Add selfstreaming feature to the runners #1657
Conversation
|
Fixes #1654 |
SebiWrn
force-pushed
the
runner/add-selfstream
branch
from
7c51588 to
0ebe546
Compare
Co-authored-by: Joscha Henningsen <[email protected]>
Co-authored-by: Joscha Henningsen <[email protected]>
Co-authored-by: Joscha Henningsen <[email protected]>
Co-authored-by: Joscha Henningsen <[email protected]>
SebiWrn
force-pushed
the
runner/add-selfstream
branch
from
8e9e776 to
dc7caa9
Compare
SebiWrn
force-pushed
the
runner/add-selfstream
branch
from
dc7caa9 to
515502a
Compare
joschahenningsen
left a comment
joschahenningsen
left a comment
There was a problem hiding this comment.
Sorry for being so verbose reviewing some of the old worker code, I think it's worthwile to re-think how it was done back then now and get rid of some of that tech depth :)
Overall, I really love the approach of having gocast be the initiator of self-streams as though they were normal lecture-hall streams!
joschahenningsen
left a comment
joschahenningsen
left a comment
There was a problem hiding this comment.
One concern with this, otherwise looks great! 💯
| return nil, err | ||
| } | ||
|
|
||
| resp, err := m.requestStreamVersion(ctx, stream, client, model.LectureHall{}, protobuf.StreamVersion_STREAM_VERSION_COMBINED) |
There was a problem hiding this comment.
We need to make sure, this is requested at the same runner, that this request originates from, otherwise it will break as soon as we have more than one runner
There was a problem hiding this comment.
You're right, I forgot to implement this. An alternative would be to have only one mediaMTX instance at a time and then distribute the load of selfstreams evenly to the runners as every runner could fetch the stream from this instance and we wouldn't rely on only one machine for a stream (except for the mediaMTX). What do you think about this?
There was a problem hiding this comment.
I'd like all runners to be treated the same, let's design this to work without magic knowledge of what which runner does
There was a problem hiding this comment.
If we need load balancing to multiple ingest servers, let's use traefik and DNS rr to distribute streams to multiple runners
There was a problem hiding this comment.
This means we only have one mediamtx for all instead of having one for each runner?
|
Whole functionality of requesting a selfstream now moved to main instance (gocast), so that every runner can handle the selfstreaming requests. |
Motivation and Context
This PR adds the selfstreaming to the runners
Important
Please first merge #1656, as these PRs build on top of each other