Reduce metrics cost without losing the signal needed for incidents.

Instruction: Choose labels and explain where high-cardinality detail belongs.

Context:

A fictional HTTP metric has labels for method, route template, status class, user_id and request_id. Each request_id is unique. Metrics storage and query latency grow rapidly. Responders need service-level alerts and a way to investigate one failed request.

Updated

Official answer available

Read the opening below, then unlock the full answer and practical guidance.

I would remove unbounded user and request identifiers from the metric labels and retain bounded dimensions such as route template, method and status class...

Upgrade to view official answer

Related Questions