My Mom told me that Git doesn't scale

Transcript

1. View Slide

2. Thanks
   for being here
   

   View Slide

3. View Slide

4. github
   

   View Slide

5. github
   Git hosting:
   No longer a pain in the ass
   

   View Slide

6. github
   Git hosting:
   No longer a pain in the ass
   for you.
   Not for us.
   Because, goddamnit,
   if I ever find the guy who invented
   this thing I’m going to hang him from
   

   View Slide

7. Let’s host some
   Git repos!
    file.c
    src
    file.h
    README.md
    COPYING.md
    .git
    Bare Repository
    HEAD
    index
    objects
    refs
   git-daemon
   

   View Slide

8. OK, now about
   the web...
   grit Ruby - Git
   interface
   

   View Slide

9. OK, now about
   the web...
   grit  Bare Repo
    Bare Repo
    Bare Repo
   Ruby - Git
   interface
   

   View Slide

10. 1VM
    grit
    
    storage
    rails app
    

    View Slide

11. nVM
    
    storage
    

    View Slide

12. nVM
    
    storage
    (GFS)
    

    View Slide

13. Rails was making us slow.
    

    View Slide

14. View Slide

15. Literally.
    

    View Slide

16. Time to move to
    Real Hardware
    

    View Slide

17. fileservers
    frontends
    
    db
    

    View Slide

18. fileservers
    frontends
    
    db
    ?????????
    

    View Slide

19. smoke
    

    View Slide

20. View Slide

21. bert
    (binary Erlang term)
    

    View Slide

22. bert
    (binary Erlang term)
    ernie
    (not an acronym)
    

    View Slide

23. 
    chimney
    (Redis)
    frontend
    fileserver
    smoke
    grit
    ernie
    grit
    

    View Slide

24. Horizontal Scaling
    Vertical Scaling
    

    View Slide

25. Horizontal Scaling
    Vertical Scaling
    problums wit them gigabits
    problums wit them gigahurtz
    

    View Slide

26. them
    gigabits
    

    View Slide

27. bummer
    x4180 =
    A LOT.
    

    View Slide

28. NetShard
    
         ...
    alternate
    network
    

    View Slide

29. NetShard
    
         ... 
    alternate
    network
    

    View Slide

30. them
    gigahurtz
    

    View Slide

31. bottleneck:
    grit
    

    View Slide

32. bottleneck:
    grit
    solution:
    git
    shell out to
    

    View Slide

33. bottleneck:
    git
    shell out to
    

    View Slide

34. bottleneck:
    git
    solution:
    git
    shell out to
    shell out to
    

    View Slide

35. bottleneck:
    git
    solution:
    git
    shell out to
    shell out to
    properly
    

    View Slide

36. posix_spawn
    

    View Slide

37. posix_spawn
    Seriously.
    < 1ms
    

    View Slide

38. posix_spawn
    Seriously.
    < 1ms
    The issue is not in
    “shelling out”,
    the issue is in the
    spawned process.
    

    View Slide

39. View Slide

40. GUISE
    

    View Slide

41. GUISE
    GUISE
    

    View Slide

42. GUISE
    GUISE
    GUISE
    

    View Slide

43. GUISE
    GUISE
    GUISE
    ...what?
    

    View Slide

44. View Slide

45. Why don’t we take
    

    View Slide

46. Why don’t we take
    the Git binary...
    

    View Slide

47. Why don’t we take
    the Git binary...
    yeah?
    

    View Slide

48. Why don’t we take
    the Git binary...
    yeah? and compile it as
    

    View Slide

49. Why don’t we take
    the Git binary...
    yeah? and compile it as
    a library
    

    View Slide

50. Why don’t we take
    the Git binary...
    yeah? and compile it as
    a library
    oh... go on...
    

    View Slide

51. Why don’t we take
    the Git binary...
    yeah? and compile it as
    a library
    oh... go on...
    and link that into
    

    View Slide

52. Why don’t we take
    the Git binary...
    yeah? and compile it as
    a library
    oh... go on...
    and link that into
    our server
    

    View Slide

53. Scientific
    Graph™
    

    View Slide

54. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

55. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

56. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

57. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

58. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

59. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

60. View Slide

61. Well, we didn’t think about
    

    View Slide

62. Well, we didn’t think about
    freeing memory, but...
    

    View Slide

63. Well, we didn’t think about
    freeing memory, but...
    THIS IS THE KIND
    OF PROBLEM
    WE COULD SOLVE
    WITH CGI
    

    View Slide

64. Well, we didn’t think about
    freeing memory, but...
    THIS IS THE KIND
    OF PROBLEM
    WE COULD SOLVE
    WITH CGI
    IN 1995
    

    View Slide

65. Scientific
    Graph™
    

    View Slide

66. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

67. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

68. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

69. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

70. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

71. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

72. Memory Usage
    Time
    Scientific
    Graph™
    

    View Slide

73. What do you mean
    the server died?
    

    View Slide

74. die("BUG: non-INDEX attr direction
    in a bare repo");
    What do you mean
    the server died?
    

