Good day!
There is a problem in the ACI fabric with horizontal interaction between endpoints in one bridge domain, but in different EPGs, between which there are the necessary contracts. Conditionally, endpoints (virtual machines in VMware. There is vmm integration in the fabric with VMware) in EPG-1, when accessing endpoints in EPG-2, after a while lose connection with them. The EPG-1 endpoints have the Linux OS installed and when accessing the EPG-2 endpoints is unsuccessful, we receive the message "no route to host", and in the arp records we see "incomplite" in the mac-address field (the mac-address of the bridge domain should be displayed, since the endpoints belong to EPG-2).
If you reconnect the network adapter on the problematic VM in EPG-1, the problem goes away for a while.
APIC version 5.2(3e)
We are inclined to believe that there is most likely a problem with L2 flooding. The current flooding configurations on Bridge Domain are made based on the recommendations for connecting the HPE Synergy blade cluster. For the correct operation of the HPE Synergy basket with its VC switch in tunnel mode, we have the Flood in Encapsulation parameter configured within the bridge domain.
In the current BD configuration, this feature cannot be disabled " For L2 Unknown Unicast is "Flood", ARP Flooding has to be set to true."
