Batch pattern

Usecase

Architecture

If you don’t have to run the prediction real-time, you may choose the batch pattern to run the prediction regularly. You can schedule a batch prediction to a bulk of data regularly, such as daily, and store the results. Of course it can be hourly or monthly depending on your usecase. The pattern requires job management server to trigger the batch job. The server will launch the job by your defined rule. The prediction server will be deployed with the batch job launch. If you are using cloud service or Kubernetes, starting and deleting the server based on the job will enable your to cost reduction.

Diagram

diagram

Pros

Cons

Needs consideration