to the backup WAN when the primary WAN has gone down. (Failover has worked as expected)
The primary WAN has come back up (Status > Gateways confirms this is up/online).
The states (VoIP sessions for phones) are still showing in the state table 12hrs later going over the backup WAN.
No new or refreshed sessions from the phones go over the primary connection.
Current state table (filtered by the phone with the IP of 10.10.30.55) looks like this:

WAN_EFM udp 135.196.xxx.xxx:41809 (10.10.30.55:49679) -> 185.83.xxx.xxx:5060 MULTIPLE:MULTIPLE 201.589 K / 102.513 K 125.60 MiB / 39.52 MiB
30VOICELAN udp 185.83.xxx.xxx:5060 <- 10.10.30.55:49679 MULTIPLE:MULTIPLE 99.293 K / 99.502 K 61.87 MiB / 38.35 MiB

To clarify:
WAN_EFM - is the backup WAN connection
30VOICELAN - is the LAN network for the phones
135.196.xxx.xxx - is the IP of my backup WAN connection
185.83.xxx.xxx - is the IP of my externally hosted VoIP platform
I have then "Reset the firewall state table"
At this point SOMETIMES the states will clear and obey the correct Gateway fail-over rule and be sent back over the primary WAN.
SOMETIMES they will stay where they are (on the backup WAN)

I can understand the argument that it is a feature request to have the states clear on the re-establishment of the primary wan connection.
However, why have I seen the following…

Primary connection has been down for a length of time and has since come back online.
A brand new device which has never connected to the network (so therefore has no open states) is connected.
This new device states are sent over the backup WAN - even though the primary wan is available
"Reset the firewall state table" and the new device has states over the primary wan (as it should have done when it first connected to the network)

I also ran a test of this in a virtual environment and simulated the primary WAN connection dropping and re-connecting.
I was using a Linux machine as a test client and just running a PING and TRACEROUTE to use as example states on the firewall (eliminating the VoIP aspect).
Sometimes, when you bought the Primary WAN connection back online, a new TRACEROUTE to a different IP address could go over the primary WAN, and other times it would remain over the backup WAN.
I have not been able to prove what causes this - it appears random.

In my mind, if the primary wan connection is reconnected and online, then any NEW state that hits the firewall should always follow the gateway group rule and go over the Tier1 connection.

Why does running a pfctl and targeting the relevant hosts/network not force clearing of the states just for the VoIP devices (without clearing the whole state table)?

Another simple way of putting it….........

If your primary connection goes down for an hour and then comes back online. At what point should your traffic start to reuse that connection again. What if your "backup" connection has a very data usage charge?

Bit of history for you…..

I used to use Draytek equipment for all my client sites, on their old 2830 series of routers, they had the WAN failover options, but the same applied… if the primary went down everything would failover to the backup and then never fail back again when the primary connection returned.

On their newer 2860 series of routers, they added one simple check box labelled "Failback" and it moved your sessions/states back to the correct primary connection when it was available again.

However on the Draytek I never had the issue where a NEW state/session would still go over the backup WAN when the primary was available. If it was a new session it always followed the rules correctly.

I hope that makes sense to some of you :)

Derelict

You still didn't show the Tier 1 being back online and new states still being created on Tier 2. I think if you really take a look at this you will find that is not happening.

And nothing can "move a state" back to the original connection. All you can do is kill the old state and let a new one be established on a reconnection.

jmonline

Ok lets simplify it even more…

Take the traceroute facility in Diagnostics > Traceroute

My primary wan is back online.
I enter a hostname to trace (Google - 8.8.8.8)
I pick the Source Address as 30VOICELAN
I get the following result:
-1 135.196.xxx.xxx 7.671 ms 6.869 ms 7.008 ms
-2 135.196.xxx.xxx 7.016 ms 7.195 ms 7.164 ms
-3 5.57.80.136 7.218 ms 7.199 ms 11.125 ms
-4 216.239.54.243 7.922 ms
-5 216.239.58.95 9.010 ms
-6 8.8.8.8 8.139 ms 8.010 ms 8.626 ms
The first line on the traceroute with the IP starting 135.196 is my backup internet connection. Not my primary.

