Showdown of the serverless cloud, orchestrating functions @ Serverlessdays.amsterdam

Serverless and event-driven architectures are vastly growing in popularity. Companies are jumping on the bandwagon fast, especially now it is so easy and cheap to create them in the bigger cloud vendors. But what cloud vendor can we best pick for a specific problem? What are the capabilities and limitations of each cloud vendor? What are the differences between serverless functions implementations? We are fans of serverless and event-driven architectures and we wanted to see how easy it was to create a serverless application in both AWS, Azure and Google cloud. In specific, we wanted to compare how we can orchestrate functions. Our use case, replacing our current car park management solution. During this demo, we will show you our solutions for this use-case in the different clouds. It is based on AWS Lambda with Step Functions and DynamoDB, Azure Durable Functions and Azure Table Storage, and Google Cloud functions with Google Cloud Datastore. Finally ending up with the pros and cons of each cloud vendor. So are you thinking about moving to the serverless cloud and not quite sure which vendor fits you the best, come to join this session!

Transcript

1. Showdown of the serverless cloud, orchestrating functions
2. Kenny Baas-Schwegler @kenny_baas Baasie.com xebia.com/blog/author/kbaas/ Marc Duiker Azure Consultant - I ❤ Serverless Strategic software consultant - Domain-Driven Design @marcduiker blog.marcduiker.nl
4. Let’s look at the specs!
5. Type​ AWS Lambda​ Azure Functions​ Google Cloud Functions​ Price 1M Executions​ ($) 0.20​ 0.20​ 0.40​ Price GB-s​ ($) 0.00001667​ 0.000016​ 0.0000025​ Price GHz-s ($) - - 0.0000100 Free executions/month​ 1M​ 1M​ 2M​ Free GB-s/month​ 400K​ 400K​ 400K​ ​Free GHz-s/month ​- ​- ​200K Total​ cost/month ($) ​78.48 75.40 92.70 Comparing Cost* * For 10M function executions of 1 sec each using a 512MB / 800MHz machine. @kenny_baas @marcduiker
6. Runtimes AWS Lambda​ Azure Functions​ Google Cloud Functions​ Node.js v6 & v8 v6, v8, v10 v6 & v8 (beta) .NET Framework ​- v4.7​ ​- .NET Core v1, v2.x v2.x -​ Python v.2.7, v3.6 & v3.7 v3.6 (preview)​ v3.7.1​ (beta) Java ​v8 v8 (preview) -​ Ruby v2.5 - -​ Go v1.x -​ v1.11(beta) Comparing Runtimes @kenny_baas @marcduiker
7. Comparing Cold Start Source: https://mikhail.io/2018/08/serverless-cold-start-war/ @kenny_baas @marcduiker
9. Parking Garage Use Case
10. You get a FAAS @kenny_baas @marcduiker
11. Everybody gets a FAAS @kenny_baas @marcduiker
12. Flying Spaghetti Functions https://clearlens.org/do-you-believe-in-a-flying-spaghetti-monster/
13. @kenny_baas @marcduiker
16. @kenny_baas @marcduiker
17. Question: How do we split systems into functions? Heuristic: Within a bounded context modules are a perfect candidate for splitting up into functions. Heuristic: Use one source control repository per bounded context. @kenny_baas @marcduiker
18. @kenny_baas @marcduiker
19. SDK Portal Monitoring Orchestration Documentation Testing
20. “Everything we hear is an opinion, not a fact. Everything we see is a perspective, not the truth.” ― Marcus Aurelius , Meditations @kenny_baas @marcduiker
21. Documentation - AWS @kenny_baas @marcduiker
22. Documentation - AWS @kenny_baas @marcduiker
23. 23 Documentation - Azure @kenny_baas @marcduiker
24. 24 Documentation - GCloud @kenny_baas @marcduiker
25. 25 Documentation + - AWS Lambda • Extensive, detailed documentation. • Everything you need! • Extensive wall of text crits you Azure Functions • Quickstarts for novices • Detailed reference material for the more experienced. • Not 100% complete across all languages. GCP Functions • Quickstarts for novices. • Hard to find specific information @kenny_baas @marcduiker
26. SDK - AWS https://www.npmjs.com/package/@types/aws-lambda @kenny_baas @marcduiker
27. SDK - AWS @kenny_baas @marcduiker
28. SDK - AWS @kenny_baas @marcduiker
29. SDK - AWS @kenny_baas @marcduiker
30. 30 SDK - Azure @kenny_baas @marcduiker
31. 31 SDK - Azure @kenny_baas @marcduiker
32. SDK - GCP @kenny_baas @marcduiker
33. SDK - GCP https://www.npmjs.com/package/@types/google-cloud__datastore @kenny_baas @marcduiker
34. SDK - GCP @kenny_baas @marcduiker
35. 35 SDK + - AWS Lambda • Flexible SDK, extensive. • Can also create your own runtime. • Quirks in the gateway body as string. Need to do validation yourself, also typed. Azure Functions • Easy to use SDK for C#. • Be careful of certain NuGet package versions the framework is using (e.g. NewtonSoft.Json). GCP Functions • Using standard 3th party libraries. @kenny_baas @marcduiker
