Gap in multi-period MPD causes hang

[l-law]

Shaka Player v2.0.0-beta2-debug (http://shaka-player-demo.appspot.com/demo/) on Chrome 49, Windows 7, failed to play a static MPD with multiple periods, it will hang after playing a period.
It is possible to seek outside the current period, eg. seek to period 2 while playing period 1, then shaka-player will start from the seek point and play until end of period 2, but will not play period 3.

Test assets can be found at https://drive.google.com/file/d/0BymF_2aNp_hBWmtmaUprNmZ2M2c/view?usp=sharing
manifest.mpd immediately inside the zip file is the MPD that caused playback hang.

Not sure if it is related to #381

[l-law]

Sorry there is one attribute put in the wrong place in the uploaded mpd, in Event tag, "timescale" should not be put there but should be put in EventStream.

Correct mpd should look like this:

        <EventStream timescale="10000000" schemeIdUri="urn:scte:scte35:2014:xml+bin">
            <Event presentationTime="60000000" duration="100000000" id="1">
                <scte35:Signal>
                    <scte35:Binary>
                        /DAvAAAAAAAA///wFAVAAAABf+/+A5OHAP4F9eEAADwAAAAKAAhDVUVJAAAAAECJXy8=
                    </scte35:Binary>
                </scte35:Signal>
            </Event>
        </EventStream>

But this doesn't affect the outcome, it will get stuck at period end.

[joeyparrish]

Thanks, we'll take a look. Odds are good that there is a gap at the start of the second period, which is something we do not currently support.

[joeyparrish]

Sorry we lost track of this one. It appears that you have a gap in your content. According to Chrome's buffered time ranges, Period 0 ends at 89.061333, but Period 1 begins at 89.24. The references from the manifest seem to indicate that the final segments of Period 0 should go all the way to 89.24, but they do not.

Gaps in content is something we do not support at this time. Any gap larger than what the browser will jump automatically will currently cause a hang such as this one. As far as we know, a gap of 150ms or less should be skipped by all major browsers, so that's our effective tolerance.

[l-law]

Thanks for the information.

I tried to edit the manifest such that Period 1 to start at 89.061333, then shaka can play through the transition from Period 0 to Period 1. Therefore this "gap" should be the main reason that caused playback hang.

However, this "gap" only exists in audio streams, for video stream in Period 0 it should end exactly at the time Period 1 starts. If there is only one single period, then all content could be played even if there is "gap" between audio and end of period(ie. only video is played in the end). I would expect shaka-player should also play all content in the case with multiple periods.

Here I have a few questions:

1. Sometimes adaptation sets have different end time, in order to use shaka-player to smoothly play across periods, period start time should be set to the earliest end time of any adaptation set of previous period?
2. If there is overlap between periods, will the content be skipped or played?
3. Will shaka-player support gaps in content in the future?

[joeyparrish]

1. Yes, for gapless playback, the next Period would start at the earliest end time of all AdaptationSets. However, if there are large (> 150ms) differences in the length of AdaptationSets, you may have an issue with your encoder. It may make more sense to fix the encoder than to hide the problem by changing the start time of the next Period.
2. If there is overlap between Periods, the browser should skip it because MSE's appendWindowEnd should cut off everything past the end of the Period.
3. We do not currently support gaps in content and we do not currently have it on our roadmap.

[joeyparrish]

@l-law, is there anything else we can do for you on this topic?

[joeyparrish]

Closing due to inactivity.