Architecting for the Cloud using NetflixOSS - Codemash Workshop
•
83 likes•40,398 views
This document provides an overview and agenda for a presentation on architecting for the cloud using the Netflix approach. Some key points: - Netflix has over 40 million members streaming over 1 billion hours per month of content across over 40 countries. - Netflix runs on AWS and handles billions of requests per day across thousands of instances globally. - The presentation will discuss how to build your own platform as a service (PaaS) based on Netflix's open source libraries, including platform services, libraries, and tools. - The Netflix approach focuses on microservices, automation, and resilience to support rapid iteration on cloud infrastructure.
Report
Share
Architecting for the Cloud using NetflixOSS - Codemash Workshop
- 1. Architecting for the Cloud - The Netflix Way! Codemash PreCompiler Workshop - Jan 8th 2014 Sudhir Tonse Manager, Cloud Platform Infrastructure @stonse
- 2. NETFLIX Netflix is the world’s leading Internet television network with more than 40 million members in >40 countries enjoying more than one billion hours of TV shows and movies per month, including original series.
- 3. NETFLIX SCALE • 100s of Mid-tier services and applications • Billions of Requests per day • ~70 Billion Events per day • 10,000s of Ec2 Instances in use in multiple AWS Regions/ Zones • Cassandra NoSQL database in a Global Ring spanning regions: Terabytes of data • At peak consumes ~1/3 of US Internet Bandwidth
- 4. I am PaaS (and so can You!) Your Cloud PaaS @stonse
- 6. Agenda • PaaS? What is it good for? • NetflixOSS -> Build your own PaaS the Netflix Way! • What you need to build, deploy and operate a Cloud Native set of applications • Cloud Native Architecture • Platform Services -> Application Services on top of IaaS • Platform Libraries -> Application Framework Libraries • Platform Tools/Operational Utilities -> Helps deploy and manage your Cloud Apps • @stonse Build your own Web Service using NetflixOSS libraries
- 7. PaaS? What is it? Libraries, Tools and Frameworks that enable one to Build, Deploy and Manage Software Services on Ephemeral, On Demand Virtual Machines (IaaS) @stonse
- 8. @stonse
- 9. Netflix Cloud Stack (IaaS, PaaS, SaaS, Cloud Management) @stonse Virtualization Technologies
- 10. Netflix Cloud Stack Clients Browsers, Mobile, Televisions … Applications/Web Services PaaS Execution Env (JVM), Web/App Servers, Frameworks, Tools IaaS Virtual Machines, Networking, Load Balancers … @stonse Netflix Apps/Services
- 11. @stonse
- 12. You can too! @stonse Virtualization Technologies
- 13. Build Your Own PaaS @stonse
- 14. Build Your Own PaaS • Yes you can! • Others have done it • Yes, on non-AWS environments too!! @stonse
- 17. The Netflix Way!? Bus + Freedom Dev + Ops Responsibility Culture @stonse BusDevOps
- 18. Think … Cloud Native Architecture @stonse More at http://www.siia.net/presentations/software/ AATC2013/OpenSourceAtNetflix.pdf - Adrian Cockroft (@adrianco)
- 19. Fine-Grained Reliable, Available, Distributed Service Oriented Architecture - that supports Rapid Iteration/innovation @stonse
- 20. Architectural Insights @stonse Slide courtesy @adrianco
- 22. TYPICAL DEPLOYMENT • Multiple Availability Zone • Multiple Regions @adrianco
- 25. • RPS • Load Average Slow Growth Predictable Bursts Compute • … Time Time Compute • Use CloudWatch Time Periodic Jobs Compute Scale UP & Down based on Demand Compute • Compute ON DEMAND AUTO SCALING! Time Unpredictable Bursts Time Steady State
- 28. LIFECYCLE • Think, Implement, Deploy, Test, Operate
- 29. AWS SERVICES (IAAS) As of Jan 2014 @stonse
- 30. Platform Services + Application Framework Libraries + Operational Tools
- 31. Your Cloud PaaS!! OR @stonse Other IaaS
