Some months ago we upgraded our two node Server 2022 Hyper-V cluster to new servers with 25G ports and upgraded our backbone to two CRS518 which are connected via a bond of the two QSFP28 ports. Each cluster node is connected to both switches via 2 LAN ports (Port 1/2 one one switch, port 3/4 on the other), on each cluster node these 4 ports are configured as "embedded teaming switch". So no bonding on the individual Mikrotik switch. This setup worked fine on the Dell N3048P switches before that were configured as stack.

Even on the stable channel we have RouterOS upgrades on a monthly basis and over the last months I experienced connection problems from VMs on one cluster node afterwards. Not always, but in most cases. The cluster nodes itself seem to work fine, no entries in the event log. All the VMs are on the same virtual switch. After firmware upgrade or sometimes a few days later the VMs can't be reached or even pinged on one node, but I can migrate them to the other node and everything is fine again. So far no fancy protocols involved, just connection with DAC between switches. This problem is solved if I restart the node.
This gets even more weird because the VMs in the server VLAN can ping each other but computers on other VLANs can't. We have Sophos UTM as VM on both cluster nodes in HA configuration which are used as routers between our VLANs. The node that is still working fine after firmware update is hosting the active UTM VM.

I am new to the Mikrotik world so it may be a misconfiguration. Any ideas what causes this problems?