Make sure we terminate an ingress session in case sending on the app source doesn't cause the pipeline to emit an EOS event #379
Conversation
|
Were these causing some of the e2e failures which said something like |
| } | ||
|
|
||
| s.result <- err | ||
| select { |
There was a problem hiding this comment.
trying to understand this part better - I assume this is done to avoid blocking - and since the channel is already buffered (1) - where else we have writing to the channel? Is whip different in that regard as try send isn't used there?
There was a problem hiding this comment.
This change should not be needed with the code as it stands. However, without this the code is not idempotent, and we have a risk of getting a deadlock if we ever (potentially after a code change somewhere else) have a case where we write to the channel more than once.
There was a problem hiding this comment.
👍 - updating whip appsrc with the same pattern sounds good then
|
would also be good to have pipeline state changes logged so we get more info when this happen to narrow down the source of the issue - I created a small PR for that: #380 |
Yes. e2e is catching the pipeline not shutting down with RTMP, as it should. |
No description provided.