Keep getting disconnected from the bus

[masterd89]

Ciao Massimo,
I've been using KNX Ultimate for quite a while, and sometimes i noticed telegram being dropped randomly ( i.e. light does not turn on or off )
I set up the watchdog node, and so i started monitoring the bus and receive notifications when events happened.

So far i noticed i get a lot of (1-2 per hour)

• Send write Error: The socket is not connected. Unable to access the KNX BUS
• Disconnected by error Invalid Connection ID
  And sometimes
• Watchdog elapsed with no response.

As already mentioned on other issues, i tried the following:

• increase the gateway delay btw telegram to 200ms
• set Auto as IP Protocol
• set explicitly eth0 as interface

which helped reducing the frequency of the events, but did not solve the issue.

my setup is the following:

• latest NR and KNX Ultimate version ( dockerized )
• host mode for the network
• spacelogix KNX-ip interface
• Ubiquiti switches

Not sure if that's related, but another issue i noticed is that the tunneling interface GroupAddress of the KNX-IP device does not correspond to the one i set in the Gateway setup. It the node reserving the specific address or is it taking the first free tunnel it finds? Sorry if i'm getting that wrong, i'm not expert on the tunneling mechanism .

Grazie,
Davide

[Supergiovane]

Hi
increase the gateway delay btw telegram to 200ms please revert to 25ms. 200ms is too much and the spacelogic will give you a timeout error.
Are you dockerizing on raspberryPi? If yes, what model of RaspberryPi?

[masterd89]

Ok thanks,
reverted to 25ms.

Yes, i am on a RPi. I was using an old RPi2 iniitally, then i moved to a RPi4 when i started seeing the Watchdog sending out errors.

[Supergiovane]

Hi
please be aware, that dockerizing things on a Rpi, even Rpi4 could be a pain in the ass. The KNX protocol is much based on telegram's timing. By definition, the operating systems aren't "realtime"; the rPi doesn't even have an RTC Clock.
For this reason, i recommend to install node-red on raspberry, without anything else (the node-red website has a script, that installs everything you need).
As instance, i tell you what i've done at home: Proxmox with a VM running node-red. On that, i run exclusively KNX-Ultimate and some other nodes i've developed. On a second VM, i've installed node-red and all other third party nodes.
So even if the second VM crashes, the main VM running KNX/Philips HUE will still be alive and rock solid.

A thing you can try, is to enable the options indicated by the red arrows, and see if it helps:

[masterd89]

Oh i see. i did know that!

I can move the container on a Proxmox VM on another server, but ultimately my goal would be to have an "independent device", which is separated from the "services" server ( to keep the house up and running no matter what ). In my view, that would be a more cost effective ( and more capable tbh ) alternative to a Gira X1, to name one.
I will for sure install NR directly on the RPi as first step, but if in your experience you found the system not adequate for the task, i would rather move to a different solution. Would a N100 mini pc be better for the task? Or another ARM sbc that you can recommend ( i always find RPi alternative lack OS support though) ?

[Supergiovane]

Hi
Rpi4 is fine for the task. Remember to buy an Industrial SD Card (the standard SD Cards will fail at certain point in time). You'll find that on Amazon as well.
;-)

[Supergiovane]

PS: There are also DIN RAIL enclosures for the RPi, so you can put it in the standard italian DIN cabinets.

[masterd89]

Hello,
Sorry to bother you again.
I am still seeing the errors even after having installed node red on the RPi. Any other advice?

[masterd89]

Adding a log around the time the last error occurred:
https://pastebin.com/ue345d7g

[Supergiovane]

Hi
Have you selected “suppress ack request..” as the above schreenshot?
Please also restart the spacelogic

[masterd89]

Have you selected “suppress ack request..” as the above schreenshot?

Yes and got the errors this morning. Then i tried also without and got errors as well. Now i reverted the config as indicated above ( both options selected )

Please also restart the spacelogic

Ok done. I'll let it run for a while and then report here.

However, i am still seeing the telegram's source being 1.1.255, even though i set in the gateway configuration the address 1.1.249. Is it expected/related ?

[Supergiovane]

Hi
No, the source address must be 1.1.249.
Do you have other third party KNX nodes installed?

[masterd89]

not in nodered. i have Homeassistant KNX integration, but that is bound to address 1.1.252 ( and i correctly see its source in ETS monitor )

[Supergiovane]

That should be not an issue. I'm running also HA and other KNX Ultimate instances in the VM's.
Can you please paste here, the debug info gathered by pressing this button?

