Navigating the service mesh ecosystem George Miranda @gmiranda23 Ghent Kubernetes Meetup CfgMgmtCamp 2018 

No content

No content

No content

What we’ll focus on — Beyond the hype — What is the problem? — How/why was the service mesh born? — What you should be asking — Product philosophies 

George Miranda Director of Community @BuoyantIO @gmiranda23 Previously: Chef Software, Inc (2012-2017) 

Microservices can be abstracted as highly distributed application architecture @gmiranda23 

data web web web worker worker worker data data data worker @gmiranda23 

@gmiranda23 

The problem @gmiranda23 

@gmiranda23 

@gmiranda23 

@gmiranda23 

@gmiranda23 

The (old) solution @gmiranda23 

Circuit breakers Timeouts Your code looks like this now Biz Logic Timeouts Telemetry RPC code Distributed state Client-side Service Discovery @gmiranda23 

Timeouts Your infrastructure looks like this now Timeouts Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side Circuit Biz Logic Telemetry RPC code Distributed Client-side 

A better solution @gmiranda23 

@gmiranda23 

The service mesh @gmiranda23 

@gmiranda23 Control Plane Data Plane 

Navigating the ecosystem @gmiranda23 

Remember: I work for Buoyant — Opinions are my own! @gmiranda23 

Remember: I work for Buoyant — Opinions are my own! — Be skeptical of my advice @gmiranda23 

Remember: I work for Buoyant — Opinions are my own! — Be skeptical of my advice — Still, I’m watching out for you @gmiranda23 

What you should be asking yourself @gmiranda23 

1) Am I ready for a service mesh? @gmiranda23 

2) What problems am I having today? @gmiranda23 

3) What level of observability do your services have today? @gmiranda23 

4) What functions of a service mesh have you already built? How will they interact if you introduce a service mesh? @gmiranda23 

5) What platform(s) do you need to support? @gmiranda23 

6) What does division of team responsibilities look like today? How will that work when managing the service mesh? @gmiranda23 

7) Centralized or decentralized functionality? @gmiranda23 

8) Support expectations? @gmiranda23 

Product philosophies @gmiranda23 

Remember: I work for Buoyant — Opinions are my own! — Be skeptical of my advice — Still, I’m watching out for you — Covered in chronological order (not an exhaustive list) @gmiranda23 

@gmiranda23 

@gmiranda23 Linkerd — Productized version of Finagle library — Many available components & integrations — Data plane (linkerd) & control plane (namerd) — Multi-platform & feature rich — Production ready, battle tested — Built on Scala & JVM — CNCF hosted project — Commercial services & support 

@gmiranda23 

@gmiranda23 Envoy — Developed for use by Lyft — “Universal data plane” — Can be used on its own (distributed config) — Paired with Istio by default (May 2017) — 3rd party control planes available (not required) — C/C++ — CNCF hosted project 

@gmiranda23 

@gmiranda23 Istio — Joint partnership: Google, IBM, Redhat — Highly configurable control plane — Uses Envoy by default, but can swap data planes — On the road to production — Some additional platform support — Robust architecture & coverage of use cases — Growing 3rd party vendor ecosystem 

(Other data plane solutions exist here) @gmiranda23 

@gmiranda23 

@gmiranda23 Conduit — Launched Dec 2017 — Minimum viable service mesh — Ultralight, zero config, performant, simple — Data plane (Rust), Control plane (Go) — Specific to Kubernetes — Built on lessons from Linkerd in production — Intent to join CNCF 

Where should I start? @gmiranda23 

1) Readiness 2) Problems today 3) Current observability 4) Current functionality 5) Platform support 6) Team responsibilities & roles 7) Centralized or decentralized 8) Commercial support needs @gmiranda23 

@gmiranda23 Try one! Give the projects feedback. Join the community. 

Thank you! Q & A @gmiranda23