Serverless - Lessons Learned - Nike Brown Bag

Brown Bag @ Nike on 03/2019.
- Serverless Deployment Framework (Serverless, Terraform, AWS SAM)
- Deployment Best Practises
- Serverless Performance (Coldstart, VPC, Runtimes, GraalVM)

Transcript

1. Serverless on AWS LESSONS LEARNED Driss Amri @drams88
2. Who is Driss Amri? Blogger at drissamri.be Traveler, Diver, Developer
3. Agenda • Deployment - Frameworks - Best practises • Performance - Cold start - VPC - Runtimes
4. “Everything we hear is an opinion, not a fact. Everything we see is a perspective, not the truth.” - Marcus Aurelius, Mediations
5. Deployment toolkits • Terraform • Serverless Framework • AWS Serverless Application Model (SAM)
6. Serverless Framework • Most widely used Serverless toolkit • Released in 2015 • Simple Developer Experience • Focus on Functions, not Cloud Infrastructure • Cloud-agnostic - AWS provider: - Compiles to CloudFormation - deployed as CloudFormation Stack
7. Serverless Framework
8. Serverless Framework CLI serverless create --template aws-java-gradle --name sls-app serverless deploy --region eu-west-1 serverless remove
11. AWS SAM • Official AWS deployment framework • Released in 2016 • Built on top of CloudFormation - Deployed as CloudFormation Stack
12. AWS SAM
13. AWS SAM CLI sam init --runtime java8 sam package \ --output-template-file packaged.yaml \ --s3-bucket bucketname sam deploy \ --template-file packaged.yaml \ --stack-name sam-app \ --capabilities CAPABILITY_IAM \ --region eu-west-1 aws cloudformation delete-stack --stack-name sam-app
14. AWS Serverless Application Model (SAM)
15. Deployment Frameworks AWS SAM Terraform + Easy to use + Official AWS Support + SAM Local Docker - Missing basic features - AWS only + Great for infrastructure + Cloud agnostic - Verbose - No extra Serverless support - Steap learning curve Serverless Framework + Easy to use + Cloud agnostic + Biggest community + Easy local testing* - Limited local testing for Java
16. Deployment best practises • Two phase deployment & destroy - Stateful resources (DynamoDB, RDS, SQS, …) - Stateless resources (Lambda, Gateway, Cloudwatch, ...) • Safe deployments - Canary - Linear - AllAtOnce
23. Performance
24. Latency UX • Synchronous requests - User waiting: < 1 sec - AWS & Google: 100-500 ms delays has huge impact • Asynchronous requests - Depends on requirements - Batch reports < 30 sec - Any request > 5 sec should provide feedback
26. When do coldstarts happen? • First invocation instance • Concurrent invocations • After provider resource clean up • After deployment & configuration change
27. The Burning Monk - I’m afraid you’re thinking about AWS Lambda cold starts all wrong
28. The Burning Monk - I’m afraid you’re thinking about AWS Lambda cold starts all wrong
29. The Burning Monk - I’m afraid you’re thinking about AWS Lambda cold starts all wrong
30. The Burning Monk - I’m afraid you’re thinking about AWS Lambda cold starts all wrong
35. - https://youtu.be/QdzV04T_kec - https://www.slideshare.net/AmazonWebServices/a-serverless-journey-aws-lambda-un der-the-hood-srv409r1-aws-reinvent-2018 AWS re:Invent 2018
36. When to use a VPC? • Access to Amazon RDS, Elasticache, RedShift, … • Private API Gateway • On-premise services over VPN/Direct Connect • ...
39. Runtime Coldstart Performance
41. The Burning Monk - AWS Lambda Compare Coldstart
42. Minimize bootstrap • Keep function single purposed • Minimize dependencies & use lightweight frameworks - serverless-optimizer-plugin, webpack,… • Prefer low overhead runtime (node.js, Python, Go) • Choose the best memory size - 2x Memory = > 2x CPU = Possibly Faster & Cheaper
43. AWS Lambda Power Tuning AWS Lambda Power Tuning
44. Real world example
45. Runtime: Java Global app init: Static + Constructor Java runtime initialization: ~200 ms Function execution • Java 8 + Gradle • DynamoDB AWS SDK 2 + AWS HTTP Client + Dagger 2 Latency: ~3.9s drissamri/aws-lambda-runtime-coldstart - java
46. Runtime: node.js • Node.js v8 + TypeScript + Webpack • DynamoDB Client drissamri/aws-lambda-runtime-coldstart - nodejs Node.js runtime initialization: ~200 ms Global app init Function execution Latency: ~560ms
48. AWS Lambda Custom Runtime • Implement AWS Lambda in ANY language • Runtime triggers the Lambda function handler • Can be included in the Function deployment - Executable file: bootstrap
50. GraalVM
51. First.. What is Graal?
52. a dynamic just-in-time compiler that improves efficiency and speed
53. A (JIT) compiler for Java transforms bytecode to machine code
55. -XX:+UnlockExperimentalVMOptions -XX:+EnableJVMCI -XX:+UseJVMCICompiler
56. What about GraalVM?
59. GraalVM • Create native images: self contained executable • Lightning fast startup • Low memory consumption • Polyglot programming
60. GraalVM
61. GraalVM Limitations
62. Cloud-Native Frameworks
63. Quarkus
64. GraalVM Lambda • Possible with AWS Lambda Custom Runtime • Framework support: - Micronaut: workable - Quarkus: experimental - Spring Fu: experimental
65. GraalVM Lambda
66. Conclusion • Deployment ✅ Serverless Framework or AWS SAM ✅ Safe deployments with CodeDeploy ✅ Separate stateful and stateless parts of build • Performance ✅ Critical latency: nodejs ✅ Keep an eye on GraalVM
67. Learn...
68. Get started! Amazon Web Services — a practical guide https://github.com/open-guides/og-aws AWS Certification preparation https://bit.ly/2aroFYb Serverless framework https://serverless.com AWS Well-Architected https://aws.amazon.com/architecture/well-architected The Burning Monk - Yan Cui https://theburningmonk.com/
69. Links AWS re:Invent 2017: Become a Serverless Black Belt https://www.youtube.com/watch?v=oQFORsso2go A Serverless Journey: Under the Hood of AWS Lambda https://youtu.be/QdzV04T_kec