    View Slide

75. die("BUG: non-INDEX attr direction
    in a bare repo");
    die("a bad revision is needed");
    What do you mean
    the server died?
    

    View Slide

76. die("BUG: non-INDEX attr direction
    in a bare repo");
    die("a bad revision is needed");
    die("'%s' is not a valid
    branch name.", name);
    What do you mean
    the server died?
    

    View Slide

77. die("BUG: non-INDEX attr direction
    in a bare repo");
    die("a bad revision is needed");
    die("'%s' is not a valid
    branch name.", name); die("Empty patch.
    Aborted.");
    What do you mean
    the server died?
    

    View Slide

78. die("BUG: non-INDEX attr direction
    in a bare repo");
    die("a bad revision is needed");
    die("'%s' is not a valid
    branch name.", name); die("Empty patch.
    Aborted.");
    die("unable to read index file");
    What do you mean
    the server died?
    

    View Slide

79. libgit
    

    View Slide

80. libgit2
    the “2” means this one
    frees memory
    

    View Slide

81. libgit2
    the “2” means this one
    frees memory
    NOT ENOUGH
    ABSTRACT
    FACTORIES
    

    View Slide

82. JGit
    the “J” means this one
    is in Java
    ...not our thing.
    

    View Slide

83. Java
    a brief timeline
    New companies
    don’t use Java
    because it’s
    not like Unix
    1995
    New companies
    use Java
    because it’s
    new and shiny
    1997
    New companies
    don’t use Java
    because it’s
    ooooooold
    2005
    New companies
    use the JVM
    because
    2011
    

    View Slide

84. Java
    a brief timeline
    New companies
    don’t use Java
    because it’s
    not like Unix
    1995
    New companies
    use Java
    because it’s
    new and shiny
    1997
    New companies
    don’t use Java
    because it’s
    ooooooold
    2005
    New companies
    use the JVM
    because
    2011
    github
    

    View Slide

85. If you think you understand
    the JVM, you are either:
    

    View Slide

86. If you think you understand
    the JVM, you are either:
    a) Very smart
    

    View Slide

87. If you think you understand
    the JVM, you are either:
    a) Very smart
    b) Very wrong
    

    View Slide

88. If you think you understand
    the JVM, you are either:
    a) Very smart
    b) Very wrong
    

    View Slide

89. Some people think
    that
    github
    is a
    Rails shop
    Ruby shop.
    or even a
    

    View Slide

90. Some people think
    that
    github
    is a
    Rails shop
    Ruby shop.
    or even a
    github
    is a
    Unix shop
    and everything else is
    just a detail.
    

    View Slide

91. libgit2
    So,
    

    View Slide

92. libgit2
    a brief timeline
    Shawn
    Pearce
    The Past
    

    View Slide

93. libgit2
    a brief timeline
    Shawn
    Pearce
    myself
    The Past
    

    View Slide

94. libgit2
    a brief timeline
    Shawn
    Pearce
    myself
    myself
    (about to have a
    mental breakdown)
    The Past
    

    View Slide

95. libgit2
    a brief timeline
    Shawn
    Pearce
    myself
    myself
    (about to have a
    mental breakdown)
    myself
    (having a mental
    breakdown)
    The Past
    

    View Slide

96. libgit2
    a brief timeline
    Shawn
    Pearce
    myself
    myself
    (about to have a
    mental breakdown)
    myself
    (having a mental
    breakdown)
    myself
    (reaching Git nirvana)
    The Past
    

    View Slide

97. libgit2
    a brief timeline
    Shawn
    Pearce
    myself
    myself
    (about to have a
    mental breakdown)
    myself
    (having a mental
    breakdown)
    myself
    (reaching Git nirvana)
    The Past
    Russell
    Belfer
    Carlos
    Martín
    Michael
    Schubert
    Ben
    Straub
    real contributors
    

    View Slide

98. libgit2
    a brief timeline
    ?
    

    View Slide

99. libgit2
    a brief timeline
    ?
    

