* Send `participant_connection_aborted` when participant session is closed
without becoming `ACTIVE`.
There is one sticky case. If there is a migration and the migration
fails, this will send `participant_connection_aborted` even though the
participant may have connected properly on the previous node.
* depsg
Proper message id checks.
Some refactoring for common code in rtcv2service handlers.
Instantiate signalling based on synchronous local candidates flag, but
needs better defined code.
* More v2 signalling changes
- Use wire message in relay signal
- Generate offerId for SDP for sync mode also
- more checks for SDP offer/answer state to see if there are cases of
missed/duplicate messages
* clean up
* Split signal segmenter and reassembler.
As reassembler may need to run a goroutine for pruning, splitting it up
so that we do not have unnecessary goroutines. Keeping the UT unified
though.
Also, for now, trying to see if `Prune` can be called from external
place to avoid creating a goroutine and have extra state/functions to
close the reassembler and clean up goroutine etc. May still have to do
goroutine, but have not created one for now.
* fix: do not increase packetId for each fragment of a packet
* HTTP PATCH on participant sid
* rename
* signal handler
* new files
* clean up
* clean up reassembly
* handle messages on server side
* deps
* staticcheck
* clean up
Currently, it is a bit of a mish-mash
- some compose the message fully and just call send()
- some give parameters and the message is composed in
participant_signal.go
Was thinking about making an interface for signalling and have v1/v2
impls, but did not want to repeat composing messages if there are common
messages. And some of those function reach into `ParicipantImpl` object
and use information (simple example of p.IsReady()) which would become
more elaborate if the signaller is split out into its own struct.
Maybe, just need to make an interface for the sink and send to the
correct sink based on v1 /v2 signal transport.
But, for now, just grouping all signal messaages in one file
so that it is easier to manage later.
Normalize the rids in SDP to known patterns.
Currently,
- LK protocol uses q;h;f
- Sean's OBS WHIP uses 0;1;2
As the ordering in SDP could be different, normalize to known order.
For RIDs not in the known set, just use it as is.
Back up codecs will not have pending track. The update needs to happen
on published track.
Have added a note about updating in published track. Can happen when we
have separate layers for primary and backup.
* SVC with RID -> spatial layer mapping
There are cases where an SVC track comes in with a RID.
As there is no RID announced in SDP, it maps to invalid layer.
Seems to happen with older browsers.
* test
The browser could send rtp packets of svc encoding without
DD extension while the sdp negotiates it, sfu detects extension
in rtp packet for this case.
Raja Subramanian