While trying to run metallb operator and creating LB in hypershift we are loosing connectivity from external host to metallb vip after vip is moving to worker. This is caused by failure in reaching nodeport service through the worker.
Example:
nc -zv 10.6.135.6 30332
Ncat: Version 7.92 ( https://nmap.org/ncat )
Ncat: TIMEOUT. 
In OCP dpu-host mode runs with disable-forwarding (default for all modes) that adds FORWARD DROP iptable rule and set 
 sysctl -w net.ipv4.ip_forward=0
sysctl -w net.ipv6.conf.all.forwarding=0

This disables port forwarding that is required for offloading to work

    Create cluster with DPF operator, set worker to dpu-host

    1. nc any to any nodeport on the one of the dpu hosts
    2.
    3.
    

    nc will fail with timeout

    Success