    View Slide

100. libgit2
     a brief timeline
     ?
     1.0 release
     

     View Slide

101. libgit2
     

     View Slide

102. libgit2
     

     View Slide

103. Good Heavens,
     just look at the
     time.
     It’s NoSQL o’clock
     NoSQL
     NoSQL NoSQL
     NoSQL
     NoSQL NoSQL
     NoSQL
     NoSQL
     

     View Slide

104. View Slide

105. ...do you even
     

     View Slide

106. ...do you even
     mongo?
     

     View Slide

107. Key-Value
     Stores
     The Magic of
     If you wish upon a star,
     and have a pure heart...
     

     View Slide

108. Key-Value
     Stores
     The Magic of
     If you wish upon a star,
     and have a pure heart...
     Anything can be
     a Key-Value store!
     

     View Slide

109. id name state lat
     13 San Francisco CA 24
     24 Phoenix AZ 33
     7 Denver CO 40
     8 Caribou ME 47
     2 Los Angeles CA 22
     SELECT * FROM CITIES
     WHERE name = ‘San Francisco’
     Key-Value
     Stores
     The Magic of
     

     View Slide

110. Git is queried like
     a Key-Value Store
     But it is not a
     Key-Value store
     git show
     f3c896c1949476e85abc0d75bb2143656a9580a6
     

     View Slide

111. a b r i e f i n t r o d u c t i o n
     t o t h e G i t d a t a m o d e l
     

     View Slide

112. View Slide

113. View Slide

114.  file.c
      src
      file.h
      README.md
      COPYING.md
     

     View Slide

115.  file.c
      src
      file.h
      README.md
      COPYING.md
     tree
     src/
     README.md
     COPYING.md
     tree
     file.c
     file.h
     blob
     blob
     blob
     blob
     

     View Slide

116. commit
     parent
     tree T
     metadata
     

     View Slide

117. commit
     T
     commit
     T
     commit
     T
     commit
     T
     commit
     T
     commit
     T
     Behold,
     a graph.
     

     View Slide

118. View Slide

119. Well that was easy.
     

     View Slide

120. View Slide

121. master
     Oh
     god
     kill
     me
     

     View Slide

122. Li le known
     torture
     methods:
     

     View Slide

123. View Slide

124. warning:
     the rabbit
     hole is
     pretty deep
     

     View Slide

125. View Slide

126. warning:
     git totally
     wasn’t designed
     for this
     

     View Slide

127. Git doesn’t give
     a #!%$ about CAP
     

     View Slide

128. View Slide

129. Number of hops on a complex query
     1,000,000
     

     View Slide

130. Number of hops on a complex query
     1,000,000
     Required hops for a successful query
     1,000,000
     

     View Slide

131. Number of hops on a complex query
     1,000,000
     Required hops for a successful query
     1,000,000
     Replica count to ensure 100% availability
     a metric shitton
     

     View Slide

132. We could fix it.
     

     View Slide

133. We could fix it.
     But we won’t.
     

     View Slide

134. GitRPC
     

     View Slide

135. GitRPC
     Less.
     

     View Slide

136. GitRPC
     Rugged
     libgit2
     server
     Ruby
     Ruby
     C
     

     View Slide

137. 
     chimney
     (Redis)
     frontend
     fileserver
     smoke
     grit
     ernie
     grit
     GitRPC GitRPC
     

     View Slide

138. 
     chimney
     (Redis)
     frontend fileserver
     GitRPC GitRPC
     server
     client
     

     View Slide

139. New
     serialization
     protocol
     Banana
     Pack
     MessagePack
     + more
     

     View Slide

140. New
     serialization
     protocol
     Banana
     Pack
     MessagePack
     + more
     mochilo
     

     View Slide

141. New
     serialization
     protocol
     Banana
     Pack
     MessagePack
     + more
     mochilo
     Banana Phone
     

     View Slide

142. evolutionary
     (disappointing?)
     

     View Slide

143. Summary:
     

     View Slide

144. Ruby
     C
     Unix
     Unix
     Unix
     Unix
     Unix
     Unix
     Unix
     Unix
     Boring
     Boring
     Boring
     Boring
     Boring
     Summary:
     

     View Slide

145. Use the
     most reliable
     tools you know.
     

     View Slide

146. Challenge yourself to build
     the simplest thing.
     Not because it’s easy,
     but because it works.
     

     View Slide

147. Innovate
     where it really ma ers.
     

     View Slide

148. revolutionary
     product
     create a
     revolutionary
     backend.
     not a
     

     View Slide

149. View Slide

150. Q: Does Git scale?
     

     View Slide

151. Q: Does Git scale?
     A: Who cares?
     

     View Slide

152. View Slide