node.js? done.scala!days2011.scala-lang.org/sites/days2011/files/56. node...implicit def...

node.js?done.scala!Implementing Scalable Async IO using

Delimited Continuations

Tiark Rompf, EPFL

Saturday, June 4, 2011

who has heard about node.js?


6/3/11 7:18 AMnode.js

Page 1 of 4http://nodejs.org/

Evented I/O for V8 JavaScript.

An example of a web server written in Node which responds with"Hello World" for every request.

var http = require('http');http.createServer(function (req, res) { res.writeHead(200, {'Content-Type': 'text/plain'}); res.end('Hello World\n');}).listen(1337, "127.0.0.1");console.log('Server running at http://127.0.0.1:1337/');

To run the server, put the code into a file example.js and executeit with the node program:

% node example.jsServer running at http://127.0.0.1:1337/

Here is an example of a simple TCP server which listens on port1337 and echoes whatever you send it:

var net = require('net');

var server = net.createServer(function (socket) { socket.write("Echo server\r\n"); socket.pipe(socket);});

DownloadChangeLogAboutv0.4.8 docs

WikiBlogCommunityDemoJobs

Hash InsightsChickRx


purely asynchronous

callback driven

no blocking operations

good scalability


6/3/11 7:22 AMGoogle Trends: erlang, node.js, clojure,

Page 1 of 1http://www.google.com/trends?q=erlang%2C+node.js%2C+clojure%2C&ctab=0&geo=all&date=all&sort=0

[email protected] | Signout

erlang, node.js, clojure,

Search Trends

Tip: Use commas to compare multiple search terms.

Searches Websites All regions All years

Scale is based on the average worldwide traffic of erlang in all years. Learn more

erlang 1.00 node.js 0.08 clojure 0.12

Rank by erlang

Scaling Large Projects With ErlangSlashdot - Jul 6 2008

The AZ of Programming Languages: ErlangComputerworld - Jun 16 2009

The AZ of Programming Languages: ClojureComputerworld - Aug 10 2009

Erlang Solutions, Ltd. and Basho Technologies, Inc. EnterBroad Partnership to Deliver Scalable, Fault TolerantApplications to a Global MarketCNNMoney.com - Mar 30 2010

Programming ClojureSlashdot - May 17 2010

Erlang and OTP in ActionSlashdot - Dec 8 2010

More news results »

Regions

1. Sweden

2. RussianFederation

3. SouthKorea

4. SouthAfrica

5. India

6. Singapore

7. Ukraine

8. Indonesia

9. China

10. Philippines

Cities

1. Goteborg,Sweden

2. Stockholm,Sweden

3.Moscow,RussianFederation

4. Beijing,China

5.SanFrancisco,CA, USA

6. Shanghai,China

7. Guangzhou,China

8. Seattle,WA, USA

9. Delhi, India

10. Bogota,Colombia

Languages

1. Swedish

2. Korean

3. Russian

4. Chinese

5. English

6. Indonesian

7. Danish

8. Hungarian

9. Japanese

10. German

Export this page as a CSV file

Google Trends provides insights into broad search patterns. Please keep in mind that several approximations areused when computing these results.

©2008 Google - Discuss - Terms of Use - Privacy Policy - Help


why would Scala programmers care?


#1: it’s about scalability


#2: no equivalent Scala libs


a little experiment: port it over to Scala!


done.scala

• a minimal scala port of node.js

• only net, http and supporting modules

• very incomplete (and not at all done!)


an interesting exercise:

JavaScript Scaladynamic language statically typed


• added type annotations in a few places

• helper classes for anonymous objects{ host: ‘google.com’, port: 80 }

• overloading, default arguments and varargs instead of dynamic typeof

quite smooth overall:


use Java NIO underneath


http.createServer(function (req, res) { res.writeHead(200, {'Content-Type': 'text/plain'}) res.end('Hello World\n')}).listen(8000, "127.0.0.1")console.log('server running')

http.createServer { (req, res) => res.writeHead(200, "Content-Type" -> "text/plain") res.end("Hello World\n")}.listen(8000, "127.0.0.1")println("server running")

JavaScript

Scala


now what?


node’s conscious choice:only callback-driven apis

but is it the right choice?


small examples are fine but does it scale?


println(“hello”)setTimeout(1000) {

println(“world”)}




hello<pause>world


setTimeout(1000) {println(“world”)

}println(“hello”)


setTimeout(1000) {println(“world”)

