gstreamer/docs/README

README 
------

(Last updated on Fri 30 jan 2009, version 0.10.1.1)

This HOWTO describes the basic usage of the GStreamer RTSP libraries and how you
can build simple server applications with it.

* General

 The server relies heavily on the RTSP infrastructure of GStreamer. This includes
 all of the media acquisition, decoding, encoding, payloading and UDP/TCP
 streaming. We use the gstrtpbin element for all the session management. Most of
 the RTSP message parsing and construction in the server is done using the RTSP
 library that comes with gst-plugins-base. 

 The result is that the server is rather small (a few 1000 lines of code) and easy
 to understand and extend. In its current state of development, things change
 fast, API and ABI are unstable. We encourage people to use it for their various
 use cases and participate by suggesting changes/features.

 Most of the server is built as a library containing a bunch of GObject objects
 that provide reasonable default functionality but has a fair amount of hooks
 to override the default behaviour.

 The server currently integrates with the glib mainloop nicely. It is also a
 heavy user of multiple threads. It's currently not meant to be used in
 high-load scenarios and you should probably not put it on a public IP address.

* Initialisation

 You need to initialize GStreamer before using any of the RTSP server functions.

   #include <gst/gst.h>

   int
   main (int argc, char *argv[])
   {
     gst_init (&argc, &argv);

     ...
   }
 
 The server itself currently does not have any specific initialisation function
 but that might change in the future.


* Creating the server

 The first thing you want to do is create a new GstRTSPServer object. This object
 will handle all the new client connections to your server once it is added to a
 GMainLoop. You can create a new server object like this:

   #include <gst/rtsp-server/rtsp-server.h>

   GstRTSPServer *server;

   server = gst_rtsp_server_new ();

 The server will by default listen on port 8554 for new connections. This can be
 changed by calling gst_rtsp_server_set_service() or with the 'service' GObject
 property. This makes it possible to run multiple server instances listening on
 multiple ports on one machine.

 We can make the server start listening on its default port by attaching it to a
 mainloop. The following example shows how this is done and will start a server
 on the default 8554 port. For any request we make, we will get a NOT_FOUND
 error code because we need to configure more things before the server becomes
 useful.

   #include <gst/gst.h>
   #include <gst/rtsp-server/rtsp-server.h>

   int
   main (int argc, char *argv[])
   {
     GstRTSPServer *server;
     GMainLoop *loop;

     gst_init (&argc, &argv);

     server = gst_rtsp_server_new ();

     /* make a mainloop for the default context */
     loop = g_main_loop_new (NULL, FALSE);

     /* attach the server to the default maincontext */
     gst_rtsp_server_attach (server, NULL);

     /* start serving */
     g_main_loop_run (loop);
   }

 The server manages two other objects: GstRTSPSessionPool and
 GstRTSPMediaMapping. 

 The GstRTSPSessionPool is an object that keeps track of all the active sessions
 in the server. A session will usually be kept for each client that performed a
 SETUP request for a certain media stream. It contains the configuration that
 the client negotiated with the server to receive the particular stream, ie. the
 transport used and port pairs for UDP along with the state of the streaming.
 The default implementation of the session pool is usually sufficient but
 alternative implementation can be used by the server.

 The GstRTSPMediaMapping object is more interesting and needs more configuration
 before the server object is useful. This object manages the mapping from a
 request URL to a specific stream and its configuration. We explain in the next
 topic how to configure this object.


* Making url mappings

 Next we need to define what media is attached to a particular URL. What we want
 to achieve is that when the user asks our server for a specific URL, say /test,
 that we create (or reuse) a GStreamer pipeline that produces one or more RTP
 streams. 
 
 The object that can create such pipeline is called a GstRTSPMediaFactory object. 
 The default implementation of GstRTSPMediaFactory allows you to easily create
 GStreamer pipelines using the gst-launch syntax. It is possible to create a
 GstRTSPMediaFactory subclass that uses different methods for constructing
 pipelines.

 The default GstRTSPMediaFactory can be configured with a gst-launch line that
 produces a toplevel bin (use '(' and ')' around the pipeline description to
 force a toplevel GstBin instead of the default GstPipeline toplevel element).
 The pipeline description should contain elements named payN, one for each
 stream (ex. pay0, pay1, ...). Also, for increased compatibility each stream
 should have a different payload type which can be configured on the payloader.

