Issues with valves

[Alexey Chernov]

On Wednesday 16 March 2011 17:19:08 Hector Canto wrote:
> Thanks a lot for the answer. I'll take your advise into account.
> 
> To follow up the issue I have to say that I have improved my recording by
> changing the valves
> the moment the pipeline is on pause, these way I don't stop the
> initialization and I don't have to
> take measures such as unblocking, flushing or whatever. The only problem I
> have left it's that some
> data is able to pass the valve before the dropping is on. Further the
> valves will work properly.

Yes, I've also been experimenting with valves and in some cases it's the right 
way to reduce the code size without any complex blocking logic.

> I've been thinking on using the output-selector element - along with a
> fakesink as one of the
> selections -, but I couldn't make it work, neither findany suitable example
> around the web.

I think use of fakesink in some kind of relay is quite common practise, I did 
the exact this way. The problem seems to be in the fact that output-selector 
as also tee don't generate anything on activating or deactivating outputs so 
the user should take care of it by himself.

> About changing elements dynamically, I agree, it would bring new problems
> but I may use it on
> future improvements of the pipeline. I was able to get it work following
> the instructions on this link
> from GST:
> http://cgit.freedesktop.org/gstreamer/gstreamer/tree/docs/design/part-block
> .txt

I think it's some kind of base guideline to write such pipelines, I used it 
intensively, too. In terms of video capture to file I have a couple more links, 
maybe you find it interesting for your projects:

http://bugzilla.gnome.org/show_bug.cgi?id=561224 - report about the resettime 
plugin which can help to keep proper segment stuff on turning some branch on 
and off (it can be used to create correct seekable file with correct timestamps)

http://gstreamer-devel.966125.n4.nabble.com/Handling-EOS-in-a-branched-
pipeline-td2965563.html - about sending EOS event at the end of recording. I 
should add that one should pay extra attention to prevent any data after EOS 
events as some elements (theoraenc, for instance) don't expect it and they 
simply crash. And it seems that there's no rule saying they won't.

Anyway, there's always a working method: to stop pipeline, change it and start 
with changed structure again. In many cases it helps.

> Anyway, I have learnt that many of the problems related to the states can
> be solved forcing flushing
> events but I need to know where - and if- a blockade has happened, so I
> will have to research on how
> gstreamer messages works.

It's very interesting. Many times I wanted to try flush events to solve 
blockade and starvations but never could use it right. Could you describe the 
way you use flush events? Do you send flush-begin and flush-end in a row or wait 
some time?

> 
> Cheers
> Hector
> 
> 
> 2011/3/16 4ernov <4ernov at gmail.com>
> 
> > Hello, Hector,
> > 
> > > Hi everybody,
> > > 
> > > I'm new in GStreamer but trying to move faster through it.
> > > 
> > > I'm recording video with basic pipelines and now I'm trying to use
> > > valves
> > 
> > to
> > 
> > > regulate the record (stop/pause/cut...).
> > > The problem is that the pipeline won't pass Prerolling when the valves
> > 
> > are
> > 
> > > initialised dropping.
> > 
> > I'd say using valves is not the best way for your case as it's very
> > easy to drop some useful data (e.g. events or buffers) with them.
> > Dynamic changes in pipeline is better but they also have a bunch of
> > problems you need to solve.
> > 
> > > I've also try to change sinks on-the-move (Playing) but it didn't work
> > > as
> > 
> > I
> > 
> > > expected.
> > > 
> > > Any tips?
> > 
> > If I get it right what you're trying to do is to change the pipeline
> > dynamically. Well, it's not the easiest part of work with GStreamer, I
> > should say. There're several posts in the list about it here and
> > there, but to sum everything гз here're the problems one need to solve
> > to change the pipeline dynamically in proper way:
> > 1. Branch synchronization if some of the sinks is in sync (e.g.
> > streaming video to ximagesink and capturing it to file
> > simultaneously): it's necessary to use queue for each branch.
> > 2. To reconnect elements on the fly one need to look after the
> > remaining pipeline. It's done by either blocking pads of changing
> > branch or using valves (so dangerous).
> > Besides that you should take into account that with every problem the
> > pipeline is stalled hard without any messages even on DEBUG log level.
> > It make the debugging difficult very much. Also these instructions
> > work quite stable only for PLAYING state and have many issues for
> > others. For example, if you block pads in PAUSED state, it will stall
> > the pipeline forever as no buffers are streamed through the pipeline
> > in this state. And finally, remember that elements are not guaranteed
> > to be reusable i.e. if you create pipeline from scratch, set it to
> > PLAYING state, then set it to READY and again to PLAYING, it's not
> > guaranteed that the behavior is the same as if you create a new one
> > instead of setting the existing pipeline to READY and then to PLAYING.
> > 
> > So, there's some hope that it would be changed in the future releases,
> > but now it's not easy at all to change the pipeline dynamically.
> > 
> > Hope it helps.