API: Support limit total connections for WebRTC and live streams.

[elin72215]

Description

I have looked at the wiki but couldn't find the architecture of SRS implementing WebRTC push-pull streaming. I originally thought that the push stream would be uniformly converted to RTMP and then the pull stream would be converted to WebRTC, so that during live streaming, multiple connections would be occupied when multiple clients pull the same stream.

So, I set max_connections to 3 to test what would happen when the connections are full, because I also wanted to test this article: https://www.zhihu.com/question/25497090. Someone said that WebRTC can only support less than ten connections and is not suitable for live streaming. In fact, I opened more than ten players to play the same stream, and all the players played normally, without being affected by the max_connections limit.

So, the questions are, when I use WebRTC for push-pull streaming:

1. Is the pull stream not limited by max_connections? Or is there any other built-in limit?

2. Is there a limit for the push stream?

3. How large does the WebRTC pull stream scale have to be before the picture becomes blurry, or has SRS already optimized this part?

4. Are there any documents or architecture diagrams for the WebRTC push-pull streaming architecture?

5. SRS version: 4.0release

6. SRS log:

[2021-10-01 06:03:10.883][Trace][1][1n885wu3] client identified, type=flash-publish, vhost=srs-origin-0.socs, app=live, stream=999, param=?eip=35.221.153.50, duration=0ms
[2021-10-01 06:03:10.883][Trace][1][1n885wu3] connected stream, tcUrl=rtmp://srs-origin-0.socs/live, pageUrl=, swfUrl=, schema=rtmp, vhost=__defaultVhost__, port=1935, app=live, stream=999, param=?eip=35.221.153.50, args=(obj)
[2021-10-01 06:03:10.883][Trace][1][1n885wu3] source url=/live/999, ip=127.0.0.1, cache=1, is_edge=0, source_id=712p9886/712p9886
[2021-10-01 06:03:10.884][Trace][1][1n885wu3] TCP: before dispose resource(RtmpConn)(0x351cdb0), conns=3, zombies=0, ign=0, inz=0, ind=0
[2021-10-01 06:03:10.884][Error][1][1n885wu3][11] serve error code=1028 : service cycle : rtmp: stream service : rtmp: stream /live/999 is busy
thread [1][1n885wu3]: do_cycle() [src/app/srs_app_rtmp_conn.cpp:217][errno=11]
thread [1][1n885wu3]: service_cycle() [src/app/srs_app_rtmp_conn.cpp:414][errno=11]
thread [1][1n885wu3]: acquire_publish() [src/app/srs_app_rtmp_conn.cpp:949][errno=11](Resource temporarily unavailable)
[2021-10-01 06:03:10.884][Trace][1][j11ma521] TCP: clear zombies=1 resources, conns=3, removing=0, unsubs=0
[2021-10-01 06:03:10.884][Trace][1][1n885wu3] TCP: disposing #0 resource(RtmpConn)(0x351cdb0), conns=3, disposing=1, zombies=0
[2021-10-01 06:03:11.118][Trace][1][712p9886] 34B video sh,  codec(7, profile=Baseline, level=2.1, 320x240, 0kbps, 0.0fps, 0.0s)
[2021-10-01 06:03:11.233][Trace][1][156vh788] Hybrid cpu=3.00%,41MB, cid=85,75, timer=62,0,639, clock=0,47,2,0,0,0,0,0,0, free=1, objs=(pkt:1612,raw:758,fua:851,msg:1774,oth:2,buf:50)
[2021-10-01 06:03:11.233][Trace][1][156vh788] **RTC: Server conns=13,** rpkts=(69,rtp:0,stun:5,rtcp:64), spkts=(1528,rtp:1523,stun:5,rtcp:0), rtcp=(pli:4,twcc:0,rr:0), rnk=(1,1,h:26,m:0), loss=(r:0,s:538), fid=(id:0,fid:69,ffid:0,addr:5,faddr:69)

1. SRS configuration:

listen              1935;
max_connections     3;
daemon              off;
http_api {
    enabled         on;
    listen          1985;
}
http_server {
    enabled         on;
    listen          8080;
}
rtc_server {
enabled on;
# Listen at udp://8000
listen 8000;
#
# The $CANDIDATE means fetch from env, if not configed, use * as default.
#
# The * means retrieving server IP automatically, from all network interfaces,
# @see https://github.com/ossrs/srs/wiki/v4_CN_RTCWiki#config-candidate
candidate *;
}
vhost __defaultVhost__ {
#   cluster {
#        origin_cluster  on;
#        coworkers       127.0.0.1:1985;
#    }
    http_remux {
        enabled     on;
    }
    hls {
        enabled         off;
    }
    rtc {
      enabled     on;
      rtc_to_rtmp on;
    }
}

[winlinvip]

I've looked through your questions, and it seems like you are expecting the following: For WebRTC push-pull streaming, you want to use max_connections to limit the total number of connections, right?

1. This feature has not been implemented yet, but it should be done. It is used to limit the system connection count, which is generally for preventing system overload, such as 1000 connections. If it exceeds this number, the connection will be rejected.
2. You set max_connections 3; to limit the number of connections. Is this for authentication purposes? If you want to verify whether each person has permission, don't use this. Instead, use the HTTP Callback, which has already been implemented.

[elin72215]

Hello, I set max_connections 3 to test what would happen when the connection limit is exceeded. I just happened to find that WebRTC push-pull streaming is not restricted by this variable. My main concern is to understand the architecture of SRS's WebRTC push-pull streaming to determine whether to use it for live streaming. Suppose live streamer A uses WebRTC to push the stream, and client B uses WebRTC to pull the stream:

1. Does SRS act as a signaling server only to assist both parties in establishing a peer-to-peer connection, with B directly pulling the stream from A?
2. Or is it that A first establishes a peer-to-peer connection with SRS to push the stream, and after SRS receives the stream, SRS establishes a peer-to-peer connection with B for B to pull the stream?

In other words, are the peer-to-peer connection targets between the two users each other or SRS?

[duiniuluantanqin]

1. SRS uses an SFU architecture, and all streams are relayed through SRS. So, the process is that A pushes the stream to SRS, and B pulls the stream from SRS.
2. SRS did not support WebRTC before and was an RTMP server. max_connections was used to limit the number of RTMP connections. Since WebRTC uses UDP transmission, do we need to limit the number of connections?

[elin72215]

So, to implement WebRTC live streaming with WebRTC push streaming and a large number of WebRTC pull streams (for example, hundreds of thousands or millions of clients), as long as there are not too many push streams, a single SRS server can handle it? If we need to expand the push streaming scale, we can use the Origin Cluster mode, right? On the other hand, using RTMP push streaming and WebRTC pull streaming with a cluster would be limited because the transmitted stream remains RTMP until it actually goes out?

[duiniuluantanqin]

Yes, I talked with Mr. Yang yesterday, and this should be done. You are welcome to submit a PR (pull request).