[FFmpeg-devel] Possible leak in RTSP to disk copy writing

[Stas Oskin]

Hi.

Can anyone advice on below?

I'd be happy to receive at least a sign this is indeed a leak/overly long
kept objects, rather then it something by design (some buffers?).

Thanks again.

On Sun, Mar 14, 2010 at 2:56 PM, Stas Oskin <stas.oskin at gmail.com> wrote:

> Hi.
>
> I'm trying to track a leak in my application, and after long inspection,
> found out that it seems to be on the FFMPEG side.
>
> When I connect via RTSP to any video source, and copy the video to disk
> (not re-coding it) the memory consumption slowly but continuously will grow.
>
> Here is an example command that will trigger it:
> ffmpeg -i rtsp://xxx.xxx.xxx.xxx:554/live.sdp -vcodec copy -acodec none
> test.mp4
>
> It's important to notice that valgrind doesn't complain in such case, so it
> seems to be a case of very long lived objects in memory. They do get
> released, but only at the end of the program.
>
>
> This case not happens if I re-stream the packets to another RTP channel in
> this method:
> ffmpeg -i rtsp://xxx.xxx.xxx.xxx:554/live.sdp -vcodec copy -acodec none -f
> rtp rtp://127.0.0.1:50000
>
> So it's related to disk writing only.
>
>
> Is this a known issue?
> If not, what would be the best way to locate it so I could submit a patch?
> Is there any tool that can help locate the long lived objects?
>
> Thanks in advance!
>