Présenté à la SGCIB, par Adrian COCKCROFT (@adrianco), VP Cloud Architecture Strategy chez AWS.
Chez Netflix, Adrian était Lead Architecte en charge de 4 transformations majeures :
-
transition d’un modèle d’envoi de DVD par voie postale vers une plateforme de streaming globale
-
migration d’une infra traditionnelle vers Amazon Web Services
-
transformation d’une architecture monolithique vers une architecture microservices complète, dont il a contribué à mettre à disposition en Open Source de nombreux composants au travers de Netflix OSS
-
création de standards de qualité, de disponibilité et de sécurité inconnus jusqu’alors grâce à la mise en place du Chaos Engineering
Egalement passionné par l’intrication entre culture d’entreprise et technologie, et la manière dont ces 2 éléments peuvent permettre à une organisation de se développer très rapidement.
Au cours de la keynote et de la session de questions-réponses, il partagera sa vision de le l’évolution du cloud et son impact global sur la technologie et la culture d’entreprise.
Adrian :
Chez Netflix, importance de la feedback loop pour améliorer en continue le stream de data fourni au client.
Culture de Netflix pour l’extrem Agility : moving much faster than any other competitors → think and develop very fast.
Some financing companies try to go as fast as them right now.
Purpose driven architecture: the purpose is the most important, not the technical part.
-
Freedom & responsibility
2 personnes à l’architecture pour la partie Cloud chez Nefflix!
migration sequence for Netflix from monolithic to microservices :
-
start with the simplest possible API service
-
next to the simplest Web page
-
then web pages one by one
big challenge: to learn developpers that they do not have one relational DB, but data spread between several different independant DBs.
1 single function by service BUT guarantying reasonable response time.
Evolution des archi microservices → serverless !!!! FaaS
"microservices to Ephemeral Functions": when the system is idle, it shuts down and cost nothing to run.
Adrian: you have to move to serverless NOW ! It is disrupting our behaviors, the market, the speed with which we can now build a PROD efficient application.
Evolution of Business Logic : Monolithic → Microservices → Functions (Serverless)
Opposition: Pay up front and depreciate over 3 years AGAINST Pay a month later for the number of seconds used.
-
Pay for what you used last month.
Not for what you guess you will need next year. -
serverless very good if workload goes sometimes to 0 (cost nothing)
-
turn off the system when it’s idle (huge savings)
-
garbage collection of "old" applications having no more traffic
Never kill a whole datacenter.
You need to automate this kind of tests.
Doing so make everybody knows that it can happen.
The more frequently you test this failover, the less painful it becomes.
question: who runs the fire drill for IT ? Because never ever EVERYBODY will be trained.
→ a team must be created: Chaos Engineering team
Lire Chaos Engineering (free) chez O’Reilly
Gremlin is a company specialized in this business (former employees from Netflix and Amazon)
-
Adrian: "a lot of clients coming back from Azure to Amazon because Azure doesn’t scale well yet…"
-
Amazon is very resilient, but the faster way to break something is to break the DNS…
-
Amazon ECS scales better than Kubernetes (???), because bigger (???). Amazon architecture is multi-tenant, whereas Kubernetes is not. You can use Kubernetes if you want to own your own server, BUT for online services, Amazon is better fitted.