Arc Forumnew | comments | leaders | submitlogin
2 points by shawn 2141 days ago | link | parent

We tried. It was almost a disaster.

We started out by making an EventSource endpoint and using JS EventSource object to get updates. Seemed to work great.

Pushed it live.

To everyone on the site. Not just to the people on /l/place. Oops. Mistake #1.

Mistake #2: I was aware that Arc kills threads that take longer than 30 seconds, but I was unaware that EventSource automatically tries to reconnect. Welcome to the DoS party.

With no way to tell the clients to stop. Mistake #3.

It was only live for about 10 minutes, but it took around two hours before the last few clients finally refreshed their browser and stopped DoSing the endpoint. The server absorbed the extra traffic like a champ, but the server logs were ridiculous for about an hour. Lots of "client took too long" error messages spit out by Arc every second.

Sooo yeah, submitting changes via JS is easy. Getting updates is the hard part.

Could poll. But then that's still hammering the server. Moreso than it already is.

It's probably a feature that it's painful to use. Only the determined will use it for art.

That said, if anyone has ideas for how to tweak arc for performance for this scenario – and remember to calculate how much memory is used by how many simultaneous sockets are open – I'd be willing to try again.



3 points by i4cu 2141 days ago | link

Reads like a really good horror story :)

> I was aware that Arc kills threads that take longer than 30 seconds...

> Sooo yeah, submitting changes via JS is easy. Getting updates is the hard part.

That's why I suggested a post request...

Just deliver the page and create an api end point, on the server, that gives you the last x changes from the change log.

With that ajax request, you could return these net updates for the client and use client side js to apply them to the dom. You could even create a trivial js debounce[1] algo to throttle requests if you wanted to go even further.

1. 'debounce' for those who don't know -> https://davidwalsh.name/javascript-debounce-function Only I'm suggesting it as a way to throttle ajax reqs.

-----

3 points by shawn 2139 days ago | link

Yeah, it was worth the effort to make the drawing code client side. Someone showed up and made a software rasterizer.

Timelapse: https://www.youtube.com/watch?v=G0sG08jKCTI

-----

2 points by krapp 2137 days ago | link

why not go all the way and render client-side in a canvas as well?

As it is, the more "pixels" you have, the slower the page loads. You could probably have just a single canvas and a single event handler.

You could preserve functionality for non-js users by using an imagemap (assuming browsers still even support those) or sticking the existing code in noscript tags.

-----

2 points by krapp 2136 days ago | link

OK... I don't know where I got "single event handler from," that obviously wouldn't be true, but it would still be simpler than having each pixel be a separate HTML form....

-----

1 point by shawn 2136 days ago | link

Wanna hop into discord and we can chat about the design? https://discord.gg/qaqkc9z

The repo's open, and I could help get you set up with it if you wanted to work on some of this. https://github.com/laarc/laarc

It should be a matter of:

  git clone https://github.com/laarc/laarc
  cd laarc
  make
  rlwrap bin/arc
  ((load "news.arc"))
and then http://localhost:8080/place

-----

2 points by krapp 2136 days ago | link

Let me actually see if I can come up with a POC first, at least for the canvas and javascript. Realistically, since it's just GET requests that don't need authentication, I shouldn't even need laarc running, I can just send the same requests by AJAX.

-----

2 points by shawn 2136 days ago | link

https://www.laarc.io/place.json is the API, by the way! I should probably mention that somewhere.

-----