What I am trying to understand is HOW watchdog reboots the system.
If the device becomes unresponsive, then it might be due to really severe reason (like kernel being stuck in some loop) and in such case it's impossible to perform proper shutdown. I guess it's not up to watchdog routine to discover the severity of unresponsiveness ... watchdog reboot is the last possibility to get device responsive again after all.
But then, one really wonders if the system was completely stuck after all before watchdog started reboot ... if you'd have logs saved to a USB storage (remote syslog probably doesn't help here), you'd see something like this:
Oct/07/2018 15:52:34 watchdog,error,critical watchdog cannot ping address 192.168.42.11, rebooting
Oct/07/2018 15:52:48 missed 21 messages while could not open log file
The second line is there because there were 21 log messages after devices started to boot before USB storage got available again. What I'd really like is that those messages are actually flushed to the disk log (they are available in memory) instead of notice that there were that many messages ...