[whatwg] ServiceWorker bottle-neck design flaw - MS Edge to the rescue?

Richard Maher maherrj at googlemail.com
Sat Mar 3 22:42:09 PST 2018


EDIT 1

Ok, it might have been Safari/Webkit but reading this
<https://webkit.org/blog/8090/workers-at-your-service/> it looks like 1
process for all ServiceWorker threads plus another to monitor/start/stop,
but only one simultaneous SW instance per domain (plus that dodgy iFrame
logic webkit has got going on)

So if neither Webkit or Edge do what my memory tells me then is it not a
good performance idea? (Barring reduced hit rate on SW recycling)

EDIT 2

On third thoughts, the fact that everything in a ServiceWorker is
Asynchronous or verbotten means Fetch, Push, Travel, events can all
multiplex to the same SW instance without much performance degradation.

Oh Well, I'm back to "Stuffed if I know why no one is implementing the
TravelManager :-("

FWIW The WebKit implementation of shared memory buffer must surely
necessitate a Synchronous API? Breaking one of the major premises of SW
design?

EPILOG

I apologise for spamming the recipients here even though I did it
knowingly. I won't do it anymore. I didn't believe I was reaching an
audience so I just kept pulling leavers and pushing any and all buttons
available.

I have been made aware that, unbeknownst to me, some of the right people
were hearing what I had to say and the TravelManager idea was gaining
traction. To those people let me just say that I spent a fair bit of time
on the Brotkrumen Web App which demarcates UA Developer and Web Developer
contributions succinctly, provides smooth Google Map marker movement via
CSS transitions to get the mug punters looking and, most importantly, is a
fully working, fully documented, complete, sscce which provides real-time
feedback on the ratio of ServiceWorker instances to Geolocation updates!

If there is a open/tolerant forum where people discuss such things then
please let me know. Once again, *no one* can tell me there's anything wrong
with the idea. OTOH I can tell you heaps of things that are wrong with the
"alternative" WakeLock or Sensor ideas.

I also have a few good ideas for permissions/visibility on Background
Geolocation if anyone is interested.

BTW when I said repeatedly that I am happy to "go away" it was on the
proviso that the two issues I've been banging on about for two years were
implemented.

1) Broadcast notifications where the client/UA can subscribe to a Topic.
Once again W3C let us down but thanks to FCM we can access this much needed
functionality so I let it go.

2) Background Geolocation. As ShamWow guys says "It sells itself". Why,
WHY, *WHY* are people saying NO to this?


Cheers Richard


On Fri, Mar 2, 2018 at 11:22 AM, Richard Maher <maherrj at googlemail.com>
wrote:

> As you may be aware, I have been lobbying strongly here
> <https://stackoverflow.com/questions/44233409/background-geolocation-serviceworker-onmessage-event-order-when-web-app-regain>,
> and here
> <https://stackoverflow.com/questions/45206149/expected-ratio-of-serviceworker-instances-to-geolocation-updates>,
> and especially over there
> <https://github.com/w3c/ServiceWorker/issues/745> for the introduction of
> a background geolocation capability with Web Apps.
>
> So frustrated had I become that I spent the time to develop and document a
> POC example with everything UA developers and Web App developers need to do
> to implement background geolocation - documented sscce here
> <https://drive.google.com/open?id=0B7Rmd3Rn8_hDNW1zSWRoXzBTclU> and, to
> date, no one can tell me there’s anything wrong with it.
>
> In this technical feed-back vacuum my mind began to ponder why, and then
> drifted to “Cui Bono?”, and ultimately to conspiracy theories. For example:
> - Phonegap/Cordova developers bribing W3C to sandbag BackgroundGeolocation
> efforts.
>
> But only today did it dawn on me like a bolt of lightning to the brain!
> There is a fundamental design flaw with the current ServiceWorker
> design/implementations in that there is only ONE per UA. Jake Archibald et
> al have been so blinkered with offline-first, cache, and the browser being
> a proxy-server that they will not tolerate anything else monopolizing the
> SW thread to the detriment of their fetch performance stats.
>
> But God Bless Edge, notwithstanding their tardy arrival on the scene, who
> I believe have specialist SW instances for PUSH, FETCH, A.N.Other. This is
>  *fantastic*!!!
>
> Just add another specialist instance “GeoLocation” and we’re away.
>
> Please don’t let self-interest stand in the way of this essential
> functionality. Let’s go!
>
> But my question here is this: - Can someone please confirm that Edge
> implements multiple, specialist SW instances?
>
> Cheers Richard Maher
>
>


More information about the whatwg mailing list