The rocky road from monolithic to microservices architecture
summary
Vue.ai evolves from an expensive monolith into dozens of cooperating microservices that serve millions of users daily. The team builds catalog ingestion, image processing that outputs 21,000‑dim vectors, an in‑memory image index and 20–30 ms visual search, plus per‑user personalization glued by REST APIs. Infrastructure uses AWS, SQS and Rabbit queues, Redis and Dynamo for caches, Postgres for truth and occasional GPUs and Lambda. Lessons: prefer asynchronous queues, make services immutable, caching has limits; the redesign proves scalable and operationally complex but successful.
AI-written summary · about
chapters (14)
- 0:00Monolith to microservices intro
- 2:07Product overview monolith
- 6:15Project Gotham breakdown
- 10:06Nightwing image pipeline
- 12:33User events engine
- 16:22API layer caching
- 18:54Migration lessons
- 21:21Containerization challenges
- 22:45Immutable microservices
- 23:28Sync vs async comms
- 24:30Non-containerized tooling
- 25:33Logging with transaction IDs
- 26:59Naming microservices
- 28:54Q&A
related talks
51mMicroservices and Microfrontends: Scaling, Learning, Growing
35mScaling RAG Applications to serve millions of users
25mFrom Millions of Customers to Petabytes of Data: A Journey to Microservices
35mLegacy Monolith to Microservices
29mMonolith to microservices transformation at Broadridge Advisor Solutions
23mMicroservices – Diptanu Choudhury
36mMicroservices Lessons Learned
51mCompetitive retail architecture with microservices