QuestionQ82

Custom Model Serving

A Machine Learning Engineer is developing an application that needs low-latency data lookups in response to a user’s question after a RAG-based search. They want to ensure users receive the most recent data possible for urgent requests, so the data must be no more than a few minutes late. The underlying data is a large table that can contain hundreds of gigabytes of data.

Which data-serving approach best suits this use case?

  • A Online tables with snapshot sync mode
  • B A fast database hosted in MLflow model serving
  • C Online tables with continuous sync mode
  • D Online tables with triggered mode and a time series key
Explanation

Continuous sync uses a streaming pipeline to incrementally update an online table as new data is written to the source table. It is designed for near-real-time freshness and low-latency online lookups, whereas snapshot synchronization is a one-time copy and triggered synchronization depends on explicit scheduled or manual updates.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!