yondonfu
commented
1 year ago Context: We expect Mist to be the primary source of segments and receiver of transcoded renditions for a go-livepeer B (as a part of Catalyst which is Mist + go-livepeer B bundled together).
Mist currently ignores the audio tracks of transcoded renditions because it only uses the source audio track. Given this, I see this as a non-issue since the audio tracks of the transcoded renditions will be ignored anyway. In fact, I actually don't think we need the audio verification code as well. In the scenario, where we support audio transcoding in the future, the audio verification code would need to look different anyways vs. the code right now that tries to fuzzy match to ensure that the audio tracks in transcoded renditions are the same as the audio track in the source.
Describe the bug Sometimes, renditions will have some audio packets missing, and, potentially, mixed. The reason for that needs to be investigated. As a temporary workaround, fuzzy audio verification is implemented. That verification approach is vulnerable to trivial packet mixing attack.
To Reproduce There are some samples in failed verifications bucket.