Software Architecture Conference 2018, London (UK): Talk by Mario-Leander Reimer (@LeanderReimer, Principal Software Architect at QAware) === Please download slides if blurred! === Abstract: In a microservice world, things become more complex. Platforms such as Kubernetes address a lot of the complexity; they handle resource isolation and utilization, networking, and deployments nicely. But a lot of the involved complexity such as load balancing, rollout scenarios, circuit breaking, retries, rate limiting, observability, tracing and transport security is still left up to the development teams. Of course, you can address all of these challenges in your microservices programmatically using popular open-source components such as Hystrix, Ribbon, Eureka, the EFK Stack, Prometheus or Jaeger. But, unfortunately, this approach can quickly lead to excessive library bloat and suddenly your microservices are not quite so micro anymore. All this might seem acceptable if you’re on a single, consistent development stack like Java EE or Spring Boot. But tackling these complexities becomes even more challenging if you’re dealing with multiple stacks and multiple frameworks, to say nothing about dealing with legacy applications that you can’t modify to retrofit these requirements. In comes Istio to the rescue. It is a so-called service mesh that addresses many of the cross-cutting communication concerns in a microservice architecture. Think of Istio as AOP (aspect-oriented programming) for microservice communication. Instead of implementing everything directly within your services, Istio transparently injects and decorates the desired concerns into the individual communication channels. Mario-Leander Reimer offers an overview of Istio and explains how it addresses the inherent complexities in microservice architectures. He briefly discusses the conceptual architecture and the main building blocks of Istio before diving into several examples deployed on a live Kubernetes cluster to demonstrate the different traffic management features, as well as diagnosability and security.

1. Mario-Leander Reimer, QAware GmbH
mario-leander.reimer@qaware.de
Putting microservices
on a diet with Istio
London, 29th October 2018

2. Mario-Leander Reimer
Principal Software Architect, QAware GmbH
Mail: mario-leander.reimer@qaware.de
Twitter: @LeanderReimer
Github: https://github.com/lreimer/
Slides: https://speakerdeck.com/lreimer/
29.10.2018
2
Developer && Architect
20+ years of experience
#CloudNativeNerd
Open Source Enthusiast
Speaker && Author

3. Fork me on Github.
https://github.com/lreimer/microservice-diet-with-istio

4. 4
https://imgur.com/gallery/LGAZEqu
The Early
Code Monkey
The Monolith

5. loosely coupled stateless
bounded contexts
makeameme.org

6. Essential Cloud-native Design Principles.
6
Design for Distribution: Containers; microservices; API driven development.
Design for Configuration: One image, multiple environments.
Design for Resiliency: Fault-tolerant and self-healing.
Design for Elasticity: Scales dynamically and reacts to stimuli.
Design for Delivery: Short roundtrips and automated provisioning.
Design for Performance: Responsive; concurrent; resource efficient.
Design for Automation: Automated Dev & Ops tasks.
Design for Diagnosability: Cluster-wide logs, metrics and traces.
Design for Security: Secure Endpoints, API-Gateways, E2E-Encryption

7. 7
Atomic Architecture

8. Atomic Microservice Blueprint.
8

9. Concrete Blueprint Incarnation with Spring Cloud Netflix.
9
Some Facts:
58 MB Uberjar
192 Dependencies
3 KB Classes

10. A polyglot microservice architecture suffers from severe
library bloat and bad maintainability in the long run.
10

11. Istio is like AOP, but for
microservice communication.

12. Istio to the Rescue!
12

13. Conceptual View on a Kubernetes Cluster.
13

14. Demo

15. Pods are the smallest unit of compute in
Kubernetes
Labels are key/value pairs used to identify
Kubernetes resources
Replica Sets ensure that the desired
number of pod replicas are running
Deployments are an abstraction used to
declare and update pods, RCs, …
Services are an abstraction for a logical
collection of pods providing DNS name
Ingress routes traffic from outside the
cluster to services and ports based on URL
patterns and host
Kubernetes Glossary.
15

16. GoF in the Cloud: Container Orchestration Patterns.
16http://blog.kubernetes.io/2015/06/the-distributed-system-toolkit-patterns.html
1. Sidecar Container:
Extend container behaviour
Log Extraction / Reformating (fluentd, logstash)
Scheduling (cron, quartz)
2. Ambassador Container:
Proxy communication
TLS Tunnel (Stunnel, ghostunnel, Istio)
Circuit Breaking (linkerd, Istio)
Request Monitoring (linkerd, Istio)
3. Adapter Container:
Provide a standardized interface
Monitoring (Prometheus)
Configuration (ConfigMaps, Secrets, …)

