New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
MSE-in-Workers: Cross-thread function and tests #26109
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The review process for this patch is being conducted in the Chromium project.
EcoInfra sheriff: this PR needs the src change to not be flakey, which hasn't landed in Dev; please admin-merge it. |
Ack. Will do once its merged in Chromium. |
dd3e9e0
to
34f0dab
Compare
Includes implementation of CrossThreadMediaSourceAttachment (CTMSA) and updates to MediaSource and SourceBuffer to use it. CTMSA owns a lock that it takes on most operations. It also provides a utility method (RunExclusively()), which is used by the MSE API to take that same lock during most operations on the API. The same thread attachment provides a basic version of the same, though no lock is used. CTMSA understands whether or not it is attached, has ever started closing, and whether or not either of the attached contexts has ever destructed. It conditions cross-thread operations to safely behave when one or both contexts is being destructed. MediaSource is given its own small lock that it uses to guard its reference to the attachment instance, when attached. This is required because attachment start is synchronous on the main thread, even if the MediaSource is owned by worker context. CTMSA and MediaSource cooperate to ensure mutex and safety of the two-stage attachment start. In MediaSource::ContextDestroyed(), further care is taken to prevent attachment start success when worker context destruction is possibly racing the main thread's attempt to start attachment. `context_already_destroyed_`, protected by MediaSource's lock, is used to prevent that start from succeeding. MediaSource's context destruction can no longer always assume that accessing the demuxer via the WebMediaSource (and WebSourceBuffers) is safe. The worker version of MediaSource context destruction uses the attachment's RunExclusively() to safely know if it can cleanly Close the underlying demuxer, or if instead it must do a simpler shutdown. Future specification work will likely determine some signal to the media element when the worker-owned MediaSource's context is shutdown, yet the element itself is still alive. For now, sane results are returned by the attachment (nothing buffered nor seekable), with no other error provided. Possible app workarounds might include main and worker watchdogs and being careful about when the main thread explicitly terminates the worker. This experimental implementation attempts to retain BackgroundVideoOptimization, even for MSE-in-Workers, but does not support AudioVideoTracks in the worker MediaSource or SourceBuffer. Plumbing of appropriately-identified track metadata parsed from initialization segment in the worker is used to populate (and remove) media element AudioVideoTracks that should agree with the track id's used internally in WebMediaPlayerImpl to accomplish BackgroundVideoOptimization. As a simplification, CTMSA assumes it can be used for at most one attachment. Effectively, this assumes that the currently on-by-default MediaSourceRevokeObjectURLOnAttach feature behavior will be always-on soon. If CTMSA is not unregistered automatically after successful attachment start (e.g., if that feature is disabled) and if the app attempts to re-use the corresponding object URL for a later attachment, that attachment will be rejected by CTMSA. Note that the same-thread attachment would still allow such re-use when that feature is disabled. Updated web-test: mediasource-worker-attach.html changed to ...-play.html and fixed New web-test: mediasource-worker-play-terminate-worker.html Manual test: page refreshing while running mediasource-worker-play*.html BUG=878133 Change-Id: I21f6542b90d51bdc28096500fb1441d202ab4ee8 Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2459431 Commit-Queue: Matthew Wolenetz <wolenetz@chromium.org> Reviewed-by: Will Cassella <cassew@google.com> Cr-Commit-Position: refs/heads/master@{#818174}
34f0dab
to
96adc48
Compare
Includes implementation of CrossThreadMediaSourceAttachment (CTMSA) and
updates to MediaSource and SourceBuffer to use it.
CTMSA owns a lock that it takes on most operations. It also provides a
utility method (RunExclusively()), which is used by the MSE API to take
that same lock during most operations on the API. The same thread
attachment provides a basic version of the same, though no lock is used.
CTMSA understands whether or not it is attached, has ever started
closing, and whether or not either of the attached contexts has ever
destructed. It conditions cross-thread operations to safely behave when
one or both contexts is being destructed.
MediaSource is given its own small lock that it uses to guard its
reference to the attachment instance, when attached. This is required
because attachment start is synchronous on the main thread, even if the
MediaSource is owned by worker context. CTMSA and MediaSource cooperate
to ensure mutex and safety of the two-stage attachment start. In
MediaSource::ContextDestroyed(), further care is taken to prevent
attachment start success when worker context destruction is possibly
racing the main thread's attempt to start attachment.
context_already_destroyed_
, protected by MediaSource's lock, is usedto prevent that start from succeeding.
MediaSource's context destruction can no longer always assume that
accessing the demuxer via the WebMediaSource (and WebSourceBuffers) is
safe. The worker version of MediaSource context destruction uses the
attachment's RunExclusively() to safely know if it can cleanly Close the
underlying demuxer, or if instead it must do a simpler shutdown.
Future specification work will likely determine some signal to the media
element when the worker-owned MediaSource's context is shutdown, yet the
element itself is still alive. For now, sane results are returned by the
attachment (nothing buffered nor seekable), with no other error
provided. Possible app workarounds might include main and worker
watchdogs and being careful about when the main thread explicitly
terminates the worker.
This experimental implementation attempts to retain
BackgroundVideoOptimization, even for MSE-in-Workers, but does not
support AudioVideoTracks in the worker MediaSource or SourceBuffer.
Plumbing of appropriately-identified track metadata parsed from
initialization segment in the worker is used to populate (and remove)
media element AudioVideoTracks that should agree with the track id's
used internally in WebMediaPlayerImpl to accomplish
BackgroundVideoOptimization.
As a simplification, CTMSA assumes it can be used for at most one
attachment. Effectively, this assumes that the currently on-by-default
MediaSourceRevokeObjectURLOnAttach feature behavior will be always-on
soon. If CTMSA is not unregistered automatically after successful
attachment start (e.g., if that feature is disabled) and if the app
attempts to re-use the corresponding object URL for a later attachment,
that attachment will be rejected by CTMSA. Note that the same-thread
attachment would still allow such re-use when that feature is disabled.
Updated web-test:
mediasource-worker-attach.html changed to ...-play.html and fixed
New web-test:
mediasource-worker-play-terminate-worker.html
Manual test: page refreshing while running mediasource-worker-play*.html
BUG=878133
Change-Id: I21f6542b90d51bdc28096500fb1441d202ab4ee8
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2459431
Commit-Queue: Matthew Wolenetz <wolenetz@chromium.org>
Reviewed-by: Will Cassella <cassew@google.com>
Cr-Commit-Position: refs/heads/master@{#818174}