Why fg command within an infinite while loop works like a "keep alive" watchdog?

Question

Recently, I've run into this implementation of a handicraft "keepalive" watchdog for a custom HTTP server:

|
|   keepalive.sh
|   init.sh
|_  main.py

There are three files. The first one being keepalive.sh: it is run as root from rc.local when the system starts up.

# keepalive.sh

while true
do
    if curl -s "localhost:8080/alive.txt" | grep -q "ALIVE"; then
        echo "UP
    else
        echo "Restarting..."
        pkill -u www-data -SIGKILL
    fi
    sleep 30
done

The second one is init.sh which is run as www-data user.

# init.sh

set -m
/usr/bin/python3 /var/www/main.py &
while true; do fg; sleep 1; done

The third one is the custom server itself main.py which needs to be kept alive.

from server import Server

HOST = '0.0.0.0'
PORT = 8080

if __name__ == "__main__":
    Server(HOST,PORT).listen()

I don't quite understand how init.sh works. With -m flag we enable job control in order to use fg from script file. What is the purpose of sending main.py to background and then doing fg in an infinite while? I understand that & is needed here to get to the loop after the server was started but why exactly fg (and not just true, for example) is used?

The other thing I do not understand is why init.sh restarts when it is killed with -SIGKILL. The Python script is launched in a separate process group with init.sh's PID as its own PPID. main.py gets killed when his parent (init.sh) gets killed, I'm all right with that. For me the question is why init.sh keeps restarting main.py (and also itself) after it's being killed with pkill.

Thanks!