17. Conceptual Istio Architecture and Components.
17

18. 18
Envoy: Sidecar proxy per microservice that handles inbound/outbound traffic within each Pod. Extended
version of Envoy project.
Gateway: Inbound gateway / ingress. Nothing more than a managed Envoy.
Mixer: Policy / precondition checks and telemetry. Highly scalable.
Envoy caches policy checks within the sidecar (level 1) and within envoy instances (level 2), buffers
telemetry data locally and centrally, and can be run in multiple instances.
Mixer includes a flexible plugin model.
https://istio.io/blog/2017/mixer-spof-myth.html
Pilot: Pilot converts high level routing rules that control traffic behavior into Envoy-specific configurations, and
propagates them to the sidecars at runtime.
Watches services and transforms this information in a canonical platform-agnostic model (abstracting away
from k8s, Nomad, Consul etc).
The envoy configuration is then derived from this canonical model. Exposes the Rules API to add traffic
management rules.
Citadel: CA for service-to-service authx and encryption.
Certs are delivered as a secret volume mount. Workload identity is provided in SPIFFE format.
https://istio.io/docs/concepts/security/mutual-tls.html

19. Demo

20. Gateway configures a load balancer for
HTTP/TCP traffic, enables ingress traffic into the
service mesh
Virtual Service defines the rules that control
how requests for a service are routed within
the service mesh
Destination Rule configures the set of policies
to be applied to a request after VirtualService
routing has occurred
Service Version aka Subset allows to select a
subset of pods based on labels
Service Entry enables requests to services
outside of the service mesh
Istio Glossary.
20

21. apiVersion: networking.istio.io/v1alpha3
kind: Gateway
metadata:
name: hello-istio-gateway
spec:
selector:
# use istio default controller
istio: ingressgateway
servers:
- port:
number: 80
name: http
protocol: HTTP
hosts:
- "hello-istio.cloud"
Example Istio Gateway and VirtualService Definitions.
21
apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
name: hello-istio
spec:
hosts:
- "hello-istio.cloud"
gateways:
- hello-istio-gateway
http:
- match:
- uri:
exact: /api/hello
route:
- destination:
host: hello-istio
port:
number: 8080
Exact URI
Routing

22. Hello Istio
Demo

23. Different release patterns can easily be applied.
23

24. Different release patterns can easily be applied.
24

25. apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
name: hello-istio
spec:
hosts:
- "hello-istio.cloud"
gateways:
- hello-istio-gateway
http:
- route:
- destination:
host: hello-istio
subset: v1
weight: 70
- destination:
host: hello-istio
subset: v2
weight: 30
Examples for different routing configurations.
25
apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
name: hello-istio
spec:
hosts:
- "hello-istio.cloud"
gateways:
- hello-istio-gateway
http:
- match:
- headers:
user-agent:
regex: ".*Chrome.*"
route:
- destination:
host: hello-istio
subset: v2
- route:
- destination:
host: hello-istio
subset: v1
Weighted
Traffic
Routing
Header
basedTraffic
Routing

26. Alphabet
Demo

27. apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
name: alphabet-service
spec:
hosts:
- alphabet-service
http:
- fault:
delay:
fixedDelay: 2s
percent: 50
abort:
httpStatus: 500
percent: 50
route:
- destination:
host: alphabet-service
subset: v1
Examples for fault injection and circuit breaker policy.
27
apiVersion: networking.istio.io/v1alpha3
kind: DestinationRule
metadata:
name: alphabet-service
spec:
host: alphabet-service
trafficPolicy:
connectionPool:
http:
http1MaxPendingRequests: 1
maxRequestsPerConnection: 1
tcp:
maxConnections: 1
outlierDetection:
baseEjectionTime: 5.000s
consecutiveErrors: 1
interval: 1.000s
maxEjectionPercent: 100
subsets:
- name: v1
labels:
version: v1
Circuit
Breaker
Policy
Fault
Injection

28. Istio has built-in support for service mesh diagnosability.
28
Diagnosability
Triangle
Metrics
LogsTraces

29. Not all Istio features are marked Stable yet, but Beta can
already be used in Production.
29
Istio v1.0.3 is deemed production ready.
Core: 4 Stable, 3 Beta, 6 Alpha
Traffic Management: 1 Stable, 5 Beta, 1 Alpha
Security and Policy Enforcement: 5 Stable, 2 Beta, 4 Alpha
Telemetry: 4 Stable, 2 Beta, 7 Alpha
Other Service Mesh technologies are emerging.
Linkerd and Conduit
Consul Connect
See https://istio.io/about/feature-stages/

30. Mario-Leander Reimer
mario-leander.reimer@qaware.de
@LeanderReimer xing.com/companies/qawaregmbh
linkedin.com/company/qaware-gmbh slideshare.net/qaware
twitter.com/qaware
youtube.com/qawaregmbh
github.com/qaware