Skip to content

Latest commit

 

History

History
25 lines (19 loc) · 1.38 KB

design_en.md

File metadata and controls

25 lines (19 loc) · 1.38 KB

Web single pattern

Usecase

  • When you want to quickly release the predictor in the simplest architecture.

Architecture

The web single pattern is an architecture that packs all the artifacts for prediction model in a web server. Since the single server REST (or GRPC) interface, preprocess, and trained model in one place, you can create and deploy as a simple predictor.
If you want to deploy multiple replications, you need to deploy with a load balancer or proxy. In case you are using GRPC for the interface, you need to consider client side load balancing or layer-7 load balancer.
To build your model into the web server, you can apply either model-in-image pattern or model-load pattern.

Diagram

diagram

Pros

  • Able to use one programming language, such as Python, for the web server, preprocess and inference.
  • Easy to manage with its simplicity.
  • Troubleshooting will not be complex.
  • Suggested to start with deploying in the web single pattern to deploy a model in synchronous system.

Cons

  • Since all components are packed in a server or docker image, applying a small patch will require whole update.

Needs consideration

  • Update and maintenance procedure for each component.
  • Scale change management of the web server.