Use it when
- You want a general-purpose lightweight library to log metrics about your system available in many languages.
- You want a server that automatically collects and stores metrics data from any number of targets.
- You want a long list of tools that export existing metrics as Prometheus metrics with out-of-the-box support.
- You want to store metrics locally or have many integration options for remote storage.
Watch out
- You will likely use Prometheus with complementary tools that help calculate data and model metrics. You may also use a tool to create dashboards - Grafana is a popular option.
- If tools in your ML stack automatically collect metrics and provide support for exporting them to Prometheus, integration is easy. Otherwise, you will need to define which metrics to log and add custom code to log that data.
Example stacks
Airflow + MLflow stack