DRM with key rotation: Black flicker observed when keys rotate

[ojw28]

Spun out from #1298. When playing a stream with key rotation, the display flickers black at the points when keys rotate. Every three minutes in this sample:

{
  "name": "Sintel - 3 minutes key rotation",
  "uri": "https://storage.googleapis.com/playerinfra/wv/sintel_3min_rotate.mpd",
  "drm_scheme": "widevine",
  "drm_multi_session": true,
  "drm_license_url": "https://widevine-proxy.appspot.com/proxy"
}

This is because the secure video decoder is (incorrectly) released and re-instantiated.

There also appears to be an audio underrun on some devices, although this may be more of a performance problem with some devices as opposed to an ExoPlayer issue.

[ueno-yuhei]

Thank you for the new issue.
If you have found something here I will send comments and actual code to this issue.

[jboisjo]

any updates for this issue?

[ojw28]

We will provide an update when there is one (lack of update == there is no update). Thanks.

[kvillnv]

Same issue here, looking forward for an update as well. Thanks

[adelati]

After reading the document about Key Rotation on Android (https://storage.googleapis.com/wvdocs/Widevine_DRM_Android_Using_Key_Rotation.pdf). to the best of my knowledge, the bug is related to the moment of releasing the initial sessionId1 which lead to tearing down the MediaCodec and creating new one. So, I think somewehere in Exoplayer library, you destruct the initial session.

[linhai326]

Originally, i thought @cizarco was right: obviously, the current implementation closes the primary/first session when opening a new one, which doesn't match the document's design. But even if I don't release any session, this issue still happens. So it seems the cause was somewhere else.

[ojw28]

@cizarco is right. Just commenting out releasing of sessions isn't sufficient to fix it. More in depth changes are required (e.g. to ensure MediaCodecRenderer doesn't tear down the MediaCodec).

[ojw28]

Fixed in dev-v2 for API level 23 and above. There's still a little pause when key rotation occurs, however.

[ojw28]

The pause is tracked by #4133, so closing this one.

[ojw28]

As per my comment above, this is fixed in dev-v2 for API level 23 and above. The issue is tricky to fix in a general way on earlier API levels, and we will likely not be investing time into doing so.

Note there's still a small pause when key rotation occurs. This is tracked by #4133.

---

Aside: I've marked the extended discussion that occurred in this issue after it was closed as off topic, so that it's easy to understand for those reading it.