Codemotion_Berlin_-_6_weeks_to_migrate.pdf

Few months ago I met a team: 8 developers working on a B2C app of car renting. They were tired, so tired ... Their vagrant was slow, the server crashed two or three times a week without restarting and without any logs, a roll out to production took 2 days to be tested, by hand ... Together we decided to react, let's have the Roll's Royce of the development environnement! Thus I've dockerized the app, I've created a Kubernetes cluster, I've added Spinnaker to the loop and some other stuff. We had some up and down, some smiles and tears, but we finally did it and we have no regrets.

About:

Aurore Malherbes, CTO - Padok

Working as a developer, I encountered many issues along with my different projects: it took 1 day and a half to install the app on my personal laptop, the deployment is not working and I do not understand why ... That's why I decided to understand the ops world and I co-founded Padok a company helping you to take the advantages of the cloud to develop faster.

Maxime Sraïki, Architect Developer & Coach - BAM

My name is Maxime Sraïki and I fell in love with development in a startup weekend coding an objective C application. I've been working in web development @Theodo for one year and then joined BAM to build awesome mobile applications using React-Native where I've been happily working as an architect for the past three years. A few things about me: - I've developed the new Linkee (redistribution of unsold food to those in need) mobile application & backend voluntarily - I've always been attracted to far east lands & languages - I sing in musicals on spare time ;)

