UCM/KNX race condition

[ekarak]

Hi all,

There looks to be a race condition within Comfort (7.0.92) and/or UCM/KNX (7.0.30). If I send two consecutive telegrams to switch on and off some load, then Comfort starts emitting an endless stream of on-off telegrams and I need to reset the unit. I\'ve noticed this today when I heard that relay clicking on and off twice per second.

In the attached screenshot a Zennio Z38i touchscreen (phys. addr. = 1.1.11) is sending two telegrams, an ON and an OFF, after which Comfort starts sending an endless stream of on and off telegrams. This looks to be a race condition of sorts.

The group address is mapped to counter #139 (\'KNX-3-0-3\')

Code:

     <Counter Number=\"139\" Name=\"KNX-3-0-3\" />
...
     <UcmKnxToComfort Type=\"OneBitSwitching\" GroupAddress=\"6147\" ItemNumber=\"139\" />

The relevant counter is only mapped to a control menu item, and it has two simple responses to turn the load on or off as follows:

Code:

    <Response Fixed=\"false\" Number=\"109\" Description=\"anakykloforia ZNX\" Name=\"KNX_3_0_3_ON\" ActionCodeString=\"197,18,2,24,3,1, 255\">
      <ActionSource>
        <string>Knx On UCM02_KNX 6147 ! 3/0/3</string>
      </ActionSource>
      <Errors />
    </Response>
    <Response Fixed=\"false\" Number=\"110\" Description=\"anakykloforia ZNX OFF\" Name=\"KNX_3_0_3_OFF\" ActionCodeString=\"197,18,2,24,3,0, 255\">
      <ActionSource>
        <string>Knx Off UCM02_KNX 6147 ! 3/0/3</string>
      </ActionSource>
      <Errors />
    </Response>
 

This is difficult to reproduce (it doesn\'t happen always), however whenever it happens its very annoying as it clogs down the bus and it can cause premature wear to the relays.

[ekarak]

Important to note: I\'ve also logged this happening today on another KNX group address that I only have a counter mapped, and nothing else (ie no control menu item or reference by any response script). So I guess this bug got to do with Comfort\'s internal representation of the counter?

[ekarak]

Here is the 2nd example counter\'s \"referenced by\" screenshot where nothing else is mapped to the counter, hence no response could have triggered the endless toggling (on/off/on/off/..).

[slychiu]

I believe we have encountered this race condition in other ucms eg Zwave

When you send a command to turn on a KNX address, the counter is changed to on
The UCM/KNX sends the counter change to Comfort which broadcasts the counter changed to 255 after a slight delay
But before this, you send an Off command to the same address which turns off the KNX device. This causes the counters to report Off, but in the meantime the previous broadcast on turns on the device, thus it causes  a race condition until it settles down

There is already a fix for this (in ucm common version 7.071) but the KNX firmware was not released yet. We shall confirm it next week

[ekarak]

As always, stellar support! Thank you!