Re: Audio Workers - please review

On Thu, Sep 11, 2014 at 11:54 AM, Joseph Berkovitz <joe@noteflight.com>
wrote:

> Hi Ehsan,
>
> It’s great that you provided this critique of what will be one of the most
> important evolutions of the spec. Chris has provided what I think is a good
> fine-grained initial response to your thoughts, and I hope the group
> devotes some teleconference time to any outstanding issues that remain.
>
> For my part I want to provide a coarse-grained, developer-oriented opinion
> on some of the issues you raised. Some of these may require more work in
> the proposed spec because I think your questions cast light on some
> problems in the language we’re using in the current AudioWorkerNode
> proposal.
>
> 1. Processing model and timing issues. I don’t think we need to nail down
> much about the exact semantics of onaudioprocess callbacks, nor do I think
> developers can really discover (or usefully exploit) any sort of global
> timing coordination between onaudioprocess and on message/postMessage(). I
> think it may be useful to shift our perspective, and start talking about
> onaudioprocess as being an asynchronous message from a developer’s
> perspective, not an synchronous message. The spec currently uses the word
> “synchronous” to describe onaudioprocess, which leads down a dangerous path
> for developer thinking.
>
> The fact that onaudioprocess is synchronous within something we currently
> call the “audio thread” is really an optimization, not an aspect of the
> specification. I think we should perhaps have the spec say something more
> like, “The UA calls onaudioprocess in a manner that attempts to best
> optimize latency and freedom from glitching."
>
> When developers implement a node, onaudioprocess can be considered as a
> special type of async message to the AudioWorker originating from the UA,
> with no need for guarantees about its timing or notions of synchronicity
> with an invisible concept like the audio thread. Developers should be
> encouraged to write nodes in a style that considers each onaudioprocess to
> be an asynchronous, granular little processing job, not something that is
> happening in some known, definite time relationship to anything else. I
> think it’s dangerous to say anything that attempts to nail down the
> behavior and timing of the audio thread (and in the future, probably that
> will be audio threadS plural). Onaudioprocess “quasi-messages” simply
> arrive in a guaranteed order and cover a contiguous sequence of time
> ranges, starting at some context time when the node starts and ending when
> the node stops.
>
> I also think we need to say something about the sequencing and
> monotonically-extending time ranges covered by successive onaudioprocess
> calls.
>

Firstly, please note that DOM events have synchronous semantics.  But of
course that tells us nothing about when these events get dispatched.  As
Chris previously described, his intention is that the audio worker
dispatches these events to all of the worker nodes in the graph
sequentially for each block.  That would take away the chance of the UA
running some worker nodes in parallel if the output of neither one is a
direct or indirect input to the others, but it's also a good thing, in that
dispatching these events asynchronously will create latency that we cannot
avoid.  That latency is an unfortunate property of the current
ScriptProcessorNode.  How do we avoid such latency if we adopt an
asynchronous processing model as you described above?


> 2. Exposure of the WebWorker interface. It seems to me that we can follow
> two paths here: a) eliminate postMessage/onmessage support, do not expose
> WebWorker, and rely on AudioParams alone for all main-thread communication
> with scripted nodes, or b) retain postMessage/onmessage support, in which
> case it seems to me we are actually better off inheriting the full Web
> Worker interface to avoid building a new animal which looks like a
> WebWorker but doesn’t walk or talk like one.
>
> Since the recent inclusion of add/removeParameter in the proposal, did
> anyone (especially Chris :-) consider whether we still truly need
> postMessage/onmessage support? If we removed it, it would render moot a lot
> of arguments about what happens when nodes try to talk to each other and
> might simplify everything a lot. Internode communication seems to me a way
> to cause a lot of mistaken assumptions re: synchronicity (see above).
> AudioParams seem cleaner and more in line with what native nodes do.
>

There are definitely use cases for sending arbitrary messages to the
worker.  Such messages can contain information such as "the user fired a
gun on the main thread", so that the worker can start outputing a
synthesized gunshot noise in the case of a game, for example.  But I
definitely agree that the current postMessage() API is too permissive (it
effectively makes it possible for you to post arbitrary MessagePorts around
on these workers, for example.)


