Building a modern SaaS in 2020

Building a modern Software as a Service platform brings a lot of interesting engineering challenges. During this talk, I’m going to share my team’s journey of building a SaaS from scratch in 2020. First, we are going to start with the technologies and the architecture we picked. Then, we’ll go over the interesting challenge of implementing multitenancy. And we'll see how we benchmarked three different options and picked one. And last but not least, we’ll explore how every startup can use open source technologies to build observability infrastructure. And how to run their SaaS in production.

Transcript

1. Building a modern SaaS in 2020 Nikolay Stoitsev, Engineering Manager @ Halo DX
5. Language Architecture Repo Structure Builds Running in Production Observability Multitenancy
6. Language Architecture Repo Structure Builds Running in Production Observability Multitenancy
7. Language evaluation Python JavaScript/Node.js Java C#/.NET Core
8. Domain Specifics FIPS 140-2
9. Node.js Can build with special OpenSSL
10. Java Package with verified implementation of cryptographic algorithms
13. DICOM
14. Imaging Center
15. Imaging Center Hospital
16. HL7 Imaging Center Hospital
17. HL7 DICOM
18. Java and Spring Boot
19. Angular
20. Language Architecture Repo Structure Builds Running in Production Observability Multitenancy
21. Microservices Discovery
22. Microservices Discovery Load Balancing
23. Microservices Discovery Load Balancing Circuit Breaking
24. Microservices Discovery Load Balancing Circuit Breaking Testing
25. Microservices Discovery Load Balancing Circuit Breaking Testing Debugging
26. Monolith
27. Failure Isolation
28. 6 services
29. Hexagonal Architecture Web Adapter Messaging Adapter Input Port Input Port Biz Logic Output Port Output Port Persistence Adapter Messaging
30. Language Architecture Repo Structure Builds Running in Production Observability Multitenancy
31. Multi Repo vs Mono Repo
32. Mono Repo Easily introduce and refactor global features Ease code sharing and reuse Simple dependency management Flexible team boundaries and code ownership Collaboration across teams Atomic commits across services
33. Build System Bazel • Fast and incremental builds • One tool to build multiple languages • Extensible • Google, Pinterest, Uber, Stripe Buck • Fast and incremental builds • One tool to build multiple languages • Helps you better understand your dependencies • Facebook, Airbnb
34. Multi-Project Builds in Gradle
35. Multi-Project Builds in Gradle • One root project • One subproject for each service • One subproject for shared libs • https://docs.gradle.org/current/userguide/multi_project_builds.html
36. Multi-Project Builds in Gradle
37. Nx for front-end monorepo https://github.com/nrwl/nx
38. Gradle wrapper for front-end builds and running tests
39. Language Architecture Repo Structure Builds Running in Production Observability Multitenancy
40. Separate Dockerfile for each service
41. Can have one Docker Compose for everything
42. Nexus https://www.sonatype.com/nexus/repository-oss
43. Why Nexus? Can store and distribute Maven/Java, npm, Helm, Docker Widely adopted Single source of truth for every component Easy to work with
44. Build Process CI Code
45. Build Process CI CD Code
46. Build Process CI CD Build Docker Image Code
47. Build Process CI CD Nexus Build Docker Image Code Docker Image
48. Language Architecture Repo Structure Builds Running in Production Observability Multitenancy
49. Kubernetes https://kubernetes.io
50. Why Kubernetes? Deployment Scaling Self-healing Inter-service communication Storage orchestration Scalable and battle tested
51. Great for engineers • Each service is accessed via a single domain • Logs are easy to read • A lot of materials on how to learn it
