J-Fall 2019: What's a service mesh and why do i need one?

You’ve been creating this cloud-native microservice based architecture. Continuous delivery pipelines, cloud-based deployments, and Kubernetes managed Docker containers. You are ready to scale beyond your wildest dreams. Now, while taking a step back, you notice that your services contain more than just the business logic you intended to write. Proper communication is key in a distributed system, but do you really need these extra libraries that increase the size of your microservice? Should the responsibility for reliable communication live within your application or can this be abstracted to a higher level? In this session, we will look at the concept of a Services Mesh and how it helps you put the responsibilities at the right layer. After this presentation you might have an answer to the question if you really need a service mesh.

Transcript

1. What’s a Service Mesh and why do I need one? Jeroen Reijn #jfall
2. About me: • (Java) Programmer and architect • Big fan of the DevOps culture • Enjoys building cloud native solutions • Community member and emeritus committer at Apache Jeroen Reijn @jreijn /jeroenreijn
3. Monolith? Microservices? Kubernetes? Cloud?
4. Service mesh, ... istio, … service mesh
5. Have you heard about a service mesh before? +
8. So what is a ‘Service Mesh’ and what problem does it solve?
9. “A service mesh is a dedicated infrastructure layer for handling service-to-service communication”
10. Why a dedicated layer?
11. Microservices Distributed systems Network communication
13. complex Reliable communication is
14. Evolution networking
15. The evolution of networking Computer B Computer A Service A Service B Networking Stack Networking Stack Business Logic Business Logic
16. The evolution of networking Computer B Computer A Service A Service B Networking Stack Networking Stack Business Logic Flow control Business Logic Flow control
17. The evolution of networking Computer B Computer A Networking Stack Service A Service B Networking Stack Business Logic Flow control Business Logic Flow control
19. The 8 Fallacies of Distributed Computing 1. The network is reliable 2. Latency is zero 3. Bandwidth is infinite 4. The network is secure 5. Topology doesn’t change 6. There is one administrator 7. Transport cost is zero 8. The network is homogeneous Composed by Peter Deutsch and his fellow engineers at Sun Microsystems
20. Critical functions for microservices Fast, reliable & safe microservices microservice microservice microservice Routing Dynamic discovery Load balancing Resiliency Circuit breaking Retries Rate limiting Observability Metrics Logging Tracing Security Policy Enforcement
21. Routing - Service discovery Registry client Registry client Registry client Registry client Registry client Registry client Registry client Registry client Registry client Registry Registry client Registry client Service A Service B Service C Service D Service D Service A Service A Service C Service C Service B Service B Service D Registry-aware HTTP client Service Registry
22. Resilience
23. Resilience - Cascading failure Service 1 Service 2 Service 3 Service 4
24. The Circuit Breaker pattern “A service client should invoke a remote service via a ‘proxy’ that functions in a similar fashion to an electrical circuit breaker” https://microservices.io/patterns/reliability/circuit-breaker.html
25. Circuit breaker Half Open Failure threshold exceeded Set breaker Failure threshold exceeded Set breaker Try reset after timeout Success Reset breaker Open Closed Success Fail (under threshold)
26. Observability of your services Golden triangle of monitoring Metrics Logs Traces
27. Security of microservices • OAuth / JWT Tokens • Mutual TLS / certificates
28. Computer B The evolution of networking Computer A Service A Service B Networking Stack Networking Stack Business Logic Flow control Flow control Circuit Breaker Service Discovery Business Logic Circuit Breaker Service Discovery Logs, metrics, traces Security Logs, metrics, traces Security
29. Computer B The evolution of networking Computer A Service A Service B Networking Stack Networking Stack Business Logic Flow control Flow control ??? ??? Circuit Breaker Service Discovery Business Logic Circuit Breaker Service Discovery Logs, metrics, traces Security Logs, metrics, traces Security
30. Computer B The evolution of networking Computer A Service A Service B Networking Stack Networking Stack Business Logic Flow control Flow control Library Library Circuit Breaker Service Discovery Business Logic Circuit Breaker Service Discovery Logs, metrics, traces Security Logs, metrics, traces Security
31. Libraries resilience4j hystrix
32. Drawbacks of libraries • Glue linking the libraries: expensive • Limiting tools, runtimes, languages • Versioning hell • Teams should not forget to add them
33. Computer B The evolution of networking Computer A Service A Service B Networking Stack Networking Stack Business Logic Flow control Flow control Library Library Circuit Breaker Service Discovery Business Logic Circuit Breaker Service Discovery Logs, metrics, traces Security Logs, metrics, traces Security