}println(“hello”)

hello<pause>world


for (x <- List(1,2,3) {setTimeout(1000) {

println(“found ” + x)}

}


for (x <- List(1,2,3) {setTimeout(1000) {

println(“found ” + x)}

}

<pause>found 1found 2found 3


try {setTimeout(1000) {

throw new Exception}

} catch {case e =>

println(“caught: ” + e)}


try {setTimeout(1000) {

throw new Exception}

} catch {case e =>

println(“caught: ” + e)}

?Saturday, June 4, 2011

many things no longer work as expected


callback-driven programmingis really hard


val host = “http://search.twitter.com/”val query = “#scaladays”

http.get(host,80,“search.atom?q=”+query) { res => println(“STATUS:” + res.statusCode) res.onData { chunk => println(“BODY:” + chunk) } res.onEnd { println(“DONE”) }}


http://search.twitter.com


val host = “http://search.twitter.com/”val query = “#scaladays”

http.get(host,80,“search.atom?q=”+query) { res => println(“STATUS:” + res.statusCode) res.onData { chunk => println(“BODY:” + chunk) } res.onEnd { println(“DONE”) }}

STATUS: 400BODY: <data>BODY: <more data>DONE




val host = “http://search.twitter.com/”val queries = List(“#scala”,“#scaladays”, “@odersky”)

val results = for (q <- queries) yield { http.get(host,80,“search.atom?q=”+query) { res => ... }}

println(results)





val results = for (q <- queries) yield { http.get(host,80,“search.atom?q=”+query) { res => ... }}

println(results)

List(ClientRequest, ClientRequest, ClientRequest)




can we do something about it?


execute asynchronously

program synchronously


scalac -P:continuations:enable


def foobar: Int @suspendable = shift { retrn: (Int=>Unit) => retrn(7)}

def abort: Unit @suspendable = shift { retrn: (Unit=>Unit) => // just don’t return...

}

reset {println(“A”)abortprintln(“B”)

}println(“C”)


def foobar: Int @suspendable = shift { retrn: (Int=>Unit) => retrn(7)}

def abort: Unit @suspendable = shift { retrn: (Unit=>Unit) => // just don’t return...

}

reset {println(“A”)abortprintln(“B”)

}println(“C”)

AC


def setTimeout(delay: Int)(callback: => Unit): Unit = ...







hello<pause>world



def sleep(delay: Int) = shift { retrn: (Unit=>Unit) => setTimeout {

retrn()}

}



println(“hello”)sleep(1000)println(“world”)

hello<pause>world



def sleep(delay: Int) = shift { retrn: (Unit=>Unit) => setTimeout {

retrn()}

}



println(“hello”)sleep(1000)println(“world”)

hello<pause>world

hello<pause>world


http.get(...) { res => res.onData { chunk => println(“BODY:” + chunk) } res.onEnd { println(“DONE”) }}



val res = http.get(...)

for (chunk <- res.data) { println(“BODY:” + chunk)}

println(“DONE”)



val res = http.get(...)

for (chunk <- res.data) { println(“BODY:” + chunk)}

println(“DONE”)

def data = new { def foreach(yld: Buffer=>Unit) = shift { retrn: (Unit=>Unit) =>

onData(yld)onEnd(retrn)

}def mkString = {

val s = new StringBuilder; foreach(s+=_); s.result}

}Saturday, June 4, 2011


val results = for (q <- queries) yield {val res = http.get(host,80,“search.atom?q=”+query)val xml = XML.loadString(res.data.mkString)xml \ "entry" \ "title" text

}

println(results)





val results = for (q <- queries) yield {val res = http.get(host,80,“search.atom?q=”+query)val xml = XML.loadString(res.data.mkString)xml \ "entry" \ "title" text

}

println(results)

<console>:11: error: no type parameters for method map: (f: (java.lang.String) => B)(implicit bf: scala.collection.generic.CanBuildFrom[List[java.lang.String],B,That])That exist so that it can be applied to arguments ((java.lang.String) => String @scala.util.continuations.cpsParam[Unit,Unit])





val results = for (q <- queries.suspendable) yield {val res = http.get(host,80,“search.atom?q=”+query)val xml = XML.loadString(res.data.mkString)xml \ "entry" \ "title" text

}

println(results)





val results = for (q <- queries.suspendable) yield {val res = http.get(host,80,“search.atom?q=”+query)val xml = XML.loadString(res.data.mkString)xml \ "entry" \ "title" text

}

println(results)

List(List(“#Scala highlighted in article from @TheEconomist on parallel programing and multicore: http://econ.st/jkGchu @typesafe”, ...), List(“There are so many great talks today at #scaladays; it's hard to decide which to attend!”, ...), List(“@odersky is there any info on Cascade up somewhere?”, ...))




http://twitter.com/#!/search?q=%23Scala

http://twitter.com/#!/search?q=%23Scala

http://twitter.com/TheEconomist

http://twitter.com/TheEconomist

http://econ.st/jkGchu

http://econ.st/jkGchu

http://twitter.com/typesafe

http://twitter.com/typesafe

http://twitter.com/#!/search?q=%23scaladays

http://twitter.com/#!/search?q=%23scaladays

implicit def richIterable[A,Repr](xs: IterableLike[A,Repr]) = new { def suspendable = new {

def foreach(yld: A => Unit @suspendable): Unit @suspendable = {val it = xs.iteratorwhile (it.hasNext) yld(it.next)

}def map[B, That](f: A => B @suspendable)

