Unstructured text logs are cheap to write and miserable to query. Structured, sampled, and tiered logging gives you fast answers and a predictable bill.
Structure first, sample second
Every log line is JSON with a stable set of fields. High-volume debug lines are sampled aggressively; errors and audit events are kept at full fidelity.
- Consistent fields across every service for instant correlation
- Tail-based sampling that keeps the traces that matter
- Hot storage for a week, cold object storage after that
Make it queryable
Because the shape is consistent, a single query works across the whole fleet — no per-service parsing rules to maintain.