ExoPlayer stucks when you plug/unplug headset for a little time

[vshcryabets]

Issue description

I'm using it to play live video content from live cameras. And i found small issue - when you plug/unplug headset ExoPlayer stucks for a little time (probably 100-200ms). Video captured from phone - https://drive.google.com/file/d/16Q2kLvAcF10cWw0akB0SW-AXotSc71Ky/view?usp=sharing. Video stucked at 00:04-00:10

Reproduction steps

• open ExoPlayer demo app
• click on YouTube DASH
• click on Google Play (MP4, H264)
• app will strat playback, try to plug/unplug headset

Link to test content

Test content used from ExoPlayer demo app - http://www.youtube.com/api/manifest/dash/id/3aa39fa2cc27967f/source/youtube?as=fmp4_audio_clear,fmp4_sd_hd_clear&sparams=ip,ipbits,expire,source,id,as&ip=0.0.0.0&ipbits=0&expire=19000000000&signature=A2716F75795F5D2AF0E88962FFCD10DB79384F29.84308FF04844498CE6FBCE4731507882B8307798&key=ik0

Version of ExoPlayer being used

2.6.0

Device(s) and version(s) of Android being used

Device: Moto X Pure Edition XT1575 with Android 6.0

A full bug report captured from the device

bugrepoprt.txt

[ojw28]

@erdemguven - Did you manage to reproduce this? @vshcryabets - Do you see this only on the Moto X, or on all devices? Thanks.

[vshcryabets]

@ojw28 same behaviour on - Nexus 5x and Nexus 6P

[erdemguven]

I have reproduced it on Nexus 6 with Lollipop.

[ojw28]

I think this is most likely to be caused by the different audio output paths have different latencies. Since we need to synchronize the video to the audio, if the latency of the audio output path changes, it's correct that the video should adjust itself so that it's still in sync with audio playing through the new path.

@andrewlewis has been looking a bit at audio output path latency, so might be able to say more (or confirm + close this issue).

[andrewlewis]

The platform audio track position can stall briefly when the audio routing changes, which is probably what's going on here. As part of fixing #3841 the duration of the freeze may be reduced, as we should poll the timestamp more frequently after getTimestamp returns false, so please follow that issue for updates.