How to Provide a Reliable Ridesharing Service
summary
Didi faces 500% annual request growth, 400+ cities, and 400+ releases per day and reduces downtime from about 40 to about 22 minutes per month. Capacity testing uses a virtual country simulation that creates virtual passengers, drivers, phone numbers and offset IDs to replay tens of thousands of orders per minute and uncover bottlenecks. Delivery uses a standard monitor library and automated blue‑green/rollback procedures. Traffic is routed by city and region for HA, and a company‑wide reliability competition assigns measurable quotas, penalties and rewards to drive follow‑up fixes.
AI-written summary · about
chapters (13)
- 0:00Start
- 1:48Introduction of DiDi Chuxing
- 3:14Challenges of service reliability in DiDi Chuxing
- 5:19Full-link stress test
- 5:43Stress test - strategies for simulation
- 8:23Stress test-data storage and cleaning
- 10:26Stress test-traffic generation
- 12:30Our Standardization works
- 13:00Standardization for Configuration
- 14:01Standardization for Monitor
- 15:14Standardization for Delivery
- 17:50Multi-clusters challenges
- 18:22Multi-cluster implementation
related talks
40mServiceability under high demand
29mPedidosYa: Behind the Delivery
31mScaling delivery management for 50 teams
56mFrom garage project to the No. 1 Taxi App in Europe
42mHow to stay alive even when others go down:...
23mDevOps at Uber
31mKubernetes on Supporting 1 Million Bike-Taxi Drivers in Indonesia
34mHow to Survive to High Demanding Services