Functional Stream Processing with Scalaz-Stream

Functional Stream
Processing
with Scala
Adil Akhter
1

The next 45 minutes
... is about Scalaz-Stream:
• Objectives.
• Design Philosophy.
• Core Building Blocks & their Semantics.
2

Task[A]
• Purely functional & compositional.
• Task[A] is a wrapper around Future[Throwable / A].
• Asynchronous: uses scalaz.concurrent.Future[A].
• Handles exceptions: with scalaz./, aka Either.
5

Task[A]
• Task separates what to do with the monadic context from when to do.
scala> Task{ println("Hello World!") }
res0: scalaz.concurrent.Task[Unit] = scalaz.concurrent.Task@2435b6e8
scala> res0.run
Hello World!
6

Task[A]
• Task separates what to do with monadic context from when to do.
scala> Task{ println("Hello World!") }
res0: scalaz.concurrent.Task[Unit] = scalaz.concurrent.Task@2435b6e8
scala> res0.run
Hello World!
scala> res0.run
Hello World!
7

Task[A]
• Example: Lifting computation into a Task:
import twitter4j._
val twitterClient = new TwitterFactory(twitterConﬁg).getInstance()
def statuses(query: Query): Task[List[Status]] =
Task {
twitterClient.search(query).getTweets.toList
}
8

Task Constructors
• def now[A](a: A): Task[A]
• def fail(e: Throwable): Task[Nothing]
• def fork[A](a: => Task[A]): Task[A]
• ...
9

Objectives
• Incremental IO & Stream Processing.
• Modularity & Composability.
• Resource-safety.
• Performance.
11

Canonical Example
• File IO using Scalaz-Stream1
:
val streamConverter: Task[Unit] =
io.linesR("testdata/fahrenheit.txt")
.ﬁlter(s => !s.trim.isEmpty && !s.startsWith("//"))
.map(line => fahrenheitToCelsius(line.toDouble).toString)
.intersperse("n")
.pipe(text.utf8Encode)
.to(io.ﬁleChunkW("testdata/celsius.txt"))
.run
streamConverter.run
1
Source: https://github.com/functional-streams-for-scala/fs2/blob/topic/redesign/README.md
12

Scalaz-Stream
... provides an abstraction to declaratively specify on how to obtain
stream of data.
16

Scalaz-Stream
... centers around one core type: Process.
17

Process[F, O]
... represents a stream of O values
which can interleave external requests
to evaluate expressions of form F[_].
18

Process[F[_], O]
• F[_]: a context/container (e.g., scalaz.concurrent.Task)
• effectful or not,
• in which a computation runs.
19

Process[F[_], O]
• Streams of Os is obtained from the context: F.
20

Process[F[_], O]
• Streams of Os is obtained from the context: F.
• Process can accept different input types (e.g., F[_]).
21

Execution Model
scala> :paste
// Entering paste mode (ctrl-D to ﬁnish)
import scalaz.stream._
import scalaz.concurrent.Task
import scala.concurrent.duration._
// Exiting paste mode, now interpreting.
22

Process
Process = Halt | Emit | Await
trait Process[F[_], O]
23

Halt
case class Halt(cause: Cause)
extends Process[Nothing, Nothing]
scala> Process.halt
res19: scalaz.stream.Process0[Nothing] = Halt(End)
24

Emit[F[_],O]
case class Emit[F[_],O](
head: Seq[O],
tail: Process[F, O]) extends Process[F, O]
25

Emit[F[_],O]
scala> import Process._
import scalaz.stream.Process._
scala> emit(1)
res1: scalaz.stream.Process0[Int] = Emit(Vector(1))
scala> emit(1).toSource
res2: scalaz.stream.Process[scalaz.concurrent.Task,Int] = Emit(Vector(1))
26

Emit[F[_],O]
scala> emit(1)
res1: scalaz.stream.Process0[Int] = Emit(Vector(1))
scala> emit(1).toSource
res2: scalaz.stream.Process[scalaz.concurrent.Task,Int] = Emit(Vector(1))
scala> val p: Process[Task, Int] = emitAll(Seq(1,5,10,20))
p: scalaz.stream.Process[scalaz.concurrent.Task,Int] = Emit(List(1, 5, 10, 20))
27

