A Netflix case study: Edge Service Architecture

| Paul Bakker

Paul BakkerEdge Developer Productivity @ Netflix

Contents.

What and why Edge?Edge architectures and their tradeoffsAPI data access patternsDeveloper experience

Architecture patterns

How we do things

This talk is not a template telling you how to do micro services.

We want an architecture that allows us to innovate.

What is Edge Service Architecture?

Lets start with How do we expose Micro Services to devices?

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

http

Multiple calls from the clientPossibly fetching duplicate dataError handling and fallbacks are up to the clientTight coupling between client and backend servicesEvery micro service has to think about client needsDoes this allow rapid innovation?

This is bad, but not horrible

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

http

Now it’s horrible

Multiple calls from the clientPossibly fetching duplicate dataError handling and fallbacks are up to the clientTight coupling between client and backend servicesEvery micro service has to think about client needsDoes this allow rapid innovation?

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

httpGateway service

REST

Clients only need a single callNo coupling between clients and backend servicesGateway takes care of error handling and fallbacksStill a one size fits all APIDoes this allow rapid innovation?

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

httpGateway service

REST

Edge

Micro Service

REST

DRM

Playback

Micro Service

REST

Streaming logs

Edge

Zuul

Edge Developer Experience

APIServer

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

http

Gateway service

REST

http

Gateway service

REST

Gateway service

REST

http

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

Micro Service

REST

http

Gateway service

REST

http

Gateway service

REST

Gateway service

REST

http

Who develops and manages these?

Now every gateway service solves the same problems

Rapid innovation?

Current Edge architecture(s)

Micro Service

Micro Service

Micro Service

Micro Service

Micro Service

http

http

httpAdapterScript

AdapterScript

AdapterScript

Client libraries

API Server

ratings

profiles

videocatalog

AdapterScript

Client libraries

{ “userProfile”: “Paul”, “videoTitle”: “Narcos”, “rating”: “THUMBS_UP”}

The adapter scripts fetch data from multiple services and combine the results

ratings

profiles

videocatalog

Client libraries

The Netflix API server

Adapter scrips written in GroovyUploaded to the API serverFaaS from the developer point of view?

ratings

profiles

AdapterScript

Client libraries

The micro service monolith

Micro Service

Micro Service

Micro Service

Micro Service

Micro Service

http

http

httpAdapterScript

AdapterScript

AdapterScript

Client libraries

API Server

Monoliths are scary.

Single point of failureThe API server is BIGOne bad script, config or client library can affect availability

Micro Service

Micro Service

Micro Service

Micro Service

Micro Service

http

NodeContainer

Adapter scripts

NodeContainer

Adapter scripts

NodeContainer

Adapter scripts

API

http

http

Each set of endpoint scripts is deployed as a lightweight serviceUsing containers to decrease deployment overheadBetter deployment isolationBetter Developer Experience!

Breaking things up

Adapter scripts

ratings

profiles

videocatalog

AdapterScript

Client libraries

{ “userProfile”: “Paul”, “videoTitle”: “Narcos”, “rating”: “THUMBS_UP”}

How to write adapter scripts?

How to deduplicate data?

How to represent downstream errors and timeouts?

Is JSON ideal from the client’s perspective?

Micro Service

Micro Service

Micro Service

Micro Service

Micro Service

NodeContainer

Adapter scripts

NodeContainer

Adapter scripts

NodeContainer

Adapter scripts

API

There is no API

Just declare what data you need.

getVideoTitleById(…)getProfile(…).getName()getRatingForVideoId(…)

[ ["currentProfile", "allVideosRated", 0,["title", "thumbRating"]] ["currentProfile", "name"]]

Treat multiple data sources as one logical model

{ "allVideosRated": { "28": { "$type": "ref", "value": [ "profilesByGuid", "UXXGMAOJ3VGGLGWTQ6NMN7OPGU", "videosById", 80190279 ] } }, "videosById": { "80190279": { "thumbRating": { "$type": "atom", "value": { "rating": "THUMBS_UP", "timestamp": 1514346641221 } }, "title": "The Indian Detective" } } }

get(["allVideosRated", 28, [“title”, “thumbRating"]]);

JSON Graph - refs.

Deduplicate dataAccess data as a logical model

API Discovery

If “there is no API”….

How do I know what to do?

Discovering paths and putting them together

Edge PaaS - Developer Experience

The right balance between serverless and developer experience?

“API Script” developersDisappointed

Can’t run code locallyNo step debuggingNo unit testing

But spoiledDeployment is trivialEasy to support many versionsNothing to manage!

Developer Experience

Projects runs in container on developer machineLocal editing and debuggingExternal routing to developer machineRun locally what runs in production

NodeQuark demo

MetricsAlertsStreaming loggingDeployment pipelinesCanary analysis

Out of the box.

Thank you.

Paul Bakker@pbakker