This brief presentation covers the basis of writing asynchronous code using callbacks, promises and coroutines.

1. The Art of Concurrent
Programming
Past, Present and Future
Iskren Chernev

2. What is this talk about
● What is parallel, concurrent, asynchronous
● Brief history
● Run Loops
● Callbacks
● Promises
● Coroutines

3. Parallel, Concurrent, Asynchronous
● Parallel - 2 pieces of code running at the same time (threads/processes
running different CPUs, cores)
● Concurrent - the begining and end of 2 pieces of code overlap, but not
necessarily run at the same time (could be parralel, but not necessarily). So it
appears to overlap, but it doesn’t have to
● Asynchronous - an operation that will complete at a later time, but its
execution doesn’t block the thread/process
● We’ll cover Asynchronous programming in this lecture

4. Brief overview of the environment
● Obviously this talk is opinionated, so here’s what I have in mind
● server-side business logic applications
● little amount of computation
● A lot of IO (talking to other (micro)-services and databases)
● A huge volume of “easy” requests
● Client side is also mostly event driven with little computation so most of talk
applies there too

5. Brief History
● Single Threaded software
● Mutli-core brought with it Multi Threaded Software
○ (+) Its very close to single threaded software
○ (-) slow context switch, creation
○ (-) cache trashing
○ (-) need for synchronization primitives -- this is complicated
● With Node.js (2009) came the hype around asynchronous, single threaded
programming
● After that most other popular languages added asynchronous primitives and
libraries as standard, or popularized and extended existing asynchronous
primitives (Java, C#, Objective-C, Python)

6. So what is this Node.js all about
var server = http.createServer((request, response) => {
response.end( 'It Works!! Path Hit: ' + request.url);
})
server.listen(PORT, () => {
console.log("Server listening on: http://localhost:%s" , PORT);
})

7. Node.js - basic
function doSomethingSync(args) {
r = doSlow1(args);
return doSlow2(r)
}
function doSomething(args, done) {
doSlow1(args, (err, res) => {
if (err) return done(err);
doSlow2(res, done);
});
}
● Easy function creation with “closures” is a must
● Every async computation calls a given function after finished

8. What else - let’s try an if
function doSomethingSync(args) {
if (cond) {
r = doSlow1();
} else {
r = 5;
}
doFast(r);
return doSlow2();
}
WTF
function doSomething(args, done) {
var doRest = (r, done) => {
doFast(r)
doSlow2(done)
}
if (cond) {
return doSlow1((err, res) => {
if (err) return done(err)
doRest(arg, done)
})
}
doRest(5, done);
}

9. Key takeaways from Node.js
● Its great because it showed the world
what asynchronous code looks like and
how to do it with callbacks
● It sucks to actually code it (there is
async.js but its still pretty painful)
● Is here a better way?
● (pro) its very fast
● (pro) all libraries support callback async
interface
● (cons) writing correct async-callback
code is tricky to do right (easy to
double call the callback, or not call it at
all, or forget to return when calling done
● (cons) and its ugly (deep nesting for a
chain of async operations)

10. Run Loops
● Something like a thread manager
● Operations a run loop should implement
○ schedule(fn) -- this function only is enough
to implement basic concurrent systems
with multithreaded blocking IO (main
thread)
○ scheduleLater(when, fn) -- similar to
schedule but executes at a later point
○ addSocketWithEvents(sock, evmask, fn)
- register for socket events
○ NOTE: that linux support for async DISK io
is spotty, and is mostly wrapped with
thread pool under the hood
● Run loops are the basis of asynchronous
programming
● You can have one or many, and
communicate between them using
schedule
● Node.js has a run-loop built in, but other
languages added those later
● Libraries should explicitly support run
loops (if there are many), and most “old”
languages, “old” libraries are blocking
● You could shim any blocking operation
with a thread pool and a single main
run-loop

11. What is a Promise
● Called in different ways : Future,
CompletionStage (Java), Task (C#) etc
● Its an object representing the result of
an asynchronous operation
● It supports chaining
function doSomethingSync(args) {
if (cond) {
r = doSlow1();
} else {
r = 5;
}
doFast(r);
return doSlow2();
}
function doSomething(args) {
if (cond) {
r = doSlow1();
} else {
r = 5;
}
Promise.resole(r).then((r) => {
doFast(r);
return doSlow2();
});
}

12. Promises -- loops
function doSomethingConc(n) {
var futures = []
for (var i = 0; i < n; ++i) {
futures.push(doSlow(i));
}
// returns a new future, which
// resolves to an array of results
return Promise.all(futures);
}
function doSomethingSeries(n) {
var future = Promise.resolved(null);
var every = (i) => {
future = future.then(() => doSlow(i));
}
for (var i = 0; i < n; ++i) {
every(i);
}
return future;
}

13. Promisses - notes
● Error propagation
● If you need to exit early from a chain you have to use exceptions
● Compared to callbacks -- promisses guarantee listener is called at most once
● So callbacks are still necessary if you need to call listener many times
● In some languages (C++, some JS libs), promise and future are separate object
● In typed languages (C++, Java, C#) you need a few different APIs on the promise to handle async
and sync returns
● Normally the listener is called right after its produced, on the same thread, but you can change
that
● Promises are hard to implement -- use an official/stable lib, do NOT do it yourself

14. Coroutines
● Coroutines are mostly syntax sugar on top
of promises, but it pays off!
● The observation is that mostly you have a
linear chain of promises, so in that case
you can use a keyword await (C#, python)
/ yield (python)
● It handles linear code, ifs and serial
loops
● You can always fall back to promises if
what you’re doing is more complicated
(branches off, parralel loops)
● To implement you need support in the
language
async def doStuff(x):
if cond:
r = async doSlow(x)
else:
r = 5
doFast(r)
return async doSlow()