It ain't about the callbacks, it's about the flow control (2013) (idea.popcount.org)

I think that, like programming in general, there are different styles that click for different people. Different analogies, like the "push" vs "pull" described in this article come with different trade offs, and different people see them more easily than others.

In reality, I just want more options. I know that synchronous IO on UI threads is bad, but I also know that I can avoid bottlenecks by not pulling from slow IO sources. Choice means that a problem can be expressed in the most natural way to a programmer or a team, which ultimately helps convey the ideas better than trying to work in an unnatural convention.

I love Ruby primarily because you can usually accomplish tasks multiple ways. You can usually find the most natural way _to you_ and use that, over, say a prescribed "best" way.

I don't quite get the mapping of "pull" and "push" and "synchronous" and "callbacks".

    foo.asyncPush(callback);
    foo.asyncPull(callback);

and

    const result = await pull();
    await push(data);

Both feel similar in terms of idioms, capabilities, etc... I know that async/await was not a thing in 2013 JavaScript, but as a pattern (promises, futures, etc...) it's been around for some time.

the important part of flow control is to accept enough work to keep yourself busy, but not more than you can eventually retire at that rate, and not more than would require holding more intermediate state that you have room for. [edit: those are really two sides of the same coin...ideally you would also come close to saturating some throughput limit (network/disk/memory bandwidth). and even better come close to saturating all of them]

if the work includes tasks with variable delay (using the disk, or an external network service), or uses variable amount of compute, then you need to look at creating appropriate adaptation and overall scheduling strategies (i.e. retire older work first)

so while having a queue depth of 1 and implementing everything as fully blocking in the naive threading case is maybe a better place to start than having an infinite queue depth in the naive callback implementation....neither one really gets you to where you need to go without further structure.

the only real frustration I have with the threaded approach in trying to build the 'right' system is that at some point you have to deal with the system scheduler being largely workload oblivious.

once you build you own scheduler, then whether you are assuming synchronous returns from a programming perspective really is a matter of style, and probably involves more language-provided safety and less boilerplate.

I still like highland.js to solve these issues

I agree that not having a natural flow of control is more prone programming debugging. However, he does say things that are wrong, ex: "It's impossible to slow down the pace of accepting new incoming client connections." You can slow down the pace by doing the following:

     const http = require('http');
     let count = 0;
     const THRESHOLD = 1;
     const server = http.createServer((req, res) => {
       if (count++ < THRESHOLD) {
         // do work
         res.write('success');
         res.end();
         setTimeout(() => count--, 10 * 1000);
       } else {
         res.write('limit');
         res.end();
         count--;
       }
     });
     server.listen(10000);