warn on event loop blocking

[davidmarkclements]

http://npm.im/blocked is nice

@mcollina which was yours again?

there's also http://npm.im/toobusy but it's a native compile mod

let's use whichever adds minimum overhead (and maybe ignore native compile)

[davidmarkclements]

or.. should this not be in mu

cc @mcollina @mcdonnelldean @pelger

[mcollina]

You can use https://github.com/mcollina/loopbench.

[davidmarkclements]

you know if we do think this is worth doing in mu,

I think we need something that's light browserside - that means no event emitter (I'd like to avoid EE and streams for basic cases browser side) - loopbench uses EE, blocked does not and it's smaller

blocked does use process.hrtime though, easier to polyfill (maybe even use performance API in modern browsers, fallback to Date)

[mcollina]

This sounds more a transport issue to me, and I would not want to have this running in the background in the browser.

A transport should not receive a message if the event loop is busy.

[davidmarkclements]

ah - the way I was thinking about was actually just as a warning system - as feedback for user implementation (say they spend too long in a define callback or something) - this would apply to both browser and server - and only be on in dev mode

if the event loop is busy - how would a transport even be able to receive the message?

[mcollina]

If the event loop is busy, it's better not to accept any new messages from outside, otherwise it will become more busy. Eventually, all the user will start reissuing commands, creating even more load.

[davidmarkclements]

ah I see - throttling incoming data if if the event loop reaches a certain threshold - isn't that basically backpressure in a different way?

hmm - setting the threshold for that will be hard, I think we need benchmarks etc. before we can be realistic about that part

so there's really two issues

1. event-loop... backpressure(?)
2. dev mode event loop usage user feedback

[mcollina]

We can pick 500ms. If the event loop has not run for 500ms, something is definitely wrong.