Introduction

Goryu is a standalone SRT relay. It listens on one SRT port, accepts the encoder's connection, and forwards the MPEG-TS payload to a downstream SRT server, copying the caller's streamid onto the outgoing connection.

It exists because SRT bonding (one stream sent over several links at once) is implemented in libsrt as a socket group, and most SRT servers cannot accept a group. Goryu terminates the group, so the server behind it only ever sees an ordinary single SRT publish.

Bonded and plain input

The listener is bound with SRTO_GROUPCONNECT=1, so it accepts both kinds of caller:

• Bonded. The encoder connects two or more redundant legs (for example two NICs or uplinks) as one socket group. libsrt merges and deduplicates the legs before Goryu reads anything, so Goryu always reads one clean stream per group. A group with a single leg connected is still bonded.
• Plain. A caller using one ordinary SRT socket is accepted by the same listener and handled by the same session code, just without legs.

The status API reports which one a stream is (input.bonded) and, for a plain input, the encoder's address (input.peer).

Data path

        encoder / publisher

   leg A: SRT caller + streamid
   leg B: SRT caller + same streamid
              |
              v
   +-----------------------------+
   | goryu listener              |
   | SRTO_GROUPCONNECT=1         |
   +-----------------------------+
              |
              | SRT group socket
              | libsrt merges and deduplicates legs
              v
   +-----------------------------+
   | session thread per streamid |
   | reads one MPEG-TS payload   |
   +-----------------------------+
              |
              +----> srt:// output, same streamid

With Bunryu

Bunryu runs Goryu as a shared relay in front of SRS: encoders publish bonded SRT to Goryu's port (10081), and Goryu forwards each stream to SRS's SRT port (10080). Bunryu generates goryu.json and polls the status API for per-leg telemetry.

Limits

Goryu forwards; it does not decode, transcode, or buffer beyond SRT's own latency window, so CPU use is low and scales with total bitrate.