Cheating Gall's Law: MediterraneaJS edition

Or, how we split a monolith and lived to tell the tale. This is the story of how npm replaced its monolithic registry service with microservices. We rolled out the registry without any users noticing this huge change behind the scenes. How did we manage this?

TLDR: Put a proxy on it!

Transcript

1. Cheating Gall's Law
2. C J Silverio director of engineering, @ceejbot
3. Cheating Gall's Law How we split a monolith and lived to tell the tale
4. Gall's Law A complex system that works is invariably found to have evolved from a simple system that worked.
5. Gall's Law in full “A complex system that works is invariably found to have evolved from a simple system that worked. A complex system designed from scratch never works and cannot be patched up to make it work. You have to start over with a working simple system.”
6. "A simple system may or may not work."
7. Systemantics: How Systems Really Work and How They Fail
8. starting simple often means starting with a monolith
10. monolith everything in one process
12. npm's monolith: embedded in couchdb
13. monoliths work just fine
14. whatever it takes to build a system that satisfies your users
15. success! now scale it.
17. scaling monoliths many copies of the full thing
19. Exponential growth of node resulted in exponential growth of the npm registry
20. exponential monoliths were going to be expensive
21. splitting the monolith
22. yay microservices?
23. just rewrite everything! what's the problem?
24. Your monolith is complex. A split system is more complex.
25. what did that Gall guy say about complex working systems?
26. "If a system is working, leave it alone. Don't change anything."
27. Gall's Law warns us about that rewrite
28. how do you split a monolith successfully?
29. Let's cheat.
30. Q: How do you cheat? A: By not rewriting the whole thing.
31. slice off a part of the system into a module with a clearly-defined interface
32. then write a second implementation of that interface
33. send requests to that second implementation with a proxy
38. now let's make our proxy smart Replace Varnish with a node service
41. first step just send everything through
42. record what the proxy does how do people use your service? measure performance
43. second step divide & conquer
46. beat Gall's Law with modularity aka information hiding
47. each service is a module a simple, testable system when viewed in isolation
48. best part: now you can change everything
49. your platform your database
52. why stop there?
55. the awesome
56. Each piece is a simple system
57. more scaling dials to turn finer-grained perf data
58. isolating tasks let us debug even without splitting the monolith
59. the system overall is complex but loosely coupled
60. you have a working system every step of the way
61. a proxy service lets you flip back & forth between implementations
62. the not awesome
63. now you've got a distributed system
64. @architectclippy says “I see you have a poorly structured monolith. Would you like me to convert it into a poorly structured set of microservices?”
65. Second system syndrome "this time we'll do it right"
66. Generalize only when forced
67. let's recap! how do you survive a rewrite?
68. simple working service first scale it later
69. respect Gall: rewrite in pieces
70. use a proxy to divide & conquer
71. Be bold you can change your system
72. Nobody noticed when we changed the entire npm registry
73. loves you npm install -g npm@latest