[mythtv] [mythtv-commits] mythtv branch master updated by stuartm. v0.28-pre-2290-g0a0872e

[Stuart Morgan]

On Sunday 12 Oct 2014 13:03:27 John Pilkington wrote:
> On 12/10/14 12:10, Stuart Morgan wrote:
> > On Sunday 12 Oct 2014 11:28:33 John Pilkington wrote:
> >> 'm now running v0.28-pre-2291-gd97aeef and see the same system
> >> slowdown as before.  All seems ok until moving to the display of
> >> the list of content, when the disk activity light comes on and
> >> stays on.  atop shows near/above 100% action in roup-lv_swap,
> >> roup-lv_root and sda, which continues without anything in the upnp
> >> debug log, and even after unplugging the ethernet cable.   Backend
> >> activity around 3%, system response very sluggish.  Without tv
> >> involvement disk action eventually stops and things gradually
> >> return to normal.
> > 
> > As I mentioned off-list, I'll need some logs with -v upnp
> > --log-level=debug
> > 
> > I need to know what your TVs UPnP server is doing differently to all
> > the others.
> > 
> > When you say you haven't had content since 21e9ec4, do you mean that
> > version worked and ones after it do not, or that it broke with that
> > version?
> 
> That was the last version that worked.
> 
> I wonder if the disk activity could be SG-related again? Copying before
> access?  But no evidence for that as far as I can see.  This particular
> instance was a 'local' Video, but Recordings don't play either.
> 
> tv menu navigation remains rapid during the hang, but eventually 'can't
> read file' or 'contact lost'-ish.
> 
> I'll send this before trying again.  *** marks the spot.

OK, I'm able to reproduce the issue with a manual GET. The problem seems to be 
that QAbstractSocket doesn't update the status when the remote host closes the 
connection. We think the socket is still connected and continue to write the 
entire file to it. The socket buffer is now huge and we start to run out of 
memory, this causes the slowdown.

I'm looking for a solution.
-- 
Stuart Morgan