(implicit bf: CanBuildFrom[Repr, B, That]): That @suspendable = {val b = bf(xs.repr)foreach(x => b += f(x))b.result

}}

}

val results = for (q <- List(1,2,3).suspendable) yield {sleep(1000); println(“tick”)

}


implicit def richIterable[A,Repr](xs: IterableLike[A,Repr]) = new { def suspendable = new {

def foreach(yld: A => Unit @suspendable): Unit @suspendable = {val it = xs.iteratorwhile (it.hasNext) yld(it.next)

}def map[B, That](f: A => B @suspendable)

(implicit bf: CanBuildFrom[Repr, B, That]): That @suspendable = {val b = bf(xs.repr)foreach(x => b += f(x))b.result

}}

}<pause>tick<pause>tick<pause>tick

val results = for (q <- List(1,2,3).suspendable) yield {sleep(1000); println(“tick”)

}



val results = for (q <- queries.suspendable) yield {println(“start ”+q)val res = http.get(host,80,“search.atom?q=”+query)val xml = XML.loadString(res.data.mkString)println(“done ”+q)xml \ "entry" \ "title" text

}





val results = for (q <- queries.suspendable) yield {println(“start ”+q)val res = http.get(host,80,“search.atom?q=”+query)val xml = XML.loadString(res.data.mkString)println(“done ”+q)xml \ "entry" \ "title" text

}

start #scaladone #scalastart #scaladaysdone #scaladaysstart @oderskydone @odersky




def spawn(body: => Unit @suspendable) = {eventLoop.submitTask(() => reset(body))

}

class DataFlowCell[A] {private[this] var value: Option[A] = Noneprivate[this] var queue: List[A=>Unit] = Nil

def apply() = shift { retrn: (A=>Unit) =>value match {

case Option(v) => retrn(v)case None => queue ::= retrn

}}

def set(v: A) = value match {case Option(_) => assert(false, “can’t set value twice”)case None => value = Some(v); queue.foreach(f => spawn(f(v)))

}}


def future[A](body: => A @suspendable) = {val cell = new DataFlowCell[A]spawn { cell.set(body) }cell

}

def par[A,B](a: => A @suspendable)(b: => B @suspendable) = {val (u,v) = (future(a), future(b))(u(),v())

}

par { sleep(1000); println(“a”)

} { sleep(1000); println(“b”)

}println(“done”)


def future[A](body: => A @suspendable) = {val cell = new DataFlowCell[A]spawn { cell.set(body) }cell

}

def par[A,B](a: => A @suspendable)(b: => B @suspendable) = {val (u,v) = (future(a), future(b))(u(),v())

}

par { sleep(1000); println(“a”)

} { sleep(1000); println(“b”)


<pause>badone


implicit def richParIterable[A,Seq,Repr](xs: ParIterableLike[A,Seq,Repr]) = new {def suspendable = new {

def foreach(yld: A => Unit @suspendable): Unit @suspendable = {val futures = xs.seq.map(x => future(yld(x)) // sequential list of futuresfutures.suspendable.foreach(_.apply())

}def map[B, That](f: A => B @suspendable)(... bf ...): That @suspendable = {

val futures = xs.seq.map(x => future(yld(x))futures.suspendable.map(_.apply())

}}

}

val results = for (q <- List(1,2,3).par.suspendable) yield {sleep(1000); println(“tick”)



implicit def richParIterable[A,Seq,Repr](xs: ParIterableLike[A,Seq,Repr]) = new {def suspendable = new {

def foreach(yld: A => Unit @suspendable): Unit @suspendable = {val futures = xs.seq.map(x => future(yld(x)) // sequential list of futuresfutures.suspendable.foreach(_.apply())

}def map[B, That](f: A => B @suspendable)(... bf ...): That @suspendable = {

val futures = xs.seq.map(x => future(yld(x))futures.suspendable.map(_.apply())

}}

}

<pause>tickticktickdone

val results = for (q <- List(1,2,3).par.suspendable) yield {sleep(1000); println(“tick”)




val results = for (q <- queries.par.suspendable) yield {println(“start ”+q)val res = http.get(host,80,“search.atom?q=”+query)val xml = XML.loadString(res.data.mkString)println(“done ”+q)xml \ "entry" \ "title" text

}





val results = for (q <- queries.par.suspendable) yield {println(“start ”+q)val res = http.get(host,80,“search.atom?q=”+query)val xml = XML.loadString(res.data.mkString)println(“done ”+q)xml \ "entry" \ "title" text

}

start #scalastart #scaladaysstart @oderskydone #scaladone #scaladaysdone @odersky




Conclusions

• Scala needs good networking libs

• asynchronous execution does not meancallback-driven programming

• continuations can re-introduce synchronictiy


node.js? done.scala!days2011.scala-lang.org/sites/days2011/files/56. node...implicit def...

Documents

Transcript of node.js? done.scala!days2011.scala-lang.org/sites/days2011/files/56. node...implicit def...