That’s about TCP messages getting split across more than one packet. I think now it’s unlikely, and in any case the binding should handle it. (as for the mystery slave, who knows, but it’s really unlikely to get the smaller poll message in fragments)
message='I/O exception: EOFException null'
A bit of googly about that in other contexts suggests it may be about the remote end closing TCP connection unexpectedly.
Your modbus TCP is all default parameters, which mostly works for most devices, but can sometimes need tweaking.
I would recommend you change to keep the connection open, to reduce the open/close churn at your slave. It might just be slow dealing with previous disconnects. And/or a little time between reconnects may help.
See -