Netflix Recommendations: Detailed Component Design

Step 7: Detailed Component Design

Step 6 built the tables. This step goes inside each component that reads and writes them. We start with the job that fills the busiest one. It runs before any user opens the app.

Recommendation Computation Workflow

1. Offline Batch Recommendation Generation

The offline pipeline does the heavy computation in this design. It builds personalized recommendations from the full history of data. That happens in bulk, on a schedule. Nothing here has to be fast. Here is how one run works.

A pipeline job runs on a cluster first