- 32. Platform Runtime Services Metadata Registry Eureka Monitoring Services Turbine Distributed Coordinator/Registry (Zookeeper) Distributed Cache Exhibitor EVCache Metrics Services Servo + AWS Resources Lifecycle Service AWS CloudWatch Edda SQS @stonse
- 33. Platform Libraries - Application Container/Framework Function NetflixOSS Library REST Framework/Bootstrapping/DI Karyon/Governator Functional Reactive Programming RxJava Resiliency/Fallback Hystrix RPC (Routing/LB) Ribbon/Eureka Distributed Co-ordination (Zookeeper) Curator Distributed Caching EVCache NoSQL (Cassandra) Persistence Data Access/Caching Astyanax Monitoring Turbine Metrics Servo Logging Blitz4J Properties/Configuration @stonse Service Requests Archaius Config/Insights
- 34. Platform Big Data/Caching & Services Cassandra NoSQL @stonse Astyanax Priam CassJMeter Hadoop Platform As a Service Genie Lipstick
- 35. Platform Tools & Services Deployment/Workflow @stonse Asgard Frigga Glisten Aminator Availability, Resiliency, Costing, Monitoring (Ops Tools) SimianArmy ICE
- 36. Soon …. @stonse
- 37. How do I get Started? Workshop Session …
- 38. How do I create an App/Service? Instructions at http://bit.ly/1dlDyVC @stonse
- 39. Sample Application – RSS Reader http://techblog.netflix.com/2013/03/introducing-first-netflixoss-recipe-rss.html @stonse
- 40. STEP 1 Choice A: Bare bones Web Service Template $git clone https://github.com/Netflix/karyon.git $cd karyon $./gradlew clean build $./gradlew :karyon-examples:hello-netflixoss:jettyRun Choice B: Full blown Recipe (RSS Reader Recipe) $git clone https://github.com/Netflix/recipes-rss.git $cd recipes-rss $./gradlew clean build @stonse
- 41. STEP 2 • Customize your App using .war of previous step as a template • TIPS (optional) • Use Ribbon for RPC calls • Use Hystrix/RxJava for wrapping concurrent calls/RPC calls (Latency/Fault Tolerance) @stonse
- 42. STEP 3 • @stonse Bake an AMI (Amazon Machine Image) using NetflixOSS Aminator
- 43. STEP 4 • Deploy to EC2 using Asgard Deploy & Manage AWS Applications/Infrastructure • TIP: Employ Simian Army to test Resiliency @stonse
- 44. Building Apps and AMIs @stonse
- 45. Building Apps and AMIs App Karyon Build WAR Deploy Aminator Asgard ASG /Cluster Launch Instances App AMI ASG/Cluster @stonse
- 46. Your Cloud! Metadata Registry Eureka App AMI Billing ASG @stonse Ribbon (RPC) App AMI Credit P ASG App AMI Order P ASG
- 47. Libraries and Services in (some) Detail ...
- 48. Eureka: Metadata Registry as a Service @stonse
- 49. Eureka: Metadata Registry as a Service
- 50. Ribbon: RPC Library (with Software Load Balancers) @stonse
- 51. RIBBON (RPC) CALL try { RestClient client = (RestClient) ClientFactory.getNamedClient("FlixmashClient"); HttpRequest request = HttpRequest.newBuilder().uri(new URI("/ flixmash/rest/v1/movies/popularMovies")).build(); response = client.executeWithLoadBalancer(request); String movies = response.getEntity(String.class); return movies; } catch (Exception exc) { throw new RuntimeException("Exception", exc); } finally { if (response!=null){ //hand the connection back to the pool response.close(); } }
- 52. Archaius: Configuration (properties) as a Service • Property Driven Development for (int i=0; I < numRowsProperty.get(); i++){ fetchMovieForRow(i, numMoviePerRowProperty.get()); } @stonse
- 53. Archaius: Configuration (properties) as a Service @stonse
- 54. Archaius: Configuration (properties) as a Service @stonse
- 55. Curator + Exhibitor: Distributed Sync & Coordination Services (Zookeeper) http://curator.incubator.apache.org/ @stonse
- 56. Hystrix: Defend your App @stonse
- 57. Hystrix: Defend your App @stonse
- 64. Servo (+ AWS CloudWatch): Metrics as a Service JMX CloudWatch @stonse
- 65. Turbine: Hystrix Dashboard (Real Time Metrics Aggregator) • • @stonse Low Latency, High Throughput Stream Processing Engine Real Time Metrics Dashboard
