"Microservices" is one of the hottest buzzwords and, as usual, everyone wants them, but few know how to build them. In this talk we will offer our interpretation of microservice architecture, and show how we are implementing these ideas: using Scala, Akka, sbt and Docker, we modularized Akka applications, Spark jobs and Play servers. In the talk we will discuss design trade-offs and challenges that we faced in the process, and how we have overcome them. The focus is not on particular features of Scala language or a library, but on building modern applications using the Typesafe stack and other open-source tools.

1. Building Micro-Services with Scala
Lior Shapsa, Yardena Meymann
September 22, 2014
© 2014 VMware Inc. All rights reserved.

2. Traditional Architecture

3. Traditional Architecture

4. Problems with Monoliths
• Size
• Integration (Conway’s law)
• Modules Change at Different Rates
• Dependencies Collisions
• Scalability requirements of modules
FEAR OF CHANGE
FEAR OF INNOVATION

5. Micro Services

6. Thought Leaders
Netflix, Amazon,
LinkedIn, Google,
ThoughtWorks, Gilt,
…
Fred George
Martin Fowler
Kit Colbert
Ben Fathi
Stefan Tilkov
VMware

7. What are micro-services?
A set of narrowly focused,

8. What are micro-services?
A set of narrowly focused,
independently deployable services,

9. What are micro-services?
A set of narrowly focused,
independently deployable services,
talking via uniform interfaces

10. What are micro-services?
A set of narrowly focused,
independently deployable services,
talking via uniform interfaces

11. Micro-services
• Each running in its own process
• Communicating with lightweight
mechanisms, often an HTTP
resource API
• Built around business capabilities
• Independently deployable
– fully automated deployment
• May be in a different programming language and use
different data storage technologies.

12. Micro-services
DEPLOYING A CHANGE IS LOW RISK

13. #NonBlocking

14. Threaded vs. Evented Servers
• Monolithic
– Most modules communicate in-process
• Micro-services
– communicate remotely, often over the network
– need to support cheap, lightweight remote communication
• Threaded servers use thread per connection
• Evented servers use non-blocking IO and callbacks
– Netty
– Node.js
– Play
– Spray

15. services vs. seconds
http://www.eecs.berkeley.edu/~rcs/research/interactive_latency.html
https://www.youtube.com/watch?v=1-vcErOPofQ

16. Threaded servers
Service A Service B Service N

17. Threaded servers
Service A Service B Service N
Thread Thread Thread

18. Threaded servers
Service A Service B Service N
Thread Thread Thread
Thread Thread Thread

19. Threaded servers
Service A Service B Service N
Thread Thread Thread
Thread Thread Thread
Thread Thread Thread

20. Thread
Service A Service B Service N
Thread
Thread
Threaded servers
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread Thread Thread

21. Thread
Service A Service B Service N
Thread
Thread
Threaded servers
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread
Thread Thread Thread

22. #NonBlocking – Evented servers
Service A Service B Service N
Thread Thread Thread
callback callback

23. #NonBlocking – Play/Spray
• Based on Akka using Netty
– Non-blocking programming much easier
• Easy deployment
– Embedded Server
• Rich JSON and HTTP support
• Plugins
– Secure Social
– Caching
– ….
• SCALA!

24. #NonBlocking – Play example
…

25. #AKKA

26. Backend
Pre-process Analyze

27. Spray
Node
Dispatch
Crawling
library
Netty
Price Engine
Head
Actor
…
Pages
Level1 Level1
Level2
…
Level2
File
writer
Head
Actor
…
k/v
Node
Akka Cluster
curl http://AkkaCluster/site_dot_com?op=start

28. #AKKA
• Based on the Actor Model
– http://www.reversim.com/2014/04/summit-2014-scale-up-your-thinking.html
• Resilience
• Location Transparency
– Cluster sharding
– Cluster clients
• Load Balancing
• Message Queue Integration

29. #Containers

30. What is ?
• Dockers are like lightweight VM for a single process
– Self contained
– No hypervisor
– Shared kernel, but Isolated and content agnostic
– No lib conflict
– No OS boot time
– Distributing a change is easy, using δ
• A Clean, Safe, Isolated and Portable Micro Service

31. Docker File
• Docker Image is Built From a DockerFile
• Consists of Descriptive Set of Instructions
– Start from a base image
– Run and command
– Add a file or directory
– Create an environment variable
– What process to run on launch

32. SBT
• Assemble “Dockerized” micro services
• We started with sbt-native-packager
– Poor Docker functionality
• We ended up with sbt-docker plugin
– https://github.com/marcuslonnberg/sbt-docker
– Still Using sbt-native-packager to package
– Ends up with Docker file

33. Price Engine Service

34. #Containers - DockerFile

35. #Containers – Using SBT

36. #Containers – Using SBT
• Using Sequences to optimize
– Reduce build time
– Smaller Containers

37. #Containers – Orchestration
• CoreOS
– Linux for massive server deployments
– Cluster management with Fleet
– Service discovery with etcd
• Evaluating
– Kubernetes
– Flynn
– Consul
– Ambassadord

38. #AKKA
Summary
#NonBloc
king
#Containers

42. Lior Shapsa
lshapsa@gmail.com
Twitter: @liorshapsa
Yardena Meymann
ymeymann@gmail.com
Twitter: @ymeymann