Closed GoogleCodeExporter closed 9 years ago
This crash happens only when:
* A large concurrent amount of threads (200+) is writing to disk
* The CPU waiting time (disk access) climbs to 30% - 40%
Might be some missing mutex?
Original comment by Stas.Os...@gmail.com
on 8 Sep 2009 at 2:28
Can you provide some sample code/ This really LOOKS like a thread corruption
issue
has occurred earlier, and then a lib-c dynamic cast is failing. I'm asking for
a non
RTP-based test case because I suspect it's actually in the RTP reading/writing
code.
Original comment by art.cla...@gmail.com
on 9 Sep 2009 at 3:37
Hi Art.
I see what you mean. I'm not quite sure about this, as the packet not written
to disk
but being processed before, so if there was some corruption / thread entangling
before, it probably would become visible then.
I will try to repro it, but again it needs a specific situation to have, and
only
becomes visible on high amount of loads.
Original comment by Stas.Os...@gmail.com
on 9 Sep 2009 at 7:35
I forgot to update that it happens even earlier then dynamic_cast. Here is a
sample dump:
GDB:
#6 <signal handler called>
#7 0x00002aaace4df8b5 in com::xuggle::xuggler::Packet::make () from
/JavaCore/xuggler/lib/libxuggle-xuggler.so.3.2.879
#8 0x00002aaace4f83a5 in
Java_com_xuggle_xuggler_XugglerJNI_IPacket_1make_1_1SWIG_12 ()
from /JavaCore/xuggler/lib/libxuggle-xuggler.so.3.2.879
JCrash:
Thread 2696: (state = IN_NATIVE)
- com.xuggle.xuggler.XugglerJNI.IPacket_make__SWIG_2(long,
com.xuggle.xuggler.IPacket, boolean) @bci=0 (Compiled frame; information
may be imprecise)
- com.xuggle.xuggler.IPacket.make(com.xuggle.xuggler.IPacket, boolean) @bci=6,
line=460 (Compiled frame)
So here the crash point happens on stage that is even earlier then dynamic_cast.
Original comment by Stas.Os...@gmail.com
on 9 Sep 2009 at 7:38
Yeah; if it's a thread corruption issue the stack traces are next to useful.
That's
why the repro case matters so much.
Original comment by art.cla...@gmail.com
on 9 Sep 2009 at 8:44
Closing with a repro case. Provide one and I'll reopen.
Original comment by art.cla...@gmail.com
on 16 Jan 2010 at 2:57
I think it got solved by some work-around - need to check.
Original comment by Stas.Os...@gmail.com
on 17 Jan 2010 at 2:00
Original issue reported on code.google.com by
Stas.Os...@gmail.com
on 8 Sep 2009 at 2:00