- 66. Edda: Lifecycle (AWS resource) Blogger as a Service @stonse
- 67. ICE : AWS Resource Usage (Cost) @stonse
- 69. Simian Army: Availability/Resiliency as a Service • Chaos Monkey • Kills instances • Janitor Monkey • Clean up unused Resources • Conformity Monkey • Confirms adherence to Best Practices • Others (yet to be open sourced) • Latency Monkey • Chaos Kong • … @stonse
- 71. Advanced Topics … Predictive Auto Scaling Engine http://techblog.netflix.com @stonse
- 72. Genie: Hadoop Platform as a Service @stonse Jeff Magnusson: Tomorrow at QConSF
- 73. Other Open Source Components (you can use) + Many Many More …. @stonse
- 74. Further Reading … • http://techblog.netflix.com • http://slideshare.net/netflix • http://netflix.github.io • http://fluxcapacitor.com • https://github.com/cfregly/fluxcapacitor/wiki/NetflixOSS-FAQ @stonse
- 75. Workshop … Hands On: Building Hello Netflix OSS application • Concepts • • Configuration • Build Script (gradle) • Application Container/Bootstrap • JSR-311 Web Resource • Properties via Archaius • Metrics via Servo • Dependency Injection via Governator/Guice • RPC calls via Ribbon/Eureka • Resiliency via Hystrix (wrappers on top of Ribbon) • Invoking a Mid-tier Web Service • @stonse Project Layout AdminConsole (port 8077)
- 76. Workshop … Hands On: Building Hello Netflix OSS application • @stonse Instructions at http://bit.ly/1dlDyVC
- 78. Project Layout
- 79. Application Container /Bootstrap Located at https://netflix.box.com/s/suk5qlis3sn9f2sz44dl public class HelloWorldBootstrap extends ServerBootstrap { @Override protected void beforeInjectorCreation(LifecycleInjectorBuilder builderToBeUsed) { builderToBeUsed.withAdditionalModules(new JerseyServletModule() { @Override protected void configureServlets() { Map<String, String> params = Maps.newHashMap(); params.put(PackagesResourceConfig.PROPERTY_PACKA GES, "com.netflix.hellonoss, com.netflixoss.workshop"); serve("/rest/v1/ *").with(GuiceContainer.class, params); binder().bind(GuiceContainer.class).asEagerSingleton(); } }); } }
- 80. JSR-311 Web Resource 1. Hello World Resource -> https://netflix.box.com/s/ nyspqi1an095d4p5xd7k 2. Flixmash Resource -> https://netflix.box.com/s/ nb1nd7bqk7qcmxnwuol0 @Path("/flixmash") public class FlixmashResource { @GET @Path("popularMovies") @Produces({MediaType.APPLICATION_JSON}) public Response movies() { // Start timer Stopwatch stopwatch = getMoviesTimer.start(); JSONObject response = new JSONObject(); try { getMoviesCounter.increment(); response.put("Movies", flixmashManager.getPopularMovies()); // obtain movies from our flixMovies component return Response.ok(response.toString()).build(); } catch (JSONException e) { getMoviesErrorCounter.increment(); logger.error("Error creating json response.", e); return Response.status(Response.Status.INTERNAL_SERVER_ERROR).build (); } finally { stopwatch.stop(); } }
- 81. ARCHAIUS: PROPERTIES flixmash.popularMovieList=Lord of the Ring,Matrix,Harry Potter,Gravity private static final DynamicStringListProperty popularMovieList = new DynamicStringListProperty("flixmash.popularMovieList", (List<String>) null); List<String> movies = popularMovieList.get();
- 82. SERVO METRICS // JMX: com.netflix.servo.COUNTER.FlixMash.getMovies private static final Counter getMoviesCounter = new BasicCounter(MonitorConfig.builder("FlixMash.getMoviesCounter").build ()); DefaultMonitorRegistry.getInstance().register(getMoviesCounter); @GET @Path("popularMovies") @Produces({MediaType.APPLICATION_JSON}) public Response movies() { try { getMoviesCounter.increment(); … }catch (Exception e){ } }
- 83. SERVO METRICS IN JMX
- 85. Takeaways … • Cloud Computing and Cloud based Services requires embracing Cloud Native - Adrian Cockroft (@adrianco) • NetflixOSS is here to help! • You can build your own custom PaaS the Netflix Way! http://netflix.github.io @stonse Yes, we are Hiring!