If Anaconda starts and completes its %pre check-in but reboots before the %post check-in, the system will get into an infinite reboot loop and the external watchdog will never be triggered (instead it is perpetually extended by the %pre check-in).
+++ This bug was initially created as a clone of Bug #953543 +++
--- Additional comment from Dan Callaghan on 2013-04-22 10:35:29 EST ---
I'm guessing the panic loop is because we now extend the watchdog for 10 minutes after panic, to allow for kdump or other post-panic activities. Really after that has happened, we should (a) disable any further panic detection, and maybe (b) prevent any further watchdog extensions (although it's possible that some people's post-panic activities might actually be extending the watchdog?).
The ./start loop is a separate issue, but similar. When Anaconda checks in during %pre we extend the watchdog and record the ./start result. After that Beaker should not accept any more %pre check-ins, since that just indicates Anaconda bailed out and rebooted.
The latter problem has been known for a long time, we have discussed it before but I can't find an open bug for it. I will clone this one.
This patch will ensure that the installation start is recorded only once and hence the watchdog isn't extended everytime Anaconda checks in: http://gerrit.beaker-project.org/#/c/2168/
Adjusting target milestone to make the changes backported to 0.14.2 easier to identify. 0.15.0 has enough significant regressions that it shouldn't be used, so the change means that 0.15.1 can be effectively reidentified as the union of that tag and the 0.14.2 target milestone.