36. Orchestration - AWS https://github.com/binxio/aws-cfn-update @kenny_baas @marcduiker
37. Orchestration - AWS @kenny_baas @marcduiker
38. Orchestration - AWS @kenny_baas @marcduiker
39. Orchestration - Azure
40. 40 SDK - Azure @kenny_baas @marcduiker
41. Orchestration - GCP Not supported OOTB :( @kenny_baas @marcduiker
42. Orchestration + - AWS Lambda • Visualisation in the portal • Uses JSON/YAML configuration. Azure Functions • Flexibility because the orchestration is in code. • The orchestration needs to be deterministic, so don’t use code which is not (e.g. new GUIDs, DateTime, new threads). GCP Functions • N/A • N/A @kenny_baas @marcduiker
43. Deployment Cycle - AWS @kenny_baas @marcduiker
44. Deployment Cycle - AWS @kenny_baas @marcduiker
45. Deployment Cycle - AWS https://xebia.com/blog/building-an-elixir-runtime-for-aws-lambda/ @kenny_baas @marcduiker
46. Deployment Cycle - Azure
47. Deployment Cycle - Azure
48. Deployment Cycle - GCP @kenny_baas @marcduiker
49. 49 Deployment Cycle + - AWS Lambda • Stateful Cloudformation. • Sometimes manual interference when config is false. Azure Functions • Multiple ways to deploy your functions to the cloud. • Prefer CLI over ARM. • Watch out for breaking changes in orchestrations. GCP Functions • Terraform to the rescue! • Google Cloud deployment configuration counter intuitive. @kenny_baas @marcduiker
50. Portal - AWS @kenny_baas @marcduiker
51. Portal - AWS
52. Portal - Azure
53. Portal - Azure
54. Portal - GCloud @kenny_baas @marcduiker
55. 55 Portal + - AWS Lambda • Most parts are simple and easy to discover. • No real boundaries of projects. • Some parts are hard to navigate. Azure Functions • Resource Groups are useful containers for multiple services. • It is very slow to use. GCP Functions • Projects is good concept for grouping services. • Portal is quick to use. @kenny_baas @marcduiker
56. 56 Testing - AWS @kenny_baas @marcduiker
57. 57 Testing - AWS AWS SAM LOCAL @kenny_baas @marcduiker
58. 58 Testing - Azure
59. 59 Testing - Azure @kenny_baas @marcduiker
60. 60 Testing - Azure @kenny_baas @marcduiker
61. 61 Testing - Azure @kenny_baas @marcduiker
62. 62 Testing - Azure @kenny_baas @marcduiker
63. 63 Testing - GCP @kenny_baas @marcduiker
64. 64 Testing + - AWS Lambda • Unit testing is easy • SAM Local seems promising • SAM Local still in beta • Mocking dynamo takes some time to understand Azure Functions • Unit testing is easy • Local runtime is available for running without cloud connection. • Local runtime depends on a separate storage emulator (not all versions are compatible). GCP Functions • Node Emulator available for local testing. • Emulator is in alpha version • Supports Node v6 only @kenny_baas @marcduiker
65. Monitoring - AWS @kenny_baas @marcduiker
66. Monitoring - Azure
67. Monitoring - GCP
68. 68 Monitoring + - AWS Lambda • CloudWatch is mature, gives the right amount of stats Azure Functions • Basic logging OOTB and easy integration with Application Insights. GCP Functions • Basic monitoring OOTB @kenny_baas @marcduiker
69. Key takeaways Question: When do we use AWS? Heuristic: When we want to use many different function runtimes. Heuristic: When we want to use the latest runtimes. @kenny_baas @marcduiker
70. Key takeaways Question: When do we use Azure? Heuristic: When we’re familiar with the Microsoft ecosystem. Heuristic: When we want to orchestrate functions. @kenny_baas @marcduiker
71. Key takeaways Question: When do we use Google Cloud? Heuristic: When we want the best functions portal experience. @kenny_baas @marcduiker
72. Key takeaways Question: When should we use Orchestrations? Heuristic: Use orchestrations within a bounded context. Heuristic: Use orchestrations when having processes over multiple modules in your bounded context. Heuristic: Use orchestrations for long running processes. @kenny_baas @marcduiker
73. 73 Pricing https://aws.amazon.com/lambda/pricing/​ https://azure.microsoft.com/en-us/pricing/details/functions/​ https://cloud.google.com/functions/pricing​ Comparison: https://docs.google.com/spreadsheets/d/1Q6vllVYe1CfHAK6MP6Uz6mTYPiU4DljwL8XN4E5FQdw/edit?usp=sharing Runtimes & languages​ https://docs.aws.amazon.com/lambda/latest/dg/lambda-runtimes.html https://docs.microsoft.com/en-us/azure/azure-functions/supported-languages​ https://cloud.google.com/functions/docs/writing/​ Cold Start https://mikhail.io/2018/08/serverless-cold-start-war/ ​ Sources Github Repository with our demo code https://github.com/Baasie/serverless-showdown @kenny_baas @marcduiker