52. https://github.com/rancher/rancher
53. Monitoring and Grafana out of the box
54. Configuring a service in Kubernetes • Deployment • Service • Config Map • Service Account • RBAC • Pod disruption budget • ....
55. Helm https://github.com/helm/helm
56. Helm • Abstraction over k8s that simplifies working with it • The building block in helm is called a helm chart • Help you define, install, and upgrade applications in k8s
57. How to work with Helm • Search for “{technology} helm chart” • Download helm chart • Update values.yaml • $ helm install {release-name} {chart-name} --set {var}={val}
58. Deploy Process CI CD Nexus Build Docker Image Docker Image
59. Deploy Process CI CD Nexus Kubernetes Kubernetes Kubernetes
60. Deploy Process CI CD Nexus Helm Update Kubernetes Kubernetes Kubernetes
61. Deploy Process CI CD Nexus Helm Update Kubernetes Kubernetes Kubernetes Pull Image
62. Language Architecture Repo Structure Builds Running in Production Observability Multitenancy
63. SLF4J backed by Log4j2 https://github.com/qos-ch/slf4j
64. Backend logging EFK - Elasticsearch, Fluentd, Kibana Pod Pod Pod
65. Backend logging EFK - Elasticsearch, Fluentd, Kibana Pod Pod Pod Fluentd Fluentd Fluentd
66. Backend logging EFK - Elasticsearch, Fluentd, Kibana Pod Pod Pod Fluentd Fluentd Fluentd Elasticsearch
67. Backend logging EFK - Elasticsearch, Fluentd, Kibana Pod Pod Pod Fluentd Fluentd Fluentd Elasticsearch Kibana
68. Front-end logging
69. Front End Logging Library
70. Front End Logging Library
71. Front End Logging front end lib front end lib
72. Front End Logging Backend Service Backend Service front end lib front end lib
73. Front End Logging Backend Service Backend Service Fluentd Fluentd front end lib front end lib
74. Front End Logging Backend Service Backend Service Fluentd Fluentd Elasticsearch Kibana front end lib front end lib
75. Language Architecture Repo Structure Builds Running in Production Observability Multitenancy
76. Subtenancy Org East West Sales Front Desk Sales Front Desk
77. Materialized path (1) Org (2) East (3) West (4) Sales (5) Front Desk (6) Sales (7) Front Desk ID Path Name 1 1/ Org 2 1/2/ East 3 1/3/ West 4 1/2/4 Sales East 5 1/2/5 Front Desk East 6 1/3/6 Sales West 7 1/3/7 Front Desk West
78. Closure table (1) Org (2) East (3) West (4) Sales (5) Front Desk (6) Sales (7) Front Desk ID Name 1 Org 2 East 3 West 4 Sales East 5 Front Desk East 6 Sales West 7 Front Desk West Ancestor Descendant 1 2 1 3 1 4 1 5 1 6 1 7 2 4 2 5 3 6 3 7 4 Null
79. Keeping hierarchy and data separated (1) Org (2) East (3) West (4) Sales (5) Front Desk (6) Sales (7) Front Desk • Each records get a tag - 1, 2, 3, etc. • Each tag points to a node in the tree • Tag hierarchy is loaded separately • Queries are decorated with OR statements to match by tag
80. Benchmark
81. Benchmark
82. Language Java and Spring Boot
83. Language Architecture Java and Spring Boot Microservices
84. Language Architecture Repo Structure Java and Spring Boot Microservices Monorepo
85. Language Architecture Repo Structure Builds Java and Spring Boot Microservices Monorepo Gradle and Docker
86. Language Architecture Repo Structure Builds Running in Production Java and Spring Boot Microservices Monorepo Gradle and Docker Kubernetes
87. Language Architecture Repo Structure Builds Running in Production Observability Java and Spring Boot Microservices Monorepo Gradle and Docker Kubernetes EFK stack
88. Language Architecture Repo Structure Builds Running in Production Observability Multitenancy Java and Spring Boot Microservices Monorepo Gradle and Docker Kubernetes EFK stack Separate hierarchy and data
89. Thank you! Q&A Nikolay Stoitsev Engineering Manager at Halo DX