Await[F[_], I, O]
case class Await[F[_], I, O](
req: F[I],
recv: I Process[F, O],
fallback: Process[F, O] = halt,
cleanUp: Process[F,O] = halt) extends Process[F, O]
28

Await[F[_], I, O]
• Example: Build a Twitter Search Process to get tweets with #spark
Tag.
• Defined earlier:
val twitterClient: Twitter = ???
def statuses(query: Query): Task[List[Status]] = ???
29

Await[F[_], I, O]
// Builds a Twitter Search Process for a given Query
def buildSearchProcess(query: Query): Process[Task, Status] = {
val queryTask: Task[List[Status]] = statuses(query)
await(queryTask){ statusList
Process.emitAll(statusList)
}
}
//In essence, it builds: Process: Await (_ Emit Halt)
val akkaSearchProcess: Process[Task, Status] =
buildSearchProcess(new Query("#spark"))
30

Await[F[_], I, O]
• Example: Build a stream of Positive Integers.
import Process._
scala> def integerStream: Process[Task, Int] = {
| def next (i: Int): Process[Task,Int] = await(Task(i)){ i
| emit(i) ++ next(i + 1)
| }
| next(1)
| }
integerStream: scalaz.stream.Process[scalaz.concurrent.Task,Int]
31

Notable Approaches to Construct Process
• Process.eval
• Process.repeatEval
• scalaz.stream.io and scalaz.stream.nio
• scalaz.stream.tcp
• scalaz.stream.async
• ...
32

Running a Process: run
• def run: F[Unit]: constructs the machinery that will execute the Process.
• Reduces the sequence of operations to a single operation, in the context of Task. Given:
scala> val intsP = integerStream.take(10)
intsP: scalaz.stream.Process[scalaz.concurrent.Task,Int] = ...
scala> val task = intsP.run
task: scalaz.concurrent.Task[Unit] = scalaz.concurrent.Task@fbce6f9
scala> task.run
// outputs nothing?
33

Running a Process: runLog
• def runLog:F[Vector[O]]: Gets all the intermediate results.
scala> intsP.runLog.run
res4: Vector[Int] = Vector(1, 2, 3, 4, 5, 6, 7, 8, 9, 10)
34

Running a Process: runLog
• What does happen when we execute integerStream.runLog.run?
def integerStream: Process[Task, Int] = {
def next (i: Int): Process[Task,Int] = await(Task(i)){ i
emit(i) ++ next(i + 1)
}
next(1)
}
35

Other Approaches
• def runLast: F[Option[A]]
• def runFoldMap(f: A => B): F[B]
36

Transformations
// map
scala> integerStream.map(_ * 100).take(5).runLog.run
res5: Vector[Int] = Vector(100, 200, 300, 400, 500)
//ﬂatMap
scala> integerStream.ﬂatMap(i emit(-i) ++ emit(-i-1)).take(5).runLog.run
res6: Vector[Int] = Vector(-1, -2, -2, -3, -3)
// Zip
scala> val zippedP: Process[Task, (Int, String)]
| = integerStream.take(2) zip emitAll(Seq ("A", "B"))
scala> zippedP.runLog.run
res2: Vector[(Int, String)] = Vector((1,A), (2,B))
38

Process1[-I,+O]
• aka. Pipe
• Pure Transformations
• Process[F, A].pipe(Process1[A, B]) => Process[F, B]
scala> import process1._
scala> integerStream |> ﬁlter(i => i > 5) |> exists(_ == 10)
//res8: scalaz.stream.Process[Task,Boolean] = ...
40

Process1[-I,+O]
• type Process1[-I,+O] = Process[Env[I,Any]#Is, O].
• process1 object facilitates ﬁlter, drop, take and so on.
41

Channel[+F[_],-I,O]
• type Channel[+F[_],-I,O] = Process[F, I => F[O]].
• Process[F, A].through(Channel[Task, A, B]) =>
Process[Task, B].
42

Channel[+F[_],-I,O]
scala> val multiply10Channel: Channel[Task, Int, Int] =
| Process.constant { (i: Int) =>
| Task.now(i*10)
| }
scala> integerStream.take(5) through multiply10Channel
scala> res40.runLog.run
43

