Knee-Deep Into P2P: A Tale of Fail (non-Elixir)

The slides for my talk at Lambda Days 2018 (www.lambdadays.org/lambdadays2018)

Transcript

1. Knee-Deep Into P2P
   A Tale of Fail
   @fribmendes
   

   View Slide

2. Knee-Deep Into P2P
   A Tale of Fail
   @fribmendes
   

   View Slide

3. “This is SOOO boring…”
   - me, circa last year or smth
   

   View Slide

4. View Slide

5. View Slide

6. View Slide

7. “This is SOOO boring…”
   - me, 10 months ago
   

   View Slide

8. View Slide

9. View Slide

10. View Slide

11. View Slide

12. View Slide

13. “
    ”
    - me, whenever I get to play
    with distributed systems
    

    View Slide

14. View Slide

15. I don’t know how to
    smart office
    

    View Slide

16. I know how to
    web development
    

    View Slide

17. I know how to
    web development
    … what now?
    

    View Slide

18. View Slide

19. Step 1: receive new connections
    

    View Slide

20. Step 1: receive new connections
    Step 2: accept and send messages
    

    View Slide

21. Step 1: receive new connections
    Step 2: accept and send messages
    Step 3: do a bunch of Steps 1 and 2
    

    View Slide

22. Step 1: receive new connections
    

    View Slide

23. View Slide

24. defmodule Gossip.Server do
    def listen(pid, port) do
    {:ok, server_socket} = :gen_tcp.listen(port, @socket_opts)
    accept_loop(pid, server_socket)
    end
    defp accept_loop(pid, server_socket) do
    {:ok, client} = :gen_tcp.accept(server_socket)
    :inet.setopts(client, [active: true])
    :gen_tcp.controlling_process(client, pid)
    Gossip.accept(pid, client)
    accept_loop(pid, server_socket)
    end
    end
    

    View Slide

25. defmodule Gossip.Server do
    def listen(pid, port) do
    {:ok, server_socket} = :gen_tcp.listen(port, @socket_opts)
    accept_loop(pid, server_socket)
    end
    defp accept_loop(pid, server_socket) do
    {:ok, client} = :gen_tcp.accept(server_socket)
    :inet.setopts(client, [active: true])
    :gen_tcp.controlling_process(client, pid)
    Gossip.accept(pid, client)
    accept_loop(pid, server_socket)
    end
    end
    

    View Slide

26. Step 1: receive new connections
    Step 2: accept and send messages
    

    View Slide

27. View Slide

28. defmodule Gossip.Worker do
    def recv_loop(pid, socket) do
    continue = receive do
    {:tcp, _port, msg} ->
    Gossip.recv(pid, msg)
    true
    {:tcp_closed, port} ->
    :gen_tcp.close(port)
    Gossip.disconnect(pid, self())
    false
    {:send, msg} ->
    :gen_tcp.send(socket, msg)
    true
    end
    continue and recv_loop(pid, socket)
    end
    end
    

    View Slide

29. defmodule Gossip.Worker do
    def recv_loop(pid, socket) do
    continue = receive do
    {:tcp, _port, msg} ->
    Gossip.recv(pid, msg)
    true
    {:tcp_closed, port} ->
    :gen_tcp.close(port)
    Gossip.disconnect(pid, self())
    false
    {:send, msg} ->
    :gen_tcp.send(socket, msg)
    true
    end
    continue and recv_loop(pid, socket)
    end
    end
    

    View Slide

30. Step 1: receive new connections
    Step 2: accept and send messages
    Step 3: do a bunch of Steps 1 and 2
    

    View Slide

31. Raspberry Pi #1 Raspberry Pi #2
    

    View Slide

32. View Slide

33. View Slide

34. “Does it scale?”
    

    View Slide

35. View Slide

36. Time to copy off the internet
    find a creative solution
    

    View Slide

37. View Slide

38. g
    

    View Slide

39. Gnutella
    

    View Slide

40. Gnutella
    

    View Slide

41. Gnutella
    

    View Slide

42. Gnutella
    

    View Slide

43. Gnutella
    

    View Slide

44. g
    (gnutella2)
    

    View Slide

45. Gnutella
    

    View Slide

46. G2/Gnutella2
    

    View Slide

47. G2/Gnutella2
    

    View Slide

48. G2/Gnutella2
    

    View Slide

49. G2/Gnutella2
    

    View Slide

50. View Slide

51. View Slide

52. View Slide

53. View Slide

54. View Slide

55. HyParView
    

    View Slide

56. View Slide

57. View Slide

58. View Slide

59. View Slide

60. View Slide

61. View Slide

62. Plumtrees
    

    View Slide

63. Optimal number of messages
    

    View Slide

64. But you can’t afford to lose nodes
    

    View Slide

65. View Slide

66. View Slide

67. View Slide

68. View Slide

69. View Slide

70. View Slide

71. View Slide

72. View Slide

73. View Slide

74. View Slide

75. “Aha! It works on my computer!”
    

    View Slide

76. “Great but we need
    something to show”
    

    View Slide

77. “Great but we need
    something to show”
    (aka Raspberry Pi time)
    

    View Slide

78. “Sure.
    Seems legit…”
    — @iampfac
    

    View Slide

79. “Wait.
    He works here!?”
    — @naps62
    

    View Slide

80. “Hey, I can borrow™ someone else’s code”
    

    View Slide

81. View Slide

82. View Slide

83. View Slide

84. you shall not pass!
    

