Why did Lyft reject a shared monolithic model serving service?
A shared monolith forced all teams to agree on a single set of Python dependency versions. Upgrading PyTorch or LightGBM for one pricing model would break fraud models, creating blocked deployment queues and ambiguous incident ownership.
Answered in
How Lyft Built LyftLearn Serving: Generated Microservices and Model Self-Tests at ScaleHow Lyft serves millions of ML predictions per second: decoupling data and control planes, generating per-team microservices, and model self-tests.
Read the full analysisOther questions this article answers
More system design & architecture questions
- What is the architectural difference between an API Gateway and a Reverse Proxy?
- When should an architecture use Layer 4 instead of Layer 7 load balancing?
- How do edge ingress layers prevent backend cascading thundering herds?
- What algorithm is best for sticky sessions without breaking even distribution?
- How can a CDN improve performance for dynamic requests?
- What is the hardest part of operating behind a CDN?
- How can a CDN help with traffic surges or DDoS attacks?
- What is the difference between a Push CDN and a Pull CDN?
Every answer on Crashtech is written by the editor of the article it comes from — never auto-summarised. Browse all answers or the System Design & Architecture beat.