3 millions users in 10 countries - Techincal history of BlaBlaCar

Discover the technical history of BlaBlaCar. 10 years of history, 10 countries, 3 versions...
Any comments ? https://joind.in/talk/view/9349

Transcript

1. 3 MILLIONS USERS IN 10 COUNTRIES TECHNICAL STORY OF BLABLACAR Created by / Olivier Dolbeau @odolbeau
2. 3 MILLIONS USERS ?
3. NO !
4. 5 MILLIONS USERS !
5. THE STORY How Blablacar is born ?
6. A MAN
7. WHO WANTS TO SEE HIS FAMILY...
8. WITHOUT TICKETS...
9. SEEING PLENTY EMPTY CARS...
10. HAD AN IDEA !
11. THE CONCEPT How Blablacar works ?
13. DRIVER: EMPTY CAR ?
14. PASSENGER: LOOKING FOR A VEHICLE ?
15. TRAVEL TOGETHER !
16. LET'S TECH ABOUT TECHNIQUE ! Because it's sexy !
17. V1 Quick & dirty !
18. "DEVELOPMENT" START IN 2004
19. U KNOW JOOMLA ?
20. BUT HAVE YOU MET MAMBO ?
21. DON'T BE JEALOUS !
22. OK THE PROTOTYPE WORKS ! Now let's move !
23. V2 Plain old school PHP
24. DEVELOPMENT START IN JANUARY 2008
25. CONTEXT From scratch Use the same database than the V1
26. SEPTEMBER 2008 Launch in France
27. JANUARY 2010 Launch in Spain
28. JUNE 2011 Launch in UK
29. MAY 2012 Launch in Italy
30. 1 million users use website every month
31. 8 680 Average rpm during last month
32. 329 Average response time during last month (in ms)
33. 4 683 commits on Git (in the 3 last years)
34. 285 588 lines of code...
35. V2 WORKS FINE France (with booking) Spain UK Italy
36. BUT IT'S NOT GOOD ENOUGH.
37. NO CODE CONVENTION
38. HOME MADE CODE
39. NEED MORE DEVELOPERS And still need ! =)
40. NOT SCALABLE To search trip ? SQL !
41. HARD TO MAINTAIN Imagine if YOU have to update the lib.trip.php file !
42. LIB.TRIP.PHP 3770 lines 69 functions 926 lines... for 1 function !
43. IT LOOK LIKE THIS i f ( . . . ) { i f ( ) { f o r ( ) { i f ( ) { } e l s e { } } i f ( ) { f o r ( ) { i f ( ) ; e l s e } } } i f ( ) { } }
45. NO TESTS !
47. BETTER THAN V1 RIGHT ?
48. V3 Let's do beautiful code !
49. CONTEXT
50. BLABLACAR IN EUROPE Less than 2 millions members Online booking in France only 5 countries From 5 (at the begin) to 20 people in tech team today
51. BUSINESS REQUIREMENTS We can't stop the site (yeah, really !) We can't stop improving v2
52. TECHNICAL NEEDS
53. WELL KNOWN FRAMEWORK To hire developers To avoid long learning curve To not reinvent the wheel (i18n, forms, templates, etc.) To code quickly
54. SYMPHONY
57. GOOD CODE Readabability Fiability Speed Scalability
58. READABILITY ? PSR-2 Code reviews
59. FIABILITY ? Unit tests with PHPUnit Functionnal tests with Behat Continous integration with Bamboo
60. SPEED ? Asynchronous job Cache (thanks sf2)
61. SCALIBILITY ? ElasticSearch RabbitMQ
62. DEVELOPMENT START IN MARCH 2012
63. AND AFTER 6 MONTHS...
64. OCTOBER 2012 Launch in Belgium
65. OCTOBER 2012 Launch in Netherlands
66. AUGUST 2012 Launch in Portugal
67. OCTOBER 2012 Launch in Luxembourg
68. OCTOBER 2012 Launch in Poland
69. MIGRATION TO V3
70. OCTOBER 2012 UK
71. NOVEMBER 2012 Italy
72. NOVEMBER 2012 Spain
73. AND OUR LAST COUNTRY... Released in 1 week only
74. APRIL 2013 Launch in Germany
75. WHAT DO YOU THINK ? :)
76. NOT ONLY A PHP REFACTORING !
77. FOR SEARCH ? No more SQL !
78. FOR ASYNCHRONOUS TASKS ? No more SQL !
79. FOR LOCKS ? No more SQL !
80. FOR LOGS ? No more tail on servers !
81. FOR SERVERS ? No more manual installation !
82. AND ALSO Between 5 and 15 deployments / day New Relic for application monitoring S3 for photo EC2 (logstash, newsletters, ...) 30 slots for dev / preprod env
83. SO NOW ?
84. IS V3 BETTER THAN V2 ?
85. AVERAGE RPM DURING LAST MONTH V2: 8 680 V3: 2 790
86. AVERAGE RESPONSE TIME DURING LAST MONTH (IN MS) V2: 329 V3: 278
87. LINES OF CODE V2: 285 588 V3: 199 774
88. NOT REALLY RELEVANT RIGHT ?
89. WHAT DO WE THINK ABOUT SF2 ? To hire developers To avoid long learning curve To not reinvent the wheel (i18n, forms, templates, etc.) To code quickly
90. WHAT DO WE THINK ABOUT OUR CODE Readabability Fiability Speed Scalability
91. READABILITY
92. TOO MANY EVENTS Inside sf2 !
93. WE MADE THIS
95. IT MADE THIS
96. WORKERS, WORKERS, WORKERS ! (you're going to hear about it in the next slides)
97. FIABILITY
98. SHARED DB BETWEEN V2 & V3 Very hard to maintain
99. ARE YOU DBA ?
100. SPEED... IN WORKERS
101. EFFECTIVELY, V3 IS FASTER THAN V2 BUT...
102. WORKERS INSIDE APPLICATION ? We have to boot the whole sf2 kernel ! Lot of code to deploy just for a worker Lot of code = memory leaks
103. NOW WE USE CILEX Still have a DIC (Pimple) Clone a skeleton, write your own logic and you're down Fast to deploy blablacar/worker-lib
104. SCALABILITY
105. 1 - DIRECT ACCESS TO DB The front should not access directly to DB
106. THE FRONT SHOULD NOT ACCESS DIRECTLY TO DB
107. WE'RE WRITTEN JAVA... Simple REST API Multiple backends (MySQL, CouchBase, ElasticSearch, ...) No more direct access to data Normalize data for v3 (and all applications ?) Wait & See. :)
108. 2 - V3 IS TOO MONOLITHIC.
109. IT'S NOT ONLY ABOUT WORKERS !
110. SOA !
111. DECOUPLED APPLICATIONS
112. WORKERS, WORKERS, WORKERS ! mails sms user actions user ips ...
113. BUT ALSO... geoproxy phonenumber waypagg
114. SO... YOU'RE DOWN ?
115. MIGRATE FRANCE TO V3
116. MIGRATE MOBILE TO V3
117. EXPLODE OUR APPLICATION (I really want do to that ! \o/)
118. BECAME AN INTERNATIONAL COMPANY ?
119. QUESTIONS ? Follow me: @odolbeau You want to work with us ? cv@blablacar.com