[mythtv-users] mythbackend still eats memory: the current status

[Alex Butcher]

On Fri, 16 Jan 2009, Jan Ceuleers wrote:

> The point of my test was to demonstrate that memory leaks are indeed present 
> in mythbackend, even while it's not recording, even on a less-complicated 
> setup than Udo's, even on a system that seems to work as designed. I'm not 
> complaining though. Reading this thread got me interested in trying to run 
> mythbackend under valgrind and when it spit out error messages I thought it 
> silly not to report them here.

FYI, an earlier encounter with Udo's problems encouraged me to set up an
hourly log of mythbackend's memory usage.

I've put the graph of two months of data online at
<http://www.assursys.co.uk/people/alex/mythbackend-memory.pdf>. It shows a
slow but undeniable rise in VSZ over that period from about 300MB to about
376MB. RSS also rose at a similar rate, but for some reason there was a
sudden drop over a number of hours about 5 weeks in. RSS rose from about
25MB to 55MB before dropping back down to about 37MB and rising again to
about 50MB.

This was Axel's ATrpms build of MythTV 0.21, revision 194 (svn fixes 19114),
running on FC8 updated as of roughly January 15 2009. I'm aware that both
the OS and MythTV are rather old, but hopefully there's still some value in
the data.

My hardware is a P4 2.53GHz with 768MB of RAM, a Hauppauge Nova-T and a
Nova-T-500 (so three DVB-T physical tuners altogether). I use multirec to
provide two virtual tuners per physical tuner. The UK has 6 multiplexes. I
probably record at least 4 hours per day most days. EIT is enabled. The
kernel is 2.6.26.6-49.fc8 using the standard cx88_dvb and dvb_usb_dib0700
drivers supplied by that kernel.

> Cheers, Jan

HTH,
Alex