Sink[+F[_],-O]
• Process[F, A].to(Sink[Task, A]) => Process[Task,
Unit]
• type Sink[+F[_],-O] = Process[F, O => F[Unit]]
44

Sink[+F[_],-O]
scala> val printInts: Sink[Task, Int] =
| Process.constant { (i: Int) =>
| Task.delay { println(i) }
| }
scala> val intPwithSink: Process[Task, Unit]
| = integerStream to printInts
scala> intPwithSink.run.run
1
2
3
4
5
45

Example
Sentiment Analysis of Tweets
46

Source
• Building a source that queries Twitter in every 5 seconds.
def buildTwitterQuery(query: Query): Process1[Any, Query]
= process1 lift { _ query }
val source = awakeEvery(5 seconds) |> buildTwitterQuery(query)
47

Channel
• Building the Twitter Query Channel.
// deﬁned earlier
def statuses(query: Query): Task[List[Status]] = ???
// Lift the Task to construct a channel
val queryChannel: Channel[Task, Query, List[Status]]
= channel lift statuses
48

Channel
• Building the Sentiment Analysis Channel.
import SentimentAnalyzer._
def analyze(t: Tweet): Task[EnrichedTweet] =
Task {
EnrichedTweet(t.author, t.body, t.retweetCount, sentiment(t.body))
}
val analysisChannel: Channel[Task, Tweet, EnrichedTweet] =
channel lift analyze
49

Source |> Channel
def tweetsP(query: Query): Process[Task, Status] =
source through queryChannel ﬂatMap {
Process emitAll _
}
// source = awakeEvery(5 seconds) |> buildTwitterQuery(query)
50

Source |> Channel
def twitterSource(query: Query): Process[Task, String] =
tweetsP(query) map(status Tweet(status))
through analysisChannel // ... to analysis channel
map (_.toString)
51

Source |> Channel |> Sink
// http4s Websocket Route
case r @ GET -> Root / "websocket"
val query = new Query("#spark")
val src1: Process[Task, Text] = twitterSource(query)
.observe(io.stdOutLines)
.map(Text(_))
WS(Exchange(src1, Process.halt))
52

Example
Sentiment Analysis of Tweets
using Twitter's Streaming API
54

async.boundedQueue
val twitterStream: TwitterStream = ???
val tweetsQueue: Queue[Status]
= async.boundedQueue[Status](size)
def stream(q: Queue[Status]) = Task{
twitterStream.
addListener(twitterStatusListener(q))
twitterStream.sample()
}
55

async.boundedQueue
• Producer
Process.eval_(stream(tweetsQueue))
.run
.runAsync { _ println("All input data was written") }
56

async.boundedQueue
• Consumer
tweetsQueue.dequeue.map(status
Tweet(
author = Author(status.getUser.getScreenName),
retweetCount = status.getRetweetCount,
body = status.getText)
) through analysisChannel map (_.toString)
57

Tee
• type Tee[-I,-I2,+O] = Process[Env[I,I2]#T, O]
• Deterministic
• Left and then Right and then Left ...
scala> import tee._
scala> val p1: Process[Task, Int] = integerStream
scala> val p2: Process[Task, String] = emitAll(Seq("A", "B", "C"))
scala> val teeExample1: Process[Nothing, Any] = (p1 tee p2)(interleave)
scala> teeExample1.runLog.run
res2: Vector[Any] = Vector(1, A, 2, B, 3, C)
scala> val teeExample2: Process[Nothing, Int] = (p1 tee p2)(passL)
scala> teeExample2.take(5).runLog.run
60

Wye
• type Wye[-I,-I2,+O] = Process[Env[I,I2]#Y, O]
• Non-deterministic
• Left, right or both
61

Next Version: FS2
• Functional Streams for Scala
• New algebra.
• Improved abstraction.
• Better performance.
62

“There are two types of
libraries: the ones people hate
and the ones nobody use"
— Unknown
63

• Creadits: Edward Kmett, Runar Bjarnasson & Paul Chuisano et al.
• Sides and code samples will be posted at http://github.com/
adilakhter/scalaitaly-functional-stream-processing.
65

Functional Stream Processing with Scalaz-Stream

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Functional Stream Processing with Scalaz-Stream (20)

Recently uploaded (20)

Functional Stream Processing with Scalaz-Stream