Introduction to Elixir

Why am I excited about it? A Ruby developer's perspective.

This was presented to colleagues at Deliveroo internally.

Transcript

1. Introduction to Elixir Why am I excited about it?
2. Please ask questions
3. μs
6. Production
10. Tabs vs Spaces
13. • dynamic • functional • concurrent • compiled Elixir is..
14. • distributed • scalable (horizontal and vertical) • fault-tolerant • highly resilient What can I make with it?
16. Erlang VM (BEAM) Proven record since 1986
17. Erlang VM (BEAM) Almost as old as I am
18. • Amazon • AOL • Basho • Bet365 • Chef • Ericsson • Facebook • GitHub • Goldman Sachs • Huffington Post • Klarna • Rackspace • Rakuten • WhatsApp • Whisper • Yahoo! • …. Companies using Erlang source: Wikipedia entry for Erlang
19. • Alpha Sights • Bleacher Report • BT • Curadora • Discored • DockYard • Discord • Puppet Labs • Pinterest • Thoughtbot • Brightcove • Evercam • Pronto • …. Companies using Elixir source: github.com/doomspork/elixir-companies
20. Elixir extends Erlang • Call Erlang functions directly • Use Erlang libraries • Elixir is a superset of Erlang
21. Elixir additions Syntax • Uniform • more accessible • personal preference*
22. Elixir additions Protocols • contract for manipulating data • Data polymorphism
23. Elixir additions Metaprogramming • Macros • Compile time
24. Elixir additions Records datatype • native pattern matching
25. Elixir additions Better structure for code • Namespaces • private functions
26. Elixir additions Consistent APIs • Zero based access • Naming of functions
27. Elixir additions Less boilerplate • Simplified creating GenServers • created shortcuts for menial tasks
28. Elixir additions Better String handling • Introduced String type • UTF-8 binaries
29. Elixir additions Better tooling
30. Elixir additions |> operator
31. – Joe Armstrong “ It's not Erlang and it's not Ruby and it has ideas of its own.”
32. – everyone in the room “That’s all well and good but why all the hype about Elixir?”
33. Scalability Vertical scaling has a limit
34. Mhz 0.1 1 100 100000 4004 8086 i386 P5 Netburst Nehalem Haswell Intel CPU clock rate
35. Scalability (Horizontal) Increasing number of CPU cores
36. Scalability (Horizontal) Multiple hardware instances
37. Scalability (Horizontal) We need a way to use all available resources efficiently
38. – Faheem, 2 years ago “Oh I know, I will run multiple instances of Ruby VM”
39. Ruby VMs Wasteful on memory
40. Ruby VMs Could be blocked on I/O, while CPU idles by
41. BEAM (Erlang VM)
42. Solution (BEAM) •Distributed •Concurrent •Parallel
43. Solution (BEAM) • Distributed •Concurrent •Parallel
44. Solution (BEAM) •Distributed • Concurrent •Parallel
45. – Faheem (while shamelessly stealing from Rob Pike) “Concurrency is the bread and butter of Elixir. Parallelism is just a nice side effect”
46. So how does Elixir do it?
48. Not OS processes Elixir Processes are..
49. Not threads either Elixir Processes are..
50. Much lighter in weight Elixir Processes are..
51. Completely isolated in execution time and space Elixir Processes are..
52. Elixir Process.. No shared state on launch
53. Elixir Process.. Communication is through messages only
54. Elixir Process.. All messages are async
55. Elixir Process.. No guarantees on message order
56. Elixir Process.. No way to directly access data
57. Elixir Process.. Responses are also just messages to sender's inbox
58. Elixir Process.. What does all that give us?
59. Elixir Process.. 1. Easier to reason about system
60. Elixir Process.. 2. Isolated crashes
61. Elixir Process.. 3. Inherently distributed system
62. Elixir Process.. 4. Clean consistent communication method for local or remote processes
63. Elixir Process.. 5. No need for stop-the-world garbage collection
64. Elixir Process.. 6. Much easier for process scheduler
65. Elixir Process.. Lastly, simple and solid building block
66. “Concurrency is the way to structure things. So (may be) you can use parallelism to do a better job” – Rob Pike (creator of golang)
67. Whatsapp (Erlang VM) • 1 million TCP connections per machine in 2011 • 2.8 million TCP connections in Feb 2012 • 571K packets/sec, > 200K distinct messages/sec • Goal was Resilient against disruptions • 450 million active users with only 32 engineers source: blog.whatsapp.com/196/1-million-is-so-2011
68. Elixir/Phoenix * Sources: www.phoenixframework.org/v1.0.0/blog/the-road-to-2-million-websocket- connections
69. Elixir/Phoenix * Source: www.phoenixframework.org/v1.0.0/blog/the-road-to-2-million-websocket- connections
70. Sent an article to all connected users Elixir/Phoenix
71. Done in under 3 minutes Elixir/Phoenix
72. So, how do I create an elixir process?
73. Create process
