Add an initial revision of an audio server.

services/media/audio/audio_output_manager.h

+// Copyright 2015 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef SERVICES_MEDIA_AUDIO_AUDIO_OUTPUT_MANAGER_H_
+#define SERVICES_MEDIA_AUDIO_AUDIO_OUTPUT_MANAGER_H_
+
+#include <set>
+
+#include "base/synchronization/lock.h"
+#include "base/threading/sequenced_worker_pool.h"
+#include "mojo/services/media/common/interfaces/media_common.mojom.h"
+#include "mojo/services/media/common/interfaces/media_pipe.mojom.h"
+#include "services/media/audio/audio_output.h"
+#include "services/media/audio/fwd_decls.h"
+
+namespace mojo {
+namespace media {
+namespace audio {
+
+class AudioOutputManager {
+ public:
+  explicit AudioOutputManager(AudioServerImpl* server);
+  ~AudioOutputManager();
+
+  // Initialize the output manager.  Called from the service implementation,
+  // once, at startup time.  Should...
+  //
+  // 1) Initialize the mixing thread pool.
+  // 2) Instantiate all of the built-in audio output devices.
+  // 3) Being monitoring for plug/unplug events for pluggable audio output
+  //    devices.
+  MediaResult Init();
+
+  // Blocking call.  Called by the service, once, when it is time to shutdown
+  // the service implementation.  While this function is blocking, it must never
+  // block for long.  Our process is going away; this is our last chance to
+  // perform a clean shutdown.  If an unclean shutdown must be performed in
+  // order to implode in a timely fashion, so be it.
+  //
+  // Shutdown must be idempotent, and safe to call from the output manager's
+  // destructor, although it should never be necessary to do so.  If the
+  // shutdown called from the destructor has to do real work, something has gone
+  // Very Seriously Wrong.

[jeffbrown, 2015/11/04 23:43:33]

Instead of this comment, maybe just make it blow up.

[johngro, 2015/11/06 02:20:25]

I already have some DCHECK protection for the idempotency requirement.  Concrete
suggestions for a universal DCHECK approach to enforcing safe-to-call from
output manager destructor requirement are welcome.  I'm all about DCHECKing my
internal invariants, but this one seems like a difficult one to universally
express in code.

+  void Shutdown();
+
+  // Select the initial set of outputs for a track which has just been
+  // configured.
+  void SelectOutputsForTrack(AudioTrackImplPtr track);
+
+  // Schedule a closure to run on our encapsulating server's main message loop.
+  void ScheduleMessageLoopTask(const tracked_objects::Location& from_here,
+                               const base::Closure& task);
+
+  // Shutdown the specified audio output and remove it from the set of active
+  // outputs.
+  void ShutdownOutput(AudioOutputPtr output);
+
+ private:
+  // TODO(johngro): A SequencedWorkerPool currently seems to be as close to what
+  // we want which we can currently get using the chrome/mojo framework.  Things
+  // which are missing and will eventually need to be addressed include...
+  //
+  // 1) Threads are created on the fly, as needed.  We really want to be able to
+  //    spin up the proper number of threads at pool creation time.  Audio
+  //    mixing is very timing sensitive...  If we are in a situation where we
+  //    have not hit the max number of threads in the pool, and we need to spin
+  //    up a thread in order to mix, we *really* do not want to have to wait to
+  //    create the thread at the OS level before we can mix some audio.  The
+  //    thread needs to already be ready to go.
+  // 2) Threads in the pool are created with default priority.  Audio mixing
+  //    threads will need to be created with elevated priority.
+  // 3) It is currently unclear if explicitly scheduling tasks with delays will
+  //    be sufficient for the audio mixer.  We really would like to be able to
+  //    have tasks fire when some handle abstraction becomes signalled.  This
+  //    will let us implement mixing not only with open loop timing, but also
+  //    with events which can come from device drivers.  Note: this seems to be
+  //    an issue with the TaskRunner architecture in general, not the
+  //    SequencedWorkerPool in specific.
+  // 4) The resolution of posted delayed tasks may hinder the low latency goals
+  //    of the system.  Being able to know the underlying achievable resolution
+  //    of dispatching delayed tasks is a minimum requirement.  From that, we
+  //    can compute our worst case overhead which can be communicated to the
+  //    user and will have an effect on overall latency.  Hitting something on
+  //    the order of 10s of microseconds is what we really should be shooting
+  //    for here.  Single milliseconds is probably too coarse.
+  // 5) Not a requirement, but a sure-would-be-nice-to-have... Scheduling
+  //    delayed tasks using absolute times.  Having to schedule using delta
+  //    times from now means that we need to take the time it takes to schedule
+  //    the task (along with its jitter) into account when scheduling.  This can
+  //    lead to additional, undesirable, latency.
+  scoped_refptr<base::SequencedWorkerPool> thread_pool_;

[jeffbrown, 2015/11/04 23:43:33]

Alternately if the number of audio outputs is small then perhaps it would be ok
to give each output its own dedicated thread.  That's kind of nice from the
perspective of treating audio outputs as autonomous agents so that init and
shutdown can also be managed asynchronously just like processing work.

[johngro, 2015/11/06 02:20:24]

Acknowledged.
We may end up at this point.  Per-our face to face discussion, this does not
eliminate OSAL issues.  I'm making a good faith effort here to use the exiting
Mojo OS abstractions, and to use the message-centric architecture that the Mojo
team seems to prefer and has encouraged me to try.

If it turns out that hard data shows that this message-centric architecture is
not sufficiently performant, and I cannot make it so by working with the Mojo
team, I will move in the direction of a message free event/signal architecture
(generally, my preferred architecture).  Currently, Mojo has limitations when it
comes OS abstraction which prevent me from doing this.  This is orthogonal to
the thread pool vs. not thread pool issue.  Even if I do not use a thread pool,
I already have this problem with attempting to use Mojo as my OSAL.  Before
resorting declaring the Mojo framework to be incapable of the task, I am going
to attempt to use it as is, and work with the team to bring the framework to the
point where it will meet my requirements.

This list above is an initial list of where I have some issues with the support
provided by the framework.  While I am willing to make a best effort attempt to
meet in the middle regarding performance and timing requirements, I do have some
hard requirements.  I feel that these are requirements not frivolous.  If they
cannot (or will not) be met by the framework's capabilities, I will reluctantly
embark upon the task of rolling-my-own solution.  It has not come to that point
yet, and I hope it does not.

+
+  // A pointer to the server which encapsulates us.  It is not possible for this
+  // pointer to be bad while we still exist.
+  AudioServerImpl* server_;
+
+  // Our set of currently active audio output instances.
+  //
+  // Contents of the output set must only be manipulated on the main message
+  // loop thread, so no synchronization should be needed.
+  AudioOutputSet outputs_;
+};
+
+}  // namespace audio
+}  // namespace media
+}  // namespace mojo
+
+#endif  // SERVICES_MEDIA_AUDIO_AUDIO_OUTPUT_MANAGER_H_