Hitchhiker's Guide to Serverless - Node Interactive

Transcript

1. The Hitchhiker’s Guide to “Serverless” Javascript Steve Faulkner, Bustle
2. Steve Faulkner @southpolesteve
6. Director of Platform Engineering
7. hype! serverless 101
 why? why not? serverless @ bustle tools
9. The people who are serving websites on Lambda or something are not only doing it wrong, they are wasting money while locking themselves in to an architecture that doesn't fit their app - @scrollaway on HN
10. doesn’t scale not production ready too slow too expensive
12. serverless for 99% of user requests
13. 10-20 million calls per day
14. < 200ms
15. < $$$
16. Serverless 101
17. there are still servers…
18. …but what if we abstracted them away?
19. as a Service
20. Where does the business logic go?
23. Functions as a Service
24. run only when needed
25. pay only when running
26. don’t bother me with details
27. request myApp() response
31. re:invent happening right now…
32. app.get(‘/‘, sayHello) function sayHello(req,res){
 res.end(‘hello!’) }
33. app.get(‘/‘, sayHello) function sayHello(req,res){
 res.end(‘hello!’) }
34. app.get(‘/‘, sayHello) function sayHello(req,res){
 res.end(‘hello!’) } Functions Routing
35. Functions-aaS + Routing-aaS
36. Lambda + API Gateway
37. Lambda
39. write code zip upload fn()
40. node java python
41. node shim + rust/go/haskell
42. function(event,context,cb){ cb(null, ‘Hello World’) }
43. versioning aliases RAM + CPU logging non-http events
44. API Gateway
46. click around a lot
47. cache/throttle authorization api keys logging/metrics custom domains ssl swagger import/export
48. why serverless?
50. ops
51. 2011 startup ops: heroku down? get lunch!
52. 2016 startup ops:
 have a plan
53. serverless ops: LOL! WAT SERVERS?
54. serverless ops: LOL! WAT SERVERS?
55. #noops
56. “serverless” ops: have a plan
57. benchmarks
 fallbacks load testing monitoring
 #lessops
58. scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale
59. 1 -> 1000 RPS
60. 1000 -> 0 RPS
61. iteration
62. single function deployments
63. <1s deploys
64. $$$$
65. events API
66. ec2 ~$2500/mo
67. lambda api gateway ~$400/mo
68. why not?
69. lock in
70. const express = require(‘express’) const fn = require(./lambda-fn/) const app = express() app.get('/', (req, res) => { fn.handler(null, null, (err, result) => { res.json(result) }) }); app.listen(3000)
71. cold functions
72. const data = loadBigData() // Takes 15s module.exports.handler = function(e,ctx,cb){ processBigData(data, cb) }
73. ~3.5 hours - Eric @ iopipe.com
74. long tasks
75. 5 min max
76. s3(3GB) Lambda Kinesis Firehose Elasticsearch
77. testing
78. test in the cloud?
80. frontend
82. API Gateway S3 Lambda(SSR) html + js
83. backend
85. API Gateway Lambda JSON dynamodb redis
86. tools!!!
87. serverless framework node-lambda apex claudia.js
89. npm install -g shep
90. JavaScript Only no CF,terraform
 swagger
 webpack
 environments
91. shep new my-api shep generate endpoint shep deploy production
92. coming soon… shep server shep docs shep logs —stream
93. questions? Steve Faulkner @southpolesteve