How is that possible?

The firewall rule on the 30VOICELAN has the Gateway set as the Gateway Group named "DSLFirst".
The Gateway group "DSLFirst" has the (Primary) DSL WAN connection as Tier1 and the (Backup) EFM WAN connection as Tier2.
Status > Gateways shows both gateways online.

Derelict

Show me the states, bro. pfctl -vss

jmonline

@Derelict:

Show me the states, bro. pfctl -vss

Ok so perfect time for a test :) Last night looks like BT did their usual maintenance on the DSL network around 1am so the ADSL line was down for 5mins. This morning I have the following in the states table for the phone on IP 10.10.30.27.

30VOICELAN tcp 185.83.xxx.xxx:5060 <- 10.10.30.27:55778 ESTABLISHED:ESTABLISHED 8.933 K / 10.417 K 3.13 MiB / 3.43 MiB
WAN_EFM tcp 185.3.xxx.xxx:40781 (10.10.30.27:55778) -> 185.83.xxx.xxx:5060 ESTABLISHED:ESTABLISHED 8.933 K / 10.417 K 3.13 MiB / 3.43 MiB

To clarify:
185.83.xxx.xxx is the external VoIP pbx.
185.3.xxx.xxx is the IP of the WAN_EFM (backup) connection.
30VOICELAN is my internal network with a subnet of 10.10.30.0/24

pfctl -vss shows the following:

igb1_vlan30 tcp 185.83.xxx.xxx:5060 <- 10.10.30.27:55778 ESTABLISHED:ESTABLISHED
[1594456643 + 42272] wscale 8 [1007765254 + 183296] wscale 5
age 05:58:04, expires in 119:59:52, 8954:10441 pkts, 3290011:3604569 bytes, rule 119

igb2 tcp 185.3.xxx.xxx:40781 (10.10.30.27:55778) -> 185.83.xxx.xxx:5060 ESTABLISHED:ESTABLISHED
[1007765254 + 183296] wscale 5 [1594456643 + 42272] wscale 8
age 05:58:04, expires in 119:59:52, 8954:10441 pkts, 3290011:3604569 bytes, rule 96

Our of interest, what should be the correct pfctl command to run to force killing these states (so all states on the WAN_EFM connection from the subnet 10.10.30.0/24)?

If I can get a command to successfully kill these states when they get stuck here, I am happy for that as a work around until someone can work out how to automate it. I don't want to be Resetting the whole state table every time since that kills sessions which should be legitimately open.

Thanks

Derelict

You probably want to kill all connections to the PBX. That would be:

pfctl -k 0.0.0.0/0 -k 185.83.xxx.xxx

That will kill everything even phones that are connected out the Tier 1.

You can try just killing one side of the connection that is tied to WAN_EFM with:

pfctl -i igb2 -k 0.0.0.0/0 -k 185.83.xxx.xxx

If, when the phones reconnect, they use the Tier1 connection, great. In my testing they continued to use the other connection so it doesn't look like you can do that.

jmonline

Ok so the following command cleared the sessions stuck on the failover WAN.

pfctl -i igb2 -k 0.0.0.0/0 -k 185.83.xxx.xxx

This is a good step forward since I can now manually force the sessions back when I know they haven't moved on their own.

I presume I may be able to scheduled this via some sort of script to run a specified period of time after the primary connection comes back online…...?

Thanks for help so far Derelict :)

devmaybe

Hello,

I know this is an old thread but I have the same problem and now I am able to reliably reproduce the behavior in a test environment.

If the "primary" WAN is a PPPoE connection and the secondary WAN is a "standard" static or DHCP assigned IP address connection when the primary goes down failover to the secondary work as expected but when the primary comes back up no traffic will flow through it.

In such cases on my production systems I usually edit my default gateway entry in System->Routing.
I uncheck the "Default Gateway" mark, re-check it and then save and apply.
Traffic starts flowing again through the PPPoE connection.

The same always works in my virtual machines test environment too.

I hope this can help in tracking down the source of the problem or at least in finding some solution.

Thanks