Deploying node.js at scale - Maraschi, Collina - Codemotion Amsterdam 2016
2 likes951 views
The document describes a developer's journey in deploying and scaling a Node.js application from a small monolith to a large distributed system using microservices. It details how the developer Bob initially deployed a simple monolithic MEAN stack application but encountered performance issues as traffic grew. Bob then refactored to use modules, microservices, and other distributed patterns to improve scalability, allowing the application and team to continue growing through various challenges.
![NoSQL is hard!
Relational vs. NoSQL
It is not Node fault!
Deduplication is accepted
Don’t shoot in your own foot
{
"name": "Matteo",
"surname": "Collina",
"phone": "+39 347123456",
"emails": [{
"type": "university",
"address": "matteo.collina@unibo.
it"
}]
}](https://image.slidesharecdn.com/c89oabf9srkwwdopjq49-signature-97ac0cba55d83fa6f2a2f253a387270677bf11da9177d0b7d555fb911fc34ba2-poli-160526084214/85/Deploying-node-js-at-scale-Maraschi-Collina-Codemotion-Amsterdam-2016-15-320.jpg)
Deploying node.js at scale - Maraschi, Collina - Codemotion Amsterdam 2016
- 1. Deploying Node.js at scale A developer journey @lucamaraschi @matteocollina
- 2. DISCLAIMER Nobody in this story, and no outfit or corporation, is based upon an actual person or outfit in the real world. The tale we are going to tell is completely fictional, but based upon the combined experiences of the speakers in building Node.js applications.
- 3. Meet bob He is a .LAVA developer He writes applications on the L2EE stack, using SQLobster as a database.
- 4. Bob reads:
- 6. Bob decides: Let’s rewrite my HUGE application in node.js Using the MEAN stack
- 7. .js
- 8. How do I deploy my Javascript monolith? I read StackOverflow! SSH into the virtual machine ● snowflake servers ● git clone ● forever! No Integration tests, but 100% code coverage
- 9. The new product has huge success Bob is happy
- 10. And the performance fall Bob is SAAD
- 11. Such speed So enterprise Wow JS
- 12. But my schema is relational... One MongoDB collection for Student One MongoDB collection for Email populate
- 13. Facts MongoDB has no concept of relationship We need relationships in our application Let’s build a JOIN engine inside our application! Taken from the Mongoose website
- 14. Facts MongoDB has no concept of relationship We need relationships in our application Let’s build a JOIN engine inside our application! Let’s do TWO queries to Mongo If we used find() instead of findOne(), we would have done 1+N queries Taken from the Mongoose website
- 15. NoSQL is hard! Relational vs. NoSQL It is not Node fault! Deduplication is accepted Don’t shoot in your own foot { "name": "Matteo", "surname": "Collina", "phone": "+39 347123456", "emails": [{ "type": "university", "address": "matteo.collina@unibo. it" }] }
- 16. The fixed product has huge success Bob is happy again
- 17. ...but my servers stop responding!!!!
- 18. Such speed So enterprise Wow JS
- 19. Exhausted file descriptors Every incoming HTTP request is a file descriptor Every outgoing HTTP request is a file descriptor 1. You receive an HTTP request call to /faulty 2. In faulty, you call service A on /doSomething 3. You pipe the result of /doSomething to the HTTP response 4. An error happens, and you do not close the HTTP response Fixed amount
- 20. Exhausted file descriptors: solution Replace a.pipe(b) With pump(a, b)
- 21. We are back online! Bob is happy again again
- 22. ● Scaling the development team from 1 (Bob) to 5 ● Delivering more features in less time ● 1000 unit tests! ● Serve more customers! But the Business (always) wants more...
- 23. Such speed So enterprise Wow JS
- 24. The new team does not deliver Bob is SAAD
- 25. MVC architecture Really monolithic Adding a feature means changing all “layers” in an application Does not scale with the team size Equivalent VMs need to be dimensioned to handle the most demanding feature
- 26. Don’t push anything till tomorrow..
- 27. Developing with modules Really monolithic Adding a feature means changing all “layers” in an application Does not scale with the team size up to a certain point Equivalent VMs need to be dimensioned to handle the most demanding feature
- 28. Productivity goes up again! Bob is happy again again again
- 29. ● Scaling the development team from 5 to 20 ● Delivering even more features in less time ● 10000 unit tests! ● Serve more customers! But the Business (always) wants more...
- 30. The NEW new team does not deliver Bob is DESPERATE
- 31. ===
- 34. A pre-made “stack” might not be good
- 35. Let’s throw everything away... But .LAVA42 does it better! ;-)
- 38. Microservices architecture Really monolithic Adding a feature means changing all “layers” in an application Does not scale with the team size up to a certain point Equivalent VMs need to be dimensioned to handle the most demanding feature
- 40. Queue in between How we discover the queues? How are the back channels / response handled?
- 41. The answer to life universe, and everything
- 42. The answer to life universe, and everything The answer to everything is service discovery
- 43. Discovery with DNS: Meet Consul
- 44. Etcd… but CoreOS
- 47. ● Weakly consistent vs. strongly consistent ● Bornt for distributed systems ● Failure detection over the cluster ● Information dissemination ● Equally distribution of the workload Advantages
- 49. Common implementations Hashicorp SERF If you are into GO mrhooray/swim-js If you are into JS ;-)
- 50. But a new cluster node is added...
- 51. How to rebalance the cluster topology?
- 54. PROBLEM How can I find a server over an unstable network? SOLUTION Each server and keys are hashed in a circular hashspace
- 56. Advantages ● Easier to avoid hotspots ● Enable partitioning ● Predictable elasticity ● Easy replication ● Scalability and Availability ● Local state
- 57. mcollina/upring
- 58. What if Bob was working for Uber? ;-)
- 61. The fourth secret of Fatima
- 62. “Hyperbahn enables service discovery and routing for large-scale systems comprised of many microservices. Distributed, fault tolerant, and highly available, it lets one service find and communicate with others simply and reliably without having to know where those services run.”
- 63. Featuring... ● Configuration Discovery ● Retries ● Load Balancing ● Rate Limiting ● Circuit Breaking ● Distributed tracing
- 64. But what about HTTP?
- 66. TChannel is a networking framing protocol used for general RPC, supporting out-of-order responses at extremely high performance where intermediaries can make a forwarding decision quickly
- 67. Tracing as a first class resident... https://github.com/openzipkin/zipkin
- 68. WHY RPC?
- 73. “SERVERLESS” is the new Hipster
- 74. Such servers So hipster Wow functions Very reactive
- 75. “FUNCTIONAL” is the new Hipster
- 76. “DISTRIBUTED FUNCTIONS” are the future!
- 77. Don’t be too late!
- 78. @lucamaraschi
- 81. Conclusions ● Think about choosing a technology ● Node.js works at scale ● You can implement any architecture with Node.js