> 3-5. AudioParam transferrability: I can’t really see the use case for
> AudioParam transferability. AudioParams seem to be the preferred channel
> for communication between the main thread and scripted nodes, and they
> stand alone in supporting that communication. Why would we transfer them in
> a separate message?
>

As I stated before, what happens when you modify the state of an AudioParam
on the main thread after handing it off to a worker node?  At least, the
semantics in that case need to be specified!



>
> .            .       .    .  . ...Joe
>
> *Joe Berkovitz*
> President
>
> *Noteflight LLC*
> Boston, Mass.
> phone: +1 978 314 6271
> www.noteflight.com
> "Your music, everywhere"
>
>
> On Sep 10, 2014, at 12:29 PM, Ehsan Akhgari <ehsan@mozilla.com> wrote:
>
> Hi Chris, and everyone else on the WG!
>
> I took a look at the merged version of this proposal and have some
> feedback, I would appreciate if you can please let me know what you think.
> As you know, I haven't been closely involved with the spec for a while, so
> if the below reflects my ignorance on some of the previously discussed
> matters, please accept my apologies in advance.  And of course, sorry that
> this is a *long* list.  :-)
>
> 1.  I think the processing model can use some further clarification.  The
> current text of the spec is very vague on a number of different issues, so
> I would appreciate if you could please clarify what you intended to happen
> in the following cases:
>
> a) It is not really clear what should happen after you call terminate() on
> AudioWorkerNode, since that seems to be orthogonal to whether the node
> participates in the graph.
>
> b) The interaction between worker closing/termination and the audio
> processing is not clear at all.  Specifically, dedicated workers typically
> have a long time that is as long as the last pending task (assuming that
> the main thread doesn't hold a reference to them), however, for these
> workers, we need to keep firing audioprocess events.
>
> c) How would different AudioWorkerNodes allocated through different
> AudioContext objects interact with each other?  (Here is an example of
> something that is not clear, should it be allowed to send an AudioParam
> belonging to a node form context 1 to a worker node from context 2?  The
> spec is currently silent about this.)
>
> d) What is the exact semantics of the audioprocess events dispatched to
> these workers?  Do they block running all audio processing on the audio
> processing thread?  Note that because of the halting problem, we cannot
> even verify that the worker script will ever terminate, let alone in a
> reasonable amount of time, so in practice the UAs need to preempt the
> execution of the script.  It would be nice if this was somehow mentioned in
> the spec, at least in terms of what the UA needs to do if it decides to
> abort the execution of one of these scripts.
>
> e) What is the order in which these worker nodes receive the audioprocess
> events (assuming obviously that one such node is not an indirect
> input/output of the other)?  Note that with the currently specified API, I
> think it is possible for one of these workers to send a MessagePort through
> the main thread to another one, and therefore be able to communicate with
> the other AudioWorkerNode workers through MessagePort.postMessage, so the
> order of execution is observable to script.  (I admit that I don't have a
> good solution for this -- but I also don't know what use cases for
> transferring information back and forth between these nodes and the main
> thread we're trying to address here.)
>
> 2. I feel very strongly against exposing this node type as a web worker.
> I think that has a number of undesired side effects.  Here are some
> examples:
>
> a) Even though the HTML web worker termination algorithm says nothing
> about killing an underlying worker thread, I think that is what authors
> would typically expect to happen, but obviously doing that would not be an
> option here.  It is also not specified what needs to be output from the
> node after terminate() has been called on it (the input unmodified?
> silence? something else?)  Also, given the fact that you can already
> disconnect the node from the graph, why do we need the terminate() method
> in the first place?
