[mythtv-users] How do I get coverart/etc in watch recordings?

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

    > Date: Thu, 17 Sep 2009 08:45:57 +1000
    > From: Jean-Yves Avenard <jyavenard at gmail.com>

    > The NFS server was 4 minutes ahead of time... even with ntpd running ;

ntpd will punt if the difference gets too large and will just let your
clock free-run.  I believe the typical limit is 30 seconds, but I
don't remember.  Basically, it decides it can't keep its phase-locked
loop running with that much drift; increasing the gain would likely
cause other stability problems, so...

I have 3 identical motherboards which occasionally (5%? 1%?) boot with
the clock running at least 5% fast; this is enough to cause NTP to lose
sync within 5 to 60 minutes and drop out.  Needless to say, when this
happens on the backend and it starts gaining up to 5 minutes every
hour, it plays merry hell with recordings.  It's obviously a BIOS bug,
but there hasn't been an update in years and I value the boards in
part because they have 5 PCI slots each and in part because they are
otherwise working fine.  [It would do this routinely if the spread-
spectrum stuff was enabled on the board, but even with it disabled,
it sometimes happens; maybe sometimes the SS stuff comes up enabled
even though the BIOS says not to.]

So I wrote a script that crosschecks each machine against a reference
every minute using sntp and notifies me if the difference exceeds one
second.  Once that happens, it kills NTP and starts disciplining the
clock by calling ntpdate on an external reference once every minute
instead.  This is hard on the reference and causes a 3 second clock
jump every minute, but it keeps things glued together until I can
notice and reboot in between recordings.

If anyone else is having problems with excessive clock drift, I can
post the scripts.  Note that you need an internal (*) reference that
isn't likely to be drifting, or at least you need two machines that
aren't both likely to drift at the same time, since you don't want to
be pinging an external (not run by you) time server every minute
forever.

(*) By "internal", I mean, "a machine that doesn't mind being
interrogated for its time via NTP once per minute", which can
easily be any other machine on your internal network but probably
shouldn't be any random NTP pool server out on the greater Internet.