[masterd89]

Sure here it is ( i removed the GA imported from the project ) :

-> id: 2dbbb489898e8413
-> type: knxUltimate-config
-> info: undefined
-> users: [object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object]
-> _config: [object Object]
-> host: 10.89.16.64
-> port: 3671
-> physAddr: 1.1.249
-> hostProtocol: Auto
-> suppressACKRequest: true
-> csv: "Group name" "Address" "Central" "Unfiltered" "Description" "DatapointType" "Security"
********** REMOVED ***************

-> KNXEthInterface: eth0
-> KNXEthInterfaceManuallyInput: eth0
-> stopETSImportIfNoDatapoint: skip
-> loglevel: debug
-> name: KNX Gateway
-> localEchoInTunneling: true
-> delaybetweentelegrams: 25
-> ignoreTelegramsWithRepeatedFlag: true
-> keyringFileXML: undefined
-> knxSecureSelected: undefined
-> autoReconnect: yes
-> label: function () {
return typeof this.name === undefined ? (this.host + ":" + this.port) : this.name + " " + (this.host + ":" + this.port);
}
-> _def: [object Object]
-> : function(){var e=Array.prototype.slice.call(arguments,0),t=e[0],o=(-1===e[0].indexOf(":")&&(e[0]=i+":"+e[0]),RED..apply(null,e));return o=o===e[0]?t:o}
-> valid: true
-> validationErrors:
-> dirty: true
-> credentials: [object Object]
-> infoEditor: [object Object]

[Supergiovane]

Hi
are you sure about the IP Address "10.89.16.64"? I'ts the first time i see the 10.x subnet.
Please set the protocol to "Tunnel UDP".
If the Schneider is also a KNX router, set the IP of KNX Ultimate to "224.0.23.12" and the protocol to "Auto" and restart node-red.

There is no other things that comes in my mind, but if the log says
sendKNXTelegramToKNXEngine: the KNXEngine is busy or is waiting for a telegram ACK with seqNumner 133. Delay handling queue.
it means that the schneider is not sending the ACKnowledge of the telegrams.
This should be fixed by enablind "Suppress ACK...." and restart node-red.
Had you restart node-red after setting such option?

[masterd89]

Hello,
yes the ip address is correct! i have a 10.XX.XX.XX private network.
the SE is an ip interface, not a router.

i enabled the flags back after posting the log. here's a new log where i saw the usual error at 3:00:01pm https://pastebin.com/qHcVSMJi

i restarted NR after the change.

[Supergiovane]

It’s not connecting.
Let’s try the dummy things:

Replace the ethernet cable
If unsuccesfull:
Temporary connect the raspberry Ethernet cable, to the spacelogic and see what happens

[masterd89]

Ciao Massimo,

unfortunately the cable did not solve the problem.
i have now connected the RPi on the same switch where the Spacelogic is: previously there was one extra hop between the two, let's see if that was causing the issue.

[Supergiovane]

Hi
The udp packet’s TTL is high. It shouldn’t be blocked by the switch.
If the switch, however, blocks udp storms, the packets can be blocked (i’ve some Unifi Switches doing that).

[Supergiovane]

Hi
Just to let you see, how many VM i'm running and how many nodes (about 300 every VM), across one KNX/IP Gateway with 5 tunnel and one KNX/IP Router.

Registrazione.schermo.2025-08-05.alle.08.44.27.mov

[masterd89]

Thanks for the hint,Is there a setting in the UniFi switch to prevent the blockage of UDP packets?Il giorno 5 ago 2025, alle ore 08:32, Massimo Saccani ***@***.***> ha scritto: Hi The udp packet’s TTL is high. It shouldn’t be blocked by the switch. If the switch, however, blocks udp storms, the packets can be blocked (i’ve some Unifi Switches doing that). —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>

[Supergiovane]

Hi
this is how my settings appears:

GLOBAL NETWORK CONFIG

CLICK ON YOUR LAN, THEN IT SHOULD BE AS MINE:

[masterd89]

Thanks for that.
My settings are basically the same. I just had Jumbo Frames disabled, but i don't know about that...
It's still giving errors, later today i'll try connect the RPi directly to the Spacelogic and monitor via wifi.

The only thing left to debug is the spacelogic itself... i can try to factory reset it ( i am still seeing the source address 1.1.255 or 1.1.254, sometimes it changes when i disconnect and reconnect the spacelogic )

[masterd89]

And be sure, i know KNXUltimate is rock solid, i just want to understand where the issue is on my system at this point !

[Supergiovane]

Hi
if you have a spare KNX/IP Interface laying around, you can try it. I've always a spare KNX/IP Interface, just in case the main one fails.
Or maybe you can check, if there is a new firmware for the spacelogic.
But... if HA is running fine without errors, there must be something else causing the issue...
What about installing node-red as HA plugin and try installing KNX-Ultimate on that? I've done this way and it works.

[masterd89]

Ciao,
As per your suggestion, yesterday i installed nodered as homeassistant plugin ( which is a proxmox VM in a separate server) and let it run.
I only instantiated a watchdog node, and i am getting no errors whatsoever from that install ( i am monitoring both instances with a notification node, so i have the real time error feed ). However, i still see the source address being different from what i set in KNXUltimate gateway config

Moreover, i added a ping check every 10s on the RPi instance, to verify if the link is broken or only the connection. Maybe i expected to see the ping fail when the watchdog gave error, but to my surprise that is not the case, and i never saw ping error messages.