74. What happens.. when an isolated process die
75. process
76. Well, VM is there
77. VM can tweet the death note if processes are linked
78. P1 P2 Explicit link P3
79. P1 P2 P3 P4
80. P1 P2 P3 P4 Supervisor
81. P1 P2 P3 P4 Supervisor crashed
82. Erlang mantra Don’t plan against failure
83. No defensive programming
84. Let it crash
86. Plan for failure
87. Plan for recovery Plan for failure
88. Makes for a resilient system
89. P1 P2 P3 P4 Supervisor (One for one) P5 crashed
90. P1 P2 P3 Supervisor (One for one) P5
91. P1 P2 P3 Supervisor (One for one) P5 P0
92. P6 P7 P8 Supervisor (One for one) P0 P9
93. Supervisor
94. P6 P7 P8 P0 P9 P12 P11 P10 Supervisor Supervisor Supervisor
96. Who supervises the supervisor?
97. Applications A set of processes that are started or stopped together
98. Applications A VM can have many applications
99. Elixir • Processes • Data types • Modules
100. Elixir • Processes • Data types • Modules
101. Elixir • Processes • Data types • Modules
102. Modules • Pipe operator (|>) • Pattern matching
103. Modules • Pipe operator (|>) • Pattern matching
104. Pipe operator
105. Pipe operator
106. Pipe operator (2)
107. Pipe operator (3)
108. Pipe operator • Data transformation • Small functions into pipelines • Increase readability
109. Modules • Pipe operator (|>) • Pattern matching
110. Pattern Matching
111. Pattern Matching # Elixir defmodule Fib do def fib(n) do case n do 0 -> 0 1 -> 1 n -> fib(n-1) + fib(n-2) end end end
112. Pattern Matching
113. Pattern Matching (2)
114. Pattern Matching (2)
115. Pattern Matching (2)
116. Pattern Matching (2)
117. Pattern Matching (3)
118. Pattern Matching (4)
119. Pattern Matching (5)
120. Modules • Pipe operator (|>) • Pattern matching
121. Elixir • Processes • Data type • Modules
122. Productivity • Tools • Docs
123. Productivity • Tools • Docs
124. Tools • Mix • IEx • Observer • Packages via hex.pm
125. Mix # Rake and bundler $ mix new foo $ cd foo $ mix test # compile and test $ mix hex.publish # Make and publish a package $ mix hex.docs # Publish docs
126. Tools • Mix • IEx • Observer • Packages via hex.pm
127. Tools • Mix • IEx • Observer • Packages via hex.pm
128. Tools • Mix • IEx • Observer • Packages via hex.pm
129. Tools • Mix • IEx • Observer • Packages via hex.pm
130. Hex.pm • 2_400 packages • 11_000 versions • 1.5 Million downloads this week • 41+ Million downloads ever
131. Productivity • Tools • Docs
132. Productivity • Tools • Docs
133. Docs • hexdocs.pm • Doctest • Good guides • Lots of books
134. Productivity • Tools • Docs
135. Moar stuff
136. Cut because of time • Lazy streams • Process pipelines • GenStage • Protocols
137. Cut because of time • Tests/CI support • Release • Deployment • Hot code update
138. Cut because of time • Phoenix • Channels • Presence • Concurrent tests • No lazy loading associations
139. Case Study
140. Case study • API endpoint (proprietary code) • Authentication • JSON data • Postgres 9.4 • Hosted on Heroku source: http://sorentwo.com/2016/02/02/caching-what-is-it-good-for.html
141. Case study • Ruby/Rails • Two performance M dynos • Elixir/Phoenix • Single 1x Production dyno source: http://sorentwo.com/2016/02/02/caching-what-is-it-good-for.html
142. Case study • Ruby 2.3 / Rails 4.2.5.1 with ActiveRecord • Fronted by a Redis cache • Tuned to fetch as little data as possible from the database • All associations are preloaded • All data is generic, not customized to the current user • The request is paginated to 100 primary records, without a limit on side loads • The payload is a hefty 160k, un-gzipped source: http://sorentwo.com/2016/02/02/caching-what-is-it-good-for.html
143. Case study • Elixir 1.2.1 / Phoenix 1.1 • No entity cache • All fields are fetched from the database, SELECT * • All JSON responses are serialized on the fly, directly in views • Includes customized data based on the current user • The request isn’t paginated at all, there are 250 primary records • The payload is a massive 724k, un-gzipped source: http://sorentwo.com/2016/02/02/caching-what-is-it-good-for.html
144. Case study source: http://sorentwo.com/2016/02/02/caching-what-is-it-good-for.html
145. Case study • Serving up 2.5x the records • With 4.5x the data, • Without any caching • Response times are 1.5x-2.5x faster. source: http://sorentwo.com/2016/02/02/caching-what-is-it-good-for.html
146. Recap
147. Recap.. 1. Elixir is fast
148. Recap.. 2. Elixir does not stop the world for garbage collection
149. Recap.. 3. Elixir can utilise all available resources (CPU/memory)
150. Recap.. 4. Pattern matching is awesome!!!
151. Recap.. 5. Developer happiness might exist beyond Ruby
153. Recap.. 6. We can have both developer happiness and performance
154. Thank you
155. Questions?