[mythtv-users] myth hangs + locks whole system

[Ross Boylan]

On Thu, 2012-08-23 at 10:51 +0200, Jean-Yves Avenard wrote:
> On 23 August 2012 03:55, Ross Boylan <RossBoylan at stanfordalumni.org> wrote:
> > Finally on 0.25.2 I've had myth hang the entire system twice today.  The
> > system eventually (c 20 minutes) came back.  I was hoping that the
> > upgrade would fix such problems, but it seems not.
> 
> Let's be clear..
> 
> Myth is a user application, it can *never* hang the kernel.
> So stating myth hang your PC makes no sense at all. Your PC crash, it
> just turned out that you were using myth at the time.
That seems a rather defensive and unhelpful response.  I ran myth and my
system became totally unresponsive.  Only when running myth has that
happened.  There are various messages in the logs relating to myth,
including an apparently infinite loop of "No more queue slots!" from
mythtranscode.

How would you propose I describe that, or, more to the point, do
something about it?

Interestingly, I viewed that same segment that caused the hang before
and it did not hang again.  So it is clearly a "whole system"
phenomenon; the load was lighter the second time.  Processor is a
Pentium 4.

To clarify, when I say "myth" I mean the application; the problem could
easily lie in some supporting library.  Who knows; it could even be a
kernel bug.

At your implicit suggestion I'll try running memtest and see if it turns
up anything.  The system is a few years old (5?).

> 
> If the kernel crash (like it did today) your issue is either a bug in
> the kernel/drivers or faulty hardware. When it's hardware it's usually
> bad RAM.
> 
> You had a kernel page fault.
> _______________________________________________
If this is a reference to the "pageout: orphaned page" messages I can
only repeat what a kernel developer told me: it's a random diagnostic,
not a problem.  I'm also not sure if you mean fault as in the normal
process of bringing swapping in memory or fault as in error.

The other action I see in the logs consists of mythbackend processes
being blocked and the out of memory killer dispatching "random"
processes.  The latter presumably indicates a memory leak somewhere.

Ross