Open darkdarkdragon opened 5 years ago
We have to also take into consideration the assumptions we're making for the operating environment.
If neither of these are true, then the transcoding environment is not appropriate for live streaming. There's only so much we can (or should) do to fix that. That includes ensuring the ingest setup is reliable.
Now, there are cases where we may indeed want to support "batch" transcoding, eg VOD. But that calls for a different type of workflow, and right now I feel pretty strongly that we should optimize for those cases separately from the live workflow.
Describe the bug Orchestrator was made with assumption that segments coming in with realtime speed, in steady manner. That means, that orchestrator always have
segment length
seconds to process segment before next one arrives. But there are legitimate cases - see here (and probably it can happen in real networks, with data buffer somewhere along the way and then arrived to broadcaster in big batch) - when this is not the case, and we don't handle these cases.This will not allow us to achieve our goal of 99.999% reliability, especially given that we're also aiming at enterprise user's use cases.