[mythtv-users] excessive crashing of 0.22 backend

Ronald Frazier ron at ronfrazier.net
Fri Jan 8 18:47:20 UTC 2010


On Thu, Jan 7, 2010 at 9:35 PM, Michael T. Dean <mtdean at thirdcontact.com> wrote:
> Couldn't really say without a backtrace.  Any chance you can get a bt
> with MythTV debug symbols?  If you don't want to run your production
> backend under gdb, you could set up the system to write a core file and
> run gdb on that.

Thanks a lot. I didn't know you could generate backtraces that way.
That is much more convenient than running under gdb for me.

So I did the rebuild, got a core dump, and ran it through gdb. I was
gonna post it but thought maybe I could solve it myself. Tracked
through the code and found the cause was a QUERY_RECORDING TIMESLOT
request coming over the socket. Problem was, the request was missing
the 4th parameter. So I looked through the code and saw that there
were only a few places that issued a QUERY_RECORDING request. One of
them was the perl bindings, and that instantly threw up a red flag,
because I've got a few perl scripts that utilize the binding. One of
them I haven't had a chance to get working yet (been busy with other
0.22 upgrade issues). So I disabled it...or at least I though I had.
Turns out it was restarting each reboot and then crashing (both itself
and the backend) the first time I watched a recording.

So anyway, I'm gonna accept most of the blame for this. I'll pass
along a little bit of blame because requests that come in from outside
should probably be sanity checked by the backend (some are, but a lot
aren't). I know myth is more or less a closed system, but for backend
stability it's probably still a really good idea. Mike...I'm pretty
sure I know your next words: patches welcome. I'm already working on
it.

Thanks again for the suggestion

-- 
Ron


More information about the mythtv-users mailing list