At this point, i would say the Spacelogic, the LAN and KNXUltimate itself are verified working as expected. I noticed i had some wrong settings in the eth0 config, lets see if fixing that helps.

Next step is to move all the NR install in a VM so to check if the whole thing is functional, and thus this leaves the RPi as only culprit.

Just one question: when moving the nodered install folder from docker to my homedir, i had to reconfigure most of the configuration nodes. Is there a way to avoid that when relocating to the VM?

[Supergiovane]

Hi

Just one question: when moving the nodered install folder from docker to my homedir, i had to reconfigure most of the configuration nodes. Is there a way to avoid that when relocating to the VM?

Yes, but it takes longer than re-insert all passwords in the config nodes! You basically need to have a VM with the exact name of your RPi, as well as same username and password. Then copy the entire .node-red folder, including hidden files.

Do you have turned off the WiFi on your Raspberry? Try with WiFi turned off (don't leave it active, even if not connected, turn it off). The operating system couldn't route the incoming UPD telegram correctly, even if you have selected a specific ethernet card on the KNX-Ultimate's config node.

[masterd89]

Hi Massimo,
i think i reached some solid conclusions.

• Cloning the nodered folder on the server and running the project there has not produced a single error so far since yesterday. i am 100% confident it will continue to work properly.
• Having excluded every other sources, the raspberry seems to be the issue.
• i found multiple sources lamenting poor RPis performance when handling UDP packets. At this point i can assume the hardware is the limitation. in my experience, it got slightly worse over time ( now i have lights that are managed by nodered that at least once a day fails to turn on or off, i don't recall such a bad situation in the past).
• Maybe using a better SD card or only running nodered alone can improve the situation, but in general i don't think its a good idea to use a raspberry for this purpose.
  I think i will get a mini pc or something and run nodered there: it's unfortunate though, the RPi are really good in terms of power efficiency and form factor...

Regarding the incorrect GA seen on the ETS monitor: I'm not sure if this is specific to Spacelogic, but apparently, when tunneling with TCP (as ETS or Home Assistant do), the group address is correctly reserved on the interface. With UDP communication, the interface seems to automatically use the first free channel. Is there any chance that TCP connection will be implemented in KNXUltimate?

[Supergiovane]

No, the TCP connection is still used only for KNX Secure (that it's on my todo list).

[masterd89]

A mini PC could be also fine, but if you have the chance to use a VM, you'll have for sure a server with ECC RAM and industrial motherboard, so why don't use that?

Redundancy being the main reason. if the server is down or broken, i would like that the lights/shaders/rain sensors/ ecc to keep working, and i can do the repairs without the pressure of not having a functional electrical system in the house !

[Supergiovane]

Yes but... what if the miniPC fails? You could install proxmox on the miniPC as well as in your server.
If your server goes down for whatever reason, you could simply import the Proxmox server's VM backup in the miniPC Proxmox instance and you're done.
You can for sure make also the opposite (miniPC with proxmox as main home server and, it this fails, import the VM into your server's proxmox instance).

I'm currently done the first (main node-red on server's proxmox VM and, if server fails, import the backup into an intel apple mac mini running proxmox).

[masterd89]

Exactly! In the end, i still need a main server and a "secondary" device as you said.
So far, the latter has been the raspberry, with everything dockerized and backupped, so i could quickly spin up a VM in the main server and import all the containers there in case of a failure. With the mini pc, as you suggested i would have proxmox and the VM directly there, and migrate with 2 clicks if anything goes south.
I did not mention that the server is located in the basement, which is separated from the main part of the home. The server runs homeassistant, NVR, and other non prioritary services. Sometimes i need to power it off for some hours for maintenance on the electrical system there, and i would like the main part of the house not to be impacted.
I don't want to experience my wife's wrath if it starts raining and the roof window does not close because the server is down !!!!

[masterd89]

Thanks for the suggestion anyway, i had not realized that having proxmox also on the "Home server" would save me a lot of hassles and reduce dramatically the downtime in case of a failure. With the rpi it wasn't possible, i'll make sure to get a device which can support it for its replacement

[Supergiovane]

Ultima cosa: iscriviti al mio canale https://www.youtube.com/@maxsupervibe
Prossimamente caricherò video interessanti.
Ciao!

[Noschvie]

@masterd89 have a look to the Proxmox VE cluster manager

[masterd89]

Nice! I knew it existed, i missed the part where each node can control the cluster, which is great if one node is down ( and it happened to be the "master" as in other similar architectures).
Actually i have a second device used for NAS and backups, which is proxmox based as well, so this will definetly come handy!

Thanks for the advice

[Supergiovane]

I think you need at least 3 nodes (3 Proxmox PC) to be able to do the clustering, otherwise you should hack some proxmox config.