 The following code snippet illustrates how to create a media factory that
 creates an RTP feed of an H264 encoded test video signal.

   GstRTSPMediaFactory *factory;

   factory = gst_rtsp_media_factory_new ();

   gst_rtsp_media_factory_set_launch (factory, 
                "( videotestsrc ! x264enc ! rtph264pay pt=96 name=pay0 )");

 Now that we have the media factory, we can attach it to a specific url. To do
 this we get the default GstRTSPMediaMapping from our server and add the url to
 factory mapping to it like this:

   GstRTSPMediaMapping *mapping;

   ...create server..create factory..

   /* get the default mapping from the server */
   mapping = gst_rtsp_server_get_media_mapping (server);

   /* attach the video test signal to the "/test" URL */
   gst_rtsp_media_mapping_add_factory (mapping, "/test", factory);
   g_object_unref (mapping);

 When starting the server now and directing an RTP client to the URL (like with 
 vlc, mplayer or gstreamer):

   rtsp://localhost:8554/test 

 a test signal will be streamed to the client. The full example code can be
 found in the examples/test-readme.c file.

 Note that by default the factory will create a new pipeline for each client. If
 you want to share a pipeline between clients, use
 gst_rtsp_media_factory_set_shared().


* more on GstRTSPMediaFactory

 The GstRTSPMediaFactory is responsible for creating and caching GstRTSPMedia
 objects. 

 A freshly created GstRTSPMedia object from the factory initially only contains a
 GstElement containing the elements to produce the RTP streams for the media and
 a GArray of GstRTSPMediaStream objects describing the payloader and its source
 pad. The media is unprepared in this state.

 Usually the url will determine what kind of pipeline should be created. You can
 for example use query parameters to configure certain parts of the pipeline or
 select encoders and payloaders based on some url pattern.

 When dealing with a live stream from, for example, a webcam, it can be
 interesting to share the pipeline with multiple clients. This must be done when
 only one instance of the video capture element can be used at a time. In this
 case, the shared property of GstRTSPMedia must be used to instruct the default
 GstRTSPMediaFactory implementation to cache the media.

 When all objects created from a factory can be shared, you can set the shared
 property directly on the factory.

* more on GstRTSPMedia

 After creating the GstRTSPMedia object from the factory, it can be prepared
 with gst_rtsp_media_prepare(). This method will put those objects in a
 GstPipeline and will construct and link the streaming elements and the
 gstrtpbin session manager object.

 The _prepare() method will then preroll the pipeline in order to figure out the
 caps on the payloaders. After the GstRTSPMedia prerolled it will be in the
 prepared state and can be used for creating SDP files or for streaming to
 clients.

 The prepare method will also create 2 UDP ports for each stream that can be
 used for sending and receiving RTP/RTCP from clients. These port numbers will
 have to be negotiated with the client in the SETUP requests.

 When preparing a GstRTSPMedia, a multifdsink is also constructed for streaming
 the stream over TCP when requested.


* the GstRTSPClient object

 When a server detects a new client connection on its port, it will call its
 accept_client vmethod. The default implementation of this function will create
 a new GstRTCPClient object, will configure the session pool and media mapper
 objects in it and will then call the accept function of the client.

 The default GstRTSPClient will accept the connection and will attach a watch to
 the server mainloop. In RTSP it is usual to keep the connection
 open between multiple RTSP requests. The client watch will be dispatched by the
 server mainloop when a new GstRTSPMessage is received, which will then be
 handled and a response will be sent.

 The GstRTSPClient object remains alive for as long as a client has a TCP
 connection open with the server. Since is possible for a client to open and close
 the TCP connection between requests, we cannot store the state related
 to the configured RTSP session in the GstRTSPClient object. This server state
 is instead stored in the GstRTSPSession object.


* GstRTSPSession

 This object contains state about a specific RTSP session identified with a
 session id. This state contains the configured streams and their associated
 transports.
 
 When a GstRTSPClient performs a SETUP request, the server will allocate a new
 GstRTSPSession with a unique session id from the GstRTSPSessionPool. The pool
 maintains a list of all existing sessions and makes sure that no session id is
 used multiple times. The session id is sent to the client so that the client
 can refer to its previously configured state by sending the session id in
 further requests.

 A client will then use the session id to configure one or more streams,
 identified by their url. This information is kept in a GstRTSPSessionMedia
 structure that is refered to from the GstRTSPSession. 


