A Watchdog timer can be used by the hypervisor to verify if an Instance is still alive. If not, for example due to a kernel panic the HV can reset the Instance so that it boots again.
Inside the Instance the 'watchdog' daemon has to run to provide this. If the Watchdog is not running the HV can't verify if the Instance has crashed.
This is supported by Libvirt and Qemu and can be configured in the XML: https://libvirt.org/formatdomain.html#elementsWatchdog