reactive reference architecture -...

Reactive Reference Architecture

Markus Jura @markusjura

Nilanjan Raychaudhuri @nraychaudhuri

s

Big dataMassive Learning Platform

http://allthingsd.com/files/2013/07/minions-‐640x327.jpg

3

Principles of Reactive Systems

Architecture

4

Web (Play)

backend (Akka)

Journal (Cassandra)

Fast data processing (Spark)

http://cdn1.ssninsider.com/wp-‐content/uploads/2013/08/despicable-‐me-‐2-‐minions-‐e1375818315418.png

http://cdn1.ssninsider.com/wp-content/uploads/2013/08/despicable-me-2-minions-e1375818315418.png

Problem: Splitting application

Monolithic

7

Splitting MLP

8

Frontend nodes

Backend nodes

Akka Distributed Pub/Sub• Location transparency

• Publish/Subscribe

• Dynamically add/remove

9

Distributed Pub/Sub Mediator

10

Frontend nodes

Backend nodes

M M

M M M

Frontend

12

// Starting up the mediatorval mediator = DistributedPubSubExtension(system).mediator

// Sending message through mediatormediator ? DistributedPubSubMediator.Send(“/user/course-aggregate”, CreateCourse(course), localAffinity = true)

Backend

13

// Starting up the mediatorval mediator = DistributedPubSubExtension(system).mediator

// Setting up the servicesval courseAggregate = system.actorOf(CourseRepository.props, "course-aggregate")mediator ! DistributedPubSubMediator.Put(courseAggregate)val levelAggregate = system.actorOf(LevelRepository.props, "level-aggregate")mediator ! DistributedPubSubMediator.Put(levelAggregate)

Problem: Distributed State

State on single node

16

Frontend nodes

Backend nodes

State across nodes

17

Frontend nodes

Backend nodes

Akka Cluster Sharding

• Partition actors across Cluster

• Location transparency

• Rebalancing

18

How does it work?

19

Shard Coordinator

Shard Region 1 Shard Region 2

Backend: Create ShardRegion

21

// Backend: Start cluster sharing in non-proxy mode ClusterSharding(system).start( "users", Some(User.props), UserSharding.idExtractor, UserSharding.shardResolver())

Backend: Shard Resolver & Id Extractor

22

// Extract the user shard// Rule of thumb: 10 times the max numbers of sharding nodesdef shardResolver(shardCount: Int = 20) = { case UserService.Envelope(email, _) => email.hashCode % shardCount.toString }

// Extract the id and payload of the user actorval idExtractor: ShardRegion.IdExtractor = { case UserService.Envelope(email, payload) => (email, payload)}

Frontend

23

// Send message via shard regionshardRegion ? UserService.Envelope(email, GetUser(email))

// Start sharding in proxy mode on every frontend node val shardRegion: ActorRef = ClusterSharding(system).start( "users", None, UserSharding.idExtractor, UserSharding.shardResolver())

Problem: Loosing State

State in memory

25

Frontend nodes

Backend nodes

Persist state

26

Frontend

Backend

Akka Persistence

• Stores state in Journal DB

• Event Sourcing

• Restore state

27

Write user state

29

class User extends PersistentActor {

// User id / sharding extractor id override def persistenceId: String = self.path.name

// Persist UserCreated event to Cassandra journal override def receiveCommand: Receive = { case UserService.CreateUser(userProgress) => persist(UserCreated(userProgress)) { userCreated => // Update user state onUserCreated(userCreated) sender() ! UserService.UserCreated } } ...

Recover user state

30

...

// Replay UserCreated message to recover stateoverride def receiveRecover: Receive = { case e: UserCreated => onUserCreated(e)}

Problem: Replicate Cache

Singleton Cache

33

Frontend nodes

Backend nodes

C

What is CRDT?

34

Conflict-free replicated data type is a type of specially designed data structure used to achieve strong eventual consistency

Great talk on CRDT by Sean Cribbs: Eventually Consistent Data Structures

https://vimeo.com/43903960

CvRDT

35

CvRDT - Example

36

CmRDT

37

Akka Data replication• CRDT for Akka cluster

• Replicated in-memory data store

• Supports many types: (GSet, ORSet, PNCounter…)

38

CRDT Replicator

39

Frontend nodes

Backend nodes

C CR R

R R R

Frontend

41

//starting up the replicatorval replicator = DataReplication(system).replicator

//get cache entryval f = replicator ? Get(key, ReadLocal, None)

Backend

42

//starting up the replicatorval replicator = DataReplication(system).replicator

//updating the cache replicator ! Update(Course.cacheKey, GSet.empty, WriteLocal)

(_ + courseCreated.course)

Q & Option[A]

@nraychaudhuri@markusjura

44

React.js

frontend

REST API Actors

backend

Cache

PubSub CRDTPersistenceCluster Sharding

journal

reactive reference architecture -...

Documents