* GstRTSPSessionMedia and GstRTSPSessionStream

 A GstRTSPSessionMedia is identified by a URL and is referenced by a
 GstRTSPSession. It is created as soon as a client performs a SETUP operation on
 a particular URL. It will contain a link to the GstRTSPMedia object associated
 with the URL along with the state of the media and the configured transports
 for each of the streams in the media.

 Each SETUP request performed by the client will configure a
 GstRTSPSessionStream object linked to by the GstRTSPSessionMedia structure.
 It will contain the transport information needed to send this stream to the
 client. The GstRTSPSessionStream also contains a link to the GstRTSPMediaStream
 object that generates the actual data to be streamed to the client.

 Note how GstRTSPMedia and GstRTSPMediaStream (the providers of the data to
 stream) are decoupled from GstRTSPSessionMedia and GstRTSPSessionStream (the
 configuration of how to send this stream to a client) in order to be able to
 send the data of one GstRTSPMedia to multiple clients.


* media control

 After a client has configured the transports for a GstRTSPMedia and its
 GstRTSPMediaStreams, the client can play/pause/stop the stream.

 The GstRTSPMedia object was prepared in the DESCRIBE call (or during SETUP when
 the client skipped the DESCRIBE request). As seen earlier, this configures a
 couple of multiudpsink and udpsrc elements to respectively send and receive the
 media to clients.

 When a client performs a PLAY request, its configured destination UDP ports are
 added to the GstRTSPMediaStream target destinations, at which point data will
 be sent to the client. The corresponding GstRTSPMedia object will be set to the
 PLAYING state if it was not allready in order to send the data to the
 destination.

 The server needs to prepare an RTP-Info header field in the PLAY response,
 which consists of the sequence number and the RTP timestamp of the next RTP
 packet. In order to achive this, the server queries the payloaders for this
 information when it prerolled the pipeline.

 When a client performs a PAUSE request, the destination UDP ports are removed
 from the GstRTSPMediaStream object and the GstRTSPMedia object is set to PAUSED
 if no other destinations are configured anymore.


* seeking

 A seek is performed when a client sends a Range header in the PLAY request.
 This only works when not dealing with shared (live) streams.

 The server performs a GStreamer flushing seek on the media, waits for the
 pipeline to preroll again and then responds to the client after collecting the
 new RTP sequence number and timestamp from the payloaders.


* session management

 The server has to react to clients that suddenly disappear because of network
 problems or otherwise. It needs to make sure that it can reasonable free the
 resources that are used by the various objects in use for streaming when the
 client appears to be gone.

 Each of the GstRTSPSession objects managed by a GstRTSPSessionPool has
 therefore a last_access field that contains the timestamp of when activity from
 a client was last recorded.
 
 Various ways exist to detect activity from a client:

  - RTSP keepalive requests. When a client is receiving RTP data, the RTSP TCP
    connection is largely unused. It is the client responsability to
    periodically send keep-alive requests over the TCP channel.

    Whenever a keep-alive request is received by the server (any request that
    contains a session id, usually an OPTION or GET_PARAMETER request) the
    last_access of the session is updated.

  - Since it is not required for a client to keep the RTSP TCP connection open
    while streaming, gst-rtsp-server also detects activity from clients by
    looking at the RTCP messages it receives.

    When an RTCP message is received from a client, the server looks in its list
    of active ports if this message originates from a known host/port pair that
    is currently active in a GstRTSPSession. If this is the case, the session is
    kept alive.

    Since the server does not know anything about the port number that will be
    used by the client to send RTCP, this method does not always work. Later
    RTSP RFCs will include support for negotiating this port number with the
    server. Most clients however use the same port number for sending and
    receiving RTCP exactly for this reason.

 If there was no activity in a particular session for a long time (by default 60
 seconds), the sessionpool will destroy the session along with all related
 objects and media as if a TEARDOWN happened from the client.


* TEARDOWN

 A TEARDOWN request will first location the GstRTSPSessionMedia of the URL. It
 will then remove all transports from the streams, making sure that streaming
 stops to the client. It will then remove the GstRTSPSessionMedia and
 GstRTSPSessionStream structures. Finally the GstRTSPSession is released back
 into the pool.

 When there are no more references to the GstRTSPMedia, the media pipeline is
 shut down and destroyed.