Tinder: Using the Elastic Stack to Make Connections Around the World

Tinder relies on the Elastic Stack to analyze, visualize, and predict not only which people a user will swipe right on, or which people will swipe right on that user, but when there’s a mutual swipe match. Hear how the service is growing into a global platform for social discovery in many facets of life.

Maria Zhang l VP of Engineering l Tinder

Transcript

1. Maria Zhang VP of Engineering, Tinder March 8, 2017
2. IT’S A MATCH! Tinder and Elastic liked each other.
4. 17,000,000,000 event counts daily 280,000,000+ Elasticsearch queries daily 100,000,000,000+ Elasticsearch queries annually Nearly 6,000,000 Elasticsearch queries just during the time I’ve been presenting to you today ELASTIC SEARCHES
5. 17,000,000,000 event counts daily 280,000,000+ Elasticsearch queries daily 100,000,000,000+ Elasticsearch queries annually Nearly 6,000,000 Elasticsearch queries just during the time I’ve been presenting to you today ELASTIC SEARCHES
6. 17,000,000,000 event counts daily 280,000,000+ Elasticsearch queries daily 100,000,000,000+ Elasticsearch queries annually Nearly 6,000,000 Elasticsearch queries just during the time I’ve been presenting to you today ELASTIC SEARCHES
7. 17,000,000,000 event counts daily 280,000,000+ Elasticsearch queries daily 100,000,000,000+ Elasticsearch queries annually Nearly 6,000,000 Elasticsearch queries just during the time I’ve been presenting to you today ELASTIC SEARCHES
8. Location - SF Gender - Woman Maximum Distance Age Range - 23-33 SEARCHING FOR PEOPLE Name and Age Mode
9. SAN FRANCISCO
10. BI-DIRECTIONAL QUERY
11. “IT’S A MATCH”
12. USER PERCEPTION ENGINEERING REALITY Minimalist UI Global Scale Data Science Machine Learning Bi-Directional Ranking
13. 196 countries 52 languages 1B + swipes daily 20B matches 17B + event counts daily GLOBAL SCALE
14. 60,000 doc hits per query for ranking 20,000 index updates per second 10s ranking dimensions 60 elasticsearch nodes REAL-TIME SEARCH CHALLENGES
15. iPhone Android n g i n x Auth Elasticsearch Clusters (main) Cache E L B A L B Cache Read Federation Integrity Core recs Dispatch Blending Render Hydrating Suppression Elasticsearch Clusters (augmented) Cache Write Data Source1 Augmentation Precomputed Query Builder Inference Post Ranking Events Pipeline Imputation Internal Service External API ARCHITECTURE DIAGRAM
19. Controller Clients Coord ES Cluster PERFORMANCE IS TESTING IS THE KEY FINDER
20. ◦ Less GC pressure (released in 5.1.1) ◦ Improve concurrency for ShardCoreKeyMap (released in 5.1.2) ◦ Acknowledge source filtering in remote reindex (released in 5.2.0) ELASTICSEARCH COMMUNITY
21. FUTURE STATE
22. GROWTH Sep | 2012 Official release Dec | 2012 20,000 downloads Jan | 2013 450,000 downloads Feb | 2014 Best new startup of 2013 Jun | 2014 Launched Moments Jul | 2014 Officially arrived on Android Mar | 2015 Released the paid version of Tinder Sep | 2014 6 billion matches made Jul | 2016 Launch of Sep | 2016 Spotify Release Sep | 2016 Boost launches In select markets Oct | 2016 Introduces Smart Photos & launches Tindev tech blog Nov | 2016 Introduces more Genders through Inclusive update