Catalogue / uber-2023-airport-etr-forecasting

Uber builds modular airport ETR forecasting on Michelangelo and fixes entity-queue Kafka and DocStore bottlenecks

Uber. Airport demand and ETR forecasting system.

Organisation
Uber
Evidence
Reported from production
Source
Demand and ETR Forecasting at Airports

Uber Blog, published 2023-03-23

Records

Decision#d1

Modularize airport ETR into separate supply and demand models plus a backend heuristic instead of direct prediction

Decision#d2

Use near-real-time Palette features and a single non-partitioned MVP training pipeline across all airports

Decision#d3

Switch driver-queue Kafka event emission from emit-on-write to emit-on-peek under client volume limits

Failure#f1

Unbalanced queue_id DocStore partitioning and unscaled entity-queue hosts caused errors, rate limiting, and timeouts

Sign in to read what was chosen, what was rejected and what it cost, each with its place in the source. An account is free.