Network Watchdog

Posted to the developer's mailing list on July 11th, 2005 by David Young.

I have added a "network watchdog" to the CUWiN station software. The way it works is that every time hslsd receives a Hello on the wireless interface, it runs a shell script, /usr/sbin/tickle, to "tickle" a shell "watchdog" daemon. The "tickling" is rate-limited to once every 15 seconds or more.

As long as the shell daemon, /usr/sbin/hellowdog, is getting tickled at least once every 5 minutes, it lays dormant. (I realize I'm mixing metaphors here.) If 5 minutes do elapse without a tickle, hellowdog wakes up and tries to restart hslsd. If that doesn't work (because, for example, hslsd is already running!), then it puts wdogctl to sleep. wdogctl is in charge of tickling the hardware watchdog timer. At most 32 seconds after wdogctl is put to sleep, the system will reboot.

In theory, if the wireless adapter or hslsd goes "out to lunch," the network watchdog will bring the router back onto the network in 5 minutes plus 32 seconds plus however long it takes for the node to reboot.