[mythtv-users] Need ideas for auto-recovery of locked up backend

[f-myth-users at media.mit.edu]

Watchdog timer.

Connect something like a 555 in monostable mode to a serial-out pin.
Have it trip the reset line if it's not constantly kept from doing so.
Adjust timeout to however long it takes the machine to boot and to
start running your watchdog code, plus a good margin.

Since the 555 has a very large range of operating voltages, you
probably don't need a level converter, though if you're accepting full
RS-232 levels, be careful not to apply similar voltages to the reset
pin directly.  (But an FET across it should be fine, and you're
probably just pulling it to ground anyway.)

Remember to disconnect it when doing anything that puts the machine in
single-user mode. :)

Might also be a good idea to OR it with the HDD activity LED.  Saves
you from a reboot during fsck in which replaying the log isn't enough
and is therefore slow.  Doesn't save you if you're running fsck by
hand and don't answer often enough.  Also doesn't save you if you're
paging so much that the machine is dead and otherwise wouldn't be
running your serial-port watchdog.  This is why good watchdog timers
can be hard to engineer. :)

P.S.  There are also commercial prodcts which are Ethernet-controllable
outlets.  Some have built-in watchdogs---fail to ping them often enough,
and they powercycle the outlet.  I seem to recall finding a $100 one
with four outlets around a while ago; semi-hobby build quality.  But
I don't recall specifics.