>
> b) The API gives you the illusion that multiple AudioWorkerNode's will run
> on different DedicatedWorkerScopes.  That, in Web Workers world, would mean
> different threads, but that will not be the case here.  I think that
> discrepancy is problematic.
>
> c) Using DedicatedWorkerGlobalScope is a bit weird in terms of APIs that
> are exposed to workers.  For example, should these workers support
> onlanguagechange?  What about IndexedDB on workers?  What about nested Web
> Workers?
>
> d) At least on Gecko, Web Workers have specific implementation concerns in
> terms of their message queue, event processing model and so on.  It might
> be a lot of effort for us to properly implement the observable behavior of
> these workers running inside our audio processing thread code (which has a
> completely different event processing model, etc.)
>
> e) The topic of whether or not synchronous APIs must be allowed on workers
> is being debated on public-script-coord, and it seems like there is no
> consensus on that yet.  But I find the possibility of running synchronous
> XHR on the audio processing thread unacceptable for example, given its
> realtime requirements.
>
> I think tying this node to Web Workers opens a huge can of worms.  It will
> also keep biting us in the future as more and more APIs are exposed to
> workers.  I am wondering if we can get away with a completely different API
> that only tries to facilitate the things that authors would typically need
> to do during audio processing without attempting to make that a Web Worker?
>
> I think at the very least, if we really want to keep tying these concepts
> to Web Workers, it might be worthwhile to bring that up on
> public-script-coord, since it is at least bending the original use cases
> that Web Workers were designed for.  :-)
>
> 3. What is the purpose of addParameter and removeParameter?  It seems to
> me that if we defined a structured clone algorithm for AudioParam (which is
> another detail missing from the spec that needs to be clarified anyway) and
> make it transferable, then the author would be able to postMessage() an
> AudioParam just as easily.  Is there a good reason to have a specialized
> method for what is effectively posting an AudioParam to the worker?  (Note
> that we'd probably need to add methods to AudioParam for extracting a-rate
> and k-rate values at any given time in that case, so that the worker script
> can get the right values when it needs them.)
>
> 4. More on the idea of transferring an AudioParam to a worker, that will
> probably involve some kind of neutering the object.  It might make sense to
> introduce a clone() method on AudioParam as a way for authors to be able to
> keep a copy of the object around on the main thread.  That could of course
> be a future enhancement idea.
>
> 5. Still more on the idea of transferring an AudioParam, another thing
> that we need to worry about is what happens if you try to transfer an
> AudioParam that is currently being used somehow (either through being a
> property on another AudioNode, or being used as an input to one.)
>
> 6. I think addressing #3 above will allow us to completely eliminate
> AudioProcessEvent, and just use AudioProcessingEvent.
>
> I hope the above is useful.  I think there is much more to think about
> especially given what we end up deciding on the first two items above, so I
> hope to provide further feedback as we make progress here.
>
> Please let me know what you think!
>
> Cheers,
> Ehsan
>
>
> On Mon, Aug 25, 2014 at 11:29 AM, Chris Wilson <cwilso@google.com> wrote:
>
>> I've done some tweaking to the Audio Worker (issue #113
>> <https://github.com/WebAudio/web-audio-api/issues/113>) proposal, and
>> most significantly added the ability to create AudioParams on Audio Workers
>> (issue #134 <https://github.com/WebAudio/web-audio-api/issues/134>).
>>
>> The fork is hosted on my fork (http://cwilso.github.io/web-audio-api/).
>>  Start here
>> <http://cwilso.github.io/web-audio-api/#widl-AudioContext-createAudioWorker-AudioWorkerNode-DOMString-scriptURL-unsigned-long-numberOfInputChannels-unsigned-long-numberOfOutputChannels>
>> to review the creation method, and the bulk of the text begins at
>> http://cwilso.github.io/web-audio-api/#the-audio-worker.
>>
>
>
>
>


-- 
Ehsan