The attached patches implement a service watchdog based on timers and a
custom SIGRT signal (of which there are 30/32 available to use) and
removes the ping based solution.
In case a child gets stuck in a tevent loop the timer will eventually
kill it (in 30 sec. by default) and the monitor will catch the child has
terminated (via SGICHLD) and restart it. This makes the ping based
infrastructure obsolet so the monitor now stops setting it up.
In order to avoid changes to the dbus interface the ping method is still
in places for responders/providers, but simply never invoked.
Resolves:
https://fedorahosted.org/sssd/ticket/2921
--
Simo Sorce * Red Hat, Inc * New York