|
Asheesh
The
problem you listed below, as well as the earlier crash in the SipMessage dtor,
are symptoms of your application corrupting the heap. I know this because
that is what my application was doing. This is most likely occurring in
some load-initiated error handling scenario. I can only tell you to try
and find the problem under valgrind. I know it really slows things down,
but in running slower you may be able to simulate the overloaded
situation. Perhaps even instrument some code to force retries/packet
loss/etc?
It
took me nearly 2 solid weeks to completely rid my code of these issues. As
an example, there was one problem in which I was writing to an object that
contained a UDP packet just after sending the packet. Under loading the
writing thread would block in the sendto syscall. While it was blocked,
the UDP response packet was received on a different thread and the original sent
object was released in that thread. When the sending thread finally got
scheduled again, the write to the object was after it was back on the heap (and
possibly allocated again). These types of problems are very hard to find,
but they will occur when you have >30000 context switches per second as you
might with RTP processing and very high call setup rates.
Good luck.
Dennis
Hi,
Can anyone help me as to
when one would get such an error ?
Free( ) : Invalid pointer:
0xxxxxxxxx
Is it compiler dependent ? While
running my Resip-B2BUA, the logs
showed up such an error after
50,000 calls. I have never seen this thing
before.
Thanks in
advance
Asheesh.
|