    View Slide

85. Solution: stick everything on Raspberry Pi’s
    

    View Slide

86. Things running on one Raspberry Pi
    

    View Slide

87. Things running on one Raspberry Pi
    ✓BEAM
    

    View Slide

88. Things running on one Raspberry Pi
    ✓BEAM
    ✓thebox (sensors)
    

    View Slide

89. Things running on one Raspberry Pi
    ✓BEAM
    ✓thebox (sensors)
    ✓Phoenix app
    

    View Slide

90. Things running on one Raspberry Pi
    ✓BEAM (x2)
    ✓thebox (sensors)
    ✓Phoenix app
    

    View Slide

91. Things running on one Raspberry Pi
    ✓BEAM (x2)
    ✓thebox (sensors)
    ✓Phoenix app
    ✓Postgres
    

    View Slide

92. Things running on one Raspberry Pi
    ✓BEAM (x2)
    ✓thebox (sensors)
    ✓Phoenix app
    ✓Postgres
    ✓Cassandra
    

    View Slide

93. Things running on one Raspberry Pi
    ✓BEAM (x2)
    ✓thebox (sensors)
    ✓Phoenix app
    ✓Postgres
    ✓Cassandra
    it works!
    

    View Slide

94. View Slide

95. View Slide

96. @antenna = Satellite::Antenna.new(host, port)
    @antenna.on(“data") do |data|
    if data.avg_temp > 25
    slack.send_msg(“people, it's too hot")
    end
    end
    @antenna.on("lights") do |payload, data|
    payload == "on" ? turn_lights_on : turn_lights_off
    end
    @antenna.watch
    

    View Slide

97. “Looking good! Everything’s working!”
    

    View Slide

98. View Slide

99. State of each node:
    

    View Slide

100. State of each node:
     • Last sensor readings
     

     View Slide

101. State of each node:
     • Last sensor readings
     • Network map (MAC-IP)
     

     View Slide

102. State of each node:
     • Last sensor readings
     • Network map (MAC-IP)
     • Target values
     

     View Slide

103. State of each node:
     • Last sensor readings
     • Network map (MAC-IP)
     • Target values
     

     View Slide

104. View Slide

105. How do we handle concurrency?
     

     View Slide

106. View Slide

107. View Slide

108. No database locks.
     No transactions.
     You’re on your own, kiddo.
     

     View Slide

109. Vector Clocks
     

     View Slide

110. View Slide

111. View Slide

112. View Slide

113. View Slide

114. View Slide

115. View Slide

116. View Slide

117. Shared state
     and why it sucks
     

     View Slide

118. Vector = (1, 0)
     Vector = (0, 1)
     

     View Slide

119. CAP Theorem
     

     View Slide

120. CAP Theorem
     “you’re a programmer.
     you can’t have nice things.”
     

     View Slide

121. consistency
     availability partitioning
     

     View Slide

122. consistency
     availability partitioning
     

     View Slide

123. Eventual Consistency
     

     View Slide

124. CRDTs
     

     View Slide

125. Operation-Based CRDT
     

     View Slide

126. Operation-Based CRDT
     commutative but not idempotent
     update exactly once
     

     View Slide

127. no CRDTs
     

     View Slide

128. no CRDTs
     

     View Slide

129. no CRDTs
     

     View Slide

130. no CRDTs
     

     View Slide

131. Op-based CRDTs
     

     View Slide

132. Op-based CRDTs
     

     View Slide

133. Op-based CRDTs
     

     View Slide

134. Op-based CRDTs
     

     View Slide

135. State-Based CRDT
     

     View Slide

136. State-Based CRDT
     commutative and idempotent
     heavier on the network
     

     View Slide

137. State-based CRDTs
     

     View Slide

138. State-based CRDTs
     

     View Slide

139. State-based CRDTs
     

     View Slide

140. State-based CRDTs
     

     View Slide

141. View Slide

142. View Slide

143. View Slide

144. View Slide

145. View Slide

146. View Slide

147. View Slide

148. Wrapping up
     

     View Slide

149. System resources matter
     

     View Slide

150. System resources matter
     your algorithms should
     account for them
     

     View Slide

151. There are models.
     Use them.
     

     View Slide

152. Distributed System Checklist
     

     View Slide

153. Distributed System Checklist
     •Is the number of processes known or finite?
     

     View Slide

154. Distributed System Checklist
     •Is the number of processes known or finite?
     •Is there a global notion of time?
     

     View Slide

155. Distributed System Checklist
     •Is the number of processes known or finite?
     •Is there a global notion of time?
     •Is the network reliable?
     

     View Slide

156. Distributed System Checklist
     •Is the number of processes known or finite?
     •Is there a global notion of time?
     •Is the network reliable?
     •Is there full connectivity?
     

     View Slide

157. Distributed System Checklist
     •Is the number of processes known or finite?
     •Is there a global notion of time?
     •Is the network reliable?
     •Is there full connectivity?
     •What happens when a process crashes?
     

     View Slide

158. It really doesn’t change
     that much
     

     View Slide

159. CRDTs aren’t a golden hammer
     

     View Slide

160. Reinventing the wheel is stupid
     

     View Slide

161. View Slide

162. Knee-Deep Into P2P
     A Tale of Fail
     @fribmendes
     

     View Slide