[Ffmpeg-devel] FFServer and RTSP
Mon Apr 16 11:08:14 CEST 2007
Alex Beregszaszi wrote:
>> Even not considering the crashes, ffserver has some other "funny
>> behaviour" when streaming RTP: for example, the first two frames read
>> from the file have the same PTSs and DTSs, sometimes resulting in a
>> "error, non monotone timestamps 45000 >= 45000 st:1" error when trying
>> to stream them (I say sometimes, because if the first PTS is 0 then the
> The timing is broken in .ffm. I'm cleaning it up, but didn't managed to
> share all the patches yet.
Well, for my test I did not use an ffm. I was just trying to stream a
.mpg file (containing MPEG1 video and mp2 audio) via RTP.
Anyway, after some other testing I discovered that this problem is due
to the fact that av_read_frame() is called before calling
After fixing this, there are other problems due to the fact that the
MPEG file contains 2 streams, sent over 2 different RTP streams... And
ffserver is kind of confused in deciding when to send an RTP packet (it
does not consider the "interleaving" between the PTSs of the two streams).
>> error does not occourr...). I think the error is due to the fact that
>> the file is opened (with av_open_input_file()), but
>> av_find_stream_info() is not invoked...
>> Is it legal to open a file for reading (av_open_input_file()), and start
>> reading frames from it without performing an av_find_stream_info()?
> Did you found that in the HTTP part too?
Sorry, I did not look at HTTP... I think HTTP is used with .ffm feeds,
right? So, frames are read using a different code (if I understand the
ffserver code correctly), and I do not know about this different
codepath (anyway, I just saw that av_find_stream_info() is called on the
feed file... So maybe HTTP is ok).
>> Anyway, it seems to me that ffserver is in a very bad shape regarding
>> RTSP/RTS streaming...
>> So I'd like to have an opinion from other people (an opinion from the
>> official maintainer would be appreciated :): is there any hope in fixing
>> ffserver for RTP streaming? (did it ever work, at least one time?).
> Sadly I don't really know the RTP/RTSP part of the code, was fixing only
> overall bugs and HTTP part, which works for me quite reliable. We are
> usin git now for 3 months, without any crashes.
Yes, after looking again at the code in this weekend I saw that the
problems I found seem to be "RTP only" things.
More information about the ffmpeg-devel