Keeping Infrastructure Costs Sane While DoorDash Kept Growing
As order volume climbed, DoorDash treated infrastructure cost as an engineering metric teams could see and act on, not just a line on a finance report.
Engineering at scale
11 articles · 43 min of reading
As order volume climbed, DoorDash treated infrastructure cost as an engineering metric teams could see and act on, not just a line on a finance report.
How DoorDash extracted its logistics and dispatch systems from a Django monolith first, and built a service platform to avoid microservices sprawl.
DoorDash built an internal experimentation platform, Curie, so any team can run statistically sound A/B tests without a data scientist gatekeeping every test.
How DoorDash designs for graceful degradation and load shedding so a spike or partial outage on one side of its marketplace doesn't take down the rest.
DoorDash's search ranking has to satisfy consumer relevance, merchant fairness, and operational reality like store hours and prep time all at once.
How DoorDash built a feature store to serve consistent, low-latency machine learning features across dispatch, search, and fraud detection.
How DoorDash built a unified, Kafka-based event pipeline named Iguazu to feed analytics and machine learning from a sprawling microservices fleet.
Inside DoorDash's dispatch system, which uses machine learning to balance delivery time, cost, and fairness when assigning orders to Dashers.
Behind DoorDash's live delivery map is a pipeline that balances Dasher battery life against location freshness across millions of concurrent deliveries.
DoorDash adopted CockroachDB to get horizontal scalability and strong consistency for order and payment data without the operational pain of manual sharding.
As DoorDash broke apart its Python monolith, it picked Kotlin over Go and plain Java for the new microservices, and ran the cutover service by service.