Transcript

2. Maxime Sraïki VP of Engineering maximes@bam.tech @AuroreMalherbes @MaximeSraiki
3. Aurore Malherbes Co founder - CTO aurorem@padok.fr @AuroreMalherbes @MaximeSraiki
4. Let’s set the scene @AuroreMalherbes @MaximeSraiki
6. 2 year long POC mindset 2017 2018 2019 March Start with a series of few weeks POC August 2 months to deliver free floating February 140 bugs Wake up call @AuroreMalherbes @MaximeSraiki
7. 3.25 2 50 @AuroreMalherbes @MaximeSraiki
8. 3.25/5 2d 50s @AuroreMalherbes @MaximeSraiki
9. Development experience @AuroreMalherbes @MaximeSraiki
10. @AuroreMalherbes @MaximeSraiki
11. @AuroreMalherbes @MaximeSraiki 50s
12. @AuroreMalherbes @MaximeSraiki
13. Docker vs Vagrant ? https://searchservervirtualization.techtarget.com/answer/Containers-vs-VMs-Whats-the-difference @AuroreMalherbes @MaximeSraiki
14. Docker - one recipe @AuroreMalherbes @MaximeSraiki 1) Do not use the latest image
15. Docker - one recipe @AuroreMalherbes @MaximeSraiki 1) Do not use the latest image 2) Move your good old monolithe entirely
16. Docker - one recipe @AuroreMalherbes @MaximeSraiki 1) Do not use the latest image 2) Move your good old monolithe entirely
17. Docker - one recipe @AuroreMalherbes @MaximeSraiki 1) Do not use the latest image 2) Move your good old monolithe entirely 3) Changes will be needed
18. Docker - one recipe @AuroreMalherbes @MaximeSraiki 1) Do not use the latest image 2) Move your good old monolithe entirely 3) Changes will be needed - Assets storage - Clients’ IP checking - Templates’ sharing - ...
19. Docker - one recipe @AuroreMalherbes @MaximeSraiki 1) Do not use the latest image 2) Move your good old monolithe entirely 3) Changes will be needed - Asset storage - Clients’ IP checking - Templates’ sharing - ... 4) Ask for developers expectations first
20. Docker - one recipe @AuroreMalherbes @MaximeSraiki 1) Do not use the latest image 2) Move your good old monolithe entirely 3) Changes will be needed - Asset storage - Clients’ IP checking - Templates’ sharing - ... 4) Ask for developers expectations first 5) Migrate your local env once you’ve migrate everything
21. Deployment experience @AuroreMalherbes @MaximeSraiki
22. @AuroreMalherbes @MaximeSraiki
23. @AuroreMalherbes @MaximeSraiki One and only tool
24. @AuroreMalherbes @MaximeSraiki A practical example A developper updates the nginx configuration with the certificates path As we were created our first web interface, we decided to setup https endpoints He does this modification only on staging via ssh and doesn’t modify ansible roles System is down
25. @AuroreMalherbes @MaximeSraiki
26. @AuroreMalherbes @MaximeSraiki Our new flow Push to master branch Build a docker image Run tests Push it to GCR Deploy new image Rolling update
27. @AuroreMalherbes @MaximeSraiki What would I do differently? Push to master branch Build a docker image Run tests Push it to GCR Deploy new image Rolling update https://github.com/GoogleCloudPlatform/cloud-builders-community/tree/master/helm
28. Debugging experience @AuroreMalherbes @MaximeSraiki
29. @AuroreMalherbes @MaximeSraiki
30. “Pros” creating cons As we were the our own sys admin, we could handle things the far west way We did not know how to monitor traffic requests and could only rely on application logs As we were always able to go and access live logs, we were not very rigourous on logs implementation @AuroreMalherbes @MaximeSraiki
31. Limited alerting We created a simple healthcheck, exposing a HEAD route and pinging it thanks to uptime robot @AuroreMalherbes @MaximeSraiki
32. A lot of tools to understand
33. @AuroreMalherbes @MaximeSraiki
34. One tool to rule them all @AuroreMalherbes @MaximeSraiki
35. One tool to rule them all @AuroreMalherbes @MaximeSraiki
36. What have we planned? @AuroreMalherbes @MaximeSraiki 1) Having all services logs 2) Having server metric 3) Having APM 4) Having external services monitoring 5) Having monitoring on critical application paths
37. What have we done? @AuroreMalherbes @MaximeSraiki 1) Having all services logs 2) Having server metric 3) Having APM
38. What have we done? @AuroreMalherbes @MaximeSraiki 1) Having all services logs 2) Having server metric 3) Having APM
39. Load experience
41. We were discerning limits of our infrastructure We were storing customers identity files and licences images on the file system Our node server was under heavier load everyday Our logs were sent via http request for each occurence We were using more and more crons harvesting our database for recurring tasks @AuroreMalherbes @MaximeSraiki
43. API What have we done? @AuroreMalherbes @MaximeSraiki
44. API What have we done? @AuroreMalherbes @MaximeSraiki
45. API Ingress API Service What have we done? @AuroreMalherbes @MaximeSraiki
46. Forest Admin API Ingress FA Service API Service What have we done? @AuroreMalherbes @MaximeSraiki
47. Forest Admin API Bucket Postgres Database Ingress FA Service API Service What have we done? @AuroreMalherbes @MaximeSraiki
48. Forest Admin API Bucket Postgres Database Cronjob Ingress FA Service API Service What have we done? @AuroreMalherbes @MaximeSraiki
49. Forest Admin API Bucket Postgres Database Cronjob Ingress FA Service API Service What have we done? @AuroreMalherbes @MaximeSraiki
50. Do some maths for your cronjobs @AuroreMalherbes @MaximeSraiki ➔ Before: • 2.5 CPUs for 3 APIs (and 1 BDD) ⇒ 1.12 CPU/API
51. Do some maths for your cronjobs @AuroreMalherbes @MaximeSraiki ➔ Before: • 2.5 CPUs for 3 APIs (and 1 BDD) ⇒ 1.12 CPU/API ➔ After: • 12 CPUs for 9 APIs ⇒ 1.33 CPU/API
52. Do some maths for your cronjobs @AuroreMalherbes @MaximeSraiki ➔ Before: • 2.5 CPUs for 3 APIs (and 1 BDD) ⇒ 1.12 CPU/API ➔ After: • 12 CPUs for 9 APIs ⇒ 1.33 CPU/API • 12 CPUs for 9 APIs and up to 10 crons !!!
53. A cronjob IS an API @AuroreMalherbes @MaximeSraiki
54. Forest Admin API Bucket Postgres Database Cronjob Ingress FA Service API Service What have been changed since? @AuroreMalherbes @MaximeSraiki Push engine PE Service Redis Worker
55. What now ? @AuroreMalherbes @MaximeSraiki
56. 3.75/5 0.5d 5s @AuroreMalherbes @MaximeSraiki
57. Key takeaways @AuroreMalherbes @MaximeSraiki ⇒ You can not plan everything, so plan it! ⇒ One thing at a time ⇒ Doing things right open up new horizons
58. @AuroreMalherbes @MaximeSraiki