transcode: add logic to force a session re-init #902
Conversation
When a sequence of segments has a discontinuity (e.g. encoded differently), the transcoder at T must be re-initialized so that transcode operations can be completed correctly. The T takes some liberties and assumes a sequence of segments are encoded the same way and thus, ends up not re-initializing certain parts of the pipeline (e.g. demuxer, etc) to save time and improve efficiency. This can result in corrupted segments with two video tracks instead of one audio and one video track. For clipping, the first segment (index=0) is encoded differently from the rest of the file. As such, a session reinit must be forced at the T using a header set in the request to B.
emranemran
force-pushed
the
emran/force-session-reinit-catalyst
branch
from
56d15cb to
a76bd68
Compare
Codecov Report
@@ Coverage Diff @@
## main #902 +/- ##
===================================================
- Coverage 51.00439% 50.95253% -0.05186%
===================================================
Files 63 63
Lines 6372 6404 +32
===================================================
+ Hits 3250 3263 +13
- Misses 2858 2874 +16
- Partials 264 267 +3
Continue to review full report in Codecov by Sentry.
|
| if transcodeRequest.IsClip && int64(segment.Index) == 1 { | ||
| transcodeConf.ForceSessionReinit = true | ||
| } else { | ||
| transcodeConf.ForceSessionReinit = false |
There was a problem hiding this comment.
Nit: I think you can skip this else since false is the default value of bool.
There was a problem hiding this comment.
Or just transcodeConf.ForceSessionReinit = transcodeRequest.IsClip && int64(segment.Index) == 1
There was a problem hiding this comment.
Do we not need a re-init for the final segment too (when that's also clipped)?
There was a problem hiding this comment.
Do we not need a re-init for the final segment too (when that's also clipped)?
That's a good point. I guess that yes!
There was a problem hiding this comment.
Do we not need a re-init for the final segment too (when that's also clipped)?
No actually, surprisingly. For some reason, the transcoder session is able to handle going from source recording segment to final segment that's also clipped (i.e re-encoded differently). Need to understand lpms better to understand why that is. Planning to follow up with clipping directly at T layer and hopefully understand this better.
emranemran
force-pushed
the
emran/force-session-reinit-catalyst
branch
from
fa5afe8 to
0f6ece2
Compare
When a sequence of segments has a discontinuity (e.g. encoded differently), the transcoder at T must be re-initialized so that transcode operations can be completed correctly. The T takes some liberties and assumes a sequence of segments are encoded the same way and thus, ends up not re-initializing certain parts of the pipeline (e.g. demuxer, etc) to save time and improve efficiency. This can result in corrupted segments with two video tracks instead of one audio and one video track.
For clipping, the first segment (index=0) is encoded differently from the rest of the file. As such, a session reinit must be forced at the T using a header set in the request to B.