34. Computer B Service B The evolution of networking Computer A Service A Service B Networking Stack Service A Networking Stack Business Logic Flow control Library ??? Flow control Library ??? Circuit Breaker Service Discovery Business Logic Circuit Breaker Service Discovery Logs, metrics, traces Security Logs, metrics, traces Security
35. Computer B Service B The evolution of networking Computer A Networking Stack Service A Networking Stack Business Logic Flow control ??? Circuit Breaker Service Discovery Logs, metrics, traces Security Proxy Flow control ??? Circuit Breaker Service Discovery Logs, metrics, traces Security Proxy Business Logic
36. OSI Model Level 7 Application: Spring, Vertx, WFSwarm Level 6 Presentation: Json, XML Level 5 Session: Http 1/2, GRPC Level 4 Transport: TCP Level 1-3 Network (IP) / Data link / Physical From here To here
37. Computer B Service B The evolution of networking Computer A Networking Stack Service A Networking Stack Business Logic Flow control Proxy Proxy Circuit Breaker Service Discovery Logs, metrics, traces Security Flow control Proxy Proxy Circuit Breaker Service Discovery Logs, metrics, traces Security Business Logic
38. Responsibility shift Development team(s) Platform team(s)
39. The evolution of networking
40. Computer B Service D First generation service mesh Computer A Service A Proxy Proxy Service B Service C
41. Second generation service mesh - Pods and sidecars • Container platforms • Kubernetes • Mesos Node Pod Pod Container Proxy Container Proxy
42. Computer B Service B The evolution of networking Computer A Networking Stack Service A Networking Stack Business Logic Flow control Proxy Sidecar Proxy Circuit Breaker Service Discovery Logs, metrics, traces Security Flow control Proxy Sidecar Proxy Circuit Breaker Service Discovery Logs, metrics, traces Security Business Logic
43. Complex micro-service architectures 450 + microservices
44. Controlling the service mesh Computer A Service A Networking Stack Business Logic Flow control Sidecar proxy Computer B Service B Networking Stack Business Logic Flow control Sidecar proxy Control plane
45. The service mesh control plane Control plane
46. Proxy based Service meshes
47. Istio • An open platform to connect, monitor, and secure microservices • Introduced by Google, Lyft, IBM and others • Manages authentication, authorization, and encryption of communication between microservices • Logging, monitoring, and keeping services operational • Traffic management and policy control
48. Istio - Architecture B
49. Envoy Proxy • Dynamic service discovery • Load balancing • TLS termination • HTTP/2 and gRPC proxies • Circuit breakers • Health checks • Staged rollouts with %-based traffic split • Fault injection • Rich metrics
50. Istio - Proxy configuration YAML
51. Istio - Discovery and Load-balancing
52. Istio - Tracing • Automatic tracing of request • Asynchronous span reporting • Multiple backends • Zipkin • Jaeger
53. Istio - Telemetry
54. Istio - Advanced routing
55. Istio - Security / Two way TLS
56. Istio Security - RBAC • Role based access control • Based on rules and for instance HTTP methods • ServiceRole (rule) • ServiceRoleBinding (assign role to set of nodes)
57. Istio gives you: • Telemetry • Security • Mutual TLS • Role based access control • Resilience • Circuit-breaker • Retry • Advanced routing
58. Demo
59. Overhead • Definitely not ‘free’, more parts in the system • Proxies are used for both inbound and outbound requests • A lot of effort going on to reduce overhead
60. Debugging • Debugging Envoy and Pilot (configuration) • Networking Issues • TLS issues • Envoy bouncing requests • …
61. Security • Many new parts of the system • Control plane components • Proxies • Envoys are everywhere • Role based access control
62. Istio • Telemetry • Security • Circuit-breaker • Retry • Advanced routing What you (want to) get What you (don’t want to) get • Overhead • Debugging • Security complexity
63. But are all service meshes equal? So we saw Istio…
64. Comparing Service Meshes Source: https://kubedex.com/istio-vs-linkerd-vs-linkerd2-vs-consul/ (Sept 2018)
65. https://smi-spec.io
66. Do I really need a service mesh?
67. Throwing more tech at the problem…
68. Do you want to configure, install and renew (mutual) TLS certificates across an entire set of applications?
69. Do you want to intercept and re-route network flows for: A/B testing, traffic shedding or failure tolerance (circuit breaking)?
70. Do you want tracing / visibility of application request flows within your micro-service network?
71. Should I just remove libraries from my apps?
72. Istio - Circuit breaking - DestinationRule
73. Istio - Circuit breaking - DestinationRule
74. Spring + Hystrix Circuit breaker fallback Note: Hystrix is deprecated and only used as an example
75. Spring + Hystrix Circuit breaker fallback Note: Hystrix is deprecated and only used as an example
76. Tracing
77. As an engineer you should still think about these concerns
78. Key take-aways from this talk • A service mesh is a dedicated infra layer for service communication • Understand the why of using a service mesh • Understand the operational complexity, but also the benefits e.g. transparently adds cross-cutting concerns to a microservices architecture • Think about where you want to solve specific problems
79. “Please rate my talk in the official J-Fall app” #jfall