Problem with TAS and gptp

[GiovanniTortia]

Hi, i'm working on a python script which can configure multiple TSN protocols automatically based on a set of parameters, and i want to configure gPTP together with TAS. I'm currently on Inet 4.5.4 (i didn't switch to 4.6 as i already developed a bunch of modules which i didn't want to have to adapt to the newer version), and i keep running in the same problem: the script creates time intervals for the gates with a margin error based on their drift rate, so that the gptp frames (in theory) always find the gates open and don't have to queue, making the correction more accurate. Initially, this works flawlessly, however, as time goes on, there inevitably comes a point where gptp frames get stuck in a queue, messing with the synchronization of all other nodes, and i get the error:
Cannot schedule message (inet::ClockEvent)selfMsgPdelay to the past
I think the problem may be with the gptp correction causing jumps in the clocks, but i'm not sure. I tried following the example of the gptpandtas showcase, but i still don't seem to get it right (and i have no idea why).
Here's an example .ini generated by my script, stripped of all purely numerical parts, where i try to configure multiple gptp domains, with the idea that the grandmaster of domain i syncs to domain j only if i>j (hot standby)

[General]
network = Demo
**.eth[*].bitrate = 25Gbps
**.hasUdp = false
**.hasTcp = false
**.hasIpv4 = false
**.numTrafficClasses = 3
**.macTable.agingTime = 9223372s

# ----------------- clocks -----------------------
*.*.clock.typename = "MultiClock"
*.t0.clock.clock[*].oscillator.driftRate = 13ppm
<oscillators of all clocks in a node have the same driftRate to simulate multiple virtual clocks with a single physical oscillator>

#------------------ GPTP -------------------------
*.*.hasTimeSynchronization = true
*.*.gptp.typename = "MultiDomainGptp"
*.*.gptp.domain[*].syncInterval = 125000us
*.*.gptp.**.pDelayReqProcessingTime = 5us

*.*.gptp.domain[0].clockModule = "^.^.clock.clock[0]"
*.*.clock.clock[0].referenceClock = "t0_0.clock.clock[0]"
*.*.gptp.domain[1].clockModule = "^.^.clock.clock[1]"
*.*.clock.clock[1].referenceClock = "t1_0.clock.clock[1]"
*.*.gptp.domain[2].clockModule = "^.^.clock.clock[2]"
*.*.clock.clock[2].referenceClock = "t2_0.clock.clock[2]"

*.t0_0.gptp.numDomains = 1
*.t0_0.clock.numClocks = 1
*.t0_0.gptp.domain[0].gptpNodeType = "MASTER_NODE"
*.t0_0.gptp.domain[0].slavePort = ""
<domain 1 spanning tree>

*.t1_0.gptp.numDomains = 2
*.t1_0.clock.numClocks = 2
*.t1_0.gptp.domain[1].gptpNodeType = "MASTER_NODE"
*.t1_0.gptp.domain[1].slavePort = ""
<domain 2 spanning tree>

*.t2_0.gptp.numDomains = 3
*.t2_0.clock.numClocks = 3
*.t2_0.gptp.domain[2].gptpNodeType = "MASTER_NODE"
*.t2_0.gptp.domain[2].slavePort = ""
<domain 3 spanning tree>

*.*.gptp.numDomains = 3
*.*.clock.numClocks = 3

**.gptp.domain[*].syncInterval = 125000us
**.gptp.domain[*].pdelayInterval = 1000000us
**.gptp.domain[0].syncInitialOffset = 11106ns
**.gptp.domain[0].pdelayInitialOffset = 0ns
**.gptp.domain[1].syncInitialOffset = 52824ns
**.gptp.domain[1].pdelayInitialOffset = 41718ns
**.gptp.domain[2].syncInitialOffset = 33504ns
**.gptp.domain[2].pdelayInitialOffset = 22398ns

#---------------- TAS ---------------------
*.*.hasEgressTrafficShaping = true
**.transmissionGate[*].clockModule = "^.^.^.^.clock"

**.transmissionGate[*].scheduleForAbsoluteTime = true
**.clock[*].defaultOverdueClockEventHandlingMode = "execute"
<gate schedules>

[levy]

Your guess may be right, this may be related to how the clock jumps around during synchronization. This may interfere with the pdelay timer being scheduled. Since gPTP has been pretty much rewritten in 4.6, it's very unlikely that we are going to hunt this down. It would be better for you to switch to the new version, because if any issue arises in the latest version, then we are happy to help.

Please take a look OMNeT++ 6.4.0 AI & LLM Technology Preview 2 at https://omnetpp.org/ because it may help you diagnose this kind of problems. See here for example: https://omnetpp.org/documentation/ai-demos/tsn/showcase